# STUDYING TREE RESPONSES TO EXTREME EVENTS

EDITED BY: Achim Bräuning, Andreas Bolte, Cristina Nabais, Sergio Rossi and Ute Sass-Klaassen PUBLISHED IN: Frontiers in Plant Science

#### *Frontiers Copyright Statement*

*© Copyright 2007-2017 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

*The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.*

*Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.*

*Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.*

*As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.*

> *All copyright, and all rights therein, are protected by national and international copyright laws.*

*The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use.*

ISSN 1664-8714 ISBN 978-2-88945-192-0 DOI 10.3389/978-2-88945-192-0

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# **STUDYING TREE RESPONSES TO EXTREME EVENTS**

Topic Editors:

**Achim Bräuning,** Friedrich-Alexander-University Erlangen-Nuremberg, Germany **Andreas Bolte,** Thünen Institute of Forest Ecosystems, Germany **Cristina Nabais,** Universidade de Coimbra, Portugal **Sergio Rossi,** Université du Québec à Chicoutimi, Canada & South China Botanical Garden, Chinese Academy of Sciences, China **Ute Sass-Klaassen,** Wageningen University and Research, Netherlands

Cover image by Britta Eilmann & Ute Sass-Klaassen (Wageningen University, The Netherlands) for FPS COST Action FP1106 STReESS

Trees are among the longest-living organisms. They are sensitive to extreme climatic events and document the effects of environmental changes in form of structural modifications of their tissues. These modifications represent an integrated signal of complex biological responses enforced by the environment. For example, temporal change in stem increment integrates multiple information of tree performance, and wood anatomical traits may be altered by climatic extremes or environmental stress. Recent developments in preparative tools and computational image analysis enable to quantify changes in wood anatomical features, like vessel density or vessel size. Thus, impacts on their functioning can be related to climatic forcing factors. Similarly, new developments in monitoring (cambial) phenology and mechanistic modelling are enlightening the interrelationships between environmental factors, wood formation and tree performance and mortality. Quantitative wood anatomy is a reliable indicator of drought occurrence during the growing season, and therefore has been studied intensively in recent years. The variability in wood anatomy not only alters the biological and hydraulic functioning of a tree, but may also influence the technological properties of wood, with substantial impacts in forestry. On a larger scale, alterations of sapwood and phloem area and their ratios to other functional traits provide measures to detect changes in a tree's life functions, and increasing risk of drought-induced mortality with possible impacts on hydrological processes and species composition of plant communities.

Genetic variability within and across populations is assumed to be crucial for species survival in an unpredictable future world. The magnitude of genetic variation and heritability of adaptive traits might define the ability to adapt to climate change. Is there a relation between genetic variability and resilience to climate change? Is it possible to link genetic expression and climate change to obtain deeper knowledge of functional genetics? To derive precise estimates of genetic determinism it is important to define adaptive traits in wood properties and on a whole-tree scale. Understanding the mechanisms ruling these processes is fundamental to assess the impact of extreme climate events on forest ecosystems, and to provide realistic scenarios of tree responses to changing climates. Wood is also a major carbon sink with a long-term residence, impacting the global carbon cycle. How well do we understand the link between wood growth dynamics, wood carbon allocation and the global carbon cycle?

Papers contribution to this Research Topic will cover a wide range of ecosystems. However, special relevance will be given to Mediterranean-type areas. These involve coastal regions of four continents, making Mediterranean-type ecosystems extremely interesting for investigating the potential impacts of global change on growth and for studying responses of woody plants under extreme environmental conditions. For example, the ongoing trend towards warmer temperatures and reduced precipitation can increase the susceptibility to fire and pests.

The EU-funded COST Action STREeSS (Studying Tree Responses to extreme Events: a SynthesiS) addresses such crucial tree biological and forest ecological issues by providing a collection of important methodological and scientific insights, about the current state of knowledge, and by opinions for future research needs.

Funded by the Horizon 2020 Framework Programme of the European Union

**Citation:** Bräuning, A., Bolte, A., Nabais, C., Rossi, S., Sass-Klaassen, U., eds. (2017). Studying Tree Responses to Extreme Events. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-192-0

# Table of Contents


*109 Trait Acclimation Mitigates Mortality Risks of Tropical Canopy Trees under Global Warming*

Frank Sterck, Niels P. R. Anten, Feike Schieving and Pieter A. Zuidema

*119 Interpreting the Climatic Effects on Xylem Functional Traits in Two Mediterranean Oak Species: The Role of Extreme Climatic Events*

Angelo Rita, Marco Borghetti, Luigi Todaro and Antonio Saracino

*130 Complex Physiological Response of Norway Spruce to Atmospheric Pollution – Decreased Carbon Isotope Discrimination and Unchanged Tree Biomass Increment* Vojteˇ ch Cˇada, Hana Šantru˚cˇ ková, Jirˇí Šantru˚cˇek, Lenka Kubištová, Meelis Seedre and Miroslav Svoboda

#### **Assessing frequency and impact of extreme events**

*142 The Imprint of Extreme Climate Events in Century-Long Time Series of Wood Anatomical Traits in High-Elevation Conifers*

Marco Carrer, Michele Brunetti and Daniele Castagneri


Alfredo Di Filippo, Joana Vieira, Cristina Nabais, Vicente Rozas and Giovanna Battipaglia


Cyrille B. K. Rathgeber, Andreas Papadopoulos and Kevin T. Smith *233 Flood-Ring Formation and Root Development in Response to Experimental Flooding of Young* **Quercus robur** *Trees*

Paul Copini, Jan den Ouden, Elisabeth M. R. Robert, Jacques C. Tardif, Walter A. Loesberg, Leo Goudzwaard and Ute Sass-Klaassen


Jean-Luc Maeght, Santimaitree Gonkhamdee, Corentin Clément, Supat Isarangkool Na Ayutthaya, Alexia Stokes and Alain Pierret

*267 Differentiated Responses of Apple Tree Floral Phenology to Global Warming in Contrasting Climatic Regions*

Jean-Michel Legave, Yann Guédon, Gustavo Malagi, Adnane El Yaacoubi and Marc Bonhomme

*280 Rhizophoraceae Mangrove Saplings Use Hypocotyl and Leaf Water Storage Capacity to Cope with Soil Water Salinity Changes*

Silvia Lechthaler, Elisabeth M. R. Robert, Nathalie Tonné, Alena Prusova, Edo Gerkema, Henk Van As, Nico Koedam and Carel W. Windt

*293 Limited Growth Recovery after Drought-Induced Forest Dieback in Very Defoliated Trees of Two Pine Species*

Guillermo Guada, J. Julio Camarero, Raúl Sánchez-Salguero and Rafael M. Navarro Cerrillo

## **New methods and tools to measure and evaluate stress markers in trees**


## **Varying adaptive capacity of populations to drought – facilitation of forest adaptation**

*371 Repeated Summer Drought and Re-watering during the First Growing Year of Oak* **(Quercus petraea)** *Delay Autumn Senescence and Bud Burst in the Following Spring*

Kristine Vander Mijnsbrugge, Arion Turcsán, Jorne Maes, Nils Duchêne, Steven Meeus, Kathy Steppe and Marijke Steenackers

*382 Early Summer Drought Stress During the First Growing Year Stimulates Extra Shoot Growth in Oak Seedlings* **(Quercus petraea)**

Arion Turcsán, Kathy Steppe, Edit Sárközi, Éva Erdélyi, Marc Missoorten, Ghislain Mees and Kristine V. Mijnsbrugge

*391 Variation in Ecophysiological Traits and Drought Tolerance of Beech (***Fagus sylvatica** *L.) Seedlings from Different Populations*

Claudia Cocozza, Marina de Miguel, Eva Pšidová, L'ubica Ditmarová, Stefano Marino, Lucia Maiuro, Arturo Alvino, Tomasz Czajkowski, Andreas Bolte and Roberto Tognetti


Anna Gershberg, Gidi Ne'eman and Rachel Ben-Shlomo

#### **Practical relevance - Consequences for forest management**

*428 Indirect Evidence for Genetic Differentiation in Vulnerability to Embolism in*  **Pinus halepensis**

Rakefet David-Schwartz, Indira Paudel, Maayan Mizrachi, Sylvain Delzon, Hervé Cochard, Victor Lukyanov, Eric Badel, Gaelle Capdeville, Galina Shklar and Shabtai Cohen

*441 Intraspecific Variation in Wood Anatomical, Hydraulic, and Foliar Traits in Ten European Beech Provenances Differing in Growth Yield*

Peter Hajek, Daniel Kurjak, Georg von Wühlisch, Sylvain Delzon and Bernhard Schuldt

*455 Desiccation and Mortality Dynamics in Seedlings of Different European Beech (***Fagus sylvatica** *L.) Populations under Extreme Drought Conditions* Andreas Bolte, Tomasz Czajkowski, Claudia Cocozza, Roberto Tognetti, Marina de Miguel, Eva Pšidová, L´ ubica Ditmarová, Lucian Dinca, Sylvain Delzon, Hervè Cochard,

Anders Ræbild, Martin de Luis, Branislav Cvjetkovic, Caroline Heiri and Jürgen Müller

# Editorial: Studying Tree Responses to Extreme Events

Achim Bräuning<sup>1</sup> \*, Andreas Bolte<sup>2</sup> , Cristina Nabais <sup>3</sup> , Sergio Rossi 4, 5 and Ute Sass-Klaassen<sup>6</sup>

<sup>1</sup> Department of Geography and Geosciences, Institute of Geography, Friedrich-Alexander-University Erlangen-Nuremberg, Erlangen, Germany, <sup>2</sup> Thünen Institute of Forest Ecosystems, Eberswalde, Germany, <sup>3</sup> Departamento de Ciências da Vida, Faculdade de Ciências e Tecnologia, Centro de Ecologia Funcional, Universidade de Coimbra, Coimbra, Portugal, <sup>4</sup> Département des Sciences Fondamentales, Université du Québec à Chicoutimi, Chicoutimi, QC, Canada, <sup>5</sup> Key Laboratory of Vegetation Restoration and Management of Degraded Ecosystems, Provincial Key Laboratory of Applied Botany, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou, China, <sup>6</sup> Forest Ecology and Forest Management Group, Wageningen University and Research, Wageningen, Netherlands

Keywords: climate change, future forests, tree, mechanistic understanding, structure-function relationships, long-term monitoring, intra-annual resolution, resilience

**Editorial on the Research Topic**

**Studying Tree Responses to Extreme Events**

## CLIMATE CHANGE AND TREE RESPONSE TO EXTREME EVENTS

There is a common understanding that climate change is a global challenge in the twenty-first century for the future of humankind (Stott et al., 2016). It is meanwhile clear that human activities have influenced the earth climate system, with substantial modifications in the frequency and magnitude of climate extreme events that occurred since the 1950s (IPCC, 2013, AR5).

Among climate extreme events, hydric as well as thermal anomalies such as droughts, flooding, heat waves, fires, and frost events play an important role for productivity and survival of trees and may cause severe disturbances in forest ecosystems (Allen et al., 2010; Teskey et al., 2015). Trees are long-living organisms with a life-span of between several hundred to thousands of years, with the oldest living tree ramets on earth having reached ages of up to 5,000 years (Stahle, 1996/1997). Thus, mature forest ecosystems may persist for many decades or centuries without considerable variation in tree species composition, also due to cyclic regeneration processes (Zukrigl et al., 1963; Fischer, 1997; Körner, 2013). Disturbances, however, may induce abrupt changes of ecosystem structure and species composition, leading to multiple and less predictable successional pathways (Swanson et al., 2011). The long-term structural persistence of forests strongly depends on the adaptive capacity/plasticity of the species, resulting from both tolerance and resilience potential of tree individuals to environmental impacts, e.g., due to climate extreme events.

These arguments inspired the present research topic, which mostly involves papers on a treecentered approach that explicitly addresses the adaptive capacity of trees at individual, sub-species, and species levels. With this, a reliable basis shall be provided for shaping and managing adaptive, climate-resilient future forests, or to restore landscapes with tree species more suitable or adapted to future environmental conditions (Millar et al., 2007; Bolte et al., 2009; Jacobs et al., 2015).

The papers presented in this research topic derived from the activities conducted during 2012 to 2016 in the EU COST Action FP1106 STReESS (Studying Tree Responses to extreme Events: a SynthesiS). The STReESS community consisted of more than 150 scientists from 34, mostly European, countries, active in the fields of wood anatomy, dendrochronology, ecophysiology, tree modeling, forest ecology, forest management, and forest genetics. This COST Action addressed and answered questions on the impact of extreme climate events on forests and trees, with a special focus on drought.

#### Edited and reviewed by:

Boris Rewald, University of Natural Resources and Life Sciences, Vienna, Austria

> \*Correspondence: Achim Bräuning achim.braeuning@fau.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 21 December 2016 Accepted: 22 March 2017 Published: 20 April 2017

#### Citation:

Bräuning A, Bolte A, Nabais C, Rossi S and Sass-Klaassen U (2017) Editorial: Studying Tree Responses to Extreme Events. Front. Plant Sci. 8:506. doi: 10.3389/fpls.2017.00506

#### Tree-Centered Perspective

The STReESS research methodology was based on a bottomup tree-centered perspective. A tree-centered approach has demonstrated to be suitable to assess impacts of extreme climatic events on trees and forest stands at different spatio-temporal resolutions, and to predict how tree species at scales ranging between the individual to the population might adapt or respond to different future climate conditions (Sass-Klaassen et al.). This framework (i) provides a basic understanding of how climate extremes affect forests in terms of individual mortality, regeneration, and physiological adaptation; (ii) forms an ideal scale to study impacts of climate on morphological, woodanatomical, and physiological responses of trees to their changing environment; and (iii) increases the perception of the legacy of climatic extremes that are imprinted in wood-anatomical structures that may affect the tree's ability to respond to future climate conditions, leading to possible aggravations of effects in case of repeated events of similar magnitude.

The COST Action was divided into specific topics dedicated to relevant aspects of tree response to climate extremes. Accordingly, the 40 papers included in this research topic can be grouped into the topical groups briefly summarized in the following sections.

#### Physiological Adaptive Traits

Extreme events, such as drought, heatwaves, flooding, or frost, affect trees via impacting vital physiological processes, such as water transport (Oberhuber et al.), carbon assimilation, carbon mobilization (Beikircher et al.), or carbon concentration (Lintunen et al.).

Physiological traits associated with water transport and drought susceptibility include specific hydraulic conductivity (Rita et al.), vulnerability to cavitation (P50 value) (David–Schwartz et al.; López et al.; Rosner et al.), stomatal conductance (Seidel et al.), and sapflow (Steppe et al., 2015). All these physiological traits closely interact with wood-anatomical characteristics which vary between and within species, but also within individuals across time (Rathgeber et al.).

These changes in functional xylem (and phloem; Gricar et al. ˇ ) anatomical traits in time clearly reflect phenotypic plasticity and define the long-term adaptive potential of physiological and wood-anatomical traits (Sass-Klaassen et al.). This is important as the plasticity of certain traits can increase the resilience of trees to climate warming (Sterck et al.).

#### Assessing Frequency and Impact of Extreme Events

An indispensable requirement for assessing the frequency of extreme climate events but also the long-term impact of climate factors on trees is the availability of long-term records. The cascading effect of extreme events on physiological and growth responses leads to distinct traces in the wood that hence can serve as biological proxies. The tree-centered approach uses long continuous tree-ring width records as well as discrete records of specific wood-anatomical features in the tree ring to precisely reconstruct past climate conditions and to exactly date extreme climate events (Carrer et al.; Kurz-Besson et al.). Specific tree-ring features such as intra-annual density variations (Zalloni et al.; Klisz et al.), flood rings (Copini et al.), or the absence of tree rings in specific years (Novak et al.) are reliable proxies ("markers") of drought, floods, or other extreme events. Advances in knowledge on the intra-annual dynamics of wood formation provide information on the species-specific timing of growth in relation to climate (Martinez Del Castillo et al.) or under particular site conditions (Gricar et al. ˇ ) and allows to increase the temporal resolution of intra-annual density variations (De Micco et al.) or flood rings (Copini et al.). Moreover, intra-annual measurements allow assessing directly the impact of the environment and its changes (Dao et al.; Maeght et al.; Legave et al.; Lechthaler et al.) or related effects such as defoliation (Guada et al.) on tree viability and growth dynamics. Within STReESS, big efforts have been made to verify the causal relationship between extreme climate events and woodanatomical markers, including the aspect of temporal precision of markers. Databases and catalogs on specific wood-anatomical markers such as intra-annual density variations (IADFs) have been generated with the potential to study temporal and spatial patterns in extreme events for various mostly Mediterranean conifer species (Zalloni et al.).

#### New Methods and Tools to Measure and Evaluate Stress Markers in Trees

Through collaboration between wood anatomists, ecophysiologists and tree modelers, huge advances have been made in developing and presenting new methods and tools to assess the mechanisms of tree responses to extreme climate events, but also to enable creation of long-term time series of cell-based wood-anatomical traits with reasonable time effort. Rathgeber et al., von Arx et al., von Arx et al., and Anfodillo et al. shed light on basic aspects of xylem-cell formation, quantification of wood-anatomical features, and important considerations for their interpretation. Other authors introduce new statistical tools for the evaluation of extreme climate events (Siegmund et al.) and highlight gaps in the understanding of the formation of specific wood-anatomical markers (Battipaglia et al.). The combination of high-time resolution monitoring of physiological processes (sapflow) and tree growth (dendrometers) forms a powerful approach to parameterise process-based tree models in real-time (Steppe et al.). Besides forming the bases for increased mechanistic understanding of tree responses to extreme events TreeWatch.net will improve the public awareness for climate-impact research on trees.

While yet only few trees are fully equipped with sophisticated monitoring tools, like TreeWatch.net, there is ample information on high-resolution tree growth available from thousands of dendrometers installed on numerous tree species in forests across the world. Firsts steps for collection and homogenisation of these measurements have been done and new analyses methods have been developed (Van der Maaten et al., 2016).

#### Varying Adaptive Capacity of Populations to Drought—Facilitation of Forest Adaptation

One focus of the COST action was the analysis of interpopulation adaptive capacity of trees to extreme events. Several studies addressed the adaptation to drought of pedunculate oak (Mijnsbrugge et al.; Turcsán et al.), European beech (Cocozza et al.; Bolte et al.; Hajek et al.), and Scots pine (Seidel and Menzel; Seidel et al.). Interestingly, a considerable variation in drought-stress tolerance or resilience among populations or provenances was observed. The results demonstrate the local adaptation of the studied tree species (Mijnsbrugge et al.; Seidel and Menzel,; Bolte et al.). However, the wide intra-population genetic variability influences the adaptive capacity of species and has to be also considered (Hajek et al.; Gershberg et al.). The outcomes strongly support the idea of the selection of droughttolerant ecotypes or even individuals for increasing the adaptive capacity of forest stands with major European tree species within or even beyond their current native range ("assistant migration" sensu Millar et al., 2007). This includes the translocation of preadapted individuals and ecotypes, so called assisted gene flow (AGF), in order to facilitate the adaptation of planted stands or mixed stands with planted and natural regeneration (Aitken and Bemmels, 2016).

#### Practical Relevance—Consequences for Forest Management

The aim of COST STReESS was to test the potential of the multi-disciplinary tree-centered approach to assess shortand long-term effects of changing climate conditions and specifically extreme events on growth responses and thresholds for mortality (e.g., Cailleret et al., 2017). The advantage of such a bottom-up approach is that through enhanced mechanistic understanding, the plasticity of functional traits and hence the adaptive potential of populations and tree species under changing climate conditions can be estimated. This forms the bases for assessing the implications of changing climate conditions for the stability and productivity of different tree species. The actual challenge is linking the theory to the application, i.e., translating the progress that is made in ecophysiological and forest ecological research to recommendations for practitioners of forest management. Concrete examples are related to species and provenance selection. Based on the tree-centered approach, it is possible to develop and point out specific ecophysiological and wood-anatomical indicators for selecting drought-tolerant provenances or tree species (David-Schwartz et al.; Hajek et al.; Bolte et al.; Kurz-Besson et al.).

## The Long Term Perspective

#### Achievements and Future Perspectives

The tree-centered approach has led to new insights on the impact of climate and extreme climate events on tree-growth processes, the structural components of wood, and the consequences for physiological performance of trees at individual to species level. The collection of articles in this research topic, together with more than 100 articles published elsewhere by the COST STReESS community on concepts (e.g., Steppe et al., 2015), metaanalysis of phenological wood traits (e.g., Rossi et al., 2013, 2016; Cuny et al., 2015), and important mechanisms behind mortality (e.g., Cailleret et al., 2017), illustrates the potential of this bottom-up approach. Major achievements include (1) enhanced understanding of relation between structure and function both on whole-tree level but specifically in woody tissues, (2) improvement of mechanistic models by parameterisation with high-time resolution measurements, (3) developing the basis for linking short-and long-term tree-growth and wood-anatomical records to assess long-term effects of extreme climate events on tree growth. Many aspects, processes, species, and traits have still to be studied in depth. Comprehensive mechanistic models linking structure, physiology, and function of tree species remain challenging, and require further multidisciplinary development of integrative conceptual and statistical approaches. The interdisciplinary network created through STReESS together with the effort that has been made on creating large databases and catalogs and advancing and harmonizing methods for data acquisition, data analyses, and modeling forms the bases for the necessary next steps.

#### Integration with Other Approaches

The findings of the COST Action STREeSS show evidence that the tree-based level provides opportunities to study cellular, genetic, physiological, anatomical, and ecological responses to climate as a whole. For example, several recent studies reported that seedlings of European beech from different climate origins over Europe performs differently in terms of drought tolerance (cf. Eilmann et al., 2014; Thiel et al., 2014; Pšidová et al., 2015; Dounavi et al., 2016). However, research activities within STREeSS showed that drought resistance is more related to local precipitation conditions at the place of origin than with geographically marginal origin (Bolte et al.). Hence, the role of local genetic variation in beech populations determining phenotypic plasticity in functional and structural traits of beech individuals that control drought adaptability need to be further evaluated (Cocozza et al.) to provide the most suitable plant material for forests adapted to future climates.

Another challenge remains the upscaling from tree-individual based drought responses to forest stands, since vegetation models are known to under-represent drought induced mortality (Steinkamp et al., 2015). Applying the tree-individual approach (Sass-Klaassen et al.) in systematically designed ways over species distribution ranges or ecological gradients in regional forested landscapes may have great potential to contribute to the debate of forest resilience to climate change (e.g., Reyer et al., 2015; Babst et al., 2017).

A third integrative pathways addresses the inclusion of highresolution data on tree diameter growth and sapflow within the TreeWatchNet (Steppe et al.) in existing large-scale forest ecosystem monitoring networks like the UN-ECE ICP Forests on air pollutant and climatic impacts (Michel and Seidling, 2016) and the ICOS network on carbon flux monitoring (Laurent, 2016). High-resolution growth and physiology monitoring provides needed data to assess dynamic response of trees and forests to stressors and functioning in carbon and nutrient cycling.

## AUTHOR CONTRIBUTIONS

All coauthors acted as topic editors for the special issue in Frontiers in Plant Science entitled "Studying Tree Responses to extreme Events" All co-authors prepared the editorial text together and equally shared the tasks of writing, proofreading, and correcting the manuscript. ABr initiated the idea of the special issue and orgnaized the writing process of the editorial, with support from ABo and US.

#### FUNDING

COST (European Cooperation in Science and Technology) is a funding agency for research and innovation networks. COST Actions help connect research initiatives across Europe and enable scientists to grow their ideas by sharing them with their peers. This boosts their research, career and innovation www.cost.eu.

#### REFERENCES


Pšidová, E., Ditmarová, L., Jamnická, G., Kurjak, D., Majerová, J., Czajkowski, T., et al. (2015). Photosynthetic response of beech seedlings of different

#### ACKNOWLEDGMENTS

This special issue is one of the final products of the COST Action FP1106 STReESS. We thank all COST FP1106 STReESS participants for an inspiring 4-year period of fruitful collaboration. Special thanks to the EU COST, and its scientific managers Fatima Bouchama, Melae Langbein, and financial manager Cassia Azevedo for supporting our Action. Besides this special issue COST STReESS resulted in many other research articles, review papers, book chapters, and films and a highly motivated network of researchers from various scientific disciplines who are highly motivated to continue working on concepts developed during our Action.

origin to water deficit. Photosythetica 53, 187–194. doi: 10.1007/s11099-015- 0101-x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Bräuning, Bolte, Nabais, Rossi and Sass-Klaassen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Tree water status and growth of saplings and mature Norway spruce (*Picea abies*) at a dry distribution limit

*Walter Oberhuber1\*, Albin Hammerle2 and Werner Kofler1*

*<sup>1</sup> Institute of Botany, Leopold-Franzens-University of Innsbruck, Innsbruck, Austria, <sup>2</sup> Institute of Ecology, Leopold-Franzens-University of Innsbruck, Innsbruck, Austria*

We evaluated the size effect on stem water status and growth in Norway spruce (*Picea abies* (L.) Karst.) occurring at the edge of its natural range in a dry inner Alpine environment (750 m asl, Tyrol, Austria). Intra-annual dynamics of stem water deficit (-W), maximum daily shrinkage (MDS), and radial growth (RG) were compared among saplings (stem diameter/height: 2.2 cm/93 cm; *n* = 7) and mature adult trees (25 cm/12.7 m; *n* = 6) during 2014. -W, MDS, and RG were extracted from stem diameter variations, which were continuously recorded by automatic dendrometers and the influence of environmental drivers was evaluated by applying moving correlation analysis (MCA). Additionally, we used Morlet wavelet analysis to assess the differences in cyclic radial stem variations between saplings and mature trees. Results indicate that saplings and mature trees were experiencing water limitation throughout the growing season. However, saplings exhibited a more strained stem water status and higher sensitivity to environmental conditions than mature trees. Hence, the significantly lower radial increments in saplings (0.16 ± 0.03 mm) compared to mature trees (0.54 ± 0.14 mm) is related to more constrained water status in the former, affecting the rate and duration of RG. The wavelet analysis consistently revealed more distinct diurnal stem variations in saplings compared to mature trees. Intra-annual RG was most closely related to climate variables that influence transpiration, i.e., vapor pressure deficit, relative air humidity, and air temperature. MCA, however, showed pronounced instability of climate–growth relationships, which masked missing temporal or significant correlations when the entire study period (April–October) was considered. We conclude that an increase in evaporative demand will impair regeneration and long-term stability of drought-prone inner Alpine Norway spruce forests.

Keywords: dendrometer, dry inner Alpine valley, maximum daily shrinkage, *Picea abies*, radial growth, stem water deficit, wavelet analysis

#### Introduction

Norway spruce (*Picea abies* (L.) Karst.) is the most widespread coniferous species in the Central European Alps ranging from low elevation to the timberline (Ellenberg and Leuschner, 2010). This tree is moderately shade-tolerant, dominates in late successional stages and shows high sensitivity to soil water supply (e.g., Lévesque et al., 2013; van der Maaten-Theunissen et al., 2013). Schuster and Oberhuber(2013a) detected increasing drought sensitivity of mature *P. abies* in recent decades,

#### *Edited by:*

*Achim Braeuning, University Erlangen-Nuremberg, Germany*

#### *Reviewed by:*

*Cristina Nabais, University of Coimbra, Portugal Bao Yang, Chinese Academy of Sciences, China*

#### *\*Correspondence:*

*Walter Oberhuber, Institute of Botany, Leopold-Franzens-University of Innsbruck, Sternwartestrasse 15, 6020 Innsbruck, Austria walter.oberhuber@uibk.ac.at*

#### *Specialty section:*

*This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science*

*Received: 22 June 2015 Accepted: 23 August 2015 Published: 07 September 2015*

#### *Citation:*

*Oberhuber W, Hammerle A and Kofler W (2015) Tree water status and growth of saplings and mature Norway spruce (Picea abies) at a dry distribution limit. Front. Plant Sci. 6:703. doi: 10.3389/fpls.2015.00703* which was related to a decline in soil moisture availability due to increasing stand density and tree size. Ecophysiological and dendroclimatological studies have shown that size-mediated changes in functional processes and climate–growth relationships occur in forest trees (Mencuccini et al., 2005; Martínez-Vilalta et al., 2007; Mérian and Lebourgeois, 2011). Therefore, sizedependent sensitivity to climate is expected to affect forest structure and succession. Furthermore, the anticipated changes in climate, which include significant warming in the following decades, changes in seasonal precipitation patterns and an increase in both frequency and intensity of severe droughts (IPCC, 2013), underline the need to evaluate differences in intra-annual development of the plant water status in saplings compared to mature adult trees.

Although plant water status is determined primarily by the physical conditions of the air and soil, trees are capable of counterbalancing temporal water deficits by withdrawing water from water storage locations within the tree, namely, the sapwood, the cell walls, and inactive vessels, the mesophyll of needles and the elastic tissues of the bark, i.e., enlarging xylem cells, cambium, phloem, and bark parenchyma (Zimmermann, 1983). It is generally assumed that the reversible diurnal changes in stem size, i.e., stem shrinking and swelling during day and night, respectively, are the result of changing water potential gradients within the plant (e.g., Whitehead and Jarvis, 1981; Herzog et al., 1995; Zweifel et al., 2001; Sevanto et al., 2011). Reversible changes in daily water content of the stem together with irreversible radial stem growth, i.e., cambial cell division and cell enlargement primarily explain stem radius variations during the growing season. The stem radius variation de-trended for growth was determined to be proportional to the water content in the living tissues of the bark and was termed the tree water deficit (-W) (Hinckley and Lassoie, 1981; Herzog et al., 1995; Zweifel et al., 2000). According to several authors (Zweifel et al., 2005; Drew et al., 2011; Köcher et al., 2013) -W is closely related to drought stress and is mainly determined by a combination of atmospheric and soil conditions (e.g., Zweifel et al., 2005; Ehrenberger et al., 2012). Maximum daily shrinkage (MDS), defined as the difference between the daily maximum and minimum stem diameter, is a measure of the proportion of water taken up at night that is lost from elastic cambial and phloem tissues during the day. Several authors reported that MDS provides information on tree water relations (Zweifel et al., 2000; Intrigliolo and Castel, 2005; Giovannelli et al., 2007). Hence, high resolution dendrometer measurements of diurnal stem diameter variations provide several parameters related to tree water status and enable the determination of radial growth (RG) (e.g., Downes et al., 1999; Deslauriers et al., 2007). -W and MDS determined in this study represent indirect measures of plant water status compared to direct methods, such as leaf water potential or relative water content (Jones, 2007). Advantages of the applied indirect method are that (i) dendrometers allow a continuous, non-destructive record of plant water status, and (ii) -W and MDS can be determined simultaneously with radial stem growth. Although especially in isohydric plants stomatal control tends to maintain tissue water status stable (homeostatic regulation), indirect measures of water status are regarded to be highly valuable for detecting physiological responses to water deficits (Naor and Cohen, 2003; Zweifel et al., 2005; Jones, 2007). Previous findings revealed that in an old-aged mixed coniferous stand that included *Pinus sylvestris*, *Larix decidua*, and *P. abies*, *P. abies* most strongly exhausted stem water reservoirs during the growing season (Oberhuber et al., 2014). Conversely, shade-tolerance, shallow rooting and isohydric behavior, i.e., early closing of stomata to stabilize water relations ('drought avoidance'), were suggested to allow *P. abies* to invade *Pinus sylvestris* stands at dry-mesic sites (Anfodillo et al., 1998; Schuster and Oberhuber, 2013b; Leo et al., 2014).

The aims of this study were therefore (i) to compare the seasonal development of tree water status indicators (-W, MDS) and growth of co-occurring saplings and mature *P. abies* growing at the distribution limit at a xeric inner Alpine site during the growing season and (ii) to determine environmental factors that are most closely related to the overall variations in parameters of tree water status and radial stem growth. We are not aware of any study comparing dendrometer derived water status indicators throughout the growing season in co-occurring saplings and mature *P. abies* within the same stand, although contrasting development of stem water status is plausible to occur. Specifically in saplings, lower stem water reserves are expected due to thinner bark and smaller sapwood area compared to mature adult trees (Zweifel et al., 2000; Dawes et al., 2014). Conversely, mature dominant trees are exposed to higher solar radiation, vapor pressure deficit (VPD) and wind velocity, which increase transpiration forcing. In accordance with previous findings that smaller trees take advantage of better rooting in the upper moist soil horizon, lower water demand of small trees and more favorable microclimates below canopy (Schuster and Oberhuber, 2013b), we hypothesized that a less distinct stem water deficit (-W) is developed throughout the growing season in saplings compared to mature trees. Furthermore, we expected that -W, MDS, and growth are most closely related to the climate variables that influence transpiration and that the temporal relationship between environmental variables and stem water indicators and RG exhibit high stability throughout the growing season.

### Materials and Methods

#### Study Site

The study site is part of a postglacial rock-slide area situated in the montane belt (c. 750 m asl) within the inner Alpine dry valley of the Inn River (Tyrol, Austria, 47◦13 53 N, 10◦50 51 E; Fliri, 1975). The mean annual air temperature is 7.3◦C and the mean annual precipitation amounts to 716 mm (1911–2010, Ötz, 812 m asl, 5 km from the study area). The soil is shallow, reaches maximum depth of 0.2 m at the measurement site and consists of unconsolidated, coarse-textured materials with low water holding capacity. It is classified as rendzic leptosol according to the FAO classification system (FAO, 2006). The dominating plant community on xeric sites is an open Spring Heath-Pine wood (Erico-Pinetum sylvestris; Ellenberg and Leuschner, 2010). *P. abies* co-occurs in the canopy under more mesic conditions, i.e., in hollows or on north-facings slopes. At the selected drymesic site a mixed stand composed of *Pinus sylvestris*, *P. abies*, and *Larix decidua* is developed, whereby the proportion referred to basal area is 60, 20, and 20%, respectively. Height and canopy coverage of the selected stand amounted to *c*. 13 m and *c*. 70%, respectively (**Table 1**). The study site was slightly facing north and the moderately shade-tolerant *P. abies* rejuvenates naturally under the mixed canopy. The number of saplings <2 m in height amounts to approximately 10 individuals in an area covering 30m × 30m (cf. Schuster and Oberhuber, 2013b). Saplings and mature trees were selected within the same stand covering an area of *c*. 0.5 ha.

#### Microclimate Records

During the study period, air temperature, relative air humidity (RH), daily precipitation and solar radiation were collected automatically (ONSET, Pocasset, MA, USA) at 2 m height on an open ridge, i.e., in a non-vegetated area close to the study area (<100 m in linear distance). Additionally, air temperature and RH were recorded below canopy within the selected stand at 2 m height. Measuring intervals for all sensors were set to 30 min and mean daily air temperatures were calculated by averaging all measurements (48 values per day). The VPD in the air was calculated from the hourly means of air temperature and RH using the equation given in Prenger and Ling (2000). Volumetric soil water content (SWC) at the 5–10 cm and 15– 20 cm soil depth layers (*n* = 3 sensors per soil depth) was recorded within the selected stand (ThetaProbes Type ML2x, Delta-T, Cambridge, England). Additionally, soil temperature in the top 5–10 cm soil depth layer was measured (S-TMB, ONSET, Pocasset, MA, USA). Measurements were taken every 60 min and mean daily SWC (Vol. %) and soil temperature ( ◦C) were calculated by averaging all measurements from three sensors.

#### Dendrometer Records

In autumn 2013 we installed temperature compensated electronic dendrometers (Ecomatik, Munich, Germany) with resolutions of <3 μm. Band dendrometers (type DC2) and diameter dendrometers (type DD-S) were mounted on healthy mature trees at 1.3 m stem height (*n* = 6) and on saplings at 15 cm stem height (*n* = 7), respectively. The temperature coefficient of the sensor amounted to <0.1 μm/K. The measuring cable of band dendrometers consisted of Invar-steel, which shows a temperature coefficient of linear expansion <1 μm/mK. To

reduce the influence of hygroscopic shrinking and swelling effects on dendrometer records (DMR), dead outermost layers of the bark (periderm) were slightly removed. Mean tree age of selected mature trees and saplings was estimated to be >100 year and <20 year, respectively (cf. Schuster and Oberhuber, 2013b). Data loggers were programmed to record measurements taken every 30 min and daily increments of stem radius were calculated by averaging all daily measurements (48 values/day). The Gompertzfunction, which is an asymmetric sigmoid curve, i.e., it accelerates more quickly than it decelerates, was used for describing the long-term development of radial increment over the whole growing season (e.g., Rossi et al., 2003). To this end, the nonlinear regression procedure included in the Origin software package (OriginLab Corporation, Northampton, MA, USA) was applied.

#### Tree Water Deficit and MDS

To distinguish between growth-related and water-status-related stem radius changes, DMR were de-trended for growth according to Ehrenberger et al. (2012). In particular, a 'growth line' was constructed by drawing a horizontal line from the daily maximum value to the next equal stem radius value ignoring periods of incomplete stem radius recovery due to drought induced stem shrinkage (**Figure 1A**). Then, the growth line followed the slope of the original DMR representing daily RG. This procedure was repeated until the end of the measurement period in early October. By applying this method it is assumed that RG, which requires an increase in cell volume and depends on high cell turgor pressure, is restricted to short periods when the stem is rehydrated rather than occurring constantly throughout the growing season (cf. Zweifel et al., 2005). In the following, tree water deficit (-W) was determined as the difference in stem size under low water availability conditions relative to the stem size under fully hydrated conditions (-W = 0). Hence, increasingly negative values of -W due to dehydration of storage pools indicate increasing drought stress (**Figure 1B**). Additionally, the magnitude of MDS (μm) defined as the difference between the daily maximum, and minimum stem radius (diurnal amplitude), was determined (**Figure 1A**). Although the maximum diameter was usually observed in the early morning, the minimum diameter was normally reached during the late afternoon, reflecting the daily cycle of water uptake and loss. Bark width (excluding periderm) was determined during a cool-moist period in the fall by sampling increment cores and stem disks from mature trees and saplings, respectively.

TABLE 1 | Site description and tree characteristics (A, aspect; CC, canopy coverage; Prz, Protorendzina (rendzic leptosol); Rh, raw humus; Sd, soil depth).


*Mean values* <sup>±</sup> *SD are shown. Tree height, stem diameter and bark width are significantly different between mature trees and saplings at P* <sup>&</sup>lt; *0.001 (Students t-test).* <sup>1</sup>*Stem diameter was determined at breast height (1.30 m) and at 15 cm height for mature trees and saplings, respectively.*

#### Wavelet Analysis of DMR

Wavelet transform is used to decompose a time series over a time-scale space, thus providing visualization of a power distribution along time and frequency. It is a powerful tool to analyze non-stationary signals and it permits the detection of main periodicities in a time series and the evolution of their respective amplitude, frequency, and phase (Percival and Walden, 2000). We used the Morlet wavelet which is a sine wave modulated by a classical Gaussian function, because it establishes a clear distinction between random fluctuations and periodic regions (Torrence and Compo, 1998). Following Grinsted et al. (2004) the dimensionless frequency (ω0) was set to 6. The generated wavelet spectrum is a time-scale plot, where the x- and y-axis represent the position along time and periodicity scale, respectively, and the color contour at each x/y point represents the magnitude of the wavelet coefficient at that point. A dark red color is assigned to the highest value of the wavelet power spectrum, whereas a dark blue color is assigned to the lowest value. As continuous wavelet analyses are applied to the time series of finite length, edge effects may appear on the wavelet spectrum, leading to the definition of a cone of influence (Torrence and Compo, 1998), which is shown in a lighter shade in the wavelet spectrum. Statistical significance levels were estimated against a red noise model with lag-1 autocorrelations estimated from the observed time series (Grinsted et al., 2004). The wavelet analysis was performed on half-hourly DMR of saplings (*n* = 7) and mature trees (*n* = 5) from April through September after removing the long-term sigmoidal growth trend from the original series. Wavelet analyses were performed using the Matlab (Mathworks, Natick, MA, USA) wavelet toolbox of Torrence and Compo (1998), provided by Aslak Grinsted (http://www.mathworks.com/matlabcentral/fileexchange/47985 cross-wavelet-and-wavelet-coherence).

#### Climate Influence on Tree Water Status and Growth

Pearson correlation statistics (*r*) were calculated to explore the relationship between daily time series of environmental variables for the period of April–October (precipitation, RH, vapour pressure deficit (VPD), air and soil temperature, SWC and time series of -W, MDS, and RG extracted from DMR. Kolmogorov–Smirnov tests were applied to check for the normal distribution of selected variables. The Spearman rank-correlation coefficient (ρ) was determined for non-normally distributed variables. Temporal changes in these relationships were evaluated by moving correlation analysis (MCA), which is based on progressively shifting the period of a fixed number of days across time to compute the correlation coefficients. To provide robust measures of association between environmental variables and -W, MDS, and RG, the length of the calibration window was set to 30 days. Correlation coefficients were arbitrarily plotted against the last day of the period. The software packages used for the analyses were Statistica, version 8.0 (StatSoft, Inc., Tulsa, OK, USA) and Matlab R2010b.

#### Results

#### Environmental Variables during the 2014 Growing Season

At the experimental site, monthly precipitation ranged from 39 mm (September) to 94 mm (August) (**Figure 2A**). Starting with frequent rainfall events in mid-April, SWC reached c. 0.25 m<sup>3</sup> m−<sup>3</sup> until mid-May, when the occurrence of sporadic low rainfall events caused a decrease of SWC to c. 0.10 m<sup>3</sup> m−<sup>3</sup> in the 5–10 cm soil depth layer for several weeks. Frequent rainfall events in late-June through early September increased SWC to c. 0.25 m<sup>3</sup> m−<sup>3</sup> in the 5–10 cm soil depth layer (**Figure 2A**). Monthly mean temperature ranged from 10.4◦<sup>C</sup> (April) to 17.7◦C (July), with an overall mean value of 14.7◦C during April–September. Maximum air temperature reached 36.4◦C in early-June. Daily mean soil temperature in the 5– 10 cm soil depth layer generally followed the trend of air temperature with minor amplitudes developed in the former (**Figure 2B**). Mean VPD amounted to 0.53 kPa from April to September and the VPD maxima reached 2.3 kPa in early June (**Figure 2C**). VPD was not significantly different above and below the canopy (data not shown), which is most likely caused by a

fairly open canopy of the selected study plot (canopy coverage is 70%).

#### Stem Radius Changes and Dendrometer-Derived Tree Water Deficit

Dendrometer traces of saplings and mature *P. abies* showing daily maximum and minimum values are depicted in **Figure 3A**. Cumulated daily RG amounted to 0.54 mm ± 0.14 and 0.16 mm ± 0.03 in mature trees and saplings, respectively, indicating radial increments more than three times in the former in 2014. Extracted daily RG was consistently lower in saplings than mature trees. RG occurred mostly concurrently in saplings and mature trees during periods of frequent rainfall (cf. **Figures 2A** and **3A**). Modeling DMR using Gompertz functions revealed that intra-annual RG peaked in mid-May (doy 137) and early June (doy 153) in saplings and mature trees, respectively (data not shown). RG stopped in mid-July (doy 193) in saplings and mid-August (doy 226) in mature trees (**Figures 3A,B**).

Extracted -W showed synchronous fluctuations among saplings and mature trees and reached –0.15 mm in saplings and –0.21 mm in mature trees, when incomplete rehydration was detected during the growing season (**Figure 3**). After the end of RG -W did not recover by mid-October indicating a permanent -W in both groups. Maximum -W values were determined in late September and corresponded to 9.4 and 6.9% of living tissues of the bark of saplings and mature trees, respectively. Bark width was twice as great in mature trees compared to saplings (**Table 1**). Mean MDS values amounted to 57 ± 22 μm and 59 ± 19 μm in mature trees and saplings, respectively (**Figure 3C**) and were not significantly different among groups. A slightly increasing trend in MDS was detected in both series during summer (June– August), when higher temperatures prevailed.

The summation of the residuals of -Wmature – -Wsapling and MDSmature – MDSsapling revealed that starting with a period of low precipitation in mid-May and decreasing SWC (**Figure 2A**) saplings developed higher -W and MDS than mature trees (**Figure 4**). Secondly, the former relationship slowly decreased with increasing precipitation and SWC at end of June and abruptly reversed after mid-August (doy 226), when permanently higher -W was developed in mature trees than saplings. Cumulative residuals of MDS, however, leveled off more quickly with the occurrence of higher precipitation events. Higher -W and MDS in saplings compared to mature trees is also indicated by higher frequencies of more negative -W and large MDS values in saplings than mature trees (**Figures 5A,B**).

Morlet wavelet analysis on the detrended DMR of saplings and mature trees revealed a statistically significant daily cycle (*<sup>P</sup>* <sup>&</sup>lt; 0.05) during a large part of the study period (**Figures 6A,B**). The diurnal cycle is, however, notably less pronounced in the latter. Significant periodicities of several days up to 2 weeks also occurred in both wavelet spectra. While these periodicities are localized in time for the saplings, periodicities of approximately 2 weeks occurred almost continuously over the study period in mature trees.

#### Influence of Environmental Variables on Tree Water Deficit and Growth

Considering the whole study period, i.e., from late April through early October, all environmental variables were highly significantly correlated with -W of saplings (**Table 2**), whereby air and soil temperature and VPD (inverse relationships) and precipitation (direct relationship) showed the highest correlation coefficients. MCA applying a window of 30 days indicated largely temporal stability of these relationships throughout the study period (**Figures 7A–C**). In mature trees, however, only precipitation and air and soil temperature showed significant relationships with -W over the whole study period, whereas MCA showed almost continuous significant relationships for VPD, RH and SWC (**Figures 7A–C**).

were smoothed based on a fast Fourier transform low-pass filter, whereby the number of points was set to 20.

Taking the whole study period into account, MDS showed the closest correlations with air temperature and SWC (direct and inverse relationship, respectively) in mature trees and saplings (**Table 2**). MCA, however, revealed decreased sensitivity of MDS to all environmental variables during periods of high rainfall occurring from July through August (**Figures 7D–F**). Although stem radial increments extracted from DMR were highly significantly related to VPD, RH, air temperature, and precipitation in saplings and mature trees (**Table 2**), MCA indicated temporal instability of these relationships (**Figures 7G–I**). Overall, based on MCA saplings and mature trees showed quite similar response of -W, MDS, and growth to environmental variables, and predominantly closer correlations were found using a 30 day window compared to taking the whole study period, i.e., late-April to early October for -W and MDS, or the growing season, i.e., late-April through July, into account (**Figure 7**; **Table 2**).

## Discussion

Total increment in 2014 differed more than threefold among saplings and mature trees, which is consistent with a previous study that showed that RG of young *P. abies* within the study area is strikingly suppressed compared to mature adult trees (Schuster and Oberhuber, 2013b). Furthermore, DMR revealed that saplings reached maximum growth earlier and also RG stopped earlier compared to mature trees. Because it was frequently found that water stress causes earlier culmination and ending of RG in *P. abies* (Pichler and Oberhuber, 2007; Levanicˇ et al., 2009; Swidrak et al., 2014) and other conifers (e.g., Gruber et al., 2010), we relate the differences in annual radial increments among saplings and mature trees primarily to the development of adverse stem water status in small trees. Our reasoning is supported by findings that water deficits affect growth earlier and more intensively than photosynthesis, i.e., sink limitation rather than source limitation of growth prevails under water deficits (for a review see Muller et al., 2011). Anderegg and Anderegg (2013) also found no significant changes in carbohydrate concentrations in saplings of two conifer species during severe experimental drought.

#### Stem Water Deficit in Saplings and Mature Trees

Results of this study are in contrast with our first hypothesis that a less distinct -W is developed in saplings compared to

mature trees. Cumulative residuals and frequency distributions of -W and MDS indicate that higher amounts of water reserves were recruited in saplings than in mature trees to sustain leaf transpiration. This result implies that during a diurnal period, saplings more strongly exploit their internal stem water reserves than mature trees. This effect is confirmed by wavelet analysis, which revealed that the daily cycle of stem shrinking and swelling is more distinct in saplings than mature trees, i.e., greater daily amplitudes are developed in the former. The daily amplitude is related to environmental variables that influence the amount and rate of daily transpiration, as reported by King et al. (2013). Furthermore, the dominance of longer-term cycles (1 week up to 2 weeks) in mature adult trees suggests that more balanced water relations exist in larger trees. Cumulative residuals (-Wmature – -Wsaplings) reached maximum values in June–July when SWC was at a minimum and the highest temperatures were recorded. High evaporative demand and long day length during a period of low SWC less strongly influenced the -W of mature trees, i.e., an insufficient replenishment of water in the expandable tissues led to higher -W in saplings compared to mature trees. Our findings revealed that residuals of MDS showed an earlier response than residual -W, which is consistent with reports of several authors that an increase in MDS is the first detectable morpho-physiological signal of changes in water status of the whole plant (e.g., Conejero et al., 2007; Giovannelli et al., 2007).

We suggest that a higher -W found in saplings compared to mature trees throughout the study period is most likely related to less efficient or competitive water uptake, i.e., mature trees have developed a more extensive root network (Mueller et al., 2005; Voelker, 2011) close to the soil surface, where SWC throughout the growing season was found to be constantly higher than in deeper soil layers. Schmid and Kazda (2002) reported that fine roots of *P. abies* are distributed primarily in upper soil layers. Furthermore, Phillips et al. (2003) found significantly greater ratios of sapwood volume to leaf area in large compared to small trees indicating that water storage capacity relative to water use increases with tree size. This corresponds to results of our study that -W was higher in saplings than mature trees.

Conversely, although *P. abies* is known to maintain a stable leaf water status over a wide range of evaporative demand or SWC through stomatal control, i.e., isohydric behavior (Buckley, 2005; Zweifel et al., 2009; Leo et al., 2014), saplings might have contrasting hydraulic behavior from mature trees. In saplings high transpiration rates presumably occur under moderate drought to maintain carbon assimilation at higher rates as SWC decreases ('anisohydric' behavior). This assumption

is consistent with findings of Cinnirella et al. (2002), who found isohydric behavior in mature *Pinus laricio*, which was not observed in potted seedlings subjected to drought. Furthermore, it has to be considered that the same individual can switch from isohydric behavior to anisohydric behavior in response to changing SWC (Domec and Johnson, 2012). Differences in size of the shrinking tissue, in the conductance of water between the bark and the xylem vessels and in RG can also affect dynamics of -W (Waring et al., 1979; Naor and Cohen, 2003) and MDS (Intrigliolo and Castel, 2005; Dawes et al., 2014). Specifically, significantly lower phloem thickness in saplings than mature trees may restrict stem shrinkage in the former, while significantly higher RG found in mature trees compared to saplings may more strongly mask stem shrinkage in mature trees.

#### Environmental Control of Tree Water Status and RG

Correlation analyses revealed that -W of saplings is more strongly affected by environmental conditions than -W of mature trees throughout the study period. This finding is in line with the development of higher -W in saplings compared to mature trees. Conversely, MDS of both saplings and mature trees showed quite similar responses to environmental variables, and low stability between environmental factors and MDS was detected. We explain this finding by the relocation of water from storage locations (sapwood, cell walls, and inactive vessels) to the elastic tissues of the bark, which might contribute to temporary decoupling of environmental variables especially from MDS (cf. Zweifel et al., 2000). Hence, results of this study are only partly consistent with our hypothesis that both stem water indicators are most closely related to climate variables that influence transpiration. Instability of the relationship between stem water indicators and environmental variables, which is in contrast to our expectation, can be explained by the alternation of unfavorable and favorable environmental conditions throughout the study period and high sensitivity of especially MDS to water availability. Similarly, Vieira et al. (2013) found varying climatic response of dendrometer derived water status parameters and radial increments in *Pinus pinaster* in a drought-prone environment. Therefore, MCA should be applied when evaluating growth and/or stem water status parameters in response to environmental variables that show high variability throughout the growing season. Coupling of -W and MDS to atmospheric conditions indicates that transpiration strongly draws upon water reserves from the living bark, especially of young trees. This observation is in line with several other studies (e.g., Zweifel et al., 2005; Turcotte et al., 2011; Ehrenberger et al., 2012). On the other hand, mature trees showed no significant correlation between -W and SWC, VPD, and RH indicating that large trees have a strong ability to regulate drought stress, because of a more extensive root system and/or greater water storage capacity. The lag observed between soil variables (SWC, soil temperature) and -W and MDS indicate a time lag between water loss by transpiration and refilling of water reserves in the stem (Meinzer et al., 2004). Higher sensitivity of -W to environmental variables compared to MDS is most likely related to the method in, which -W is calculated, i.e., periods of missing stem rehydration are represented in -W but not in MDS.

Although RG was found to be closely related to atmospheric conditions, soil parameters (SWC, soil temperature) had a marginal effect on growth. This effect is consistent with findings in previous studies at low (Oberhuber et al., 2014) and high altitude (Gruber et al., 2009), in which a close coupling of RG of mature deciduous and evergreen coniferous species to atmospheric conditions was detected. These findings imply that transpiration draws upon internal storage tissues in the stem rather than soil water (Cermak ˇ et al., 2007; Betsch et al., 2011) and in this way prevents low stem water potentials that might be caused by peaks of transpiration. As a result, high turgor in cambial cells and their differentiating derivatives favor cell division and cell enlargement. In a dendrometer study Köcher et al. (2012) also reported that atmospheric moisture status rather than other hydrological variables, especially precipitation and soil moisture, influenced cambial growth of five temperate broad-leafed tree species.

FIGURE 6 | Detrended stem variations and Morlet wavelet spectra of half-hourly DMR of saplings (A) and mature adult trees (B). The black line outlines designate the 5% significance level. The 'cone of influence,' where edge effects become important and the results should be ignored, are shown in a lighter shade. Red and blue colors indicate high and low wavelet power spectrum values, respectively (sv, stem variation).

TABLE 2 | Correlation coefficients [Pearson product-moment (*r*) and Spearman rank-correlation coefficient (**ρ**)] between environmental variables (Prec, precipation; SWC, soil water content; VPD, vapor pressure deficit; RH, relative humidity; Tsoil, mean soil temperature; Tairmin, minimum air temperature; Tairmean, mean air temperature; Tairmax, maximum air temperature) and water deficit (*-*W), maximum daily shrinkage (MDS) and growth of saplings and mature *Picea abies*.


*A one-day lag in* -*W, MDS, and growth was considered in correlations with SWC and Tsoil (*-*W and MDS: n* = *178 for both, saplings and mature trees; lag1: n* = *177; growth: n* <sup>=</sup> *84 and 117 for saplings and mature trees, respectively).* <sup>∗</sup>*<sup>P</sup>* <sup>≤</sup> *0.05;* ∗∗*<sup>P</sup>* <sup>≤</sup> *0.01;* ∗∗∗*<sup>P</sup>* <sup>≤</sup> *0.001.* <sup>1</sup>*Spearman rank-correlation coefficient (*ρ*).*

FIGURE 7 | Moving correlations (window 30 days) between the water deficit (*-*W) (A–C), MDS (D–F) and growth (G–I), and environmental variables of saplings (open symbols) and mature *Picea abies* (filled symbols). Spearman rank-correlations (ρ) were calculated for relationships between (i) precipitation and -W, MDS, and growth, and (ii) all environmental variables and growth. For all other relationships the Pearson product-moment coefficients (*r*) are shown (Prec, daily precipitation sum; VPD, vapor pressure deficit; RH, relative air humidity; Tmax, daily maximum air temperature; SWC, soil water content; Tsoil, mean soil temperature). A lag of 1 day was considered in the relationship between -W, MDS, and growth, and SWC and Tsoil. Horizontal dotted and dashed lines indicate *P* = 0.001 and *P* = 0.05 significance levels, respectively.

Notably similar RG responses of saplings and mature trees to environmental variables were detected in this study, which is in contrast to long-term analysis of climate-growth relationships in dendroclimatological studies revealing higher climate sensitivity of older and larger *P. abies* trees (Schuster and Oberhuber, 2013b). These contrasting results indicate

the influence of time scale (days vs. years) on RG response to environmental conditions. Most likely, several seasonal growth phases are integrated in annual increments (i.e., tree ring width), which might affect the climate-growth relationship, whereas high-resolution DMR allows for studying environmental factors on RG directly. Furthermore, different autocorrelation signals in inter-annual (tree rings) vs. intra-annual (DMR) time series might exist.

Results of this study revealed a more strained stem water status in saplings as opposed to mature *P. abies* trees in response to limited water availability and evaporative demand during the growing season. Size-related differences in root morphology and physiology and hydraulic capacitance (cf. Scholz et al., 2011) under drought are needed to prove the suggested processes. We conclude that productivity, structure and distribution of droughtprone Norway spruce forests will be increasingly affected if in the dry inner-Alps the projected decrease in water availability develops in future decades (Christensen et al., 2007).

## References


## Author Contributions

WO assembled data, made conception of the paper, coordinated the research project and drafted the article; all authors analyzed and interpreted the data; AH and WK discussed and commented on the manuscript.

#### Funding

This research was funded by the Austrian Science Fund (FWF), Project Nos. P22280-B16 and P25643-B16.

### Acknowledgment

We acknowledge network building in the frame of the COST Action STReESS (FP1106).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Oberhuber, Hammerle and Kofler. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Prolonged Soil Frost Affects Hydraulics and Phenology of Apple Trees

#### Barbara Beikircher\*, Claudia Mittmann and Stefan Mayr

Institute of Botany, University of Innsbruck, Innsbruck, Austria

Restoration of an adequate water supply in spring is a prerequisite for survival of angiosperm trees in temperate regions. Trees must re-establish access to soil water and recover xylem functionality. We thus hypothesized that prolonged soil frost impairs recovery and affects hydraulics and phenology of Malus domestica var. 'Golden Delicious.' To test this hypothesis, over two consecutive winters the soil around some trees was insulated to prolong soil frosting, From mid-winter to early summer, the level of native embolism, the water and starch contents of wood, bark and buds were quantified at regular intervals and findings correlated with various phenological parameters, xylogenesis and fine root growth. The findings confirm that prolonged soil frost affects tree hydraulics and phenology but the severity of the effect depends on the climatic conditions. In both study years, percentage loss of hydraulic conductivity (PLC) decreased from about 70% at the end of winter to about 10% in May. Thereby, xylem refilling strongly coincided with a decrease of starch in wood and bark. Also treated trees were able to restore their hydraulic system by May but, in the warm spring of 2012, xylem refilling, the increases in water content and starch depolymerization were delayed. In contrast, in the cold spring of 2013 only small differences between control and treated trees were observed. Prolongation of soil frost also led to a delay in phenology, xylogenesis, and fine root growth. We conclude that reduced water uptake from frozen or cold soils impairs refilling and thus negatively impacts tree hydraulics and growth of apple trees in spring. Under unfavorable circumstances, this may cause severe winter damage or even dieback.

#### Keywords: fine roots, native embolism, refilling, starch, tree hydraulics

## INTRODUCTION

Freezing winter conditions in some temperate, continental climates can severely impair water transport in the xylem by freeze-thaw induced embolism. Impaired water transport affects tree water status, water uptake from the soil and especially water supply of the leaves. The prompt restoration of a tree's water-transport system in spring is thus a prerequisite for growth and fruiting that season (Sperry et al., 1994; Ameglio et al., 2002). Water transport in the xylem is driven by a pressure gradient and requires that the water columns are hydraulically continuous (Sperry et al., 1988a; Tyree and Zimmermann, 2002; Choat et al., 2012). Drought and/or freeze-thaw events can break these columns (cavitation) resulting in air-filled conduits (embolisms; Tyree and Zimmermann, 2002). Air is soluble in water but not in ice. Thus, freezing of xylem water leads to

#### Edited by:

Andreas Bolte, Johann Heinrich von Thünen-Institute, Germany

#### Reviewed by:

Bernhard Schuldt, Georg-August-Universität Göttingen, Germany Roberto Tognetti, Università degli Studi del Molise, Italy

> \*Correspondence: Barbara Beikircher barbara.beikircher@uibk.ac.at

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 25 January 2016 Accepted: 02 June 2016 Published: 20 June 2016

#### Citation:

Beikircher B, Mittmann C and Mayr S (2016) Prolonged Soil Frost Affects Hydraulics and Phenology of Apple Trees. Front. Plant Sci. 7:867. doi: 10.3389/fpls.2016.00867

the formation of air bubbles (Hammel, 1967). Depending on the size of the bubbles and the xylem tension (negative pressure), these bubbles can expand during thawing and cause air embolisms (Sperry and Sullivan, 1992; Tyree et al., 1994; Hacke and Sauter, 1996; Cochard et al., 2001; Mayr et al., 2007; Sevanto et al., 2012). A critical factor thereby is conduit diameter, as it is determines the radius of curvature of the approximately hemispherical menisci at the two ends of a bubble (Sperry and Sullivan, 1992; Pittermann and Sperry, 2006; Choat et al., 2011). Accordingly, angiosperms are generally more prone to freezethaw-induced embolisms than gymnosperms (e.g., Hacke et al., 2001; Sperry et al., 2006). In addition, winter embolisms can also be induced by drought. When xylem tensions exceed speciesspecific thresholds, air can be sucked into water-filled conduits from the air-filled intercellular spaces or adjacent embolized conduits (e.g., Sperry et al., 1988a; Tyree and Zimmermann, 2002; Christensen-Dalsgaard and Tyree, 2014). In contrast to freezethaw induced cavitation, relationships between xylem anatomy and drought-induced cavitation are far more complex with pit membrane properties playing a major role (Hacke et al., 2001; Wheeler et al., 2005; Sperry et al., 2006; Cai and Tyree, 2010).

Depending on xylem structure, winter xylem tensions and the number of freeze-thaw cycles, many tree species suffer an almost complete loss of hydraulic conductance in winter (Sperry et al., 1988b, 1994; Cochard and Tyree, 1990; Sperry and Sullivan, 1992; Hacke and Sauter, 1996; Cochard et al., 2001; Jaquish and Ewers, 2001; Ameglio et al., 2002; Christensen-Dalsgaard and Tyree, 2014). This may not be particularly damaging, however, as full water-transport capacity is not necessarily required during winter with low transpiration rates and reduced or blocked water uptake from the soil in many deciduous angiosperms. Thus, embolisms do not generally pose special problems for these deciduous plants in winter. However, circumstances quickly become critical in spring when water demand increases due to increased evaporation rates, on-setting transpiration and growth processes, (e.g., Sperry et al., 1994; Ameglio et al., 2002; Hao et al., 2013). Therefore, tree species that are unable to avoid winter embolism, must have mechanisms that allow rapid restoration of functionality of their water transport systems in order to avoid growth limitation or dieback through shoot desiccation.

Recovery of hydraulic conductance in spring can occur through the refilling of embolized xylem conduits and/or through the differentiation of new ones (e.g., Sperry and Sullivan, 1992; Sperry et al., 1994; Cochard et al., 2001; Jaquish and Ewers, 2001; Ameglio et al., 2002; Christensen-Dalsgaard and Tyree, 2014; Mayr et al., 2014). It is known that many diffuse-porous trees are able to refill some of their embolized conduits in spring (Sperry et al., 1988b, 1994; Sperry and Sullivan, 1992; Hacke and Sauter, 1996; Cochard et al., 2001; Christensen-Dalsgaard and Tyree, 2014). Although such refilling has been demonstrated for many species (see Brodersen and McElrone, 2013, for a detailed review), the mechanisms involved are still a matter of debate. Thermodynamic principles suggest that the refilling of an embolized xylem conduit should be possible only when xylem water pressure is near to or above zero, relative to the atmosphere (Yang and Tyree, 1992; Cochard et al., 2001; Nardini et al., 2011; Delzon and Cochard, 2014). Several angiosperm genera are able to generate positive xylem pressures in the root or stem prior to leaf flushing, and these would be sufficient to dissolve embolisms (Sperry et al., 1987, 1994; Cochard et al., 2001; Ewers et al., 2001; Ameglio et al., 2002; Brodersen and McElrone, 2013; Hao et al., 2013). However, there also seem to be many species which are able to refill embolized conduits in the absence of positive root or stem pressures (Tyree et al., 1999; Cochard et al., 2001; Salleo et al., 2006; Nardini et al., 2011; Brodersen and McElrone, 2013). Several studies suggest that this may be related to starch depolymerisation in xylem parenchymal cells (Salleo et al., 2004; Zwieniecki and Holbrook, 2009; Nardini et al., 2011; Secchi and Zwieniecki, 2012). These cells may also provide some of the water required for the refilling. This idea is supported by the observation of droplets in the refilling grapevine conduits visualized using high-resolution x-ray computer tomography by Brodersen et al. (2010). However, if the xylem is highly embolized, the amount of available water in the xylem parenchyma cells may not be sufficient. Observations of alterations in carbohydrate metabolism and related enzyme activities as well as the up-regulating of aquaporin genes responsible for the water movement from phloem to xylem support the hypothesis of phloem as second source for sugar and water (Salleo et al., 2006; Zwieniecki and Holbrook, 2009; Nardini et al., 2011; Secchi and Zwieniecki, 2011, 2012; Brodersen and McElrone, 2013).

However, water stored in xylem and phloem can only be the first source of water for refilling. In spring, thus sufficient water supply from the soil is not only essential for imminent transpiration and growth but also for xylem refilling. Frozen soils and stem bases prevent water uptake and flow, respectively, and can lead to severe water stress when water is lost by stomatal and/or peridermal transpiration (Larcher, 2003; Beikircher and Mayr, 2013). Water uptake from the soil, though, is also impaired at positive but low soil temperatures. The prime reason therefore is the strong increase in water flow resistance caused by structural changes in the biomembranes of roots which change from a liquid-crystal to a solid-gel state leading to a decrease in cell permeability (Grossnickle, 1988; Ameglio et al., 1990). Additionally, also the increased transfer resistance between soil and root due to a higher viscosity of water at low temperatures may play a role (Larcher, 2003; Pallardy, 2008). Last but not least, the growth of roots is strongly reduced at low soil temperatures, whereby the critical temperature limits vary among species and habitats (Kozlowski et al., 1991; Larcher, 2003). As it is assumed that water uptake occurs mainly over fine roots (Pallardy, 2008), inhibited fine root growth in spring may thus negatively affect embolism repair.

Several studies have shown that delayed soil thawing at contemporaneously high air temperatures negatively affect water relations in terms of sapflow and lead to severe or even lethal damage in conifers of boreal regions: water shortage by delayed or inhibited water uptake can lead to xylem cavitation and in consequence to reduced potential efficiency of Photosystem II and foliar injuries (e.g., Repo et al., 2005, 2008). However, to our knowledge there are no studies dealing with xylem functionality and spring recovery under prolonged soil frost in angiosperm species. In apple orchards in northern Italy, so called 'winter

damage' occurs at irregular intervals. Symptoms show in early spring and take the form of delayed leaf-flushing, dieback of parts of the crown or even of whole trees. Most affected by this phenomenon are young apple trees having reduced water storage capacity and growing on sites with delayed soil warming (northexposed, shaded). Based on climatic conditions in the study area as well as measurements on frost hardiness from autumn to spring (Pramsohler et al., 2012; Pramsohler and Neuner, 2013) on trees of the present study, frost damage of buds and branches can be excluded. We hypothesized that prolonged soil frost in spring delays the recovery of the hydraulic system and thus negatively impact tree growth. To test this idea, soil frosting was artificially prolonged by insulating the soil under selected apple trees in two successive years. From mid-winter to early summer, the level of embolism, the water contents of wood, bark and buds, and the starch contents of wood and bark were measured at intervals in treated and control trees. Tree phenology was also determined including bud phenology, current year ring width and differences in fine root growth.

## MATERIALS AND METHODS

## Plant Material, Climate, and Experimental Design

The field experiment was carried out in a commercial orchard in northern Italy (Tarsch, South Tyrol, 850 m a.s.l.; 46◦ 370N, 10◦ 520NE) on the apple cultivar Malus domestica var. 'Golden Delicious' trees [about 13 years old, 3 m tall, mean diameter at breast height (DBH) 64 mm].

The orchard is situated in an inner alpine dry valley with exceptionally high sunshine duration (315 days), high annual mean temperature (9.6◦C) and low precipitation (450– 550 mm). The field treatments were imposed from February 2012 (about mid-winter) to March 2012 (spring), and repeated on eight other trees in the adjacent tree row from February 2013 to April 2013. In both years, the frozen soil around eight trees was insulated with Styrofoam roof-insulation sheets (RoofmateTM, 40 mm × 600 mm × 1250 mm) to prolong soil frosting. The sheets were mounted along the row over an area of 5 m × 0.6 m. Along the margins of the insulated area, vertical stripes of insulation sheets were inserted to a depth of 0.15 m to reduce lateral heat inflow from the surrounding un-insulated areas (see **Figure 1**). Gaps around stems and between sheets were filled with polyurethane foam (Soudafoam All SeasonsTM, Soudal). For control trees, eight un-insulated trees in the respective adjacent row were chosen. This experimental design may lead to masked responses if investigated parameters are highly variable within the orchard. However, in the framework of another study trees were chosen randomly over the same orchard and no significant differences in water relation and phenology between trees observed (Beikircher, unpublished).

Air temperature and relative humidity (sensor EMS 33) at 2.5 m (in the upper crown) and soil temperatures (sensor Pt 100) at 10 cm depth (the main rooting depth) of both plots were measured at 1 min intervals and the 15 min means

were accumulated in a data logger (ModuLog 3029; sensors and data logger of Environmental Measuring System, Brno, Czech Republic). In 2013, soil water potential in 10 cm depth was measured with two gypsum block sensors connected to a data logger (MicroLog SP, EMS, Brno, Czech Republic). The insulation sheets were removed as soon as the soil temperatures beneath them rose above freezing. From January to July of each year, embolisms and the water contents of bark, wood, and buds and the starch contents of bark and wood were measured at regular intervals (about every 14–21 days) and shoot phenology was monitored. In April 2013, the length of white roots was analyzed immediately after the removal of the insulation.

## Sampling and Preparation of Branches

At each sampling date, five west-exposed branches (several years old) were chosen randomly out of the eight trees per treatment. Because of earlier pruning management, branches were highly branched and crooked and contained several long and short shoots. Branches were cut at the base, immediately re-cut under water above the first annual shoot and left in water under a dark plastic bag for 30–45 min. This procedure was of particular importance to avoid entrance of air at the cut surface and thus artificial embolism in (I) winter, when frozen branches were thawing and (II) after bud break due to transpiration. Branches were then wrapped in dark plastic bags and transported to the laboratory. There, three samples per branch were cut under water from shoots developed in the preceding growing season (i.e., 2012 and 2013, respectively) for analyses of native embolism, water, and starch content. In a precedent study (Beikircher and Mayr, 2015), this harvesting and sampling protocol was proved to be appropriate to preserve the hydraulic state in the shoots used for measurements by avoiding artifactual embolization (Wheeler et al., 2013) as well as refilling (Trifilo et al., 2014).

## Native Embolism, Water Content, and Midday Water Potential

Xylem recovery was assessed by measuring the level of native embolism from winter to spring. Samples about 40 mm long were de-barked, ends were re-cut several times with a sharp wood-carving knife and sealed under water in tubes connected to a "Xylem" hydraulic conductance and embolism measurement system (Bronkhorst, Montigny les Cormeilles, France). The level of native embolism was expressed as the percentage loss of hydraulic conductance (PLC) and measured as the hydraulic conductance at 4.5 kPa before (ki) and after (kmax) removal of embolism by repeated high pressure flushes at 95 kPa for 20 min (Eq. 1):

$$\text{PLC} = 100 - (k\_{\text{i}}/k\_{\text{max}} \times 100) \tag{1}$$

For measurements, distilled, filtered (0.2 µm) and degassed water was used, containing 0.005% (v/v) 'Micropur Forte MF 1000F' (a mixture containing Ag+ and sodium hypochlorite for water sterilization and preservation; Katadyn Products Inc., Wallisellen, Switzerland) to prevent microbial growth (Sperry et al., 1988a; Beikircher and Mayr, 2008).

For water content analyses, about 100 mm long samples were de-barked and the fresh weights (FW) of bark, wood, and terminal buds were measured with an analytical balance (Sartorius BP61S, 0.0001 g precision, Sartorius AG, Göttingen, Germany). After oven drying at 80◦C for 48 h, dry weight (DW) was determined and the water content expressed as a percentage of DW (WC%DW) as:

$$\text{WC}\_{\% \text{DW}} = (\text{FW} - \text{DW}) / \text{DW} \times 100 \tag{2}$$

From bud break to June, midday water potential were measured on selected days between 11:00 and 12:00 CET. Measurements were made on at least five end twigs per treatment and date. Prior to bud break, measurement of xylem water potential was not possible due to the small portion of living tissues.

#### Starch Contents of Wood and Bark

For starch analyses, the bark was removed from samples (∼40 mm long) and cut into thin strips with scissors. The wood was sliced with a microtome (Sledge Microtome G.S.L. 1, Schenkung Dapples, Zurich, Switzerland). After oven drying to constant weight at 80◦C for 48 h, samples were finely ground with a micro-dismembrator (Braun Biotech, Melsungen, Germany). The materials thus obtained were incubated twice in 80% ethanol (v/v) at 75◦C, with polyvinylpyrrolidone (PVP 40, Sigma Chemicals, USA) being added to bind polyphenols during the first incubation. The supernatants were then removed and the solvent evaporated in an oven. The dry residue containing the starch was incubated with sodium hydroxide (0.5 N) at 60◦C for 1 h, neutralized with hydrochloric acid (0.5 N), treated with amylase (amyloglucosidase, Sigma–Aldrich, USA) dissolved in citrate buffer (pH 4.6) to break the starch down to glucose and incubated for 30 min at 60◦C. The supernatants were then mixed with NADP/ATP and HK/G6P-DH-solution (EnzytecTM E1210, r-biopharm, Germany) and the absorption of NADPH (dependent on starch content) measured with a UV/Vis

spectrophotometer (Lambda 20, Perkin Elmer, USA) at 340 nm before and after the addition of the latter solution. Starch content (SC; %) was calculated as:

$$\text{SC} = \mathcal{c} \times 100 \times V/m \tag{3}$$

where c is the starch concentration (g/L) in the measured solution, m is the mass of plant material (g) and V (L) is the volume of the solution in the spectrophotometer (also see Mayr et al., 2014).

## Shoot Phenology and Length of White Roots

The phenology of the shoot was monitored from winter to the initial stages of fruit development. Phenological stages were classified according to the BBCH-scale (Chapman and Catlin, 1976; Meier, 2003; see **Table 1**). To investigate xylem development, cross-sections were made with a microtome and current year ring width was measured on five shoots per treatment. In 2013, after removal of the insulation, 8–10 roots from three trees per treatment were harvested at 10 cm depth (the main root zone) and taken to the laboratory. The length of each visible white root was analyzed with a stereo microscope (Olympus SZ61, Olympus Austria, Vienna, Austria) interfaced with a digital camera (Sony Cyber-shot DSC-W17) and image analysis software [ImageJ, 1.37, National Institutes of Health (NIH), Bethesda, MD, USA, public domain].

TABLE 1 | Phenological stages of apple trees according to the BBCH-scale (Chapman and Catlin, 1976; Meier, 2003).


### Statistics

Differences between treatments in the levels of native embolism, water and starch contents and lengths of white roots were tested with the Student's t-test (normal distribution and equal variances) or the non-parametric Mann–Whitney U Test (no normal distribution and/or unequal variances). All tests were made pairwise at a probability level of 5% using SPSS version 21. Correlation coefficients were tested with the Pearson's productmoment coefficient.

## RESULTS

### Weather Conditions and Insulation

Weather conditions differed considerably between the 2 years of the study. The following description is based on the monthly climate report of the South Tyrolean Weather Service as well as our own measurements with the meteorological station in the orchard. The first half of February 2012 was exceptionally cold with mean air temperatures about 1.5◦C below the longterm average. The cold period was followed by a sudden increase in temperature and extraordinarily warm temperatures in the second half of February (**Figure 2A**). This caused an early onset of growth and one of the earliest bud breaks ever observed in the study area (South Tyrolean Advisory Service for Fruit and Winegrowing, pers. comm.). Besides temperature extremes, February 2012 was also extremely dry due to low precipitation and foehn winds. Also, March 2012 was exceptionally warm with mean air temperatures about 3.2◦C above the long-term average. In April, air temperatures were close to the long-term average and precipitation was significantly higher.

January 2013 also experienced above-average air temperatures and precipitation but (in contrast to 2012) cold periods occurred in February and March. Although relatively mild temperatures at the end of February and the beginning of March induced the onset of bud swelling, the low temperatures in the second half of March stopped further development until temperatures again increased at the end of March and beginning of April (**Figure 2D**). Because of the air temperature differences between 2012 and 2013, soil temperatures also differed considerably. In 2012, the soil in the control plots thawed during the second half of February and by mid-March, mean soil temperatures were consistently above 4◦C (**Figure 2C**). In 2013, soil thawing also started at the end of February but soil temperatures remained low until the end of March. Due to the early irrigation start in February, in both study years soil water potential never fell below −0.30 MPa from soil thawing in January/February to freezing in November/December (**Figure 3G**, also see Beikircher et al., 2013).

Insulation of the soil with the Styrofoam sheets prolonged soil frosting for several weeks. In 2012, daily mean soil temperatures in the control plots rose above zero on 23 February, while that in the insulated plots remained frozen for another 23 days. Differences in soil temperature between plots ranged from about 1 to 6◦C, with a difference greater than 3◦C on 11 days (**Figure 3G**). In contrast, in 2013, soil thawing started about 4 weeks earlier in the control plots (27th February) compared with the insulated plots. However, due to the subsequent weather conditions, temperature differences were small. Differences were less than 1◦C on 17 days and greater than 3◦C only on the last 2 days of insulation (**Figure 3G**). We cannot completely exclude that the insulation has altered soil water relations by preventing precipitation from reaching the root zone. However, as insulation was immediately removed when soil started thawing (see "Materials and Methods" section) and due to the daily irrigation from February onward, soil water availability upon thawing was high for both control and treated trees. Accordingly, soil water potential measurements in 2013 revealed similar values upon natural thawing and removal of insulation, respectively (**Figure 3G**).

## Native Embolism, Water, and Starch Content

In January of both years of the study, a hydraulic conductance loss of about 70% was measured in the control trees. Due to refilling processes, the level of native embolism decreased to less than 10% by the end of spring (**Figure 3A**). Simultaneously, the water contents of wood, bark and buds increased strongly (**Figures 3B– D**). Prolonging the soil frost by surface insulation had a strong influence on embolism reversal, particularly in 2012. Significant differences between control and insulated plots were found from 6 March to 24 April. In contrast, in 2013, significant differences were found only on one sampling date (24 April; **Figure 3A**). In 2012, simultaneous increases in the water contents of wood, bark and buds occurred in both treatments but were significantly higher in the control trees. In contrast, embolism differences had already leveled off by April 2012. In March 2013, the water contents of wood and buds were slightly higher in the control trees but significant differences were only found for the buds (**Figures 3B–D**). In contrast, midday water potential from bud break (after removal of insulation) to June was relatively constant and ranged between −0.23 and −1.45 MPa. No differences between treatments were observed. Starch contents in the bark were low during winter (ca. 2%), increased around bud break (7%) but then decreased again during refilling. No significant differences between control and treated trees were observed (**Figure 3F**). In contrast, xylem starch content was consistently high during winter (9–11%). In 2012, it decreased to almost 0% in the control trees and to about 2% in the treated trees with the differences being significant from mid-March to the end of May. In 2013, the starch content of the wood decreased to only about 4%, with treatment differences being non-significant (**Figure 3E**). In control trees, for both study years a strong correlation between PLC and starch content of the xylem was found (**Figure 4**).

#### Phenology and Length of White Roots

In 2012, bud break occurred about 3 weeks earlier than in 2013 (**Figures 2D** and **3D**) with prolonged soil frost significantly affecting phenology. Bud break of treated trees was about 9 days after that of the control trees and the subsequent phenological stages were also delayed in the treated trees, compared to the controls (**Figure 5**). In 2013, there were no significant phenological differences between treatments (data not shown).

FIGURE 2 | Mean daily air temperature (A), air humidity (B), soil temperature at 10 cm depth (C), and phenological stages of Malus domestica 'Golden Delicious' (D) of control plots from January to June 2012 (black lines and symbols) and 2013 (gray lines and symbols). Dashed vertical lines show bud break. For an explanation of the phenological stages, see Table 1.

Root growth was clearly affected by prolonged soil frost. On 24 April 2013, about 10 days after bud break in control trees, the mean length of white roots of control trees (2.03 ± 0.20 mm) was significantly higher than that of treated trees (1.02 ± 0.13 mm; **Figure 6**). In contrast to the treated trees, the control trees had developed numerous white roots with lengths greater than 4 mm (**Figure 6**).

#### DISCUSSION

During the vegetation period, irrigation management in most commercial apple orchards is optimized to ensure high quality and productivity (Naor and Girona, 2012; also see Beikircher et al., 2013). However, hydraulic limitations of growth and yield are related not only to summer drought. During autumn and

winter, the soil–plant-atmosphere continuum is interrupted by winter embolism and inhibited water uptake from the frozen soil (frost drought). Transpirational water loss by the tree due to late or failed leaf shedding or from the periderm can further lower the water status of apple trees in winter (Beikircher and Mayr, 2013). A timely restoration of the tree's hydraulic transport system (stem) and an efficient re-connection to the soil water (roots) in spring is thus crucial. In this study we analyzed the impacts of prolonged soil frost on xylem recovery and growth in M. domestica var. 'Golden Delicious.'

## Xylem Recovery and Phenology of Control Trees

Weather conditions varied considerably between the two study years. While the spring of 2012 was exceptionally warm and dry, the weather in 2013 was rather cold and wet. This had strong influences on the time course of the trees' xylem recovery and growth. In mid-winter of both years, the loss of hydraulic conductance in control trees was about 70% but this decreased to negligible values in spring (**Figure 3A**). A similar pattern of winter embolism and subsequent xylem recovery has been reported for many angiosperm species (e.g., Sperry et al., 1988a, 1994; Sperry and Sullivan, 1992; Cochard et al., 1997; Ameglio et al., 2002) inclusive M. domestica var. 'Alberta green' (Christensen-Dalsgaard and Tyree, 2014). In February 2012, a rapid increase in air and soil temperatures triggered the restoration of the hydraulic conductance in the stem with the level of native embolism decreasing steadily until May. In contrast, air and soil temperatures in 2013 remained below 5◦C until the end of March. In consequence, the level of embolism remained high (about 60%) until April when it decreased rapidly as soon as weather conditions became more favorable

FIGURE 5 | Typical phenological stages of control and treated trees of M. domestica from February to May 2012. Bars 1 cm.

(**Figure 3**). However, in both years the level of native embolism was below 10% by June. This observation is similar to that of Christensen-Dalsgaard and Tyree (2014), who also found only small differences in the refilling success of apple, willow and poplar trees, despite significantly different weather conditions during spring. The residual 10% loss of hydraulic conductance

may be due to damage, cavitation- or frost-fatigue of conduits (Hacke et al., 2001; Christensen-Dalsgaard and Tyree, 2014).

In 2012, a significant decrease in native embolism was observed prior to the growth of new conduits (level of native embolism in February differed significantly from following values; significance not shown in **Figure 3**), while in 2013 both processes occurred in parallel. Thus, in M. domestica, xylem recovery is a result not only of the differentiation of new xylem conduits but also of the refilling of embolized ones (also see Christensen-Dalsgaard and Tyree, 2014).

Refilling of embolized conduits in spring can be related to positive xylem pressures. The genus Malus has been reported to develop root pressure in spring but to a much lesser extent than maple, walnut or birch for instance (Wiegand, 1906). Thus, root pressure might have been involved in very first refilling processes. However, in Malus as in other species, root pressure ceases before or around bud break (e.g., Wiegand, 1906; Sperry et al., 1994; Cochard et al., 2001; Ewers et al., 2001; Ameglio et al., 2002; Hao et al., 2013). At that time, refilling is not yet completed for several more weeks (Sperry et al., 1994; Christensen-Dalsgaard and Tyree, 2014). Recent studies indicate that vessel refilling in the absence of positive pressures is based on starch depolymerisation and subsequent water inflow to embolized conduits (see 'Introduction'). However, to our knowledge, there is only one other study linking xylem recovery in spring with seasonal patterns in starch content: In their study, Mayr et al. (2014) found a close relationship between starch content and refilling in Picea abies. In our study trees, vessel refilling coincided with a decrease in the starch contents of both bark and wood (**Figure 3**). Thereby, decrease in native embolism was strongly correlated with the decrease of starch in the wood (**Figure 4**). In the bark, prior to the decrease in spring, an increase in starch content was observed in late winter, which is a commonly observed phenomenon in many temperate tree

species (Essiamah and Eschrich, 1985; Sauter and van Cleve, 1994; Ameglio et al., 2004). Contemporary with the formation of new conduits (increase in current year branch wood ring width), a significant increase in wood water content was observed (**Figure 3**). Strati et al. (2003) and Hao et al. (2013) found the increase in the wood's volumetric water content was correlated with the refilling of embolised conduits. The increase in the water contents of the bark and buds started earlier than of the wood and was likely related to water uptake by young cells as they extended and grew (Larcher, 2003; Neuner and Beikircher, 2010).

Weather conditions during spring influenced not only tree hydraulics but they also had significant effects on tree phenology. In February 2012, a strong increase in bud water content was observed, followed by an exceptionally early bud break on March 10th (**Figure 3D**). Obviously, this was related to favorable weather conditions lasting for several weeks. In 2013, bud swelling also started in February but due to the subsequent unfavorable weather conditions, the buds remained in this phenological stage until the beginning of April (**Figures 2** and **3**).

### Impact of Prolonged Soil Frost on Hydraulics and Phenology

Prolongation of soil frost for 3–4 weeks clearly affected apple tree hydraulics and phenology. The greatest effects were observed in 2012, when soil temperatures differed considerably between control and treated trees (**Figure 3G**). About 14 days after the first significant decrease in embolism of control trees (February), a significant but small decrease was also observed in the treated trees. However, the level of native embolism in treated trees then remained stable at around 60%. Despite starch depolymerisation in the wood at the beginning of April, the next significant decrease in embolism was not observed until the end of April. This indicates that limited uptake of water from the soil, and not starch metabolism, was the determining factor for xylem recovery in the treated trees. This hypothesis can be further proofed by the two-phase relationship between soil temperature and level of native embolism: Up to about 7◦C level of embolism remained constant and then decreases rapidly toward 0% embolism (data not shown). Studies on Scots pine in the boreal region also showed a delay in the onset of sap flow when soil thawing lagged behind the start of the growing season (Repo et al., 2008). As with recovery from embolism, the water contents of wood, bark and buds showed a delayed increase in the treated trees, even though differences had already leveled off at the beginning of April (**Figures 3B–D**). In contrast to 2012, differences in the soil temperatures of control and treated trees were less dramatic in spring 2013. Due to cold and wet weather conditions, the soil in the control plots was only slightly above freezing until the end of March. As a consequence, the loss of hydraulic conductance remained high in all trees and significant differences were found only at the end of April (**Figure 3A**). This shows that not only soil frost but also cold soils slow xylem recovery in apple trees. However, in contrast to Repo et al. (2005) who found that a 2-week delay in soil thawing led to the death of Scots pine seedlings, we did not observe any visible damage of treated trees.

Frozen or cold soils strongly limit water uptake due to reduced root water permeability, increased transfer resistances, and inhibited fine root growth (see 'Introduction'). Water uptake in trees occurs mainly in the fine roots (Pallardy, 2008). Newly formed roots are white, unsuberized and highly permeable to water. In apple trees, white roots live for from 1 to 4 weeks in summer and for up to 3 months in winter (Wells and Eissenstat, 2001; Pallardy, 2008). Therefore, fine root survival rate during winter depends strongly on the prevailing soil conditions (Psarras et al., 2000; Wells and Eissenstat, 2001; Pallardy, 2008). According to Rogers (1939), the onset of fine root growth in apple trees starts at soil temperatures above 6◦C and also according to Alvarez-Uria and Körner (2007), 6◦C is a critical temperature limit for root growth in many temperate trees. Several authors reported that the main period of root growth starts several weeks after bud break (Head, 1967; Psarras et al., 2000; Wells and Eissenstat, 2001). We analyzed the length of white roots 10 days after bud break on April 10th, 2013. At this stage, the soil temperature around the control trees lay between 4 and 6◦C for about 2 weeks but for only a few days around the treated trees. This difference had a strong effect: the white-root length of the control trees was about twice that of the treated trees (**Figure 6**). Although water uptake in spring may occur partly via overwintering fine roots and older roots (Landsberg and Jones, 1981; Wells and Eissenstat, 2001, 2003), it is likely that the observed limitation in fine root growth negatively influenced xylem recovery. However, further research is required to proof this hypothesis and determine the importance of water uptake by white roots for refilling.

Besides xylem recovery, prolonged soil frost also had strong impacts on phenology. In 2012, bud break occurred about 10 days later in treated trees than in control trees. As a consequence, subsequent phenological stages were also delayed but the differences diminished with time and were negligible by June (**Figure 5**). In 2013, a short, warm spell at the beginning of March triggered bud swelling (also evidenced by increased bud water content) in control trees but due to the later deterioration in weather, the trees remained at this phenological stage. Thus, bud break was almost contemporary in control and treated trees. Greer et al. (2006) also found bud break to be strongly correlated with soil temperatures. According to these authors, apple trees store carbohydrates predominately in the roots, and low soil temperatures affect bud break by reducing carbohydrate remobilisation. However, growth also strongly depends on water status. While water stored within the tree may be sufficient for the very earliest growth processes, further development requires access to soil water.

Overall, it is evident that prolonged soil frost critically affects both tree hydraulics and phenology but these are affected to different extents. While for the level of embolism, significant differences between treatments were observed from the beginning of March to the end of April 2012, growthrelated parameters showed less pronounced differences. For instance, bud break was delayed by only 10 days in 2012. Though water availability is undoubtedly important for growth, air temperatures also play a major role.

## CONCLUSION

fpls-07-00867 June 16, 2016 Time: 13:2 # 11

Successful restoration of the hydraulic system after winter is essential for the survival of temperate trees. Nevertheless, there are remarkably few studies dealing with the recovery of the hydraulic system in spring, its underlying mechanisms and possible limitations. There are some studies focusing on seasonal patterns in xylem embolism, some dealing with the impact of prolonged soil frost on sap flow and some on fine root growth in spring (see above). Many of those studies have been carried out on conifers. To our knowledge, our study is the first (I) showing the impact of prolonged soil frost on xylem recovery and phenology by (II) correlating xylem functionality to starch content and xylogenetic aspects and (III) including fine root growth in an angiosperm over (IV) two consecutive spring seasons with contrasting weather conditions.

Our findings show that prolonged soil frost can have strong impacts on xylem recovery and phenology of apple trees in spring. The main reason therefore is the limited water uptake from the cold soil due to increased resistance and impaired fine root growth rather than starch metabolism. However, to which extent prolonged soil frost affects xylem recovery and tree phenology strongly depends on weather conditions. Thus, prolonged soil frost may be critical when a number of unfavorable conditions coincide. For example, when soil frost persists in conjunction with early bud break and thus high water demand for transpiration and growth, or weather conditions favoring high transpirational demand, or increased water loss from the periderm of lammas shoots in winter and spring (see Beikircher and Mayr, 2013), or reduced water storage capacity. Other factors limiting water uptake are cold soil and high fine-root mortality due to earlier frost injury (Groffman et al., 2001; Tierney et al., 2001). These may lead to the severe winter damage as observed at irregular intervals in the apple orchards of northern Italy (see 'Introduction'). In particular locations, where some

#### REFERENCES


of these factors occur fairly often, apple production may not be commercially sustainable. However, facing climate change prolonged soils frost may also have implications for other trees and other ecosystems such as boreal forests when soil thawing lags behind the start of the growing season.

## AUTHOR CONTRIBUTIONS

BB organized the field experiment, carried out the measurements in the second study year, supervised the measurements of CM in the first study year and wrote the presented manuscript. In the frame of her master thesis, CM carried out all measurements of the first study year. SM is leader of the research group, wrote the project proposal financing the presented study, supervised the experiment, counseled the PostDoc BB and revised the presented manuscript.

### FUNDING

This study was financed by the Austrian Science Fund (FWF; project L556-B16; 'Winter damage on apple trees' and project T667-B16 'Hydraulics of juvenile trees') and is linked to the COST Action FP1106 'STReESS.'

#### ACKNOWLEDGMENTS

We want to thank the staff of the South Tyrolean Advisory Service for Fruit and Wine-growing, especially Martin Abler, for helpful support as well as the farmers which enabled the study in their orchards. We also thank Mag. Birgit Dämon for helpful assistance with starch analyses and Dr. Walter Oberhuber for insights into root growth.




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Beikircher, Mittmann and Mayr. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Osmolality and Non-Structural Carbohydrate Composition in the Secondary Phloem of Trees across a Latitudinal Gradient in Europe

Anna Lintunen<sup>1</sup> \*, Teemu Paljakka<sup>1</sup> , Tuula Jyske<sup>2</sup> , Mikko Peltoniemi <sup>2</sup> , Frank Sterck <sup>3</sup> , Georg von Arx <sup>4</sup> , Hervé Cochard<sup>5</sup> , Paul Copini 3, 6, Maria C. Caldeira<sup>7</sup> , Sylvain Delzon<sup>8</sup> , Roman Gebauer <sup>9</sup> , Leila Grönlund<sup>1</sup> , Natasa Kiorapostolou<sup>3</sup> , Silvia Lechthaler <sup>10</sup> , Raquel Lobo-do-Vale<sup>7</sup> , Richard L. Peters <sup>4</sup> , Giai Petit <sup>10</sup>, Angela L. Prendin<sup>10</sup> , Yann Salmon<sup>11</sup>, Kathy Steppe<sup>12</sup>, Josef Urban<sup>9</sup> , Sílvia Roig Juan<sup>2</sup> , Elisabeth M. R. Robert 13, 14, 15 and Teemu Hölttä<sup>1</sup>

#### Edited by:

Achim Braeuning, University Erlangen-Nuremberg, Germany

#### Reviewed by:

Ivika Ostonen, University of Tartu, Estonia Jürgen Kreuzwieser, University of Freiburg, Germany

#### \*Correspondence:

Anna Lintunen anna.lintunen@helsinki.fi

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 27 January 2016 Accepted: 11 May 2016 Published: 01 June 2016

#### Citation:

Lintunen A, Paljakka T, Jyske T, Peltoniemi M, Sterck F, von Arx G, Cochard H, Copini P, Caldeira MC, Delzon S, Gebauer R, Grönlund L, Kiorapostolou N, Lechthaler S, Lobo-do-Vale R, Peters RL, Petit G, Prendin AL, Salmon Y, Steppe K, Urban J, Roig Juan S, Robert EMR and Hölttä T (2016) Osmolality and Non-Structural Carbohydrate Composition in the Secondary Phloem of Trees across a Latitudinal Gradient in Europe. Front. Plant Sci. 7:726. doi: 10.3389/fpls.2016.00726 <sup>1</sup> Department of Forest Sciences, University of Helsinki, Helsinki, Finland, <sup>2</sup> Natural Resources Institute Finland, Vantaa, Finland, <sup>3</sup> Forest Ecology and Forest Management Group, Department of Environmental Sciences, Wageningen University, Wageningen, Netherlands, <sup>4</sup> Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Birmensdorf, Switzerland, <sup>5</sup> INRA, UMR 547 PIAF, Université Clermont Auvergne, Clermont-Ferrand, France, <sup>6</sup> Alterra, Wageningen University and Research Centre, Wageningen, Netherlands, <sup>7</sup> Forest Research Centre, School of Agriculture, University of Lisbon, Lisbon, Portugal, <sup>8</sup> INRA, University of Bordeaux, UMR BIOGECO, Talence, France, <sup>9</sup> Department of Forest, Botany, Dendrology and Geobiocenology, Mendel University in Brno, Brno, Czech Republic, <sup>10</sup> Department Territorio e Sistemi Agro-Forestali, Legnaro (PD), Università degli Studi di Padova, Padova, Italy, <sup>11</sup> Department of Physics, University of Helsinki, Helsinki, Finland, <sup>12</sup> Laboratory of Plant Ecology, Department of Applied Ecology and Environmental Biology, Ghent University, Gent, Belgium, <sup>13</sup> Centre for Ecological Research and Forestry Applications (CREAF), Cerdanyola del Vallès, Spain, <sup>14</sup> Laboratory of Plant Biology and Nature Management (APNA), Vrije Universiteit Brussel, Brussels, Belgium, <sup>15</sup> Laboratory of Wood Biology and Xylarium, Royal Museum for Central Africa (RMCA), Tervuren, Belgium

Phloem osmolality and its components are involved in basic cell metabolism, cell growth, and in various physiological processes including the ability of living cells to withstand drought and frost. Osmolality and sugar composition responses to environmental stresses have been extensively studied for leaves, but less for the secondary phloem of plant stems and branches. Leaf osmotic concentration and the share of pinitol and raffinose among soluble sugars increase with increasing drought or cold stress, and osmotic concentration is adjusted with osmoregulation. We hypothesize that similar responses occur in the secondary phloem of branches. We collected living bark samples from branches of adult Pinus sylvestris, Picea abies, Betula pendula and Populus tremula trees across Europe, from boreal Northern Finland to Mediterranean Portugal. In all studied species, the observed variation in phloem osmolality was mainly driven by variation in phloem water content, while tissue solute content was rather constant across regions. Osmoregulation, in which osmolality is controlled by variable tissue solute content, was stronger for Betula and Populus in comparison to the evergreen conifers. Osmolality was lowest in mid-latitude region, and from there increased by 37% toward northern Europe and 38% toward southern Europe due to low phloem water content in these regions. The ratio of raffinose to all soluble sugars was negligible at mid-latitudes and increased toward north and south, reflecting its role in cold and drought tolerance. For pinitol, another sugar known for contributing to stress tolerance, no such latitudinal

pattern was observed. The proportion of sucrose was remarkably low and that of hexoses (i.e., glucose and fructose) high at mid-latitudes. The ratio of starch to all non-structural carbohydrates increased toward the northern latitudes in agreement with the build-up of osmotically inactive C reservoir that can be converted into soluble sugars during winter acclimation in these cold regions. Present results for the secondary phloem of trees suggest that adjustment with tissue water content plays an important role in osmolality dynamics. Furthermore, trees acclimated to dry and cold climate showed high phloem osmolality and raffinose proportion.

Keywords: hexose, osmotic concentration, phloem water content, pinitol, raffinose, sucrose, starch

### INTRODUCTION

Plants have to keep osmolality levels sufficiently high in the phloem to maintain basic cell metabolism processes (see Rodríguez-Calcerrada et al., 2015) and cell turgor at levels that allow growth (Kröger et al., 2011). Phloem osmolality levels also have a role in various physiological processes in plants e.g., biomass accumulation (Simard et al., 2013; Steppe et al., 2015), control of transpiration (Schroeder et al., 2001), and maintaining xylem hydraulic integrity (Sala et al., 2012; Sevanto et al., 2014; Pfautsch et al., 2015). In addition, high osmolality decreases the wilting point (Bartlett et al., 2012a,b; Charrier et al., 2013a,b) and the ice nucleation temperature (Burke et al., 1976) of living cells thus affecting their ability to tolerate drought and freezing temperatures.

Plants may vary in phloem osmolality because they differ in the control of sugar concentrations (or other osmotic substances), or they differ in phloem water content, i.e., cell osmolality can be increased either by an increase in the amount of solutes or a decrease in the amount of water in the cell. In dry climates, the maintenance of cell turgor may require higher osmolality to compensate for low stem water potentials. In cold climate, such as in the boreal zone, high osmolality and high carbon storage may both be required to avoid symplastic freezing during winters. So far, such processes have been studied for leaves (e.g., O'Neill, 1983; Gross and Koch, 1991; Callister et al., 2008; Bartlett et al., 2014; O'Brien et al., 2014; Maréchaux et al., 2015) but only scarcely for the secondary phloem in plant stems or branches. It has been shown that sucrose concentration in the secondary phloem of Picea abies increases with increasing latitude in Finland (Jyske et al., 2015). Moreover, the comparison of studies suggests that the sugar concentration in the secondary phloem increases with increasing elevation in Larix decidua (Hoch et al., 2003; Streit et al., 2013), but we lack empirical tests over continental scale on the ability of secondary phloem of trees to osmotically adjust to different climates.

Studies for the osmolality and non-structural carbohydrate (NSC) concentration in the secondary phloem are needed, because the secondary phloem is structurally different from the primary phloem in leaves. The secondary phloem includes noncollapsed and collapsed tissue. Non-collapsed tissue is typically the youngest part of the phloem, whereas older layers in the outer part of the secondary phloem collapse and become storage tissue (Evert, 2006). Sugars are transported between loading (at C sources) and unloading sites (at C sinks) in sieve elements in the non-collapsed phloem tissue. The transport is driven by a gradient in osmotically established turgor pressure (Münch, 1930; Thompson, 2006; De Schepper et al., 2013), and is coordinated with the axial gradient of water potential developed along the xylem compartments (Hölttä et al., 2006). Secondary phloem also needs to tolerate seasonal drought and cold stresses, whereas these stresses can be avoided in the leaves of deciduous species by shedding.

Phloem osmolality is a measure of the moles of solute per kilogram of solvent (mol kg−<sup>1</sup> ), and there are different types of solutes that contribute to it: soluble sugars, ions and amino acids. Sucrose is a soluble sugar that is considered as the most important compound being translocated in phloem elements (Pate, 1976). Hexoses (i.e., glucose and fructose) are present in high amounts in all living cells, and can also be important transport sugars in the phloem for some species (Van Bel and Hess, 2008). Raffinose and pinitol occur in small amounts in phloem, but may contribute to protecting cells against environmental stress, such as drought and low temperatures (Bohnert and Shen, 1999; Zuther et al., 2004; Deslauriers et al., 2014). Starch, which is the most common storage form of non-soluble carbohydrates, contributes only marginally to the value of osmolality due to its high molar mass.

Non-structural carbohydrates (i.e., soluble sugars and starch) are constantly transformed from one form to another. Starch, for example, is formed when high levels of soluble sugars occur, and is transformed to sugars if sugar content is low (Escobar-Gutiérrez et al., 1998). Amount and composition of NSC in phloem tissue show a seasonal behavior in temperate and boreal regions (Hoch et al., 2003; Simard et al., 2013; Jyske et al., 2015), and have an important role in the development of cold hardiness: starch is converted into sugars during cold acclimation (Zwieniecki et al., 2015). Furthermore, a fraction of NSCs can be converted to defensive chemicals in some species (Kozlowski, 1992).

In this study, we aimed at showing variation in osmolality and non-structural carbohydrate composition in secondary branch phloem across a large geographical and climatic gradient. We hypothesize that (i) variability in osmolality is mainly controlled by solute content over large geographical scale, (ii) osmolality and solute content increase from the mid-latitudes toward the more drought-prone lower latitudes as well as to more coldstressed higher latitudes, and locally from moist to dry soil sites, and (iii) the share of raffinose and pinitol among soluble sugars increase from the mid-latitudes toward south and north given their role in tolerating drought and cold stress. To test these hypotheses we collected branches from four widely distributed species Pinus sylvestris, Picea abies, Betula pendula and Populus tremula from moist and dry soil sites, from boreal, temperate and Mediterranean regions across Europe. For phloem samples of those branches, we analyzed osmolality, concentrations of different sugars and water content, and discuss patterns across the studied geographic/climatic gradient.

## MATERIALS AND METHODS

#### Plant Material

We conducted a European wide study on Scots pine (Pinus sylvestris L.), Norway spruce (Picea abies (L.) Karst.), silver birch (Betula pendula Roth.) and common aspen (Populus tremula L.). These four species have a wide distribution and cover both deciduous angiosperm species and evergreen coniferous species. In total, we studied trees in seven regions along a climate gradient across Europe from northern Finland (67◦N 29◦E) to Portugal (40◦N 7◦W) (**Figure 1**, **Table 1**), and selected one moist soil site and one dry soil site per region based on soil type, ground vegetation, and soil moisture measurements. Measurements on needle lengths of Pinus and Picea showed that the needles from the moist soil sites were longer than the ones from the dry soil sites within each region with the exception of Pinus at the northern Finland (Figure S1 in the Supplementary Material). The climate gradient runs from cold and slow growth conditions in the north, through higher growth conditions at mid-latitudes, to drier and slower growth conditions in the south (**Table 1**).

For each region, we selected five trees per species and per moist and dry soil site (**Table 1**). We selected healthy trees more than 5 m in height to harvest one 0.7-m-long branch (linear distance from tip) that was fully exposed to light in order to avoid shading effects. Fixed distance from the branch tip was selected for sampling to fix the transport distance from the C source to the sampling location. Branches were cut at 1 pm or later in the afternoon to minimize the impact of confounding diurnal trends in osmolality. Two 5-cm-long branch segments between distances 70 and 60 cm from the branch tip were cut. The basipetal segment was put in 50% ethanol for anatomical analysis of phloem area. The acropetal segment was sealed in a plastic

data with forests in green, and USGS digital elevation model.



 number of growth rings in the xylem tissue at the fixed 0.7-m-sampling distance from branch tip. Especially in the case of Betula pendula, some of the five sampled trees per species and per site were removed from the dataset dueinadequate sampling material for osmolality measurements.

 to

Lintunen et al. Osmolality and NSC in Branch Phloem

tube and frozen immediately in the field in liquid nitrogen or dry ice for osmolality and water content measurements. Sampling was performed in late summer of 2014 after the end of seasonal secondary growth but before leaf senescence in the different regions.

#### Phloem Osmolality

Phloem osmolality measurements were conducted in the laboratory at the University of Helsinki. Frozen samples were brought to room temperature for 15 min to thaw. Freezing and thawing the samples rapidly breaks the cell membranes and releases symplastic contents to the apoplast (Kikuta and Richter, 1992; Callister et al., 2006). The outer bark was scraped away with a razor blade and each sample cut in two 2 cm pieces in order to have two subsamples of each branch. The inner bark (including cambium and all the tissues from cambium to the innermost periderm) was separated from xylem on the basis of the hardness and color differences between the tissues and weighted for fresh mass (FM). The samples were set in silicabased membrane collection tubes (GeneJET Plasmid Miniprep Kit, Thermo Scientific, Massachusetts, USA) into a centrifuge (Heraeus Fresco 17, Thermo Scientific, Massachusetts, USA) at 14,000 g for 10 min (Devaux et al., 2009). The liquid was collected in osmometer tubes and the osmolality of the liquid was measured with a freezing-point osmometer (Osmomat-030 Freezing point osmometer, Gonotec, Berlin, DE).

We assumed that the ratio of phloem tissue volume to whole inner bark tissue volume is large enough that it is justified to refer to the collected inner bark sap as phloem sap.

## Amount of Solutes and Water Content

In order to get comparable information of the accumulation of cellular solutes in secondary phloem in different tree individuals growing in different regions, phloem osmolality measurements either need to be analyzed at full tissue saturation (Rosner et al., 2001) or be connected with tissue water content measurements. To determine the amount of solutes (n) and water content (WC) in the inner bark tissue, we randomly selected three of the five phloem osmolality samples per species and site. The samples were dried at 80◦C in an oven for 72 h to obtain their dry mass (DM). WC (g g−<sup>1</sup> DM) was calculated as the difference between FM and DM divided by the DM, and n (mol kg−<sup>1</sup> DM) was calculated as osmolality multiplied by WC.

## Non-Structural Carbohydrate Composition

To avoid methodological artifacts (Quentin et al., 2015), all NSC measurements were done at the Natural Resources Institute Finland. Furthermore, to standardize our results, we focused on the ratio between starch and the total NSC content, and the ratio between the target sugar and the total soluble sugar content.

We analyzed NSC composition of evergreen conifer phloem by using a sub-sample (ca. 2 cm in length) of the segment collected for the measurements of osmolality and water content. NSC measurements were performed according to Jyske et al. (2015). Briefly, samples of inner bark were cut into matchsticksized pieces, freeze-dried for 72 h, and milled with a ballmill while kept frozen. About 20 mg of powder was weighed into glass test tubes and heated to 100◦C to deactivate the enzymes. The soluble sugars were extracted twice (at 100◦C) by using 80% ethanol to which m-erythtrit (Calbiochem, Merck KGaA, Darmstadt, Germany) was added as an internal standard. The sugar extracts were evaporated to dryness with nitrogen flow, silylated with 20% TMSI-pyridine mixture (i.e., 1 trimethylsilyl-imidazole; Sigma-Aldrich, Darmstadt, Germany), and analyzed with gas chromatography–mass spectrometry (GC-MS; Agilent Hewlett-Packard 6890 GC, equipped with a Zebron ZB-SemiVolatiles column (30 m × 0.25 mm i.d × 0.25 µm df) and Hewlett-Packard 5973 MSD, EI-MS 70 eV), in which helium was used as a carrier gas (flow 1.5 ml/min). The chromatographic conditions were as follows: initial temperature 110◦C; rate of temperature increase 10◦C min−<sup>1</sup> ; final temperature 320◦C maintained for 14 min; injector temperature 260◦C, and split ratio 1:20. The MS-interface temperature was 300◦C and ion source temperature was 230◦C. In the analysis, the compounds were identified on the basis of their mass spectra and retention times as verified by using the following authentic compounds (i.e., external standards): Dfructose (Merck, Darmstadt, Germany), myo-inositol (Merck), D-glucose (BDH AnalaR, VWR International Ltd, Poole, UK), sorbitol (Fluka, Sigma-Aldrich), sucrose (BDH AnalaR), Draffinose pentahydrate (Fluka). For pinitol, fructose was used as a standard. The results were calculated using an internal standard and the external standards.

The soluble-sugar-free samples obtained after extraction were used for starch analyses with a commercial starch assay kit (Total Starch Assay Procedure, Megazyme International, Wicklow, Ireland). Briefly, starch in residual pellets was hydrolyzed into maltodextrins by adding α-amylase (in MOPSbuffer, pH 7) and incubated for 6 min at 100.5◦C. Next, the samples were suspended in acetate buffer (pH 4.5) and amyloglugosidase was added to hydrolyze maltodextrins into dglucose by incubating for 30 min at 50.5◦C. The absorbance of the samples was measured colorimetrically (Shimadzu UV-2401 spectrometer at 510 nm) using glucose oxidase and peroxidase. The standard curve was made with D-glucose (BDH AnalaR).

## Phloem Area and Sample Age Measurements

The most basipetal branch segment was used for anatomical analysis. Each segment infiltrated in 50% ethanol was cut with a hand saw to have a 5–8 mm thick disk. Disks were then dehydrated with immersions in ascending ethanol concentrations until absolute ethanol, infiltrated with liquid paraffin, and embedded into paraffin blocks (Anderson and Bancroft, 2002). The blocks were trimmed and moistened with cold water for at least 2 h to soften the woody tissue and then cut with a rotary microtome (RM2245, Leica). Sections (10–15 µm in thickness) were then stained with a solution of safranine and Astra blue (1 and 0.5% in distilled water, respectively), dehydrated with alcohol (50 and 96%), rinsed with xylol and permanently fixed by mounting a cover glass with Eukitt (Bioptica, Milan, Italy). Digital images were captured at 40× magnifications with a camera mounted on a light microscope (Eclipse80i, Nikon) to cover the whole cross-sectional area and then stitched with PTGui v8.3.10 (New House Internet Services B.V., Rotterdam, The Netherlands). Stitched images were analyzed with ROXAS v2.1 (von Arx and Dietz, 2005; von Arx and Carrer, 2014) along a wedge of known angle centered at the pith to identify tree-ring boundaries and determine branch age. Proxy for the growth rate (cm year−<sup>1</sup> ) of 70 cm-long branches could be calculated from the branch age. In addition, the non-collapsed phloem area was determined from the wedge and upscaled to the total cross-section (Zhang et al., 2015). Collapsed phloem was identified as the phloem older than 1 year, characterized by bigger and stretched cells.

#### Statistical Analysis

We first analyzed the effect of water content (WC) on osmolality. A two-level mixed-effect model for explaining osmolality was created. The fixed term of the model included the explanatory variables 1/WC, species and their interaction. In addition, the model had random intercepts for levels describing the nested structure of the data: regions, and sites within regions. Random intercepts followed normal distribution. These models, and mixed-effect models described below, were fitted with the function lme of the R package nlme (Pinheiro et al., 2013). All statistical analyses were performed with R version 3.2.2 (R Core Team, 2013).

Solute content—osmolality regression was fitted but the significance of the fit was not analyzed because the osmolality was used in the calculation of the solute content, thus creating dependency of response and explanatory variables. Nonetheless fitted curves are given to guide the reader's eye. Curves were estimated with the function nls of the R package stats (Bates and Chambers, 1992) assuming a relationship y = a + x <sup>∧</sup>b between ordinate (y) and abscissa (x). Parameters a and b were fitted by species.

Secondly, we compared the differences of osmolality, n, WC, and NSC composition between species and regions. Therefore, the fixed term of the mixed-effect model included species and region, and several covariates [sample age, non-collapsed phloem area, tree height, site moisture status (moist/dry)]. The random term included sites, and, in the case of osmolality, observations within tree as we had two repetitions per tree. First, we performed the selection of covariates using AIC criterion and step AICfunction of the R package MASS (Venables and Ripley, 2002) when the site moisture status, species, and region were always in the model. Second, also the site moisture status and species were removed from the model if they did not improve AIC. ANOVA results of the model in the Result section are shown for the marginal effects, i.e., for the effects, when all other variables are already in the model. Pairwise differences between regions and species were tested with Tukey's range test (R function glht—R package multcomp; Hothorn et al., 2008), except for NSC-related variables, where ANOVA results were used to test whether the conifers significantly differ from each other.

## RESULTS

Phloem osmolality decreased with increasing water content per tissue dry mass in all studied species (**Table 2**, **Figure 2A**). In addition to tissue water content, phloem osmolality increased with increasing tissue solute content (calculated from osmolality and water content measurements) in Populus and Betula, but such a trend was either weaker (Picea) or absent (Pinus) for both evergreen conifers (**Figure 2B**).

Phloem osmolality (mol kg−<sup>1</sup> ) varied across regions between 0.38–0.60 (Pinus), 0.44–0.69 (Picea), 0.53–0.69 (Populus), and 0.49–0.64 (Betula). Phloem osmolality was on average 15% lower in Pinus than in the other species (**Table 3**, **Figure 3A**). Among the studied regions, osmolality was lowest at midlatitude (Czech Republic), increased toward the north and the south with the highest average values being measured in Southern Finland and Italy, respectively, and then decreased again in Northern Finland and Portugal (**Table 3**, **Figure 3B**). The difference in the average osmolality between the Czech Republic and Southern Finland was 37%, and between the Czech Republic and Italy 38%. There was no significant difference in phloem osmolality between dry and moist soil sites within the regions. Tree height, sample age or, the area of non-collapsed phloem were not related to phloem osmolality. The latitudinal trends were visible in all species (Figure S2 in the Supplementary Material).

Water content per dry mass was on average 91% higher in Pinus and 57% higher in Picea in comparison to the deciduous angiosperm species (**Figure 3A**). The highest tissue water content was measured at mid-latitudes, from where it decreased on average by 21% toward northern latitudes and 28% toward southern latitudes (**Figure 3B**). Tissue water content variability was high at the intermediate latitudes (**Figure 3B**). In addition to species and region, increasing non-collapsed phloem area increased water content per tissue dry mass indicating that non-collapsed phloem contains more water in comparison to collapsed phloem (**Table 3**). Water content per tissue dry mass

TABLE 2 | Mixed-effect model result for testing the effect of species, water content (WC) and their interaction on phloem osmolality.


Betula pendula is used as reference for the model estimates for the class variable species. Sample size is 208.

\*P < 0.05, \*\*\*P < 0.001.

was higher in younger samples in comparison to older samples, and higher in shorter trees in comparison to taller trees (**Table 3**). Phloem water content was slightly higher in moist soil sites

TABLE 3 | Mixed-effect model results for testing the influence of species and region on osmolality, water content (WC), and solute content (n).


Potential covariates in the model were site moisture status, tree height, sample age and non-collapsed phloem area; covariates and their order in the final model were selected with AIC. Dry site, Betula pendula and Portugal (40◦N) are used as references for the model estimates for the class variables site, species and region, respectively, in the model output. N is sample size.

\*P < 0.05, \*\*P < 0.01, \*\*\*P < 0.001.

Species-specific model fits are drawn in a based on a mixed-effect model (Table 2). In (B), power fits and 95% confidence intervals are drawn for each species based on the raw data to guide the eye although statistical tests are not justified (n is not independent from osmolality).

in comparison to dry soil sites, but this difference was not statistically significant (**Table 3**).

Solute content per phloem dry mass was on average 51% higher in the studied evergreen conifers in comparison to the deciduous angiosperm species (**Figure 3A**). Solute content showed only few statistically significant differences between the regions (**Table 3**, **Figure 3B**), and a decreasing trend with increasing sample age (**Table 3**).

The ratio of starch to total NSC increased approximately 60% from Portugal to the Finnish regions (**Table 4**, **Figure 4A**). The ratio of disaccharide sucrose to total soluble sugars was the lowest at mid-latitudes in Switzerland and the Czech Republic (**Table 4**, **Figure 4B**), whereas the ratio of monosaccharides glucose and fructose (i.e., hexoses) to total soluble sugars was the highest in these regions (**Table 4**, **Figure 4C**). Raffinose content was negligible at mid-latitudes in Czech Republic, and increased with both increasing and decreasing latitudes (**Table 4**, **Figure 4D**). The ratio of raffinose to all soluble sugars was the only soluble sugar that had significantly different values in the two studied evergreen conifers: the share of raffinose was 10% higher in Pinus in comparison to Picea (**Table 4**). In contrast, the share of pinitol to total soluble sugars showed only a few statistically significant differences across regions and there was no difference between the two conifers (**Table 4**, **Figure 4E**). No significant differences were observed in any soluble sugars between moist and dry soil sites (**Table 4**). Similar latitudinal trends were visible in NSC composition in absolute concentrations (Figure S3 in the Supplementary Material).

In addition to region, the ratios of starch to total NSC, and the ratios of sucrose and raffinose to all soluble sugars were positively linked to tree height (**Table 4**). The ratio of starch to total NSC increased with increasing sample age (**Table 4**). Similarly, sample age affected the share of sucrose positively, as did the area of non-collapsed phloem (**Table 4**). Share of hexoses, on contrary, decreased with increasing tree height, and was the lower the higher the area of non-collapsed phloem (**Table 4**). The share of pinitol decreased with increasing sample age (**Table 4**).

Branch growth rate was highest at mid-latitudes and decreased toward north and south (**Table 1**). Also needle length in the studied conifers showed similar trend (Figure S1 in the Supplementary Material).

## DISCUSSION

### Latitudinal Trends and Species Differences in Osmolality

The results showed that the major determinant of observed variation in phloem osmolality across Europe was tissue water content instead of solute content, in contrary to what we expected. Solute content played a role in explaining the variation in phloem osmolality for both deciduous angiosperm species (Betula pendula and Populus tremula), but its effect was weak for the two evergreen conifers (Pinus sylvestris and Picea abies). We thus found support for active osmoregulation, by adjusting sugar contents, for deciduous angiosperms, but not for evergreen conifers. Phloem transport, conversion of NSC from one form to another, or unloading of sugars with the xylem may contribute to such osmoregulation.

The study confirmed our hypothesis that phloem osmolality increases from mid-latitudes toward the extreme ends of the latitudinal gradient following decreasing branch growth rate (see **Table 1**). A higher osmolality of phloem sap was expected in the driest conditions as it contributes to maintaining turgor when tree water potential is low, and in cold conditions because it decreases the freezing point of living tissue (Charrier et al., 2013a) and maintains sufficient metabolism as the metabolic

efficiency decreases at lower temperatures (see e.g., Piper et al., 2006). Moreover, high osmolality may enable refilling of xylem conducting elements embolised during freezing and thawing, as has been shown forJuglans regia (Charrier et al., 2013b) and Picea abies in alpine timberline (Mayr et al., 2014).

Local soil properties and/or topography had no effect on phloem osmolality, implying that climate rather than soil water supply affected phloem osmolality and its components. The sites where selected subjectively, but the contrast between two moisture statuses was strong enough to induce differences in needle length of Pinus and Picea (see Figure S1 in the Supplementary Material). Although site moisture status did not have a direct effect on phloem osmolality or its components, it played a role via sample age and non-collapsed phloem area as these variables varied between moist and dry soil sites. These sources of variation were controlled in the analyses with statistical tests.

The latitudinal trends observed in phloem osmolality followed the latitudinal trends observed in the level of tissue water content, whereas the level of phloem solute content was surprisingly similar across Europe. Low phloem water content per tissue dry mass in the extreme ends of the latitudinal gradient can

either be caused by differences in weather conditions during sampling, phloem tissue anatomy, or elastic tissue adjustment. However, the weather cannot explain the whole water content variability, e.g., the low phloem water content observed in the two most northern regions were measured on rainy days. Phloem anatomy is one potential explanation although the literature is scarce concerning the differences in phloem anatomy between different climates (see Gricar et al., 2015 ˇ ); in xylem it is known that density typically increases with decreasing growth rate (e.g., Raiskila et al., 2006), and growth rate was indeed lower in the most extreme ends of the latitudinal gradient in comparison to the middle latitudes (see **Table 1**). Another potential explanation is latitudinal differences in tissue elasticity. It is known that elastic shrinkage and extraction of water is characteristic for phloem tissue both during cold (Zweifel and Häsler, 2000; Améglio et al., 2001; Lintunen et al., 2015) and drought (Zweifel et al., 2001; Steppe et al., 2006; Mencuccini et al., 2013) stress. Similarly, Gross and Koch (1991) and Callister et al. (2008) studied seasonal changes of leaf osmotic potential at full turgor in Picea abies and three Eucalyptus species, respectively, and concluded that the observed increase in leaf osmotic concentration in winter was mainly caused by decreased tissue water content (due to increased tissue elasticity) instead of active accumulation of solutes. On the other hand, previous studies on leaf osmotic potential under drought stress suggests that the turgor loss point in leaves is dictated by solute content via osmoregulation rather than elastic adjustment with tissue water content (Bartlett et al., 2014; Delzon, 2015; Maréchaux et al., 2015).

## Latitudinal Trends and Species Differences in NSC Composition

Although phloem solute content was rather stable between regions along the latitudinal gradient, the composition of the non-structural carbohydrate concentration (NSC) differed between regions. The proportion of starch to total NSC was 46% higher in the two regions measured in Finland in comparison to other regions. These results are in accordance with the study of Hoch and Körner (2012) on late-season NSC concentration in various tree species in the tree line ecotones, where they reported increasing NSC concentration in branch wood with elevation due to increased starch content. The results of our study and the study of Hoch and Körner (2012) imply that the TABLE 4 | Mixed-effect model results for testing the influence of species and region on the ratio of starch to non-structural carbohydrates (NSC), ratio of sucrose, hexoses (i.e., glucose + fructose), raffinose and pinitol to total soluble sugars.


TABLE 4 | Continued


Potential covariates in the model were site moisture status, tree height, sample age and non-collapsed phloem area; covariates and their order in the final model were selected with AIC. Dry site, Picea abies and Portugal (40◦N) are used as references for the model estimates for the class variables site, species and region, respectively, in the model output. N is sample size.

\*P < 0.05, \*\*P < 0.01, \*\*\*P < 0.001.

high proportion of starch reflects the build-up of osmotically inactive C reservoir to balance between periods of low C supply, and to buffer during the events of abiotic stress (Sala et al., 2012; Hartmann, 2015). Starch is typically converted into soluble sugars as a result of decreasing temperature (Améglio et al., 2004), so high starch ratio measured in the north at the end of the growing season may lead to increased soluble sugar concentrations and phloem osmolality for winter. This conclusion is in accordance with earlier studies showing that starch concentration in xylem of Picea abies (Hou, 1985) and cambium of Picea mariana (Deslauriers et al., 2014) increase in autumn, whereas Jyske et al. (2015) found contradictory results for the inner bark of Picea abies where the ratio of starch to total NSC decreased from the mid-summer toward autumn. Especially Pinus species are known to store also lipids in their woody parts in high elevations/latitudes (Hoch et al., 2002; Hoch and Körner, 2003) which also contribute to C storage dynamics in case when C fixation is extremely constrained (Hoch and Körner, 2003).

Sucrose comprised only a minor share of total soluble sugars in the middle latitudes of Switzerland and the Czech Republic, whereas hexoses represented higher proportion in comparison to other regions. Sucrose is a non-reducing sugar, it is chemically stable, and has higher molar mass in comparison to other sugars such as hexoses glucose and fructose. The high hexose to sucrose ratio in the middle latitudes might be linked to mobilization of starch after an active growing period (Witt and Sauter, 1994; Deslauriers et al., 2014) or to high respiratory losses (Strimbeck et al., 2008). Similarly, Deslauriers et al. (2014) measured sucrose levels close to zero from Picea mariana xylem and cambium during July in Quebec, and suggested that growth activities might be the cause. On the other hand, high sucrose levels have been empirically connected to high cold tolerance in Pinus sylvestris and Picea abies needles (Aronsson et al., 1976; Strimbeck et al., 2008) and in Picea abies buds (Lipavská et al., 2000). Sucrose is known to be able to retain the liquid-crystalline state of membranes under osmotic stress caused by cold, drought, and salinity (see Lipavská et al., 2000). Also, it cannot be totally excluded that the seasonal timing of sampling was not fully synchronized between regions (i.e., trees might have been in different phenological phases between regions regarding e.g., the end of secondary growth, leaf senescence, cold acclimation, etc.). In general, sucrose is considered as the most important compound being translocated in phloem elements (Pate, 1976) and it is a preferable phloem transport sugar (Lang, 1978; Aoki et al., 2012; De Schepper et al., 2013), whereas hexoses are used more for storage and are present in high amounts in all living cells.

Our hypothesis that the stress-related soluble sugars raffinose and pinitol have high share of total soluble sugar content in the most drought and cold stressed environments was confirmed for raffinose, but not for pinitol. The proportion of raffinose from all soluble sugars was higher in the most southern and northern latitudes compared to the middle latitudes as we hypothesized. Accordingly, the role of raffinose in plant cell protection during environmental stress, such as drought and low temperatures is supported by several studies (e.g., Zuther et al., 2004; Nishizawa-Yokoi et al., 2008; dos Santos et al., 2011; Deslauriers et al., 2014), but is not yet demonstrated for the secondary phloem at continental scale. Deslauriers et al. (2014) reported raffinose concentration to increase in response to drought in the cambium and xylem of Picea mariana and concluded the osmoregulatory response to be directly dependent on raffinose. Jyske et al. (2015) reported increases in raffinose content of Picea abies phloem when approaching dormant season in the south and north of Finland. Accordingly, Simard et al. (2013) found raffinose concentration to increase during winter acclimation in the cambium of Picea abies and Larix decidua and Hoch et al. (2002) in the wood of Pinus cembra. The important role of raffinose among the soluble sugars in stressful environments is that it acts as osmoprotectant and antioxidant: raffinose protects cellular structures and sustains osmotic balance in plants (dos Santos et al., 2011), inhibits the oxidation of other molecules thus protecting plant cells from oxidative damage and maintaining redox homeostasis (Nishizawa-Yokoi et al., 2008), and inhibits the tendency of sucrose to crystallize and hence to lose its protective effect under stress conditions (Caffrey et al., 1988; Lipavská et al., 2000).

Surprisingly, the proportion of pinitol from all soluble sugars was not higher in the most northern and southern regions in comparison to other regions as hypothesized. This was against our hypothesis and is contradicting previous studies on several herbaceous species (e.g., Guoa and Oosterhuis, 1997) and tree foliage (e.g., Griffin et al., 2004), which described pinitol as sugar that increase tolerance to stress by drought, salinity or low temperature (Orthen et al., 1994). However, pinitol content in wood is higher during active growth than during cold (Hoch et al., 2002; Simard et al., 2013) or drought periods (Deslauriers et al., 2014), which implies that pinitol dynamics in the woody parts of trees differ from such dynamics in tree foliage or herbaceous species. In general, raffinose and pinitol were only present in small absolute amounts in the phloem, in agreement with other studies (e.g., Simard et al., 2013).

The ratio of raffinose to total soluble sugars was higher in Pinus than Picea. Any other soluble sugar did not show differences between the studied conifers. Raffinose concentrations have been measured previously in the inner bark of Pinus sylvestris (e.g., Antonova and Stasova, 2006) and Picea abies (Jyske et al., 2015), but there does not seem to be earlier studies allowing to compare the concentrations between the two species growing in the same region (see Quentin et al., 2015). Picea abies and Pinus sylvestris are both cold-tolerant tree species, but Pinus is better adapted to dry growth conditions. In general, both tissue water content and solute content expressed per tissue dry mass are higher in the studied evergreen conifers than in the deciduous angiosperm species, which might indicate higher phloem tissue density in the angiosperm species.

## Effect of Branch Age, Tree Height and Non-Conducting Phloem Area

Our sampling was done at a fixed distance of 70 cm from the branch tip, and thus the data represents a large variation in the age of the sample and area of non-collapsed phloem. There was a clear gradient in sample age from approximately 15 years in the cold Northern Finland and dry Portugal to approximately 3 years in the Czech Republic with good growing conditions. In the phloem of older samples, there was less water and solutes per phloem tissue dry mass. Thus, the lack of age-dependency of phloem osmolality could likely be explained by these two effects canceling each other out. Rosner et al. (2001) similarly found that differences in phloem water content parameters observed in Picea abies were explained by age-dependent structural changes in secondary phloem.

Also the ratios of sucrose and pinitol in the secondary phloem were dependent on branch age: the ratio of sucrose to total soluble sugars increased with branch age, whereas the ratio of pinitol decreased. Similarly Ericsson (1979) has shown that the youngest needles in Pinus sylvestris have more pinitol in comparison to older needles. The high levels found in the developing current-year needles as well as in the young phloem indicate that pinitol may also be involved in the synthesis of new cell components (Ericsson, 1979).

Tissue water content decreased with increasing tree height. However, the nature of this relationship is not easy to interpret as while tree height varied across regions and sites, height of branch cut was more homogeneous. Xylem conduits widen from stem/branch apex basally at nearly fixed rate (Anfodillo et al., 2013), therefore samples collected at fixed distance from the branch tip were virtually characterized by similar path-length hydraulic resistance (Petit and Anfodillo, 2009) and similar drop in water potential, if leaf transpiration is assumed to be comparable.

Increasing non-collapsed phloem area increased phloem water content indicating that conductive phloem contains more water in comparison to collapsed phloem. This result is in line with the study of Rosner et al. (2001) showing that phloem water content in Picea abies was clearly higher in non-collapsed, conducting phloem than in collapsed, non-conducting phloem. Similarly in Quercus robur, magnetic resonance imaging showed that phloem water content was the highest in the conductive phloem cells (De Schepper et al., 2012). This can be attributed to the degeneration of sieve cells and Strasburger cells (Rosner et al., 2001; De Schepper et al., 2012). Also the share of sucrose and raffinose to total soluble sugars increased with increasing non-collapsed phloem, whereas the share of hexoses decreased.

### CONCLUSION

Our study was the first one presenting patterns in osmolality and NSC composition in branch phloem at continental scale. It shows that osmolality increases with stronger drought or cold stress, and that both evergreen conifers and deciduous angiosperms adjust their phloem osmolality mainly by using water while solute content is surprisingly constant across large climate gradient. This suggests that passive, elastic adjustment with phloem water content occurs in stem and branches rather than osmoregulation. This is reasonable, as phloem is close to water potential equilibrium with the xylem (Thompson and Holbrook, 2003), which is achieved by exchanging water between these two tissues. Moreover, the composition of soluble sugars indicate that controlling raffinose, but not pinitol, allows trees to resist colder and drier conditions. The starch levels show that boreal cold stressed trees store more starch to survive the long cold winter. Overall, climate together with branch age and non-conducting phloem area rather than local soil water supply affected phloem osmolality and its components.

## AUTHOR CONTRIBUTIONS

Data from different countries was collected in a framework of topic groups 1 and 6 (COST STReESS, FP1106). FS coordinated the experimental design of plots across Europe. TP and SR made the osmolality measurements in the laboratory, TJ made the NSC measurements, GP had the main responsibility of the anatomy measurements, and GV and RLP made the needle length analysis presented in the supplementary material. MP and AL were responsible for the statistical analysis. AL and TH formulated

## REFERENCES


the study questions with the help of the other authors. AL wrote most of the manuscript with the aid of TH, but all authors contributed by discussing and reviewing drafts and the final version.

## FUNDING

This article is based upon work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology). Financial support from the Academy of Finland was received by AL, TP, and TH (268342, 272041), and YS (1284701). TJ was supported by the Japan Society for the Promotion of Science (JSPS KAKENHI no. 26•04395) and MP by EU Life Programme (LIFE12 ENV/FI/000409). GV was supported by a grant from the Swiss State Secretariat for Education, Research and Innovation SERI (SBFI C14.0104). RG and JU were supported by Internal Grand Agency of Mendel University (IGA MENDELU 73/2013) and by the Ministry of Education, Sports and Youth of the Czech Republic (COST LD13017). EMRR is funded by the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 659191 and by the Research Foundation—Flanders (FWO, Flanders, Belgium).

#### ACKNOWLEDGMENTS

Dirk Schmatz provided help with extraction of climatic data from the Swiss Federal Research Institute WSL's climate database.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00726

photoperiods and thermoperiods. Physiol. Plant. 36, 127–132. doi: 10.1111/j.1399-3054.1976.tb03922.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Lintunen, Paljakka, Jyske, Peltoniemi, Sterck, von Arx, Cochard, Copini, Caldeira, Delzon, Gebauer, Grönlund, Kiorapostolou, Lechthaler, Lobo-do-Vale, Peters, Petit, Prendin, Salmon, Steppe, Urban, Roig Juan, Robert and Hölttä. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Plasticity in Vulnerability to Cavitation of Pinus canariensis Occurs Only at the Driest End of an Aridity Gradient

#### Rosana López<sup>1</sup> \* † , Francisco J. Cano<sup>1</sup>† , Brendan Choat<sup>2</sup> , Hervé Cochard<sup>3</sup> and Luis Gil<sup>1</sup>

#### Edited by:

Andreas Bolte, Johann Heinrich von Thünen-Institute, Germany

#### Reviewed by:

Louis S. Santiago, University of California, Riverside, USA Zhenzhu Xu, Institute of Botany – Chinese Academy of Sciences, China

> \*Correspondence: Rosana López rosana.lopez@upm.es

#### †Present address:

Rosana López and Francisco J. Cano, Hawkesbury Institute for the Environment, University of Western Sydney, Richmond, 2753 NSW, Australia

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 30 January 2016 Accepted: 17 May 2016 Published: 03 June 2016

#### Citation:

López R, Cano FJ, Choat B, Cochard H and Gil L (2016) Plasticity in Vulnerability to Cavitation of Pinus canariensis Occurs Only at the Driest End of an Aridity Gradient. Front. Plant Sci. 7:769. doi: 10.3389/fpls.2016.00769 <sup>1</sup> Forest Genetics and Physiology Research Group, Sistemas y Recursos Naturales, School of Forest Engineering, Technical University of Madrid, Madrid, Spain, <sup>2</sup> Hawkesbury Institute for the Environment, University of Western Sydney, Richmond, NSW, Australia, <sup>3</sup> PIAF, INRA, Université Clermont Auvergne, Clermont-Ferrand, France

Water availability has been considered one of the crucial drivers of species distribution. However, the increasing of temperatures and more frequent water shortages could overcome the ability of long-lived species to cope with rapidly changing conditions. Growth and survival of natural populations adapted to a given site, transferred and tested in other environments as part of provenance trials, can be interpreted as a simulation of ambient changes at the original location. We compare the intraspecific variation and the relative contribution of plasticity to adaptation of key functional traits related to drought resistance: vulnerability to cavitation, efficiency of the xylem to conduct water and biomass allocation. We use six populations of Canary Island pine growing in three provenance trials (wet, dry, and xeric). We found that the variability for hydraulic traits was largely due to phenotypic plasticity, whereas, genetic variation was limited and almost restricted to hydraulic safety traits and survival. Trees responded to an increase in climate dryness by lowering growth, and increasing leaf-specific hydraulic conductivity by means of increasing the Huber value. Vulnerability to cavitation only showed a plastic response in the driest provenance trial located in the ecological limit of the species. This trait was more tightly correlated with annual precipitation, drought length, and temperature oscillation at the origin of the populations than hydraulic efficiency or the Huber value. Vulnerability to cavitation was directly related to survival in the dry and the xeric provenance trials, illustrating its importance in determining drought resistance. In a new climatic scenario where more frequent and intense droughts are predicted, the magnitude of extreme events together with the fact that plasticity of cavitation resistance is only shown in the very dry limit of the species could hamper the capacity to adapt and buffer against environmental changes of some populations growing in dry locations.

Keywords: vulnerability to cavitation, hydraulic conductivity, drought, phenotypic plasticity, genetic differentiation, provenance trials, Pinus canariensis

## INTRODUCTION

fpls-07-00769 June 1, 2016 Time: 13:8 # 2

Water availability is a crucial driver of species distribution (Ramírez-Valiente et al., 2009). However, the increasing temperatures and more frequent water shortages associated with global climate change could overcome the ability of long-lived species to cope with rapidly changing conditions (Hoffmann and Sgrò, 2011; Choat et al., 2012). Ecosystem responses to these new climatic scenarios will include the interrelated processes of evolutionary change, shifts in geographic range and extinction of some populations (Nicotra et al., 2010). The relative role of each process is far from clear and will depend on how species and populations acclimate their structure and function (i.e., phenotypic plasticity) or adapt through natural selection. Another factor that could increase selective pressures in longlived species is the decrease in gene flow due to severe habitat fragmentation (López de Heredia et al., 2010). This may lead to decreases genetic diversity, limitations on the ecological benefits of plasticity, and decoupling of climate and local adaptation (Jump and Peñuelas, 2005). Taken together these factors would result in increased vulnerability to extreme climatic events and to a higher risk of mortality of trees. This is particularly important for populations in the southernmost locations in the northern hemisphere which may become extinct if they are not able to adapt or migrate. Thus, studies that quantify the ability of species to maintain their fitness and plasticity of key traits should be a priority in management and conservation programs.

Survival and growth of natural populations adapted to a given location, transferred and tested in other environments as part of provenance trials or common garden tests, can be interpreted as a simulation of ambient changes at the original location and are also valuable tools to separate the genetic component of adaptation from phenotypic plasticity. Although, provenance trials have been intensively established during the last 250 years, they have been mainly used for productive purposes avoiding sites with low fertility or at the ecological limits of the species (Mátyás, 2002). Few studies have been implemented under conditions of severe change where populations are reaching their tolerance limits. In such marginal situations, the effectiveness of adjustment through natural selection is limited and mass mortality may occur. In fact, in recent years an overall increased global frequency in reported drought-related mortality events (Allen et al., 2010) has evidenced the increasing vulnerability to forest dieback.

Hydraulic failure due to xylem embolism is broadly accepted as a key factor of drought-induced mortality, whether directly rupturing the water continuum from soil to leaves or via carbon starvation due to prolonged stomatal closure (Sala et al., 2010). The importance of changes in the plant conducting system in response to drought has been highlighted in order to maximize water uptake and reduce hydraulic failure (Sperry et al., 2002; Ladjal et al., 2005; Brodribb and Cochard, 2009). Hydraulic traits thus play a crucial role in adaptation and can be used to predict the future resilience of forested ecosystems.

Plants differ widely in their vulnerability to drought-induced cavitation and the responses to drought are species-specific and depend on a tree's hydraulic strategy (Bréda et al., 2006). Overall, conifers are more resistant to cavitation than angiosperms but less efficient in water transport (Maherali et al., 2004). The xylem of conifers, cheap to maintain but less efficient than broadleaved xylem, may confer a competitive advantage in low resource environments where photosynthesis is limited and water availability scarce (Hacke and Sperry, 2001). Despite the homogeneity of xylem structure, where tracheids make up almost 90% and the remainder is axial and ray parenchyma along with resin ducts in certain species (Plomion et al., 2001), striking interspecific variation in cavitation resistance has been reported (Piñol and Sala, 2000; Maherali et al., 2004; Martínez-Vilalta et al., 2004; Brodribb and Cochard, 2009; Delzon et al., 2010; Bouche et al., 2014). However, information about variation within species, and to what extent genotypes exhibit plasticity in hydraulic traits, remains scarce and very few studies have reported quantitative relationships between survival or growth and resistance to xylem embolism (Tyree et al., 2002; Brodribb and Cochard, 2009; Barigah et al., 2013; Urli et al., 2013). The study of such phenotypic variation is critical for both the development of a general understanding, and predicting plant responses to climate change. In this context, provenance trials are valuable tools to separate the effects of genetics and acclimation on phenotypic variation.

Pinus canariensis is endemic to the Canary Islands. Despite this restricted distribution area, volcanic destruction and successive fragmentation of populations, erosional activities and the influence of the humid Trade winds and the dry Saharan winds have created extremely diverse habitats that may exert varying selective pressures (Emerson, 2002). Current environmental conditions are very different from those in which this species evolved under a much wetter climate even during the late Holocene (de Nascimento et al., 2009). This species is therefore a good model to compare the intraspecific variation and the relative contribution of plasticity to adaptation of key functional traits related to drought resistance: vulnerability to cavitation, efficiency of the xylem to conduct water and biomass allocation. We use six populations of Canary Island pine growing in three provenance trials with contrasted climatic conditions: a wet location influenced by the Trade winds, a dry site in the leeward slopes of the Teide volcano and the most xeric site, in the very edge of the distribution of the species.

## MATERIALS AND METHODS

### Provenance Trial Experiments

We selected trees from six populations of P. canariensis growing in three provenance trials in the Canary Islands with contrasted climatic conditions. The most humid provenance trial (wet site onward) is influenced by the Trade winds which can even double the annual precipitation due to fog (795 mm MAP; **Figure 1**). This wet site and the dry site (460 mm MAP) have similar soils and temperatures but differ sharply in water availability and drought period (**Table 1**). Finally, the xeric site, located in the dryer limit of distribution of the species, combines an arid environment (320 mm MAP and the periodic gusts of the extreme dry Saharan wind) with very compact and stony soil.

The six populations included in this study, out of 21 growing in the provenance trials (more details about the establishment and populations included in these provenance trials can be found in López et al., 2007) were selected to cover both the range of the climatic envelope of the species and populations with different ages based on the chronostratigraphy of the substrate where they occur (more details in López de Heredia et al., 2014).

(numbered) described in Climent et al. (2004, 2006). Note the influence of the humid Trade Winds.

Survival, height and basal diameter were measured during the first 6 years after the establishment of the provenance trials. After this period survival rates were stable.

## Vulnerability to Cavitation

To evaluate the phenotypic plasticity of vulnerability to cavitation and hydraulic efficiency in this species we have used data from the wet and the xeric provenance trials already published in López et al. (2013). We have completed these data with unpublished measurements of the dry provenance trial. We describe in short the plant material and the methods used.

One branch exposed to the sun, longer than 40 cm and with a maximum diameter of 1 cm was sampled from 8 to 14 trees per population in each provenance trial. In the wet and xeric site,


TABLE 1 | Climate characterization of the six populations of Pinus canariensis and the three provenance trials included in this study.

MAP: annual precipitation, T: mean annual temperature, Tr: annual temperature range, Dp: drought period; ETo: Evapotranspiration calculated with Penman-Monteith equation (sp: spring; sum: summer; aut: autumn; win: winter).

we sampled branches in 2010, when trees were 11 years-old and sampled branches corresponded to the previous year growth unit in the wet site and to the last 2–3 years units in the wet site. Needles were removed and branches were wrapped in a black plastic bag with moist paper towels to prevent dehydration. In the dry site, branches were collected 2 years later following the same procedure.

Vulnerability curves of branches from the wet and the xeric sites were constructed with the Cavitron technique (Cochard et al., 2005). For a detailed description of the methods see López et al. (2013). The vulnerability curves of branches from the dry site were constructed with the standard centrifuge method. Both, the Cavitron and the static centrifuge have shown similar results in conifer species (Li et al., 2008). Branches were trimmed under water and then both ends were shaved to a final length of 14 cm. The initial hydraulic conductance (ki, mol s−<sup>1</sup> MPa−<sup>1</sup> ) was measured using a XYL'EM device (Xylem Embolism Meter, Bronkhorst, Montigny les Cormeilles, France) at low pressure (4– 5 × 10−<sup>3</sup> MPa), perfusing the samples with the same solution described above. After measuring ki, branches were spun in a centrifuge for 5 min at increasing pressure steps to achieve negative xylem pressures (Alder et al., 1997). After each step, kh was measured with the XYL'EM.

The percentage loss of hydraulic conductivity (PLC) was estimated by the step-by-step decrease of kh with regard to ki as: PLC = 100 × (1 – kh/ki). The observed curve was fitted to a logistic function (Pammenter and Vander Willigen, 1998):

$$\text{PLC(\%)} = 100/[1 + \exp(s/25(\text{P} - \text{P}\_{50}))]$$

where P<sup>50</sup> represents the value of 9 at which 50% of hydraulic conductivity is lost and s is the slope of the VC at P50. Finally, estimates of xylem water potentials at the beginning of xylem embolism (P12) and full embolism (P88) were calculated following Domec and Gartner (2001):

$$\begin{array}{rcl} \text{P}\_{12} &=& \text{P}\_{50} + 50/s \\ \text{P}\_{88} &=& \text{P}\_{50} - 50/s. \end{array}$$

#### Hydraulic Conductivity

Xylem specific hydraulic conductivity (Ks) was assessed dividing ki by the sapwood area in the middle of the branch (As) and multiplying by sample length. Leaf specific hydraulic conductivity (Kl) was calculated as the conductivity per unit of projected leaf area (Al). All leaves distal to the branch section used for constructing the vulnerability curve were collected. The projected area of 12 leaves per sample was obtained with a scanner and analyzed with WinFOLIA (Regent Instruments, Inc., Canada). Then they were dried at 60◦C for 3 days to determine leaf dry mass and leaf mass per area (LMA). The rest of the leaves were dried as previously described and total leaf area was calculated dividing the total needle mass by LMA. The Huber value (HV), i.e., the ratio of the sapwood cross sectional (As) to the distal leaf area supported (Al), was also calculated.

#### Statistical Analysis

Differences among populations and between the three provenance trials for growth, biomass allocation, vulnerability to cavitation, and hydraulic conductivity were assessed using a general linear model (GLM) with the fixed factors provenance trial, population and their interaction. Survival was analyzed with a linear logistic model with the same factors considering a binomial distribution of the data and a logit function. The percentage of variation explained by each factor was calculated with the variance components, assuming all the factors were random. We assumed that phenotypic plasticity occurred if the effect of the environment (provenance trial) in the GLM was significant and that genotypes (populations) differed in plasticity if the interaction population × provenance trial was significant.

Correlations between traits were evaluated by calculating Pearson's coefficient on the population Best Linear Unbiased Estimator (BLUE). In addition, Spearman's correlation coefficients were determined between the climatic conditions at origin and the BLUEs of hydraulic and growth traits of each population. All analyses were performed using STATISTICA v. 7.0 (StatSoft, Inc.).

#### RESULTS

### Phenotypic Plasticity and Genetic Variation in Survival, Growth, and Hydraulic Traits

As expected, survival and growth were higher in the wet provenance trial and lower in the xeric one. In this last site we found the most striking differences between provenances: those from drier locations survived better than those from more mesic habitats. This pattern was particularly striking in the xeric site although one of the arid provenances El Hierro, exhibited higher than expected mortality (**Figure 2**). We did not find any significant difference in height or basal diameter among populations but a substantial phenotypic plasticity (**Table 2**) and almost a linear increment with MAP (**Figure 2**). Six years after planting, mean height in the wet provenance trial reached almost 2 meters, in the dry 126 cm while in the xeric average height was 80 cm (**Figure 2**).

Trees growing at the wet provenance trial had a more permeable xylem, i.e., higher Ks. This was evident in two of the six populations (Tirma and El Hierro, **Figure 2**, Supplementary Figure S1). On the contrary, the leaf area supplied by a given xylem area, Kl, was significantly higher in the xeric site due to a Huber value twice as higher and decreased almost linearly with MAP (**Figure 2**, Supplementary Figure S1). In this provenance trial, only current year leaves persist in the trees, whereas in the wet and dry sites it is common to find more than three cohorts of leaves. Most of the variability of hydraulic conductivity traits remained within populations rather than between populations or between sites (**Table 2**).

In the wet and dry provenance trials, embolism began at similar water potential, i.e., similar P12, average −1.4 MPa, but followed at different rates in each population, as shown by different slope of the vulnerability curves (Supplementary Figure S1), resulting in similar values of P<sup>50</sup> and P<sup>88</sup> in both sites for a specific population, with significant differences between

populations (**Figure 2**, Supplementary Figure S1). For instance, P<sup>50</sup> varied from −3.1 to −4.1 MPa for Esperanza and Mogan, respectively. In the xeric provenance trial, P<sup>12</sup> dropped to −3.0 MPa and P<sup>50</sup> ranged from −4.6 MPa in El Hierro to −5.9 MPa in Vilaflor. Values of water potential at full embolism, P88, varied between c. −6 MPa in the wet and dry provenance trials and almost −8 MPa in the xeric one (**Figure 2**). On average, 8% of the observed phenotypic variation in P<sup>50</sup> and P<sup>88</sup> was due to between-population differences whereas 60 and 44% were due to the provenance trial, respectively (**Table 2**). Genotype × environment (i.e., population × provenance trial) interaction was only significant for P<sup>88</sup> (7.5% of the variance), reflecting the similar plasticity of all populations for hydraulic safety traits.

It is worth mentioning that we did not flush the branches before constructing the vulnerability curves. To discard the effect of variations in native state levels of embolism on the curves, we plotted changes in Ks with increasing xylem tension (Supplementary Figure S1). Initial values of Ks for a given provenance across sites were only significantly higher in the wet provenance trial for El Hierro and Tirma and it fell more steeply than in the other two trials but never converged. Moreover, we did not find any correlation between changes in Ks and changes in P<sup>50</sup> across sites for any provenance (p > 0.1; data non-shown). For the most resistant provenances the slope was steeper between −2.5 and −5.5 MPa in the wet and the dry sites and 1 MPa lower in the xeric site whereas for the others, the steepest fall in Ks was found between −2 and −4.5 MPa in the wet and dry site and also 1 MPa lower in the xeric site (Supplementary Figure S1). These results suggested that both the vulnerability curves, expressed as PLC or as Ks, showed the same information. Finally, when the loss of conductivity was expressed as Kl, we found a remarkable difference among sites, even between the wet and the dry provenance trials for some provenances, despite the

TABLE 2 | Percentage of the explained variation and significance values ( <sup>∗</sup>p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001) due to provenance trial, population and the interaction provenance trial by population for six populations of P. canariensis growing in three provenance trials.


H: height 6 years after planting; Db: basal diameter 6 years after planting; Ks: specific hydraulic conductivity; Kl: leaf specific hydraulic conductivity; HV: Huber value; P12, P50, P88: xylem water potential at 12, 50, and 88% loss of conductivity respectively.

almost overlapping vulnerability curves expressed as PLC. These differences could have been overestimated at high xylem tensions as trees adjust their foliar area to water availability.

### Relationship between Hydraulic Properties, Survival, Growth, and Climate

Pooling data from the three sites, taller trees were related to lower HV (r = −0.45, p < 0.001) and less negative P50, P12, and P<sup>88</sup> (average r = 0.35, p < 0.001). We observed weak correlations between HV and K<sup>l</sup> with P<sup>12</sup> (r = −0.23, p < 0.01; r = −0.15, p < 0.05) and P<sup>50</sup> (r = −0.20, p < 0.01; r = −0.15, p < 0.05). However, none of these relationships were found within provenance trials (**Figure 3**).

Aridity at the origin of the populations influenced survival and vulnerability to cavitation in the dry and the xeric sites. Plants from locations with less precipitation and a longer drought period survived better (ρ = −0.94) and constructed a safer xylem (ρ = −0.77). Moreover, survival in the dry provenance trial was strongly correlated with P<sup>50</sup> (ρ = −0.83) and in the xeric with P<sup>88</sup> (ρ = −0.94). On the contrary, in the wet provenance trial the length of the drought period was not related with any parameter derived from the vulnerability curve but with Kl (ρ = −0.94). Here, survival was more related with the efficiency of water transport, Ks (ρ = 0.89) and Kl (ρ = 0.83). Finally, we found a negative trend between temperature range and P<sup>50</sup> in the three provenance trials (**Figure 3**).

#### DISCUSSION

#### Plasticity in the Plant Hydraulic System

Despite pronounced site-of-origin differences in precipitation and temperature, P. canariensis populations expressed broadly similar patterns of growth and functional plasticity in three contrasting habitats (**Figure 2**). Hydraulic traits in plants from all six populations exhibited acclimation to drier and warmer conditions with more cavitation resistant xylem, increased leaf

FIGURE 3 | Pairwise correlations between climate in the origin, hydraulic traits and survival and growth of six populations of Canary Island pine planted in three provenance trials. Only significant relationships (p < 0.05) are depicted. Dashed lines indicate negative relationships. In black overall correlations (pooling data from the three provenance trials). Blue: wet provenance trial, Green: dry provenance trial, Red: xeric provenance trial. Ks: xylem specific conductivity, Kl: leaf specific conductivity, HV: Huber value (HV), P12, P50, P<sup>88</sup> xylem potential at 12, 50, and 88% loss of conductivity, respectively.

specific conductivity and structurally increased Huber value by dramatically reduced the leaf area.

The evolution of plasticity in key functional traits may determine an organism's ability to establish (Schlichting, 1986), colonize new environments (Matesanz et al., 2012) and persist in highly variable environments or over broad niches if plasticity increases that organisms' fitness. In this sense, the importance of changes in the plant conducting system in response to drought has been highlighted in order to maximize water uptake and reduce hydraulic failure (Sperry et al., 2002; Brodribb and Cochard, 2009). Within a single species, the variance of hydraulic traits can have a major impact on population demographics and responses to changes in climate extremes (Anderegg et al., 2013). One general trend found in interspecific comparisons is that taxa growing in drier habitats tend to exhibit a safer xylem (Hacke et al., 2000; Pockman and Sperry, 2000; Maherali et al., 2004; Choat et al., 2012), although such studies may confound both genetic variation and phenotypic plasticity. The few studies quantifying intraspecific phenotypic plasticity for cavitation have reported lower values of P<sup>50</sup> when plants of a given population grow in drier environments (Corcuera et al., 2011; Wortemann et al., 2011) or after water withholding (Aranda et al., 2015), but see Lamy et al. (2014) for an exception. Constructing a more cavitation-resistant xylem allow plants to maintain higher stomatal conductance despite increasing water stress in drier or warmer habitats. Interestingly, we only observed this trend in the xeric provenance trial. The vulnerability to cavitation curves in the wet and the dry provenance trials were very similar, despite remarkable variation in environmental conditions between both sites (Supplementary Figure S1). Therefore, it appears that only

trees growing at the very edge of the distribution limit exhibit plasticity of cavitation resistance and with a shift only reported for this species, almost 2 MPa (López et al., 2013 and the present work), comparing with changes lower than 1 MPa found in beech and maritime pine (Corcuera et al., 2011; Wortemann et al., 2011; Aranda et al., 2015) or in populations of Canary Island pine from La Palma (López et al., 2013). In fact, cavitation resistance of Canary Island pines growing in the xeric provenance trial is one of the highest found in pines (Bouche et al., 2014), consistent with its pioneer behavior, its capacity to colonize bole volcanic soils, and with an extremely low capacity to retain water after eruptions.

Another key change in the hydraulic architecture entails the reduction of the water potential gradient for a given transpiration rate. Increasing Kl may assist in maintaining xylem water potentials above the level that would trigger cavitation during high evaporative demand or low soil water availability (Choat et al., 2007). In the present study, the average Kl in each provenance trial was in accordance with this trend and it increased gradually from the wet to the xeric one. The bulk of the variation in Kl was driven by variation in HV rather than changes in Ks as reflected by the differences in the curves in Supplementary Figure S1, indicating that shifts in branch sapwood:leaf area allocation influence more the response of the hydraulic capacity than changes in xylem permeability. Thus, in the dry provenance trial P. canariensis relied on higher Kl to reduce water potential gradients and lower leaf area to decrease the water use, thus saving soil water, rather than depend upon greater resistance to cavitation during summer or prolonged drought periods. Changes in the ratio of conducting to transpiration tissues is common in isohydric species such as pines and more intraspecific phenotypic variability has been found in this trait that in cavitation resistance. Xeric populations of P. ponderosa have higher HV and Kl than mesic populations (Maherali and DeLucia, 2000) but like in P. canariensis, presumably as a result of phenotypic plasticity rather than ecotypic differentiation (Maherali et al., 2002). In P. sylvestris drier populations also follow this pattern of branch carbon allocation (Martinez-Vilalta et al., 2009) whereas, P. palustris and P. halepensis decreased the HV in xeric habitats or under severe drought but allocated more carbon to root production increasing the root to leaf area ratio (Tognetti et al., 1997; Addington et al., 2006).

In a recent review focused on plasticity of cavitation resistance, Anderegg (2015) found a substantial spatial variation within species and even larger within population variability in P50. In accordance, Canary Island pine populations from the windward slopes of Tenerife and El Hierro were more vulnerable than populations from the leeward slopes of Tenerife and Gran Canaria. Although constant plasticity (i.e., similar plasticity of different populations), as shown for most traits in our study, reduces the strength of diversifying selection and can alter the impact of gene flow on local adaptation in heterogeneous environments (Crispo, 2008), volcanism and aridity could have exerted selective pressures strong enough in traits related to drought resistance as to counteract for the homogenizing effect of an extensive gene flow (López de Heredia et al., 2014). This seems to be the case of genetic variation in cavitation resistance and other drought adaptive traits at the leaf level: sclerophylly, osmotic adjustment and leaf anatomy (López et al., 2009, 2010, 2013) whereas for other such Kl, Ks or growth most of the variation resides within populations (López et al., 2013; the present study). On the other hand, some authors have suggested that stressful environments are the most likely to result in the expression of higher levels of variation (Schlichting, 2008) as differences in survival in our study among populations in the drier edge of the ecological niche for the species (**Figure 2**).

#### Climate Drivers of Cavitation Resistance

Plant hydraulic architecture has evolved to changes in climate over evolutionary timescales. For instance, dry periods drove the adaptation of cavitation resistant xylem in Cupressaceae at multiple points in the past 30 million years (Pittermann et al., 2012). Evidence gathered across species from a wide range of habitats, including conifers and evergreen angiosperms have pointed to a positive correlation between vulnerability to cavitation and water availability (Maherali et al., 2004; Jacobsen et al., 2007). Although very valuable to assess general trends, phylogeny and other historical constraints such as glaciations can interfere with the trait-climate relationships when applying to lower scales such as species of the same family or intraspecific trends. A good alternative for understanding how precipitation patterns influence xylem structure and function is to consider species that occur across wide moisture gradients. Despite its restricted distribution area, P. canariensis is a good model because it inhabits a wide range of climatic conditions and it is under strong selection pressures. Our results showed that cavitation resistance was more tightly correlated with annual precipitation, drought length, and temperature oscillation at the origin of the populations than hydraulic efficiency or the HV and also directly related to survival in the dry and the xeric provenance trials, illustrating its importance in determining drought resistance. As expected, the drier populations were the more resistant to cavitation and the mesic ones the more susceptible. This pattern is in accordance with several studies showing that populations from drier environments are less vulnerable than those from wetter environments (Alder et al., 1996; Mencuccini and Comstock, 1997; Sparks and Black, 1999; Choat et al., 2007; Aranda et al., 2015) and with less drought damage in subraces of Eucalyptus globulus originating from areas with more temperature seasonality (Dutkowski and Potts, 2012).

## Xylem Resistance to Cavitation Is an Adaptive Trait but What Is the Cost?

A growing literature over the last years has linked cavitation resistance to life-story traits, demography and crown desiccation after drought and fire (Pratt et al., 2007, 2008; Nardini et al., 2013). The water potential at incipient cavitation, P12, is linked in many species to stomatal closure (Brodribb et al., 2003; Meinzer et al., 2009; Nolf et al., 2015) and carbon assimilation (Aranda et al., 2015); P<sup>50</sup> has emerged as an appropriate trait for modeling of forest die-off on a local scale under climate changes scenarios (Nardini et al., 2013) and P<sup>88</sup> seems to be the

threshold to recover normal physiological function or resprout after severe drought episodes in angiosperms (Urli et al., 2013; Li et al., 2016). Moreover, the straight correlation in our study between survival in dry and xeric environments and P<sup>50</sup> and P<sup>88</sup> highlighted the adaptive value of cavitation resistance. However, increased cavitation resistance is often thought to come at the expense of reduced plant growth (Cochard et al., 2007). Such a tradeoff is expected if increased cavitation resistance lies in the necessity to build a denser wood, which in conifers is achieved with thicker tracheids, a feature supposed to be costly in terms of carbon allocation (Enquist et al., 1999). We found a clear tradeoff between cavitation resistance and height when considering the three provenance trials (**Figure 3**) but within each one we could not detect differences in growth between populations, thus although both traits showed a strong response to water availability they could not be interrelated.

We also failed to detect a trade-off between hydraulic efficiency and safety at the tissue level within provenance trial but when pooling data Kl and HV were negatively correlated with P<sup>12</sup> and P<sup>50</sup> (**Figure 3**). This counterintuitive correlation was mainly due to the nearly universal trend of decreasing Al with climatic dryness coupled with comparatively little changes in Ks observed in P. canariensis: a 9 and 20% decreased of Ks in the dry and the xeric provenance trial respectively regarding the wet provenance trial was offset by 82 and 139% increase in HV. Moreover, the lack of a safety-efficiency trade-off could be explained in gymnosperms because of decoupling in the anatomical traits that control it; hydraulic efficiency is related to most strongly tracheid lumen diameter whereas safety is controlled by the overlap between the torus and the pit aperture (Pittermann et al., 2010; Bouche et al., 2014).

## CONCLUSION

Vulnerability to cavitation appeared to be the key factor for survival and maintenance of a positive carbon balance for P. canariensis in xeric environments. In dry areas shifts in branch sapwood:leaf area allocation influence more the response of the hydraulic capacity than changes in xylem permeability or vulnerability to cavitation. Water limitation in the leeward slopes of the Canary Islands seems to have been a powerful agent

## REFERENCES


of natural selection promoting local adaptation in this species, despite high levels of gene flow among populations. Nevertheless, in a new climatic scenario where more frequent and intense droughts are predicted, the magnitude of extreme events together with the fact that plasticity of vulnerability to cavitation is only shown in the very dry limit of the species could hamper the capacity to adapt and buffer against environmental changes of some populations growing in dry locations.

## AUTHOR CONTRIBUTIONS

LG led the establishment of the provenance trials. RL, HC, and LG designed the experiment. RL and FC carried out the field and lab measurements. All authors contributed to interpret the results. RL drafted the manuscript. All authors read and approved the final manuscript.

## ACKNOWLEDGMENTS

We are grateful to the Canary Islands Government and the Cabildos of Tenerife and Gran Canaria for longstanding support in the study of Canary Island pine. We thank all people involved in the plantation and measurements of the provenance trials and to Christian Bodet and Pierre Conchon for their assistance with the Cavitron in France and Martin Venturas with the centrifuge in Spain. RL was supported by a González Esparcia fellowship during her stay in Clermont-Ferrand and currently holds a Marie Curie fellowship-People Program (Marie Curie Actions) of the European Union's Seventh Framework Program FP7/2007-2013/ under REA grant agreement [624473]. This work was supported by the Spanish Ministry of Science in the Project AGL2009-10606 (VULCAN) and by the National Parks Authority through the project SPIP2014-01093 (PersPiCan). The inspiring discussions during the meetings of the COST action STReESS (COST-FP1106) helped to improve the manuscript.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.00769

mortality reveals emerging climate change risks for forests. Forest Ecol. Manag. 259, 660–684. doi: 10.1016/j.foreco.2009.09.001



stomatal control of xylem tension with hydraulic capacitance. Funct. Ecol. 23, 922–930. doi: 10.1111/j.1365-2435.2009.01577.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 López, Cano, Choat, Cochard and Gil. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Novel Hydraulic Vulnerability Proxies for a Boreal Conifer Species Reveal That Opportunists May Have Lower Survival Prospects under Extreme Climatic Events

Sabine Rosner <sup>1</sup> \*, Jan Svetlík ˇ 2 , Kjell Andreassen<sup>3</sup> , Isabella Børja<sup>3</sup> , Lise Dalsgaard<sup>3</sup> , Robert Evans <sup>4</sup> , Saskia Luss <sup>1</sup> , Ole E. Tveito<sup>5</sup> and Svein Solberg<sup>3</sup>

1 Institute of Botany, BOKU Vienna, Vienna, Austria, <sup>2</sup> Centre MendelGlobe – Global Climate Change and Managed Ecosystems, Mendel University, Brno, Czech Republic, <sup>3</sup> Norwegian Institute of Bioeconomy Research, Ås, Norway, <sup>4</sup> CSIRO Materials Science and Engineering, Clayton, VIC, Australia, <sup>5</sup> Norwegian Meteorological Institute, Oslo, Norway

#### Edited by:

Ute Sass-Klaassen, Wageningen University and Research Centre, Netherlands

#### Reviewed by:

Hans W. Linderholm, University of Gothenburg, Sweden Hervé Cochard, Institut National de la Recherche Agronomique, France

> \*Correspondence: Sabine Rosner sabine.rosner@boku.ac.at

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 25 February 2016 Accepted: 26 May 2016 Published: 09 June 2016

#### Citation:

Rosner S, Svetlík J, Andreassen K, ˇ Børja I, Dalsgaard L, Evans R, Luss S, Tveito OE and Solberg S (2016) Novel Hydraulic Vulnerability Proxies for a Boreal Conifer Species Reveal That Opportunists May Have Lower Survival Prospects under Extreme Climatic Events. Front. Plant Sci. 7:831. doi: 10.3389/fpls.2016.00831 Top dieback in 40–60 years old forest stands of Norway spruce [Picea abies (L.) Karst.] in southern Norway is supposed to be associated with climatic extremes. Our intention was to learn more about the processes related to top dieback and in particular about the plasticity of possible predisposing factors. We aimed at (i) developing proxies for P<sup>50</sup> based on anatomical data assessed by SilviScan technology and (ii) testing these proxies for their plasticity regarding climate, in order to (iii) analyze annual variations of hydraulic proxies of healthy looking trees and trees with top dieback upon their impact on tree survival. At two sites we selected 10 tree pairs, i.e., one healthy looking tree and one tree with visual signs of dieback such as dry tops, needle shortening and needle yellowing (n = 40 trees). Vulnerability to cavitation (P50) of the main trunk was assessed in a selected sample set (n = 19) and we thereafter applied SilviScan technology to measure cell dimensions (lumen (b) and cell wall thickness (t)) in these specimen and in all 40 trees in tree rings formed between 1990 and 2010. In a first analysis step, we searched for anatomical proxies for P50. The set of potential proxies included hydraulic lumen diameters and wall reinforcement parameters based on mean, radial, and tangential tracheid diameters. The conduit wall reinforcement based on tangential hydraulic lumen diameters ((t/b 2 ht) ) was the best estimate for P50. It was thus possible to relate climatic extremes to the potential vulnerability of single annual rings. Trees with top dieback had significantly lower (t/bht) <sup>2</sup> and wider tangential (hydraulic) lumen diameters some years before a period of water deficit (2005–2006). Radial (hydraulic) lumen diameters showed however no significant differences between both tree groups. (t/b 2 ht) was influenced by annual climate variability; strongest correlations were found with precipitation in September of the previous growing season: high precipitation in previous September resulted in more vulnerable annual rings in the next season. The results are discussed with respect to an "opportunistic behavior" and genetic predisposition to drought sensitivity.

Keywords: climatic extremes, conduit wall reinforcement, functional wood anatomy, global warming, Norway spruce, Picea abies, top dieback

## INTRODUCTION

Managed and unmanaged boreal conifer forests provide ecosystem services, such as climate regulation including carbon fixation (Gauthier et al., 2015; McDowell et al., 2016), and their role in national and rural economy of the Nordic countries is fundamental (Schlyter et al., 2006). So far, conifer forests of the northern hemisphere reacted to warming with an acceleration in growth (Kauppi et al., 2016) and some might have retained resilience to cope with current disturbances, but projected climate change scenarios (IPCC, 2013) suggest a threat to their health (Gauthier et al., 2015; McDowell et al., 2016). A positive effect of a warmer climate is that the period of net carbon uptake will be extended in the autumn, which could increase total carbon uptake in boreal forests dominated, for instance, by Norway spruce [Picea abies (L.) Karst.; Stinziano et al., 2015]. However, it is questionable if Norway spruce, an autochthonous species of the alpine timberline (Mayr et al., 2003, 2014) and of northern regions (Solberg, 2004; Andreassen et al., 2006), can cope with extreme weather events such as frequent and prolonged summer droughts (Schlyter et al., 2006). In that respect, lower latitude northern Norway spruce forests are more endangered as increase in mortality due to summer droughts has already been reported for southern Norway (Solberg, 2004; Hentschel et al., 2014). In this study, we focus on hydraulic vulnerability traits of healthy looking and declining Norway spruce trees and on the correlation of these traits to climate (extremes). We take advantage of an existing hydraulic dataset (Rosner et al., 2016) and define novel anatomical functional traits for trunkwood based on SilviScan technology (Evans, 1994, 1999).

Many conifer species have quite high hydraulic safety margins; their P50, i.e., the water potential resulting in 50% conductivity loss, is much lower than minimum water potentials measured out in the field (Choat et al., 2012). Compared to angiosperms, conifers are supposed to have a lower capacity to reverse embolism, thus to refill conduits and restore a hydraulically functional state (McDowell et al., 2008; McDowell, 2011; Meinzer and McCulloh, 2013; Zwieniecki and Secchi, 2015). Norway spruce growing at the timberline has the capacity to refill embolism induced in winter by freezing (Mayr et al., 2014). Moreover, Norway spruce has to some extent the ability to adapt the structure of wood to function optimally in local conditions (Gricar et al., 2015 ˇ ). Could this plasticity be disadvantageous under the impact of a sudden drought? Currently, there is no knowledge on the relationship among climate and anatomical traits associated to hydraulic vulnerability in Norway spruce. A precondition to establish this knowledge is the development of hydraulic predictive traits based on experimentally assessed reference data for P<sup>50</sup> (Dalla-Salda et al., 2011). Recent studies showed that e.g., radial tracheid dimensions are influenced by climate (Castagneri et al., 2015; Gricar et al., 2015 ˇ ) but we lack proof if these traits are related to hydraulic vulnerability. Moreover, in many cases, trees that did not survive extreme climatic events are not available for analysis, since they were harvested in order to avoid e.g., bark beetle outbreaks. Therefore, study sites where healthy trees are compared to declining or dead trees are helpful to learn more about functional anatomical preconditions triggering tree mortality after extreme climate events (e.g., Britez et al., 2014; Hentschel et al., 2014).

For our study we selected two sites in SE Norway, where during the last 20 years an unusual symptom of top-dieback was observed on Norway spruce [Picea abies (L.) Karst.; Solberg, 2004]. The afflicted trees were usually 40–60 years old, dominant or co-dominant in the vigorously growing stands, often on former agricultural lands. First visible symptoms appeared as a stunted top growth, followed by needle discoloration and finally drying of the top crown. The trees died within 1–4 years from the onset of the symptoms. Trees with top-dieback were typically found scattered throughout the stand. So far, we know that decrease in sap flux density can occur during a quite short period (Børja et al., 2016), trees prone to top dieback had a less strict stomatal control and produced wood with lower density prior and during a period of drought stress compared to healthy looking trees (Hentschel et al., 2014; Rosner et al., 2014). Wood density is not directly causally linked to hydraulic vulnerability (P50); it is a proxy where statistical correlations exist because some traits that have an effect on density also affect hydraulic properties (Hacke et al., 2001; Bouche et al., 2014; Lachenbruch and McCulloh, 2014). Once such indirect structure-function relationships are established, their application as screening tools is justified because density and other anatomical parameters are relatively easy to assay, whereas the underlying direct hydraulic measurements are labor intensive and prone to errors (Cochard et al., 2013). However, when taking advantage of proxies such as wood density it should be clear that they are only valid over the conditions used (Lachenbruch and McCulloh, 2014) and thus neither applicable for other species nor for other stem parts (branches) or organs (roots). Another advantage of anatomical proxies for hydraulic properties is that their assessment is independent on the current functional state of secondary xylem (i.e., wood) at the time of harvesting; other than direct hydraulic measurements, proxies allow us to make assumptions on the hydraulic vulnerability of wood that was produced many years ago and that may even not conduct sap vertically anymore (Rosner et al., 2014).

Aims of our study were, (i) to develop hydraulic vulnerability proxies for P<sup>50</sup> based on anatomical data assessed on Norway spruce trunkwood by SilviScan technology and (ii) to test these proxies for their plasticity regarding climate, in order to (iii) analyze annual variations of hydraulic proxies of healthy looking trees and trees with top dieback and hence their impact on tree survival. Our intention was to learn more about the processes related to top dieback in Norway spruce and in particular about the plasticity of possible predisposing factors.

**Abbreviations:** DBH, diameter at breast height [cm]; bh, hydraulic lumen diameter [µm]; bhr, radial hydraulic lumen diameter [µm]; bht, tangential hydraulic lumen diameter [µm]; b<sup>r</sup> , radial lumen diameter [µm]; b<sup>t</sup> , tangential lumen diameter [µm]; d<sup>r</sup> , radial tracheid diameter, from one middle lamellae to the next [µm]; dt , tangential tracheid diameter, from one middle lamellae to the next [µm]; t; double wall thickness [µm]; P50, negative of the overpressure inducing 50 % loss of hydraulic conductivity; RF, number of radial tracheid files/mm circumference; RW, ring width [mm]; SE, standard error; TF, number of tangential tracheid files/annual ring; (t/bh) 2 , conduit wall reinforcement of cells with ±10% bh; (t/bhr) 2 , conduit wall reinforcement of cells with ±10% bhr; (t/bht) 2 , conduit wall reinforcement of cells with ±10% bht.

## MATERIALS AND METHODS

## Study Sites, Plant Material, and Sampling

We studied Norway spruce [Picea abies (L.) Karst.] trees at two forest sites; Sande and Hoxmark (**Table 1**, **Figure 1**) in southern Norway where scattered individual trees showed visual signs of top dieback. Both sites were at low altitude (80 m–110 m a.s.l.) and had rather shallow soils (44–52 cm) on bare rock with marine sediment. Both sites had high clay content (21–24%) relative to sites sampled systematically across the forest area in Norway. Trees were planted on former agricultural sites about 50 years ago. Both sites have been subjected to common forest practices in Norway; the spruce is planted, and this has been followed by weed and grass removal around plants, later by pre-commercial thinning and removal of most of the naturally regenerated broadleaves, and finally thinning and occasional removing of dead trees. The sites were well stocked with living trees with 350–400 m<sup>3</sup> /ha, about 900 stems/ha. About 70–80 m<sup>3</sup> /ha dead trees were recorded (300 trees/ha) on each site. Temperature and precipitation records were available from meteorological stations near both study sites. "Potential water deficit" (mm), i.e., a parameter for water availability, was derived from the cumulative precipitation subtracted by the modeled cumulative potential evapotranspiration including parameters of relative humidity, temperature, cloud cover, and wind speed as input variables. More details about the calculation method and about the study sites can be found in Hentschel et al. (2014). Information about the mean monthly precipitation and the daily water deficit is available in **Supplement Figure 1**.

We selected 10 Norway spruce tree pairs at Sande and Hoxmark, respectively. Each pair consisted of one tree with dieback symptoms, i.e., needle yellowing, needle shortening and decrease in height increment, and the nearest healthy looking neighbor. We use the terminology "healthy looking" because trees with no visual symptoms may have lower sap flux density than expected (Børja et al., 2016). The trees were 40–50 years old, and all sampled trees were still living. Six tree pairs were harvested on each study site in September 2011. Information on growth characteristics of these trees can be found in **Table 1**. Diameter at breast height showed no significant differences across sites and between healthy looking and declining trees, thus no influence on wood structure due to cambial maturation was projected (Lundgren, 2004). For that reason, and because the two sites showed quite similar soil characteristics as well as daily water deficits (**Figure 2**), a pooled analysis for annual variability of anatomical traits was carried out. Wood boles (25 cm) were cut of the 10th whorl from the top immediately after harvesting. After de-barking in the field, boles were transported to the lab in plastic bags containing some fresh water and containing 0.01 vol. % Micropur (Katadyn Products Inc.). From all 40 trees, wood cores (13 mm) were taken at breast height in order to perform anatomical analyses with SilviScan technology. Sapwood specimen for determination of P50, i.e., an outer sapwood zone of 20 mm separated from the wood boles by means of a chisel, as well as wood cores were stored frozen (−18◦C) until further preparation steps.

## Vulnerability to Cavitation (P50)

The dataset for calculating P<sup>50</sup> values of selected sapwood specimen was available from a previous study (Rosner et al., 2014). P<sup>50</sup> was defined as the positive pressure inducing 50 % loss of hydraulic conductivity, also termed "air-seeding pressure." The method to measure P<sup>50</sup> on small trunkwood beams by

(S) and Hoxmark (H) in southern Norway.

TABLE 1 | Information on study sites and growth parameters (mean ± SE) of Norway spruce trees harvested in southern Norway.


No significant differences in breast height diameter (DBH) or tree heights between healthy looking and trees with top dieback were found.

means of the pressure collar technique is described in Domec and Gartner (2002a) and Rosner et al. (2008). We isolated outer sapwood specimens with a transverse surface area of about 9 × 9 mm<sup>2</sup> by splitting the wood along the grain with a chisel. Thereafter, small wood beams with tangential and radial dimensions of 6 mm, respectively, were produced on a sliding microtome. Samples were shortened to 130 mm on a band saw and ends were re-cut with a razor blade to a final length of 120 mm. Specimens had to be kept wet during all preparation steps. Wood beams were thereafter soaked in distilled water under partial vacuum for at least 48 h to refill embolized tracheids. Hydraulic conductivity measurements with distilled, filtered (0.22µm), and degassed water containing 0.005 vol. % Micropur were carried out under a pressure head of 5.4 kPa (54 cm water column). After measurement of the conductivity at full saturation, air overpressure was applied to the specimens by means of a double-ended pressure chamber (PMS Instruments Co., Corvallis, Oregon). Hydraulic conductivity was measured again after a relaxation period in distilled water of 30 min. Initially, the pressure chamber was pressurized to 1.0 MPa, and the pressure was thereafter subsequently increased after each conductivity measurement in steps of 0.5–1.0 MPa until more than 95% loss of conductivity was reached. P<sup>50</sup> values for each sapwood beam were calculated as described in Pammenter and Vander Willigen (1998). In most of the specimen, conductivity measurements were however performed only until 70% loss of hydraulic conductivity, since that proved sufficient to calculate reliable P<sup>50</sup> values for each tree segment (Rosner et al., 2014). In general, the P<sup>50</sup> value for a tree segment is calculated from pooled conductivity and pressure data of more than three single wood beams. It is thus not necessary, that for each wood beam a complete vulnerability curve, i.e., the conductivity loss plotted against the positive pressure, is available. For the present study, selected wood beams were used, where (a) annual rings were perfectly parallel aligned to the tangential longitudinal surfaces, where (b) complete annual rings were present and where (c) a P<sup>50</sup> value could be calculated, thus where data for a complete vulnerability curve were available (n = 19). A detailed description of the origin of these specimens can be found in **Supplement Table 1**.

## Anatomical Investigations with SilviScan Technology and Dataset of Potential Proxies

From the selected wood beams, small wood cubes with radial, tangential and longitudinal dimensions of 6 mm were sawn. Wood cores were thawed and soaked in 96% ethanol in order to avoid the development of cracks or deformation due to shrinkage processes during the drying process at ambient temperature. From dry wood cores and cubes, strips with longitudinal dimension of 7 mm and tangential dimension of 2 mm were sawn by a twin-blade saw. Wood strips were then analyzed by SilviScan technology at CSIRO Forestry and Forest Products (Australia) (Evans, 1994, 1999). X-ray microdensity, radial and tangential tracheid diameters (lumen diameter plus single cell wall, thus the dimension from one to the next middle lamella) and cell wall thickness were assessed in 25 µm radial measurement steps. After cross-dating the wood cores based on the X-ray microdensity variations, a dataset of ring widths (RW) and potential functional traits for each annual ring from 1990 to 2010 was calculated. The same dataset was created for the wood cubes.

Theoretical number of radial tracheid files of 1 mm circumference was estimated as: RF = (1000µm/mean tangential cell diameter (d<sup>t</sup> , i.e., the mean tangential lumen diameter plus single cell wall thickness inµm, **Figure 3**). We defined the theoretical number of tangential tracheid files/annual ring as TF = RW (inµm)/mean radial cell diameter (d<sup>r</sup> , i.e., radial lumen diameter plus single cell wall thickness inµm, **Figure 3**).

Hydraulic lumen diameters were calculated as 6D 5 /6D 4 , where D is the individual tracheid lumen diameter. The method

is preferable over other artificial methods for hydraulically weighting diameter distributions (Kolb and Sperry, 1999). Tracheids of this diameter should cavitate at P<sup>50</sup> in case that air-seeding progresses from wide to narrow tracheids (Hacke et al., 2001). Hydraulic lumen diameters were derived for mean lumen dimensions, i.e., the mean value of radial and tangential lumen diameters, (bh), for radial lumen dimensions (bhr) and for tangential lumen dimensions (bht).

The conduit wall reinforcement is estimated as (t/b) 2 , where t is the cell double wall thickness and b is the lumen diameter (Hacke et al., 2001). (t/b) <sup>2</sup> was calculated for tracheids with diameters ±10% of b<sup>h</sup> (Pitterman et al., 2006; Domec et al., 2009). Using a percent value is applicable across different species, since ± µm suggestions are restricted not only to a given species but also to a specific organ (needles, branch, trunk, roots) within a tree, because lumen diameters are highly variable within a trunk (Anfodillo et al., 2006, 2013; Domec et al., 2009; Carrer et al., 2015). According to different hydraulic lumen diameters presented in this study, (t/b) <sup>2</sup> was estimated for ±10% of b<sup>h</sup> ((t/bh) 2 ), for ± 10% of bhr ((t/bhr) 2 ), and for ± 10% of bht ((t/bht) 2 ).

#### Statistical Analyses and Numbers of Samples

Data are given as mean ± standard error (SE) or as box and whiskers plots. The Pearson's correlation coefficient was used to test the associations between traits. Mean values were examined for significant differences by the Student's t-test. Data were tested for normal distribution with the Kolmogorov-Smirnov test. Relationships between traits and differences in mean values were accepted as significant if P was <0.05. Analysis was carried out with SPSS <sup>R</sup> 21.0.

The relationship between P50, ring width, cell dimensions and potential hydraulic proxies was tested on 19 sapwood specimens originating from 12 trees (**Supplement Table 1**). Chronologies of ring width and selected anatomical traits from 1990 until 2010 were analyzed in 40 trees. For comparisons between healthy looking and trees with top dieback symptoms, mean tree data of the two sites were pooled. Thus, mean values of 20 healthy looking trees and 20 trees with top dieback could be statistically compared (Student's t-test). For anatomy-climate relations of years 1990–2010 (21 years), mean monthly temperature and precipitation were related to annual means of ring widths and selected anatomical parameters at Sande and Hoxmark sites, respectively. This means that each correlation analysis for healthy looking trees and trees with top dieback symptoms was performed with a set of 42 data pairs (21 years of two sites). Correlations for mean daily water deficit resembled those for precipitation, and are thus not shown.

#### RESULTS

#### Relationship between P50, Growth and Anatomical Traits

The outermost still functional sapwood of trees with top dieback symptoms was characterized by a higher hydraulic safety than that of healthy looking trees (**Figures 4A–C**). At the time of harvesting, vulnerability to cavitation (P50) was thus significantly higher in specimen of trees with wider annual rings (**Table 2**). Surprisingly, tangential lumen diameters (bt) were a better proxy

hydraulic lumen diameter ((t/b<sup>h</sup> ) <sup>2</sup>; A,D), with <sup>±</sup>10% of the radial hydraulic lumen diameter ((t/bhr) <sup>2</sup>; B,E), and with ±10% of the tangential hydraulic lumen diameter ((t/bht) <sup>2</sup>; C,F). Gray dots indicate samples from healthy looking trees, black dots from trees with top dieback. Significant relationships at the 1% level are indicated with \*\* and at the 0.1% level indicated with \*\*\* .



The significance of the Pearson's correlation coefficients are indexed with \* if P < 0.05, with \*\* if P < 0.01, and with \*\*\* if P < 0.001. Full names of the traits can be found in the list of Abbreviations. Statistically significant correlations are also indicated by bold fonts.

for P<sup>50</sup> than radial lumen diameters (br). The theoretical number of radial tracheid files (RF), which depends on tangential lumen diameters, was thus tighter correlated with P<sup>50</sup> than the number of tangential files (TF). The hydraulic lumen diameter (bh) was a good predictive trait for P50, however, the hydraulic lumen diameter calculated from radial cell dimensions (bhr) was much weaker related to P<sup>50</sup> than the tangential hydraulic lumen diameter (bht). Double cell wall thickness was tightly correlated with P<sup>50</sup> (**Table 2**). The conduit wall reinforcement of tracheids with ±10% of the hydraulic diameter ((t/bh) 2 ) was therefore strongly related to P<sup>50</sup> as well (**Figure 4A**). The best proxy for P<sup>50</sup> was the conduit wall reinforcement derived from the tangential hydraulic lumen diameter ((t/bht) 2 ), whereas the correlation between P<sup>50</sup> and (t/bhr) 2 , was much weaker (**Table 2**, **Figures 4B,C**). (t/bht) <sup>2</sup> was stronger related to mean ring wood density (**Figure 4F**) than (t/bh) 2 (**Figure 4D**) or (t/bhr) 2 (**Figure 4E**). To sum up, conduit wall reinforcement estimated from the mean hydraulic lumen diameter ((t/bh) <sup>2</sup> was a quite suitable trait for predicting P<sup>50</sup> but (t/bht) <sup>2</sup> was even better (**Figure 4C**).

## Water Deficit in Years Studied (1990–2010)

The most extreme year during the study period was 1992, where mean daily water deficit and temperature was very high in June (**Figure 2**, **Supplement Figures 1**, **2A**). Mean September precipitation gradually decreased from 1991 until 1993 (**Supplement Figure 2B**). In 1994, during the whole early summer, daily water deficit was relatively high; August was however quite wet (**Supplement Figure 2C**). Until 2004, no climate extremes could be found; except in 2001, where June was relatively dry. 2005 and 2006 were characterized by dry June and July months but also by drier September months compared to the previous years. 2007 was a year with very high water supply, and can be considered as an extreme year (**Supplement Figure 1**).

## Climatic Sensitivity of Hydraulic Traits in Healthy Looking Trees and Trees with Top Dieback Symptoms

In **Figure 5** correlations for mean monthly temperature and precipitation, annual radial growth and selected anatomical traits are given. Climatic sensitivity of ring width and anatomical traits differed marginally between healthy looking trees and trees with top dieback. Temperature had almost no effect on ring width (**Figure 5A**), whereas high precipitation in June had a positive impact on annual increment (**Figure 5B**). Cell wall thickness was not related to temperature (**Figure 5C**) and only slightly positively to precipitation in August (**Figure 5D**). Radial lumen diameters were positively influenced by high precipitation in September of the previous growing season and in June (**Figure 5F**) rather than by temperature (**Figure 5E**). Only for trees with top dieback symptoms significant influence of temperature (**Figures 5G,K**) and precipitation (**Figures 5H,L**) on tangential (hydraulic) lumen diameters were found. In both tree groups, radial hydraulic lumen diameters responded significantly positively to high precipitation in September of the previous growing season (**Figure 5J**) but weakly to temperature (**Figure 5I**). Conduit wall reinforcement traits had rather weak relationships with temperature; higher temperature in April resulted however in a slight increase in (t/b) 2 (**Figures 5M,O,Q**). In both tree groups, (t/b) 2 traits responded similarly to changes in temperature and precipitation, where September precipitation had the most significant impact on conduit wall reinforcement (**Figures 5N,P,R**).

## Chronologies of Ring Width and Anatomical Traits in Healthy Looking Trees and Trees with Top Dieback Symptoms

The year 1992, where early summer was extremely dry (**Supplement Figure 1**), was characterized by a sudden decrease in ring width (**Figure 6A**) due to a lower number of tangential tracheid files (**Figure 6B**) and mean radial lumen diameters had minimum peak (**Figure 6E**). Although, 1994 was a year with high summer water deficit, ring width increased gradually until 1995. A slight decrease in ring width as well as in radial lumen diameter (**Figure 6E**) was present in 2001. Ring width in trees with top dieback was slightly, but not significantly, higher than in healthy looking trees in almost all annual rings formed before

#### FIGURE 5 | Continued

(bht; K,L), conduit wall reinforcement based on the hydraulic lumen diameter ((t/b<sup>h</sup> ) <sup>2</sup>; M,N), conduit wall reinforcement based on the radial hydraulic lumen diameter ((t/bhr) <sup>2</sup>; O,P), and conduit wall reinforcement based on the tangential hydraulic lumen diameter ((t/bht) <sup>2</sup>; Q,R) for the period between 1990 and 2010. Blue lines and symbols represent healthy looking trees, red lines and symbols trees with symptoms of top dieback. Significant correlations at (least at) the 5% level are indicated with \* in blue for healthy looking trees and \* in red for trees with symptoms of top dieback.

2004 (**Figure 6A**). After 2005, ring width showed a different trend in trees with top dieback: annual increment decreased gradually and was significantly lower in 2010 than in healthy looking trees. In 2010, mean radial lumen diameter was therefore significantly smaller in trees with top dieback (**Figure 6E**). However, 2007, the year with the highest water supply of the investigation period (**Supplement Figure 1**, **Figure 2**), was characterized by a slight increase in ring width in both tree groups.

Whereas annual fluctuations and mean values of radial (**Figures 6E**), radial hydraulic (**Figure 6H**) and mean hydraulic lumen diameters (**Figure 6G**) were quite similar until 2009 in both tree groups, chronologies of tangential lumen diameters (**Figure 6F**) and the number of radial cell files (**Figure 6C**) exhibited totally different trends after 1996. Between 1998 and 2002, trees with top dieback produced tracheids with significantly wider tangential lumen diameters than healthy looking trees (**Figure 6F**). Tangential hydraulic lumen diameters were significantly larger (**Figure 6I**) and the number of radial cell files significantly lower (**Figure 6C**) in trees with top dieback in the period from 1999 until 2003. Wall thickness showed no significant differences between both tree groups (**Figure 6D**). Therefore, the (t/b) <sup>2</sup> based on the mean hydraulic lumen diameter and on the radial lumen diameter showed similar trends in both tree groups (**Figures 6J,K**), whereas (t/b) <sup>2</sup> derived from tangential lumen diameters was significantly lower in trees with top dieback between 2000 and 2004 (**Figure 6L**) than in healthy looking trees.

## DISCUSSION

## Conduit Wall Reinforcement: Focus on Tangential Tracheid Diameter

The best proxy for P<sup>50</sup> was (t/bht) 2 , the conduit wall reinforcement based on tangential hydraulic lumen diameters (**Figure 4C**). The parameter (t/bh) 2 , i.e., the second power of the wall (t) to span (b) ratio of tracheids which show little deviation from a calculated hydraulic lumen diameter, was introduced by Hacke et al. (2001) as an estimate for P50. Lower vulnerability to cavitation implies the need for a safer cell design with either smaller lumen or thicker walls to resist implosion, because the cell walls have to withstand higher negative xylem pressures before cavitation occurs. Several approaches to calculate this parameter and hydraulic lumen diameters have been presented in the literature so far. Radial dimensions (e.g., Domec et al., 2009; Here¸s et al., 2014; Wilkinson et al., 2015), both radial and tangential dimensions (e.g., Mencuccini et al., 1997; Jyske and Hölttä, 2015) or tracheid diameter modeled from lumen area measurements, considering the lumen either to be rectangular or circular (Mayr and Cochard, 2003; Anfodillo et al., 2006) have been used for calculating hydraulic diameters. Tracheid lumen and wall thickness measurements for t/b or (t/b) 2 are either performed in the first tangential files of earlywood (Rosner et al., 2009), in whole earlywood, since it accounts for most of the hydraulic conductance (Bouche et al., 2014; Rosner et al., 2016), or in several radial files across the whole annual ring (Hacke et al., 2001; Mayr and Cochard, 2003; Domec et al., 2009; Hacke and Jansen, 2009; Here¸s et al., 2014). In the latter case, first a hydraulic diameter is calculated based on the assumption that tracheids with this given diameter cavitate right at P50. Thereafter, (t/bh) 2 is assessed for all tracheids with e.g., ±10% (Domec et al., 2009) of the hydraulic diameter.

In Norway spruce, P<sup>50</sup> is strongly related to wood density across cambial age (Rosner et al., 2014). In the present study, we found that (t/bht) <sup>2</sup> was much stronger related to wood density than (t/bhr) 2 (**Figures 4E,F**); (t/bh) 2 is calculated as a mean of radial and tangential tracheid lumen dimensions and holds thus an intermediate position (**Figure 4D**). We do not have a sound explanation for this result yet, since studies on tangential lumen diameter changes within an annual ring (Vysotskaya and Vaganov, 1989) or with cambial age (Lundgren, 2004; Keunecke et al., 2009) are rather scarce and, moreover, no studies on the predictive quality for P<sup>50</sup> exist. There is however no doubt, that radial tracheid diameters are highly variable within an annual ring (earlywood and latewood) and they are more influenced by climate and cambial age than tangential lumen diameters, because the number of periclinal (radial: inside to outside) division is much higher than that of anticlinal (circumference: side) division (Vysotskaya and Vaganov, 1989; Larson, 1994; Keunecke et al., 2009). The stability of this trait indicated by its normal distribution within a conifer annual ring (Vysotskaya and Vaganov, 1989) might be an advantage regarding the predictive quality for P50. Since (t/bhr) 2 traits were calculated from tracheids with a mean hydraulic diameter, a masking effect of (hydraulically more sensitive but extremely dense) latewood (Domec and Gartner, 2002b; Rosner, 2013; Dalla-Salda et al., 2014) can be excluded. In the next sections, the climate dependence of the tangential (hydraulic) lumen diameter and (t/bht) 2 as well as the variability of these traits in healthy looking trees and in trees with top dieback is discussed.

## Annual Variation of Hydraulic Vulnerability Is Triggered by Climate

Ring width was slightly and positively related to spring temperatures (Churakova et al., 2014), but also to precipitation in September of the previous growing season and in current June (**Figures 5A,B**). In accordance, Mäkinen et al. (2002) found that precipitation during the growing season at low altitude sites (130 m a.s.l.) in southern Norway is positively correlated to growth. At lower latitude sites, in southern Germany, high amounts of precipitation in July were found to foster growth of Norway spruce (Zang et al., 2012). The limiting effect of

(br, E), mean tangential lumen diameter (bt , F), mean hydraulic lumen diameter (b<sup>h</sup> , G), mean radial hydraulic lumen diameter (bhr, H), mean tangential hydraulic lumen diameter (bht, I), conduit wall reinforcement based on the hydraulic lumen diameter ((t/b<sup>h</sup> ) <sup>2</sup>, J), conduit wall reinforcement based on the radial hydraulic lumen diameter ((t/bhr) <sup>2</sup>, K), and conduit wall reinforcement based on the tangential hydraulic lumen diameter ((t/bht) <sup>2</sup>, L) for the period between 1990 and 2010. Blue lines and symbols represent healthy looking trees, red lines and symbols trees with symptoms of top dieback. Significant differences between tree groups (healthy looking and trees with dieback symptoms) at (least at) the 5% level are indicated with \* .

low precipitation on Norway spruce growth decreases, however the effect of temperature increases with increasing latitude or altitude (Jyske et al., 2014). At low and intermediate elevation sites in the Italian Alps, median cell lumen area of Norway spruce annual rings benefited from August, September and October precipitation of the previous growing season and from current year June precipitation (Castagneri et al., 2015). Radial lumen diameters were positively related to high precipitation (thus low temperature) in May and June (**Figures 5E,F**). Similar results are reported by Gricar et al. (2015) ˇ for Norway spruce growing in Slovenia and Czech Republic: lumen dimension of earlywood tracheids are positively affected by precipitation in the previous autumn and early summer of the current growing season. The significant negative correlation we found between precipitation in August and ring width as well as radial lumen dimensions is difficult to explain logically, since at this time of the year cell division has ceased already (Mäkinen et al., 2003; Rossi et al., 2008, 2013; Henttonen et al., 2009; Jyske et al., 2014). An explanation may be that dry early summer periods were often followed by quite wet late summers in the period investigated (**Figure 2**, **Supplement Figure 2**). In trees with top dieback symptoms, higher precipitation in September of the previous growing season and in May of the current season resulted in larger tangential (hydraulic) tracheid diameters (**Figure 5L**). Surprisingly, similar relationships were found in healthy looking trees only for the radial hydraulic diameter (**Figure 5J**).

The most reliable proxy for P50, (t/bht) <sup>2</sup> was negatively correlated to precipitation in September of the previous growing season and current year's May (**Figure 5R**), this implies, that high water availability in late summer of the previous growing season and in spring triggered the production of annual rings with higher hydraulic vulnerability. For instance, high water deficits in September 1993 and 2003 in combination with May water deficits in 1994 and 2004 (**Figure 2**) triggered a rapid increase in (t/bht) 2 in annual rings 1994 and 2004 (**Figure 5I**). 1994 was a year with an extremely wet late summer (**Figure 1**), the outcome was wider annual rings (**Figure 6A**; Castagneri et al., 2015), tracheids with larger diameters (**Figures 6E,F**; Gricar et al., 2015 ˇ ) and thinner call walls (**Figure 6D**) and thus a sudden drop in (t/b) 2 traits in 1995 (**Figures 6J–L**). Stinziano et al. (2015) suggest for boreal forests dominated by Norway spruce, that as the climate warms, the period of net carbon uptake in needles could extend in the autumn, which could increase total carbon uptake in these forests. Warm but not too wet Septembers could result in higher carbon fixation with little impact on the hydraulic vulnerability of the annual ring formed in next growth period.

High temperatures in April had a positive impact on all (t/b) <sup>2</sup> parameters. Norway spruce might be thus able to adapt to some extent to warmer spring temperatures that have been observed during the past decades in this region (Mikkonen et al., 2015). Warmer spring temperatures bear however an increased risk of early spring frosts damage due to earlier de-hardening (Schlyter et al., 2006). In trees with top dieback symptoms, (t/bht) 2 and cell wall thickness benefited as well from high precipitation in August, however also from low precipitation and temperature in September. Wall thickening in cold climates (e.g., at the treeline) can last until end of August to mid of September (Gindl et al., 2000; Treml et al., 2015), in mild continental temperate climate up to October (Cuny et al., 2012). Wimmer and Grabner (2000) found a positive relationship between mean ring wood density, that was tightly correlated with (t/bht) 2 in our study (**Figure 4F**), and precipitation in August for Norway spruce grown in Germany at low altitudes. Due to the negative relationship between wall thickness and ring width (**Table 2**) it is however questionable to conclude logically that wet Augusts favor cell wall thickening. Note that dry early summer periods were often followed by wet late summer periods (**Supplement Figure 2**).

To sum up, climate had a poor effect on cell wall thickness, but quite a strong effect on (hydraulic) tracheid diameters. In both tree groups, high September precipitation in the previous growing season resulted in significantly hydraulically less safe annual rings. Climate-hydraulic proxy relationships shall be tested on other conifer species, since the dynamics of xylogenesis are surprisingly homogeneous among conifer species of the northern hemisphere, although dispersions from the average are observed (Rossi et al., 2013).

## Are Trees Prone to Top Dieback Opportunists?

We found striking differences in the chronologies of hydraulic proxies between healthy looking and declining tress. However wall thickness and radial (hydraulic) lumen diameters and thus (t/bhr) <sup>2</sup> did not differ significantly between the two tree groups. On contrary, mean tangential (hydraulic) lumen diameters and (t/bht) <sup>2</sup> differed extremely between 2000 and 2004 (**Figure 6**). During this period, trees with top dieback symptoms invested fewer carbohydrates in hydraulic safety, as indicated by the high predictive quality of b<sup>t</sup> and (t/bht) 2 for P<sup>50</sup> (**Table 2**, **Figure 4C**). Quite unexpected was the strategy of trees with top dieback for producing wider hydraulic lumen diameters; our results suggest that Norway spruce has two opportunities to maintain hydraulic efficiency: either fewer anticlinal cell divisions or enlargement in the radial direction of tracheids produced by periclinal divisions. Fewer anticlinal cell divisions are probably associated with a decreasing trend of annual growth with age (Keunecke et al., 2009) but might be also interpreted as "opportunistic strategy" as it is obviously less costly since trees with top dieback produced wood with lower (t/bht) 2 (6L) and thus wood density (Hentschel et al., 2014; Rosner et al., 2014). Climate correlations were found for tangential (hydraulic) lumen in trees with top dieback but not in healthy looking trees. Wall thickness was only marginally affected by climate, therefore, trees with top dieback consequently produced wood with lower (t/bht) <sup>2</sup> when e.g., water supply in previous year's September and current May was sufficient. This "opportunistic strategy" can be a risky investment regarding hydraulic safety since it can lead to lower survival prospects under the impact of an extreme sudden drought. Mature Norway spruce needs at least 10 annual rings for axial water transport (Bertaud and Holmbom, 2004). Due to irreversible embolism (Choat et al., 2015) or quite slow recovery from drought induced embolism (reviewed in Zwieniecki and Secchi, 2015), hydraulic conductivity might get lost forever in more vulnerable annual rings or in the most conductive parts of a given annual ring. This can result in an impairment of the water supply of the crown and finally to reduced growth. Healthy looking trees might have been better prepared for the dry July months in 2005/6 (**Figure 2**), since they produced hydraulically safer wood in the period 2000– 2004 as indicated by significantly higher (t/bht) 2 (**Figure 6L**). In accordance, from 2005, annual increment took a different course in trees with top dieback (**Figure 6A**).

The concept of "opportunistic strategy" is not in line with the more anisohydric behavior (Tardieu and Simonneau, 1998) of trees with top dieback symptoms since they tended to spend water; they had a predisposition to less strict stomatal control (Hentschel et al., 2014). Aguadé et al. (2015) underline the intertwining of physiological mechanisms leading to droughtinduced conifer mortality and the difficulty of isolating their contribution under field conditions. According to McDowell et al. (2008), mechanisms that cause mortality in trees are carbon starvation, i.e., failure to maintain metabolism or defense reaction due to prolonged negative carbohydrate balance, and hydraulic failure. Both processes are likely coupled when conifers become vulnerable to pests and extreme climate (McDowell et al., 2016). Anisohydric strategy should be coupled with a more hydraulically safe design in order to avoid conductivity loss during severe droughts when more negative water potentials develop in the tracheids, whereas isohydric behavior, i.e., a more strict control of water loss through stomatal closure, should demand less safe wood (Tardieu and Simonneau, 1998). However, recent studies show that shifts rather than a clear separation between isohydry and anisohydry exist (McDowell, 2011; Klein, 2014; Martínez-Vilalta et al., 2014; Sevanto et al., 2014). Carbon starvation is closely linked to water supply in the plant, since water is necessary for transport of non-structural carbohydrates (carbon reserves) in the secondary phloem (Hartmann et al., 2013a; Sevanto et al., 2014). Whereas catastrophic hydraulic failure in Norway spruce can occur in above-ground tissues, belowground tissues (roots) can die from carbon starvation, indicating that mortality mechanisms are not defined at the organism level but rather within tree compartments (Hartmann

et al., 2013b). In accordance, in trees with top dieback, higher fine-root mortality was found (Godbold et al., 2014). The abrupt change from a rather dry period (2005/6) to an extremely wet year in 2007 could have influenced root vitality negatively and could have made trees more vulnerable to biotic agents. Whatever finally caused top dieback, the process was favored by the risky combination of less strict stomatal control (Hentschel et al., 2014) and by the "opportunistic strategy" of producing more vulnerable wood when e.g., precipitation in the previous year's September and in early summer was high.

## Is breeding for Higher Hydraulic Safety Possible?

Concerning different, probably genetically determined, wood formation characteristics of healthy looking and declining trees prior to drought stress (under sufficient water supply), the question arises if it was possible to breed Norway spruce or other conifer species for higher hydraulic safety (Dalla-Salda et al., 2011; Montwé et al., 2016). Wood density (Rosner et al., 2014) or (t/bht) 2 could be easily applicable screening tools and early selection for wood density is highly effective from rings 6–7 in Norway spruce (Chen et al., 2014). Starting at that age, selecting individuals for higher hydraulic safety in trunkwood could be thus possible.

Trunkwood safety covers however only one aspect of drought sensitivity related to hydraulic architecture (Hacke et al., 2015), and selected trees shall be tested for their hydraulic performance in field experiments. In a recent study on potted 3-year-old seedlings, limited variation at the family level indicates that the response to drought is quite conservative within Norway spruce, which may limit breeding opportunities for increased drought resistance (Chmura et al., 2016). At that young age, hydraulic safety is however anyway very high and hydraulic structure-function relationships (e.g., with density) are masked by mechanical demands of the young trunk (Rosner, 2013). It must be also considered, that in mature Norway spruce wood, density (and thus (t/b) 2 ) is genetically negatively correlated with growth (Hannrup et al., 2004; Rosner et al., 2014). When selecting such conifer species for higher hydraulic safety, reduction in growth might have to be taken into account (Montwé et al., 2016). Guiding breeding programs requires also much more knowledge on stress physiology of a given species with focus on carbohydrate supply within a tree during different stress levels and the processes related to recovery from embolism at the cellular and whole plant level (Zwieniecki and Secchi, 2015).

## CONCLUSIONS AND OUTLOOK

Regarding the aims of our study, (i) to develop proxies for P<sup>50</sup> based on anatomy and (ii) to test these proxies for their plasticity regarding climate, in order to (iii) interpret annual variations of hydraulic proxies related to top dieback, we arrived at the following conclusions.

(i) The high predictive quality of (t/b) <sup>2</sup> based on tangential lumen diameters for P<sup>50</sup> implies, that, if we solely focus on radial cell dimensions as basis for this proxy we will lose a lot of information regarding hydraulic vulnerability of Norway spruce trunkwood. Calculating a mean diameter (e.g., derived from the lumen area), is preferable to the radial diameter, since information of the tangential diameter is included. The next step is to test these relationships within a given species, thus between different organs of a tree, and on other conifer species.


## AUTHOR CONTRIBUTIONS

LD, SS had the project idea, were responsible for the project design, proposal writing and project management. JS, KA, IB, LD, and SR contributed to field work (site selection, tree selection, coring, harvesting, and sample pre-preparation), in that regard, site and tree selection was one of the biggest challenges. Hydraulic measurements were conducted by SR. SL was responsible for sample preparation for SilviScan measurements. RE assessed the SilviScan dataset and gave important input concerning data analyses. OT provided climate raw data and calculated daily water deficits. The dataset was analyzed by SR after prepreparation steps done by JS, OT, and RE. All authors contributed to the interpretation of the results. SR wrote a first draft of the manuscript; thereafter all authors revised the first draft by rewriting, discussion and commenting. All authors were involved in re-writing the text and creating new graphs after the first revision of the reviewers. All authors agree on the contents of this manuscript.

#### FUNDING

This study was mainly financed by the Norwegian Research Council (project "Dieback in Norway spruce", No. 199403), by the Norwegian Forest Owners' Research Fund "Skogtiltaksfondet", six regional funds in Norway (Fylkesmannen), by the Austrian Science Fund FWF (V146-B16) and by the CZ Ministry of Education (No. 6215648902). The research leading to these results has also received funding (employment of SL) from the European Union's Seventh Framework Programme for research, technological development and demonstration under grant agreement n◦ 284181 ("Trees4Future"). The contents of this publication reflect only the authors' views and the European Union is not liable for any use that may be made of the information contained therein.

#### REFERENCES


#### ACKNOWLEDGMENTS

This work has been carried out under the framework of the COST FP1106 network STReESS. We thank Rainer Hentschel for assistance in the field and helpful discussions. We also thank both reviewers for carefully reading our manuscript and helpful suggestions. Ute Sass-Klaasen is thanked for additional important comments on the content on the manuscript and her careful editorial work.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00831

Supplement Figure 1 | Mean monthly precipitation and daily water deficit from 1957 to 2015 of Sande and Hoxmark sites (gray areas or bars). Red lines indicate the year 1992, where early summer was extremely dry, blue lines indicate the extremely wet year 2007.

Supplement Figure 2 | Mean monthly temperature (A), precipitation (B) and daily water deficit during the vegetation period (May–September) of years 1989–2010. Bars represent mean values for Sande and Hoxmark.

Supplement Table 1 | Information on the origin of the wood beam sample set for calculation of P<sup>50</sup> and anatomical proxies from the SilviScan data set (n = 19) of 12 trees, where six trees were healthy looking and six trees showed signs of top dieback.


is involved in adaptation to drought in Douglas-fir (Pseudotsuga menziesii (Mirb.)). Ann. For. Sci. 68, 747–757. doi: 10.1007/s13595-011-0091-1


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Rosner, Svetlík, Andreassen, Børja, Dalsgaard, Evans, Luss, Tveito ˇ and Solberg. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Diverging Drought Resistance of Scots Pine Provenances Revealed by Infrared Thermography

#### Hannes Seidel <sup>1</sup> \*, Christian Schunk <sup>1</sup> , Michael Matiu<sup>1</sup> and Annette Menzel 1, 2

<sup>1</sup> Department of Ecology and Ecosystem Management, TUM School of Life Sciences Weihenstephan, Technische Universität München, Freising, Germany, <sup>2</sup> Institute for Advanced Study, Technische Universität München, Garching, Germany

With recent climate changes, Scots pine (Pinus sylvestris L.) forests have been affected by die-off events. Assisted migration of adapted provenances mitigates drought impacts and promotes forest regeneration. Although suitable provenances are difficult to identify by traditional ecophysiological techniques, which are time consuming and invasive, plant water status can be easily assessed by infrared thermography. Thus, we examined the stress responses of 2-year-old potted Scots pine seedlings from six provenances (Bulgaria, France, Germany, Italy, Poland, and Spain) based on two thermal indices (crop water stress index and stomatal conductance index). Both indices were derived from infrared images during a 6-week drought/control treatment in a greenhouse in the summer of 2013. The pines were monitored during the stress and subsequent recovery period. After controlling for fluctuating environmental conditions, soil moisture or treatment-specific water supply was the most important driver of drought stress. The stress magnitude and response to soil water deficit depended on provenance. Under moderate drought conditions, pines from western and eastern Mediterranean provenances (Bulgaria, France, and Spain) expressed lower stress levels than those from both continental provenances (Germany and Poland). Moreover, pines from continental provenances were less resilient (showed less recovery after the stress period) than Mediterranean pines. Under extreme drought, all provenances were equally stressed with almost no significant differences in their thermal indices. Provenance-specific differences in drought resistance, which are associated with factors such as summer precipitation at the origin of Scots pine seedlings, may offer promising tracks of adaptation to future drought risks.

Keywords: thermal imaging, water supply, aboveground dimensions, thermal indices, tissue temperature, CWSI, climate change

## INTRODUCTION

Scots pine (Pinus sylvestris L.) forests are sensitive to drought-related dieback. In a review of global forest mortality, Scots pine forests accounted for 40% (10 out of 25 cases) of all European die-off events (Allen et al., 2010). This situation might worsen in the future as climate change simulations propose increasing temperatures and decreasing local summer precipitation even in moderate scenarios (Kirtman et al., 2013). Seedlings and young trees are more vulnerable to stress

#### Edited by:

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### Reviewed by:

Gerald Moser, University of Giessen, Germany Giovanna Battipaglia, Seconda Università degli Studi di Napoli, Italy

> \*Correspondence: Hannes Seidel hseidel@wzw.tum.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 28 January 2016 Accepted: 05 August 2016 Published: 31 August 2016

#### Citation:

Seidel H, Schunk C, Matiu M and Menzel A (2016) Diverging Drought Resistance of Scots Pine Provenances Revealed by Infrared Thermography. Front. Plant Sci. 7:1247. doi: 10.3389/fpls.2016.01247 (e.g., drought) than large trees, especially when not sheltered by dense canopies (Niinemets, 2010; Bussotti et al., 2015). This vulnerability might seriously impede forest regeneration. Thus, for successful forest management and climate change adaptation, the assisted migration of adapted tree species or the selection of suitable provenances might be necessary (Millar et al., 2007). This could be achieved by transferring seeds or plant material from drier and/or warmer climates to regions with similar future projected climates (Bussotti et al., 2015). Assisted migration might be especially appropriate for species with wide-ranging distributions and contrasting environments, such as P. sylvestris (Boratynski, 1991).

Scots pine provenances differ in their response to water availability. In a study conducted in Valais, Switzerland (Richter et al., 2012), the number of Mediterranean seedlings after a summer drought was twice the number of continental seedlings. Seedlings also differ in their shoot and/or height increments during drought periods (Taeger et al., 2013a, 2015). A dendroecological study revealed varying drought resistance among Scots pine provenances (Taeger et al., 2013b). Stomatacontrolled leaf traits of Pinus pinaster (Fernández et al., 2000) and P. halepensis (Tognetti et al., 1997; Klein et al., 2013), such as stomatal conductance, transpiration rates, and intrinsic wateruse efficiency, respond differently to water shortage in different provenances. However, no study has evaluated the provenance– drought interaction effects on the ecophysiological leaf traits in P. sylvestris.

Monitoring plant responses to climatic changes by ecophysiological techniques (e.g., water potential, xylem vulnerability to cavitation, stomatal conductance, transpiration rates, water use efficiency) is frequently time consuming and/or destructive. The applicability of these ecophysiological measurements in the field is generally reduced by limited accessibility to adult tree canopies. In contrast, crop breeding programs adopt non-invasive and high throughput techniques such as RGB imaging, chlorophyll fluorescence, thermal imaging, and imaging spectroscopy (Fiorani and Schurr, 2013). However, these techniques are mostly applied to morphologically simply structured organisms, e.g., Arabidopsis, or cereals and other crop plants.

When evaluating drought stress in plants, thermal imaging relates the actual surface temperatures of the leaves to their water availability. Plants interact with their aboveground environment by exchanging water, carbon, and energy, mostly through their stomata. One function of stomatal control is to maximize the photosynthetic gain while minimizing water loss through the leaves (Chaves et al., 2003; Jones, 2013). Meanwhile, the leaf tissue temperature depends on the stomatal conductance. As stomata close, the decreased transpiration reduces evaporative cooling and thus increases the leaf temperature (Raschke, 1960). Therefore, leaf temperature can be an indicator of stomatal closure and hence of water availability.

In recent years, the explanatory power of thermal images for drought stress responses has been improved by various approaches. Several thermal indices have been developed to normalize leaf surface temperatures under temporally changing environmental conditions. These indices have been linked to stem or leaf water potential and stomatal conductance (as reviewed in Maes and Steppe, 2012). Under field/outdoor conditions, thermal indices are especially recommended for constant (semi-) arid weather conditions as they have low variability under high vapor pressure deficits, so the changing weather conditions are relatively unimportant. Under temperate/moist conditions, the thermal indices are more problematic. They are also influenced by vegetation, canopy, and leaf characteristics. On the other hand, it is advantageous that thermal imaging can cover large spatial scales and efficiently catch the plant-to-plant variability in a single measurement (Maes and Steppe, 2012).

Most studies examining tree water status by thermal imaging have been conducted in orchards of almond, apple, citrus, olive or peach (Andrews et al., 1992; Sepulcre-Cantó et al., 2006; Ben-Gal et al., 2009; Wang and Gartung, 2010; García-Tejero et al., 2011; Gonzalez-Dugo et al., 2012; Zarco-Tejada et al., 2012; Agam et al., 2014; Virlet et al., 2014). In these studies, the thermal imaging discriminated between water-stressed and non-waterstressed individuals. The drought sensitivity of deciduous tree species has also been ranked by thermal imaging of their canopies (Scherrer et al., 2011). In general, these studies are hampered by the heterogeneity of orchard and forest trees and of the study sites themselves (Maes and Steppe, 2012). Apart from Leuzinger et al. (2010), thermal imaging of conifers is almost unreported in the literature; thus, comparison studies of conifers in different provenances by thermal indices are largely lacking.

We used thermal imaging to examine the drought stress responses of potted Scots pine seedlings from six provenances in a greenhouse experiment, assuming soil moisture as the most important driver. We investigated (i) whether Scots pines from different provenances differ in their stress responses, (ii) whether they respond differently to soil water deficit, and (iii) whether and to what extent the differences in thermal indices are explained by the plant dimension covariates. For a given water supply, the soil moisture in the pots might also depend on plant biomass (which differs among provenances), as larger individuals will probably have higher water consumption rates. Therefore, we additionally tested (iv) whether the stress levels under specific irrigation treatments differ among provenances.

## MATERIALS AND METHODS

## Experimental Setup Plant Material

Scots pine seedlings were grown from seeds in a nursery in 2011 and were potted in April 2012. Pots had a volume of 3.l, were filled with peat substrate and placed in a greenhouse to conduct an extensive seasonal drought and warming experiment involving 10 provenances from all over Europe that started in 2013. Among this larger experimental setup, we randomly selected 48 2-year-old seedlings from six provenances. The climate conditions at the origin of the seeds are quite different (6–11◦C annual mean temperature, 600–900 mm annual sum of precipitation), comprising Mediterraneancontinental (Bulgaria), Mediterranean (Spain, France, Italy), and temperate-continental (Germany, Poland) sites (**Table 1**; see


also Taeger et al., 2013a and Supplementary Figure 1). The heights and diameters of the seedlings before the experiment were similar across most provenances and treatments (**Table 1**; Supplementary Figure 2). However, the tree heights differed among the provenances (Kruskal-Wallis test, p < 0.001), being larger in the German and Polish provenances than in the Spanish (Dunn's test, p < 0.01), and the French one (Dunn's test, p < 0.05). There were no significant height differences across provenances between the two treatment groups, except in the control treatment where the German specimens were taller than the Spanish ones (Kruskal-Wallis test, p < 0.01). No significant height differences were observed among the treatments for all six provenances. Seedling stem diameters were also not significantly different between treatments and provenances except for the control treatment, in which the Italian specimens had larger diameters than the French specimens (Dunn's test following a Kruskal-Wallis test, p < 0.05).

#### Drought Treatment

Automated dripping irrigation allowed four water treatment groups in the larger experimental setup, but we implemented just two treatments under the time constraints of this study. Twenty-seven of the selected individuals assigned to the control group, and 21 were subjected to a summer drought from July 11th to August 21st 2013. In the drought treatment, irrigation was initially intermitted and only small amounts of water were added afterwards forcing the soil moisture to oscillate around the permanent wilting point. During this 42-day period each individual in the control group received 3050 ml (i.e., 190 mm) water, while individuals in the drought treatment received only 725 ml (i.e., 45 mm, **Figure 1A**). On August 22nd, all pots were saturated with water; in the subsequent recovery period until September 4th, all specimens were again well-watered with identical amounts of water. We weighted each of the 48 pots on each measuring day, and calculated the percent soil water deficit (PSWD) as the difference between the pot weight at field capacity on July 5th and the actual weight of the pots divided by the absolute water content at field capacity (see **Figures 1B,C**). The absolute water content of the pots at field capacity was derived from water retention curves following the pressure plate method by Richards (1941) and was estimated as 40% at 10 kPa soil-water matrix potential.

#### Moisture Levels during Experiment

 or

 provenances,

 were

 treatments

 provenance.

> On July 5th, all pots were fully saturated and their calculated percent soil moisture deficits (PSWDs) were zero (therefore, the first measuring day is omitted from **Figures 1B,C**). During the drought treatment period (from July 11th to August 21st), individuals in the drought group received only ∼24% of the irrigation water added to the control group (**Figure 1A**). As revealed in the overarching larger experiment, this drought treatment corresponded to conditions around the permanent wilting point. Seven days from the start of treatment (July17th), PSWD differences between the treatments became significant (Wilcoxon test, p < 0.001; **Figure 1B**). We thus define the period from July 17th until August 21st as the stress period. We could not detect any significant differences in PSWD across

provenances within each treatment (see Supplementary Figure 2C for mean PSWDs during the stress period), although the individuals of some provenances differed in height and diameter, likely causing unequal water depletion in the pots. In both treatment groups, the PSWD decreased at the beginning of the stress period and stabilized after approximately 1 week; however, PSWD of the drought treatment group remained significantly lower until August 21st (**Figure 1C**). These differences partly remained during the recovery period (from August 22nd to September 4th; measuring days 14–17), although the pots were fully saturated with water on August 22nd. Because the PSWDs of the two treatment groups were not always fully separated, we modeled the stress response in two ways; the first based on the absolute PSWD, the second based on drought treatment vs. control treatment referring to different water supply scenarios (see subsection Statistical analysis).

#### Thermal Indices

The plant surface temperatures can be related to drought stress and plant water status by several methods (reviewed in Maes and Steppe, 2012). To account for the changing environmental conditions, the surface temperatures must be normalized by reference temperatures (see Reference surfaces and plant monitoring platform). We used two thermal indices; the crop water stress index (CWSI) and stomatal conductance index (Ig).

Jones (1999) proposed the CWSI as a modification of Idso et al.'s (1981) formulation. The CWSI is known to mirror stomatal conductance and the leaf and stem water potentials. It normalizes the leaf surface temperatures by the surface temperatures of wet (Twet) and dry (Tdry) references, where Twet represents a fully transpiring leaf and Tdry a non-transpiring leaf. The CWSI is calculated as

$$\text{CWSI} = \frac{T\_{\text{canopy}} - T\_{\text{wet}}}{T\_{\text{dry}} - T\_{\text{wet}}} \tag{1}$$

I<sup>g</sup> employs the same variables as CWSI but is linearly related to the stomatal conductance (Jones, 1999). Thus, Ig is a linear function of the stomatal opening:

$$I\_{\mathcal{S}} = \frac{T\_{dry} - T\_{canopy}}{T\_{canopy} - T\_{wet}} \tag{2}$$

Higher CWSI and lower Ig values indicate higher surface/tissue temperatures and thus stomatal closing.

#### Reference Surfaces and Plant Monitoring Platform

As suggested in Meron et al. (2003) and Möller et al. (2006), we calculated the abovementioned thermal indices using artificial reference surfaces (ARSs), which are included in each picture (see Meteorological data and **Figure 2**). To mimic the maximum transpiring surface, we wrapped a white cotton fabric around a styrofoam board floating in a water-filled plastic box (wet reference). A non-transpiring leaf was represented by an opal white laminated fiberboard (dry reference) mounted on the plastic box (**Figure 2**). Jones (1999) used wet and dry leaves as the reference surfaces, but here we chose the ARSs because the thin needles of conifers (unlike leaves) easily dry out under the high greenhouse temperatures (**Figure 3A**). Additionally, the temperature information in the pixels of the reference needles might become mixed with that of non-reference needles in the background. Unfortunately, this approach might be sensitive to changing environmental conditions because the short-wave absorptances and heat capacities differ between needles and reference targets. For these reasons, environmental variables were incorporated as control covariates in the statistical models (see Statistical analysis). In addition to the wet and dry ARSs, two black painted electric heating plates with a mean temperature of 40◦C were horizontally attached to a support frame. These created a strong contrast to the plant tissue and totally masked the pot and soil in the images. A small gap in the middle edge of the plates prevented squeezing of the trunk. The handling time of taking an individual tree to the monitoring platform, mounting it into the platform and capturing up to three thermal images was approximately 3 min.

#### Thermal Imaging

Thermal images were acquired by a thermal infrared camera (VarioCam hr inspect 780, Infratec, Dresden Germany) with a resolution of 1280×960 infrared pixels using the optomechanical resolution enhancement. The thermal resolution within the images was below 0.08 K at 30◦C and the absolute measurement accuracy was ± 1.5 K. The emission coefficient was set

(1) the heating plates, (2) dry reference, (3) wet reference, and a pine located between the heating plates.

to 1 during the image acquisition and was later corrected using the emission coefficients calculated by the quotients of the contact thermocouple temperatures and the thermal image-derived temperatures of the reference surfaces and pine seedlings (performed in a dark chamber under temperature and relative humidity control and in the greenhouse, respectively). The resulting emission coefficients of the wet reference, dry reference and pines were 0.95, 0.93, and 0.90, respectively. The calculated emissivity can be considered as the apparent emissivity since we did not measure background temperature. The errors caused by this omission should be negligible inside the greenhouse, which was constantly shaded for the thermal imaging, largely corresponding to cloudy conditions (see Meteorological data, Maes and Steppe, 2012). Thermal images were taken approximately twice a week from July 5th to September 4th, leading to 17 measuring days. Ten of these days constituted the stress period with clear PSWD effects, and 4 days followed the summer drought treatment (see **Figure 1B** for exact dates). During the image acquisition (between 11 a.m. and 3 p.m.), the greenhouse was shaded to reduce the possible influence of variable solar radiation (e.g., due to scattered clouds). The camera was vertically mounted at 2.5 m above the monitoring platform attached to the scaffolding of the greenhouse. On every measuring day, each of the 48 pines was photographed 2 or 3 times (rarely once due to technical problems), together with the reference surfaces.

In the subsequent image analysis, the plant tissue was separated from the (heated) background, and interferences at the needle edges (where single pixels were mixtures of plant and background temperatures) were additionally removed. To automate the image processing, a script was written in Fiji (Schindelin et al., 2012). Images were processed by the following steps. First, the image was sharpened using the command "Unsharp Mask" with a radius of 2 and a mask weight of 0.9. From this sharpened image, two masks were created, one to remove the background (using auto thresholding based on the intermodes algorithm), the other to remove the edges (using the "Find Edges" command). The median temperature of the plants' canopy and the dry and wet references in each image were calculated from the remaining pixels. The respective mean daily temperatures were then determined from the multiple (1–3) images acquired on each measuring day. Finally, the thermal indices were calculated from the mean temperatures per measuring day and used in subsequent statistical analysis.

#### Meteorological Data

During the thermal image acquisition, an air temperature (T) and relative humidity (RH) sensor (HOBO U23 Pro v2, Hobo <sup>R</sup> , Onset Computer Corporation, Bourne, MA) was placed next to the plant monitoring platform. Data were recorded in 1-min intervals. In the greenhouse, the air temperature, relative humidity and solar radiation were measured at 10-min intervals throughout the whole study period by a meteorological weather station (Davis Vantage Pro2 PlusTM, Davis Instruments, Hayward, CA). We matched the meteorological and thermal image data with their nearest temporal counterparts. The vapor pressure deficit (VPD), defined as the difference between

saturation vapor pressure (es) and actual vapor pressure (ea), was calculated after Allen et al. (1998) with T and RH as the input variables.

The daily mean values of the air temperature, relative humidity, solar radiation, and vapor pressure deficit, collected at 10-min intervals between 11 a.m. and 3 p.m. (the time window of the thermal image acquisition on measuring days), varied during the study period (July 5th to September 4th), within the ranges 16.6–36.6◦C, 26.4–85.0%, 33.0–421.7 W/m<sup>2</sup> , and 0.3– 4.5 kPa, respectively (**Figure 3**). The minimum and maximum air temperature, relative humidity, solar radiation and vapor pressure deficit during the image-acquisition time over the 17 measuring days were 20.4 and 37.1◦C, 29 and 66%, 24 W/m<sup>2</sup> and 234 W/m<sup>2</sup> , and 1.0 kPa and 4.5 kPa, respectively. Thus, the conditions during measuring times well represented the indoor conditions at noon over the course of the study period except for the solar radiation, which was mostly determined by the additional shading in the greenhouse (generally throughout June and July, and on the measuring days after mid-August).

#### Statistical Analysis

We analyzed the effect of PSWD on thermal indices over the whole study period (from July 5th to September 4th). The analysis was performed by linear mixed-effects models (nlme; Pinheiro et al., 2016) implemented in R version 3.2.2 (R Core Team, 2015). Full models were constructed by adding the covariates provenance as factorial dummy variable, heights, and diameters of seedlings at the beginning of the experimental period, air temperature, relative humidity, vapor pressure deficit, and solar radiation and the two-way interactions of PSWD with provenance, air temperature, relative humidity, and vapor pressure deficit. Because the PSWD varies nonlinearly with the thermal indices, it was added as a linear and quadratic term to the models. All covariates besides PSWD were centered on their means (by subtracting their respective means from the discrete variable values) for easier interpretation of interaction effects. As time and PSWD were strongly correlated, they cannot be included simultaneously due to collinearity; hence, no time variable was included in the PSWD-based models. To account for the repeated measurements of individuals during the experiment, we included the individual trees as random variables.

In a first step of model selection we either chose air temperature plus relative humidity or vapor pressure deficit as covariates based on the AIC (Akaike Information Criterion) of the respective models, since these variables were strongly correlated. The full model is mathematically expressed as

$$\begin{aligned} Index\_{i,j} &= \beta\_0 + b\_{0,i} + \gamma\_1 Powername\_i + \beta\_1 PSWD\_{i,j} \\ &+ \beta\_2 PSWD\_{i,j}^2 + \beta\_3 Height\_i + \beta\_4 Diameter\_i \\ &+ \beta\_5 VPD\_{i,j} + \beta\_6 Radiation\_{i,j} + \gamma\_2 Provename\_i \\ &\* \, PSWD\_{i,j} + \gamma\_3 Provename\_i \ast \, PSWD\_{i,j}^2 + \beta\_7 VPD\_{i,j} \\ &\* \, PSWD\_{i,j} + \epsilon\_{i,j} \end{aligned} \tag{3}$$

with random intercepts b0,<sup>i</sup> ∼ N(0, σ 2 0 ) and errors ǫi,<sup>j</sup> ∼ N(0, σ 2 ǫ ). Indexi,<sup>j</sup> is the thermal index of tree i (=1 . . . 48) at measurement j (=1 . . . 17). The meteorological variables (in this case, the VPD and radiation) also vary from tree to tree, because they were measured sequentially on the measurement day, and altered throughout the course of the day. As the six provenances were modeled using dummy variables, each of γ1, γ2, andγ<sup>3</sup> is a five-dimensional coefficient vector of the dummy regressors.

To simplify the full models, we evaluated the importance of the explanatory variables/interactions using the drop1 function (stats; R Core Team, 2015). Any terms that did not improve the models' explanatory power were excluded (**Table 2**). The R<sup>2</sup> of the final models was computed by the r.squaredGLMM function (MuMIn; Barton, 2015, **Table 3**). To avoid heteroscedacity and non-normal distribution of the residuals, we examined the diagnostic plots and applied variance function structure classes. The Ig was square-root-transformed to meet these criteria. Provenances were compared by a pairwise post-hoc test using the glht function (multcomp, Hothorn et al., 2008) comparing contrasts with the Tukey's range test. By this test, we also compared the provenances under additional PSWD scenarios (0%, 50%, and 100% PSWD) after centering the PSWD values on these thresholds and refitting the models. To check the different stress behaviors of provenances in relation to PSWD, we tested the coefficients of the linear and quadratic terms of the PSWD– provenance interaction with the glht function.

Differences in thermal indices between the treatments (water supplies) and among the provenances, and in their corresponding ability of the pines to recover from the water stress, were separately analyzed over the stress period (measuring days 4–13, July 17th until August 21st) and the recovery period (measuring days 14–17, August 24th until September 4th). Here we fitted and simplified the linear mixed models as described above. The factorial dummy variable treatment (control/drought) was used rather than the PSWD, and the two-way interactions of treatment with height and diameter were added to the initial full model.

TABLE 2 | Variables included in the final linear mixed models evaluating the effect of percentage soil water deficit (PSWD) on the thermal indices CWSI and Ig, respective effect sizes (Estimate) and p-values extracted from the summary table of the models.


Effect sizes of categorical variables (provenance) and interaction terms involving this variable are not shown, but their contribution is indicated by X. The PSWD contribution involves a linear (PSWD) and a quadratic (PSWD<sup>2</sup> ) component. In the model fittings, the vapor pressure deficit (VPD) and solar radiation were centered on their means. The Ig was square-root-transformed.

As the PSWD changes over time in both treatments and study periods, we added a time variable (number of days since the observations started) as a control covariate with a linear and a quadratic term. Based on the AIC, we selected the VPD and air temperature/relative humidity as covariates. The full model is mathematically expressed as


where Treatment<sup>i</sup> is 0 if tree i received the dry treatment, and 1 for the control group. t<sup>j</sup> is the number of days after the start of the experiment (July 5th) at measurements j (=4 . . . 13) for the stress period and j (=14 . . . 17) for the recovery period. Employing the glht function, we again tested the differences among provenances and between treatments in pairwise post-hoc tests, comparing contrasts with the Tukey's range test.

Additionally, we analyzed the response magnitudes of the provenances between treatments during the stress period. The mean index values were calculated for each individual. The response magnitude was calculated as the pairwise difference in particular indices between each individual of the drought group and all other individuals of the control group. Differences in response magnitude, tree height and diameter were analyzed by the Kruskal–Wallis test (stats; R Core Team, 2015) and by Dunn's test for multiple comparisons (FSA; Ogle, 2015).

All p-values for multiple comparisons were corrected by the false discovery rate (FDR) method.

#### RESULTS

#### Relationship between Thermal Indices, PSWD, and Provenances

The two thermal indices were significantly related to PSWD, provenances and meteorological control parameters. Excluding some meteorological control covariates (temperature, relative humidity) and plant biomass covariates (diameter, height) from the full models, the final models resulted in better fits in terms of lower AIC-values and explained a proportion of variance of 0.64 and 0.62 for CWSI and Ig, respectively (**Table 2**).

To interpret the signs of the model variables (**Table 2**), we must remember that CWSI and Ig increase and decrease at higher stress levels, respectively. In both thermal indices, higher stress levels were linked to higher PSWD, indicating lower soil-water availability (**Table 2**; **Figure 4**). However, provenance and the its interaction with PSWD were also included in the models of the thermal indices (**Table 2**). The Bulgarian and especially the Spanish provenance showed a pronounced stress minimum (PSWD = 15–20%; see **Figure 4**). Although responses


TABLE 3 | P-values of the pair-wise comparisons (Tukey's range test of contrasts) of the provenances' response to PSWD estimated by linear mixed-effects models.

P-values are listed for the coefficients of linear and quadratic components of PSWD (PSWD/PSWD<sup>2</sup> ). The upper-right and lower-right triangles display the CWSI and Ig results, respectively. Provenances are abbreviated as follows: F3, France; PL9, Poland; ES1, Spain; D8, Germany; BG10, Bulgaria and I4 Italy. Bold values indicate significance at a level of 0.05.

of provenances to PSWD were varying, significant differences were detected only in the Spanish provenance. Pines in this provenance were much more sensitive to increasing water deficit than all provenances other than the Bulgarian one (**Table 3**; **Figure 4**). Increasing VPD and radiation also increased the stress levels (either by natural response of the stomata and leaf surfaces or because the ARS-derived indices were sensitive to the changing environmental conditions); **Figure 4** displays the responses of the different provenances to PSWD, for the mean VPD and radiation conditions. Note that all dimensional variables (seedling height and diameter) were excluded from the final model because they added no further explanatory power aside from provenance.

## Discrimination of Provenances under Different Soil Moisture Scenarios

Under well-watered conditions (PSWD = 0%) the provenances from Spain and Bulgaria were less stressed than those from Poland and Germany (p < 0.05; Supplementary Table 1). Marginal differences were observed between pines from France and Germany and between pines from Italy and Spain (p < 0.1; data not shown). At medium soil moisture (PSWD = 50%) the Bulgarian and Spanish seedlings were less stressed than the German and Polish seedlings and the Spanish provenance performed better than the Italian provenance (p < 0.05). Additionally, the CWSI was marginally different between the French and Spanish seedlings and between Italian and Bulgarian provenances (p < 0.1; data not shown). Under severe drought conditions (100% PSWD), only the two most extreme provenances (France and Spain) showed significant differences in both thermal indices (p < 0.01). The Ig significantly differed between pines from Poland and France (p < 0.05; Supplementary Table 1).

## Variation of Stress Level with Provenance, Treatment (Water Supply), and Seedling Dimensions

The immediate drought stress during the stress period and the conditions during the recovery period, were assessed for both treatments (defined by their water supply i.e., the amount of irrigation water) by the thermal indices. For both CWSI and Ig, the explanatory power of the corresponding linear mixed models was lower during the stress period (both indices = 0.63) than during the recovery period (0.74 and 0.75, respectively, **Table 4**).

During the stress and recovery periods, the respective final models of CWSI and Ig included provenance, time and treatment. As expected, stress was increased by the drought treatment and continued to increase throughout the stress period, then declined over the recovery period. In contrast, the sets of plant specific (dimension, provenance) and meteorological covariates explaining the variation in thermal indices differed between stress and recovery periods (**Table 4**).

Diameter was negatively associated with stress during the stress period, but was unimportant in the models of the recovery phase (**Table 4**; **Figures 5B,C**). The influence of seedlings' height on the thermal indices depended on the water supply treatment (**Table 4**; **Figures 5B,C**). Specifically, the stress levels increased with height in the control treatment but were independent of height in the drought treatment. Equally, height was unimportant in the recovery period models.

During the stress period, the effect of provenance on the thermal indices depended on the treatment (**Table 4**; **Figure 5**), whereas during the recovery period, the treatment and provenance influenced the thermal indices without interacting with each other (**Table 4**; **Figure 6**). In each provenance, the TABLE 4 | Variables included in the final linear mixed models evaluating the influence of the drought treatment (defined by water supply) on the thermal indices CWSI and Ig during the stress period (measuring days 4–13, July 17th to August 21st) and the recovery period (measuring days 14–17, August 24th to September 4th).


The effect sizes (Estimates) and p-values are extracted from the summary table of the models. The effect sizes of categorical variables with more than two levels (provenance) and interaction terms containing this variable, are not shown, but their contribution is indicated by X. The models were fitted with the continuous variables centered on their means. The Ig was square-root-transformed. T and RH denote air temperature and relative humidity, respectively. Bold values indicate significance at a level of 0.05.

Radiation 0.00086 <0.001 −0.0015 <0.001

seedlings were significantly more stressed in the drought than in the control group during the stress period (p < 0.01, Supplementary Table 2). The exception was the Italian provenance, where high stress levels were already observed in the control group (**Figure 5**). Although the treatment differences under stress decreased by ∼50% during the recovery period (**Table 4**), the overall differences between treatments persisted in the recovery period (**Figures 6B,D**; Supplementary Table 3). During the stress period, the provenance differences were not significant in the highly stressed drought treatment group. However, in the control group, whose PSWD levels imposed moderate stress during this period, the CWSI levels indicated more stress in the Italian provenance than in provenances from France, Spain and Bulgaria. However, the Ig did not capture these differences at a significance level of 0.05 (**Figures 5A,D**; Supplementary Table 2).

During the recovery period, provenances showed equal stress patterns in both treatment groups, because the provenance and treatment interaction was not significant in the statistical model (**Table 4**). According to both thermal indices, the provenances from Spain and Bulgaria were less stressed than those from Poland and Germany, and the French provenance was less stressed than the German provenance (**Figures 6A,C**; Supplementary Table 3). Furthermore, the CWSI showed higher stress levels in the Italian seedlings in than the Spanish and Bulgarian ones. Although the seedlings recovered, the significant difference between the control and experimental groups persisted into the recovery period, indicating some post-drought stress (**Table 4**; **Figures 6B,D**).

#### Response to Water Supply Treatments

The treatment effect (water supply) during the stress period clearly separated the individuals of almost every provenance. The exception was the Italian provenance, in which the negative group differences indicated overlap of the stress levels of both treatment groups (Supplementary Figure 3). The differences between treatments exhibited similar patterns for both thermal indices. The Spanish and Italian provenances demonstrated the strongest and weakest response to water supply, respectively. During the stress period, the response to the water supply treatment was stronger in the Spanish pines than in the Polish, Bulgarian, and Italian pines, as shown by both indices (Supplementary Table 4). The CWSI and Ig revealed further differences in the water supply response; specifically, the Italian seedlings demonstrated weaker CWSI response than the French, Polish, and German seedlings, and the Ig response to water supply was higher in Spanish than in German pines (Supplementary Table 4).

#### DISCUSSION

#### Methodological Considerations of Thermal Imaging

When performing thermal imaging under non-100% controlled conditions, several limitations must be considered (Prashar and Jones, 2014). Variations in environmental conditions affect the leaf temperature, so reference surfaces are required. Here we replaced wet and dry Scots pine needles with ARSs (see Materials and Methods) instead of using wet and dry Scots pine needles which would have had similar spectral and physical properties to the target leaves (Jones, 1999). Nevertheless, the ARS performance should be adequate for the following reasons: (1) the emissivity of each target surface was determined; (2) the greenhouse was shaded before and during image acquisition, meaning that all thermal images were acquired under low levels of incoming solar radiation; (3) on most measurement days, the T, RH, and VPD varied within a narrow range, suggesting that the errors introduced by differing heat capacities of the targets are also small; (4) when comparing indices across provenances and treatments, these errors can be partially compensated by incorporating meteorological covariates in the statistical models.

Additionally, the temperature within a thermal image might vary with angle of view, illumination, and distance to targets (Prashar and Jones, 2014). In our experimental design, these

parameters were maintained largely constant by a plant monitoring platform. All images were horizontally captured above the seedlings during a short time interval, ensuring a fixed distance to the seedlings. The mixed pixels on the plant edges were removed during the image processing. As all seedlings were morphologically similar, these geometric and lighting factors can also be neglected.

#### Provenance Performance

The differences in stress sensitivity and recovery of the six provenances were studied after adjusting for PSWD or (more simply) for treatment (water supply). During the stress period, the increased PSWD indicated mild drought stress even in the control group (**Figure 1B**). Therefore, the provenance performance was studied under two conditions of water stress; extreme (drought treatment) and moderate (control treatment in the stress period).

The Spanish and Bulgarian seedlings were more distressed by over-saturated than under-saturated water conditions (**Figure 4**). Excessive soil water may decrease the oxygen availability for roots. The lower root respiration reduces the root function and causes leaf dehydration (Vartapetian and Jackson, 1997). Consistent with these findings, the ecophysiological leaf traits of Scots pine (net assimilation, stomatal conductance and transpiration) are lower in short-term water-logged soils than in soils hydrated to field capacity (Repo et al., 2016). The germination rates, shoot growths and root growths of Scots pines of different provenances also respond differently to waterlogging (Mukassabi et al., 2012).

Important triggers of the drought stress response besides PSWD were treatment, provenance and the provenance–PSWD and provenance–treatment interactions. However, because of the limited sample size, significant differences in stress sensitivity between the provenances were rarely identified. Both lines of evidence (PSWD and water supply) imply that pine species' response to strong drought stress (either modeled for 100% PSWD or for the drought treatment during the stress period, when the soil moisture approximated the wilting point) is comparable among provenances. At 100% PSWD, significant differences were observed only between the most extreme provenances (Spain and France, and additionally between France and Poland in the Ig analysis). In contrast, for moderate stress (i.e., over the range of measured PSWD, for the given water supply in the control treatment, or as modeled for 50% PSWD) our study revealed significant differences. Pines in the Spanish provenance tended to be less sensitive to moderate drought stress than Italian pines. Furthermore, the Spanish and Bulgarian pines were significantly less stressed at 50% PSWD than the German and Polish ones, and the Italian provenance demonstrated higher response to mild stress under the control water supply than the French and Bulgarian provenances. Taeger et al. (2013a, 2015) investigated the stem diameter, stem length and respective relative growth rates in pines from these provenances, and reported similar differences among the provenances. The difference in stress levels between the control group (exposed to moderate stress) and the drought group (exposed to extreme drought stress) were smallest and largest in pines from Italy and Spain, respectively (see Supplementary Figure 3).

The Spanish, French, and Italian provenances experience a Mediterranean climate with minimal precipitation in summer; whereas the Bulgarian provenance (despite its similarly dry summer season) is continental (see Supplementary Figure 4 and **Table 1**). In contrast, the German and Polish provenances are clearly characterized as temperate-continental.

In moderate drought scenarios (50% PSWD or the control group during the stress period in), the pines from provenances with a summer precipitation minimum at their origin (Bulgaria, France, Spain) were more stress-resilient than pines from provenances with rainy summers (Germany, Poland). Thus, the temperate continental provenances exhibited the highest stress under moderate drought. The superior resilience of the Mediterranean type group (the Spanish and Bulgarian provenances) lacks an obvious explanation. Mediterranean Scots pines, which are considered to naturally adapt to drought events (Cregg and Zhang, 2001; Richter et al., 2012; Taeger et al., 2013a), may have adapted their stomatal control under our study conditions. This finding may also imply that southern (Mediterranean) provenances transpired less and used a smaller share of the offered water supply 3050 ml during 42 days of treatment. Tognetti et al. (1997) reported comparable results in drought-stressed Pinus halepensis, whose leaf conductance depends on the moisture content of its origin. Drought adaptation in P. sylvestris might be governed by lower investment in aboveground biomass and higher biomass allocation to roots (Cregg and Zhang, 2001; Taeger et al., 2015). However, identifying the mechanism of drought adaptation is beyond the scope of this study.

During the recovery period, pines from the Polish, German, and Italian provenances maintained significantly higher stress levels than their Spanish and Bulgarian counterparts. This trend might also reflect a differential stomatal control.

An important feature of our study was the overlap of stress levels in both treatments, despite clear PSWD differences between the treatments. This phenomenon, which appeared in some instances and was especially observed in pines from the Italian provenance, suggests a resource saving strategy as the stress levels were already high under mild drought conditions (control treatment). Such a strategy is supported by a previous mortality experiment, in which trees from the Italian provenance were least threatened by drought-induced mortality (Seidel and Menzel, 2016).

#### Influence of Seedling Dimensions on Stress Levels

The seedling diameter did not significantly differ among the provenances, but the continental individuals were taller than the Spanish and French ones. Although the seedling dimensions were not included in the final linear mixed-effects models relating PSWD to stress levels, provenance-specific growth traits might contribute to the water responses of individual trees, as mentioned above. The exclusion of tree dimensions from the PSWD models indicates that dimensional traits do not purely drive the observed differences between provenances. This is also indicated by the fact that the water supply models included both provenance and tree dimensions.

In the water supply models, the seedling dimensions (diameter and height) significantly affected the stress response. In particular, the stress levels generally decreased with increasing tree diameter (**Figures 5C,F**). Trees can store a considerable amount of water in sapwood, whose volume is closely related to stem diameter (Meinzer et al., 2001). Indeed, a simulation study of Pinus sylvestris showed a strong relationship between stored water use and tree diameter. Stored water use can contribute up to 40% of the total daily transpiration (Verbeeck et al., 2007). Thus, large diameter provides a water buffer against drought stress. Under mild drought conditions, we detected a positive relationship between stress level and seedling height (**Figures 5B,E**). A link between reduced growth of aboveground biomass and increased drought adaptation has been suggested (Alía et al., 2001; Valladares et al., 2007). Thus, the increased stress levels in higher seedlings may reflect the higher water consumption of larger sized seedlings or the higher transpiration of larger needle area. In contrast, the seedlings in the drought group exhibited uniform stress levels, most likely because they consumed the irrigation water to an equal extent.

#### CONCLUSION

The study investigated Scots pine specimens from six provenances. In practical forestry, the best provenance for a given water supply (i.e., effective stand precipitation) is an important question. The results suggest a trade-off between stress resistance and height growth; that is, higher stress tolerance is inevitably linked to smaller trees. As the dimensions drive the water consumption of an individual tree, the drought effects are less prominent under nominal soil moisture conditions than under a controlled water supply. In summary, Scots pine seedlings from different provenances respond differently to moderate drought stress, but more uniformly to severe drought stress. Thus, by regarding leaf temperature as a stress indicator, we conclude that drought sensitivity and resilience of Scots pine depends on its native provenance. Individuals from Mediterranean climates, especially from Spain and Bulgaria, are better adapted to moderate drought than pines from temperate continental regions. In practical forestry, provenance-based assisted migration may be a viable adaptation response to climate change.

#### AUTHOR CONTRIBUTIONS

HS collected data, contributed to the experimental design, analyzed, and interpreted the data and wrote the paper. CS collected data, contributed to experimental design, revised the

#### REFERENCES


paper and contributed to writing the paper. MM contributed to experimental design, contributed to data analyses, revised the paper and contributed to writing the paper. AM contributed to the conception of the work, interpreted the data and wrote the paper.

#### ACKNOWLEDGMENTS

This work was financed by the European Research Council under the European Union Seventh Framework Programme (FP7/2007–2013/ERC grant agreement No. 282250). We thank Allan Buras and especially Elisabet Martínez Sancho for fruitful discussion, Marina Gabler, Laura Stratopoulos, and Stefanie Weindler for help with thermal image acquisition and Steffen Taeger, Andreas Ludwig (BaySF) and the ASP/Teisendorf for providing plant and seed material. We further thank the team of the GHL Dürnast for carefully handling all plant material during the overarching drought experiment.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 01247


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Seidel, Schunk, Matiu and Menzel. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Plasticity in variation of xylem and phloem cell characteristics of Norway spruce under different local conditions

Jožica Gricar ˇ 1 \*, Peter Prislan<sup>1</sup> , Martin de Luis <sup>2</sup> , Vladimír Gryc<sup>3</sup> , Jana Hacurová<sup>3</sup> , Hanuš Vavrcíkˇ <sup>3</sup> and Katarina Cufar ˇ <sup>4</sup>

<sup>1</sup> Department of Yield and Silviculture, Department of Forest Techniques and Economics, Slovenian Forestry Institute, Ljubljana, Slovenia, <sup>2</sup> Department Geografía, University of Zaragoza, Zaragoza, Spain, <sup>3</sup> Faculty of Forestry and Wood Technology, Mendel University in Brno, Brno, Czech Republic, <sup>4</sup> Department of Wood Science and Technology, Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia

#### Edited by:

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### Reviewed by:

Emanuele Ziaco, University of Nevada, USA Alan Crivellaro, University of Padova, Italy

#### \*Correspondence:

Jožica Gricar, ˇ Department of Yield and Silviculture, Department of Forest Techniques and Economics, Slovenian Forestry Institute, Vecna pot 2, SI-1000 Ljubljana, Slovenia jozica.gricar@gozdis.si

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 08 June 2015 Accepted: 28 August 2015 Published: 10 September 2015

#### Citation:

Gricar J, Prislan P, de Luis M, Gryc V, ˇ Hacurová J, Vavrcík H and ˇ Cufar K ˇ (2015) Plasticity in variation of xylem and phloem cell characteristics of Norway spruce under different local conditions. Front. Plant Sci. 6:730. doi: 10.3389/fpls.2015.00730 There is limited information on intra-annual plasticity of secondary tissues of tree species growing under different environmental conditions. To increase the knowledge about the plasticity of secondary growth, which allows trees to adapt to specific local climatic regimes, we examined climate–radial growth relationships of Norway spruce [Picea abies (L.) H. Karst.] from three contrasting locations in the temperate climatic zone by analyzing tree-ring widths for the period 1932–2010, and cell characteristics in xylem and phloem increments formed in the years 2009–2011. Variation in the structure of xylem and phloem increments clearly shows that plasticity in seasonal dynamics of cambial cell production and cell differentiation exists on xylem and phloem sides. Anatomical characteristics of xylem and phloem cells are predominantly site-specific characteristics, because they varied among sites but were fairly uniform among years in trees from the same site. Xylem and phloem tissues formed in the first part of the growing season seemed to be more stable in structure, indicating their priority over latewood and late phloem for tree performance. Long-term climate and radial growth analyses revealed that growth was in general less dependent on precipitation than on temperature; however, growth sensitivity to local conditions differed among the sites. Only partial dependence of radial growth of spruce on climatic factors on the selected sites confirms its strategy to adapt the structure of wood and phloem increments to function optimally in local conditions.

Keywords: cambium, growth/climate relation, Picea abies, tracheid, tracheidogram, cell differentiation, sieve cell, phloemogram

#### Introduction

Norway spruce [Picea abies (L.) H. Karst.] is considered to have high adaptive potential and, despite the adverse effect of climate change on its growth, remains one of the most important European forest tree species (Skrøppa, 2003). Due to its major economic importance and vulnerability to climate change, several tree-ring studies have been carried out in recent decades (e.g., Bošel'a et al., 2014). Nevertheless, studies dealing with the plasticity of tree species, or climate reconstruction from tree rings, should not only focus on trees growing in relatively extreme sites (i.e., treeline or xeric habitats) but should also include various temperate habitats with moderate growing conditions, as has been recently stressed by several research groups (Carrer et al., 2012; Drew et al., 2013). Namely, most trees grow in forests that are not at the altitudinal or latitudinal limits of their distributions; consequently, large areas of the forested biomes of the world are ignored (Drew et al., 2013).

In dendrochronology, tree-ring widths are normally presented as time-series and are analyzed on an annual scale (Fritts, 1976). However, environmental information in trees is encoded in the inter- and intra-annual variability of tree-ring widths, including earlywood and latewood width, wood density, cellular wood structure, and chemical composition of cell walls (Eckstein, 2013). Cells are created in the wood formation process, which can be divided into five consecutive developmental stages: cell division, cell enlargement, cell wall thickening, lignification and programmed cell death (e.g., Rossi et al., 2014). The process depends on genetic signaling, availability of resources, temperature, tree water and nutrient status and the stage of ontogenetic development (Hölttä et al., 2010). Thus, to understand the mechanisms and the dynamics of wood formation in relation to climatic or physiological factors better, analyses on a shorter temporal scale are required (Rossi et al., 2014). Investigating the xylem anatomy level has already been demonstrated to be a promising approach in tree-ring studies (Fonti et al., 2010), while the environment–phloem relationship is still relatively unexplained. In general, the development of bark cells is very complex and not yet fully understood (Gricar et al., ˇ 2015). Although their main functions appear to be completely different, xylem and phloem are closely associated both spatially and functionally (Evert, 2006). There are two main long-distance pathways in trees, one related to the xylem conducting water and nutrients absorbed from the roots up to the leaves to sustain evapotranspiration and photosynthesis, and one related to the phloem delivering sugar solutions produced by photosynthesis that are necessary for cell respiration and growth to all the living tissues (Petit and Crivellaro, 2014). Carbon gain and whole-tree survival depend on the functioning of and interplay between these two vascular subsystems (e.g., Sevanto et al., 2014).

Research on the seasonal dynamics of phloem formation in various tree species (Prislan et al., 2013; Gricar et al., ˇ 2014b; Swidrak et al., 2014), as well as on the anatomy of phloem in relation to tree vitality (Gricar et al., 2014a ˇ ), osmotic potential (Rosner et al., 2001) or variation along the stem (Petit and Crivellaro, 2014; Jyske and Hölttä, 2015), has noticeably increased in the last decade. Nevertheless, year-to-year variations in phloem cell characteristics in trees from different environments still remain relatively unknown. Secondary changes in older phloem tissue hinder its anatomical analysis in consecutive annual increments; only the youngest phloem increments are therefore appropriate for such observations (Gricar et al., 2015 ˇ ). As a result, sampling over several years needs to be performed to acquire a time series of phloem cell characteristics. In a recent study of Norway spruce, it was shown that the seasonal dynamics of phloem formation exhibited less plasticity than xylem formation in trees growing at the same location (Gricar et al., 2014b ˇ ). However, differences in the phloem phenology among different sites suggest that phloem development is at least partly affected by local environmental conditions.

Intra- and inter-species variation in secondary tissues exists because of local adaptations and environmental conditions (Rowe and Speck, 2005). The capacity to change growth form may represent a significant advantage of plants when growing in changing environmental conditions (Agusti and Greb, 2013). There is limited information on intra-annual plasticity of variation of xylem and phloem cell characteristics of individual tree species under different environmental conditions. To increase the current knowledge about the plasticity of secondary growth that allows trees to adapt to specific environmental regimes (Rowe and Speck, 2005), we examined long-term radial growth of spruce from three contrasting locations in the temperate climatic zone by analyzing tree-ring widths for the period 1932–2010, and cell characteristics in xylem and phloem increments formed in the years 2009–2011. This joint approach of dendrochronological and quantitative-anatomical methods allowed to define climate-growth relationships from cell parameters and tree-ring widths more precisely. In order to obtain more information about the strategy of Norway spruce for functioning optimally in temperate local conditions, we addressed two main questions: (1) are there site-specific responses of wood and phloem anatomy to local conditions and (2) is year-to-year variability in environmental conditions more pronounced in the structure of xylem or phloem?

We analyzed the long-term tree ring-climate relationship, also separately for earlywood and latewood. The main novelty of this work is that it provides fundamental information on the variation in phloem anatomy over three consecutive years, in parallel with variations of xylem conduits in Norway spruce from three different sites with contrasting climate conditions.

## Materials and Methods

#### Study Sites

The study was carried out at three forest sites with different altitudes and latitudes: two in Slovenia and one in the Czech Republic. In Slovenia, sampling was performed at two unevenaged mixed forest stands differing in altitude. The low elevation site, Panška reka (PA), is located near Ljubljana and is classified as Hacquetio-fagetum typicum forest type, where Fagus sylvatica L., Acer pseudoplatanus L. and P. abies (L.) H. Karst prevail (**Table 1**, **Figure 1**). The high elevation site, Menina planina (ME), is on a pre-Alpine Karst plateau in the Kamnik-Savinja Alps and is classified as Abieti fagetum prealpinum typicum forest type, with F. sylvatica, P. abies, and Abies alba Mill. being the dominant species. The site in the Czech Republic, Rájec-Nemˇ cice (RN), is ˇ located north of Nemˇ cice, ca. 400 km away from the Slovenian ˇ sites. The site is classified as Abieto-Fagetum mesotrophicum with Oxalis acetosella L. It is spruce monoculture (the first generation after mixed forest) (Fabiánek et al., 2009).

Climatic time series (minimum, maximum and mean monthly temperatures and monthly amount of precipitation) for the


TABLE 1 | Location and characteristics of the study sites.

period 1901–2013 were obtained from the closest gridpoint of the CRU TS3.22 dataset (Harris and Jones, 2014). The general features of climatic regime at the three study sites are presented in **Table 2**.

#### Tree Selection and Sample Preparation for Xylem and Phloem Analyses

We selected six dominant or co-dominant, healthy Norway spruce trees at each of the sites. The trees were 68 ± 8 (average ± standard deviation) years old at PA, 102 ± 31 at ME and 88 ± 4 at RN, with diameters at breast height (DBH) ranging from 30 to 40 cm (36 ± 5 cm at PA; 37 ± 12 cm at ME and 34 ± 2 cm at RN) and heights of around 25–32 m (30 ± 5 m at PA; 25 ± 1 m at ME and 32 ± 2 m at RN). For anatomical analysis of xylem and phloem increments, we used micro-cores that had been collected at the end of September/beginning of October in 2009, 2010, and 2011, when annual xylem and phloem increments were fully developed. They were taken at 1.1–1.7 m above ground using a Trephor tool (Rossi et al., 2006). In order to avoid wound TABLE 2 | Climate data: mean annual temperature (T), T range, mean annual amount of precipitation (P) and P range at PA, ME, and RN for the period 1901–2010.


effects, sampling locations were separated by 10 cm. Each microcore contained phloem (non-collapsed and collapsed), cambium and at least two of the last-formed xylem rings. Immediately after removal from the trees, the samples were fixed in ethanolformalin acetic acid solution (FAA). After 1 week, the samples were dehydrated in a graded series of ethanol, infiltrated with Dlimonene and embedded in paraffin blocks (Rossi et al., 2006). Transverse sections of 8–12µm thickness were cut with a rotary microtome, using low profile microtome blades. The sections were transferred to object glass and stained with a safranin (0.04%) and astra blue (0.15%) water mixture (van der Werf et al., 2007), dehydrated and embedded in Euparal. They were observed under a light microscope using transmission and polarized light modes. Histometrical analyses were performed with a digital camera and an image processing program ImageJ (Abramoff et al., 2004).

#### Measurements of Xylem and Phloem Cell Characteristics and Data Processing

In samples taken at the end of the vegetation periods of 2009– 2011, xylem and phloem cell characteristics were measured immediately after the cessation of cambial activity and cell differentiation processes (**Figure 2**). This was important because, early phloem sieve cells already started to collapse in the autumn of the current growing season, which we tried to avoid. Vaganov (1990) used the term "tracheidogram" for plots showing the variation of tracheid parameters in the growing period of a conifer. By analogy, we called the plots of variations in phloem cell size (i.e., sieve cells and axial parenchyma cells) in the growing season "phloemograms."

Measurements were performed on transverse sections along three radial cell rows per growth ring. Only rows in which cells were cut approximately in the middle of their longitudinal length (i.e., cells with maximum tangential and radial diameter) were chosen. In the phloem, we analyzed cells that were not crushed. Mean values of the listed anatomical variables for earlywood/latewood and early phloem/late phloem separately were calculated for each cell row. Since the number of cells in

radial rows of the xylem or phloem annual rings was different, it was necessary to standardize the sample size per year in order to compare them among trees, sites and years. We therefore used the "relative position" of each cell within a radial row of an annual ring instead of an absolute value.

In xylem increments, we analyzed the following variables: ring width expressed in number of cells (mean of three radial files/tree), number of earlywood and latewood cells (mean of three radial files/trees), radial diameter of tracheids (mean of three radial files/tree), lumen diameter of tracheids (mean of three radial files/tree), and double cell wall thickness of tracheids (mean of three radial files/tree) for each cell in a radial row. Finally, the mean values of initial earlywood (i.e., first tangential row of cells at the growth ring boundary) and terminal latewood tracheids (i.e., last tangential row of cells adjacent to the cambium) were calculated. Data for each anatomical variable per file were transformed into a tracheidogram. Latewood tracheids were defined according to Denne's (1988) first interpretation of Mork's (1928) definition; i.e., when the width of the cell lumen was smaller than twice the double cell wall thickness.

In phloem increments, we assessed five anatomical variables: ring width expressed in number of cells (mean of three radial files/tree), number of early and late phloem cells (mean of three radial files/trees), radial diameter (mean of three radial files/tree) and number of sieve cells (mean of three radial files/tree) for each cell in a radial row. Finally, the mean values of initial early phloem (i.e., first tangential row of cells at the phloem growth ring boundary) and terminal late phloem sieve cells (i.e., last tangential row of cells adjacent to the cambium) were calculated. The transition from early to late phloem was identified by the appearance of the first axial parenchyma cells separating the two tissues, which is typical of Pinaceae.

#### Tree-rings and Chronology Computation

At all sites, we took samples of wood (cores or discs) for dendroclimatological analysis from 13–16 dominant or codominant Norway spruce trees growing near to each other and having approximately the same diameters, heights and ages as the trees used for analysis of xylem and phloem anatomy. The samples of wood taken from trees after the end of 2010 (PA, ME) and 2011 (RN) growing seasons, were polished and the tree-ring widths, as well as visually separated earlywood and latewood widths, were measured to the nearest 0.01 mm. TSAP Win or WinDENDRO programs were used for data acquisition and cross-dating.

The tree-ring series were assembled into local chronologies using the ARSTAN programme (Holmes, 1994). The individual raw tree-ring series were standardized in a two-step procedure. First, the long-term trend was removed by fitting a negative exponential function or a regression line to each tree-ring series. Second, more flexible detrending was achieved by a cubic smoothing spline with a 50% frequency response of 60 years, in order to reduce non-climatic variance further. Thereafter, biweight robust estimation of the mean was applied to construct local standardized chronologies (Cook and Peters, 1997).

#### Anatomical Variables and Climate-growth Relationships

Pearson's Product Moment Correlation Coefficient was used to measure the strength of association between xylem and phloem ring widths, earlywood, and latewood widths, early phloem and late phloem widths, and number of cells and measured width of xylem ring at PA, ME, and RN. The differences among years and sites in the listed xylem and phloem anatomical variables were determined using individual One-Way repeated measures ANOVA, in which the site was the treatment factor and year of growth was the repeated measure. The normality of distribution and homogeneity of variance were verified using the Shapiro-Wilk W-test and Levene's test, respectively (Quinn and Keough, 2002).

Climate–growth relationships were calculated through correlation function analysis using the program DendroClim2002 (Biondi and Waikul, 2004), whereby the standardized chronologies of the tree-ring chronology was the dependent variable and the regressors were the monthly minimum, maximum and mean temperatures and the monthly sums of precipitation for each year from the previous September to the current October. Analyses were based on the common 1931–2010 period, in which an elevated Expressed Population Signal (EPS > 0.85) was observed in all three chronologies. Relationships between climatic conditions and xylem and phloem anatomical variables for the period 2009–2011 were also calculated. For each site and year, mean values of each analyzed variable were compared with the monthly minimum, maximum and mean temperatures and the monthly sums of precipitation from previous September to current October using Pearson correlation. Procedure was the same as for dendrocronological analyses, however here we used 3 sites together and 3 years (9 values).

## Results

#### Xylem and Phloem Cell Characteristics

In general, we found statistically significant differences in xylem and phloem anatomical variables among the selected sites but not among the years (**Table 3**). Only the average double cell wall thickness of earlywood tracheids and of initial earlywood tracheids also statistically differed among the years. On the other hand, no significant differences were found in the number of earlywood cells, the radial (lumen) dimension of initial earlywood tracheids and radial dimension of terminal latewood and late phloem cells. In the case of the number of latewood and late phloem cells, as well as the radial dimensions of initial early phloem cells, we detected differences only between some sites and some years, showing an inconsistent site effect and thus high variation in the values of these anatomical parameters. Correlation between number of cells and measured width of xylem ring was 0.98, between xylem and phloem ring widths 0.98, between earlywood and latewood widths 0.23, and between early phloem and late phloem widths 0.41.

Spruce at PA thus had on average the widest xylem and phloem increments, latewood and late phloem, as well as the thickest walls of earlywood and latewood tracheids (**Figures 3**–**8**). The radial dimensions of phloem sieve cells were proportionally related to phloem width (**Figures 7**, **8**). Xylem annual increments were narrowest at ME, while phloem increments were narrowest at RN. Trees at ME had the highest proportion of earlywood, widest early phloem and widest average radial (lumen) dimension of earlywood tracheids. At RN, only the lumen dimensions of latewood cells were significantly wider than at the other two sites. Phloem rings at ME and RN were about 20 and 40%, respectively, narrower than at PA (**Figure 6**). Furthermore, spruce at RN had about 16 and 12%, respectively, narrower early phloem and late phloem sieve cells than at ME and PA.

Early phloem sieve cells were on average 30% wider than late phloem cells at all three sites. The lumen dimensions of early phloem sieve cells were 12–30% smaller than those of earlywood tracheids. In the increments formed in the second part of the growing season, it was just the opposite: lumen dimensions were 30–50% smaller in latewood than in late phloem. A tangential band of axial parenchyma that normally separates early and late phloem was discontinuous in narrower phloem rings, composed of fewer than six cell layers, which were detected only in spruce at RN in 2010 and 2011. In contrast, an additional discontinuous tangential band of axial parenchyma was detected in the late

TABLE 3 | Summary of one-way repeated measures ANOVA of anatomical variables in which "site" was the treatment factor and "year of growth" was the repeated measure.


Statistically significant values are in bold.

phloem of increments wider than 12 cell layers, as observed at PA.

Only some variations in anatomical variables in xylem and phloem were explained by seasonal climatic data (**Figures 3**–**8**). Thus, the number of earlywood cells was positively influenced by precipitation in the previous October and December, and maximum temperature in the period November–February (**Figure 3**). The wall thickness of xylem cells was mainly influenced by maximum temperature and precipitation of the previous November, and in the case of latewood cells by minimum temperature in July and precipitation in October. The lumen dimension of earlywood tracheids was affected by precipitation in the previous October and current June, and by minimum temperature in February and May. Precipitation in the previous autumn and current June, and maximum temperature in winter negatively influenced lumen dimensions of terminal latewood tracheids (**Figures 4**, **5**). The number of (late) phloem cells as well as their radial dimensions were mainly affected by precipitation in late autumn and maximum temperature in winter (**Figures 6**–**8**). The number of early phloem cells and dimension of all phloem cells were also positively influenced by precipitation in April. In addition, radial dimension of initial early phloem sieve cells was predominantly determined by maximum temperature in the period January–March.

Cell wall thickness of the tracheids along the radial files slowly increased and reached maximum values in latewood cells at a 0.7–0.9 relative position within a radial row and afterwards again decreased (**Figure 9**). Radial dimensions of xylem and phloem cells showed a different trend; cells formed at the beginning of the growing season grew larger than those created later in the growing season (**Figures 10**, **11**).

#### Long-term Xylem-climate Relations

The main descriptive statistics of the local tree-ring, earlywood and latewood chronologies of spruce at PA are presented in **Table 4**. For spruce at PA, summer (June, July) temperature (where both minimum and maximum temperature had negative

affect) proved to be the most important climatic factor explaining year-to-year variations in the widths of xylem increments (**Figures 12A,D,G**). In addition, June temperature negatively affected earlywood width, whereas July and August temperature negatively affected latewood width. Furthermore, January temperature positively influenced tree-ring and earlywood widths, while the previous September's temperature and precipitation influenced latewood widths. At ME, tree-ring and earlywood widths were positively related to March precipitation (**Figures 12B,E**). In the case of latewood, widths were influenced by the climatic conditions of the previous autumn (September and November), of early spring (March and April) and of July (**Figure 12H**). The temperature of the previous September and October was important for explaining year-to-year variations in the widths of xylem increments at RN (**Figures 12C,F,I**). Furthermore, February maximum and March minimum temperature was significant for tree-ring

and earlywood widths, whereas July maximum temperature negatively affected latewood widths at this site.

## Discussion

#### Xylem Cell Characteristics

To cope with different weather conditions, trees have evolved phenological, morphological, and physiological adaptations, which often arise from specific patterns of cambial activity and cell differentiation and are responsible for changes in xylemhydraulic conductivity and vulnerability to cavitation (Eilmann and Rigling, 2012). We showed that xylem anatomical parameters generally varied among the sites but not among the years, indicating a fairly stable structure at a given location with almost no year-to-year variation. Spruce at the lower elevation PA had on average the widest xylem increments and narrowest at the Czech site RN. These differences could not be attributed to

climatic factors as seen from climate-cellular relationship in xylem and phloem for the period 2009–2011. This confirms our previous speculations that local adaptation may also play a decisive part in the strategy of spruce for adapting the structure of wood and phloem increments to function optimally in local conditions (Gricar et al., 2014b ˇ ). Further, earlywood proportion was the highest in spruce at the Alpine site ME. The number of earlywood cells was positively influenced by climatic conditions in late autumn and winter. Because photosynthesis is possible in conifers in winter (Guehl, 1985), high temperature during the winter period, in combination with a sufficient water supply, could promote growth and improve carbon storage in the following year (Battipaglia et al., 2009; Lebourgeois et al., 2010). In addition, high winter or early spring temperatures may trigger cambial reactivation (e.g., Begum et al., 2013).

The average lumen dimension of earlywood tracheids was largest at ME, whereas the lumen diameter of initial tracheids was comparable at all sites. The lumen dimension of earlywood tracheids was affected by precipitation in the previous October and current June, and by minimum temperature in February and May. Tracheid enlargement is irreversible and drought sensitive processes, and is largely determined by the ability of the cell to generate and maintain positive turgor (Hölttä et al., 2010). Thus, the radial diameter of tracheids is directly affected by changes in water availability during the period of cell enlargement (e.g., Abe

et al., 2003), but may also be controlled by temperature months before the tissue is formed (Eckstein, 2013). In addition, the cell size can be influenced by numerous factors besides limited water availability: period of cell enlargement, the limitation of assimilates and a reduced synthesis of growth regulators, which are essential for cell enlargement (Eilmann et al., 2006). A higher earlywood proportion indicates the effective water use of trees (Domec and Gartner, 2002). Although the relative proportion of earlywood decreased with tree-ring width, the absolute number of earlywood cells was higher in wider rings (even if statistically not significant), which would explain the smaller values for the average lumen diameter of earlywood cells in spruce at PA, but similar values for initial tracheids. The more or less constant width of earlywood suggests its priority for the tree; a conduction function over a mechanical one. As stressed by Anfodillo et al. (2013), variation of wood traits is dependent on a tree's size. The conduit diameter is thus strictly related to tree height; in a basipetal direction, it is gradually wider. The distance from the stem apex is therefore crucial for comparison of the conduit diameters among different trees and sites. The trees selected for our study were already mature, with comparable heights, so the hydraulic parameters were probably not affected by changes in tracheid size.

Compared to earlywood width, latewood width, as well as the thickness of cell walls in latewood, seemed to be more variable, whereas the average lumen dimension of latewood tracheids was stable. Only lumen dimension of terminal latewood tracheids was affected by climatic conditions. A decreased latewood proportion in narrower rings would negatively affect the density and mechanical properties of wood but perhaps the need for additional strength provided by latewood becomes less crucial as

the stem increases in diameter (Rao et al., 1997). Similarly, Park and Spiecker (2005) observed that cell parameters in earlywood changed very little from site to site, whereas those in latewood were more influenced by site conditions. Trees from a warm-dry site also had more latewood cells and substantially thicker cell walls, whereas those from a cool-humid site had larger earlywood cells. Authors have ascribed these differences to the hydraulic adaptation mechanisms of trees to given site conditions (Park and Spiecker, 2005).

The wall thickness of the tracheids in the annual xylem ring followed a Weibull function pattern, indicating that terminal latewood tracheids did not have the thickest cell walls. Furthermore, the cell walls of the terminal latewood tracheids were thinnest in spruce from the lower site PA, where cambial cell production ceased later than at the other two sites (Gricar ˇ et al., 2014b). Consequently, development of the last formed cells also finished later (Gricar ˇ et al., 2005). The end of final cell maturation was thus basically proportionally related to the date of the cessation of cambial activity, and inversely related to the elevation. It has been previously reported for treeline Norway spruce that maximum wood density (and the thickest cell walls) might not be located exactly in the terminal cell row of a tree ring, because lignification of the secondary cell wall in terminal latewood tracheids is also influenced by the climatic

conditions, particularly in September (Gindl et al., 2000). It could be speculated that earlier cessation of cambial cell production would lead to thicker cell walls of terminal latewood tracheids because their development would occur earlier in autumn, when the temperature is higher. We found that the wall thickness of latewood cells were mainly influenced by maximum temperature and precipitation of the previous November, and by minimum temperature in July and precipitation in October. It is known that turgor, which is closely linked with water status in a tree (Steppe et al., 2015), has not only an impact on cell expansion, but also on biosynthesis of secondary cell wall (Proseus and Boyer, 2006). On the other hand, anatomical variations in xylem depend on rate and duration of differentiation processes (Skene, 1972), which determine the amount and properties of wood (Hölttä et al., 2010). Thus, both water and carbon affect seasonal radial growth resulted in wood anatomy, which influence xylem hydraulic conductivity and cavitation vulnerability (Steppe et al., 2015).

The thickest cell walls in earlywood and latewood were found in spruce at the lower site PA, with the widest tree rings. According to Hacke and Sperry (2001), vulnerability to cavitation depends on the mechanical strength of the conduits, so tracheids in drier environments usually have thicker cell walls. However, Gindl et al. (2001) measured the lignin content in tracheids of Norway spruce from two altitudes; 580 and 1260 m,

and observed that the low-elevation trees had wider growth rings, with thicker cell walls. In contrast, the lignin content was negatively correlated to the cell wall thickness. The authors hypothesized that trees growing at higher altitudes compensate for the thinner cell walls with an increased lignin content, which helps to maintain the mechanical integrity of the xylem (Gindl et al., 2001).

Even relatively small differences in the xylem structure of spruce from three temperate regions can have a considerable influence on the annual hydraulic properties of the tree ring (Tyree and Zimmermann, 2010). Namely, small increases in radial conduit diameter result in large changes in conductivity due to the direct relationship between flow and radius to the fourth power (Tyree and Zimmermann, 2010). Moreover, other anatomical characteristics of tracheids, such as their length (Mäkinen et al., 2008) and pit structure (Hacke and Jansen, 2009) may have further contributed to regulation of the hydraulic tradeoff. The differences could be an adaptation of trees to function optimally in local environmental conditions (Bryukhanova and Fonti, 2013).

#### Phloem Cell Characteristics

As found for the widths of the annual phloem increment (Prislan et al., 2013; Gricar et al., 2014b ˇ ), the dimensions of the sieve elements also seem to be predominantly site-specific characteristics in Norway spruce. In general, phloem anatomical


parameters varied among the sites but were fairly uniform among years. The number of late phloem cells as well as their radial dimension were mainly affected by precipitation in late autumn and maximum temperature in winter. The number of early phloem cells and dimension of all phloem cells were also positively influenced by precipitation in April. In addition, radial dimension of initial early phloem sieve cells was determined by maximum temperature or minimum in the period January– March. Favorable conditions in late autumn and winter positively affect phloem growth and carbon reserves in the following year (Lebourgeois et al., 2010).

Similarly as in wood, late phloem was also more variable than early phloem. These findings confirm again our previous assumptions that stable phloem formation patterns and its structure, as previously repeatedly reported, is found only in trees

of similar age, position in stand, vigor and vitality, which are growing in similar environments (Gricar et ˇ al., 2014a). As in xylem, the conductive capacity of phloem cells is not only affected by their anatomical structure, such as conduit size and number, but also by sieve pore size and frequency along the pathway (Mullendore et al., 2010). Jyske and Hölttä (2015) reported that phloem conduits were slightly narrower than xylem conduits. In our case, that was similar only for the early part of the annual rings, whereas in the increments formed in the second part of the growing season, it was just the opposite. Thus, tracheids in earlywood were on average 12–30% wider and in latewood 30–50% narrower than sieve cells in early and late phloem, respectively. These ratios were different among the sites but uniform among the years. There was no linkage between the radial lumen size of initial and terminal cells in xylem and phloem cells at an individual location. Thus, the variation in the structure of annual xylem and phloem increments in Norway spruce clearly shows that plasticity in the seasonal dynamics of cambial cell production and cell differentiation exists on both xylem and phloem sides. Xylem has different requirements for mechanical and hydraulic safety (Sperry et al., 2008), reflecting in larger functional areas and more rigid cell walls than in phloem. This results in lower conduit frequency per unit of conducting area but in higher mechanical support (Jyske and Hölttä, 2015). In xylem, sap transport usually occurs over several sapwood rings, with declining conductivity toward the inner sapwood (Spicer and Gartner, 2001), whereas phloem sieve elements function for only one to two growing seasons. Tree survival therefore depends on the yearly formation of new phloem to maintain and extend translocation pathways for photosynthates and biomolecules (Evert, 2006). However, the relevance of the annual hydraulic plasticity of the xylem tissue on future performance needs to be adjusted to its relative contribution to the whole sapwood and not only to the outermost annual increment (Jyske and Hölttä, 2015). In addition to sieve cells (and Strassburger cells), ray and axial parenchyma networks constitute annual phloem increments, which potentially serve as an important carbohydrate and water reserve (Rosell et al., 2014). Since the variability of late phloem is higher than that of early phloem, a variation in phloem annual increments can greatly affect the amount of axial parenchyma in the phloem. No such information exists to date. However, it deserves more in depth investigation in order to understand better the role of parenchyma cells on the whole functioning of trees in different environments.

#### Climate-growth Relationship

Analysis of the climatic impact on the radial increment of Norway spruce from the three studied locations revealed sitespecific growth sensitivities to local climatic conditions. Radial growth of spruce was on all three sites only partly dependant on climatic factors, including maximum and minimum temperature and precipitation. Tree-ring widths of spruce growing in temperate forests under average climatic conditions are known to respond less intensely to climatic variation than trees growing in extreme conditions (Mäkinen et al., 2003). Local site conditions, tree age, tree competition and tree vitality, as well as different forest management practices can be more important drivers of local-scale growth trends than the regional climate signal (Levanic et al., 2009 ˇ ).

The most important climatic parameters affecting the radial growth of spruce at the Slovenian lower site PA was found to be a positive effect of a warm January and negative affect of a hot summer (June–July). Spruce at the Alpine site ME showed a significant positive response to March precipitation, while spruce at the Czech site RN mainly negatively responded to a warm previous September and positively to a warm late winter (February–March). Other research groups have reported that the most important climatic parameters affecting spruce growth throughout Europe are summer temperatures, and the summer and autumn temperatures of the previous year, especially July and September (e.g., Savva et al., 2006; Bošel'a et al., 2014). However, the response to the same climatic driver can work in opposite directions at different sites. For spruce from higher elevations, the response to summer temperature is generally positive, since temperature is a limiting factor for growth, whereas at lower sites, the effect of summer temperature is negative, because it is usually related to a decrease in precipitation, a limiting factor for low-altitude spruce (e.g., Levanic et al., 2009; Lebourgeois ˇ et al., 2010). In our study, interestingly, growth was more weakly correlated with precipitation than with temperature. In addition, precipitation in spring seemed to be mainly important for the Alpine site ME. This could be explained by the fact that none of our sites is precipitation limited. Even at the Czech site RN with annual precipitation of only 600 mm, rain falls predominantly during the growing period.

Earlywood widths showed a similar relation to climatic conditions as tree-ring widths. Nevertheless, earlywood width was generally less dependent on climatic conditions than that of latewood. Some authors have ascribed the low dependence of earlywood width on climate to more endogenous control of its formation (Bošel'a et al., 2014). The latewood width at all three sites was negatively affected by a hot summer and also to the climatic conditions of the previous autumn. In terms of the latewood width, trees at the Alpine site ME showed the highest growth dependence on climatic conditions in July, whereby wet and less hot conditions positively affected the amount of latewood. This is probably mainly related to prolongation of wood (latewood) formation in this period. Our findings are in contrast with previous observations (e.g., Levanic et al., ˇ 2009) but, as already mentioned, the sites selected for this study were sufficiently supplied with precipitation. In addition, microtopography, soil properties and rock type, which differ among the selected sites, influence runoff, distribution and water movement in the soil. Acid granodiorite rock type at RN, for example, is able to retain more water than dolomite at PA or limestone at ME.

Other studies have also highlighted a highly negative response of tree-ring width to the summer temperature of the previous year (e.g., Andreassen et al., 2006; Bouriaud and Popa, 2009). The authors explained this relationship by the stimulation of cone production after a warm summer for trees that have reached maturity (Andreassen et al., 2006) or high respiration rates triggered by high temperatures, which would decrease the carbohydrate reserves available for needle development and growth initiation during early phases of the growing season (Bouriaud and Popa, 2009).

## Conclusions

In temperate regions, where climatic conditions are favorable for growth of Norway spruce, variation in the structure of xylem and phloem increments is mainly observed at the site level rather than on a temporal scale. In addition, xylem and phloem tissues formed in the first part of the growing season seem to be more stable in structure, which indicates their priority over latewood wood and late phloem for tree performance. Analysis of the longterm climatic impact on the radial increment of Norway spruce from the three studied locations confirmed site-specific growth sensitivities to local climatic conditions. Earlywood widths showed a similar relation to climatic conditions as tree-ring widths, although earlywood width was generally less dependent on climatic conditions than that of latewood. Radial growth was in general less dependent on precipitation than on temperature, which could be explained by the fact that none of the study sites is precipitation limited. Radial growth of spruce was on all three sites only partly dependant on climatic factors, which confirms the strategy of spruce to adapt the structure of wood and phloem increments to function optimally in local conditions.

Ability of plants to adapt to the surrounding environmental conditions is important since climate is always changing, both in terms of long-term average values and in terms of frequency and severity of extreme events. Adaptation is thus characteristic of xylem and phloem cell parameters and increment widths. Knowledge of intra-annual radial growth could provide valuable information on adaptation strategies of spruce to local environmental conditions and will help to improve interpretation of the relevance of such high annual plasticity of secondary tissues on future tree growth.

## Author Contributions

JG—developed the concept of the paper, wrote the paper, and together with KC performed wood-anatomical analysis for the ˇ Slovenian sites; PP, ML—carried out the statistical analysis, wrote the statistical parts of the paper and prepared the figures; VG, JH, HV—performed wood-anatomical analysis for the Czech site. KCˇ was responsible for the Slovenian sampling, section preparation and climate data. All authors discussed and commented on the manuscript.

## Funding

This work was supported by the Slovenian Research Agency, young researchers' program and programs P4-0015 and P4-0107, by the Spanish Science and Innovation Ministry (MICINN), the ELENA program (CGL2012-31668), by the FEDER program of the European Union and by the European Social Fund and the state budget of the Czech Republic, Project Indicators of Trees Vitality Reg. No. CZ.1.07/2.3.00/20.0265. The cooperation among the international partners was supported by COST Action FP1106, STReESS.

#### Acknowledgments

The authors gratefully acknowledge the help of Marko Beber and the Slovenian Forest Service, Milko Detmar and

#### References


Metropolitana d.o.o., as well as Luka Krže, Maks Merela, Marko Željko, Gabriela Vichrová, Jaroslav Kratochvíl, and Tomáš Kratochvíl for their immense help in the field and in the laboratory. We thank Martin Cregeen for language editing. The long-term climatic data for Rájec-Nemˇ cice ˇ site were provided by the Czech Hydrometeorological Institute.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Griˇcar, Prislan, de Luis, Gryc, Hacurová, Vavrˇcík and Cufar. ˇ This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Tree-Centered Approach to Assess Impacts of Extreme Climatic Events on Forests

Ute Sass-Klaassen<sup>1</sup> \*, Patrick Fonti <sup>2</sup> , Paolo Cherubini <sup>2</sup> , Jožica Gricar ˇ 3 , Elisabeth M. R. Robert 4, 5, 6, Kathy Steppe<sup>7</sup> and Achim Bräuning<sup>8</sup>

<sup>1</sup> Forest Ecology and Forest Management Group, Wageningen University, Wageningen, Netherlands, <sup>2</sup> Landscape Dynamics Unit, Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Birmensdorf, Switzerland, <sup>3</sup> Department of Forest Yield and Silviculture, Slovenian Forestry Institute, Ljubljana, Slovenia, <sup>4</sup> CREAF, Cerdanyola del Vallès, Spain, <sup>5</sup> Laboratory of Plant Biology and Nature Management, Vrije Universiteit Brussel, Brussels, Belgium, <sup>6</sup> Laboratory of Wood Biology and Xylarium, Royal Museum for Central Africa, Tervuren, Belgium, <sup>7</sup> Laboratory of Plant Ecology, Department of Applied Ecology and Environmental Biology, Faculty of Bioscience Engineering, Ghent University, Ghent, Belgium, <sup>8</sup> Department of Geography and Geosciences, Friedrich-Alexander-University Erlangen-Nuremberg, Erlangen, Germany

Keywords: climate change, future forests, tree, mechanistic understanding, structure-function relationships, long-term monitoring, intra-annual resolution, resilience

## INTRODUCTION

#### Edited by:

Judy Simon, University of Konstanz, Germany

> Reviewed by: Martin De Luis, University of Zaragoza, Spain

> > \*Correspondence: Ute Sass-Klaassen ute.sassklaassen@wur.nl

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

> Received: 25 May 2016 Accepted: 06 July 2016 Published: 21 July 2016

#### Citation:

Sass-Klaassen U, Fonti P, Cherubini P, Gricar J, Robert EMR, Steppe K and ˇ Bräuning A (2016) A Tree-Centered Approach to Assess Impacts of Extreme Climatic Events on Forests. Front. Plant Sci. 7:1069. doi: 10.3389/fpls.2016.01069 A major task of our society is to manage forests in a way that their resources are preserved to meet future generation needs (Forest Europe et al., 2015). Current scenarios of climate change effects are making this task extremely challenging (Kirilenko and Sedjo, 2007). Climate shifts will impact forest vitality and affect goods and services forests provide, including carbon sequestration and climate change mitigation (IPCC, 2014). To guide sustainable forest management, forest researchers are asked to provide concrete answers about forest resilience in response to expected climatic trends, and extreme climatic events (Lindner et al., 2014). This is not an easy task, because responses of trees and forest ecosystems to environmental conditions are often non-linear and moreover vary on spatial and temporal scales (Smith, 2011; Anderegg et al., 2012; Reichstein et al., 2013). For instance, although drought is one of the most frequent and widespread climatic extremes affecting forests worldwide (e.g., Allen et al., 2010), the assessment of its impact on future forests is currently under intense debate. Mechanisms behind tree growth and mortality are complex (McDowell et al., 2008, 2011; Fatichi et al., 2014; Anderegg et al., 2015; Meir et al., 2015). Besides strength or frequency of external factors, such as extreme events, also the tree's ability to resist and recover is relevant, which, in turn, is largely determined by intrinsic factors such as the tree's life stage, life history, and genetic characteristics.

In this paper, we advocate for a tree-centered approach. By providing an improved mechanistic understanding of physiological and growth responses of trees growing under various conditions we can define the tree's capacity to respond to external stress factors. This concept can valuably contribute to the debate on how to shape future forests toward resilient forest ecosystems.

#### A TREE-CENTERED APPROACH

Current spatiotemporal simulations on future forest growth responses to changing climate conditions are performed with dynamic global vegetation models (DGVMs; Wullschleger et al., 2014). These models—usually generalizing tree species as plant functional types (PFTs)—provide valuable descriptions of the evolution of natural vegetation at a grid cell level under several climate scenarios. Such approaches are powerful in assessing growth responses related to the interaction between vegetation and atmosphere (including anthropogenic impact). However, although first steps toward representation of tree species, size classes, and forest structure in a DGVM were recently made (e.g., Naudts et al., 2016) they often lack to explain the variability between and within species, and often do not adequately explain growth responses (Fatichi et al., 2014) under varying site conditions, and to climatic extremes (Anderegg et al., 2015). These aspects are however extremely relevant to evaluate plasticity of tree individuals and tree species and the resilience of forests under changing climatic conditions, especially considering changing frequencies and intensities of climatic extremes (Reyer et al., 2013).

The tree-centered approach proposed here considers the individual tree as main source of information for understanding variability in growth responses. Comprehensive investigations using well-selected trees growing under different environmental conditions foster a better understanding of projected largescale forest responses to changing climate. In comparison to generalizing approaches using PFTs, the tree-centered approach yields information with less spatial coverage but with the potential to convey more details on specific tree responses to a given climatic factor. This knowledge complements other approaches and can for instance support forest managers in tree species and/or provenance selection to better prepare specific forest stands to cope with expected challenges.

## FOUR IMPORTANT ELEMENTS

The incentive for the tree-centered approach is gaining a process-based understanding on tree responses to changing environmental conditions on temporal scales varying from shortterm responses to climatic extreme events to long time periods matching the life cycles of tree populations. This can be achieved through an ensemble of observational studies on a selection of trees from different species and life histories, growing in diverse settings (forest types, species composition, successional stages, management regimes), and exposed to different climates and extreme climatic events. Establishing such a model framework requires the following elements:


(iv) Comparative studies in selected sites experiencing extreme events, long-term manipulation studies, and experiments, including e.g., provenance trials are finally necessary to test and validate the model framework to conditions expanding even far outside today's natural range.

## IMPLEMENTATION OF THE TREE-CENTERED APPROACH

The COST Action STReESS—a 4-year European framework initiative to promote networking among researchers of several plant-research disciplines to study tree responses to extreme events—is demonstrating the potential of such an integrated bottom-up approach. Starting from an enhanced understanding of the physiological processes behind wood formation, the STReESS Action established a modular process-based approach, which will eventually result in a model framework for explaining tree responses to climate extremes (**Figure 1**).

The COST Action STReESS contributed to the following main interlinked elements of the concept along the causal path: environmental trigger—structure—function—performance.

#### Long-Term High-Resolution Monitoring

Efforts performed for long-term high-resolution monitoring of tree physiology, tree growth together with contemporary site, and climatic factors, allows quantifying causal relationships between external triggers and tree physiological and growth responses (Steppe et al., 2015, 2016). Knowledge gained from such real-time measurements has resulted in process-based plant models in which the mechanisms underlying diel water and carbon transport and their tight coupling have been integrated (see review by De Swaef et al., 2015). In addition, worldwide xylogenesis and dendrometer databases have been compiled during the STReESS Action to assess global response patterns to various specific climate and site conditions (Rossi et al., 2013) and to gain insight into processes involved in wood formation (Cuny et al., 2014, 2015; Steppe et al., 2015).

## Linking Structure to Function

Linking structure to function is vital to understand the impact of climate-caused changes on wood formation dynamics and wood structure, which strongly influences the water, and carbon household and determines actual, and future tree survival chances and growth performance. Recent studies have highlighted that tree morphological properties and related e.g., hydraulic safety properties (Delzon and Cochard, 2014) vary within individuals, among provenances and species, or along environments and stress gradients. Relevant parameters include phloem to xylem ratio (Gricar et al., 2015; Jyske et al., 2015 ˇ ) and connection (Pfautsch et al., 2015), as well as xylem and phloem-cell characteristics in stems (Anfodillo et al., 2012; Olano et al., 2013; Carrer et al., 2015; Gricar et al., 2016 ˇ ), branches (Salmon et al., 2015), and roots (Brunner et al., 2015). Such characteristics reflect functional adjustments in the tree's structures in response to changing environmental conditions. In turn, these adjustments also form a legacy by influencing future tree performance and hence reflecting the acclimation capacity

of trees and tree species (Lachenbruch and McCulloh, 2014; Rosner et al., 2016a,b; Sterck et al., 2016; Anfodillo et al., in review).

#### The Long-Term Perspective

The continuous adjustments in wood structure are permanently stored in tree rings either as annual variations in wood-anatomical characteristics, such as cell-wall thickness, cell size, or tissue percentage, or in case of extreme climatic events, as obvious wood-anatomical markers (Battipaglia et al., 2016; Bräuning et al., 2016). These characteristics enable the use of dated tree rings to reconstruct how trees have been growing and functioning in the past (Fonti and Jansen, 2012), and consequently reflect the resilience and acclimation strategies of trees in a changing climate (e.g., Breda et al., 2006). Typical cases of wood-anatomical markers considered in the STReESS action are flood rings (Copini et al., 2016), missing rings and dark rings (Novak et al., 2016), and intra-annual density fluctuations (IADFs). IADFs comprise an abrupt change in wood density in a given tree ring (Nabais et al., 2014; Campelo et al., 2015) and have been demonstrated to hold valuable high-resolution information on the timing of past droughts in Mediterranean conifers (Battipaglia et al., 2016; De Micco et al., 2016; Zalloni et al., 2016). Recently, these approaches became increasingly applicable due to methodological advances in efficiently quantifying cell-lumen size, cell-wall thickness or specific cell types as resin canals, and parenchyma cells (von Arx and Carrer, 2014; von Arx et al., 2016). The enormous potential of wood-anatomical characteristics and markers lies in the information they provide on the exact timing of climatic constraints (Fonti et al., 2010; Rathgeber et al., 2016) and in the possibility to evaluate consequences that these constraints have for xylem and phloem functioning. This creates the bridge between the elements 1 and 2 and allows to uniquely integrating a long-term functional perspective on the current assessment of tree-growth responses to the environment.

## Field Studies, Provenance Trials, Manipulation Experiments

The limit of spatial coverage in the high-resolution monitoring approach (element 1) together with the need for testing the effect of future climate-change scenarios on tree performance can be waged by comparative field and experimental studies to target specific species, provenances, or environmental conditions. De Luis et al. (2014) illustrated the potential of using tree-ring networks across species distributions to assess local adaptation and plasticity of Pinus halepensis in the Mediterranean basin. Another approach is a trans-European cross experiment where drought mortality-resistance of European beech provenances has been related to water availability and to the origin of the beech seedlings (Pšidová et al., 2015; Bolte et al., 2016). Hence, provenance trials have revealed a genetic control of wood structural properties (Eilmann et al., 2014; Nabais et al., in review), although investigations of other tree species are needed to fully evaluate the importance of genetic preadaptation to future climatic conditions. Such kinds of studies prove the added value of targeting specific situations, e.g., natural conditions or manipulated experiment, for developing, and validating the model framework.

#### Integration into a Model Framework

The four elements can be captured and integrated into process-based tree models. The first steps of this integration have already been achieved. For example, diel stem-size variations and sap-flux densities combining real-time and highresolution measurements of tree functioning under on-site environmental conditions have allowed to link environmental triggers (climate events) with the resulting tree growth and performance (e.g., Steppe et al., 2006; De Schepper and Steppe, 2010; Hölttä et al., 2010; Schiestl-Aalto et al., 2015). This means that instant responses of a tree to a drought or a heat wave (Teskey et al., 2015) can be readily assessed, and changes in its water and carbon budget quantified. Effort still needs to be invested to implement parameterizing of long-term climate-growth responses or the peculiarity of species and proveniences into the existing models to finally come up with estimates for tree plasticity, acclimation potential of tree species, and ultimately resilience of forests.

## CONCLUSION AND PERSPECTIVE

There is a fundamental difference between generalized PFTbased approaches (i.e., DGVMs) and tree-centered approaches. While PFT-based approaches perform spatially explicit "scenarios" of future global responses, the interdisciplinary process-driven tree-centered approach has potential to also provide practical support for local management decisions based on a solid understanding of tree functioning under specific site conditions. The COST Action STReESS has improved our understanding on the variability of responses to climate trends and extreme events. After 4 years of collaboration, the consortium has collected indications of the usefulness of such an integrated approach with continuous "methodological" development, and creativity. Through the integration of monitoring studies (e.g., time series of dendrometers, wood formation, and forest inventories), dendrochronological approaches, manipulation experiments (e.g., induced drought stress), and process-based models (e.g., at the cell, plant, or vegetation level) there is potential to collect valuable characterization to build process-based tree models accounting for variability between species, provenances, sites, and climatic events. This will contribute to unraveling a large set of yet unanswered questions related to processes of tree mortality (e.g., McDowell et al., 2011) or (mal)-adaptation (e.g., Martinez-Meier et al., 2008). Such a process-based approach will help to reduce uncertainty on tree performance under future environmental conditions.

Actual plans include the extension of the twittering-tree network for further development of a near real-time detection of tree processes and environmental impacts (Steppe et al., 2016) as well as the extension of global data networks and harmonization of protocols for high-resolution growth measurements (e.g., dendrometer, xylogenesis).

Despite still many implementation challenges ahead, we believe that the tree-centered approach offers an additional opportunity to assess forest management sustainability at the profit of the whole society.

## AUTHOR CONTRIBUTIONS

All authors developed the concept and structure of the manuscript. USK and PF wrote the first draft of the manuscript. EMRR developed and designed the figure in cooperation with PF and USK. AB, EMRR, KS, JG, and PC authors accomplished and checked the first version and read and approved the submitted version.

## ACKNOWLEDGMENTS

We thank all COST STReESS participants for an inspiring 4-year period of fruitful collaboration. Our opinion paper as part of this

### REFERENCES


special issue is one product of the COST Action FP1106 STReESS besides the many other common research articles, review papers, book chapters and films. Moreover the COST Action has resulted in numerous collaborations, projects and project initiatives. EMRR is funded by the Research Foundation—Flanders (FWO, Belgium) and by the EU through a Marie Curie IF fellowship (No 659191). This work profited from support from the Swiss COST project D-STReSS.ch (C12.0100).


uncertainties, and what are the implications for forest management? Environ. Manag. 146, 69. doi: 10.1016/j.jenvman.2014.07.030


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Sass-Klaassen, Fonti, Cherubini, Griˇcar, Robert, Steppe and Bräuning. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Trait Acclimation Mitigates Mortality Risks of Tropical Canopy Trees under Global Warming

Frank Sterck<sup>1</sup> \*, Niels P. R. Anten2,3, Feike Schieving<sup>3</sup> and Pieter A. Zuidema<sup>1</sup>

<sup>1</sup> Forest Ecology and Forest Management Group, Wageningen University, Wageningen, Netherlands, <sup>2</sup> Centre for Crop Systems Analysis, Wageningen University, Wageningen, Netherlands, <sup>3</sup> Ecology and Biodiversity Group, Department of Biology, Utrecht University, Utrecht, Netherlands

There is a heated debate about the effect of global change on tropical forests. Many scientists predict large-scale tree mortality while others point to mitigating roles of CO<sup>2</sup> fertilization and – the notoriously unknown – physiological trait acclimation of trees. In this opinion article we provided a first quantification of the potential of trait acclimation to mitigate the negative effects of warming on tropical canopy tree growth and survival. We applied a physiological tree growth model that incorporates trait acclimation through an optimization approach. Our model estimated the maximum effect of acclimation when trees optimize traits that are strongly plastic on a week to annual time scale (leaf photosynthetic capacity, total leaf area, stem sapwood area) to maximize carbon gain. We simulated tree carbon gain for temperatures (25–35◦C) and ambient CO<sup>2</sup> concentrations (390–800 ppm) predicted for the 21st century. Full trait acclimation increased simulated carbon gain by up to 10–20% and the maximum tolerated temperature by up to 2◦C, thus reducing risks of tree death under predicted warming. Functional trait acclimation may thus increase the resilience of tropical trees to warming, but cannot prevent tree death during extremely hot and dry years at current CO<sup>2</sup> levels. We call for incorporating trait acclimation in field and experimental studies of plant functional traits, and in models that predict responses of tropical forests to climate change.

Keywords: carbon budget, climate change, functional plant trait, mechanistic plant model, optimization, plasticity, tropical forest, water relations

## INTRODUCTION

The effect of climate change on tropical forests is highly uncertain and subject to a heated debate (Körner, 2009; Lewis et al., 2009; Clark et al., 2010; Rammig et al., 2010; Corlett, 2011). One of the prominent concerns is the risk of large-scale tree mortality when trees are gradually pushed outside their current temperature envelop (Wright et al., 2009; Schippers et al., 2015), or confronted with extreme hot and dry years (Phillips et al., 2009; Schippers et al., 2015). Given the importance of tropical forest trees for the global carbon cycle and in heat and water vapor exchange with the atmosphere (Bonan, 2008), large scale mortality of trees may have enormous consequences for global climate and has been identified as one of the tipping points of the whole earth system (Cox et al., 2004).

#### Edited by:

Achim Braeuning, University Erlangen-Nuremberg, Germany

#### Reviewed by:

Zhenzhu Xu, Chinese Academy of Sciences, China Ze-Xin Fan, Chinese Academy of Sciences, China

> \*Correspondence: Frank Sterck frank.sterck@wur.nl

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 06 January 2016 Accepted: 20 April 2016 Published: 11 May 2016

#### Citation:

Sterck F, Anten NPR, Schieving F and Zuidema PA (2016) Trait Acclimation Mitigates Mortality Risks of Tropical Canopy Trees under Global Warming. Front. Plant Sci. 7:607. doi: 10.3389/fpls.2016.00607

Several global dynamic vegetation models have predicted the conversion of moist Amazonian forest to seasonal forests or savannah under the warming projected for the coming century (Cox et al., 2004; Malhi et al., 2009). On the other hand, scientists have pointed to two factors that may buffer trees against warming: higher photosynthesis and improved water-use efficiency at high ambient CO<sup>2</sup> (Lloyd and Farquhar, 2008), and the plastic physiological and morphological responses of trees to climate change (trait acclimation) (Galbraith et al., 2010; Corlett, 2011; Smith and Dukes, 2013). There is ample empirical evidence for increasing photosynthesis at high CO<sup>2</sup> (Körner, 2009) and there are some indications of increased water-use efficiency (Hietz et al., 2005), but the implications for tree growth and survival are uncertain (Rammig et al., 2010; Corlett, 2011; Zuidema et al., 2013) and subject to debate (Körner, 2009).

In spite of its potential importance in mitigating the negative impacts of warming on tropical forest trees, trait acclimation has so far remained notoriously understudied (Wright et al., 2009; Corlett, 2011). In a recent review on tropical forests and global warming, the importance of acclimation in long-lived canopy trees was emphasized, as "many individual trees alive today will still be living in 2100" (Corlett, 2011), implying that the degree to which tropical trees can acclimate to climate change will critically determine the future of tropical forests. Studies on acclimation responses of tropical trees conducted so far, suggest that they may acclimate to increased temperatures (Way and Oren, 2010) and drought (Metcalfe et al., 2010), and under certain conditions to elevated CO<sup>2</sup> concentrations (Körner, 2009). Yet, the potential effect of trait acclimation on tree carbon gain under climate change has not been quantified so far (Corlett, 2011) and is not incorporated in current dynamic global vegetation models, DGVMs (Galbraith et al., 2010; Huntingford et al., 2013). If trait acclimation mitigates negative effects of warming for trees and forests, it may increase maximum temperatures at which tropical trees survive with potential implications for the risks of tropical forest dieback.

#### A MECHANISTIC APPROACH

There is a dire need for a mechanistic approach to projecting tropical tree growth under climate change to quantify the suggested effects of warming, CO<sup>2</sup> fertilization and trait acclimation (Malhi et al., 2009; Corlett, 2011; Hirota et al., 2011; Cox et al., 2013, Huntingford et al., 2013). Here we show the potential contribution of acclimation in functional plant traits to carbon gain of tropical forest trees. We considered the acclimation of morphological and physiological features that are plastic on time scales of weeks to years (leaf photosynthetic capacity, crown total leaf area, stem sapwood area) and that affect tree growth and survival (Violle et al., 2007). We developed and used a mechanistic, mathematical, plant model that calculates the daily carbon gain of trees based on the hydraulic tree structure accounting for the acquisition, transport and transpiration of water, leaf stomatal coordination, and chemistry and temperature dependency of photosynthesis and respiration of C3-plants (**Figure 1**, Supporting Information Text S1; Sterck et al., 2011).

FIGURE 1 | A simplified representation of the climate impacts on the functioning of trees in the used mechanistic plant model. Climate-related variables (blue) drive photosynthesis and respiration (green) and determine the water balance (gray). The innovation is the acclimation of functional traits (orange), which is realized by optimizing trait values to maximize carbon gain while maintaining the water balance (see Supporting Information). The water balance component ensures that rates of transpiration, water uptake and water transport are equal.

We used the model to evaluate the impacts of temperature and ambient CO<sup>2</sup> on the net carbon gain of rainforest canopy trees. This approach is unique in the sense that it uses optimization theory to quantify the maximum possible contribution of trait acclimation to stimulate carbon gain, tree growth, or mitigate mortality risks under climate change scenarios (Dewar et al., 2009). We simulated trees that optimally acclimate their total leaf area, stem sapwood area and leaf photosynthetic capacity to maximize net carbon gain. Those traits are highly plastic in tropical trees and involve trade-offs because their positive impacts on gross photosynthesis are accompanied by higher respiration costs, higher production costs to replace leaves and sapwood (**Figure 1**, Sterck and Schieving, 2011) and more water loss (**Figure 1**, McMurtrie et al., 2008; Dewar et al., 2009).

We simulated trait acclimation and carbon gain of trees for a wide range of temperatures (25–35◦C) and atmospheric [CO2] concentrations (390–800 ppm) which are projected for tropical forests in this century (Knutti et al., 2005, IPCC, 2007; Supporting Information Figure S1, IPCC, 2013). We simplified this first exploratory analysis by neglecting more complex diurnal or seasonal patterns in atmospheric conditions, or soil conditions. These sets of simulations allowed us to provide a first bench mark figure for the maximum attainable effect of trait acclimation on tropical canopy trees exposed to gradual warming that occurs along with rising [CO2]. In our simulations, non-acclimating trees possess traits optimized at 25◦C. We kept relatively humidity constant at 70% which results in an increased transpiration demand at higher temperatures. In our main set of simulations we kept soil water potential at 0 MPa for the entire range of climatic conditions, thus assuming that soil water was not limited. In the Supporting Information Figure S2, we

also presented simulations with drier soil conditions (soil water potential of −0.5 MPa).

#### MODEL STRUCTURE AND PROCESSES

We used a mechanistic plant model (Sterck and Schieving, 2011; Sterck et al., 2011, 2014) to simulate the carbon gain and survival of virtual tropical canopy trees in response to different climate scenarios for ambient CO2, temperature and water stress, under the assumption that trees maximize carbon gain by acclimating in their leaf area index, stem sapwood area and leaf photosynthetic capacity. The plant model captures the aboveground structure and physiology of a tree. The modeled trees consist of a cylindrical crown, with a given top height, crown bottom height, crown radius, sapwood area and total leaf area. Leaves are assumed to be uniformly distributed within the crown. The crown is assumed to have an average nitrogen concentration per unit leaf area, and this nitrogen is distributed optimally over the crown in parallel to the light gradient in the canopy following predictions made by big-leaf models (Sellers et al., 1992). Crown photosynthesis is calculated from a biochemical photosynthesis model (Farquhar et al., 1980), a stomatal conductance model and water transport model (Tuzet et al., 2003), temperature dependencies of photosynthetic (protein) and respiratory processes for C3-plants in the range of 25–35◦C (Bernacchi et al., 2001, 2003), respiratory processes of sapwood for tropical forest trees (Meir and Grace, 2002), and the scaling procedures from leaf to whole tree level (Sterck et al., 2011). A detailed mathematical description of the model is included in Sterck et al. (2011) and in the Supporting Information Text S1. We parameterized the model for a 30 m tall canopy tree exposed to open sky light conditions. Photosynthetic parameters and their temperature responses were fixed for typical C<sup>3</sup> plant values (Bernacchi et al., 2001, 2003), and mass based sapwood respiration rate, turnover rate of sapwood and leaves, specific sapwood conductivity, and stomatal sensitivity to leaf water potential were set at constant values (Supporting Information Table S1 for complete list of parameter values). The constant trait values reflect a scarcity in quantitative information on acclimation and lack of understanding in underlying tradeoffs. Given the potential impact that these traits have on plant responses to climate (Choat et al., 2012; Scoffoni et al., 2012), we call for research on their possible acclimation.

#### MODELING OPTIMIZATION PROCEDURE

We used an optimization framework to determine optimal values for crown leaf area index, stem sapwood area and leaf photosynthetic capacity that maximize carbon gain, and thus mimic the maximum contribution of acclimation in these traits to the carbon gain and survival of trees under climate change. This approach was used to simulate trees under future climate change. We determined the acclimation (i.e., through phenotypic plasticity) of trees to climatic changes such that their net carbon gain is maximized, to set a benchmark for the potential contribution of trait acclimation to future tree performance. This optimization takes the most important constraints on tree carbon gain into account: the co-limitation of photosynthesis by carboxylation and electron transport processes through leaf nitrogen partitioning between these two photosynthetic processes, and steady state for water uptake, transport and loss through coordination of stomatal conductance by tuning the leaf water potential (for details, see Sterck and Schieving, 2011). The traits which we allowed to acclimate are: crown leaf area index, stem sapwood area and leaf photosynthetic capacity (i.e., the average photosynthetic nitrogen mass per leaf area). The values of other functional plant variables were predicted by the implemented processes and include, among others, the crown water potential, intercellular CO<sup>2</sup> concentration, stomatal conductance, transpiration, nitrogen partitioning between electron transport and carboxylation processes, gross photosynthesis, respiration costs and turnover costs (Supporting Information Table S2 for complete list). Overall, this optimization approach allows us to track emergent multiple trait patterns from underlying principles. Trade-offs occur because increasing values of traits that enhance gross photosynthesis also tend to entail larger respiration costs, as well as higher production costs to replace leaves and sapwood (turnover costs).

#### MODEL ASSUMPTIONS

The modeling approach is based on a number of assumptions (see Sterck and Schieving, 2011): (i) Leaf nitrogen is optimally partitioned within leaves such that carboxylation and electron transport co-limit photosynthesis. (ii) The crown is characterized by an average leaf water potential ψ<sup>l</sup> and average intercellular CO<sup>2</sup> concentration c<sup>i</sup> inside leaves, calculated over the average vertical and horizontal distance from stem base to leaf in the crown (Sterck et al., 2011). (iii) Leaf temperature equals the air temperature. This approach does not account for vertical gradients in ψ<sup>l</sup> , c<sup>i</sup> or leaf temperature within the crown, but allows us to scale from leaf to whole tree model and solve that whole tree model for optimized values of traits. (iv) There is steady state for transpiration and stem water flow, which is a reasonable assumption on the 24-h time scale considered here. We assumed that a canopy tree is hydraulically limited in photosynthesis, replaces leaves and sapwood in steady state, and has a 15% surplus of carbon budget left for net vegetative or reproductive growth. We started all simulations from the same set of default parameters (Supporting Information Table S1), but included temperature dependencies of typical C<sup>3</sup> plants for photosynthetic and respiratory parameters (Sterck et al., 2011, Supporting Information Table S2).

#### POTENTIAL ACCLIMATION IMPACTS

We simulated carbon gain of tropical forest canopy trees in response to different combinations of ambient atmospheric CO<sup>2</sup>

and temperatures. An initial comparison of carbon gain for acclimating and non-acclimating trees reveals higher values for acclimating trees for the full range of temperatures and at two CO<sup>2</sup> concentrations (**Figures 2B,C**). Yet, at increasing temperatures, acclimating trees could not avoid declines in carbon gain (**Figure 2B**). These results are consistent with empirical observations that growth of tropical forest trees is reduced both during extremely warm and dry years (Feeley et al., 2007; Phillips et al., 2009). The results are consistently found at both current ambient and double ambient CO<sup>2</sup> level (**Figures 2B,C**), with higher values in the latter case reflecting a potential CO<sup>2</sup> fertilization effect.

A conspicuous result of our simulations is that at current CO<sup>2</sup> levels acclimation responses to temperature allowed trees to increase the temperature at which carbon gain approaches zero by an estimated 2◦C (**Figures 2A–C**). This result was robust for different scenarios of vapor pressure differences between atmosphere and leaves and different soil water conditions (Supporting Information Figure S2). This result suggests that acclimation will mitigate the effects of warming on changes in the geographical distribution of tropical forests. So far, predictions of the impact of climate change in tree performance and distribution were based on either climate envelops by current distribution patterns (Wright et al., 2009), leaf models (Lloyd and Farquhar, 2008), plant models (McMurtrie et al., 2008) or vegetation models (Sitch et al., 2008). None of these approaches considered the role of acclimation in functional traits such as total leaf area, stem sapwood area and leaf photosynthetic capacity based on biophysical principles, as we did here. Those studies may thus have overestimated the negative impact of warmer and drier conditions on tropical tree distributions. Our prediction that acclimation may extend the climate envelop of tropical forest canopy trees by a maximum of 2◦C is particularly relevant at the lower range of ambient CO<sup>2</sup> levels, and suggest that acclimation increases the resilience of tropical trees to warming predicted for the coming decades.

#### ACCLIMATION AND CO<sup>2</sup> OFFSET WARMING IMPACTS

The predicted decrease in carbon gain with rising temperature was robust over the full range of simulated CO<sup>2</sup> levels (**Figure 2A**). Similarly, the predicted increase in carbon gain at

elevated ambient CO<sup>2</sup> was robust over the full range of simulated temperatures (**Figure 2A**). Our simulations suggest that the carbon gain of acclimating trees will gradually increase over time, when considering the predicted coupled changes in ambient CO<sup>2</sup> and temperature (**Figure 2A**, solid line) and their confidence interval (i.e., based on the uncertainties in IPCC scenarios, dashed lines, **Figure 2A**). This suggests that for this century the increasing ambient CO<sup>2</sup> levels together with trait acclimation will more than offset the negative impacts of warming (**Figure 2A**). This finding is in line with the predicted responses of leaflevel photosynthesis (Lloyd and Farquhar, 2008) and observed elevated photosynthesis and leaf sugar loads under elevated ambient CO<sup>2</sup> (Körner et al., 2005). When considering tree carbon gain, our results suggest that the negative impacts of temperature and water stress on tree carbon gain will likely be more than offset by the positive impacts of trait acclimation and increase in ambient CO<sup>2</sup> under the 2–4◦C warming scenarios predicted for the end of this century. The predicted resilience of tropical rain forest trees by acclimation and CO<sup>2</sup> fertilization is consistent with the existence of species rich tropical forests in warm, CO<sup>2</sup> rich episodes 57 million years ago, when rainfall patterns were not affected (Jaramillo et al., 2010). Yet, the predicted increase in carbon gain needs to be interpreted with care. The doubts about a direct translation of carbon gain into tree (biomass) growth (Körner, 2009), the highly variable biomass responses in CO<sup>2</sup> enrichment experiments in temperate forests (Körner et al., 2005), the absence of such experiments in the tropics (Zuidema et al., 2013), and isotope/tree ring studies indicating that CO2 induced simulation of tree growth may not be valid (van der Sleen et al., 2014) call for caution in relating changes in carbon gain to changes in tree biomass growth.

#### TREE TRAIT ACCLIMATION FROM PHYSIOLOGICAL PRINCIPLES

How can acclimation mitigate the negative effects of warming or enforce the positive effects of higher ambient CO2? At constant ambient CO2, the simulated trees acclimated to a higher temperature and water stress by decreasing leaf area (**Figure 2D**) and partial closure of stomata (to limit water loss), resulting in reduced leaf photosynthesis. Moreover, a lower leaf area and corresponding lower total nitrogen and protein mass in the crown allowed trees to keep respiration costs (which are strongly determined by protein turnover rates and thus protein content) within the bounds of the photosynthetic carbon supply, i.e., maintaining a positive carbon gain. Trees also maintained a relatively constant stem sapwood area (**Figure 2E**), and the resulting increase in the ratio of sapwood area to leaf area enabled them to maintain a more favorable balance between water supply and transpiration demand, allowing high stomatal conductance and photosynthesis rates at leaf level. The reduction in crown leaf area index resulted in greater light penetration into the canopy and induced a slightly higher average leaf photosynthetic capacity (**Figure 2F**). The predicted acclimation patterns agree with empirical studies showing that tropical trees reduce their leaf area if experimentally subjected to drought (Meir and Woodward, 2010, also for shrub and grasses, Xu and Zhou, 2006; Xu et al., 2014), that trees combine a reduction in leaf area with increased leaf photosynthetic capacity when nitrogen is not limiting (McMurtrie et al., 2008), and that trees establish a lower leaf to sapwood area ratio under drier climatic conditions (Martínez-Vilalta et al., 2009).

Acclimating trees were predicted to maintain a positive carbon gain (i.e., respiration being lower than photosynthetic carbon supply) over a wider temperature range with a ∼ 2 ◦C higher maximum temperature than non-acclimating trees (**Figures 2B,C**). These 2◦C higher upper range temperature value was robust for different water stress scenarios, both when differing in the vapor pressure difference between atmosphere and leaves and when considered different levels of soil water availability (Supporting Information Figure S2). Both acclimating and non-acclimating trees were assumed to have the same increase in biochemical respiration costs of photosynthetic proteins under warming (Supporting Information Text S1, Bernacchi et al., 2001, 2003). Such protein-based temperature effects on respiration may account for the observed instantaneous increases in leaf respiration at higher temperature in experimental studies (Bernacchi et al., 2001; Gifford, 2003; Smith and Dukes, 2013; Slot et al., 2014). Our simulations predict that acclimating trees mitigate the additional, temperature-induced, respiration costs at the whole tree level by reducing total leaf area (**Figure 2D**). They thus reduced the total amount of photosynthetic active proteins in the crown, resulting in lower crown maintenance respiration costs (**Figure 3A**). However, on the short run (i.e., days to weeks) trees may not respond in crown leaf area but rather in leaf physiology. When crown leaf area index was kept constant, simulated trees reduced their leaf photosynthetic capacity and leaf respiration (**Figures 3B** and **4**), which is in line with shortterm leaf responses to temperature in experimental (Gifford, 2003; Loveys et al., 2003; Smith and Dukes, 2013), modeling studies (Dewar et al., 1999), and field studies on tropical canopy trees and lianas (Slot et al., 2014). At the level of the entire tree, both fully and partially acclimating trees tended to homeostasis in the maintenance respiration: photosynthesis ratio at increasing temperature (**Figures 3C,D**), which is an emergent property of our simulations based on modeled physiological principles and trait acclimation. Our results are also consistent with the empirical long-term respiration responses of plants to warming (Gifford, 2003; Loveys et al., 2003), which are much lower than the often assumed Q<sup>10</sup> values of 2 (but see Slot et al., 2014), and the slight increase in the ratio for plants at temperatures beyond 23◦C (Loveys et al., 2003). Based on our results, we expect that acclimation in crown leaf area index is most relevant for climate impacts over a scale of weeks or more. On a scale of days to weeks, acclimation in the leaf photosynthetic capacity and leaf respiration becomes more relevant (see also Smith and Dukes, 2013; Slot et al., 2014), probably next to substrate limitations for respiration (Dewar et al., 1999) but those were not considered here. Yet, since the mechanisms driving acclimation in tree respiration remain, poorly understood (Teskey et al., 2008), the uncertainty in the understanding and predicting tree growth in response to

warming still remains (Smith and Dukes, 2013; Slot et al., 2014).

With increasing ambient CO2, our simulated trees produced more leaf area (**Figure 2D**) due to increased photosynthetic efficiency. The higher level of self-shading, a consequence of a larger crown leaf area index, resulted in a lower optimal photosynthetic leaf capacity (**Figure 2F**). These trends in leaf area and photosynthetic capacity with increasing CO<sup>2</sup> are in line with the results of FACE experiments (Ainsworth and Rogers, 2007) and experiments on small plants (Xu et al., 2014). Since high ambient CO<sup>2</sup> allows leaves to maintain high leaf photosynthesis but reduce transpiration by a partial closure of stomata, trees increased water-use efficiency (carbon gain/water loss) and decreased sapwood area (**Figure 2E**) to reduce respiratory

and replacement costs of sapwood (Ryan et al., 2006). The reduced leaf photosynthetic capacity (**Figure 2F**) at elevated CO<sup>2</sup> is consistent with observations of lower carboxylation rates (Medlyn et al., 1999; Millard et al., 2007). The increased wateruse efficiency at elevated CO<sup>2</sup> is consistent with results of stable isotope (13C) studies in tropical tree rings (Hietz et al., 2005). The predicted reduction in stomatal conductance also agrees with the observation of a 34% decline in stomatal densities for sub-tropical tree species over the last 150 years (Lammertsma et al., 2011). Thus, the predicted acclimation responses are qualitatively supported by empirical studies, and likely contribute to enhancing carbon gain at high ambient CO2.

## EXTREMELY HOT AND DRY YEARS

Gradual climatic changes may have very different impacts on tree growth compared to short extreme events, such as the incidental dry and warm years that are affecting tropical forests (Bonan, 2008; Phillips et al., 2009) and will likely increase in frequency due to deforestation and forest fragmentation (Malhi et al., 2008). During such droughts temperatures may increase by as much as 3–5◦C (Phillips et al., 2009). We evaluate the effects of droughts in additional simulations in which soil water was drastically reduced. These simulations suggest that during dry and hot years, carbon gain is severely reduced, with high risks of zero or negative carbon gain when such droughts occur under current ambient CO<sup>2</sup> levels (see Supporting Information Figure S2). Under those conditions, the simulated trees collapsed in total leaf area, stem sapwood area and leaf photosynthetic capacity, similar to the situation at temperatures >30◦C, current ambient CO<sup>2</sup> (390 ppm) and normal soil moisture that is shown in **Figures 2D–F**. Such a collapse confirms the risks for tipping points (Phillips et al., 2010), which is consistent with the increased death rates of large forest trees during warm and dry years (Phillips et al., 2009, 2010). At doubled ambient CO<sup>2</sup> levels predicted for the end of this century, simulations for dry and hot years showed higher carbon gains and a higher survival probability (Supporting Information Figure S2). Our results thus suggest that tropical trees will become more resilient to individual extreme hot and dry years as ambient CO<sup>2</sup> increases toward the end of this century. Note, however, that these simulations do not take shifts in rainfall regimes into account, and do not account for extreme case scenarios of warming and drought (Cox et al., 2004).

## LIMITATIONS

We based our mathematical model on the recent progress made in physiological leaf models (Bernacchi et al., 2001, 2003; Lloyd and Farquhar, 2008), plant models (McMurtrie et al., 2008; Dewar et al., 2009; Sterck et al., 2011) and vegetation models (Sitch et al., 2008), and added the physiological acclimation in (plastic) functional traits like leaf area, sapwood area and leaf photosynthetic capacity within the constraints set by the biophysical principles of the water and carbon relations within plants (**Figure 1**, Sterck and Schieving, 2011). Our approach provides a step in understanding the role of trait acclimation in mitigating the effects of warming on tropical forest canopy trees, but more work is needed to fully understand this role. Using optimization techniques, our model yields the maximum attainable contribution that the acclimation in some key functional plant traits can make to tree carbon gain and survival. This is the maximum attainable value since, by allowing traits to reach optimal values, possible genetic or time constraints on acclimation are not taken into account. Our simulations also did not consider the possible warming impacts on investments in reproduction, costs of root respiration and maintenance of mycorrhiza, all of which could reduce the realized impact of trait acclimation. On the other hand, we did not include the potentially important drought-induced acclimation in stomatal regulation and cavitation vulnerability, both of which may increase the effect of acclimation. Little is known about the magnitude and the rate at which tree traits respond to increased temperature, drought and elevated ambient CO2. Our simulations thus present a first clear benchmark for the maximum attainable impact of acclimation on the climate sensitivity of tropical trees; being 10–20% in terms of tree growth and about 2 C in terms of hightemperature tolerance. The extent to which this potential will be achieved depends on the degree of plasticity in the traits that we considered, but also in traits that were considered not plastic and entered as constants in the simulations. More research on this plasticity is urgently needed.

#### CONCLUSION

Our study suggests that trait acclimation may assist tropical forest trees to survive under the climatic changes predicted

for this century. Positive effects of CO<sup>2</sup> fertilization and trait acclimation on tree carbon gain may mitigate negative impacts of warming and gradually increasing water stress, as long as they remain within the thermal limits of C3-photosynthesis of woody plants (Bernacchi et al., 2001, 2003). On the other hand, our simulations reveal that strongly reduced carbon gain and risks of tree death remain during hot and dry years when tree structure and physiology may collapse; tropical forest trees will unlikely be able to adjust to those conditions. These risks are considerable under current and near-future CO<sup>2</sup> levels, but may be smaller at the doubled CO<sup>2</sup> concentrations projected for the end of this century (IPCC, 2007, 2013). However, it remains highly uncertain whether acclimation and CO<sup>2</sup> impacts will be sufficient to mitigate the mortality risk of canopy trees that are exposed to extreme warming (up to ∼9 ◦C) and rainfall loss in largely deforested landscapes (Cox et al., 2004). Under those circumstances, large-scale tree death or crown thinning (Phillips et al., 2009) may provoke further reductions in tree cover, reduce rainfall at regional scale and bring moist forests close to tipping points of conversion to drier forests or even savannahs (Malhi et al., 2009; Hirota et al., 2011). Limiting the magnitude of warming and reducing tropical deforestation during this century will reduce chances that extreme drought events will bring tropical forests close to such tipping points.

We argue that acclimation of functional plant traits and their underlying physiology in tropical canopy trees requires considerably more attention from researchers. We simulated the consequences of optimized acclimation in functional traits that are notoriously plastic and have strong impact on the water balance and carbon gain of trees. The match between the model predictions and observations on acclimation suggests that our approach captured many of the key principles driving the responses of canopy trees to climate change. Our simulations nevertheless present a first benchmark for the attainable impact of acclimation on the climate sensitivity of tropical trees; being 10–20% in terms of tree net productivity (which is rather similar to the predictions of Slot et al., 2014) and about 2◦C in terms of high-temperature tolerance. The extent to which this potential will be achieved depends on the degree of plasticity in the traits that we considered, but also in traits that were considered not

#### REFERENCES


plastic and entered as constants in the simulations. Even when limited in their acclimation, traits like cavitation vulnerability and stomatal sensitivity to leaf water potential (Sterck et al., 2012) certainly require attention in future studies given their potential impacts on plant responses to climate (Choat et al., 2012; Scoffoni et al., 2012). Such understanding in acclimation is urgently needed to understand how trait acclimation within trees and trait variation across species will drive the resilience of canopy trees that currently dominate the tropical forest, as well as their future replacement by trees of the same or other species. In addition to the rapidly increasing body of literature on trait comparisons across species, better information on acclimation responses is needed to realistically quantify the impact of acclimation on simulated tree growth and forest biomass.

## AUTHOR CONTRIBUTIONS

All authors contributed to designing this study, drafting and revising the work. FS and FSc developed the plant model used and FS carried out the analysis for this study.

## FUNDING

FS and PZ were supported by a European Research Council grant (ERC #242955).

## ACKNOWLEDGMENTS

We thank Peter Hietz, William Laurance, and Marten Scheffer for their comments on an earlier version of the manuscript. This study benefited from the discussions during the FPS COST Action FP1106 STReESS.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.00607



on growth in Scots pine branches across Europe. Funct. Ecol. 26, 541–549. doi: 10.1111/j.1365-2435.2012.01963.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Sterck, Anten, Schieving and Zuidema. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Interpreting the Climatic Effects on Xylem Functional Traits in Two Mediterranean Oak Species: The Role of Extreme Climatic Events

#### Angelo Rita1, 2 \*, Marco Borghetti <sup>2</sup> , Luigi Todaro<sup>2</sup> and Antonio Saracino<sup>1</sup>

<sup>1</sup> Dipartimento di Agraria, Università di Napoli Federico II, Portici, Italy, <sup>2</sup> Scuola di Scienze Agrarie, Forestali, Alimentari ed Ambientali, Università della Basilicata, Potenza, Italy

#### Edited by:

Achim Braeuning, University of Erlangen-Nuremberg, Germany

#### Reviewed by:

Ignacio García-González, University of Santiago de Compostela, Spain Ze-Xin Fan, Xishuangbanna Tropical Botanical Garden, China

> \*Correspondence: Angelo Rita angelo.rita@unina.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 24 February 2016 Accepted: 14 July 2016 Published: 02 August 2016

#### Citation:

Rita A, Borghetti M, Todaro L and Saracino A (2016) Interpreting the Climatic Effects on Xylem Functional Traits in Two Mediterranean Oak Species: The Role of Extreme Climatic Events. Front. Plant Sci. 7:1126. doi: 10.3389/fpls.2016.01126 In the Mediterranean region, the widely predicted rise in temperature, change in the precipitation pattern, and increase in the frequency of extreme climatic events are expected to alter the shape of ecological communities and to affect plant physiological processes that regulate ecosystem functioning. Although change in the mean values are important, there is increasing evidence that plant distribution, survival, and productivity respond to extremes rather than to the average climatic condition. The present study aims to assess the effects of both mean and extreme climatic conditions on radial growth and functional anatomical traits using long-term tree-ring time series of two co-existing Quercus spp. from a drought-prone site in Southern Italy. In particular, this is the first attempt to apply the Generalized Additive Model for Location, Scale, and Shape (GAMLSS) technique and Bayesian modeling procedures to xylem traits data set, with the aim of (i) detecting non-linear long-term responses to climate and (ii) exploring relationships between climate extreme and xylem traits variability in terms of probability of occurrence. This study demonstrates the usefulness of long-term xylem trait chronologies as records of environmental conditions at annual resolution. Statistical analyses revealed that most of the variability in tree-ring width and specific hydraulic conductivity might be explained by cambial age. Additionally, results highlighted appreciable relationships between xylem traits and climate variability more than tree-ring width, supporting also the evidence that the plant hydraulic traits are closely linked to local climate extremes rather than average climatic conditions. We reported that the probability of extreme departure in specific hydraulic conductivity (Ks) rises at extreme values of Standardized Precipitation Index (SPI). Therefore, changing frequency or intensity of extreme events might overcome the adaptive limits of vascular transport, resulting in substantial reduction of hydraulic functionality and, hence increased incidence of xylem dysfunctions.

Keywords: tree-ring, hydraulic conductivity, quantitative wood anatomy, Mediterranean climate, pointer years

## INTRODUCTION

An accepted picture for the Mediterranean region is the ongoing trend toward higher temperature and reduced precipitation, associated to an increase in the frequency and magnitude of climatic extremes, which are expected to have detrimental effects on trees and forest biomes (Easterling et al., 2000; IPCC, 2013). Drought imposed by changesin rainfall patterns and temperature anomalies were implicated in a number of well-documented drought-induced tree mortalities and forest decline episodes, with likely consequences on species distribution and community structure (e.g., Allen et al., 2010; Anderegg et al., 2012).

Interestingly, extreme climatic events are increasingly considered to play a major role in tree mortality, and variation in xylem anatomical traits linked to tree hydraulic properties has received considerable attention in recent decades as an important plant acclimation process (for a review on this issue, see Jentsch et al., 2007). Specifically, both observational and experimental studies to date reported that variability and extremes in climate are more important drivers of ecosystem processes than mean conditions (Royer et al., 2011; Smith, 2011; Thompson et al., 2013). For example, a number of studies that experimentally imposed climate extremes via field experiments clearly described the negative impact of extreme drought on the xylem hydraulic function and productivity (Jentsch et al., 2011; Barigah et al., 2013; Urli et al., 2013). However, there are so far at least three factors limiting our understanding of the impacts of extreme events on plant hydraulic functionality: (i) the frequently loose definition of extreme events, which needs refinement because climate change involves modifications in both mean and variability (Smith, 2011; Lloret et al., 2012); (ii) the difficult comparison between case studies, particularly due to their heterogeneity in temporal and spatial scales and the disparity of response variables (Reyer et al., 2013); (iii) still incomplete information on the phenotypic plasticity of trees, that is, the potential to modify their form and function in response to environmental changes, especially to changing climate (Fonti et al., 2010). Phenotypic plasticity can reduce mortality risk when plants are exposed to new conditions. The most common plastic responses of trees to drought are a reduction in leaf area to restrict water loss (DeLucia et al., 2000) and increased root growth to enhance access to water and nutrients. In turn, the structure and hydraulic function of xylem can also vary within a single species in response to climate conditions. For instance, variation in tree-ring widths, vessel diameters, or distribution has often been used to reconstruct information about past environmental conditions and infer the hydraulic function of xylem (see Fonti et al., 2010 for a review). Nevertheless, acclimation is not instantaneous, and it is sensitive to several factors. For example, full acclimation to a temperature shift may take between a few days to weeks and might be further affected by interactions with other factors such as drought (Valladares et al., 2007). Therefore, xylem plastic adjustments may not be able to cope with the effect of rapid and extreme climatic events.

Nevertheless, for high temporal resolution, long-term xylem traits chronologies have shown to be sensitive indicators of climate variability. Promising results have been obtained from studies on water conducting elements across a range of hardwoods species (e.g., Maherali et al., 2004). In particular, several authors successfully revealed a clear signal in vessel traits of ring-porous species mainly linked to the water availability (Fonti and García-González, 2004; Campelo et al., 2010; Gea-Izquierdo et al., 2012); for sub-Mediterranean oaks it was demonstrated that most of the variability in early wood vessel size could be explained by spring precipitations (García-González and Eckstein, 2003; González-González et al., 2014). Moreover, several studies recently acknowledged the importance of the ontogenetic changes on the hydraulic design of woody plants (Olson et al., 2014), suggesting careful evaluation of the climatic information from tree-ring time series (Carrer et al., 2015).

The present study aims to disentangle the effects of mean and extreme climatic variability on functional anatomical vessel traits from long-term tree-ring series. Specifically, we hypothesized that (i) vessel traits may reflect more substantially the climatic signal than tree-ring width, (ii) a closely link between extreme values in tree-ring series (both ring with and specific hydraulic conductivity) and site-specific extreme climatic condition occurs.

We used Quercus cerris L. and Quercus pubescens Willd. as model species, both are ring-porous species with large diameter early-wood vessels that allow water movement with a minimum of hydraulic resistance (Tyree and Zimmermann, 2002). However, experimental evidences indicates higher cavitation rates and vulnerability to embolism in the former species (Borghetti et al., 1993; Lo Gullo et al., 1995; Nardini et al., 1999), whose large vessels could enhance water transport efficiency but compromise the safety of the xylem (Tyree and Zimmermann, 2002). Recent studies have reported several oak-decline episodes in the Iberian Peninsula during the 1980s and 1990s when several intense summer droughts episodes occurred (Peñuelas et al., 2001; Corcuera et al., 2004). To this aim, the Generalized Additive Model for Location, Scale and Shape (GAMLSS, Rigby and Stasinopoulos, 2005) and a Bayesian logistic simulation were used to perform a high-resolution examination of tree-ring traits and climate relationships. The ability of the aforementioned tools to handle non-linear data structures can better represent the complex relationship between xylem functional traits and environmental variables.

#### MATERIALS AND METHODS

#### Study Site and Plant Material

The study was carried out on trees sampled in a mountain forest in the Pollino National Park in Southern Italy, close to the Mediterranean coast. The climate is influenced by differences in altitude, slopes, and proximity to the sea. There is a typical Mediterranean seasonal alternation between dry and warm summers and rainy winters.

Temperatures were collected from Castrovillari (39◦ 83′ N, 16◦ 19′ E, 343 m a.s.l.) and precipitation data from San Lorenzo Bellizzi (39◦ 88′ N, 16◦ 32′ E, 851 m a.s.l.) meteorological stations (Italian Hydrographic Service, SIMI). Temperatures were corrected for altitude by applying a coefficient of −0.007◦C m−<sup>1</sup> (ICAO, 2002). The average annual precipitation is 1065 mm distributed as 39.5% in winter, 23.7% in spring, 29.2% in autumn, and 7.6% in summer. The Mediterranean sub-humid climate is characterized by warm summers (at this elevation, 18.06◦C is the average temperature for July through August) and cold winters (average 1.8◦C for December through February). Mean annual temperature is 9.4◦C and snowfalls are generally distributed from November to April.

In recent decades, a maximum of 120 days of dry weather was recorded in summer.

At the study site (39◦ 56′ 58.8′′N, 16◦ 10′ 32.4′′E, elevation 1050 m a.s.l.) the forest consist of scattered trees with a canopy height of ∼20 m, with few trees reaching a tree height of ∼28 m.

Two 5 mm diameter cores from each of 16 Q. cerris and 15 Q. pubescens tall adult trees with a diameter at breast height (DBH) >40 cm were collected with an increment borer for treering analysis. For each core sample, tree-rings were first visually cross dated (Yamaguchi, 1991) and then measured to the nearest 0.01 mm using the incremental measuring table SMIL3 (Corona et al., 1989) interfaced with data acquisition software. Finally, the COFECHA software (Holmes, 1983) was used to check for the presence of cross-dating errors and the expressed population signal (EPS) was calculated with the package "dplR" (Bunn, 2008) in the R statistical suite (R Core Team, 2015) to quantify the common variability among tree-ring series. An average tree-ring chronology spanning from 1926 to 2012 was obtained; the EPS value exceeded the suggested threshold 0.85 level (EPS = 0.92), indicating a high degree of common variability between tree-ring series (Wigley et al., 1984).

#### Xylem Anatomy and Specific Hydraulic Conductivity

A subsample of 10 cross-dated cores from the two species was investigated for xylem anatomical characteristics and xylem hydraulic conductivity after checking the presence of reaction wood or wounding (Arbellay et al., 2012).

A sliding microtome (HM 400, Microm International GmbH, Walldorf, Germany) was used to obtain 20µm thick transverse sections from split micro-sections of entire wood cores (from 1 to 1.5 cm long). Histological preparations were then obtained by staining sections with 2% astrablue and 1% safranin solutions, which resulted in unlignified cells appearing blue and lignified cells appearing red (Schweingruber and Poschlod, 2005). Sections were subsequently dehydrated using a series of ethanol solutions of increasing concentrations, washed with xylol, and embedded in Canada balsam.

Annual ring images from transverse sections were captured with a CCD digital camera (Skopkam DCM300) mounted on a reflected light microscope (AxioPhot, Carl Zeiss, Jena, Germany). Sequential images were subsequently stitched using the Microsoft Image Composite Editor (ICE 1.3.5), and analyzed with the image-analysis software ImageJ v.1.40 (National Institute of Health, Bethesda, MD, USA, http://rsb.info.nih.gov/ij). Images were first converted from 24-bit color into 8-bit grayscale and then the objects contour was produced in a threshold binary image (mask) in which only the particles of interest were retained, in our case the vessels lumen. Before any measuring, the image was calibrated from a scale bar of known length in the image. The particle analysis function led us to calculate for each treering, in a chosen surface (SXylem = W<sup>r</sup> ∗ l, where W<sup>r</sup> is ring width and l = 2 mm) bounded by rays, the vessel number (N), the vessel lumen area (A), and the Cartesian coordinates of each vessel (>480µm<sup>2</sup> ). Careful visual inspection was also performed to verify all vessel elements and non-vascular elements included.

Since vessels are not exactly circular but mostly elliptical, the diameter of each vessel was calculated as:

$$\mathbf{d} = \left(\frac{32\left(\mathbf{ab}\right)^3}{\mathbf{a}^2 \mathbf{b}^2}\right)^{\frac{1}{4}}$$

where a and b are major and minor perpendicular lumen diameters, respectively (Lewis, 1992).

Based on the vessel contribution to hydraulic conductance, we calculated the hydraulically weighted mean diameter (Dh) for each ring according to Tyree and Zimmermann (2002):

$$\mathbf{D}\_{\mathbf{h}} = \left(\frac{1}{\mathbf{n}} \sum\_{1}^{\mathbf{n}} \mathbf{d}^4\right)^{\frac{1}{4}}$$

According to the Hagen–Poiseuille equation, theoretical hydraulic conductivity (Kh, m<sup>4</sup> MPa−<sup>1</sup> s −1 ) was calculated from the vessel radii (r) as

$$\mathbf{K}\_{\mathbf{h}} = \frac{\pi \sum\_{1}^{\mathbf{n}} \mathbf{r}^{4}}{8\eta}$$

where η is the viscosity of water at 20◦C (1.002 10−<sup>3</sup> Pa s).

The tree-ring specific hydraulic conductivity (K<sup>s</sup> , kg m−<sup>1</sup> MPa−<sup>1</sup> s −1 ) was estimated by dividing the theoretical hydraulic conductivity (Kh) by the tree-ring surface area (Si) and multiplying with the density of water (ρ) at 20◦C (998.20 kg m−<sup>3</sup> ), according to the modified Hagen-Poiseuille equation reported by Tyree and Ewers (1991)

$$\mathbb{K}\_{\rm s} = \frac{\mathbb{K}\_{\rm h} \rho}{\mathbb{S}\_{\rm Xylem}}$$

Average vessel size (Aav) and vessel density (dv), determined as the ratio between the number of vessels and the area analyzed, were also calculated.

#### Data Analysis

Statistical analyses were performed for the period 1952–2012 (60 years). To explore the relationships between xylem traits (W<sup>r</sup> and Ks) and climatic variables (temperature and precipitation) we applied Generalized Additive Models for Location, Scale, and Shape (GAMLSS) proposed by Rigby and Stasinopoulos (2005) as semiparametric regression model. GAMLSS overcomes some limitations associated with Generalized Linear Models (GLMs) and GAMs by providing a flexible modeling framework that allows the use of more general distributions, such as highly skewed or kurtotic distributions, which may be more appropriate for modeling the record of interest. The number of parameters represented in the GAMLSS distributions varies from one to four, with almost all distributions represented by a location (µ) and scale (σ) parameter and some distributions represented by one or two shape parameters (υ and τ) to represent skewness and kurtosis in the response variable data. For this reason, the form of the distribution assumed for the response variable is y∼f(x| µ, σ, υ, τ). Computational implementation was performed using the package "gamlss" (Stasinopoulos and Rigby, 2007) in the R statistical suite (R Core Team, 2015). The model also included the cambial age as covariate and a random intercept term to account for variation among trees. The resulting estimations, are based on a P-Spline (ps, Penalised B-spline) smoothing function (Eilers and Marx, 1996), where the smoothing parameters (and hence the effective degrees of freedom) are estimated using the local maximum likelihood method.

The model building process consisted in comparing many competing models for which different combinations of components (i.e., Mmodel = Ddistribution, Glink function, Tpredictors, λsmoothing) were tried. Minimizing the Akaike Information Criterion (AIC) was used for the model selection (Akaike, 1998). Selected models were checked for the independence and normality of the residuals by worm plots and qq-plots (Buuren, 2007).

In addition, we used the LMS (λ-µ-σ) method (Cole and Green, 1992) to determine the age-specific trends in the xylem traits, thereby allowing examination of the temporal trends in specific percentile points of W<sup>r</sup> and K<sup>s</sup> . They were estimated via lms() in the "gamlss" R package, where the first three moments of the measurement frequency distribution were modeled as cubic smoothing spline curves, based on Box Cox-type transformations of data. The LMS method implicitly leads to non-crossing curves via a scaling function.

Given that low summer precipitation concomitantly with high temperatures were found to be the most limiting factor for the development of the Mediterranean tree species, we tested the hypothesis of an influence of climate on extreme deviation in the ring width (Wr) and specific hydraulic conductivity (Ks), by applying one of the most commonly used procedures in classical dendrochronological studies: pointer years analysis. In fact, as stated by Schweingruber et al. (1990), event and pointer years are a suitable proxy of the extreme climatic events to which trees have been exposed in the past. We calculated the pointer years on the individual series of W<sup>r</sup> and K<sup>s</sup> with the "pointRes" (van der Maaten-Theunissen et al., 2015) R package by using the normalization in a moving window (|W| = 5 years) according to Cropper (1979; cf. Schweingruber et al., 1990). Pointer years were defined as those years with absolute values above a threshold of 0.75 according to Cropper (1979). This method z-transforms tree growth in year i within a symmetric moving window of n years, thereby providing the number of standard deviations that tree growth deviates in individual years from the window average. Subsequently, positive and negative pointer years were represented by dichotomous variables coded as 0 and 1, respectively.

Hence, a Bayesian logistic regression was adopted to test the link between pointer years (where pointer years are Bernoulli distributed) and climate through JAGS + "rjags" R package cross-platform Plummer (2003). The predictor variable comprised the Standardized Precipitation Index (SPI) computed through the "SPEI" R package (Beguería and Vicente-Serrano, 2013).

Then, with W<sup>r</sup> and K<sup>s</sup> pointer years values (Y) in {0, 1} the estimated model was:

$$\begin{aligned} \text{Y} & \sim Bernoulli \left( \text{y}\_{i} \middle| \pi\_{i} \right) \\\\ \text{where, } \pi\_{i} & \equiv \Pr \left( \text{y}\_{i} = 1 \mid \text{SPI}\_{i}, \beta \right) = \frac{1}{1 + \exp(-\text{SPI}\_{i}\beta)} \end{aligned}$$

JAGS used Markov Chain Monte Carlo (MCMC) to generate a sequence of dependent samples from the posterior distribution of the parameters by assuming a weakly informative prior distribution (0, 0.5), as proposed by Gelman et al. (2008). Simulation was performed by running four chains with 20,000 total iterations per chain and 10,000 initial samples "burn-in." Convergence diagnostics (provided by the coda R package) were visually checked by the autocorrelation plot, Geweke's diagnostic and the Gelman-Rubin shrink factor (Brooks and Gelman, 1998).

Then we modeled the change in the probability of presence (1) − absence (0) of Wr and Ks pointer years at the extreme upper and lower SPI event (exceeding the 90th and 10th percentile, respectively) by running 1000 bootstrap simulation of quantities of interest (QI) from the posterior density of the Bayesian model, as suggested by King and Wittenberg (2000).

## RESULTS

Relationships between xylem traits and W<sup>r</sup> showed that the standard linear regression fit to these data was significantly different from zero for Aav, Dh, and d<sup>v</sup> (**Figure 1**, upper panels); in particular, all outcomes appeared to be negatively related to the tree-ring width. Moreover, the regression coefficient for all traits showed different significant patterns according to the quantile considered (**Figure 1**, lower panels).

Results of the fitted GAMLSS highlighted significant temperature, precipitation, and cambial age effect for K<sup>s</sup> , whilst only the age effect was depicted for W<sup>r</sup> (**Table 1**). In particular, as expected, increase of both temperature and precipitation led to decrease and increase in the K<sup>s</sup> , respectively (**Figure 2**). Interestingly, both W<sup>r</sup> and K<sup>s</sup> models indicated that these variables are strongly dependent on cambial age. In this regard, further investigation (**Figure 3**) indicated an inverse trend across the overall age distribution and in all percentiles for W<sup>r</sup> and K<sup>s</sup> , where the rate of increase in hydraulic conductivity was directly related to age. In particular, higher percentile levels were reached from 10 to 20 years for W<sup>r</sup> and from 40 years onwards for K<sup>s</sup> . Moreover, there were larger increases in the upper than the lower percentiles in K<sup>s</sup> , particularly from age >20 years. The typical monotonic increasing trend of hydraulic conductivity tends to be low when choosing the 5th percentile, i.e., selecting only the smallest conduits per year.

The summary statistics table of the Bayesian logistic model (**Table 2**) showed the marginal posterior distribution for parameter α (intercept) and βSPI (coefficient; see Supplementary Figures S1, S2 for trace and density plots). Interestingly, for K<sup>s</sup>

FIGURE 1 | Relationships between the average vessel size [Aav; β = −0.002, F(1, 375) = 10.52, p < 0.01], hydraulic diameter [Dh; β = −0.009, F(1, 375) = 18.77, p < 0.001], vessel density [dv; β = −1.808, F(1, 375) = 34.12, p < 0.001], and tree-ring width [Wr] (upper panel). Red and blue rugs inside the plots show distribution of the x and y variables, respectively. Slope of the estimated linear quantile regression for the xylem traits as a function of the τth quantile τ (lower panels). Full red lines are the least squares estimates for the coefficients and the red dashed lines are the 95% confidence intervals for the least squares estimates. The gray area represents the 95% confidence interval (1000 replicate bootstrap) for the quantile regression estimates (full circles, each 5th quantile).



Parameters of the distribution families are µ (mean, location parameter), σ (centile-based coefficient of variation, scale parameters), υ (skewness), and τ (kurtosis). Stars mean statistical significance for \*\*\*p < 0.001, \*p < 0.05. Est, Estimate; BCCGo, Box-Cox-Cole-Green-orig.; BCPEo, Box-Cox Power Exponential-orig.; AIC, Akaike Information Criterion.

confidence intervals. Gray rugs on the y-axis and x-axis represent distribution of the partial residuals and covariates, respectively. Note free scales of partial residuals.

the 95% credible interval for βSPI is positive, indicating with very high probability that the β term is positive: exposure to high SPI increases the probability of positive deviation in K<sup>s</sup> . No similar evidence was found for W<sup>r</sup> . However, simulation of quantities of interest, in terms of predicted probability of a success, was more informative than simply reporting the model estimates (**Figures 4**, **5**). In particular, results of simulation reported the predicted and the expected probability (sensu King and Wittenberg, 2000) of the presence/absence of pointer years at the 90th and 10th percentile of the SPI with 95% confidence level. Simulation in **Figure 4** confirms that there was no clear probability that formation of pointer years in W<sup>r</sup> was affected by changes in SPI. Indeed, we reported that in spite of extreme values of SPI (above the 90th percentile), there is the 53% probability of occurrence of positive pointer years. On the other hand, SPI values below the 10th percentile are expected to affect the formation of a negative pointer year at 46% probability. However, contrasting results were found for simulation of K<sup>s</sup> pointer years (**Figure 5**). We expected the 78% probability of occurrence of positive pointer years at the 90th percentile of SPI and only 22% probability of negative ones. By contrast, extreme negative values of SPI (below the 10th percentile) led to the occurrence of negative pointer years at 76% probability.

## DISCUSSION

Notable negative relationships were detected in **Figure 1** between radial growth and vessel traits; we can therefore reasonably rule out the hypothesis of a direct growth-dependent constraint on the intra-annual xylem hydraulic traits. In former studies, this inverse relationship was reported when comparing wood anatomy of both ring porous (Phelps and Workman, 1994; Fonti and García-González, 2004; Gea-Izquierdo et al., 2012) and diffuse porous hardwood species (Denne et al., 1999; Rita et al., 2015). On the other hand, according to the recent tendencies, we cannot exclude an age-dependent constraint on the whole-plant hydraulic function (Olson et al., 2014; see Section Discussion below).

As a whole, there are a number of interesting considerations that can be drawn from our GAM model. First, our findings reported a discernible climatic signal of K<sup>s</sup> compared to W<sup>r</sup> (**Table 1** and **Figure 2**), in accordance with many recent studies on the Mediterranean ring-porous (García-González and Eckstein, 2003; Campelo et al., 2010), diffuse-porous (Rita et al., 2015), and conifer trees (Olano et al., 2012). This marked link reflects the ability of trees to adjust the characteristics of their xylem hydraulic architecture, such as

FIGURE 3 | Tree-ring width (Wr) and specific hydraulic conductivity (Ks) centile curve (left and central panels, respectively) using the LMS method as a function of cambial age. Each panel shows the 5th, 25th, 50th, 75th, and 95th percentile from the bottom to the top, respectively. Scatter plot between K<sup>s</sup> and W<sup>r</sup> with non-linear [K<sup>s</sup> ∼ I(1/W<sup>r</sup> \* a) + b] quantile regression line (right panel). The black line is median (a, β = 3.505e-04, p < 0.001; b, β = 1.377e-04, p < 0.001), the red line is the 90th quantile (a, β = 0.00069, p < 0.001; b, β = 0.00018, p < 0.001), and the blue line is the 10th quantile (a, β = 0.00015, p < 0.001; b, β = 0.00008, p < 0.001).



Marginal posterior distribution of mean, standard deviation, and quantiles for α (intercept) and βSPI. SPI, Standardized Precipitation Index. Trace and density plots are on Supplementary Figures S1, S2.

the arrangement, frequency, and diameter of vessels to climate variability (Hacke et al., 2006; Sperry et al., 2008), and can provide information about the plasticity of a species under changing environmental conditions. For instance, functional relationships between xylem traits of Q. canariensis trees growing in the Mediterranean drought-prone sites exhibit both spatial and temporal plasticity in relation to climatic variability (Gea-Izquierdo et al., 2012).

As for the effects of climate on functional anatomical traits, GAMs results confirm the broadly described influence of climatic factors on the variations in wood traits structure. Accordingly, the positive influence of precipitation and the negative effect of high temperature on the specific hydraulic conductivity (Ks) are considered key features of most Mediterranean tree species (**Table 1** and **Figure 2**). Indeed, rise in temperature and reduced water availability, that concomitantly lead to an intensification in evapotranspiration, are often been reported to strongly reduce the vessel lumen area and increase their density in order to reduce vulnerability to embolism (Lo Gullo et al., 1995; Tyree and Zimmermann, 2002). Accordingly, many valuable results from long-term time series of xylem traits of sub-Mediterranean oaks emphasized greater phenotypic plasticity in response to the stressful climate conditions (González-González et al., 2014 among the others). In particular, the earlywood vessels lumen area of Q. robur were found to decrease in response to reduced spring rainfall (García-González and Eckstein, 2003); similar findings were also highlighted by for Q. ilex (Campelo et al., 2010). Further recent evidences showed correlations between earlywood vessels size and precipitation along the previous growing season for Q. petraea, whereas the number of vessels was related to winter temperature for the sub-Mediterranean Q. pyrenaica (González-González et al., 2014).

In our study, relationships with cambial age over time underlined a strong ontogenetic influence on both growth and specific hydraulic conductivity (**Figures 2**, **3**). In this regard, consistent with the pattern found by other authors for ringporous Quercus spp. (e.g., Heliñska-Raczkowska, 1994; Lei et al., 1996; Fonti and García-González, 2004; Leal et al., 2006), most of the variability in vessel size might be explained by cambial age. Their results showed an overall increasing trend in vessel size and a slight tendency for the conductive area to increase with cambial age. Such a relationship can be largely explained by functional reasons. For instance, multiple lines of evidence suggest that ontogenetic changes in wood anatomy have evolved primarily to provide hydraulic safety in long distance water transport (Anfodillo et al., 2006; Preston et al., 2006; Poorter et al., 2010). Therefore, ontogenetic trends are known to reflect an adaptive compromise between growth constraints and the environment, which is why they should be carefully modeled and interpreted, rather than routinely removed by means of standardization procedures (Carrer et al., 2015).

Indeed, the age-specific tree-ring data analyses of Voelker (2011) showed that the sensitivity of tree growth to environmental variability changes predictably with tree age and that the growth of older forests may be more resilient to climate change effects.

Despite the long-standing recognition of the importance of climate extremes on the overall functional performance of trees, the study of climate extremes is a relatively new emphasis in ecology (Smith, 2011). In this case, one advantage of Bayesian methods was the ability to directly answer specific questions in terms of probability of success. In this study we found no direct-related effect of the SPI on tree-ring growth (Wr) departures, according to the role of "compensatory process" argued by Lloret et al. (2012). This result is also supported by a short-term manipulative experiment on three young deciduous oaks species exposed to artificial air warming and drought: monitored growth reaction showed that, despite a phenological shift induced by warming, annual growth and shoot biomass were not affected by the exposed drought (Kuster et al., 2014).

On the other hand, the probability of extreme departure in specific hydraulic conductivity (Ks) rises at extreme values of SPI (**Figure 5**). What emerged in our study provides new insights into the effect of severe climate events on K<sup>s</sup> from long-term tree-ring series. Indeed, while a generic link between pointer years (sensu Schweingruber et al., 1990) and peculiar climate occurrences are fully investigated (Rolland et al., 2000; Neuwirth et al., 2007; Rita et al., 2014), no specific

temporal pattern between climate extremes and functional traits was well addressed. Therefore, by relying on our results, we believe that the occurrence of extreme SPI (below the 10th percentile) may lead to an adverse effect on the xylem water transport capability for this species at the study site. In fact, although the xylem structure can acclimate to variation during growth and development by plastically adjusting its xylem anatomical traits (Fonti et al., 2010), the presence of extreme climate events may undermine this anatomical adaptation strategy leading to embolism and related dysfunctions (Choat et al., 2012; Urli et al., 2013). In this regard, there is much experimental evidences suggesting that extreme drought stress is a trigger factor inducing hydraulic failure in trees, resulting in loss of carbon assimilation rate (Brodribb et al., 2010; Urli et al., 2013), shoot dieback (Hoffmann et al., 2011), and tree mortality (Carnicer et al., 2011; Barigah et al., 2013). In particular, our results are consistent with Fonti et al. (2013) which document a significant vessel size reduction with diminished conductivity in saplings of three oak species artificially drought-exposed over three consecutive growing seasons. Moreover, physiological measurements conducted by Nardini et al. (2013) highlighted diffuse crown desiccation in Q. pubescens trees caused by hydraulic failure during an extreme drought. Therefore, assuming departures in K<sup>s</sup> during extreme events in the current climate, it is conceivable that increased frequency or magnitude of extreme climate events with more adverse conditions would lead to higher reduction of K<sup>s</sup> and greater incidence of xylem dysfunctions. Thus, the intensification in local extremes rather than average climatic conditions might affect woody plant survival.

## CONCLUSIONS

In this paper we sought to describe the most likely effect of frequency of extreme climate events on tree hydraulic system capacity. By relying on our results, we pointed out several important aspects of how climate variability can affect xylem function of two Mediterranean oak species. Thus, our main results may be summarized as follows:


## REFERENCES


in substantial reduction of hydraulic functionality, and hence increased incidence of xylem dysfunctions.

## AUTHOR CONTRIBUTIONS

AR conceived and designed the study; carried out the measurements; performed analysis of data; wrote the manuscript; AS and LT collected plant materials and contributed to the chronology building; AR, MB, LT, AS contributed to discussing and interpreting the data at all stages.

#### ACKNOWLEDGMENTS

The research jointly funded by the MIUR-PRIN grant No. 2012E3F3LK "Global change effects on the productivity and radiative forcing of Italian forests: a novel retrospective, experimental, and prognostic analysis" granted to MB and MIUR-PRIN No. 2002075152-004 and No. 2005072877-002 granted to AS.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 01126

rates after exposure to extreme water deficit. New Phytol. 188, 533–542. doi: 10.1111/j.1469-8137.2010.03393.x


L. in Mediterranean mountain forests. Dendrochronologia 32, 220–229. doi: 10.1016/j.dendro.2014.04.001


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Rita, Borghetti, Todaro and Saracino. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Complex Physiological Response of Norway Spruce to Atmospheric Pollution – Decreased Carbon Isotope Discrimination and Unchanged Tree Biomass Increment

Vojtech ˇ Cada ˇ <sup>1</sup> \*, Hana Šantr ˚ucková ˇ 2 , Jirí Šantr ˚u ˇ cek ˇ 2 , Lenka Kubištová<sup>1</sup> , Meelis Seedre<sup>1</sup> and Miroslav Svoboda<sup>1</sup>

<sup>1</sup> Department of Forest Ecology, Faculty of Forestry and Wood Sciences, Czech University of Life Sciences Prague, Prague, Czech Republic, <sup>2</sup> Faculty of Science, University of South Bohemia in Ceské Bud ˇ ejovice, ˇ Ceské Bud ˇ ejovice, Czech Republic ˇ

#### Edited by:

Achim Braeuning, Friedrich-Alexander University Erlangen-Nürnberg, Germany

#### Reviewed by:

Marco Carrer, Università degli Studi di Padova, Italy Raquel Esteban, Consejo Superior de Investigaciones Científicas, Spain

> \*Correspondence: Vojtech ˇ Cada ˇ cada@fld.czu.cz

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 02 February 2016 Accepted: 23 May 2016 Published: 09 June 2016

#### Citation:

Cada V, Šantr ˚u ˇ cková H, Šantr ˚u ˇ cek J, ˇ Kubištová L, Seedre M and Svoboda M (2016) Complex Physiological Response of Norway Spruce to Atmospheric Pollution – Decreased Carbon Isotope Discrimination and Unchanged Tree Biomass Increment. Front. Plant Sci. 7:805. doi: 10.3389/fpls.2016.00805 Atmospheric pollution critically affects forest ecosystems around the world by directly impacting the assimilation apparatus of trees and indirectly by altering soil conditions, which subsequently also leads to changes in carbon cycling. To evaluate the extent of the physiological effect of moderate level sulfate and reactive nitrogen acidic deposition, we performed a retrospective dendrochronological analysis of several physiological parameters derived from periodic measurements of carbon stable isotope composition ( <sup>13</sup>C discrimination, intercellular CO<sup>2</sup> concentration and intrinsic water use efficiency) and annual diameter increments (tree biomass increment, its inter-annual variability and correlation with temperature, cloud cover, precipitation and Palmer drought severity index). The analysis was performed in two mountain Norway spruce (Picea abies) stands of the Bohemian Forest (Czech Republic, central Europe), where moderate levels of pollution peaked in the 1970s and 1980s and no evident impact on tree growth or link to mortality has been reported. The significant influence of pollution on trees was expressed most sensitively by a 1.88 reduction of carbon isotope discrimination (113C). The effects of atmospheric pollution h interacted with increasing atmospheric CO<sup>2</sup> concentration and temperature. As a result, we observed no change in intercellular CO<sup>2</sup> concentrations (Ci), an abrupt increase in water use efficiency (iWUE) and no change in biomass increment, which could also partly result from changes in carbon partitioning (e.g., from below- to above-ground). The biomass increment was significantly related to 113C on an individual tree level, but the relationship was lost during the pollution period. We suggest that this was caused by a shift from the dominant influence of the photosynthetic rate to stomatal conductance on 113C during the pollution period. Using biomass increment-climate correlation analyses, we did not identify any clear pollution-related change in water stress or photosynthetic limitation (since biomass increment did not become more sensitive to drought/precipitation or temperature/cloud cover, respectively). Therefore, we conclude that the direct effect of moderate pollution on stomatal conductance was likely the main driver of the observed physiological changes. This mechanism probably caused weakening of the spruce trees and increased sensitivity to other stressors.

Keywords: climate change, carbon dynamics, growth trends, soil acidification, spruce decline, tree-ring analysis, tree stress

## INTRODUCTION

fpls-07-00805 June 8, 2016 Time: 16:15 # 2

Atmospheric pollution and particularly acid sulfate and reactive nitrogen depositions influence ecosystem functioning and services such as carbon sequestration, water purification and nutrient cycling around the world (e.g., Oulehle et al., 2011). While sulfur dioxide (SO2) emissions have been successfully regulated since the end of the 20th century in Europe and North America, they are increasing in other parts of the world (Klimont et al., 2013). Emissions of nitrogen oxides (NOx) and ammonia (NH3) have also decreased since the end of the 20th century in Europe, but remain at relatively high levels compared to the pre-industrial period (Hunová, 2014 ˚ ).

Low levels of sulfur and nitrogen deposition have a fertilizing effect on plants (Meng et al., 1994; Högberg et al., 2006), while increased levels of deposition can acidify soils causing soil nutrient depletion and toxic aluminium (Al3+) mobilization that can weaken the tree root system and cause nutrient deficiency or water stress (Jentschke et al., 2001; Hruška et al., 2012). High levels of deposition directly damage foliage by entering the intercellular space through the stomata and decrease photosynthetic rate, stomatal conductance (Meng et al., 1994) and alter plant water use efficiency (Thomas et al., 2013). Evergreen conifers covering large areas of temperate forests are especially sensitive to atmospheric pollution (Greaver et al., 2012) because the relatively larger surface area of their leaves (also retained during the winter season) effectively captures the deposition of pollutants from the atmosphere (Kopácek and ˇ Hruška, 2010). In this study, we examine the extent of the pollution physiological effects on conifers subjected to moderate pollution load, which is not satisfactorily understood at present. This is achieved by characterizing tree physiology using variables derived from plant carbon stable isotope composition and tree biomass increment.

Forest biomass production also plays a fundamental role in the global carbon cycle and forests represent a major terrestrial storage and sink of atmospheric CO<sup>2</sup> (Pregitzer and Euskirchen, 2004). Therefore, mechanistic understanding of complex environmental effects (including pollution) on forest biomass production is needed (Pregitzer and Euskirchen, 2004). For example, some studies predict an increase in biomass production due to a possible fertilizing effect of increased atmospheric CO<sup>2</sup> (Peñuelas et al., 2011; Saurer et al., 2014). However, if other limiting factors such as atmospheric pollution or temperatures overwhelm the CO<sup>2</sup> effect, very complex and variable responses could be observed (Rydval and Wilson, 2012; Treml et al., 2012; Thomas et al., 2013). This study improves understanding of Norway spruce biomass production by examining its temporal trends and inter-annual variations in relation to air pollution and carbon isotope discrimination.

Environmental changes such as air pollution and climate change can alter the importance of different factors limiting tree physiological processes. Dendrochronological studies have shown for example that temperature limited trees can become insensitive to temperature as a result of pollution load (Elling et al., 2009) or can become sensitive to drought in older ages, under higher competition pressure or in warmer climate (Primicia et al., 2015). More importantly, if a change in the correlation is detected during the peak in air pollution, it may indicate changes in water stress (correlation with precipitation or drought index) and/or photosynthesis limitation by climate (correlation with temperature or cloud cover). Increased water stress could be a consequence of pollution related root weakening and decreased temperature/cloud cover response could be a consequence of pollution related photosynthesis limitation by nutrient deficiency or direct foliage damage.

In central Europe atmospheric pollution peaked in the 1970s and 1980s with highest levels in the region called the "Black Triangle," where widespread mortality and reduction in tree growth, particularly in Norway spruce stands, occurred (Kandler and Innes, 1995; Sander et al., 1995; Rydval and Wilson, 2012). Although the direct link between air pollution and forest decline is evident in the "Black Triangle," there is lack of evidence about the pollution effect from moderately polluted areas beyond the "Black Triangle" (see **Figure 1**; Kandler and Innes, 1995). The unique study from a moderately polluted area by Šantru˚cková ˇ et al. (2007) indicated increased stress of Norway spruce trees exposed to moderate pollution load by the analysis of carbon isotope discrimination. Previous studies also often focused on a single physiological parameter (e.g., Šantru˚cková et al., 2007 ˇ ; Rydval and Wilson, 2012). In this work we will expand on previous studies with the analysis of the complex response of spruce trees to air pollution as indicated by several physiological variables.

The primary goal of this study is to investigate the extent to which moderate air pollution affected the physiology of mountain Norway spruce in the Bohemian Forest in central Europe. The physiological response is represented using variables obtained from tree-ring carbon isotope composition [13C discrimination, intercellular CO<sup>2</sup> concentration (Ci) and intrinsic water use efficiency (iWUE)] and tree-ring increment (biomass increment, its inter-annual variability and correlation with climate). We specifically aim to address the following questions:


#### MATERIALS AND METHODS

#### Study Area

The study was conducted in the mountain range called the Bohemian Forest in central Europe located along the borders of the Czech Republic, Germany (Bavaria), and Austria (**Figure 1A**). This area was affected by significant pollution load, which peaked in the 1970s and 1980s, although the pollution levels were relatively lower compared to other areas of Central Europe. Sulfur emissions reached values of 10–15 µg S m−<sup>3</sup> in comparison to more than 25 µg S m−<sup>3</sup> in the most severely polluted

areas such as the Ore Mts. (**Figure 1A**) 1 . In the 1970s and 1980s, deposition of SO<sup>4</sup> <sup>2</sup><sup>−</sup> and NO<sup>3</sup> <sup>−</sup> was around 125 and 95 mmol m−<sup>2</sup> year−<sup>1</sup> , respectively, based on model results, which is more than 10 and 25 times higher compared to pre-industrial conditions (**Figure 1B**; Kopácek and Hruška, ˇ 2010). To account for this temporal pattern, we divided the data into four periods; before (1900 – 1935), during (1971 – 1989) and after (1998 – 2005 and 2006 – 2011) the period of maximum air pollution. The period after the peak in air pollution was split into two periods to analyze the pattern of forest recovery. By comparing the various tree physiological parameters between these periods we characterize their change over time (hereafter referred to as temporal trend) and association with air pollution.

Our study focuses on natural mountain Norway spruce [Picea abies (L.) Karst.] forest, which are present at high elevations of the mountain range. Norway spruce dominates the tree layer (>90%) with minor components of Sorbus aucuparia L., Abies alba Mill., and Fagus sylvatica L.. The understorey is mostly dominated by Calamagrostis villosa (Chaix) J. F. Gmel., Vaccinium myrtillus L. and Athyrium distentifolium Tausch ex Opiz (Svoboda et al., 2006). A large part of the forest comprises of natural old-growth forest, which developed under a natural disturbance regime and which was recently affected by extensive disturbances (Cada et al., ˇ 2016).

We selected two glacial-lake catchments 65 km apart; Certovo ˇ (49◦ 100N, 13◦ 120E, 1027 – 1343 m.a.s.l.) in the NW and Plešné (48◦ 470N, 13◦ 510E, 1089 – 1378 m.a.s.l.) in the SE part of the mountain range. These lakes and catchments are subjected to

<sup>1</sup>http://www.emep.int

long-term ecosystem monitoring and research (e.g., Kopácek and ˇ Vrba, 2006; Vrba et al., 2014; Kana et al., 2015 ˇ ). The bedrock of the Certovo and Plešné catchments is characterized by poor mica ˇ schist and richer granite, respectively, (Cháb et al., 2007). Soils in the area mostly include acidic leptosols, podsols, and dystric cambisols with a higher level of base saturation in the Plešné catchment (Šantru˚cková et al., 2007 ˇ ). The climate is cold and humid with mean annual temperatures of around 4◦C (Turek et al., 2014), and mean annual precipitation above 1200 mm year−<sup>1</sup> (CRU TS3.10; Harris et al., 2014).

#### Data Collection

Samples for the stable carbon isotopic analyses of wood were collected along a slope transect in both catchments. We extracted increment cores at breast height (1.3 m) from 25 and 6 dominant and healthy (without signs of injury or defoliation) trees in the Certovo and Plešné catchment, respectively. Samples ˇ were analyzed separately for each individual tree. Cores were sectioned into 1–5-year segments, dried and homogenized to a fine powder in a ball mill (MM200 Retsch, Haan, Germany). Carbon isotope composition was determined using an elemental analyzer (EA1110, ThermoQuest, Italy) linked to DeltaXLplus (ThermoFinnigan, Bremen, Germany) for each 1–5-year sample of bulk wood, because the bulk wood provides unbiased estimates in comparison to cellulose composition (Harlow et al., 2006). The <sup>13</sup>C/12C isotopic ratio was calculated relative to the Vienna-Pee Dee Belemnite (VPDB) standard to obtain δ <sup>13</sup>C (McCarroll and Loader, 2004). Trees sampled for isotopic analysis recruited to breast height between 1728 and 1878 (median 1828). We therefore believe that the isotopic composition in the studied period after 1900 was not affected by a juvenile effect.

Samples for biomass increment analysis were collected using a regular grid set across the forest stand. We utilized part of the published datasets of Seedre et al. (2015) and Svoboda et al. (2012) for Certovo and Plešné catchment, respectively. The ˇ regular grid was set across the Certovo catchment, while the ˇ Plešné catchment samples come from a 20 ha plot 2 km from the lake. Increment cores were taken at breast height from 105 canopy trees in both localities. Cores were air-dried, glued to wooden mounts and cut with a razor blade. Ring widths were measured to the nearest 0.01 mm using a sliding table LINTAB and TsapWin software (RINNTECH, Heidelberg, Germany)<sup>2</sup> . Each tree-ring series was cross-dated visually and also using statistical tests implemented in Past4 (Knibbe, 2007; SCIEM, Vienna, Austria). We only used samples which could be reliably cross-dated. All series began before 1900 and median recruitment to breast height was 1859 for the Certovo and 1814 for the ˇ Plešné catchment. Mean sensitivity was 0.21 and 0.20 and mean first order autocorrelation reached 0.86 and 0.88 at the Plešné and Certovo catchment, respectively. Distance to the pith for ˇ cores that did not intersect the pith was estimated using the curvature of the innermost rings and concentric circles printed on transparent foil. Tree age was then estimated by subtracting from the innermost dated ring the estimated distance to the pith divided by the width of the five innermost rings along with 10 additional years to roughly account for the time that a spruce takes to grow from the stump height (30 cm) to the coring height (130 cm).

#### Physiological Parameters

Plants obtain their carbon from the atmosphere, yet their <sup>13</sup>C/12C isotopic ratio is reduced relative to CO<sup>2</sup> in air. This process, called carbon isotope discrimination (against heavier <sup>13</sup>C isotopes), yields variable isotope ratios depending on the specific plant response to the environment (e.g., irradiance, drought, temperature). CO<sup>2</sup> diffusion through stomata and photosynthetic carbon fixation are the main processes involved in carbon isotope discrimination. The lighter <sup>12</sup>C isotope diffuses more easily than the heavier <sup>13</sup>C and is preferred during fixation by the carboxylation enzyme (Farquhar et al., 1982; McCarroll and Loader, 2004). The leaf internal CO<sup>2</sup> concentration is also intimately related to <sup>13</sup>C discrimination because the relative proportion between the CO<sup>2</sup> influx through stomata and photosynthetic rate control both variables. For example, if the photosynthetic rate is higher than stomatal CO<sup>2</sup> influx, the internal CO<sup>2</sup> concentration decreases, and <sup>13</sup>C discrimination decreases likewise because internal proportion of <sup>13</sup>C increases (Farquhar et al., 1982; McCarroll and Loader, 2004). By opening their stomata, plants not only obtain CO<sup>2</sup> for photosynthesis but also loose water vapor. Therefore, plant carbon isotope composition is proportional to water-use efficiency (Farquhar et al., 1982; McCarroll and Loader, 2004).

The carbon isotope ratio of sugars synthetized during photosynthesis is imprinted in plant tissues produced during a given season and we reconstructed the annually integrated spruce physiological parameters (i.e., discrimination against <sup>13</sup>C, Ci and iWUE) from the carbon isotope ratio of wood. We reconstructed carbon isotope composition of tree foliage using the relationship (δ <sup>13</sup>Cfoliage = 1.0523 ∗ δ <sup>13</sup>Cwood − 0.205) presented in Gebauer and Schulze (1991) and expressed in terms of discrimination against <sup>13</sup>C in the atmosphere (113C) while accounting for the anthropogenic increase in <sup>12</sup>C atmospheric concentrations (McCarroll and Loader, 2004):

$$\Delta^{13}\text{C}\ \left(\%\right) \ = \frac{(\\$^{13}\text{C}\_{\text{air}} - \\$^{13}\text{C}\_{\text{foliage}})}{(1 - \\$^{13}\text{C}\_{\text{foliage}})/1000} \tag{1}$$

where δ <sup>13</sup>Cair and δ <sup>13</sup>Cfoliage is the relative isotopic composition of the atmosphere and foliage, respectively, for each 5-year segent. Values of δ <sup>13</sup>Cair were interpolated from the Law Dome ice cores (McCarroll and Loader, 2004). To reveal the amount of CO<sup>2</sup> supply for photosynthesis in the conditions of increasing atmospheric CO<sup>2</sup> concentrations, we calculated the seasonally integrated Ci according to Farquhar et al. (1982):

$$\text{Ci} \begin{pmatrix} \mu \text{mol CO}\_2 \text{ mol}^{-1} \text{air} \end{pmatrix} = \frac{\text{Ca} \times \begin{pmatrix} \Delta^{13} \text{C} \ - \ \text{a} \end{pmatrix}}{\text{(b} \ - \ \text{a} \end{pmatrix}} \quad \text{(2)}$$

where Ca is the atmospheric CO<sup>2</sup> concentration in the relevant years, 'a' and 'b' are constants representing the fractionation during diffusion of CO<sup>2</sup> through the stomata (4.4h) and during carboxylation (27h); values of Ca were obtained from the Law Dome ice cores (McCarroll and Loader, 2004). iWUE reflects the proportion of carbon assimilated in photosynthesis (A) in relation to the potential water loss through stomata (stomatal conductance to water vapor, gw), i.e., iWUE = A/gw. iWUE was calculated according to McCarroll and Loader (2004):

$$\text{iWUE (\mu mol CO}\_2 \text{ mol}^{-1} \text{H}\_2\text{O)} = \text{ (Ca } - \text{ Ci) } \ast \text{ 0.625} \quad \text{(3)}$$

where the constant 0.625 reflects the ratio of CO<sup>2</sup> and H2O diffusivity based on the assumption that A = (Ca – Ci) ∗ gc, where g<sup>c</sup> is the stomatal conductance to CO2.

Biomass increments were calculated for each calendar year between 1900 and 2006 (2007) for the Certovo (Plešné) ˇ catchment based on ring width series, which were converted to diameters of each year proceeding from pith to bark (including the estimated distance to the pith). The diameters were multiplied by a constant of 1.096 to account for bark thickness and water loss. This value was obtained by comparing the diameters in the final year with actual diameter measurements (**Table 1**). Total tree biomass (needles + branches + dry branches + stem + roots) was calculated using the obtained diameters, ages and modeled heights and crown lengths (**Table 1**) based on best available models developed by Wirth et al. (2004). We used two separate models for height/diameter relationships for the Certovo ˇ and Plešné catchments using data from Svoboda and Pouska (2008) and Seedre et al. (2015), respectively, and one model for the crown length/diameter relationship using data published in Seedre et al. (2015) from the Certovo catchment, which ˇ was considered sufficient because the trends between the two catchments were expected to be similar and because crown length has a relatively minor influence on total biomass. Finally, the biomass increment was obtained by subtracting the biomass of

<sup>2</sup>http://www.rinntech.com

Frontiers in Plant Science | www.frontiersin.org June 2016 | Volume 7 | Article 805 |



Total tree biomass is calculated as the sum of biomass of needles, branches, dry branches, the stem and roots.

the preceding year from that of the current year. Mean correlation between individual tree biomass increment series for 1900–2007 was 0.20 for the Certovo and 0.23 for the Plešné catchments. The ˇ inter-annual variability of biomass increment was evaluated using a standard dendrochronological metric called sensitivity (Fritts, 1976), which was calculated as:

$$\text{sensitivity}\_{\text{i}} = \frac{2 \ast |\text{increment}\_{\text{i}} - \text{increment}\_{\text{i}-1}|}{(\text{increment}\_{\text{i}} + \text{increment}\_{\text{i}-1})} \qquad \text{(4)}$$

where |increment<sup>i</sup> – incrementi−1| is the absolute value of the difference between the increment of the current and preceding year.

For the biomass increment-climate correlation analysis we removed the decadal and longer scale trends from the biomass increment series in order to remove the influence of confounding effects such as age and competition dynamics from the data. The biomass increments were first power transformed and the optimal power (p) was computed as

$$
\mathfrak{p} = \mathfrak{l} - \mathfrak{m}, \tag{5}
$$

where m is the slope of the regression of the log10 median ring width against the log10 interquartile range of ring width based on non-overlapping, 10-year segments (Emerson, 1983). We obtained an optimal power of p = 0.14 (N = 3779, r = 0.82). We fit a 30-year cubic smoothing spline with a 50% frequency cutoff to each transformed tree biomass increment series and subtracted the spline from the transformed increments (Cook and Peters, 1997). The detrending procedure was performed using the package dplR (Bunn, 2008) in the R statistical software (version 3.1.1; R Development Core Team, 2014). This procedure removes not only the above-mentioned confounding effects, but also decadal scale climate-related trends (e.g., Wilson et al., 2005). To account for this trend removal, the climatic data were detrended following the same procedure.

Gridded monthly climatic data from the CRU TS3.10 database (Harris et al., 2014) were used for the biomass incrementclimate correlation analysis. We selected two pairs of variables that could potentially limit photosynthesis or water availability [i.e., temperature, cloud cover, precipitation and Palmer drought severity index (PDSI)]. PDSI is a normalized index based on a water balance model of soil moisture where values around zero represent normal conditions and increasingly negative (positive) values represent progressively drier (wetter) conditions. We then performed a preliminary increment-climate correlation analysis (for the whole 1901 – 2007 period) to determine for which months each of the climatic variables show a significant relationship with the annual biomass increment. The months with strongest relationships for each variable were used to examine potential changes in the relationship between each climatic variable and biomass increment over time. If relevant the individualmonths with strong correlation were averaged (or summed for precipitation) into periods to obtain stronger relationship. Highest correlations were identified with growing season temperatures (May to September), cloud cover of the early growing season (May to July), precipitation in the late growing season of the previous year (previous-year July to

previous-year September) and drought PDSI in September of the previous year, which is consistent with other studies of the same forest type (Levanic et al., 2009 ˇ ; Treml et al., 2012; Primicia et al., 2015). Mean correlations of average biomass increment series were 0.38 with temperature, 0.36 with cloud cover, 0.24 with precipitation and 0.19 with drought index (all relationships were statistically significant at p < 0.05 except drought index for the Certovo catchment). Finally, we calculated ˇ the Spearman correlation coefficient between the detrended climatic variable series of the above-specified months and the detrended biomass increment series of each individual tree for the periods defined using temporal trend of air pollution (**Figure 1B**). The use of Spearman correlation was suitable here because of the relatively short periods used and the robustness of Spearman correlation against the effect of single extreme values. Nevertheless, the interpretation of increment-climate correlations should be viewed with caution considering the short periods examined. To support our interpretations we also present running Spearman correlations in the Appendix to acknowledge the temporal variability of the increment-climate relationship.

#### Statistical Analysis

To assess the temporal trend of the physiological parameters in relation to air pollution, we averaged all of the parameters into periods defined by the air pollution trend (**Figure 1B**) so that for each period we obtained one value for each individual tree (see Supplementary Figures S1, S2 and S3 for the original time series). Using R software (version 3.1.1; R Development Core Team, 2014) for statistical analysis, we applied linear mixed effect models in the package lme4 (Bates et al., 2015) to test if the physiological parameters differed between the defined periods. The analysis was followed by Tukey's pairwise comparison in the 'lsmeans' package (Lenth, 2015). We included tree individuals as a random effect and the time periods and catchment as a fixed effect. The catchment effect was significant in some models (see **Table 2**) and we present the effect in figures only for those cases where catchment effect was found to be significant. For 113C we also calculated a more complex model which included the biomass increment, inter-annual variability of the biomass increment, elevation, age, and diameter. Of these variables only the biomass increment showed a significant relationship. We selected the best model using the lowest Akaike Information Criterion (AIC) values and ANOVA comparison. We also calculated a pseudo-R<sup>2</sup> based on Nakagawa and Schielzeth (2013) using the package piecewiseSEM (Lefcheck, 2015). Calculated pseudo-R<sup>2</sup> comprises of marginal (R2m) and conditional (R<sup>2</sup> c) values that account for the proportion of the variance explained by the fixed factors and whole model (i.e., fixed plus random factors), respectively.

#### RESULTS

We found significant temporal changes in spruce physiological parameters in response to the peak in air pollution (**Table 2**), but the temporal trend of the individual parameters was different. The most distinct deviation during the air pollution period was observed in the discrimination against heavier carbon isotope (113C, **Figure 2A**) that transiently decreased and recovered after the pollution period. This result indicates a relatively lower internal leaf CO<sup>2</sup> concentration (in comparison to atmospheric CO2) and most likely stressful conditions due to polluted air. Model estimates (**Table 2**) show a 10% decrease in 113C in response to the pollution and a subsequent recovery of 7%. Surprisingly, considering the atmospheric CO<sup>2</sup> concentration increase (**Figure 1B**), it seems that intercellular CO<sup>2</sup> availability remained relatively stable during the peak in pollution compared to the earlier period (**Figure 2B**). We observed a slight, though insignificant, decrease of Ci by 2% in the pollution period, but a sharp increase in the recent periods by 17 – 24% likely as a result of release from polluted conditions and higher CO<sup>2</sup> availability. The combined effect of air pollution and CO<sup>2</sup> resulted in changes of spruce water use efficiency (**Figure 2C**), which sharply increased (by 40%) during the peak in air pollution and decreased slightly (by 4%) after pollution levels declined. Water use efficiency was highest during the pollution period, but the effect of increased atmospheric CO<sup>2</sup> can be observed in the difference between the iWUE before and after the pollution (34% increase).

Mean tree biomass increment did not show any consistent temporal trend with air pollution or 113C since the periods before and during pollution did not differ significantly (**Figure 3A**). However, we observed a slight increase in the recent period after the pollution diminished, though significant at Plešné (26% increase) and insignificant at Certovo (12% ˇ increase) catchment. On the other hand, inter-annual variability of the biomass increment followed a similar trend to 113C and air pollution (**Figure 3B**), which may suggest increased sensitivity to environmental factors induced by pollution. We observed a 53% increase in inter-annual variability during the pollution period relative to the pre-pollution period and a subsequent decrease of 19% after pollution levels decreased. The biomass increment generally reflects diverse environmental factors (such as age or competition) as displayed in the lowest explained variance in the models of biomass increment and the involvement of catchments ant catchment–period interactions.

At the individual tree level, biomass increment was significantly related to variables obtained from isotopic analysis. **Figure 4** shows the relationship of individual tree biomass increment with 113C and indicates that the growth of trees which fixed more of the heavier isotope was more vigorous. The relationship was similar for periods before and after the peak in air pollution, but was non-significant for the air pollution period (as the model included interactions; **Table 2**). This result suggests that the increased fixation of <sup>13</sup>C during air pollution was more prominent in trees, which grew slowly, and the influence of air pollution on 113C likely overrode other environmental effects (expressed in the biomass increment).

We found no clear trend consistent with air pollution when examining the increment-climate relationships (**Figure 5** and Supplementary Figure S3). Correlations with variables potentially


#### TABLE 2 | Parameter estimates of the best mixed–effect models including tree–individual as a random effect.


The temporal trend was assessed in periods defined based on the temporal pattern of atmospheric pollution (Figure 1B), i.e., before (1900–1935), during (1971–1989) and after (1998–2005 and 2006–2011) the peak in air pollution. The model shows parameter estimates relative to the 1900–1935 period and Certovo catchment. Only ˇ significant effects are shown.

<sup>1</sup>Marginal (proportion of variance explained by the fixed factor, R2m) and conditional (proportion of variance explained by fixed plus random factors, R2c) pseudo-R<sup>2</sup> values were calculated following Nakagawa and Schielzeth (2013).

<sup>2</sup>113C, Ci and iWUE represent carbon isotope discrimination, intercellular CO<sup>2</sup> concentration and intrinsic water use efficiency.

3 Inter-annual variability of biomass increment calculated as the dendrochronological metric called sensitivity (Fritts, 1976).

<sup>4</sup>Spearman correlations of detrended biomass increment-climate relationship.

limiting photosynthesis (i.e., temperatures and cloud cover) were weaker at the beginning of the 20th century but remained stable in later periods. Correlations between biomass increment and variables potentially indicating water stress (i.e., precipitation and drought) increased in the last few years, indicating that severe drought has an effect on the increment. However, we observed no significant effect of air pollution. Catchment effect and catchment–period interactions were again significant and Plešné catchment trees showed generally better climate relationships. Older ages or lower competition pressure at Plešné catchment are the possible explanations for this pattern.

## DISCUSSION

We investigated the extent to which moderate air pollution affected the physiology of Norway spruce. We found that the physiology was significantly influenced, but different

physiological parameters showed variable responses because of the interaction with increasing atmospheric CO<sup>2</sup> concentrations and increasing temperatures due to climate change. As a result, the biomass increment did not change significantly during the peak in air pollution, yet increased slightly in the recent period. Biomass increment was inversely related to carbon isotope discrimination at the individual tree level, but the relationship was lost during the pollution period. As our results did not indicate increased water stress or decreased influence of climate on photosynthesis in response to the pollution (because tree biomass increment did not become more or less sensitive to drought/precipitation or temperature/cloud cover, respectively), we suggest that the direct influence of pollution on stomatal conductance was the main driver of the observed physiological changes (see below).

Carbon isotope discrimination is a sensitive indicator of the physiological response of plants to air pollution (particularly SO<sup>2</sup> deposition; Savard, 2010). We found a decrease of 1.88h 113C in response to the deposition of, particularly, 124 mmol SO<sup>4</sup> <sup>2</sup><sup>−</sup> m−<sup>2</sup> year−<sup>1</sup> and 95 mmol NO<sup>3</sup> <sup>−</sup> m−<sup>2</sup> year−<sup>1</sup> . This 113C decrease is in the middle of the range of values reported in other studies, where the decrease ranged from 1h to more than 3h in both conifers and deciduous species (Niemelä et al., 1997; Rinne et al., 2010; Savard, 2010; Thomas et al., 2013). The 113C decrease exceeds the typical range of natural variation (usually less than 0.5h) and is proportional to the pollution load (Niemelä et al., 1997; Savard, 2010), which was at moderate level in our study area. Our results demonstrate that changes in 113C can be used as a sensitive indicator of acid pollution stress far earlier than tree mortality, reduction of growth or possibly even before reduction in the rate of photosynthesis occurs (Thomas et al., 2013). We also provide evidence that air pollution significantly affected tree physiology throughout central Europe and not only in the most polluted regions such as the "Black Triangle" as suggested by Kandler and Innes (1995).

The effects of air pollution on tree physiology interacted with other environmental changes such as the increase in atmospheric CO<sup>2</sup> or temperature. Observed 20th century increases of iWUE in many ecosystems can be related not only to the increase in atmospheric CO<sup>2</sup> (Peñuelas et al., 2011; Saurer et al., 2014), but also to increased atmospheric pollution. The iWUE significantly increased by 3.8 µmol CO<sup>2</sup> mol−<sup>1</sup> H2O in the polluted conditions with 339 ppm CO<sup>2</sup> in comparison to the recent postpollution conditions when the concentration of atmospheric CO<sup>2</sup> reached 381 ppm. Increased atmospheric CO<sup>2</sup> can compensate for the negative effects of air pollution because the Ci can remain similar (Thomas et al., 2013). Despite the major iWUE increase in the second half of the 20th century in our study location, the biomass increment did not increase correspondingly, suggesting that there is no causal relationship between the two variables and that an increase in iWUE does not necessarily mean increase in the carbon storage of the forest (Peñuelas et al., 2011).

Compensation of the negative effect of pollution by increased ambient CO<sup>2</sup> concentrations which produced comparable Ci before and during the pollution period could partly explain why the biomass increment did not change significantly. Similarly to CO2, air temperature, which is a limiting factor for spruce biomass increment in our study area, increased during the 20th

century and could have therefore also mitigated the negative effect of air pollution (Levanic et al., 2009 ˇ ; Treml et al., 2012; Primicia et al., 2015). It is also possible that the carbon partitioning pattern of the trees may have changed in favor of above-ground biomass during the pollution period resulting in unchanged biomass increment derived from stem diameter increments. The negative effects of high nitrogen load on trees (Högberg et al., 2006) and particularly on below-ground carbon allocation (root growth and mycorrhiza; Jentschke et al., 2001; Nilsson and Wallander, 2003) have been documented. Similarly, rather than affecting biomass increment in terms of volume, air pollution could affect wood density (Sander et al., 1995). Generally, the stem volume increment is probably affected when air pollution exceeds some threshold value, as occurred in the most polluted part of central Europe (Hauck et al., 2012; Rydval and Wilson, 2012).

However, the sensitivity of different tree species to pollution is likely variable. Silver fir (A. alba) was found to be highly sensitive with growth significantly affected even in the moderately polluted region of southern Germany near our study area (Elling et al., 2009). In addition to the effect on the growth increment, the fir carbon isotope ratio changed by about 3h in this area and the oxygen isotope ratio also showed a distinct effect of pollution on fir physiology (Boettger et al., 2014). The physiology of other coniferous (e.g., pine Pinus sylvestris) and deciduous (e.g., oak Quercus robur) species was also found to be influenced by pollution (Rinne et al., 2010). Pinus sylvestris was observed to react sensitively even to low pollution loads (Niemelä et al., 1997). However, direct comparisons of the response of different tree species to pollution, excluding the influence of site, climatic and environmental conditions and variability in pollutant deposition, are required, particularly because these factors significantly influence the specific reaction of each species to the pollution load. For instance, mountain areas with shallow and acidic soils are the most sensitive to acid pollution and conifers, which have a relatively large leaf area that remains exposed during the winter, are subjected to higher total amounts of pollutant – a factor unrelated to their physiological resistance (Kopácek and Hruška, ˇ 2010; Greaver et al., 2012).

The biomass increment increase in the recent period may not be related only to the direct effects discussed above, but also to the complex ecosystem release from the acidic conditions. These are related for example to increased decomposition of organic matter accumulated during the pollution period, its utilization (cessation of nitrate leaching) and improved nutrition as indicated by needle nitrogen concentrations (Oulehle et al., 2011). Changes in between tree competitions related to disturbance events could also play a role in determining changes in biomass increment and particularly in the higher recent growth increase at Plešné catchment, which was affected by bark beetle outbreak since 1990s (Svoboda et al., 2012; Janda et al., 2014). As our results indicate that extreme droughts periodically influenced the increment, an increase in the frequency of drought years due to climate change can also lead to the increasing importance

of drought as a stress factor that negatively influences spruce biomass increment (Primicia et al., 2015).

Inter-annual variability of the biomass increment increased significantly during the pollution period. This may be related to the increased sensitivity of pollution affected trees to other stress factors such as frost, insects or drought (Kandler and Innes, 1995). Moderate pollution loads probably do not directly cause widespread tree damage as observed in most polluted regions, but increase tree vulnerability to other stressors, which can in combination cause forest decline (Kandler and Innes, 1995). Increased variability is thought to be an indicator of increased stress or an "early warning signal" for a critical state transition (Scheffer et al., 2009). At the same time we cannot discount the possibility that this result was related to a series of favorable and unfavorable growth years which may have occurred during the air pollution period.

Air pollution can influence trees via several pathways, i.e., directly through foliage or indirectly through the soil. Soil acidification decreases soil nutrient availability and often mobilizes aluminum, which is toxic to plants and can obstruct nutrient uptake (Jentschke et al., 2001; Hruška et al., 2012). Our biomass increment-climate correlation analysis does not support the hypothesis that possible root weakness caused water stress in our study site. Nutrient deficiency can lead to a reduction in photosynthesis rate. While decreased capacity of photosynthesis alone would lead to an increase in 113C (followed by increased Ci), we instead observed a decrease in 113C and therefore refute the notion that soil acidification would be the main driver of the observed physiological changes (Viet et al., 2013).

A direct effect of pollution on foliage can decrease the rate of photosynthesis and/or stomatal conductance (Meng et al., 1994). Our results, which show a reduction of 113C during the air pollution period, do not support the pollution influence via reduced photosynthesis alone, but are instead in line with decreased stomatal conductance (Farquhar et al., 1982; McCarroll and Loader, 2004). No consistent trend in the correlation between biomass increment and temperature or cloud cover, which would have also indicated photosynthesis limitation, supports the idea that stomatal conductance was the main driving force of the observed 113C decrease. Minor limitation of photosynthesis in response to either the direct or indirect pollution impact could be indicated by physiological patterns such as non-significantly decreased biomass increment and Ci or greatest iWUE during pollution period. Increased dark respiration was also proposed to be related to air polluted conditions and could have also played a role (Savard, 2010). We suppose that photosynthesis limitation and root weakening are of higher importance in locations with different site conditions or a higher pollution load as suggested by some studies showing decreased responses to temperature (Sander et al., 1995; Elling et al., 2009) or in more sensitive species (Boettger et al., 2014). Our interpretation is also supported by Kandler and Innes (1995) that reviewed in situ measurements of gas exchange done around the year 1990 and found no influence on photosynthesis rate in moderately polluted areas, but significant effect in most polluted areas. We therefore conclude that moderate pollution levels have the greatest effect directly on stomatal conductance (Thomas et al., 2013), which increases tree stress and sensitivity to other stressors.

The negative relationship between 113C and biomass increment was significant in the periods before and after the pollution, while it was lost during the pollution period. The positive or negative relationship between 113C and growth was used to indicate stomatal or non-stomatal (i.e., carboxylation rate) limitation of assimilation in trees (Viet et al., 2013). The relationship was positive in P. strobus limited by soil water availability and, presumably, stomatal conductance (McNulty and Swank, 1995), while it was negative in P. densiflora, where photosynthesis was limited by soil acidity. In our case, air temperature is the main climatic variable driving physiological processes of mountain Norway spruce growing in non-polluted conditions (Levanic et al., 2009 ˇ ; Treml et al., 2012; Primicia et al., 2015) and the negative relationship between 113C and biomass increment supports the statement about the dominant control of the rate of photosynthesis on 113C in non-polluted conditions. This implies that trees exposed to more sunlight grow better, utilize intercellular CO<sup>2</sup> faster and their 113C is therefore lower in comparison to more slowly growing shaded trees. We therefore suggest that the loss of the negative relationship between biomass increment and 113C during the pollution period in our study area indicates a loss of photosynthetic rate as the dominant control of 113C in favor of stomatal conductance. This supports our conclusion that moderate pollution predominantly influenced studied physiological processes by reducing stomatal conductance.

## CONCLUSION

Moderate levels of atmospheric pollution and particularly sulfate and reactive nitrogen acid deposition significantly affected the physiology of mountain Norway spruce in central Europe. We observed a complex response of spruce trees to pollution with 113C being the most sensitive indicator of the pollution effect. The atmospheric pollution effects interacted with those of increasing atmospheric CO<sup>2</sup> and temperature resulting in unchanged intercellular CO<sup>2</sup> concentrations abruptly increased water use efficiency and unchanged biomass increment, which could also partly result from changes in carbon partitioning. Our biomass increment-climate correlation analyses did not indicate any clear pollution-related changes in water stress or photosynthetic limitation due to climate. We therefore concluded that the direct effect of moderate pollution on stomatal conductance was likely the main driver of the observed physiological changes. We also assume that the shift from the dominant control of the photosynthetic rate to the stomatal conductance on 113C during the pollution period caused the loss of the negative relationship between biomass increment and 113C. This mechanism probably caused weakening of the spruce trees and increased sensitivity to other stressors, as indicated by increased inter-annual variability of biomass increment.

### AUTHOR CONTRIBUTIONS

fpls-07-00805 June 8, 2016 Time: 16:15 # 11

HŠ, LK and MSv designed the research and data collection. JŠ and LK performed the laboratory analysis. VC, HŠ, LK and ˇ MSe did the calculations and statistical analysis. VC organized ˇ the manuscript preparation with the contribution of all the authors.

#### FUNDING

The study was funded by the Ministry of Education (project COST CZ no. LD13064) and the Czech Science Foundation (project GACR no. P504/12/1218). JŠ was supported by the ˇ project GACR no. 14-12262S of the Czech Science Foundation. ˇ

#### REFERENCES


#### ACKNOWLEDGMENTS

The study was performed under the framework of the COST Action FP1106 "STReESS - Studying Tree Responses to extreme Events: a SynthesiS." We thank the staff of the National Park and Protected Landscape Area Šumava for their permission to conduct this research. We are grateful to M. Rydval for language editing. Suggestions of both reviewers greatly improved the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.00805

SO<sup>2</sup> levels in the Harz Mountains, Germany. Environ. Pollut. 164, 132–141. doi: 10.1016/j.envpol.2012.01.026



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Cada, Šantr ˇ u˚ˇcková, Šantru˚ˇcek, Kubištová, Seedre and Svoboda. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Imprint of Extreme Climate Events in Century-Long Time Series of Wood Anatomical Traits in High-Elevation Conifers

#### Marco Carrer<sup>1</sup> \*, Michele Brunetti<sup>2</sup> and Daniele Castagneri<sup>1</sup>

<sup>1</sup> TeSAF Department, Università degli Studi di Padova, Padova, Italy, <sup>2</sup> Institute of Atmospheric Sciences and Climate, National Research Council, Bologna, Italy

#### Edited by:

Achim Braeuning, Friedrich-Alexander University Erlangen-Nürnberg, Germany

#### Reviewed by:

Eryuan Liang, Institute of Tibetan Plateau Research, China Ze-Xin Fan, Xishuangbanna Tropical Botanical Garden, China

> \*Correspondence: Marco Carrer marco.carrer@unipd.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 12 February 2016 Accepted: 03 May 2016 Published: 18 May 2016

#### Citation:

Carrer M, Brunetti M and Castagneri D (2016) The Imprint of Extreme Climate Events in Century-Long Time Series of Wood Anatomical Traits in High-Elevation Conifers. Front. Plant Sci. 7:683. doi: 10.3389/fpls.2016.00683 Extreme climate events are of key importance for forest ecosystems. However, both the inherent infrequency, stochasticity and multiplicity of extreme climate events, and the array of biological responses, challenges investigations. To cope with the long life cycle of trees and the paucity of the extreme events themselves, our inferences should be based on long-term observations. In this context, tree rings and the related xylem anatomical traits represent promising sources of information, due to the wide time perspective and quality of the information they can provide. Here we test, on two high-elevation conifers (Larix decidua and Picea abies sampled at 2100 m a.s.l. in the Eastern Alps), the associations among temperature extremes during the growing season and xylem anatomical traits, specifically the number of cells per ring (CN), cell wall thickness (CWT), and cell diameter (CD). To better track the effect of extreme events over the growing season, tree rings were partitioned in 10 sectors. Climate variability has been reconstructed, for 1800–2011 at monthly resolution and for 1926–2011 at daily resolution, by exploiting the excellent availability of very long and high quality instrumental records available for the surrounding area, and taking into account the relationship between meteorological variables and site topographical settings. Summer temperature influenced anatomical traits of both species, and tree-ring anatomical profiles resulted as being associated to temperature extremes. Most of the extreme values in anatomical traits occurred with warm (positive extremes) or cold (negative) conditions. However, 0–34% of occurrences did not match a temperature extreme event. Specifically, CWT and CN extremes were more clearly associated to climate than CD, which presented a bias to track cold extremes. Dendroanatomical analysis, coupled to high-quality daily-resolved climate records, seems a promising approach to study the effects of extreme events on trees, but further investigations are needed to improve our comprehension of the critical role of such elusive events in forest ecosystems.

Keywords: cell diameter, cell number, cell-wall thickness, dendroanatomy, extreme climate events, tree ring, xylem anatomy

## INTRODUCTION

fpls-07-00683 May 18, 2016 Time: 13:42 # 2

It has become manifest that climate change is producing significant effects on natural systems and human society worldwide. This condition will likely worsen as, despite different emission scenarios, all current projections indicate a warming trend in the future, associated to a corresponding change in the frequency, severity, and nature of extreme events (IPCC, 2013). Small changes in the mean or variance of a climate variable may lead to disproportionally large changes in the frequency of extremes, representing a severe challenge for living organisms to respond adaptively (Gutschick and BassiriRad, 2003). These events are now recognized as major drivers of current and future ecosystem dynamics (Smith, 2011; Frank et al., 2015). Biological responses to extreme weather events can vary, and even be reversed among different species or growth stages in the same habitat, but the impact can be pervasive and propagates through the ecosystem with a cascade of side effects (Parmesan et al., 2000).

Characteristic infrequency and stochasticity challenges investigations of extreme events, and is the reason for the lack of a comprehensive, precise and biologically meaningful definition of them (Gutschick and BassiriRad, 2003). Besides, extreme events occur in a wealth of divergent types (e.g., heat waves, hurricanes, droughts, ice storms, etc.), at multiple time scales, are highly context dependent, and their effects, rather than linear and monotonic, are usually non-linear and threshold based (Knapp et al., 2008; Smith, 2011; Bahn et al., 2014; Frank et al., 2015). This calls for research focused on extreme events and their consequences in different ecosystems at multiple time, spatial and magnitude scales by collecting evidence in both natural and controlled conditions (Jentsch et al., 2007; Reyer et al., 2015). Up to now, most studies have been conducted in controlled environments, on short-lived species or early life stages (Marchand et al., 2005; Wang et al., 2008; De Simon et al., 2013), or tracking the effects of a single event (Ciais et al., 2005; Piayda et al., 2014; Balducci et al., 2015; Julio Camarero et al., 2015). Due to differences in longevity among growth forms and climatic sensitivities among early and adult life stages, and the idiosyncratic behavior of any extreme events, it is difficult to extrapolate these findings to forest ecosystems (Teskey et al., 2015). To assess the effect of extreme climate events on trees and forests our inferences should be based on long-term observations just to cope with the typical life cycle of these organisms and the paucity of the events themselves (Jentsch et al., 2007; Anderegg et al., 2015).

To date, working with time series seems the most straightforward approach to deal with the effect of extreme climate events on long-lived organisms such as trees (Jentsch et al., 2007). Within this context, tree-ring analysis can deliver most of its potential considering the intrinsic association with annually resolved and absolutely dated chronologies of different tree-ring parameters. Building up to millennia-long tree-ring series allows a vast time dimension, an essential perspective to soundly assess the effects and potential changes in frequency, intensity, and nature of extreme events (Babst et al., 2012). Actually, the study of extreme event effects on tree-ring width and structure has a long tradition in tree-ring science, thanks to relevant tree-ring features that mark a direct signature of climate extremes such as frost rings (Rhoads, 1923; Harris, 1934; Lamarche and Hirschboeck, 1984), intra annual density fluctuations (Schulman, 1938), white- or blue-rings (Waito et al., 2013; Piermattei et al., 2015), resin-duct density (Wimmer and Grabner, 1997) – see Schweingruber et al. (1990) and Schweingruber (1996) for a thorough review. However, despite the wealth of studies there are still two major facets that could be improved: (i) the lack of a rigorous causal relationship of climate extremes on wood structure, which has led to an anecdotal rather than systematic analysis in most of the investigations and (ii) the time resolution, still mostly bounded at yearly or monthly level at best, leaving the extreme events occurring at shorter time scales largely unscrutinized.

Dendroanatomy, the study of wood-anatomical traits with a dendrochronological approach, has been expanding recently thanks to technological and technical advances (Wegner et al., 2013; von Arx and Carrer, 2014). Working at anatomical level can provide new clues for understanding climate extreme effects on tree growth and functioning: first, detailed assessment of xylem traits gets closer to the functional and physiological tree aspects, making the link between pattern and process more apparent. Second, investigating intra-ring anatomical traits, the time resolution of the analyses can potentially significantly increase compared to a typical tree-ring approach, i.e., reaching the sub-monthly level. In this study, we test the potential of dendroanatomy to deal with extreme climate events. In particular, our research question was: is it possible to detect the imprint of past temperature extremes occurred during the growing season at wood anatomical level? To address this question, we collected wood samples from two high-elevation conifers (Larix decidua and Picea abies) in the Alps and analyzed some xylem anatomical parameters, namely the number of tracheids per ring, their diameter and wall thickness, for any potential associations with extreme temperature events.

#### MATERIALS AND METHODS

#### Study Site

Analyses were performed on two conifer species: Picea abies (L.) Karst. (Norway spruce), evergreen, and Larix decidua Mill. (European larch), deciduous. Both the species are widespread in the Alps, and reach the treeline, which in the Eastern Italian Alps occurs at around 2200 m a.s.l. The study site was located at an elevation of 2100 m a.s.l., close to Cortina d'Ampezzo (46◦ 300 N, 12◦ 07<sup>0</sup> E). At the valley bottom, mean annual precipitation is 1080 mm, with a maximum in June. Daily maximum temperature averages 20.8◦C during July, and 3.1◦C in January (Cortina d'Ampezzo meteorological station, 1275 m a.s.l., 1926–2011).

#### Instrumental Climatological Data

The availability of long and reliable temporal series of meteorological variables at a fine space-time resolution is crucial when the analysis target goes beyond the common climate-ring width associations and aims at investigating climate

influence on xylem cell structure. However, global or regional climatological datasets frequently lack representativeness at local scale, especially in areas with rugged terrain. We therefore reconstructed climate variability more accurately taking into account the relationship between meteorological variables and the topographical settings of the region. The climate information comes from the daily minimum and maximum temperature series of the Cortina D'Ampezzo station, covering the 1926–2011 period, and from synthetic records of monthly minimum, mean, and maximum temperatures covering the period 1800–2011 reconstructed for the specific site location.

As well as for any other meteorological measure, physical signals in raw temperature data series are often hidden behind non-climatic noise caused mainly by station relocation and changes in instruments, in the environment around the station or in the observing conventions. The noise represented by nonclimatic disturbances in the raw data is often of the same order of magnitude as the target climate signal, or even greater. For this reason, data homogenization (i.e., the procedure to remove nonclimatic signals) is crucial to ensure the reliability of the dataset in representing the true climatic signal.

The homogenization approach used in this study was the same as that discussed in Brunetti et al. (2006), but adapted to daily resolution. We checked monthly minimum and maximum temperature series of Cortina d'Ampezzo separately, by means of a multiple application of the Craddock test (Craddock, 1979), using as references the nearest series available from Brunetti et al. (2006) and Simolo et al. (2010). Monthly correcting factors were estimated using at least three reference series among the neighboring most correlated ones and performing a trigonometric smoothing of the correcting factors. Daily adjustments were then calculated by fitting a trigonometric function to monthly factors, resulting in 366 daily correcting factors.

Synthetic records of monthly minimum, mean and maximum temperatures covering the period 1800–2011 were reconstructed to be representative of the specific location of the sampled site by means of the anomaly method (New et al., 2000; Mitchell and Jones, 2005) as described in Brunetti et al. (2012). The spatio-temporal structure of the signal of a meteorological variable over a given area can be described by the superimposition of two fields: the climatological normals over a given reference period (i.e., the climatologies), characterized by remarkable spatial gradients, and the departures from them (i.e., anomalies), generally characterized by higher spatial coherence and linked to climate variability. Climatologies and anomalies were reconstructed in a completely independent way from each other and based on different data sets (high spatial density and limited temporal coverage for the climatologies, and low spatial density but long temporal coverage and accurate homogenization for the anomalies). Climatologies and anomalies were reconstructed estimating the local temperature-elevation relationship (Brunetti et al., 2014) and using weighted averages of high-quality and homogenized neighboring series (Brunetti et al., 2006), respectively. Finally, the two fields were superimposed to obtain a temporal series in absolute values representative of the site location.

## Samples Collection and Processing

Two increment cores were extracted at breast height with a Pressler borer from 15 trees of each species. Ring widths were measured to the nearest 0.01 mm using TsapWin (Rinntech, Heidelberg, Germany) and then crossdated to match each tree ring with its year of formation (Stokes and Smiley, 1968; Holmes, 1983).

Eight cores for spruce and six for larch (corresponding to 14 trees) were then selected among those without visible faults such as nodes, reaction wood, rotten or missing parts. These cores were split in 4–5 cm long pieces for anatomical measurements. A rotary microtome (Leica, Heidelberg, Germany) was used to obtain 15–20 µm thick transversal micro-sections, which were stained with safranin (1% in distilled water) and fixed on permanent slides with Eukitt (BiOptica, Milan, Italy). Digital images were captured with a light microscope at 40× magnification (Nikon Eclipse 80 mounted with distortionfree lenses), and stitched together with PTGui software (New House Internet Service B.V., Rotterdam, The Netherlands). The images were then processed with the image analysis software ROXAS v2.1 (von Arx and Carrer, 2014). After a brief manual fine-editing to remove objects wrongly identified as cells, the software automatically provided the lumen and wall size together with the relative position within the dated annual ring of each cell in the image. With this information, we divided each ring into 10 sectors of equal width along the tangential direction and created the tree-ring anatomical profiles. These profiles represent the variation of different anatomical parameters within each ring (**Figure 1**). As an example, sector 1 comprises all the cells with a distance from the initial earlywood ring border to 10% of the total ring width. For each sector, we computed (1) mean cell wall thickness (CWT) and (2) mean cell diameter (CD). For each tree, we therefore built 10 CD and 10 CWT time series, which ranged from earlywood to latewood, representing distinct time windows within the growing season. We also computed (3) the total number of cells per ring (CN), and corresponding individual CN series.

For each parameter, species and sector, we built mean chronologies by computing the bi-weight robust mean from the detrended individual time series. Indeed, most of the anatomical parameters (**Figure 2**) presented a typical age/size trends (Carrer et al., 2015), which can alter correlations with climate (Cook et al., 1990). Individual series were therefore detrended by fitting a stiff function (power function or 200 years cubic spline with 50% frequency cut-off) to raw data, and dividing observed by expected values.

Some descriptive statistics, within the 1800–2011 common period, have been adopted to compare and describe the resulting chronologies. These are the mean sensitivity (MS) and mean correlation between the series (Rbar) to assess the strength of the high frequency variability within the series and the level of yearby-year growth variations shared by trees of the same species, respectively.

FIGURE 1 | Tree-ring anatomical profiles during a cold (1926, the first year with an extreme cold June in the 1926–2011 period) and a warm (1931, the first year with an extreme warm June in the 1926–2011 period) year in a spruce (PA04) and a larch (LD04) tree. X-axis represents the relative distance from the ring border, and the corresponding sector (upper axis and different colors). Dots are individual cell values, solid and dotted lines represent the mean and standard deviation values. Note different Y-axis scales for the four anatomical traits.

## Anatomical-Traits Response to Climate Extremes

Previous analyses demonstrated that maximum spring-summer temperature is the main climate factor affecting spruce and larch radial growth processes at this elevation (Carrer and Urbinati, 2004; Rossi et al., 2007; Castagneri et al., 2015b). We therefore investigated anatomical-traits response to springsummer temperature at two distinct scales of detail.

In the first analysis, we investigated how climate extremes, assessed at monthly scale over the period 1800–2011, influence the anatomy of tree rings. We initially explored mean responses to inter-annual climate variability in order to identify the month (or monthly aggregate) when climate has the largest influence on CN, CD, and CWT and computed correlation between chronologies (whole ring CN and the 10 sectors CD and CWT) and temperature from May to October. The monthly period considered largely comprises the beginning and the end of the typical growing season for both species in the area (Rossi et al., 2008). Then, we computed the yearly tree-ring profiles of CD and CWT from the first to the last sector together with the CN value for all the rings. Finally, we contrasted the median CN value and median CD and CWT profiles for all the years in 1800– 2011, with those computed for the 10 coldest and 10 warmest years for the month (or monthly aggregate) selected with the correlation analysis. To test for a potential carry over effect, we also considered the anatomical traits in the years following the extreme events.

In the second analysis, using daily temperature records over the period 1926–2011, we were able to refine the analyses to better cope with the short-term climate influence on cell parameters, and to go beyond the artificial aggregation of climate variability into months (Fonti et al., 2013). In particular, we scrutinized how CN, CD, and CWT positive and negative extremes within intra-ring sectors were associated to short-term specific climate conditions, First, we calculated for each parameter the correlation with temperature to identify the period within the season with the highest sensitivity to climate variability. For CD, determined by cell enlargement which lasts 14–25 days in the first earlywood tracheids to very few days in the last cells (Rossi et al., 2008), daily maximum temperature data were averaged over a 15-days moving window shifted at daily step, and running Pearson correlations with CD sector chronologies were computed between May 1st and October 31st of the ring formation year. For CWT, related to the wall thickening phase that lasts up to 40 days in these species (Rossi et al., 2008), the same procedure was applied using 30-days moving windows. Similarly, as CN depends on growth during the whole growing season, 30-days moving windows were used. This allowed identification with 1-day precision, of the 15- or 30-days period when the temperature sensitivity of each anatomical parameter peaks. We then considered the first and last 10 values (extremes in anatomical traits) of CN, CD and CWT, corresponding to the tails of their distributions, and pinpointed the corresponding calendar year. In parallel, we ranked the years from warmest to coldest, based on the temperature during the selected time window and divided the distribution in four quartiles. Lastly, we applied a χ 2 test to evaluate the match between temperature quartiles and the anatomical extremes. This allowed us to assess whether extreme values in anatomical traits were associated with corresponding short-term extreme temperature fluctuations during the growing season.

## RESULTS

#### Tree-Ring Anatomical Traits

Dendroanatomical analysis allowed us to investigate xylem anatomical variations in Norway spruce and European larch over the last two centuries (**Figure 2**). CN was much higher in spruce than in larch rings while larch had larger cells (CD)

#### TABLE 1 | Descriptive statistics for mean cell number (CN), mean cell diameter (CD), and mean cell wall thickness (CWT) chronologies in spruce and larch over the period 1800–2011.


CN statistics refer to the whole ring while CD and CWT are related to each of the 10 sectors. MS is chronology mean sensitivity; Rbar is the mean correlation between the series.

in the first part of the ring (**Table 1**; **Figure 1**). Nonetheless, profile comparison evidenced some similarities between species. The largest CD usually occurred in sector 3 (i.e., at 1/4 – 1/3 of the ring profile) and slightly decreased until sector 7, however, toward the latewood, CD decline was sharper in larch. In the two species, CWT was similar in the first sectors, but larger in larch

for the last sectors. CWT profile in both species increased around sectors 6–8. However, this increase was smoother in spruce than in larch, which evidenced a sharp wall thickening from sectors 6 to 9, while CWT in sector 10 was, on average, similar to sector 9 (**Table 1**). Descriptive statistics of anatomical series indicate that, in general, both CD and CWT MS and inter-tree correlation (Rbar) increased along the ring profile (**Table 1**).

Tree-ring anatomical profiles showed strong inter-annual variability, as exemplified in **Figure 1**. Within the same individual, tree-ring anatomy (profile configuration, cell number, cell size, and wall thickness within the sectors) can considerably change from years with cold (e.g., 1926, the first of 10 coldest June years in the 1926–2011 period) to warm (e.g., 1931, the first of 10 warmest June years in the 1926–2011 period) climate conditions.

## Ring Profiles and Monthly Climate Extremes

Considering climate variability with a monthly resolution during 1800–2011, CN in both larch and spruce was mainly related to June–July temperature (**Figure 3**). Furthermore, associations between climate and CD were quite similar in the two species. May temperature had a slight negative correlation with CD, while June temperature had a stronger positive correlation in all tree-ring sectors. Temperatures during spring and summer months were positively associated with CWT in both species, especially for the last sectors. In spruce, the highest correlations occurred for August and September temperature, while in larch they occurred for July temperature.

Compared to the 1800–2011 median value, CN was larger in years with warm June–July, and smaller in cold years for both species (**Figure 4**). Deviation from the median was much stronger for spruce during cold years. Very cold June temperature induced a reduction in the CD of both species with respect to the median profile (1800–2011) and this deviation was more evident and consistent along the whole profile in larch. On the contrary, CD profiles during years with extremely warm June barely differed from the median ones. Cold conditions (during August and September for spruce, July for larch) affected CWT more than warm conditions inducing a clear divergence from the median in the last sectors toward the latewood, especially in spruce.

The year following the extreme event, all the anatomical profiles were similar to the median one (**Figure 5**). Only CN in larch, the year after an extremely warm June, was sensibly above the median.

#### Intra-Ring Anatomy and Short-Term Climate Extremes

Investigation on climate association with ring anatomy at a finer scale (1926–2011 moving windows with 1-day step) showed a time shift of correlations along the ring profile, with maxima in the first sectors occurring earlier compared to maxima in the late sectors (correlations with daily maximum temperature are shown in Supplementary Figure S1). That is, detailed analysis was able to capture intra-ring differences in the xylem response to climate. After identification of the period with the greatest response to temperature in each sector, we scrutinized how anatomical traits were related to temperature conditions in that specific period. This analysis revealed that the 10 largest CN, CD, and CWT were usually associated with warm conditions, while smaller values were associated to cold (see Supplementary Figure S2 for results at the single sector level). On average, considering results for both species, all anatomical traits and sectors, 79% of positive extremes

fell in years with temperature above the median, and 52% in the warmest quartile, while the lowest CD and CWT values mainly occurred during cold years (83% of negative extremes fell in years with temperature below the median, and 60% in the coldest quartile; **Figure 6**; p < 0.001 for all the cases, χ 2 test). Looking at the single traits, CN extremes in larch were more strictly related to temperature than those in spruce. For CD and CWT, negative extremes were more related to cold conditions than the opposite. In both species, about half of the negative CD extremes occurred in the first (i.e., coldest) quartile of temperature distribution (**Figure 6**). About 2/3 of negative CWT extremes occurred in the first temperature quartile, and 92% (spruce) and 84% (larch) in years with temperature below the median.

#### DISCUSSION

The increasing occurrence and intensity of future extreme climate events is expected to have important aftereffects on forest ecosystems and the carbon cycle worldwide; this explains why a better knowledge on tree response strategies at different level is a scientific priority (Allen et al., 2010; Anderegg et al., 2015; Bussotti et al., 2015). This work represents a pilot study where we introduced some novel methodological and analytical aspects to gain a better understanding of the effects of climate extremes on xylem traits, especially in temperature-limited environments.

We sharpened the focus on several anatomical traits that are independent of one another and mostly related to the different xylem functions of water transport (CD) and mechanical support (CWT), with CN related to both. These traits are also related to different phases of xylogenesis: cell production via cambial division (CN), cell enlargement (CD), and wall thickening (CWT).

#### Anatomical Response to Monthly Climate Extremes Over Two Centuries

Investigation on anatomical traits along tree-ring series allowed us to cover a long time period. In particular, we were able to assess the xylem-anatomical responses of two species to the 10 coldest and warmest years in the last two centuries, providing a robust assessment of the effect of extremes on xylem structure. Indeed, climate extremes are rare by definition, therefore studies not covering a long time span can hardly investigate more than one or very few events. This could hamper a complete mechanistic view of the cause and effect relationships, as the frequently idiosyncratic responses to rare events such as droughts or heatwaves has already been demonstrated by previous studies comparing two or more events (Lloret et al., 2011; Castagneri et al., 2015a; George et al., 2015).

Despite differences in their ecology and physiology, both species, typical of the subalpine environment, are well adapted to cold conditions (Tranquillini, 1979). Nonetheless, temperature and climate extremes influence wood formation at this elevation (Körner, 2006; Vaganov et al., 2006). With climate assessed at monthly resolution, in general we observed a rather similar

response to mean and extreme events, with a common bias toward cold extremes, particularly evident for CD. This kind of unidirectional sensitivity to extreme cold seems in contradiction to the common knowledge of tree responses in temperature-limited conditions. However, this can be explained considering the above-mentioned main function of hydraulic transport related to the CD trait. Along and across the stem, monotonic change in lumen size has been reported to be almost universal within vascular plants, being strictly connected with stem elongation. It represents the most efficient anatomical adjustment to balance hydraulic path-length resistance with the progressive growth in height (Anfodillo et al., 2006; Carrer et al., 2015). The tight link between stem height and lumen size explains the relative unresponsiveness of CD to high temperature. Larger conduits would be unnecessary to efficiently transport water in trees with the same height but, on the contrary, they would increase the risk of winter embolisms (Mayr et al., 2006). Therefore, to observe a positive effect of high temperature on CD a fairly long series of warm years would be needed to permit the trees to grow taller and not just a single extreme event. On the contrary, low temperature in early summer can likely limit the mobilization and fixation of photosynthates (Petit et al., 2011; Körner, 2015), which affects the enlargement of primary wall and results in cells with reduced diameter.

Both the species and most of the anatomical traits did not show any carry-over effect respect previous-year extreme climate events. Previous growing season conditions can significantly affect tree growth of many conifer species, as demonstrated by several studies (Frank and Esper, 2005; Babst et al., 2013). Similarly, extreme climate events can leave a long-lasting legacy effect on the growth rates of many tree species, especially in harsh environmental conditions (Babst et al., 2012; Anderegg et al., 2015). However, most of past studies investigated just the treering width, mostly related to cell number. Looking at different anatomical traits, it seems that larch and spruce trees close to the treeline did not present any carry-over effect in relation to unusual warm and cold short-term spells occurred in the previous year growing season. Probably the extreme events we defined were too short to induce inertial effects able to extend their impact beyond the short time span when they occurred.

Considering CWT, the impact of warm and cold monthly climate extremes seems overall quite balanced and more evident in spruce in the last ring sectors. This is likely due to the longer time of wall thickening phase in the xylogenetic process, especially in the last cells of the ring, and to the higher plasticity of this anatomical trait with respect to lumen size. The better capacity of tree-ring maximum density, a parameter highly related to CWT, to track temperature variability with respect to ring width (Briffa et al., 2002), which is closely correlated to CN, seems to be in line with these results.

#### Anatomical Response to Short-Term Climate Extremes

In this study, we endeavored to assess the effect of short-term climate extremes on adult trees by investigating xylem anatomy at intra-ring level coupled with a high-quality daily-resolved climate dataset, an aspect still neglected in the current literature (George et al., 2015).

In general, CWT and CN extremes were more clearly associated to climate than CD, confirming what was observed with the lower-resolution monthly analysis. Difficulties in matching mainly CD extremes with the corresponding shortterm climate variability might be related to year-to-year variations in cambial reactivation, and consequently in the timing of xylogenesis phases. Indeed, the peak of the 15-days window in climate sensitivity results from the simplistic assumption, common to many dendroanatomical studies (e.g., Liang et al., 2013; Castagneri et al., 2015b), that tree phenology is related to Julian days, i.e., anatomical responses to climate are fixed on the same days in all years. However, different phenological phases are usually tuned to the course of present or closely prior climate conditions (Seo et al., 2008; Rossi et al., 2014). The high-resolution time window that we introduced is likely a step forward for detecting short-term climate extremes effect on wood anatomy, but is still not enough to cope with the corresponding phenological plasticity in xylogenesis, and other approaches to detect the "key" time window during the season are advisable. CN, determined by cell division throughout the cambial activity period, and CWT, related to the wall thickening phase that can last up to 40–50 days (Rossi et al., 2008) seem to perform better in relation to the longer period of formation. In this case, inter-annual variations of xylogenesis phases were probably less appreciable considering that these traits integrate climate information on a longer time window.

The high significance of the results indicates that most of the extreme values in anatomical traits fall on the correct side of the temperature distribution (**Figure 6**). However, there is a margin of 0–34% of occurrences that did not match the expected temperature in their corresponding time window. Provided that short-term heatwaves or cold spells affect xylem structure only if they occur in (or immediately before) the period of active xylogenesis, in most of the events tree response could present a certain delay (Babst et al., 2012). Retrospectively identifying the cause/effect time lag together with a better focus on the past course of xylem phenology will surely improve the quality of the inferences. Moreover, the simultaneous effects of many climatic variables on growth could likely mask the impact of individual extreme climatic events on growth response (Ettl and Peterson, 1995). Lastly, detailed analyses on growth-climate interactions probably enhance differences in the individual responses related to microsite and surrounding conditions, genotypic variability, local disturbances, etc. (Carrer, 2011; Rozas, 2015), compared to extensive approaches that capture seasonal-level response of the tree population. These peculiar growth responses usually tend to converge, providing more consistent outcomes in the presence of a clear environmental factor limiting growth on a longer time scale.

#### CONCLUSION

Despite the rising attention within the scientific community, and the prospected increasing role in future climate scenarios, shortterm extreme events are still rather elusive. This is a challenge for the retrospective detection of such events in the tree-ring series and only the introduction of approaches that investigate different temporal scales could lead to progressive results. In this pilot investigation, we proposed some strategies for assessing the role of extreme events in a long-term perspective. Dendroanatomical measurements, with tree-ring sectors partitioning, and the use of high-quality daily-resolved in situ temperature records should go in this direction and has shown a good potential. Our analyses focused on a temperature-limited environment: here, the shortterm extreme events detection appears rather challenging; dealing with other factors with a discrete nature (e.g., precipitation or winds) should be more straightforward.

#### REFERENCES


Forest ecosystems are dominated by long-lived organisms, where extreme events can extend their effects from when they occur, even with a legacy of several years (Babst et al., 2012). There is much to be learned about the critical role of contingency and hysteresis in forest ecosystem response after climate extremes (Anderegg et al., 2015), and a better integration of empirical and modeling studies is needed, given that climate and vegetation models still do not realistically simulate most climate extremes and their effects (Reichstein et al., 2013). Wood represents an ideal archive to test for extreme events effects as it can provide a long temporal record; with tree-ring anatomy we have another powerful tool to improve our comprehension on such critical phenomena.

#### AUTHOR CONTRIBUTIONS

MC and DC were responsible for the study design, data analysis, and results interpretation. MC and DC provided the xylem anatomical data. MB analyzed and provided the climate records. All authors contributed to manuscript writing and revision and finally read and approved the submitted version.

#### ACKNOWLEDGMENTS

DC was supported by the University of Padova (Research Project D320.PRGR13001, Senior Research Grants 2012). The authors are grateful to Francesco Natalini, Giai Petit, and Angela Luisa Prendin for their help in field and lab work, and Georg von Arx for the continuous support with the software ROXAS. This research is linked to activities and benefited from the fruitful discussions conducted within the COST FP1106 'STReESS' network.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.00683

extreme events in Europe. Environ. Res. Lett. 7:045705. doi: 10.1088/1748- 9326/7/4/045705



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Carrer, Brunetti and Castagneri. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Effects of Recent Minimum Temperature and Water Deficit Increases on Pinus pinaster Radial Growth and Wood Density in Southern Portugal

Cathy B. Kurz-Besson<sup>1</sup> \*, José L. Lousada<sup>2</sup> , Maria J. Gaspar3,4,5, Isabel E. Correia4,6 , Teresa S. David4,6, Pedro M. M. Soares<sup>1</sup> , Rita M. Cardoso<sup>1</sup> , Ana Russo<sup>1</sup> , Filipa Varino1,7 , Catherine Mériaux1,8, Ricardo M. Trigo<sup>1</sup> and Célia M. Gouveia<sup>1</sup>

1 Instituto Dom Luiz, Faculdade de Ciências, Universidade de Lisboa, Lisboa, Portugal, <sup>2</sup> Centro de Investigação e de Tecnologias Agro-Ambientais e Biológicas, Universidade de Trás-os-Montes e Alto Douro, Vila Real, Portugal, <sup>3</sup> Biosystems and Integrative Sciences Institute, Universidade de Lisboa Faculdade de Ciências, Lisboa, Portugal, <sup>4</sup> Centro de Estudos Florestais, Instituto Superior de Agronomia, Universidade de Lisboa, Lisboa, Portugal, <sup>5</sup> Departamento de Genética e Biotecnologia, Universidade de Trás-os-Montes e Alto Douro, Vila Real, Portugal, <sup>6</sup> Instituto Nacional de Investigação Agraria e Veterinária, Oeiras, Portugal, <sup>7</sup> Centre National de Recherches Météorologiques, Meteo-France/CNRS, Toulouse, France, <sup>8</sup> School of Physics and Astronomy, Monash University, Clayton, VIC, Australia

#### Edited by:

Achim Braeuning, University of Erlangen-Nuremberg, Germany

#### Reviewed by:

Ignacio García-González, University of Santiago de Compostela, Spain Veronica De Micco, University of Naples Federico II, Italy

#### \*Correspondence:

Cathy B. Kurz-Besson cbbesson@fc.ul.pt cathybesson@gmail.com

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 24 February 2016 Accepted: 20 July 2016 Published: 12 July 2016

#### Citation:

Kurz-Besson CB, Lousada JL, Gaspar MJ, Correia IE, David TS, Soares PMM, Cardoso RM, Russo A, Varino F, Mériaux C, Trigo RM and Gouveia CM (2016) Effects of Recent Minimum Temperature and Water Deficit Increases on Pinus pinaster Radial Growth and Wood Density in Southern Portugal. Front. Plant Sci. 7:1170. doi: 10.3389/fpls.2016.01170 Western Iberia has recently shown increasing frequency of drought conditions coupled with heatwave events, leading to exacerbated limiting climatic conditions for plant growth. It is not clear to what extent wood growth and density of agroforestry species have suffered from such changes or recent extreme climate events. To address this question, tree-ring width and density chronologies were built for a Pinus pinaster stand in southern Portugal and correlated with climate variables, including the minimum, mean and maximum temperatures and the number of cold days. Monthly and maximum daily precipitations were also analyzed as well as dry spells. The drought effect was assessed using the standardized precipitation-evapotranspiration (SPEI) multi-scalar drought index, between 1 to 24-months. The climate-growth/density relationships were evaluated for the period 1958-2011. We show that both wood radial growth and density highly benefit from the strong decay of cold days and the increase of minimum temperature. Yet the benefits are hindered by long-term water deficit, which results in different levels of impact on wood radial growth and density. Despite of the intensification of long-term water deficit, tree-ring width appears to benefit from the minimum temperature increase, whereas the effects of long-term droughts significantly prevail on tree-ring density. Our results further highlight the dependency of the species on deep water sources after the juvenile stage. The impact of climate changes on longterm droughts and their repercussion on the shallow groundwater table and P. pinaster's vulnerability are also discussed. This work provides relevant information for forest management in the semi-arid area of the Alentejo region of Portugal. It should ease the elaboration of mitigation strategies to assure P. pinaster's production capacity and quality in response to more arid conditions in the near future in the region.

Keywords: maritime pine, climate change, wood radial density, wood radial growth, dendrochronology, SPEI drought index, IADFs

## INTRODUCTION

fpls-07-01170 August 12, 2016 Time: 10:58 # 2

Trees are continuously responding physiologically to the prevailing climatic conditions. Climate and environmental factors are well known to affect wood formation at different scales, from the cell to the entire tree ring (Fritts, 1976). These factors have an impact on the amount and characteristics of the xylem cells, as well as the properties of wood rings and their density (Fritts, 1976; Vieira et al., 2014). For temperate woody species the combination of temperature and water availability mostly controls wood radial increment (Carrer and Urbinati, 2006). In drier regions such as the Mediterranean, it is, however, often assumed that tree growth is mostly limited by water availability (Cherubini et al., 2003; Gouveia et al., 2008). For example, Lebourgeois et al. (2012) succeeded in linking narrower conifer trees to years with drier springs. Also positive effects of spring precipitation changes on tree cambial activity have been reported for several Mediterranean ligneous species, such as the stone pine (Pinus pinea L.), the Aleppo pine (P. halepensis Mill.) and the maritime pine (P. pinaster Ait.) (Vieira et al., 2010; Campelo et al., 2013; De Luis et al., 2013). Nevertheless, Correia et al. (2008) found that water use efficiency (WUE) assessed by the variation of δ <sup>13</sup>C, was better explained by the mean minimum temperatures of the coldest month than by annual precipitation for a Maritime pine provenance test in Portugal. P. pinaster trees from contrasting altitudes presenting significant variations of the phenotypic traits at a mature stage also differed in their cone, seed and germination traits, which could be explained by a response to different minimum temperatures at the provenance origin (Correia et al., 2010, 2014).

Most of the studies evaluating tree response to climate changes are focusing on tree rings width and wood growth. Yet wood density is a highly relevant property of wood quality traits. It provides an excellent indicator of strength and yield. It is also related to the optimization of wood conversion processes (cutting, gluing, finishing, drying, and paper making) (Barnett, 2003). Besides wood density is also useful to assess carbon stored in tree stems, which is essential for the estimation of the carbon footprint of the forestry sector (Ilic et al., 2000). Wood density fluctuates intra-annually from less dense earlywood to denser latewood (Fritts, 1976). It can also vary at a lower scale within earlywood or latewood due to the development of latewood-like cells within earlywood or earlywood-like cells within latewood (Fritts, 2001). Those variations are called IADFs and are mostly the result of tree genetics and aging and/or responses to environmental conditions, including climate, soil, and rooting depth (Campelo et al., 2007, 2013; Novak et al., 2013; Nabais et al., 2014; Battipaglia et al., 2016; De Micco et al., 2016). The variation in density in hardwoods is to some extent linked to wood growth rate as well as wood fibers and vessel dimensions. In the case of gymnosperm softwoods and the pine gender Pinus sp in particular, lower densities are a result of a higher proportion of tracheids with a larger extend of cavities where the amount of cell wall material is lower. This results in lower wood density (Tsoumis, 1991; Haygreen and Bowyer, 1996; Hoadley, 2000). In addition, significant correlations were found between wood density, tracheid radial diameter and climate variables such as temperature and standardized precipitation-evapotranspiration (SPEI) drought index for the endemic long-lived conifer Huon pine in Australia (Drew et al., 2013). In the Mediterranean area, Pasho et al. (2012) observed that Aleppo pine trees presented a bigger proportion of latewood tracheids in response to decreased precipitation and high temperatures. In a dry inner Alpine valley, however, an irrigation experiment revealed that non irrigated trees presented tracheids with a wider lumen than control trees (Eilmann et al., 2011). In general, xylem anatomy response to climate changes appears to change considerably between environments and/or species (Fonti and Jansen, 2012).

In Europe, maritime pine is widely spread over the Mediterranean basin, particularly in France, Iberian Peninsula, Italy and North Africa (Carrión et al., 2000). In Portugal, the species covers about 23% of the national forested area, (ICNF, 2013). The forest species contributes notably to the Portuguese economy representing roughly 11% of the total value of forest product exports (Aguiar et al., 2003; Correia et al., 2004), as it is mostly used for carpentry, construction, chipboard, pulp, and paper production, floor boards and palettes, and high quality resin (Oppen and Hone, 1995; Alía and Martín, 2003; Gaspar et al., 2009). This species can cope with a large range of climate, altitude and soil types. Nevertheless, the consequences of recent climate changes on tree-ring traits have not been fully assessed.

Climate models consistently predict marked increases in temperature associated to decreases in precipitation in the Mediterranean basin (Diffenbaugh and Giorgi, 2012; Lionello, 2012). The occurrence of extreme events, including extreme precipitation or extreme lack of precipitation associated to high temperature is also expected to increase, leading to high-impact droughts, and heatwaves (Kundzewicz et al., 2006; Lionello, 2012; IPCC, 2013). The increase of drought incidences and intensities together with a higher variability of hydrological cycles are thus expected to affect water-supply patterns and reduce stored water availability. Such remarkable changes will prejudice forests and their related ecological, economical, sociocultural and landscape services (Schröter et al., 2005), especially in the Mediterranean basin amplifying significantly the fire risk (Sousa et al., 2015).

A significant change of the precipitation in the Iberian Peninsula has especially been characterized by a noticeable higher intra and inter-annual variability over the last three decades (García-Barrón et al., 2011; Santo et al., 2014b). According to Fragoso and Gomes (2008) the Alentejo region in southern Portugal has increasingly suffered from severe drought conditions during summer with significant drying trends observed in spring (Santo et al., 2014b). Since the 1940s, Santo et al. (2014a) observed in mainland Portugal a significant increase in the frequency and duration of heat waves, as well as an increase in the frequency of hot days especially in spring and summer.

**Abbreviations**: Cold, Number of days with minimum temperature <5 ◦C; Dryd, Number of dry days per month; Drys, maximum number of consecutive dry days or Dryspell; EWD, earlywood mean density; EWG, Earlywood growth; HS1yr and HS2yr, Heat sum: daily mean T accumulated over 1 or 2 years; IADFs, intra-annual density fluctuations; LWD, latewood mean density; LWG, Latewood growth; P, monthly accumulated precipitation; PET, potential evapotranspiration; Pmax , maximum daily precipitation; T, monthly mean temperature; Tmax, Monthly average of maximum temperature; Tmin, monthly average of minimum temperature; TWD, ring mean density; TWG, Total annual ring width.

This explains why Alentejo is drifting toward an arid climate (Nunes and Seixas, 2011) expressed by a fast increase of aridity indexes since the recent past, which has highlighted a tendency toward drier climatic conditions in this region (Costa and Soares, 2012; de Figueiredo, 2013).

The objectives of this study were (1) to characterize recent climate changes in the Alentejo, a semi-arid region of Portugal, focusing on agronomical relevant climate variables and drought indexes and (2) to assess the effect of such changes on wood treering width and density in P. pinaster trees. We report the results of over 32000 Climate−Wood relationships and analyze the impact of the recent climate variability on dendrochronological traits of P. pinaster in southern Portugal between 1958 and 2011.

We used an innovative approach based on the work by DeSoto et al. (2014). The temporal evolution of climate versus monthly growth and density relationships were assessed by calculating correlations framed by a moving 15-y window. The results presented as contour plots offer a 24-month window screening of the evolution of climate−growth and climate−density relationships from 1958 to 2011. Besides temperature and precipitation, further derived relevant agronomical climatic factors were also appraised. The effect of drought was assessed using the standardized precipitation-evapotranspiration index (SPEI) with temporal scales from 1 to 24 Months.

#### MATERIALS AND METHODS

#### Sampling Location

Wood ring cores were sampled in the northern-east part of the Alentejo region (**Figure 1A**). The sampling area was located in Companhia das Lezirias (38◦ 470 24.01 N; 8◦ 540 11.10 W) at 10−20 m above the sea level and on a gentle 1.6% slope. Long-term mean annual rainfall in the sampling location was 683 mm with a mean annual temperature of 16.07◦C and 831 mm mean annual PET, using the Thornsthwaite approach (E-Obs ECA&D, 1950−2012). The Alentejo region is typically spread over a raised plain from the center to the south of Portugal. It is representative of the typical Mediterranean climate characterized by hot and dry summers and wet and cold winters as described in Köppen's classification. Rainfall occurs predominantly from October to April, while summer water deficit prevails from June to September (**Figure 1B**). This geographic area of Portugal is considered a semi-arid region (**Figure 1A**) with a dry sub-humid climate (Nunes and Seixas, 2011). Temperature presents minimum values in January and maximum in August. In this region, the monthly mean maximum temperatures have an intra-annual range of 12 to 18◦C in January and 25 to 36◦C in August; whilst the minimum monthly mean temperature in January fluctuates between 3 to 9◦C and 13 to 19◦C in August (Soares et al., 2012). The maximum monthly daily precipitation mean occurs in December with circa 4 mm/day and the minimum occurs during the summer months, with almost no precipitation (Soares et al., 2012; Cardoso et al., 2013). The annual precipitation cycle shows another minimum in March (1.5 mm/day).

## Dendrochronological Datasets

This study presents a preliminary exploratory analysis, which is part of a large framework aiming at assessing P. pinaster's plasticity response to contrasting climatic conditions in Portugal. Companhia das Lezirias is one of eleven selected sites in the framework. For this reason the sampling size in each location was limited to ten cores and only one core by tree.

Dendrochronological series considered here are the average of ten 60-year old maritime pine trees with a mean diameter at breast height of 36.9 cm and a mean height of 21.3 m. The material was collected at breast height (1.3 m), extracting one increment core per tree, from bark to pith. Using a twin-blade circular saw, a 2 mm-thick radial strip segment was sawn from each increment core and then conditioned at 12% moisture content. These radial samples were X-rayed perpendicularly to the transverse section, and their images scanned and analyzed by microdensitometric equipment (Joyce Loebl MK3) in order to compute the density components. The details of the method can be found in Gaspar et al. (2008).

Growth ring boundaries were identified in the radial density profiles by locating the sharp variations in density. Crossexamination was occasionally required, and this was done by visual assessment of the macroscopic anatomical features of the wood strip. The earlywood–latewood boundary was assigned for each ring by the average of the minimum and maximum density values within each ring (Gaspar et al., 2009). Thus, all density points within each ring with values higher than the defined boundary value for that ring were considered to be latewood.

Tree-ring cores were dated and synchronized against the reference of two characteristic wet years (1990 and 1998) and two characteristic dry years (1995, 2005) (Supplementary Figure S1). Age-related trends were removed from dendrochronological series through the application of an appropriate standardization methodology. This was achieved by fitting a negative exponential or polynomial smoothing spline functions on tree-ring width time-series and linear functions on density time-series (Cook and Peters, 1981; Cook, 1987) for each tree-ring core (see example of tree n◦ 7 in Supplementary Figure S2). Standardized indices of tree-rings were obtained by dividing the original observed data of width or density by the best fitted exponential or linear function for each core. The final indices used to evaluate the impact of recent climate change on P. pinaster wood ring traits were calculated as the average of the 10 sampled core standardized indices. Here we considered the final indices of the total, early and latewood radial growth (TWG, EWG, LWG) and the mean radial density of the TWD covering the period 1958−2011. The final indices of both early and latewood mean radial density of (EWD, LWD) started after 1965 due to the removal of several outliers, which prevented the standardization function from being correctly fitted.

IADFs in P. pinaster were evaluated by the gradual transition in cell size and color and wall thickness within previously identified annual tree-ring boundaries (Nabais et al., 2014). IADFs were classified based on the radial position within the treering (Vieira et al., 2014): IADFs classified as type E were identified as latewood-like cells within earlywood; IADFs with type L were

identified when earlywood-like cells were observed within the latewood. L+ type IADFs were distinguished from L type when the gradient color of earlywood-like cells in latewood presented a homogeneous aspect. The identification of IADFs was made visually on synchronized dated cores using a binocular magnifyer and cross-validated against microdensitometric profiles. The relative frequency of IADFs was calculated as f = ni/N, where n<sup>i</sup> is the number of trees showing an IADF in the year i and N is the total number of trees observed the same year i (Nabais et al., 2014).

#### Meteorological Datasets

Meteorological datasets covering the period 1950−2012 were retrieved from the public database of the Observational station data of the ECA&D European Climate Assessment & Dataset (Klein Tank et al., 2002) and the Observations gridded dataset form the EU-FP6 project ENSEMBLES (E-OBS, Haylock et al., 2008). Daily temperature and precipitation were extracted from the closest grid point 9.7 km away from the sampling location. Further climate variables and indexes were calculated from the E-OBS retrieved datasets on a monthly basis.

In order to maintain the analysis within a reasonable length we restricted the wide range of potential meteorological variables to the most interesting plant growth climate factors including the following precipitation-based monthly variables: the accumulated precipitation (P), the maximum daily precipitation (Pmax), the number of dry days (Dryd) and the maximum of consecutive dry days or dryspell (Drys).

Monthly temperature-based variables were also calculated and evaluated, such as the monthly average of daily mean temperature (T), the monthly averages of daily minimum (Tmin) and maximum temperatures (Tmax), the number of cold days when Tmin was lower than 5◦C (Cold), the growing degree days (GDD), the sum of daily mean temperature since the 1st of January of the current year (HS1yr) and the sum of daily mean temperature since the 1st of January of the previous year (HS2yr).

Drought effects have been assessed through the SPEI multiscalar drought index, which has become widely used in the last decade and is considered to be more appropriate for the Mediterranean type of climate, than the standard PDSI (Sousa et al., 2011; Vicente-Serrano et al., 2013). This index based on the difference between precipitation and PET is comparable in time and space (Hayes et al., 1999) and can be computed at different time scales, from one month to several years, to monitor droughts. Nonetheless, SPEI is a sitespecific drought indicator of deviations from the average water balance. Thus it integrates the effect of temperature increase on droughts. SPEI was computed to analyze the effect of drought severity and short to long-term water balance deficits on tree growth and density preceding or co-occuring the wood ring formation.

#### Data Processing and Statistics

Climatic series of precipitation and temperature were decomposed into seasonal and trend components to highlight long-term climatic patterns. This was achieved by computing the "STL" function using the default configuration in the R programming language, according to the Seasonal Trend decomposition procedure from Loess (Cleveland et al., 1990). Since the seasonal component of climatic variables was

stationary, it was not removed from the climatic dataset before the computation of Pearson correlations coefficients.

The temporal evolution of the climate-growth relationship was analyzed according to DeSoto et al. (2014), by computing the Pearson correlation coefficients between the dendrochronological time-series and the monthly climatic timesseries for the common 54-year period of 1958−2011. Pearson coefficients (r) were calculated considering a moving window of 15-year intervals. Each correlation coefficient provided for a specific year y<sup>i</sup> on the y-axis of dendrochronology/climate figures was calculated as the correlation between dendrochronological variables observed from year yi−<sup>7</sup> to yi+<sup>7</sup> and the climatic variable of the corresponding 15-year period. As an example, r values for 1970 presented on those figures represents the Pearson correlations from the relationship of tree-ring traits versus climate variables observed during the period 1963−1977. We considered a shorter moving window (15 years) compared with the 30-year interval of DeSoto et al. (2014) for two reasons: (1) Pearson correlations with n = 15 allowed the identification of clear climate response patterns; (2) a 30-year interval would have provided a much shorter time span on dendrochronology/climate figures, thus not allowing such a clear identification of the transition from juvenile to mature wood, nor the effect of the most recent and speeding up climate changes observed since the late 1980s.

A temporal framework of 24 consecutive months was considered as the x-axis of the dendrochronology/climate figures to identify the relationships between wood properties and key climatic factors from the previous to the current year of treering xylogenesis. Though correlations were calculated for each climatic variable initially considered, here we only present graphs of the main common key climatic factors involved in P. pinaster's wood ring growth and density.

We computed monthly SPEI with time-scales spanning from 1 to 24 months, which were related with maritime pine radial growth and density. We then calculated the correlations between SPEI and the corresponding dendrochronological time-series over the period 1965−2010 and for each time-scale (1 to 24 months). Therefore positive values of SPEI correspond to wet conditions while negative values indicate drought.

The impact of climatic variables on dendrochronological traits was ranked by scoring the total number of significant correlations (P < 0.05, |r| > 0.515) divided by the total number of correlations obtained in dendrochronology/climate figures, The results were expressed as the percentage of significant correlations obtained over a total of 912 (24 months<sup>∗</sup> 38 years = 912) calculated correlations in each dataset.

## RESULTS

#### Dendrochronological Time-Series

The time-series of standardized tree-ring width and density measurements performed on P. pinaster cores are presented in **Figure 2**. Tree radial growth in TWG, earlywood, and LWG showed particularly high value in 1990, 1998 and 2010 while the lowest values were observed in 1995 and 2005 (**Figure 2A**). These observations, respectively, matched the characteristic wet and dry years used for core dating synchronization. Tree radial density (TWD) presented higher inter-annual variability with less obvious climatic pattern (**Figure 2B**).

## Recent Climate Evolution in Southern Portugal

The inter-annual evolution of climatic variables (Tmax, Tmin, Cold, P, Drys, PET, and SPEI) from 1950 to 2012 and according to Loess trends is presented in **Figure 3**. Monthly basis changes of temperature-based variables (HS1yr, Tmax, Tmin, Cold) and precipitation-based variables (P, Pmax, Dryd, Drys) are shown in **Figure 4**.

Minimum daily temperature (Tmin) was the most affected by recent climate evolution in our sampling location (**Figure 3B**). After Tmin significantly dropped from 1950 to 1972, the slope inverted between 1972 and 1975 and was followed by an increase of ∼3.5◦C until 2012 (**Figure 3B**). Tmin showed the same behavior within each month of the monthly time-series, with larger amplitudes in fall and winter (**Figure 4C**). This led to the

FIGURE 3 | Trend component of climate time-series according to the seasonal decomposition by Loess (STL) over the period 1950−2010. A 3rd degree polynomial regression was fitted over STL trends to highlight multi-decadal changes (dotted curve). SPEI values are presented with a 12-month time-scale. (A) Maximum temperature (Tmax), (B) Minimum temperature (Tmin), (C) Number of cold days (Cold), (D) Precipitation (P), (E) Dryspell (Drys), (F) Potential evapotranspiration (PET), (G) 12-Month SPEI index .

FIGURE 4 | 2D representation of the recent evolution of monthly climatic variables (1958−2011) smoothed with a 15-year moving average window. The colored bar above each graphic represents the long-term average over the period 1950−2012. (A) Heat sum of daily mean temperature from 1st of January in ◦C (HS1yr). (B) Monthly average of maximum temperature in ◦C (Tmax). (C) Monthly average of minimum temperature in ◦C (Tmin). (D) Number of cold days with Tmin <5 ◦C (Cold). (E) Monthly accumulated precipitation in mm (P). (F) Monthly maximum precipitation (Pmax). (G) Maximum number of dry days (Dryd). (H) Maximum number of consecutive dry days (Drys).

progressive decline of the number of cold days (Cold) since the 1980s, to reach the point of disappearing (**Figures 3C** and **4D**). Changes in Tmax were much less perceptible, except for a slight positive anomaly in spring since the 1990s (**Figure 4B**). There was only a slight increase of +0.5◦C as compared to the average of the studied period. As a consequence, the mean temperature (T) showed an intermediate behavior between Tmin and Tmax, with an increase of ∼1.5◦C with respect to the period average (not shown).

Precipitation did not show any significant trend over the study period (**Figure 3D**), fluctuating around an average annual amount of 683 mm. Nevertheless, a recent increase in the variability of precipitation expressed by a raising amplitude of fluctuations around the trend could be noticed between 1995 and 2010. After the extreme drought of 2005, there was a large increase in precipitation, which reflected in the SPEI index (**Figure 3G**).

Seasonally, there was a sharp decline of spring precipitation progressively spreading from spring to late winter from 1950 to 2012 (**Figure 4E**). The number of dry days (Dryd, Drysp) presented the expected opposite behavior as compared to precipitation (**Figures 4G,H**). The depletion of precipitation in spring resulted in the raise of the dry periods length mostly in March since the 1980s (**Figures 4E,G,H**). Over the same period the SPEI index showed decreasing values (**Figure 3G**) emphasizing the occurrence of more frequent droughts. These droughts were linked to the accentuated raise of temperature since the late 1970s inducing a higher evapotranspiration rate (PET) (**Figure 3F**) and driving the local water balance toward an unfavorable growing deficit. By contrast, precipitation tended to intensify in fall since the 1980s (**Figure 3D**). This was also reflected on maximum daily precipitation, which showed an increment in the months of October−November, concurrent with a clear drop in February and March during the same time lapse (**Figure 4F**).

#### Discarded Climate Variables

The GDD and the heat sum since 1st of January (HS1yr) resulted in nearly identical correlation patterns when linked to wood xylogenesis. This also occurred with the maximum daily precipitation (Pmax) and the 95th precipitation percentile (P95). Therefore and to avoid redundant information we further discarded GDD and P95 variables in the rest of the manuscript.

#### Effects of Recent Climate Changes on Wood Radial Growth

Tree radial growth and density versus monthly climate variables relationships are, respectively, presented in **Figures 5**−**7**. **Figures 8** and **9** expose dendrochronology/SPEI relationships. The comparative effect of each climate variable is shown in **Figure 10**.

Tree radial growth was more sensitive to temperature than precipitation (**Figures 5** and **10**). TWG was mostly positively correlated with temperature and more strongly and frequently with Tmin (**Figures 5** and **10**). Significant correlations were stronger and more frequent with Tmin during most of the entire year preceding the ring formation. The temperature effect was cumulative, as reflected by the more solid and frequent correlations obtained with heat sums (HS1yr, HS2yr) of the year preceding the ring formation (**Figures 5** and **7**). The sign and level of correlation between TWG and Tmin (**Figure 5**) followed the evolution of Tmin trends since 1958, dropping from 1958 to ∼1982 and increasing again afterward (**Figure 3**). LWG response to temperature showed a more drastic alteration in ∼1982, shifting from slightly negative to positive correlations. Tree radial growth (TWG, EWG, and LWG) positively reacted to the disappearance of cold days (Cold) since 1986 with an increasing significance until 2012 (**Figure 5**).

Tree radial growth was consistently and positively correlated to winter precipitation and its intensity (**Figures 5** and **7**). Significant positive correlations with P and Pmax (and negative with Dryd and Drys), were mostly concentrated in November-January preceding the ring formation. TWG and LWG were the most sensitive to winter precipitation (**Figure 5**). TWG positive response to Pmax intensified since the 1990s with a shift from October to December, matching Pmax seasonal changes (**Figures 4** and **7**).

Tree radial growth was negatively affected by water deficit (**Figures 5**, **7**, and **10**). It showed hardly any significant correlation with SPEI (**Figure 8**). Nevertheless, significant relationships with SPEI accumulated over 7−8 months intensified since the late 1980s (**Figure 9**). LWG was the most affected by water deficit, while EWG showed an intermediate response to SPEI, as compared to TWG (**Figures 8** and **9**). TWG also negatively reacted to the reduction of rainy days in spring (Dryd, Drys) observed in the 1990s (**Figures 4** and **5**).

## Effects of Recent Climate Changes on Wood Radial Density

Maritime pine wood density severely responded to the inversion of temperature trend (**Figures 6** and **7**). Also the number of significant correlations and their magnitude were proportional to the amplitude of temperature change (**Figures 6** and **7**). Within temperature-based variables, TWD was more sensitive to heat sum and Tmin (**Figures 6** and **10**). There was a radical cleavage in the correlation pattern linking TWD to temperature-based variables (HS2yr, HS1yr) in ∼1980 (**Figure 6**). Before 1980, wood density was mostly negatively correlated to T, Tmin, and Tmax and essentially during the spring period of the previous and current year of the growing ring. After 1980, significant positive relationships became dominant in winter preceding ring xylogenesis (**Figure 6**) and appeared to be linked to the disappearance of cold days (**Figures 3, 4**, and **6**). The relationships with cold days was more consistent for LWD (**Figure 6**). After ∼1997, correlations between EWD and temperature weakened (**Figure 6**) while T and Tmin stabilized during the same time lapse (**Figures 3** and **6**).

Positive correlations were observed between TWD and P-based variables in July of the current and previous years of wood xylogenesis, but only until ∼1975 (**Figure 6**). After ∼1972, significant positive correlations appeared during the winter period preceding ring growth (**Figure 6**). After 1972,

at a more mature stage of the trees, there were strong and consistent positive (negative) relationships between TWD and winter P (winter Dryd-Drys) (**Figures 6** and **7**). EWD was the most sensitive to the lack of water availability during summer at the juvenile stage (**Figures 6** and **7**). LWD, however, tend to benefit from from the lack of summer precipitation before reaching 20 years old, yet it was more consistently linked to winter precipitation since the juvenile stage, as compared to EWD (**Figure 6**).

Interestingly, EWD was more sensitive to P-based variables during the coldest period (1972−1986) and increasingly less sensitive afterward despite the aggravation of water deficit (**Figures 6** and **9**). Latewood density showed the opposite pattern, demonstrating an increasing sensitivity to winter precipitation (**Figure 6**) as well as an increasing benefit of long-term water deficit until 2012 (SPEI-12, **Figure 9**). After the 1980s, latewood density was mostly controlled by winter T and P preceding the ring growth. The synergy between T and P effects on LWD seems to be linked to the strong relationships observed between LWD and the SPEI index (**Figures 6** and **9**) and intensifying since the 1980s.

Overall, TWD was more sensitive to SPEI-13, EWD to T-based variables and especially heat sum and LWD was more sensitive to SPEI accumulated over 12−15 months and P-based variables (**Figure 10**).

#### Intra-Annual Density Fluctuations

The different types of IADFs E, L, and L+, distinguished in maritime pine cores are identified in **Figure 11**. The relative frequencies of their occurrence is presented in **Figure 12** and **Figure 13** shows a comparison of years with IADFs to all years covering the period 1958−2011.

IADFs located in earlywood (type E) were rarely observed in our samples (**Figure 12A**). They only accounted for 2% of the total number of tree-rings observed. IADFs type E occurred on years 1967, 1975, and 1995 (**Figure 12**) with particularly high LWG and LWD (percentiles 75 and 60, respectively) (**Figure 13A**). Those years were also exceptionally cold and dry (**Figure 13B**).

Intra-annual density fluctuations type L and L+, located in latewood, were much more frequent. They, respectively, accounted for 11 and 26% of the total number of tree-rings. IADFs type L occurred in years with low radial growth (TWG, EWG, and LWG), and years with low LWD (**Figure 13A**). They also tend to happen under colder conditions (**Figure 13B**) though not as cold as for IADFs type E.

Intra-annual density fluctuations type L+ were the most frequently observed. They occurred on 35 out of 53 years, in years with low radial growth (TWG), such as IADFs types L. However, they do not seem to be linked to climate variables, at least on an annual basis (**Figure 13B**). Taking into account the contrasting climatic conditions along mainland Portugal, the results related with IADFs highlighted the need of a detailed monthly basis analysis, which is nevertheless out of the scope of the present work.

#### DISCUSSION

#### Recent Climate Evolution in Central Portugal

The recent evolution of climate variables in southern Portugal is in agreement with the recent Fifth Assessment Report of the IPCC (2013). The latter shows similar temperature increase trends at the global scale since 1958, with a decline until 1978 followed by an increasing trend since. This led to the disappearing of cold days in our sampling location. This trend was confirmed by the IPCC report describing the decline of cold days since 1950 and attributed to an anthropogenic cause with a very high likelihood.

While a slight annual precipitation decline has often been reported in mainland Portugal between 1951 and 2010 (Sousa et al., 2011; Santo et al., 2014b), our results, using the ECAD climatic dataset, do not present a clear evidence of such trend. Nevertheless, we noticed significant changes in the seasonal patterns of precipitation, namely a descent of precipitation conveyed by a raise of dryspell in spring and the intensification of precipitation in fall since the 1980s. Those changes have also been previously reported by Miranda et al. (2006) and more recently by Santo et al., 2014a,b).

The decline of SPEI index observed since the late 1980s is in agreement with the results of Vicente-Serrano et al. (2010) in Spain. It is most likely the result of the increasing evaporative demand due to the raise of temperature, since annual precipitation did not change significantly.

## P. pinaster Wood Growth Response to Recent Climate Changes

Consistent correlations patterns in our results showed that wood ring growth and density in maritime pine appeared to be more

affected by temperature-based than precipitation-based variables when considering climate variables independently on a monthly basis (**Figure 10**). Wood radial increment and density both showed strong response to heat sum and Tmin (**Figures 5**−**7** and **10**), clearly following the changes of Tmin trends (**Figures 3, 6**, and **7**). In addition, wood xylogenesis reacted very positively to the recent disappearance of cold days (**Figures 5** and **6**). The latter resulted in a larger number of days with Tmin above 5 ◦C allowing cambium activity (Vieira et al., 2014), which led to the extension of the growing period. Overall our results expose the benefits of the increase of accumulated temperature and especially minimum temperature on P. pinaster's wood dendrochronological traits, in particular since the 1980s.

Winter precipitation preceding the ring formation displayed consistent positive correlations with wood radial growth (**Figures 5** and **7**). These results agree with those of Vieira

et al. (2009, 2010) who showed that a drier winter prior to growth had a negative impact on maritime pine tree ring width. Also P. pinaster wood growth seemed to be more sensitive to the seasonal changes of precipitation patterns rather than to annual fluctuations since 1958. The species seemed affected by the recent precipitation deficit in May of the year preceding the ring formation, mostly since the 1990s (**Figures 4, 5**, and **7**) and though spring precipitation has been shown to decline since the 1960s (Paredes et al., 2006). The decline of spring precipitation might affect water storage, inducing precocious water deficit for the following year. Hence the relationships found between wood ring properties and SPEI drought index accumulated over several months identify the increasing long-term water deficit as a keyfactor of the loss of productivity for P. pinaster in southern Portugal.

The response of P. pinaster's wood radial increment to precipitation-based variables emphasize the importance of the storage of precipitation water from fall to winter (and even recently spring) preceding the growing period.

## P. pinaster Wood Density Response to Recent Climate Changes

According to Campelo et al. (2007), climate can explain up to 76% of the variability of pine ring density in Portugal. The changes in temperature trend observed in our work seems to affect severely the ring mean density, though with correlations showing less consistency and more complexity than those obtained with ring radial growth. Ring density was more affected by the temperature of the entire year before the 1980s, and mostly by winter temperature afterward.

Overall, EWD showed a less consistent and more puzzling response to climate variables than latewood. This is likely because earlywood is rather formed at the expense of stored carbohydrates than current photosynthesis (Hill et al., 1995). Besides, EWD undergoes a strong genetic control comparatively to latewood components (Nicholls et al., 1980; Zhang and Jiang, 1998). This is also true for maritime pine in Portugal (Louzada and Fonseca, 2002; Gaspar et al., 2008). In our study, EWD climatic response strongly changed from the juvenile to the mature stage. The juvenile phase of trees appears to be characterized by a high phenotypic variance (Louzada and Fonseca, 2002), in which the genetic potential is fully expressed between 7 and 10 years of cambial age (Gaspar et al., 2008). This can be related to a higher sensitivity of young trees to climate fluctuations and land establishment. Here EWD exhibited sensitivity to spring water stress, as previously reported by Nabais et al. (2014). Our results also agree with those of Campelo et al. (2007) studying P. pinea in Portugal. The authors showed that earlywood formation was mostly pre-determined at the beginning of the growing season. EWD correlation to spring precipitation intensified since the late 1990s, suggesting an increasing negative effect of the

decline of spring precipitation. Yet this intensification could also be due to a raise of genetic heritability with tree age according to Gaspar et al. (2008). Although age-related effect on dendrochronological traits was minimized by standardization, it should be noted that the intensification of P. pinaster's response to climate change can also be linked to a higher sensitivity of wood xylogenesis to climate factors through time (Zhang and Jiang, 1998; Aguiar et al., 2003; Dorado Liñán et al., 2012; Campelo et al., 2013, 2015; De Micco et al., 2016).

(A) TWG, (B) EWG, (C) LWG, (D) TWD, (E) EWD, (F) LWD.

By contrast, LWD was more consistently and strongly correlated with climatic conditions than EWD. This is in agreement with Campelo et al. (2007), who stated that latewood development was more sensitive to climate variations compared to earlywood. Latewood formation mainly depends on current photosynthesis products, more closely controlled by current climate conditions (Zhang, 1997; Lebourgeois, 2000). We found that LWD was positively correlated with climatic conditions favoring wood growth. Propitious years with sufficient stored water at the beginning of the growing season tend to increase the length of the growing period and, therefore, the duration of tracheid maturation and carbon deposition, which in turn results in thicker cell walls and thus higher wood density (Wodzicki, 1971).

On the other hand, LWD was increasingly affected by longterm water deficit since the 1980s, negatively responding to the decline of SPEI index. Water deficit appears to affect the physiological processes involved in the allocation and utilization of carbohydrates stored toward the end or even after the growing season (Kozlowski and Pallardy, 1997). Nabais et al. (2014) suggested that the reduction of carbon assimilation due to lower precipitation in winter and spring might prevent the completion of cell wall deposition in latewood tracheids, leading to less dense earlywood-like cells. In addition, water stress has been shown to influence ring density fluctuations by a direct effect on cell volume affecting the lumen diameter and linked to the tradeoff existing between hydraulic safety and hydraulic efficiency (Wilkinson et al., 2015). Besides restricting evapotranspiration and carbon assimilation through the regulation of stomatal conductance, water stress can also induce alterations in carbon allocation toward roots then becoming a priority carbon sink (Kurz-Besson et al., 2006). This would prevent further cell wall thickening in the trunk and reduce wood density.

Severe abiotic conditions during the growing season such as water stress may generate the production of latewood-like cells within earlywood or earlywood-like cells within latewood affecting the tree-ring density profile (Olivar et al., 2015). Our results show that false rings (IADFs) mostly occurred on years with low radial wood growth, pointing out non optimal, or stressful environmental conditions. IADFs in earlywood occurred very rarely in our location. They appear to be quite unusual in Mediterranean pine species (Battipaglia et al., 2016). Vieira et al. (2010) also reported rare earlywood IADFs in 100 yr old P. pinaster from western Portugal. This is attributed to the rarity of drought events during earlywood formation (Vieira et al., 2010; De Micco et al., 2016). We found that earlywood IADFs coincided with extreme cold and dry years. This is in agreement with our results on EWD showing stronger relationships with climate during the coldest years of the studied period (1972−1986). IADFs types L were much more frequent than IADFs\_E. This was also reported by De Micco et al. (2016), who considered them as an extension of the wood formation promoted by a combination of summer drought and favorable conditions in late summer and early autumn. In our study L-IADFs appeared under less extreme cold climate conditions than type E. This is in contradiction with Campelo et al. (2013) who found that latewood false rings happened on years characterized by warm December. The effect of climate variables (considered on an annual basis) on IADFs type L+ was more elusive in our results, probably due to a higher effect of biotic factors such as tree heritability or aging, or abiotic factors such as nutrient availability or soil heterogeneity influencing root distribution and tree competition. Battipaglia et al. (2016) considered IADFs type L+ as transitional wood rather than true density fluctuations, after analyzing carbon and oxygen isotopic

cells within latewood (type L) and as earlywood-like cells at the end of latewood and presenting a band with homogeneous color (type L+).

signals of the different kinds of fluctuations. On another hand, Vieira et al. (2010) and Campelo et al. (2013) showed that the IADFs type L+ were preferentially triggered by seasonal climate variables, namely favorable previous winter precipitation and favorable conditions before summer. The authors suggested that IADFs constitute a potential mechanism for individual trees to adapt to drought in P. pinaster, as they are mediated by stem size. On the other hand, tree radial growth and stem size is also strongly influenced by rooting depth, tree access to deep water sources during the dry years and groundwater fluctuations (Sun et al., 2000; Ford and Brooks, 2003)

## Groundwater Recharge as a Main Driver of P. pinaster's Productivity and Quality

It is assumed that juvenile wood turns into mature wood 15– 20 year from the pith (Fries and Ericsson, 2009). Our results agree with this last statement by showing a shift of significant correlations between climate and EWD (as well as LWD) from spring of the current year to winter of the previous year while changing from the juvenile to a more mature stage. In our study, the anatomical transition occurred around 1972, when trees were about 20 years of age (considering that 14 rings were formed between 1958 and 1972 and that trees needed 6 years to reach breast height where the tree-ring core were sampled).

We attribute those anatomical shifts to an alteration of tree water source from recent shallower water sources to older and deeper stored ones. This is in agreement with the study by Danjon et al. (2013) on P. pinaster in the South−West of France, showing the emergence of sinker roots on 5 years old trees and the expansion of deep roots after 12 years, becoming more abundant, thick and deep after 19 years. The authors also show that deep roots have a high overall tapering rate, especially during drought periods. This has also been shown for co-occurring ligneous species such as P. halepensis and Quercus suber (Filella and Peñuelas, 2003; David et al., 2013; Kurz-Besson et al., 2014). Maximum rooting depth reported for the three cooccurring species was lower than 7m depth (Canadell et al., 1996). According to Gómez and Viñas (2011), most of the P. pinaster root volume remained in the soil unsaturated zone, and only deep roots connected to the water table (wet season) or the capillary fringe (dry season). The latter used to show important seasonal fluctuations following the groundwater table depth (Ronen et al., 2000).

We showed that winter precipitation provides the largest amount of water in our location (Supplementary Figure S3). Since

water availability in shallow soil horizons is not a limiting factor during winter and since the physiological activity of P. pinaster is minimum at this time of the year (Vieira et al., 2014), our results strongly implies that P. pinaster's growth is favored by an optimum recharge of deep water sources during the rainy period. This statement is further strengthened by several studies performed in nearby locations of the Portuguese Alentejo region, which have shown that the aquifer recharge usually occurs 3 to 4 months after the onset of rainfall following the summer drought period (Francés, 2008; Ribeiro and Veiga da Cunha, 2010).

The fluctuation of aquifer recharge could also have been responsible for the stronger relationships observed between LWD and accumulated SPEI since the 1980s as the increase of temperature and PET most probably led to declining water storage in the region. Different studies predicting a significant decline of runoff process and river flow in the near future also predict a drop of the aquifer recharge in the southern Mediterranean basin within the near future (Fiseha et al., 2014; Mourato et al., 2015). For example, a decline of 40 to 68% of the annual groundwater recharge has also been predicted within 2050 in northern Morocco, associated to a drop of the groundwater level down to -5m depending on the heterogeneity of the region orography (Van Dijck et al., 2006). In South Portugal runoff is expected to fall 13 to 90% by the end of the century (Mourato et al., 2015). According to our results, such drastic changes of the ecosystem water balance are expected to induce dramatic consequences for P. pinaster's wood productivity and quality for commercial purposes in the Alentejo region in the near future.

## CONCLUSION

In this manuscript, we analyzed and synthetized over 32000 correlations resuming the impact of the recent climate variability on dendrochronological traits of P. pinaster in southern Portugal between 1958 and 2011. Among the profusion of very interesting imbricated correlations exposed, we can highlight the two most striking ones:


Mediterranean droughts and are favored by better groundwater recharge at the end of the winter period.

Since the 1960s, southern Portugal has faced a significant increase of the aridity, drifting from a dry sub-humid to a semiarid climate (Costa and Soares, 2012). Using best performing multi-model ensembles, Soares et al. (2015, 2016) predicted consistent rainfall reductions between 20 and 30% in this geographical area, according to the A1B and RCP8.5 greenhouse gas emission scenarios, for the end of the 21st century. With the same scenarios, Jacob et al. (2014) pointed out to an increase of the mean annual air temperature over 3◦C in the southern Mediterranean area by 2100. Those predictions agree with the predicted collapse of multi-scalar drought indexes and the consequent aggravation of dryness in southern Europe by the end of the 21th century (Stagge et al., 2015). Such changes will likely exacerbate the aridification process, severely affect groundwater recharge and induce an acute drop of the capillary fringe and groundwater levels. According to our results, P. pinaster's productivity and suitability for commercial purposes could be severely affected in the semi-arid area of the Alentejo region in southern Portugal.

Understanding how P. pinaster tree-ring traits are affected by climate changes under contrasting climatic conditions and according to the genetic variability of the species is still under assessment in Portugal. Outcomes of ongoing investigations will allow the characterization of the potential geographic distribution of maritime pine according to updated climate change scenarios. This will be crucial for plant breeders to select the most adapted proveniences and for policy makers and water managers to concept adaptation strategies to guaranty P. pinaster's productivity and commercial benefits in Portugal in the near future.

## AUTHOR CONTRIBUTIONS

CB's achieved the calculation of meteorological indexes and the analysis of the climate evolution trends, the computation of dendrochronological/climate correlations, the interpretation and the redaction of the manuscript. IC provided the framework and part of the funding for the work to be undertaken. IC, JL, and MG all participated in the dendrochronological sampling. JL and MG performed dendrochronological measurements. TD participated in the conceptual design of the study. PS defined the meteorological datasets to be used for the study and the approach used for the meteorological analysis. RC assessed and shared the meteorological time-series. AR shared and

adapted her Matlab algorithm for the computation of the SPEI multi-scalar drought index. FV shared and adapted the methodological approach used in a previous work. RT conceived the moving correlations methodology and supervised the team effort. CG provided complementary fundings for the work to be completed. She supervised the use of SPEI index in dendrochronological correlations and their related interpretation. CM performed a thorough detached analysis of the data interpretation, spotting unclear contents. Each of the co-authors performed a thorough revision of the manuscript, provided useful advices on the intellectual content and improved the English language.

### FUNDING

The publication was supported by the Fundação para a Ciência e a Tecnologia (FCT) through the projects UID/GEO/50019/2013- Instituto Dom Luiz and PIEZAGRO\_PTDC/AAG-REC/7046/ 2014.

#### REFERENCES


#### ACKNOWLEDGMENTS

The data was host-processed in the computational facility in the framework of the FCT SHARE project (RECI/GEO-MET/0380/2012). The authors also wish to acknowledge the FCT projects QSECA (PTDC/AAG-GLO/4155/2012), PINUS RAIN (PTDC/AGR-CFL/099614/2008) and SOLAR (PTDC/GEOMET/7078/2014), which shared preliminary data without which this work would not have been possible. The authors additionally thank FCT for Ana Russo's grant (SFRH/BPD/99757/2014) and Eng. Rui Alves for all logistic support in the field of Companhia das Lezírias S.A.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.01170



Fritts, H. C. (1976). Tree Rings and Climate. London: Academic Press, 567.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Kurz-Besson, Lousada, Gaspar, Correia, David, Soares, Cardoso, Russo, Varino, Mériaux, Trigo and Gouveia. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Does the Genotype Have a Significant Effect on the Formation of Intra-Annual Density Fluctuations? A Case Study Using Larix decidua from Northern Poland

#### Marcin Klisz <sup>1</sup> \*, Marcin Koprowski <sup>2</sup> , Joanna Ukalska<sup>3</sup> and Cristina Nabais <sup>4</sup>

<sup>1</sup> Department of Silviculture and Genetics, Forest Research Institute in Poland, Sekocin Stary, Poland, <sup>2</sup> Department of Ecology and Biogeography, Faculty of Biology and Environmental Protection, Nicolaus Copernicus University, Torun, Poland, ´ <sup>3</sup> Biometry Division, Department of Econometrics and Statistics, Faculty of Applied Informatics and Mathematics, Warsaw University of Life Sciences, Warsaw, Poland, <sup>4</sup> Department of Life Sciences, Centre for Functional Ecology, University of Coimbra, Coimbra, Portugal

#### Edited by:

Jian-Guo Huang, Chinese Academy of Sciences, China

#### Reviewed by:

Rubén Retuerto, Universidad de Santiago de Compostela, Spain Patrick R. N. Lenz, Natural Resources Canada, Canada

> \*Correspondence: Marcin Klisz m.klisz@ibles.waw.pl

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 25 January 2016 Accepted: 05 May 2016 Published: 20 May 2016

#### Citation:

Klisz M, Koprowski M, Ukalska J and Nabais C (2016) Does the Genotype Have a Significant Effect on the Formation of Intra-Annual Density Fluctuations? A Case Study Using Larix decidua from Northern Poland. Front. Plant Sci. 7:691. doi: 10.3389/fpls.2016.00691 Intra-annual density fluctuations (IADFs) can imprint environmental conditions within the growing season and most of the research on IADFs has been focused on their climatic signal. However, to our knowledge, the genetic influence on the frequency and type of IADFs has not been evaluated. To understand if the genotype can affect the formation of IADFs we have used a common garden experiment using eight families of Larix decidua established in two neighboring forest stands in northern Poland. Four types of IADFs were identified using X-ray density profiles: latewood-like cells within earlywood (IADF-type E), latewood-like cells in the transition from early- to latewood (IADF type E+), earlywood-like cells within latewood (IADF-type L), and earlywood-like cells in the border zone between the previous and present annual ring (IADF-type L+). The influence of explanatory variables i.e., families, sites, and years on identified density fluctuations was analyzed using generalized estimating equations (GEE). We hypothesized that trees from different families will differ in terms of frequency and type of IADFs because each family will react to precipitation and temperature in a different way, depending on the origin of those trees. The most frequent fluctuation was E+ and L types on both sites. The most important factors in the formation of IADFs were the site and year, the last one reflecting the variable climatic conditions, with no significant effect of the family. However, the relation between the formation of IADFs and selected climate parameters was different between families. Although, our results did not give a significant effect of the genotype on the formation of IADFs, the different sensitivity to climatic parameters among different families indicate that there is a genetic influence.

Keywords: IADF, genetics, G × E, European larch, generalized estimating equations, wood density

## INTRODUCTION

Radial growth of trees reflects the interactions between external (environmental) and internal factors (physiological processes determined in relation to genotype; Savva et al., 2002). The influence of environment (mainly climatic) has been very welldocumented in numerous studies (Wimmer et al., 2000; Rigling et al., 2001; Masiokas and Villalba, 2004). Much work has also addressed the influence that biotic and abiotic environmental variables exert in generating anomalies in the course of xylogenesis (Dmuchowski et al., 1997; Eilmann et al., 2013; Vieira et al., 2015).

Trees' plasticity expressed in terms of stress-induced growth reactions is known to differ in line with the expression of the genotype determining the physiological processes (López-Maury et al., 2008). Differences in the radial reaction are most probably dependent on trees' degrees of relatedness at the level of the species, provenance, or genotype. The few long-term provenance experiments that have been carried out confirm inter-population variability of the radial reaction as expressed in terms of tree ring width (McLane et al., 2011; Kalliokoski et al., 2012; Wilczynski ´ and Kulej, 2013).

The temperature increase and precipitation anomalies observed in recent years, assuming the form of droughts, heatwaves, and heavy rainfall (Lindner et al., 2014), also finds reflection in the anatomical structure of annual rings (Masiokas and Villalba, 2004; Rozas et al., 2011; Seo et al., 2011). The typical anatomical structure, with a clear distinction between early- and latewood, can be modified in a growing season with climatic conditions considered non-typical (Wimmer, 2002). In this sense, latewood-like cells can appear within earlywood, accounted for as a physiological reaction of plants to stress associated with a shortage of water, to prevent risks of cell embolism (Battipaglia et al., 2014). On the other hand, at the end of a growing season, which has featured a water shortage, the arrival of precipitation can lead to the appearance of modifications in the anatomical structure of latewood cells, with the formation of earlywood-like cells (Vieira et al., 2015). These types of modifications are classified as intra-annual density fluctuations (IADFs). Campelo et al. (2013) proposed a classification of four types of IADFs based on their location within the tree-ring: IADF-type E, with latewood-like cells within earlywood; IADF-type E+, representing a smooth transition between early- and latewood cells; IADF-type L, with earlywood-like cells within latewood; and IADFs-type L+, with earlywood-like cells between the end of the latewood and the earlywood of the next tree ring. This classification for wood density fluctuations has been used for the purposes of the work described here.

Previous attempts to explain differences in trees' reactions to climatic anomalies, as manifested in the development of density fluctuations in wood, have concentrated on the aspect of the variability characterizing trees differing in terms of size or age (Vieira et al., 2009; Campelo et al., 2013, 2015; Novak et al., 2013), or aspects reflecting intraspecific competition (biosocial position or growth rate; Copenheaver et al., 2006). Differences in the radial-growth reaction relating to soil-moisture conditions have also been investigated (Battipaglia et al., 2010; de Luis et al., 2011). Schweingruber (1980) emphasized that wide rings have more fluctuations than narrow ones. According to this, we can assume that climate conditions favoring growth can also affect the frequency of density fluctuations. The use of IADFs for environmental studies in different parts of Europe were published recently by Battipaglia et al. (2016). However, few works have taken into account the genotype effect on the formation of IADFs. It is known that the wood density of annual rings and its component elements (i.e., density of earlywood and latewood) are strongly determined genetically, with a high degree of heritability (Pâques, 2004; Fujimoto et al., 2008; Klisz, 2011). In conditions extremely unfavorable to growth (a water shortage induced by drought and high temperatures), it is possible to observe a high level of genetic control of the density profiles of annual rings, including the location of peak density within earlywood or the transition zone (Rozenberg et al., 2002). This findings support the idea that the frequency of occurrence of the different types of density fluctuation in wood might also be determined by a genetic factor.

The aim of this work was to determine the influence of the origin of trees on the frequency of different types of IADFs. We have used a common garden experiment (seed orchard) using half-sib families of Larix decidua established in two neighboring forest stands in northern Poland since 1985. We hypothesized that trees from different families will differ in terms of frequency and type of IADFs because each family will react to precipitation and temperature in a different way, depending on the origin of those trees.

## MATERIALS AND METHODS

## Study Site and Samples

Two study areas (even-aged seed orchards) were established in 1985 in northern Poland, within the Gdansk Coast (Pobrze ´ ze˙ Gdanskie) Macroregion, under similar site conditions within the ´ Forest Districts of Młynary (fresh broadleaved forest; N: 54◦ 13′E: 19◦ 54′ ; 55 m a.s.l.) and Zaporowo (fresh mixed/broadleaved forest; N: 54◦ 24′E: 19◦ 51′ ; 10 m a.s.l.). The climate is temperate with a mean temperature of 7.6◦C and annual precipitation of 742 mm, mostly distributed in June–November [climate data from European Climate Assessment & Dataset (ECA&D) project, weather station in Elblag and Kaliningrad, reference period 1948– 2014; (Klein Tank et al., 2002)]. Site condition characterized sandy clay soil in Mlynary and brown earth soil in Zaporowo. Both of the seed orchards were laid out in line with a randomized complete block design represented by the progeny of 33 plus trees of European larch. The 33 families in turn represented four seed zones for the European larch in northern Poland. i.e., nos. 103, 106, 157, and 205 (**Figure 1**). Twenty-five families of European larch were represented by at least 20 specimens at the two experimental sites. Based on the microdensity profiles of these 25 families, eight wood density classes were established. Afterwards, a selection of eight families was made to study the frequency of the different types of IADFs (Klisz, 2011). The criteria to select eight families among the 25 were based on the representation of each wood density class. At each experimental

site (seed orchard), at least 20 sample trees representative of each of the eight families of European larch were selected. A total of 188 trees were selected at the Młynary site, and 201 at Zaporowo. In 2007, by means of Pressler borer, two cores of 5 mm diameter were taken from each tree at a height of 1.3 m above the ground, from the eastern and southern directions.

## Wood Properties

Each wood sample was subject to the standard preparatory procedure for X-ray scanning. Uniformity of wood-sample width was achieved by way of lengthwise cuts using two circular saws (Larsson et al., 1994). Resin extraction was carried out in distilled water (Grabner et al., 2005), while stabilization of sample humidity levels at 15% was achieved in the Itrax Scanner density scanner. To obtain each of the required microdensity profiles, samples of wood were X-ray scanned using the Itrax Scanner at constant parameters: voltage 35 V, exposure 50 ms and resolution 25 µm. The exact procedure for X-ray scanning of wood samples using an Itrax Density Scanner (Cox Analytical Systems) is described in Lindeberg (2004), Bergsten et al. (2001), and Fries and Ericsson (2006). The images obtained were analyzed using the WinDENDRO 2009b program to establish the density profiles. To define the limits of earlywood and latewood, a constant value of 70% of the maximum density of wood in the given annual ring was adopted. This was the value used in comparisons with sample material—for European larch, and with samples of Scots pine (constant value for 50% of maximum wood density in a given annual ring), as analyzed using the same method at the same laboratory, by Fries and Ericsson (2009). The tree-rings analyzed covered the period from 1990 to 2006. All sample trees had the same cambial age.

## Determination of Different Types of IADF

In line with the locations of density fluctuations in annual rings, Campelo et al. (2013) identified four types of fluctuations. Our study improves on this classification, incorporating measured values for wood density. As a limit value allowing the occurrence of fluctuations to be identified, we adopted a microdensity local maximum >0.1 g/cm<sup>3</sup> of the average early or latewood (**Figure 2**). The classification was then into E—high values for density in earlywood; E+—high values for wood density in the transition zone between earlywood and latewood; L low values for density in latewood; and L+—low values for

density in the latewood adjacent to the next annual ring. Determination of IADF occurrence was done for each core separately, and then the frequency value (present, absence) for each pair of cores was verified in order to avoid data duplication.

#### Statistical Analysis The Impact of the Environment and Genotype on IADF Frequency

To determine the influence of provenances of the studied trees, site conditions, and years (the quality variables) on the presence of different types of IADF (the binary presenceabsence dependent variable) we have used generalized estimating equations methodology (GEE). Proposed by Liang and Zeger (1986) as an extension of GzLM generalized linear models (Nelder and Weddenburn, 1972), this allows for the analysis of correlated data (Stokes et al., 2000). Use is made of a model assuming the form:

$$\mathcal{g}(\mu\_{ijk}) = \mathcal{S}\_i + F\_{\vec{f}} + \mathcal{S} \times F\_{\vec{\imath}\vec{f}} + Y\_{k\vec{f}}$$

where g(µijk ) is the logit link function, µijk is the mean for the ijth site × family combination in the kth year, S<sup>i</sup> is the effect of the ith site (i = 1, 2), F<sup>j</sup> is the effect of the jth family (j = 1, ..., 8), S × Fij is site × family interaction effect, and Y<sup>k</sup> is the kth repeated measures year effect (k = 1, ..., 17). The significance of model effects was tested using Wald statistics for the type 3 analysis. Decomposition of variance (DOV) was performed to further quantify how much variance in the predicted IADF can be attributed to site, family, site × family interaction, and year effects (Huang et al., 2014). The contrast analysis was in turn applied for comprehensive comparisons of significant model effects. Differences between tested families in the IADFs frequencies were shown on diffograms (mean-mean scatter plot). The analyses were carried out using SAS 9.3 software (SAS Institute Inc, 2011), and the GENMOD procedure (Stokes et al., 2000) was followed.

#### Relationships between Climate and IADF

To investigate climate-IADF relationships, a GzLM model was applied using the MASS package (Author: Brian Ripley) from R (R Development Core Team, 2007). Climate data from two meteorological stations (Elblag and Kaliningrad) were obtained from the European Climate Assessment & Dataset (ECA&D) project (Klein Tank et al., 2002). For the purposes of the study, a dependent relationship between earlywood and climate data from March to June was assumed, as well as between latewood and the climate between May and September. The most parsimonious model according to AIC criterion was applied to the different IADF types, for each family and site (Akaike, 1974). A chi-square test for the type 3 analyses was applied to check for the significance of relationships. On the basis of Wald estimates a percent change in the odds of the occurrence of IADF for each 1-unit increase of climatic variables was determined (Allison, 2012).

## RESULTS

#### IADF Frequency

At the Młynary and Zaporowo site, 9.1 and 11.1% of all rings showed the occurrence of IADFs, respectively (**Table 1**). Taking into account all tree-rings with IADFs, the dominant type at both experimental sites was type E + accounting for 44.8% in the case of Młynary, and 42.6% in the case of Zaporowo. The second most frequently identified fluctuation at both sites was type L (35.8% in Młynary and 31.8% in Zaporowo). Taken together, the remaining two types of IADFs (E and L +) accounted for just 19.2% of fluctuations in trees from Młynary, and 26% of those recorded in trees from Zaporowo (**Table 1**).

From 1990 to 2006, trees at the Młynary site showed IADFs every year, while in Zaporowo no IADFs were observed in 1992 (**Figure 3**). At the Młynary site, the highest frequencies occurred in 1996, 1998, and 2000, with a high percentage of IADF type E+, accounting for 59.6, 68.9, and 38.9%, respectively (**Figure 3**). In turn, the highest frequency of IADFs observed in Zaporowo occurred in 1997 and 2003 dominated by IADF type L and type E+, respectively (**Figure 3**). At Młynary the highest frequencies of IADFs type L occurred in the years 1991, 1992, 1995, 2002, 2005, and 2006, while at Zaporowo, occurred in 1994, 1996, 1997, 2002, and 2004. The IADFs type L + occurred in 1992 and 1994 at the Młynary, and in 1994 at Zaporowo.

#### IADFs vs. Site, Family, and Year

GEE analysis allowed the identification of the significant factors (site, family, year, and their interactions) determining the frequency of the different types of IADFs. The most significant factors determining the formation of IADFs type E were the site and year, with P = 0.001 and P < 0.001, respectively (**Table 2**). Although, the families did not have a significant impact in the formation of IADFs type E ( P = 0.099), it was possible to identify pairs of families that differed significantly in the frequency of IADFs type E. Family 2818 was found to differ significantly from families 2514, 2817, 2860, and 2864 ( P = 0.018, 0.015, 0.014, and 0.021, respectively; **Table 3** , **Figure 4**).

In the case of IADFs type E +, the GEE analysis revealed that the only significant factor was associated with the calendar year (P < 0.001; **Table 2**). Nonetheless it was possible to distinguish pairs of families differing significantly in the frequency of IADFs type E +. Family 2818 showed a lower frequency of IADFs type E + compared with 2549, 2817, 2860, 3076, and 3078 (**Table 3** , **Figure 5**). Additionally, family 2860 showed a higher frequency of IADFs type E + compared with family 2864.

The frequency of IADFs type L was significantly associated with the year ( P < 0.001; **Table 2**). In the pairwise comparison of the different families, the family 2860 showed a higher frequency of IADFs type L compared with family 2818, and between family 2514 and 2860, it was marginally significant (**Table 3** , **Figure 6**).

The overall variation in IADFs frequency was strongly explained by the year effect, as suggested by a higher percentage of DOV, namely 82.4, 84.1, and 89.5% for IADFs type E, E +, and L, respectively. The site effect represents 6.55, 0.03, and 0.01%, and the family effect represents 7.76, 7.96, and 4.25% of the variation in the frequency of IADFs


Total

 190 (314) 200 (330)

 230

 251

 4500

 4194

 9.1 (410) 11.1 (465) 13.9 (57) 23.2 (108)

 44.9 (184)

 42.6 (198)

 35.9 (147) 31.8 (148)

 5.4 (22)

 2.4 (11)

TABLE 2 | Wald statistic for type 3 GEE analysis, the significance of the models' effects and the decomposition of variance DOV (%) for different IADFs' types.


type E, E+, and L, respectively. Although, IADFs type L+ occurred in quite low frequencies, the overall variation in their frequency was explained by the family (64.4%) and site effect (35.7%).

The low number of observations of IADFs type L+ did not allow the factor year to be included in the model for the statistical analysis (**Table 2**). The site showed a marginal significance in the frequency of IADFs type L+. In the pairwise comparison of families, family 2864 showed a higher frequency when compared with family 2817 (**Figure 7**).

#### IADFs and Climate

In general, IADFs type E showed an increase with high precipitation in May and temperature in April, with more significant relationships occurring in Młynary, compared with Zaporowo (**Table 4**). IADFs type E of the family 2860 showed no climatic signal, in both sites. Concerning IADFs type E+ in general the most prominent significant relationships with climatic conditions were an increase of type E+ occurrence with high precipitation in March, April, and May (**Table 5**). IADFs type E+ also showed, in general, a decrease with temperature in March and an increase with temperature in April and May (**Table 5**). All families showed significant relationships with some of the climatic parameters. Interestingly, families 2864, 3076, 3078, 2860, and 2514 seem to respond earlier to climatic conditions, with an increase with precipitation in March and a decrease with the temperature also in March (**Table 5**). IADFs type E+ of the family 2860 growing in Zaporowo showed no significant climatic relationships.

Generally, IADFs type L also showed an increase with precipitation in May and a decrease with precipitation in June and September (**Table 6**). IADFs type L also occurs more often with temperature in May. The families 2864, 3076, 3078, 2860, and 2514 present more significant relationships with climatic parameters, compared with families 2818, 2817, and 2549, as observed for IADFs type E+. Interestingly, an increase of IADFs type L occurrence with temperature in May was only found in Zaporowo (**Table 6**).

Given the limited frequency of the L+ type, no significant relationships were found.

## DISCUSSION

Our results showed that the IADF frequency was largely dependent on the climatic conditions, with IADF type E also showing a significant influence of the site. However, our results did not find a significant effect of the genotype on the formation of the different types of IADFs. Nonetheless, different families appear to show different sensitivities to climatic conditions, reflected in the different correlations between IADFs and climate.



P, p-value; OR, odds ratio. OR estimates are obtained by exponentiation of the differences of Family Least Squares Means (LSM) estimates. If OR > 1 Family i has OR higher odds of IADF presence than Family j.

endpoints of the confidence interval are either positive or negative i.e., they

FIGURE 5 | Diffogram of differences in the IADF-type E+ frequencies between the European larch families. The point of intersection of the horizontal and vertical line emanating from both axes for the two families represents the value of their difference (in log scale) and segments on both sides of the crossing point show the confidence interval for this difference. The difference between pair of families is significant when lower and upper endpoints of the confidence interval are either positive or negative i.e., they both are above or below the dotted line of equality.

#### IADFs and Genotypes

The presence of IADFs in the earlywood indicates the capacity of trees to adapt to stress conditions, lowering the risk of xylem cells embolism through a reduction in the lumen area and an increase in cell-wall thickness (Hacke et al., 2001). The presence of density fluctuations within earlywood seems to be influenced by both tree genotype and environmental conditions (Rozenberg et al., 2002). Experiments with clones of Norway spruce showed a strong genetic effect on the occurrence of IADF type E as a reaction to water shortages (Rozenberg et al., 2002). The use of specimens with a high degree of relatedness increases the chances to observe the influence of the genotype on certain traits (Rozenberg and Cahalan, 1997). In our study we have used half-sib families of European larch and this may explain why we could not see a strong effect of the genotype on the occurrence of IADFs type E and E+. This indirectly points out a high level of heritability of the wood density components of the European larch families (Klisz, 2011), has also observed in other tree species (Hylen, 1999; Louzada and Fonseca, 2002; Louzada, 2003). Nonetheless, some of the families of the European larch significantly differ in terms of the frequency of IADFs type E and E+ (**Table 3**).

both are above or below the dotted line of equality.

The heritability of the latewood density in conifers would predict that the occurrence of IADF type L and L+ was determined by the trees genotypes (Hylen, 1999; Hannrup et al., 2004). However, the frequency of density fluctuations in latewood was similar among the analyzed families of European larch. While some authors point to a higher heritability of earlywood as opposed to latewood (Rozenberg and Cahalan, 1997; Louzada and Fonseca, 2002), others have observed the reverse trend (Goggans, 1964; Hylen, 1999; Hannrup et al., 2004). In our study, the occurrence of IADFs type L was much more dependent on climatic conditions, namely humidity conditions at the end of the growing season. Trees at both experimental sites were characterized by very low frequencies of IADF L+, probably linked to an absence of an autumn growing season that favor the generation of this type of fluctuation (Campelo et al., 2006). Vieira et al. (2009) suggests that this kind of fluctuation is due to the occurrence of climatic conditions unfavorable for the complete maturation of xylem cells.

#### IADFs and Climate

The frequency of IADFs in annual rings is associated, inter alia with the type of climate shaping the course of the processes by which wood cells arise, grow, and differentiate. The low

frequency of IADFs in the tree rings of European larch growing in northern Poland, with 9.1 and 11.1% of the annual rings showing IADFs, at the Młynary and Zaporowo sites, respectively, is within the range observed by other authors under temperate and boreal climates (Wimmer et al., 2000; Rigling et al., 2001; Franceschini et al., 2013). Relatively high frequencies of such fluctuations are observed in conifers growing under Mediterranean or Atlantic climate (15–32 and 30–52%, respectively; Campelo et al., 2006; Bogino and Bravo, 2009; Vieira et al., 2009; Rozas et al., 2011). This is probably related with a wider length of the growing season under Mediterranean or Atlantic climate, increasing the probability of environmental variations affecting wood development processes, with the consequent formation of IADFs (Campelo et al., 2015). Nonetheless, the frequency of IADFs in European larch was quite high in specific years, namely in 1998 and 2000, with more than 15% of the trees showing IADFs in Młynary, and in 1997 with more than 40% of the trees showing IADFs at Zaporowo. This could be related with very specific climatic conditions of those years that induced the formation of IADFs in more trees. However, because Młynary and Zaporowo are close by and under similar climatic conditions, and if the formation of IADFs is mainly controlled by climatic conditions, then we would expect that years with a higher frequency of IADFs would be the same among the sites, which was not observed in the mentioned years above. Thus, other site


TABLE 4 | Percent change (positive value indicates increase; negative value indicates decrease) in the odds of the occurrence of IADF type E for each 1-unit increase in monthly climatic data (mean temperature and total precipitation) from March to June, for the period 1990–2006.

Grayscale indicate the level of significance based on chi-square test—white p < 0.05, light gray p < 0.01, dark gray p < 0.001.

differences (e.g., microclimatic conditions, soil differences) could explain the observed year-lag of IADFs frequency among sites.

Conifers growing in temperate climatic conditions usually present a higher frequency of earlywood IADFs (Rozenberg et al., 2002; Hoffer and Tardif, 2009), while under boreal or alpine climate, latewood fluctuations tend to prevail (accounting for 69–100% of the total; Rigling et al., 2001) as well as under Mediterranean climate (Campelo et al., 2006; Vieira et al., 2009; Nabais et al., 2014). Our results showed that the dominant types of wood density fluctuations in European larch growing in the North of Poland were E+ and L, accounting for 42.6–44.9 and 31.8–35.9% of all reported fluctuations, respectively. This intermediate behavior might be related to the fact that the study sites were located in the transition area from temperate to boreal climate.

In general, all the main types of IADFs found in European larch (type E, E+, and L) showed a significant positive correlation with the precipitation in May. Thus, it seems that higher precipitation levels in May increases the probability of IADFs formation of any type. High water availability during the growing season can increase the division rate of cambial cells (Schuppler et al., 1998; Rossi et al., 2009). Cells must go through a process of growth and maturation, and an increase in the amount of cells means that more cells are available to integrate variations in the environmental conditions, increasing the probability of the occurrence of an IADF (Carvalho et al., 2015). In fact, precipitation in May–July stimulates the growth of larch (Oleksyn and Fritts, 1991; Koprowski, 2012) and as a result wide rings are present. Previous studies have also shown that density fluctuations are more common in wider rings than in narrow ones (Schweingruber, 1980).

The formation of IADFs type E is usually determined by water shortages during the growing season, with the formation of latewood-like cells within earlywood (Rozas et al., 2011). The relation between IADFs type E and climatic conditions among sites and families was quite scattered and thus with no clear pattern. Even so, there was a positive relationship of IADFs type E with the temperature in April and precipitation in May. Although, high precipitation in May does not indicate water stress, high temperatures in April could indicate a potential increase in evapotranspiration and thus water stress. Higher temperatures during the growing season could also increase the respiratory rate, reducing the hexose pool in cambium and xylem (Deslauriers et al., 2016). Hexose is produced to increase cell osmotic potential to generate turgor pressure for cell expansion (Koch, 2004). If there is a lower availability of carbon for growth, particularly for cell enlargement (Deslauriers et al., 2016), then an IADF type E could be formed. Cuny et al. (2014) have also showed that cell enlargement duration contributed to 75% of changes in cell diameter, while the amount of wall material per cell was quite constant. Thus, the thicker walls of an IADF type E do not represent a higher amount of carbon for cell wall thickening but is just a consequence of a lower lumen diameter. Interestingly the site significantly affected the formation of this


TABLE 5 | Percent change (positive value indicates increase; negative value indicates decrease) in the odds of the occurrence of IADF type E+ for each 1-unit increase in monthly climatic data (mean temperature and total precipitation), from March to June, for the period 1990–2006.

Grayscale indicate the level of significance based on chi-square test—white p < 0.05, light gray p < 0.01, dark gray p < 0.001.

type of IADF, with a higher frequency in Zaporowo compared to Młynary. A potential environmental factor modeling the reaction of the cambium to drought-associated stress conditions may be the capacity of the soil to retain water (Rozenberg et al., 2002). Soils with a high water storage capacity, mainly determined by soil type and depth, may weaken the impact of shortfalls in precipitation in the first half of the growing season, thereby reducing the frequency of earlywood density fluctuations (Rozenberg et al., 2002). The soil-types were different between the two sites, with a sandy clay soil in Młynary, and a brown earth soil in Zaporowo. It is known that sandy soils retain less water but under water stress conditions plants retrieve water more easily, compared with soils richer in organic matter, like the brown earth soils. Thus, the soil-type might explain the higher frequency of IADFs-type E in Zaporowo, compared with Młynary.

IADFs type E+ can be understood as a smooth transition between early- to latewood, i.e., a smooth decreasing of lumen diameter and increase in cell wall width. Although, there were no significant effects of the family on the formation of IADFs type E+ (P = 0.064), the correlations with climatic conditions seems to separate two groups of families, with one group showing positive correlations with precipitation in March and May, and a negative correlation with temperatures in March, and the other group showing positive correlations with precipitation in April and temperatures in April and May. This could indicate a different sensitivity of these two groups of families toward climatic conditions. However, further research is necessary to understand if these dissimilarities are due to genotypic differences.

The mechanism behind the formation of IADFs type L, commonly found in trees from the Mediterranean region, is associated with the cambium reactivation following a summer drought (Vieira et al., 2010; Novak et al., 2013; Nabais et al., 2014). The fact that this kind of fluctuation is prevalent under the circumstances of a Mediterranean climate is related with the bimodal growth pattern present among trees growing in this climatic zone (Cherubini et al., 2003). While the generation of IADF type E is interpreted as a reaction to unfavorable growth conditions in spring, the formation of an IADF-type L is indicative of better conditions for growth at the end of the growing season (Campelo et al., 2007; Nabais et al., 2014). In the case of the European larch growing in Poland, the formation of IADFs type L is generally associated with a positive correlation with precipitation and temperature in May. As observed for IADFs type E+, it appears that some of the families are more sensitive to climatic conditions at the end of the growing season, as indirectly revealed by the occurrence of more significant correlations with climatic conditions and IADFs-type L. Although, there was no significant effect of the site, at least some families growing in Zaporowo showed a negative correlation between IADFs-type L and the precipitation in June and the temperature in August. While the negative correlation with temperature in August makes sense from the point of view of the formation of IADFs type L, the negative correlation with June


TABLE 6 | Percent change (positive value indicates increase; negative value indicates decrease) in the odds of the occurrence of IADF type L for each 1-unit increase in monthly climatic data (mean temperature and total precipitation) from May to September, for the period 1990–2006.

Grayscale indicate the level of significance based on chi-square test—white p < 0.05, light gray p < 0.01, dark gray p < 0.001.

precipitation is counter-intuitive. To further understand how and when climatic conditions can be imprinted in wood anatomy, studies of xylogenesis of European larch are necessary to see if the timings of the different phases of wood formation differ among families and/or sites.

## CONCLUSION

The frequency of IADFs under temperate climates is quite low when compared with Mediterranean or Atlantic climates. This is probably related to differences in the length of the growing season. Most of the IADFs found in the European Larch growing under a temperate climate in Northern Poland were of the type E+ and type L. Genetic determination underpinning the adaptation of the radial growth process to anomalies arising cyclically in a temperate climate should mainly concern earlywood and the transition zone with latewood, with a focus on related individuals. Although, we did not find a clear effect of genotype on the frequency of the IADFs, differences occurred in the pairwise comparison of the different families. The influence of a tree's genotype on its predisposition to generate density fluctuations within the annual rings would seem to be dependent on the degree of relatedness. In the European larch, anomalies in the anatomical structure of earlywood and wood of the transition zone are expressed most clearly in clones (i.e., trees that are identical genetically) and weakly among the families has in our study. Besides a general positive correlation of May precipitation and the frequency of IADFs, no clear pattern of the climatic signal of the different types of IADFs could be found. This can be due to the fact that we have used different families weakening a clearer climatic signal. Nonetheless, two groups of families could be distinguished based on their different sensitivity to climatic parameters, although further studies are necessary to confirm this observation.

## AUTHOR CONTRIBUTIONS

MK, MKo, JU, and CN gave a substantial contribution to the conception and design of the study, MK performed x-ray density analyses, MK and JU were in charge of genetic analyses, MKo and JU performed climatic analyses, MK wrote the first draft of the manuscript, MK, MKo, JU, and CN contributed to writing specific sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

#### ACKNOWLEDGMENTS

This research was performed under the Forest Research Institute statutory aid No. 24.12.17 of the Ministry of Science and Higher Education in Poland. This research is linked to activities conducted within the COST FP1106 "STReESS" network.

#### REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Klisz, Koprowski, Ukalska and Nabais. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Climatic Signals from Intra-annual Density Fluctuation Frequency in Mediterranean Pines at a Regional Scale

Enrica Zalloni<sup>1</sup> \*, Martin de Luis<sup>2</sup> , Filipe Campelo<sup>3</sup> , Klemen Novak<sup>2</sup> , Veronica De Micco<sup>1</sup> , Alfredo Di Filippo<sup>4</sup> , Joana Vieira<sup>3</sup> , Cristina Nabais<sup>3</sup> , Vicente Rozas<sup>5</sup> and Giovanna Battipaglia6,7

<sup>1</sup> Department of Agricultural Sciences, University of Naples Federico II, Portici, Italy, <sup>2</sup> Department of Geography and Regional Planning, Environmental Sciences Institute, University of Zaragoza, Zaragoza, Spain, <sup>3</sup> Department of Life Sciences, Centre for Functional Ecology, University of Coimbra, Coimbra, Portugal, <sup>4</sup> Department of Agricultural and Forestry Science, Tuscia University, Viterbo, Italy, <sup>5</sup> Departamento de Ciencias Agroforestales, Escuela Universitaria de Ingenierías Agrarias, Universidad de Valladolid, Campus Duques de Soria, Soria, Spain, <sup>6</sup> Department of Environmental, Biological and Pharmaceutical Sciences and Technologies, Second University of Naples, Caserta, Italy, <sup>7</sup> Laboratoire Paléoenvironnements et Chronoécologie, Ecole Pratique des Hautes Etudes, Institut des Sciences de l'Evolution – UMR 5554, Université de Montpellier, Montpellier, France

#### Edited by:

José M. Grünzweig, Hebrew University of Jerusalem, Israel

#### Reviewed by:

Jesús Julio Camarero, Consejo Superior de Investigaciones Científicas, Spain Ze-Xin Fan, Xishuangbanna Tropical Botanical Garden – Chinese Academy of Sciences, China

> \*Correspondence: Enrica Zalloni enrica.zalloni@unina.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 22 December 2015 Accepted: 14 April 2016 Published: 02 May 2016

#### Citation:

Zalloni E, de Luis M, Campelo F, Novak K, De Micco V, Di Filippo A, Vieira J, Nabais C, Rozas V and Battipaglia G (2016) Climatic Signals from Intra-annual Density Fluctuation Frequency in Mediterranean Pines at a Regional Scale. Front. Plant Sci. 7:579. doi: 10.3389/fpls.2016.00579 Tree rings provide information about the climatic conditions during the growing season by recording them in different anatomical features, such as intra-annual density fluctuations (IADFs). IADFs are intra-annual changes of wood density appearing as latewood-like cells within earlywood, or earlywood-like cells within latewood. The occurrence of IADFs is dependent on the age and size of the tree, and it is triggered by climatic drivers. The variations of IADF frequency of different species and their dependence on climate across a wide geographical range have still to be explored. The objective of this study is to investigate the effect of age, tree-ring width and climate on IADF formation and frequency at a regional scale across the Mediterranean Basin in Pinus halepensis Mill., Pinus pinaster Ait., and Pinus pinea L. The analyzed tree-ring network was composed of P. pinea trees growing at 10 sites (2 in Italy, 4 in Spain, and 4 in Portugal), P. pinaster from 19 sites (2 in Italy, 13 in Spain, and 4 in Portugal), and P. halepensis from 38 sites in Spain. The correlations between IADF frequency and monthly minimum, mean and maximum temperatures, as well as between IADF frequency and total precipitation, were analyzed. A significant negative relationship between IADF frequency and tree-ring age was found for the three Mediterranean pines. Moreover, IADFs were more frequent in wider rings than in narrower ones, although the widest rings showed a reduced IADF frequency. Wet conditions during late summer/early autumn triggered the formation of IADFs in the three species. Our results suggest the existence of a common climatic driver for the formation of IADFs in Mediterranean pines, highlighting the potential use of IADF frequency as a proxy for climate reconstructions with geographical resolution.

Keywords: IADF, tree rings, climate, Pinus halepensis, Pinus pinea, Pinus pinaster

## INTRODUCTION

fpls-07-00579 May 2, 2016 Time: 16:43 # 2

Tree-ring width is a powerful proxy of past environmental conditions able to record fluctuations of biotic and abiotic factors during the tree's lifetime (Fritts, 2001). Tree rings reveal physiological response to environmental fluctuations because the latter affect xylogenesis which in turn can lead to peculiar anatomical features, such as intra-annual density fluctuations (IADFs). IADFs are defined as a layer of cells within a tree ring identified by different shape, size, and wall thickness (Kaennel and Schweingruber, 1995), and characterized by the occurrence of latewood-like cells within earlywood or earlywood-like cells within latewood (Fritts, 2001). They can occur in several species in different environments and are often irregularly found in time and space (Cherubini et al., 2003; De Micco et al., 2016). IADFs constitute a useful tool to reconstruct intra-annual changes in climatic factors, providing detailed information at the seasonal level (Rigling et al., 2001, 2002; Copenheaver et al., 2006; Campelo et al., 2007, 2013; de Luis et al., 2007, 2011a; Bogino and Bravo, 2009; Hoffer and Tardif, 2009; Battipaglia et al., 2010; Edmonson, 2010; Vieira et al., 2010; Rozas et al., 2011; Olivar et al., 2012; Novak et al., 2013a,b; Nabais et al., 2014; Olano et al., 2015). IADF formation can be considered as a strategy of trees to adjust wood anatomical traits to short-term variations in environmental conditions maintaining the balance between hydraulic efficiency and safety against embolism during wet and dry periods, respectively (Campelo et al., 2007; De Micco et al., 2007; De Micco and Aronne, 2009; Wilkinson et al., 2015). Numerous studies reported high IADF frequency in species growing in the Mediterranean area, which is considered one of the most vulnerable regions to climate changes. According to the Intergovernmental Panel on Climate Change [IPCC] (2014), higher irregularities in the intra-annual precipitation patterns and increasing temperature are expected in the Mediterranean Basin in the next decades (Giorgi and Lionello, 2008). The expected climate changes will likely have an impact on tree growth and thus IADF frequency.

Most dendrochronological studies on IADF occurrence in the Mediterranean area have been conducted on Pinus species, since Mediterranean pines are quite sensitive to climate fluctuations and are prone to form IADFs (Campelo et al., 2007, 2013, 2015; de Luis et al., 2007, 2011a; Carvalho et al., 2015; De Micco et al., 2007; Vieira et al., 2009, 2010, 2015; Rozas et al., 2011; Olivar et al., 2012; Novak et al., 2013a,b; Nabais et al., 2014; Carvalho et al., 2015). Despite the variety of climatic conditions throughout the Mediterranean Basin, Pinus is a widespread genus (Barbéro et al., 1998), allowing to compare the climate response of different species at a regional scale.

IADF formation is reported to depend on tree age, sex, size, and/or width of the formed tree-ring (Rigling et al., 2001; Wimmer, 2002; Campelo et al., 2007, 2013, 2015; Bogino and Bravo, 2009; de Luis et al., 2009; Vieira et al., 2009; Olivar et al., 2012; Nabais et al., 2014; Olano et al., 2015). As a consequence, a wide variability in the occurrence of IADFs across species distribution is commonly described (Rigling et al., 2002; Edmonson, 2010; Novak et al., 2013b; Nabais et al., 2014). A higher frequency of IADFs has been found in young trees of Pinus pinaster growing under Mediterranean climate compared to older ones (Bogino and Bravo, 2009; Vieira et al., 2009). A similar age-relation has been observed in Pinus halepensis stands throughout its natural distribution area in the Iberian Peninsula (Olivar et al., 2012; Novak et al., 2013b). In the Iberian Peninsula, an age and size dependency of IADF frequency in P. halepensis and P. pinaster trees has been reported: the maximum frequency of IADFs was observed during the juvenile stages (about 27 years-old trees), and more IADFs were found in wider than narrower tree rings (Novak et al., 2013b; Campelo et al., 2015). In P. pinaster from east-central Spain, the presence of IADFs has been negatively correlated with radial growth rates (Bogino and Bravo, 2009), while no significant relationships of IADF frequency with age and treering width have been found in young trees (<55 years) from the wetter north-western Spain (Rozas et al., 2011). Aside from Mediterranean pines, significant relationships between IADF frequency and either tree-ring age (negative) or tree-ring width (positive) have been found in Pinus sylvestris trees growing in dry sites in the central Alps (Rigling et al., 2001, 2002). Analyzing tree rings of Pinus banksiana and Picea mariana from eastern Manitoba, Hoffer and Tardif (2009) showed a higher frequency of IADFs in juvenile rings than in older ones, but no significant relation between IADF occurrence and tree-ring width was found.

A few studies have been performed on the geographical variation of IADF occurrence. A significant variability in the frequency of IADFs across the range of P. halepensis was found in Spain (Novak et al., 2013b), with a higher frequency of IADFs in coastal sites than inland or mountain sites. Moreover, Rozas et al. (2011) found that IADF frequency of P. pinaster under Atlantic climate depends strongly on elevation, with more abundant IADFs at low elevations. Rigling et al. (2002) showed a higher mean IADF frequency in P. sylvestris growing at a drier than moderate wet sites in Switzerland. A recent study on P. pinaster and Pinus pinea comparing a Mediterranean and a temperate site in Portugal highlighted that local adaptation and site-specific climatic conditions can play an important role in the formation of IADFs regardless of the species (Nabais et al., 2014).

The literature survey reveals that available data about the relations between IADFs and climate were based on single species or when more than one species was used they were restricted to a single or a few sites. Indeed, studies based on a network of IADFs covering a broad geographical area would likely help to gain information on the ability of tree species to adjust their hydraulic architecture and physiology in response to intra-annual environmental changes on a larger geographical scale.

In the present study, we used a network of IADF frequency covering a broad geographical area with the aim to analyze whether the occurrence of IADFs in Mediterranean pine species is triggered by common regional climatic drivers. In order to reach this aim, we investigated the relationships between IADF frequency and tree-ring age, tree-ring width and climate in three widespread Mediterranean pine species, namely P. halepensis, P. pinaster and P. pinea, growing along their distribution ranges.

Our specific goals were: (1) to characterize the regional patterns of IADF frequency in P. halepensis, P. pinaster, and P. pinea growing along their distribution range, (2) to determine if and how the relationships between IADF frequency and treering age/width vary between the three species, and (3) to identify the large-scale climatic factors driving the formation of IADFs under Mediterranean climate, by analyzing the relationships between IADF frequency and monthly maximum temperature (Tmax), mean (Tmean) and minimum temperature (Tmin), as well as total precipitation.

#### MATERIALS AND METHODS

#### The Dataset: Species and Sites

The database consists of 55 previously published and 13 newly processed chronologies of tree-ring width and series of IADF frequency from: (a) P. pinea trees growing at 10 sites (i.e., 2 in Italy, 4 in Spain, and 4 in Portugal), (b) P. pinaster trees from 19 sites (i.e., 2 in Italy, 13 in Spain, and 4 in Portugal), and (c) P. halepensis trees from 38 sites in Spain. Details of each site are reported in the supporting material (Supplementary Table S1).

Climatic time series of monthly temperature and total precipitation for the period 1901–2013, for all the sites, were derived from the Climatic Research Unit (CRU) TS v. 3.22 dataset with 0.5◦ grid resolution (Harris et al., 2014). CRU gridded data were chosen for comparative purposes because of its complete coverage of all studied sites and high correlations with the local weather stations. Mean monthly temperature and total precipitation as average of all the sites for each species are shown in climate diagrams in **Figure 1**. The overall climate regime is Mediterranean-like, with the occurrence of mild winter and spring, and a period of summer drought followed by an increase in precipitation concomitant to a decrease of temperature during autumn. The mean monthly temperature for all the study sites ranged from 7◦C in January to about 24◦C in August. The sites with the highest amount of precipitation were those where P. pinaster is dominant (**Figure 1B**): at these sites, the highest monthly values of precipitation during the entire year were recorded with a maximum in December (132.3 mm) and a minimum in July (18.8 mm). The lowest values of mean precipitation throughout the year were recorded for P. halepensis sites with a maximum of 55.6 mm in October and a minimum of 13.3 mm in July (**Figure 1A**). Finally, P. pinea trees grow in sites with the lowest amount of summer precipitation (**Figure 1C**) with July as the driest month (10.8 mm), and a maximum of rainfall in December (73.2 mm).

#### Identification of IADFs

The database includes three species (P. halepensis, P. pinaster, and P. pinea) from a wide variety of sites. Despite the great potential of IADFs as proxies, the methods for their objective classification in different types (e.g., based on their relative position within the tree ring) have not been standardized yet. At present, to study IADFs, tree-ring series are mainly analyzed visually. Although, tree-ring width measuring and IADF identification were performed by different operators, all followed a common protocol and all were trained to adopt the same criteria for IADF identification. This allowed unambiguous identification of the presence/absence of IADFs, but the classification of different types of IADFs still suffered from subjectivity. Consequently, to build series of IADF occurrence, we adopted a conservative criterion using only the presence/absence of IADFs in tree rings visually identified on dated cores with the help of a stereomicroscope.

## Relationships between IADF Frequency and Tree-Ring Age and Width

The age of individual tree rings (here, defined as "tree-ring age") was indicated in ascending order starting from the most juvenile ring to the oldest one within each core. To study the influence of tree-ring age on the likelihood of IADF formation, a logistic binomial model was applied using tree-ring age as the independent variable and the presence (1) or absence (0) of IADFs in the corresponding tree ring as the dependent variable. The analysis was conducted independently for each species and was limited to tree-ring ages with at least 20 tree rings. Data from a total of 84,794 tree rings ranging from tree-ring age 1 to 169 were included in the analysis for P. halepensis. The total number of tree rings analyzed for P. pinaster and P. pinea was 30,792 and 16,028, while the range of tree-ring age was from 1 to 125 years and from 1 to 108, respectively. Predicted values of IADF frequency obtained for each age class were used as reference series for detrending purposes.

The influence of tree-ring width on the likelihood of IADF formation was analyzed with a similar procedure and the same dataset by using a new set of logistic binomial models with the width of each individual tree ring as the independent variable. Predicted and detrended IADF values for each individual treering width were calculated with the same method previously described for tree-ring age.

## IADFs Frequencies and Geographical Pattern

The geographical pattern of IADF frequency was analyzed using age-detrended IADF values obtained from the logistic binomial models including tree-ring data from all three species. To obtain detrended IADF values, the ratio between observed (0 or 1) and predicted IADF frequencies ([0,1]) was calculated for each individual IADF observation. Then, to obtain a robust estimation of the frequency of IADFs, independent from the age structures of the studied populations, the average of all individually obtained ratios (thereafter referred as IADF\_r) were calculated for each study site. For each species, IADF\_r equal to 1 represents years in which the frequency of IADFs is equal to the expected speciesspecific average. IADF\_r of 2 and of 0.5 indicates that IADF frequency was twice and half the expected average, respectively. IADF\_r were then rescaled to allow intra- and inter-species comparison. To do that, IADF\_r obtained for each population was multiplied by the average IADF as predicted from age 1 to age 100 of the specific logistic model. The obtained rescaled frequency (IADF\_f) represents the estimated frequency of IADFs

for each population which is independent from the population age structure and comparable between sites and species.

## Replication Depth and a New Approach to Study Climatic Signal in IADFs

The principle of replication represents one of the keys of dendrochronological research highlighting the need to use more than one stem radius per tree and more than one tree per site to obtain reliable tree-ring chronologies. Different statistics based on mean inter-correlation among tree-ring series, as the expressed population signal (EPS), which determines how well a chronology established on a finite number of trees approximates the theoretical population chronology (Wigley et al., 1984; Briffa and Jones, 1990), are often used to identify well-replicated periods for different types of dendrochronological series (e.g., width, density, or chemical composition). Sampling strategies in dendrochronology are often designed to ensure such replication requirements.

However, the presence of other anatomical features like IADFs cannot be measured but just characterized as a binary variable of 0 and 1 (dummy variable), based on its absence or presence in a specific tree ring. In these cases, criteria to define the appropriate number of samples to obtain an accurate representativeness of IADF frequency cannot be based on the same approach used for tree-ring chronologies, due to the binary nature of the data. To determine the appropriate sample size needed to estimate the proportion of a population that possesses a particular property (i.e., IADF occurrence), a specific calculation needed to be computed (Eq. 1).

This equation allowed calculating the required sample size in order to estimate a proportion (prevalence) with a specified level of confidence and precision. For example, the number of required samples to estimate IADF frequency for a given site and a specific year, with a 95% of confidential level (z = 1.96) and a precision of 10% (e = 0.1), is 97 (Eq. 1).

$$n = (z^2 \* p(1-p))/e^2 = [1.96]^2$$

$$\* 0.5(1-0.5)/[0.1]^2 = 97\tag{1}$$

Indeed, this number is substantially higher than the number of samples that are commonly collected in dendrochronological research (based usually on 15 trees and 2 samples per tree). Thus, replication depth issue represents an important challenge aimed to obtain reliable estimations of the frequency of anatomical variables such as IADFs, especially when the aim is to identify the main climate factors promoting their formation. A well-defined sampling strategy could be the perfect solution to reach this purpose.

Nevertheless, to deal with this challenge we adopted an alternative analytical approach which allowed us to use datasets already available (previously collected for other dendrochronological purposes), but solving the problem associated to the high replication depth required.

Our approach was based on a global analysis by combining information from all the study sites and years. To study the influence of annual precipitation on IADF formation in a given species, all available individual tree rings were grouped in 100 classes according to the percentile positions of the local annual precipitation of the year of their formation. Tree rings were grouped in classes ranging from the ones formed under drier to those formed under wetter conditions. Then, mean annual precipitation and mean standardized IADF frequencies were calculated for each class. The statistical normality of the obtained IADF series was verified using the Kolmogorov–Smirnov's test, then Pearson's correlation coefficient was computed to study the association between the two series. By using such procedure, IADF frequencies were not calculated independently for any specific calendar year but estimated for different ranges of annual precipitation conditions. The estimation of frequency associated to each precipitation class was based on at least 158 samples (as for P. pinea), in agreement with replication requirements, since dataset including IADF quantification and climate data (1901–2013) included 79901, 30736, and 15889 tree rings for P. halepensis, P. pinaster, and P. pinea, respectively.

Furthermore, since IADF frequencies were not calculated on time series of tree rings in chronological order, but by grouping rings in classes according to the climate conditions occurring during their formation, autocorrelation did not affect the significance level of the results.

The same procedure as explained for annual precipitation was also applied to mean annual temperature (Tmean), minimum (Tmin) and maximum temperature (Tmax) and total precipitation at monthly and seasonal scales from September of previous year to December of the current year. The correlations with temperature and precipitation of previous autumn months were performed to investigate the effect of growth conditions of the previous year on IADF frequency. The months of the whole calendar year were chosen for correlations between IADF frequency and current growth conditions, since cambial activity under Mediterranean climate was found to be active up to December (de Luis et al., 2009, 2011a,b).

### RESULTS

Descriptive statistics and a summary for the measured variables from the three species are shown in **Table 1**. A total of 139,342 rings were analyzed for the three species considered together, of which 24,143 showed IADFs. Mean age varied among species and ranged between 38 and 48 years. Mean tree-ring width ranged between 1.77 mm for P. halepensis and 2.76 mm for P. pinaster.

#### IADF Frequency and Tree-Ring Age

An age-dependent trend was found in the distribution of IADF frequency for all the analyzed species (**Figures 2A–C**). Higher frequency of IADFs was found in juvenile than older rings in the three Mediterranean pines, with the peak shifting to different ring ages depending on the species. The logistic binomial regression between tree-ring age and the IADF frequency showed an asymmetric bell-shaped distribution with a maximum of 12% at the age of 26 years in P. halepensis, of 45.9% at the age of 19 years in P. pinaster and of 26.9% at the age of 38 years in P. pinea. Sample depth per each species is shown in **Figure 2**.

#### IADF Frequency and Tree-Ring Width

The analysis of IADF frequency related to tree-ring widths showed a similar tendency of the three pine species with the occurrence of more IADFs in wide rings than in narrow



or very large rings, especially in P. halepensis and P. pinea. Conversely, in P. pinaster, despite the decline showed in the largest tree rings, the frequency of IADFs was maintained above 40% in rings wider than 1 cm (**Figures 3A–C**). The highest IADF frequencies were observed for tree rings representing the percentiles 0.91, 0.89, and 0.80 of their ring widths for P. halepensis, P. pinaster, and P. pinea, respectively. The distributions of IADF frequency in relation to tree-ring width

are bell shaped for all the three species. The comparison of IADF frequency and tree-ring width with sample depth showed that the highest values of IADF frequency are in the same range of ring widths (3–5 mm) for the three species regardless of the different growth rates, and it shows the increase in tree-ring width moving from the narrowest rings of P. halepensis to the widest ones of P. pinaster (**Figures 3A–C**).

#### Geographical Pattern of IADF Frequency

The spatial distribution of mean IADF frequencies standardized by age showed a geographical pattern of fluctuations in the entire network (**Figure 4**). Data of raw and detrended frequency for each site are shown in the supporting material (Supplementary Table S2). P. halepensis in Spain was the species with the narrowest range of IADF frequencies, with a minimum of 0.3% and a maximum of 34.9% (**Figure 4**). P. pinaster showed the widest range of frequencies of IADFs ranging between 6.8 and 93.2%, with the highest values of frequency in north-west Spain (**Figure 4**). Finally, P. pinea IADF frequencies ranged between 2.2 and 53.6%, with the lowest values in Spain and maximum values in Portugal and in Italy (**Figure 4**).

#### IADF Frequency and Climate

In all the three species, autumn precipitation of the current growth year seemed to be the main climatic condition triggering IADF formation (**Figures 5A–C**). Correlation coefficients between precipitation in autumn and IADF frequency were 0.4 in P. halepensis, 0.8 in P. pinaster, and 0.7 in P. pinea (p < 0.05). Significant negative correlations with precipitation were found in June in P. halepensis (r = −0.3) and in July in P. pinea (r = −0.3), while P. pinaster IADF frequency was positively correlated with precipitation during the whole year (p < 0.05). IADF frequency was positively correlated with temperature throughout the year in P. halepensis, with values of 0.5–0.7 for Tmin, 0.5–0.7 for Tmax, and 0.5–0.7 for mean temperature. IADF frequency was also positively correlated with temperature throughout the year in P. pinea, with values ranging of 0.3–0.7 for Tmin, 0.1–0.7 for Tmax, and 0.3–0.7 for mean temperature. By contrast, a highly significant negative correlation with summer temperatures (from June to September) was observed in P. pinaster (r = −0.5 with Tmin, r = −0.6 with Tmax, and r = −0.6 with mean temperature), where the most negative correlations were found in July (r = −0.5 with Tmin, r = −0.6 with Tmax, and r = −0.6 with mean temperature; **Figure 5C**). Maximum autumn temperature also showed a negative correlation with IADF frequency in P. pinaster (r = −0.3), with a high negative

FIGURE 4 | Map of the mean IADF frequencies (IADF\_f) of the sites of the network detrended by age (x axis: longitude –W, +E). The amplitude of the circles is directly proportional to the frequency: wider circles are related to higher IADF frequency compared to smaller ones (PIHA, Pinus halepensis; PIPI, Pinus pinaster; PIPN, Pinus pinea).

correlation in September (r = −0.2 with Tmin, r = −0.6 with Tmax, and r = −0.4 with mean temperature).

#### DISCUSSION

#### IADF Frequency – Tree-Ring Age Relationship

This study based on a large number of samples throughout the western Mediterranean Basin confirmed the presence of a strong relationship between IADF frequency and tree-ring age in all the analyzed pine species. An age trend toward a higher formation of IADFs in juvenile rings than in older ones was in agreement with previous studies showing that both tree-ring width and IADF frequency are age-dependent (Rigling et al., 2001, 2002; Bogino and Bravo, 2009; Hoffer and Tardif, 2009; Vieira et al., 2009; Olivar et al., 2012; Novak et al., 2013b; Campelo et al., 2015). A different timing and duration of xylem formation may explain the age-dependent IADF frequency. Indeed, the high frequency of IADFs in juvenile tree rings could be due to an earlier reactivation of the cambium and the consequent longer growing season, together with a fast physiological and morphological response to changing factors within the growing season (Villalba and Veblen, 1994; Vieira et al., 2009). On the other hand, high IADF frequency in young individuals could also be attributed to a higher sensitivity to environmental fluctuations: the shallower root systems of younger trees would favor IADF formation in response to changing water availability (Ehleringer and Dawson, 1992; Battipaglia et al., 2014). The relationship between a higher IADF frequency in tree rings and a shallower root system was also found by Pacheco et al. (2016), suggesting a higher sensitivity of the shallower rooted Spanish juniper to summer and autumn rains compared to the deeper rooted Aleppo pine in northeastern Spain. The strong relationship between IADF frequency and age highlights the necessity to overcome age trends in order to have an independent reconstruction of climate from IADFs (Novak et al., 2013b). In this paper, we show the importance of using a standardization method to obtain IADF series without the effect of the population structure and comparable among sites and species. Novak et al. (2013b) applied a standardization procedure to IADFs in P. halepensis to remove the effect of age, whereas Campelo et al. (2015) adopted a different method in P. pinaster to remove the effect of tree-ring width from IADF series. Here, a new approach was used to remove the effect of tree-ring age from IADF series across several species. This approach could be extended to other species across different geographical ranges and environments to facilitate the comparison of results and to gain univocal information.

#### IADF Frequency Geographical Pattern and Tree-Ring Width Relationship

The map of mean detrended frequencies of IADFs of the studied sites helped to show the spatial distribution of frequency among species and geographical location, pointing out a potential climatic influence. The highest values of IADF frequency were found in the sites located at longitudes where Mediterranean

Zalloni et al. IADFs in Mediterranean Pines

southern Portugal and south-central Spain (Peel et al., 2007). The inter-specific analysis highlighted the relationship between IADF frequency and growth rate, and its relation with climate: P. pinaster was the species with the highest frequency of IADFs and the widest tree rings at the same time, growing in sites with the highest mean precipitations throughout the year and mild winter conditions. On the opposite, the lowest frequencies of IADFs were recorded in tree rings of P. halepensis that was also the species with the highest percentage of narrow rings, growing in sites characterized by the lowest values of mean precipitation throughout the year. In all the species studied, the relationship between IADF frequency and tree-ring width showed that IADFs tend to be more frequent in wide but not in the widest rings. This result is mainly in agreement with the recent finding of Carvalho et al. (2015) who suggests that the formation of IADFs in latewood of P. pinaster is predisposed by higher rates of cell production in spring which, in turn, increases the number of cells under enlargement after the summer drought, leading to the formation of wider rings. This could explain the case of P. pinaster in Portugal and north-west Spain, and P. pinea in Italy, under temperate and Mediterranean conditions, respectively, showing the highest IADF frequencies and the widest tree rings. Regarding the reason why IADF frequency decreases or remains stable in extreme wider tree rings, we hypothesize that extremely wider rings are usually formed during years with favorable conditions for tree growth throughout the growing season (Fritts, 1966) without fluctuations in environmental conditions, which is considered to be the main triggering factor of IADF formation (Carvalho et al., 2015). On the opposite, the low frequency of IADFs in narrow tree-rings may be attributed to particularly unfavorable conditions during the growing season (Fritts, 1966; Zubizarreta-Gerendiain et al., 2012). In this case, trees may not have enough reserves to allow the resumption of cambial activity as a response to favorable climatic conditions after the summer drought. This could explain the situation mainly found in P. halepensis and P. pinea sites in eastern Spain under a semiarid Mediterranean climate (Peel et al., 2007), showing the lowest values of IADF frequency.

#### Climatic Signal in IADF

The climate correlations allowed us to standardize the large amount of data from the three species growing under different microclimatic conditions and in populations with different structures, with the aim of analyzing common regional patterns over local ones. A common large-scale climatic factor driving IADF formation was autumn precipitation, with high values of correlation coefficients for all the analyzed species. These results agree with the hypothesis that the formation of IADFs in Mediterranean pines is mainly triggered by the resumption of cambial activity in response to the return of favorable conditions, such as autumn precipitation after summer drought (Campelo et al., 2007; de Luis et al., 2007, 2011a,b; Camarero et al., 2010; Novak et al., 2013a,b; Hetzer et al., 2014; Carvalho et al., 2015). A decrease in precipitation during summer, associated with an increase in temperatures, may influence cell division and expansion, slowing down cambial activity (Antonova and Stasova, 1997; Deslauriers and Morin, 2005; Vieira et al.,

climate is affected by oceanic influences, as Portugal and northwest Spain. The lowest values were found in the sites located in eastern Spain, where the climate ranges from Mediterranean to semiarid. High values of IADF frequency were also found for sites with warm Mediterranean climate as the ones located in Italy,

and lowercase letters refer to the previous year. Blue horizontal lines indicate the limits of the correlation coefficient (±0.196) for significance at p < 0.05

(n > 100).

2014). Tracheids with narrow lumen and thick wall (latewood or latewood-like cells) are formed in response to low cell turgor during summer drought (Domec and Gartner, 2002). When water availability increases after autumn precipitation, differentiating cells can promptly re-acquire enough pressure (Wimmer et al., 2000; Abe et al., 2003; Rossi et al., 2009) for the enhancement of lumen enlargement (Wimmer et al., 2000), leading to earlywood-like cells (Carvalho et al., 2015; Vieira et al., 2015).

Late summer/autumn precipitation preceded by a drier period seemed to be the common triggering factor of IADFs in the three species. However, the species appear to be differently predisposed to the formation of IADFs depending on the geographic influence by peculiar climatic conditions experienced throughout the year.

Temperature seemed to play an important role for IADF formation in P. halepensis and P. pinea as confirmed by positive correlations between temperature and IADF frequency throughout the year. Favorable conditions of growth with high temperatures throughout the year could be a predisposing factor for the formation of IADFs in these two species, allowing them to efficiently react to seasonal fluctuations of precipitation. Moreover, temperatures suitable for growth throughout the year can induce a longer growing season, resulting in wider rings that are generally more prone to form IADFs (Deslauriers et al., 2008; Campelo et al., 2015). Regarding P. pinaster, living in the wetter sites of our network, precipitation seems to have a major role for IADF formation, as suggested by the positive correlations between precipitation and IADF frequency throughout the year. Furthermore, a mild dry summer was found to be a significant factor leading to the formation of IADFs in this species, as showed by the highly significant negative correlations of IADF frequency with summer-early autumn temperature. Favorable conditions for growth with wet conditions at the beginning of the growing season could facilitate P. pinaster to have a second period of cambial activity during autumn, after the summer drought, as suggested by Pacheco et al. (2016) for Spanish juniper. On the contrary, severe drought periods may prevent the formation of IADFs leading P. pinaster to an earlier stop of the growing season, unable to resume cambial activity in response to increased water availability (Vieira et al., 2014).

The methodological approach used in this study helped to link IADF frequency to climate conditions on a regional scale, but does not solve the question of the species-specific nature of their formation. There is evidence of a geographical/environmental gradient for P. halepensis, with more frequent IADFs in coastal than in inland or high elevated sites (Novak et al., 2013b). In our analysis, this trend was evidenced in P. halepensis but not in the other two pine species. The expansion of the network, especially by addition of new inland sites, could be useful to gain further insight on species-specific microclimate associations in IADF formation.

However, distribution ranges of the three species do not completely overlap, with wide geographical areas where the species do not coexist, making the comparison between species growing under the same weather conditions difficult. Thus, further research is needed using complementary approaches (including modeling) in order to disentangle the species-specific nature of IADF formation from the dependence on climate at a regional level.

Furthermore, several studies have shown the importance of the position of IADFs within the tree rings since it could reflect different climatic triggering factors of IADF formation and could also influence their frequency (Campelo et al., 2007; Battipaglia et al., 2010; De Micco et al., 2012). In the present work, we did not evaluate the relative position of IADFs within rings, since their identification is not straightforward as different methodological approaches exist (Campelo et al., 2007; De Micco et al., 2012, 2014). However, this kind of information is valuable to better understand the temporal link between IADFs and environmental factors, thus further improvements are needed to include this information in this type of analysis on a broad geographical scale.

## CONCLUSION

To date, a few studies have analyzed climatic influences on IADF formation across environmental gradients, and, to our knowledge, there is no study that combines spatial and temporal variations of IADF frequency and climate across a wide geographical range. Our results showed that tree-ring age has to be taken into account when analyzing IADFsclimate relationships because it plays an important role in IADF formation in Mediterranean pines.

A common interval of tree-ring width presenting the highest frequency of IADFs was found for the three species, with a more frequent formation of IADFs in tree rings moderately wide. Moreover, a common large-scale climatic factor driving IADF formation was found in autumn precipitation, demonstrating the potential of IADFs for climatic reconstructions. IADF formation was found to be lower in species living under Mediterranean to semi-arid conditions, where the frequency of narrow rings is higher (e.g., P. halepensis in eastern Spain). However, the highest frequency of IADF formation was found in species living in temperate sites with oceanic influence, where wetter conditions throughout the year associated with a moderately dry period in summer lead to the formation of wider tree rings (e.g., P. pinaster in Portugal and in north-west Spain). A proxy record of the intraannual plastic response of wood traits on a large scale could add insights on global change studies as highlighted for anatomical parameters by Fonti et al. (2010). IADFs could be analyzed in a large tree-ring network and used as efficient indicators to predict the plastic adjustment of tree species to changing environmental conditions, especially in the climate hotspot of the Mediterranean ecosystems. Their occurrence in several Mediterranean species, particularly conifers as Pinus spp. enabled this pioneering study of IADFs on a wide network which might be further expanded to the entire Mediterranean Basin.

## AUTHOR CONTRIBUTIONS

EZ, ML, and GB gave a substantial contribution to the conception and design of the study. All authors contributed to the supply of

data for the network. ML and EZ were in charge for statistical analyses. EZ, VM, FC, and GB contributed to data analysis. EZ, ML, FC, VM, and GB gave substantial contribution to the interpretation of data. EZ wrote the main part of the manuscript. ML, FC, VM, GB, and EZ performed the critical revision of the work. All authors contributed to manuscript revision, read and approved the submitted version.

#### FUNDING

Collection of datasets used for this work was supported by the projects: ELENA (CGL2012-31668) and CGL2015-69985- R funded by the Spanish Science and Innovation Ministry (MICINN) and FEDER funds; PGIDIT06PXIB502262PR funded by Xunta de Galicia; POCI/CLI/58680/2004 and PTDC/AAC-AMB/111675/2009 funded the Portuguese Foundation for Science and Technology (FCT) and the European Union (POCI 2010).

#### REFERENCES


The work of EZ was partially funded by COST Action STREeSS (COST-FP1106) through a Short-Term Scientific Mission (STSM).

#### ACKNOWLEDGMENTS

This article is based upon work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology). We thank the numerous collaborators for field and laboratory assistance.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00579

adjusting lumen size instead of wall thickness. PLoS ONE 10:136305. doi: 10.1371/journal.pone.0136305



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Zalloni, de Luis, Campelo, Novak, De Micco, Di Filippo, Vieira, Nabais, Rozas and Battipaglia. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Living on the Edge: Contrasted Wood-Formation Dynamics in Fagus sylvatica and Pinus sylvestris under Mediterranean Conditions

Edurne Martinez del Castillo<sup>1</sup> \*, Luis A. Longares<sup>1</sup> , Jožica Gricar ˇ 2 , Peter Prislan<sup>2</sup> , Eustaquio Gil-Pelegrín<sup>3</sup> , Katarina Cufar ˇ <sup>4</sup> and Martin de Luis<sup>1</sup>

<sup>1</sup> Department of Geography and Spatial Management, University of Zaragoza, Zaragoza, Spain, <sup>2</sup> Department of Yield and Silviculture, Department of Forest Techniques and Economics, Slovenian Forestry Institute, Ljubljana, Slovenia, <sup>3</sup> Agrifood Research and Technology Centre of Aragon, Instituto Agroalimentario de Aragón (IA2), Unidad de Recursos Forestales, Zaragoza, Spain, <sup>4</sup> Department of Wood Science and Technology, Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia

#### Edited by:

Andreas Bolte, Johann Heinrich von Thünen-Institute, Germany

#### Reviewed by:

Ignacio Garcí-a-González, Universidade de Santiago de Compostela, Spain Ivika Ostonen, University of Tartu, Estonia Dieter Eckstein, University of Hamburg, Germany

\*Correspondence:

Edurne Martinez del Castillo edurne@unizar.es

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 02 December 2015 Accepted: 10 March 2016 Published: 23 March 2016

#### Citation:

Martinez del Castillo E, Longares LA, Gricar J, Prislan P, Gil-Pelegrín E, ˇ Cufar K and de Luis M (2016) Living ˇ on the Edge: Contrasted Wood-Formation Dynamics in Fagus sylvatica and Pinus sylvestris under Mediterranean Conditions. Front. Plant Sci. 7:370. doi: 10.3389/fpls.2016.00370 Wood formation in European beech (Fagus sylvatica L.) and Scots pine (Pinus sylvestris L.) was intra-annually monitored to examine plastic responses of the xylem phenology according to altitude in one of the southernmost areas of their distribution range, i.e., in the Moncayo Natural Park, Spain. The monitoring was done from 2011 to 2013 at 1180 and 1580 m a.s.l., corresponding to the lower and upper limits of European beech forest in this region. Microcores containing phloem, cambium and xylem were collected biweekly from twenty-four trees from the beginning of March to the end of November to assess the different phases of wood formation. The samples were prepared for light microscopy to observe the following phenological phases: onset and end of cell production, onset and end of secondary wall formation in xylem cells and onset of cell maturation. The temporal dynamics of wood formation widely differed among years, altitudes and tree species. For Fagus sylvatica, the onset of cambial activity varied between the first week of May and the third week of June. Cambial activity then slowed down and stopped in summer, resulting in a length of growing season of 48–75 days. In contrast, the growing season for P. sylvestris started earlier and cambium remained active in autumn, leading to a period of activity varying from 139-170 days. The intraannual wood-formation pattern is site and species-specific. Comparison with other studies shows a clear latitudinal trend in the duration of wood formation, positive for Fagus sylvatica and negative for P. sylvestris.

Keywords: xylogenesis, European beech, Scots pine, microcore, cambial activity, Moncayo Natural Park

## INTRODUCTION

A forest community can prosper only on sites where the environmental conditions are within the niche volumes of each species (Reed and Clark, 1978). The distribution of different species is limited by a combination and interaction of biotic and abiotic factors (Mcinerny and Etienne, 2012); outside such conditions, the species cannot survive (Kearney, 2006).

The widespread forest species European beech (Fagus sylvatica) and Scots pine (Pinus sylvestris) have both high ecological relevance and economic values. European beech forests are spread all over central Europe, from central Poland, the south of Scandinavia and British Isles to the southernmost locations in the mountain ranges of Greece, Italy and Spain (Jalas and Suominen, 1973). Scots pine forests are distributed from the Alps to northeast Europe, covering all Scandinavia and Russia beyond 55◦ northern latitude (Jalas and Suominen, 1976). In the Mediterranean region are isolated patches of both species, climatically constrained by the warmer and drier conditions compared to the greater part of Europe. In these areas, extreme climatic events, such as summer droughts, heat waves or late frosts, restrict forest expansion on the edge of the distribution limit, leaving some populations isolated in mountain shelters.

Climate not only affects forest distribution but also tree growth. The study of cambial activity and tree-ring formation provides valuable information for understanding how trees respond to different climatic conditions (de Luis et al., 2011a; Gricar et al., 2014 ˇ ). In temperate ecosystems, climatic variability leads to an annual periodicity of cambial activity, with winter dormancy and an active period during the growing season.

Monitoring wood formation at the edge of a species' distribution is therefore especially relevant, since these trees are most sensitive to limiting climatic factors and respond most distinctively to any change (Fritts, 1972; Gruber et al., 2010; González-González et al., 2014). Knowing how these species grow may help to predict the distribution of tree species in the context of the expected climate change scenarios (de Luis et al., 2011a). In particular, more extreme climatic conditions are expected to affect tree-species' distribution (Richter et al., 2012; Eilmann et al., 2014).

Although the wood-formation patterns of Fagus sylvatica and Pinus sylvestris have been studied on different sites in Europe, studies along the western – southern distribution limits are still missing. Rossi et al. (2013) and Cuny et al. (2015), in comprehensive studies, compiled data on cambium phenology and wood-formation dynamics for several conifer species, including P. sylvestris, growing in different biomes. In Austria, P. sylvestris trees growing at xeric and dry-mesic sites were studied by Gruber et al. (2010), Oberhuber et al. (2011) and Swidrak et al. (2014). Similar studies were performed by Rathgeber et al. (2011a) and Cuny et al. (2012, 2014) in France, as well as by Seo et al. (2011) and Jyske et al. (2014) in Finland. These studies highlighted the plasticity of tree-ring formation of P. sylvestris in response to contrasting climatic conditions. Different key phenological dates showed distinct variability among study sites and years. It was shown that trees at northern sites initiate tree-ring formation later than trees at southern sites.

Compared to P. sylvestris, there is less information available on F. sylvatica. Cambial productivity of this species has been monitored at different sites and during several growth seasons in Slovenia by Cufar et al. (2008b) ˇ and Prislan et al. (2013) and also in Romania (Semeniuc et al., 2014). In addition, other studies have been performed for one growing season in the Netherlands (van der Werf et al., 2007), France (Michelot et al., 2012), Czech Republic (Vavrcík et al., 2013 ˇ ), and in north Germany (Schmitt et al., 2000). Previous studies indicated the importance of photoperiod and leaf phenology for the onset of xylem production, and the influence of climatic conditions in June, which was proved to be the most important month for wood formation (Cufar et al., 2014 ˇ ). Studies performed during several years showed that year-to-year variations in tree-ring formation can be explained by climatic conditions or environmental factors, although in some cases the relationship between variation in xylogenesis and weather conditions can be very complex (Prislan et al., 2013).

In order to better understand the growth adaptations and limitations of F. sylvatica and P. sylvestris at one of the Mediterranean edges of their distribution, we studied the dynamics of xylem-growth formation at one of the southernmost sites of the two species during three years. Cambium phenology (onset and cessation of cambial cell production) and the timing of xylem formation was compared between the two species and sites to evaluate the adaptation strategies under different environmental conditions. The duration of the xylogenesis was compared with data from other studies of the same two species performed all over Europe.

## METHODOLOGY

#### Study Site

The study was carried out in the Moncayo Natural Park, a mountain area in the northeast of the Iberian Peninsula in the province of Zaragoza (41◦ 480 31<sup>00</sup> N, 1◦ 490 10<sup>00</sup> W). This natural park is considered as a biodiversity hotspot; the contrasting climate conditions along the altitudinal gradient of the mountain allow growth of various vegetation types, from Mediterranean to Eurosiberian species (Martinez del Castillo et al., 2015). This site is one of the southernmost forest stands in Europe for F. sylvatica and P. sylvestris. The mean annual temperature and average annual precipitation for the last 37 years were 11◦ and 710 mm, respectively [according to the Spain02 database (Herrera et al., 2012)].

Two pure stands of different altitudes were selected for each species, corresponding to the lower and higher altitudinal limits of F. sylvatica forest on this mountain. The low elevation site was located at 1180 m a.s.l. and the high elevation site at 1560 m a.s.l.

## Sample Preparation

The sampling of tissues for xylem-formation monitoring was performed biweekly from mid-March until late November from 2011 to 2013. At each sampling date, six trees were randomly selected per species and site in a sampling plot of around 50 m × 50 m. The selected trees were similar, healthy and dominant, with a stem diameter at breast height of 40–55 cm and an age of around 80 years for P. sylvestris and 35– 50 cm and 120 years for F. sylvatica. From each tree, two microcores containing phloem, cambium and the last formed xylem growth ring were collected at breast height with a Trephor tool (Rossi et al., 2006a). The sample had a diameter of 2 mm

and was up to 15 mm long. The sampling followed a helical arrangement around the stem to avoid wound effects from previous samplings, with sampling locations separated by at least 10 cm. After sampling, the microcores were immediately transferred in Eppendorf microtubes filled with formaldehydeethanol-acetic acid (FAA) fixative solution for one week and later stored in 70% ethanol.

The microcores were processed following the protocol described by Rossi et al. (2006a). The microcores were first dehydrated in a graded series of ethanol (70, 80, 90, and 100%) and infiltrated with D-limonene and paraffin using a Tissue Processor Leica TP1020. After infiltration, samples were embedded in paraffin blocks. Transverse sections of 8–10 µm thickness, depending on the species, were cut with a Leica RM 2245 rotary microtome. The sections were afterward stained with safranin and astra blue (Gricar et al., 2007 ˇ ; van der Werf et al., 2007; Prislan et al., 2013), mounted in Euparal and examined with a Nikon Eclipse E800 light microscope equipped with polarized light mode.

## Xylem Phenology Measurements and Data Processing

For F. sylvatica, the width of the cell layers in the cambium was measured with the NIS Elements BR3 image analysis system. Moreover, the width of growth-ring increments and also the width of tissues containing xylem cells in various differentiation phases were measured, i.e., post-cambial growth (enlarging cells), cells undergoing secondary wall thickening and mature cells. For P. sylvestris, the cambial cells were counted as well as the xylem cells in the three different aforementioned phases.

Cambial activity was identified and interpreted within the context of the multi-seriate concept, that the vascular cambium comprises both the cambial initial cells and xylem and phloem mother cells (Plomion et al., 2001). Thin-walled cambial cells were identified based on their small radial dimensions compared to xylem and phloem cells in the enlarging phase (post-cambial growth), with larger radial dimensions. The polarized mode of the light microscope enabled the discrimination between enlarging cells and secondary wall-thickening cells as described in Rossi et al. (2006b).

The number of cells and the width of tissues in each phase varied between and within trees due to the variation of the treering width around the tree circumference. The number of cells in the previous xylem ring was therefore counted for P. sylvestris to normalize the measurements according to Rossi et al. (2003). In the case of F. sylvatica, the normalization formula was adapted using width measurements instead of cell number, as described in Prislan et al. (2013).

Extreme values were filtered and xylem-formation dynamics was analyzed with the Gompertz function (Rossi et al., 2003). Cambium phenology and timing of xylem formation were assessed using R package CAVIAR (Rathgeber et al., 2011a,b). We defined: beginning and end of the enlarging phase (bE, cE), beginning and end of the thickening phase (bW, cW) and beginning of cell maturation phase (bM). All the dates were computed with a dedicated function using logistic regressions. Differences in the different xylogenesis phases were determined applying repeated measurements ANOVA analysis (de Luis et al., 2011a; Prislan et al., 2013). The effects of fixed factors such as species and sites and the effect of time were evaluated. The total duration of the xylogenesis period was calculated by subtracting the beginning of the enlarging phase from the cessation of the thickening phase. Mean xylogenesis duration was calculated to compare the results with other studies.

## RESULTS

## Wood Formation

In all years, locations and species, the cambium was still dormant on the first sampling date in the last week of March. Despite site and annual variations in weather conditions, the different xylogenesis phases followed a common pattern during the growing season. The cell enlarging and cell-wall thickening curves follow a characteristic bell shape, while the cell maturation curve follows a sigmoid shape (**Figure 1**).

Overall, the onset of cell enlargement in P. sylvestris occurred between the last week of March and the first week of April, followed by the onset of the wall-thickening phase around 2 months later. The beginning of cell maturation occurred around the summer solstice, before the enlarging of cell ends in early September. Over more than 2 months, the currently forming tree-ring contained cells in different developmental stages. Completely mature xylem growth rings were observed in the first half of November.

Xylem-formation dynamics patterns differed between F. sylvatica and P. sylvestris. In F. sylvatica the onset of enlarging and wall-thickening phases occurred in the second half of May and in June, respectively. The beginning of the maturation process occurred from the last week of June to mid-July, followed by an immediate ending of the enlarging phase. Finally, the xylogenesis ended around mid-August.

#### Phenology of Xylem Formation

The critical dates for the xylogenesis of P. sylvestris and F. sylvatica were summarized on three levels, shown in **Figure 2**. Dates significantly differed between the two species (ANOVA bE, cE, bW, cW: p < 0.001; bM: p = 0.005). The cell enlargement started first in P. sylvestris, around 31 March (DOY 90) and 50 days later, around 20 May (DOY 140) in F. sylvatica. Cessation of cell enlargement was observed between 9 July and 6 August (DOY 190-218) for F. sylvatica and in P. sylvestris between 10 August and 1 October (DOY 222-274). Cell-wall thickening and lignification began up to ca. one month earlier and ended around three months later in P. sylvestris than in F. sylvatica. The first mature cells were observed around 25 June (DOY 176) in P. sylvestris and around 7 July (DOY 188) in F. sylvatica.

The beginning of the enlargement phase was highly variable and significantly different among the years for both species at both high and low elevations (ANOVA species<sup>∗</sup> site: p < 0.001). In F. sylvatica, it began between 7 May and 20 June (DOY 127–171), with noticeable differences within years (**Figure 2**). In P. sylvestris, cell enlargement began between 23 March and

17 April (DOY 82–107). Although the variability was lower in the latter species, in 2011 there was a delay in the beginning of cell enlargement. Focusing on the elevation differences, cell enlargement started earlier at lower elevation in the P. sylvestris than in F. sylvatica. The end of this phase highly varied among years, whereby higher variability was observed between the years then among the sites for each species (ANOVA site: p = 0.31).

The onset of the secondary wall formation was highly variable (**Figure 2**); in all cases the cell-wall thickening phase started earlier in 2011 and later in 2013, followed by the same temporal pattern (within-subjects ANOVA time<sup>∗</sup> species<sup>∗</sup> site: p = 0.561) (**Figure 2**). The first mature cells (5%) were formed earliest in P. sylvestris at low elevation in 2012 (around June 4, DOY 155), and in 2011 (around 11 June, DOY 162) for F. sylvatica.

#### Duration of the Growing Season

The total duration of the xylogenesis of F. sylvatica was significantly shorter than in P. sylvestris (ANOVA species: p < 0.001) (**Figure 3**). The cell production period during the three study years took 48–75 days for F. sylvatica, in contrast to P. sylvestris, with a growing period lasting from 140 to 170 days. Trees growing at low elevation had a longer growing period in the case of P. sylvestris, whereas the growing period of F. sylvatica, in contrast, was shorter at low elevation than at high elevation.

The mean duration of the xylogenesis was compared with other studies all over Europe (**Table 1** and **Figure 4**). In P. sylvestris, the duration of xylogenesis was shorter at high latitude and longer at low latitude, with a range of 49 days in Finland (Seo et al., 2011) to 217 days in Spain. In contrast, wood formation process in F. sylvatica was longer at high latitudes, over 163 days in the Netherlands (van der Werf et al., 2007) and only 67 days in Spain. In both species, the average duration of the xylem formation follows a linear pattern along the latitudinal range; however, whether it is directly or inversely proportional to the latitude depends on the species (**Figure 5**).

## DISCUSSION

## Dynamics of Xylogenesis

Tree growth is largely affected by different climatic conditions, which become more limiting in adverse climatic conditions, such as in a Continental Mediterranean climate (Camarero et al., 2010; de Luis et al., 2011a; Pasho et al., 2012). Different tree species are differently affected by climate: e.g., evergreen or deciduous species, or early-successional or late-successional species. In this context, previous studies suggest that evergreen species adapt better to Mediterranean environmental and climatic conditions than deciduous species (Blumler, 1991), while early-successional species adopt riskier life strategies (Körner and Basler, 2010), making them more adaptive but also more vulnerable to the


TABLE 1 | Mean xylogenesis duration values of Fagus sylvatica and Pinus sylvestris from various wood-formation studies in Europe.

highly variable Mediterranean climate. These differences may trigger a different phenology of xylem formation. Our results suggest that F. sylvatica and P. sylvestris respond differently to local Mediterranean conditions. Accordingly, the phenology of xylem formation was significantly different between the two species, the period of all P. sylvestris developmental phases being significantly longer.

Our results demonstrate that Mediterranean climate has less impact on P. sylvestris than on F. sylvatica, despite this earlysuccessional condition. The Pinus genus has been established as very plastic and capable of adapting its growth to changing climatic conditions (Camarero et al., 2010; de Luis et al., 2011a; Novak et al., 2013; Vieira et al., 2014) and the bimodal growth pattern as an adaptation to Mediterranean climate has been frequently described (Camarero et al., 2010; Campelo et al., 2015). Specifically, P. sylvestris has recently been determined as a plastic species in the Mediterranean area (Sánchez-Salguero et al., 2015).

Several studies performed on F. sylvatica under Mediterranean conditions have highlighted the growth limitation due to summer high temperatures and drought (Robson et al., 2013; Rasztovits et al., 2014; Chen et al., 2015; Rozas et al., 2015). In addition to climatic constrictions, F. sylvatica trees are more limited during the year in terms of plasticity because, with the activation of a leaf senescence mechanism, trees inexorably enter a dormant period. Despite this, our results reveal differences in the altitudinal gradient in agreement with the results shown in Prislan et al. (2013): who found similar patterns but different timing in two F. sylvatica forests with different climatic regimes.

The most striking result of the present study is the great differences in growth patterns among the years, highlighting a plastic response of radial growth in F. sylvatica, similarly as in P. sylvestris.

#### Occurrence of Xylem Phenology

High variability in xylem phenology between years and sites demonstrates high plasticity of the species. The timing of different developmental phases significantly varied between the two species. Even though the variability of the critical dates was high among years and sites, the most remarkable disparity was found between the two tree species.

Xylogenesis, starting with cambial division and cell enlargement, is triggered by an increase in air temperature in spring. Several studies have demonstrated this positive relationship (Rossi et al., 2008; Prislan et al., 2011; Vieira et al., 2014), which has also been supported by stem heating experiments (Gricar et al., 2007 ˇ ; Begum et al., 2010). Under the same climatic conditions, we showed a difference in the onset of xylogenesis between the two species of over 50 days, especially in 2013, when the difference was about 72 days at both elevations. These differences suggest that climatic conditions for the onset of xylogenesis are species-specific. Moreover, the same weather conditions resulted in a completely different response of the tree species in terms of the temporal dynamics of xylogenesis, as can clearly be seen in 2011, when F. sylvatica started earlier than in the other two study years, while P. sylvestris showed the latest onset of growth in the same year.

The end of the cell-wall thickening phase seems to be a key date, since it defines the end of xylogenesis. Overall, the thickening phase in P. sylvestris continued until mid-November, whereas in F. sylvatica it ceased in mid-August, i.e., about three months earlier. Mild temperatures during early autumn may result in an extension of the growing period for P. sylvestris but not for F. sylvatica, since by that time leaf senescence has also already started. Cessation of cell-wall thickening was first observed in the lower part of the mountain in both species, as was similarly reported by Moser et al. (2009) and Oladi et al. (2011). This indicates that the end of xylogenesis is possibly influenced by temperature as well as the length of the photoperiod, as proposed by Plomion et al. (2001).

A common pattern is an evident delay in all developmental phases in 2013, except the beginning of wood formation in P. sylvestris. This may be explained by the late frost that occurred in 2013, after the onset of cell enlargement in P. sylvestris. This event presumably affected the temporal dynamics of wood formation in both species, although with different magnitudes. Menzel et al. (2015) showed that spring late frost events cause considerable damage in F. sylvatica. Nonetheless, a more detailed study of the climatic-growth relationship would be needed to confirm this hypothesis.

FIGURE 5 | Xylogenesis duration variations by latitude. The black lines represent linear regressions: (P. sylvestris) y = −6.25x + 470.14, P < 0.001; (F. sylvatica) y = 7.28x −222.56, P = 0.010.

## Growing-Season Length under Mediterranean Conditions

Although both species are growing at their southern distribution limit and, consequently, their radial growth is somewhat constrained, the duration of the growing period of the two species significantly varied. The duration of xylogenesis highlights the differences between the species; it was two months for F. sylvatica and more than five months for P. sylvestris. Because of the early start and late end, P. sylvestris on Moncayo showed the longest xylogenesis duration in this species recorded in the various studies to date. Moreover, the shortest xylogenesis duration for F. sylvatica was also captured on this mountain.

Cambial resumption in P. sylvestris occurred earlier than on numerous Central European sites (Gruber et al., 2010; Oberhuber et al., 2011; Cuny et al., 2012, 2014; Swidrak et al., 2014), which could be explained by the warmer spring in Moncayo. These results are also in accordance with observations of Rossi et al. (2013) on various sites and Jyske et al. (2014) in Finland. On the other hand, the end of xylogenesis occurred later at Moncayo and resulted in a longer duration of xylogenesis than in other places in central and northern Europe. An extension of the growing season has also been described for other pine species, such as P. halepensis (de Luis et al., 2011b) or P. pinaster

(Vieira et al., 2014). The extension of xylogenesis could be caused by the Mediterranean mild temperatures, despite water restrictions. However, in F. sylvatica, the results showed the opposite response, the beginning was later than in colder locations in Europe, such as in France (Michelot et al., 2012), Slovenia (Cufar et al., 2008b ˇ ; Prislan et al., 2013), Romania (Semeniuc et al., 2014), Czech Republic (Vavrcík et al., 2013 ˇ ) and the Netherlands (van der Werf et al., 2007). This indicates that warmer and drier conditions at Moncayo negatively affect the duration of xylogenesis in F. sylvatica.

#### CONCLUSION

It appears that the temporal dynamics of xylogenesis is considerably different in F. sylvatica than in P. sylvestris growing at the edge of their southern spatial distribution. This shows that intra-annual radial growth patterns in the studied species are differently affected by the Mediterranean conditions. The annual variation of the critical xylogenesis dates indicates a high speciesspecific plasticity for adapting to changing climatic conditions. As a result, the period of xylogenesis in F. sylvatica was around 2 months, while for P. sylvestris it was more than 5 months. Our findings are in accordance with our hypothesis of contrasting growth strategies and adaptations of the two species at the edge of their spatial distribution.

Furthermore, we compared our observations with those of other authors working on the same two species in different climatic environments, especially along a latitudinal range. A clear north–south trend was found in the xylogenesis duration over the distribution range of both species. P. sylvestris showed a positive xylogenesis duration trend on southern locations. F. sylvatica, in contrast, showed a shorter xylogenesis duration in the south of Europe than that shown in northern locations. These findings demonstrate that a deciduous and latesuccessional species such as F. sylvatica is negatively affected by Mediterranean climatic conditions, resulting in a shorter xylogenesis, whereas in the evergreen and early-successional P. sylvestris, xylogenesis is shown to be longer in Mediterranean environments.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

EMDC did the fieldwork, wrote the paper and together with ML carried out the statistical analysis and prepared the figures. ML and LA developed the idea of this work and get the necessary funds. EMDC together with KC, JG, and PP did the laboratory ˇ work and measurements. EG-P and KC provided facilities and ˇ material support and gave technical support during laboratory work. All authors without exception helped to improve the work, specially the discussion and conclusions.

#### FUNDING

Financial support for this research was provided by the Spanish Ministry of Economy and Competitiveness (Government of Spain). Project ELENA: CGL2012-31668: "Eventos climáticos extremos: Caracterización, variabilidad espacio-temporal e impacto en los sistemas naturales" [Extreme climate events: characterization, spatio-temporal variability and impacts in natural systems], project CGL2015-69985 "Variabilidad, tendencias y extremos del clima en la vertiente Mediterránea de la península Ibérica" [Climate variability, trends and extremes in the Mediterranean side of the Iberian peninsula] and the Slovenian Research Agency (ARRS), programs P4-0015 and P4-0107 and project V4-1419 funded by ARRS and the Ministry of Agriculture, Forestry and Food of the Republic of Slovenia. EMDC benefited from a Ph.D. contract grant (no. BES-2013-064453) funded by the Spanish Ministry of Economy and Competitiveness and also from a Short-Term Scientific Mission (STSM) financed by COST Action STREeSS (COST-FP1106).

#### ACKNOWLEDGMENTS

This article is based on work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology). We thank Martin Cregeen for language editing and three reviewers for their helpful suggestions which improved the paper.




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Martinez del Castillo, Longares, Griˇcar, Prislan, Gil-Pelegrín, Cufar and de Luis. This is an open-access article distributed under the terms ˇ of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Timing of False Ring Formation in Pinus halepensis and Arbutus unedo in Southern Italy: Outlook from an Analysis of Xylogenesis and Tree-Ring Chronologies

Veronica De Micco<sup>1</sup> \*, Angela Balzano<sup>1</sup> , Katarina Cufar ˇ <sup>2</sup> , Giovanna Aronne<sup>1</sup> , Jožica Gricar ˇ 3 , Maks Merela<sup>2</sup> and Giovanna Battipaglia4,5

<sup>1</sup> Department of Agricultural Sciences, University of Naples Federico II, Naples, Italy, <sup>2</sup> Biotechnical Faculty, Department of Wood Science and Technology, University of Ljubljana, Ljubljana, Slovenia, <sup>3</sup> Slovenian Forestry Institute, Ljubljana, Slovenia, <sup>4</sup> Department of Environmental, Biological and Pharmaceutical Sciences and Technologies, Second University of Naples, Caserta, Italy, <sup>5</sup> Laboratoire Paléoenvironnements et Chronoécologie, École Pratique des Hautes Études, Université de Montpellier, Montpellier, France

#### Edited by:

Ute Sass-Klaassen, Wageningen University, Netherlands

#### Reviewed by:

Joana Vieira, University of Coimbra, Portugal Annie Deslauriers, Université du Québec à Chicoutimi, Canada

> \*Correspondence: Veronica De Micco demicco@unina.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 03 March 2016 Accepted: 06 May 2016 Published: 24 May 2016

#### Citation:

De Micco V, Balzano A, Cufar K, ˇ Aronne G, Gricar J, Merela M and ˇ Battipaglia G (2016) Timing of False Ring Formation in Pinus halepensis and Arbutus unedo in Southern Italy: Outlook from an Analysis of Xylogenesis and Tree-Ring Chronologies. Front. Plant Sci. 7:705. doi: 10.3389/fpls.2016.00705 Mediterranean tree rings are characterized by intra-annual density fluctuations (IADFs) due to partly climate-driven cambial activity. IADFs are used as structural signals to gain information on relations between environmental conditions and eco-physiological processes during xylogenesis, with intra-annual resolution. To reach an unbiased synchronization of the IADF position within tree rings and seasonal fluctuations in environmental conditions, it is necessary to know the timing of cambial activity and wood formation, which are species- and site-specific processes. We applied the microcoring technique to analyze xylogenesis in Pinus halepensis and Arbutus unedo. To the best of our knowledge, this is the first attempt to study xylogenesis in a hardwood species forming frequent IADFs. Both species co-occur at a site in southern Italy characterized by a Mediterranean climate. To facilitate tree-ring dating and identification of IADFs, we performed traditional dendroecological analysis. We analyzed xylogenesis during summer, which is considered a constraint for xylogenesis and a trigger for IADF formation. We followed the different phases of cell development in the current wood increment with the aim of evaluating whether and which type of IADFs were formed. We additionally analyzed the same phases again in September and in winter to verify the possible formation of IADFs in fall and whether cell production and differentiation was completed by the end of the calendar year. Both species formed the same type of IADFs (earlywood-like cells within latewood), due to temporary growth restoration triggered by rain events during the period of summer drought. At the end of the calendar year, no cells in the phases of enlargement and secondary cell wall deposition occurred. A. unedo was more sensitive than P. halepensis because IADFs were formed earlier in the season and were more frequent in the tree-ring series. The dendro-anatomical approach, combining analysis of tree-ring series and of xylogenesis, helped to detect the period of IADF

formation in the two species. Results are discussed in functional terms, highlighting the environmental conditions triggering IADFs, and also in methodological terms, evaluating the applicability of xylogenesis analysis in Mediterranean woods, especially when the formation of IADFs is not uniform around the stem.

Keywords: cambial activity, cambial phenology, mediterranean climate, intra-annual density fluctuations, tree rings

#### INTRODUCTION

Tree rings are well-established climate proxies: environmental information can be extracted from dated tree-ring series by analyzing the variability in tree-ring width, earlywood and latewood widths, wood density and functional anatomical traits (e.g., Eckstein and Schmidt, 1974; Schweingruber, 1978; Eckstein et al., 1979; Tardif, 1996; Cherubini et al., 2003; Grudd, 2008; Fonti et al., 2010; Esper et al., 2012; Beeckman, 2016). The analysis of tree-ring series is relatively easily applied in plants from temperate regions characterized by a clear seasonality inducing a dormancy in cambial activity once a year (e.g., Prislan et al., 2013a). Under such conditions, each ring corresponds to one calendar year, with earlywood and latewood, respectively, linked to spring and summer climatic conditions (Fritts, 1976).

The advancement of tools of digital image analysis has raised new interest in the application of quantitative wood anatomy to tree-ring series to study a plant's response to environmental changes (Fonti et al., 2010; von Arx and Carrer, 2014). The analysis of tree-ring series is more and more applied in various climatic regions worldwide and has a great potential to reconstruct environmental information with seasonal or intraseasonal resolution, especially under conditions promoting an alternation of growth flushes and dormancy during the year (De Micco et al., 2016a). Within this context, wood of Mediterranean species is particularly interesting because frequent fluctuations in climatic factors exert a control on cambial activity, thus triggering the formation of intra-annual density fluctuations (IADFs) in tree rings (Cherubini et al., 2003; De Micco and Aronne, 2009; De Micco et al., 2016a). The increasing drought and changes in the frequency of precipitation and extreme events forecasted for the Mediterranean basin (IPCC Working Group I et al., 2013) will likely influence trends in cambial phenology and xylogenesis, thus the frequency and structural features of IADFs (Vieira et al., 2010). Since different species can show different sensitivities to fluctuating environmental conditions and can be differently prone to form IADFs in various environments, understanding the patterns and processes of xylem formation in response to variable environmental conditions is valuable for forecasting species growth fitness and adaptation capability (Camarero et al., 2010), which are ultimately linked to forest dynamics, biomass production and biogeochemical cycles (Cuny et al., 2015; Xia et al., 2015; Pacheco et al., 2016).

IADFs have been considered a constraint in dendrochronology until recently but they have been finally accepted as "positive anomalies" in tree rings because their analysis furnishes information on the relations between environmental conditions and eco-physiological processes during wood formation, with intra-annual resolution (Campelo et al., 2007a,b, 2013; de Luis et al., 2007, 2011a; De Micco et al., 2007, 2012, 2014; Battipaglia et al., 2010, 2014; Vieira et al., 2010; Rozas et al., 2011). In the last decade, numerous studies have analyzed IADFs in Mediterranean softwoods and hardwoods, also raising hypotheses on the factors responsible for their formation (De Micco et al., 2016a). Several classifications of IADFs have been proposed based on their position within the ring and on anatomical traits (e.g., lumen diameter and cell-wall thickness) of the xylem conduits in the IADF zone (Campelo et al., 2007a,b, 2013, 2015; Battipaglia et al., 2010, 2014; De Micco et al., 2012, 2014). In Mediterranean conifers growing at coastal sites in south-eastern Spain, the most common IADFs are classified as type-L (large-lumen and thin-walled earlywood-like cells within narrow-lumen and thick-walled latewood conduits), whose formation has been linked to the reactivation of cambial activity, due to favorable conditions in fall after a period of summer drought (de Luis et al., 2011a,b; Campelo et al., 2013; Novak et al., 2013a,b, 2016; Carvalho et al., 2015; Vieira et al., 2015). Type-E IADFs (narrow-lumen and thick-walled latewood-like cells within large-lumen and thin-walled earlywood conduits) have been described in Pinus pinaster growing in Italy and are considered a response to summer drought conditions inducing stomata closure (De Micco et al., 2007). Both types of IADFs have also been found in a few Mediterranean hardwoods (Campelo et al., 2007a, 2010; Battipaglia et al., 2010, 2014; De Micco et al., 2012, 2014).

Hypotheses on the reason for IADF formation derive from indirect evidence, namely correlations with climate variables (i.e., temperature and precipitation) but knowledge gaps still remain to be filled (Battipaglia et al., 2016; De Micco et al., 2016a; Zalloni et al., 2016). The formation of IADFs in Mediterranean species has been mainly linked to water availability, which affects the turgor-driven expansion of xylem cells (Sperry et al., 2006; De Micco et al., 2016a). However, to confirm such a hypothesis, the study of IADFs during their formation is needed through analysis of xylogenesis aimed at unraveling how and when wood with specific anatomical traits is formed (Vaganov et al., 2006; Camarero et al., 2010; de Luis et al., 2011b; Vieira et al., 2014; Novak et al., 2016). The analysis of xylogenesis in woods forming IADFs with high frequency is useful for achieving precise synchronization of the IADF position within the tree ring and specific environmental fluctuations triggering them. Analysis of cambial activity has been widely applied through microcoring techniques, mostly in conifers and hardwoods growing in temperate climates (Rossi et al., 2003, 2006, 2008, 2012; Cufar ˇ et al., 2011; Prislan et al., 2013b; Gricar et al., 2014 ˇ ; Pérez-de-Lis et al., 2016). Analysis of xylogenesis has also recently been applied

to tree rings forming IADFs in Pinus species (de Luis et al., 2011a; Vieira et al., 2014, 2015; Novak et al., 2016). However, to the best of our knowledge, there are no reports dealing with the study of IADF-genesis in hardwood species.

In this study, we analyzed cambial activity in a softwood and a hardwood species, Pinus halepensis Mill. and Arbutus unedo L., co-occurring at a site in southern Italy, characterized by a Mediterranean climate. The work aimed at: (1) evaluating whether and which type of IADFs were formed during summer, and (2) highlighting which weather conditions were concomitant or closely preceding IADF formation. Together with the two ecological aims, we also pursued a third methodological issue. In view of the fact that the formation of IADFs is a variable phenomenon along the stem circumference in Mediterranean woods (Cherubini et al., 2003), we aimed to verify the degree of applicability of the micro-coring technique to the two species, especially considering that A. unedo is a hardwood species. Thus, we applied common dendroecological techniques to analyze tree-ring chronologies and evaluate their variability within and between plants.

We performed microcore sampling and microscopy analysis of thin cross sections during the period of summer aridity (which is considered a factor limiting plant growth and predisposing the occurrence of IADFs) to detect the time of IADF formation. In order to do this, we first evaluated the onset of latewood formation during the development of tree rings with and without IADFs, then continued the analysis of xylogenesis, searching for earlywood-like cells, until we detected the formation of an IADF in the 2014 tree-ring. This was based on the hypothesis that IADFs in both species are formed during the summer months, triggered by water stress followed by the temporary restoration of growth as a consequence of favorable conditions. We therefore hypothesized that both species experience a bimodal pattern of cambial activity, as reported in Camarero et al. (2010), thus completing their ring growth by the end of the calendar year. To verify this hypothesis, we also analyzed xylogenesis at additional dates until the end of the calendar year. This helped us to verify that ring growth had been completed by the end of the calendar year, as commonly assumed in traditional dendrochronology, and to verify the possible formation of other IADFs primed by rain events in fall.

#### MATERIALS AND METHODS

#### Species and Study Area

The study was conducted in 2014 on plants of Pinus halepensis Mill. and Arbutus unedo L. co-occurring at a site at Quisisana, Castellammare di Stabia (Naples) in southern Italy. The sampling site (40<sup>0</sup> 683 N, 14<sup>0</sup> 481 E, 346 m a. s. l) is characterized by typical Macchia vegetation, with shrub and tree species including Quercus ilex L., P. pinaster Aiton, P. halepensis Mill., Castanea sativa Mill., Fraxinus ornus L., Acer opalus Mill. subsp. neapolitanum, Erica arborea L., Laburnum anagyroides Medik, A. unedo L., and Ruscus aculeatus L. The climate at the site is Mediterranean, with hot, dry summers followed by mild, wet winters. According to data recorded at the closest meteorological station, 10 km from the sampling site (Pimonte, 400 672N, 14<sup>0</sup> 50E, 370 m a. s. l), during the year of sampling 2014, annual mean temperature was 15.5◦C with the hottest month being August (monthly average mean temperature of 23.3◦C) and the coldest month being January (monthly average mean temperature of 9.8◦C). The cumulative annual precipitation was 1020 mm; the wettest month was January, with a cumulative monthly precipitation of 447 mm, while the lowest value was reached in August (cumulative precipitation of 1.4 mm). The worst aridity period lasted from June to the beginning of September (**Figure 1**). Longer meteorological series are not available for the Pimonte site. Meteorological series from other nearby stations in the Campania region were analyzed and showed that meteorological conditions in 2014 did not deviate considerably from those registered in the period 2005– 2014.

#### Tree-Ring Data

Tree-ring chronologies were built through common dendroecological techniques to facilitate the synchronization of tree rings and IADFs in the collected microcores.

Core sampling was carried out in March 2015 on 15 dominant trees of P. halepensis and 15 plants of A. unedo. Diameter at breast height (DBH) was measured and two cores were taken at breast height as well from each tree (west and east directions) with an increment borer (diameter 5 mm). The cores were transported to the laboratory and air dried. The surface of the cores was polished using sand paper of different grain-sizes, and tree-ring width (TRW) measurements were made at a resolution of 0.01 mm, using LINTAB measurement equipment fitted with a stereoscope and equipped with TSAP Win software (Frank Rinn, Heidelberg, Germany). Tree-ring series were visually cross-dated and compared using standard dendrochronological techniques (Stokes and Smiley, 1968). The

cross-dating accuracy was then checked using the program COFECHA (Holmes, 1983). The program ARSTAN (Cook, 1985) was used to remove growth trends related to tree age and competition, producing standardized tree-growth indices. Series were detrended with a 10-year spline to remove longterm growth trends embedded in the raw tree-ring series, which were thought to be induced by non-climatic influences, such as aging and competition between trees (Fritts, 1976). Once all series had been validated, tree-ring chronologies were constructed. Descriptive statistics were computed, including standard deviation (SD), which estimates the variability of measurements and the expressed population signal (EPS), thus indicating the level of coherence of the constructed chronology and how it portrays a hypothetical perfect population chronology.

The occurrence of IADFs was quantified in each core by considering each ring as a growth increment and distinguishing the true annual rings from IADFs through visual analysis of the features of boundaries (e.g., abruptness of changes between earlywood and latewood cells) and considering data from cross dating in cases of doubt (Cherubini et al., 2003; De Micco et al., 2016a). We finally calculated the frequency of IADFs in each plant as the ratio between the number of IADFs and the total number of increment growth. The Chi-square test was used to compare the occurrence of IADFs between the two species (two-way contingency table).

#### Microcore Sampling and Microscopy

Microcores (1.8 mm in diameter) were collected with a Trephor tool (Rossi et al., 2006) from six trees of P. halepensis and eight plants of A. unedo at breast height, following a spiral with a distance of 2 cm between consecutive samples. Since we were interested in following xylogenesis throughout the aridity period, microcores were collected at weekly intervals starting from June. Sampling was interrupted in August because the observation of microcores (as reported below) showed that IADFs had been already formed. Further microcore collections were done after rain events at the beginning of September 2014 and at the end of the calendar year to check whether other IADFs had been formed in August or during fall.

Microcores were immediately fixed in 70% ethanol and stored at 4◦C. Microcores were then embedded in paraffin using a Leica TP 1020-1 (Nussloch, Germany) tissue processor for dehydration in alcohol series (70, 90, 95, and 100%) and bio-clear (d-limonene) for paraffin infiltration. Paraffin blocks were cut with a semi-automatic rotary microtome RM 2245, Leica, (Nussloch), thus obtaining cross sections (9 µm thick), which were flattened on slides pre-treated with albumin. The slides were dried at 70◦C for 30 min and cleaned of residual paraffin by washing with bio-clear and ethanol. The sections were then stained with a water solution of safranin and astra blue (Werf van der et al., 2007) and permanently mounted on glass slides in Euparal (Bioquip Rancho Dominguez, California). The sections were observed under a BX61 transmission light microscope (Olympus, Hamburg, Germany), while the quantification of anatomical parameters was performed through a Nikon Eclipse 800 microscope equipped with Nis Elements BR3 (Melville, NY, USA) image analysis software on microphotographs captured with a Nikon DS-Fi1 digital camera.

In P. halepensis, we focused on the development of tracheids, using visual criteria based on lumen size and wall thickness (de Luis et al., 2007, 2011a; Novak et al., 2016) measured by means of the eyepiece micrometer while looking through the microscope. The distinction between earlywood and latewood was based on the application of Mork's definition (Mork, 1928). In A. unedo we focused on the development of vessels and imperforate tracheary elements, using visual criteria based not only on lumen size and wall thickness but also on vessel frequency; the transition between earlywood and latewood was often diffuse but latewood could be distinguished by the presence of narrower (halved lumen diameter) and less frequent vessels than earlywood (De Micco et al., 2016b). The following phases of cell development were considered according to de Luis et al. (2007) and Cufar ˇ et al. (2008, 2011): cambial cells (CC), post cambial cells (PC), cells with developing secondary wall (SW) and mature cells with lignified secondary wall (MC) (**Figure 2**). CC were radially flattened, with thin cell walls that stained blue. PC were enlarging cells in the phase of postcambial growth, which also stained blue. SW were immature xylem derivatives with developing (thickening and lignifying) secondary walls. SW cells showed birefringence under polarized light and stained blue and light red, depending on the progress of lignification. MC were cells without any trace of protoplast in the lumen and had fully deposited and lignified cell walls that colored intense red by safranin. The cambium was considered productive when PC were detected.

In P. halepensis, the number of cells in different developmental phases was counted. In A. unedo, since xylem consisted of conduits and fibers that were not arranged in ordered radial rows, the width of the cambium zone and developing xylem corresponding to the various developmental stages was measured as in Cufar et al. (2008, 2011) ˇ . Measurements were taken in the cambium and in the developing xylem ring along three radial rows. Measurements along three radial rows were averaged.

Finally, in order to detect the time of IADF formation, we analyzed the microsections from subsequent microcores by focusing on cells in SW and MC phases to highlight changes in cell lumen size and wall thickness marking the transition from earlywood to latewood and vice versa. For each series of microcores, we classified SW and MC cells into four categories: earlywood (EW), latewood (LW), earlywood-like (EW-like), and secondary production of latewood (SLW) to distinguish them from customary LW. When the transition from earlywood to latewood was detected only once, and latewood formation was maintained until the end of the calendar year, the ring was classified as "not having an IADF". In contrast, when the transition from earlywood to latewood was followed by additional transition from latewood to earlywood (EW-like) and from EW-like to latewood (SLW), then the ring was classified as "having an IADF". We calculated the percent of plants showing SW and MC cells in each category per each date.

FIGURE 2 | Developing xylem in Pinus halepensis (A) and Arbutus unedo (B). Moving from the cambial zone toward the center of the stem, the following cells are encountered: cambial cells (CC), enlarging post cambial cells (PC), cells developing secondary walls (SW), and mature cells with lignified secondary wall (MC). Scale bars = 100µm.

#### RESULTS

## Tree-Ring Chronologies and IADF Occurrence

The P. halepensis trees had a DBH of 52.24 ± 5.41 cm (mean value ± SD) and belonged to the same age class, with a mean of 90 ± 12 years (**Figure 3A**). High EPS values (>0.85) for the period of 1921–2014 indicated that the mean chronology was representative of radial growth variations of the whole population of trees (Wigley et al., 1984). The MS value (0.25) and the r bar value (0.82) showed a strong common growth signal among individuals. Thus, the variability among individuals and between twin cores from the same tree was not high. Despite the occurrence of IADFs, it was still possible to recognize, measure and cross-date the rings and build a robust mean chronology (**Figure 3A**). The percentage of IADFs in P. halepensis was 20.19 ± 2.63% (mean value ± SE), with a minimum of 6.19% and a maximum of 34.56%. In 2014, 58.3% of the trees formed IADFs appearing as earlywood-like cells within latewood (**Figures 4A,B**).

A different situation was found in the A. unedo plants. They were younger than the P. halepensis trees, with DBH of 7.97 ± 1.78 cm and related tree-ring chronologies spanning from 1995 to 2014 (**Figure 3B**). Very high variability was found among individuals and between twin cores from the same plant, as shown by high values of standard deviation (**Figure 3B**). Furthermore, the A. unedo cores contained a very high frequency of IADFs, hampering cross-dating. In this shrub hardwood, the percentage of IADFs was 36.65 ± 3.02% (mean value ± SE), with a minimum of 21.43 and maximum of 48.39%. In some cases, the tree-ring from the same calendar year contained more than one IADF. In 2014, 72.3% of the plants formed IADFs appearing as earlywood-like cells within latewood (**Figures 4C,D**).

The analysis of the two-way contingency table showed that the percentage of IADFs was significantly higher (p < 0.0001) in A. unedo than in P. halepensis tree-ring series.

#### Cambial Productivity

As expected, xylem formation in P. halepensis and A. unedo had started prior to the first sampling on 5 June. At the beginning of June, in P. halepensis the cambial zone (CC) consisted of 5.81 ± 0.14 cells (mean value ± SE) and the current xylem growth ring consisted of 11.72 ± 1.65 cells in different phases of differentiation (PC and SW) and included 1.99 ± 0.81 fully differentiated cells (MC). The number of CC slightly increased in July (**Figure 5A**). The number of PC was highest on 12 June, remained more or less stable until the end of July and was very low at the beginning of September. The number of SW and MC cells varied due to the usual variability around the stem. In December, CC consisted of 5.00 ± 0.23 cells and no differentiating PC or SW cells were observed, so the current tree-ring consisted of MC cells only.

At the time of the first sampling, the CC of A. unedo was 18.35 ± 0.82 µm wide (mean value ± SE) and on average consisted of three cells (**Figure 5B**). Its width slightly increased thereafter and reached a maximum on 26 June (34.10 ± 1.10 µm). The thickness of the PC zone reached its maximum on 12 June (63.57 ± 18.67 µm), whereas almost no PC cells were observed on 31 July or at the beginning of September. The zone of SW cells remained wide throughout the summer. In September, the current tree-ring mainly consisted of SW with a small proportion of MC cells. At the end of December, we could observe no cell production and almost all cells of the current ring were fully differentiated. The width of the currently formed tree ring also varied around the stem in A. unedo.

#### Wood Formation and IADFs

The overall analysis of the microcore data showed that at the time of first sampling, only 33.4 ± 4.9% (mean value standard error) and 46.5 ± 6.2% of the xylem increment (tree ring) of the current year had been already formed in P. halepensis and A. unedo, respectively. In June, it consisted of mainly SW cells with earlywood characteristics.

The analysis of anatomical characteristics of wood formed during summer 2014 was much easier in P. halepensis than in

tree-ring boundaries; long arrows indicate the IADF. Bars = 100 µm.

A. unedo. In P. halepensis, two (out of six) analyzed trees, had a "normal" 2014 tree ring, consisting of earlywood followed by latewood with no IADFs (**Figures 4A** and **6A**). In the other four trees, IADFs were formed during the summer (**Figures 4B** and **6B**). As summarized in **Table 1**, in the two trees without 2014-IADF, MC in the increment growth consisted of earlywood tracheids until mid-July (**Figures 4A** and **6A,E**). The first latewood tracheids appeared on 24 July and the production of latewood tracheids continued until the completion of the annual ring (**Table 1**). In the other four trees, the first latewood tracheids appeared completed between 24 and 31 July, and latewood production was followed again by earlywood-like cells, with large lumina and thin cell walls on the successive dates. In samples collected in September, the first mature earlywood-like tracheids were observed (**Figures 6F,G**; **Table 1**) while current PC and SW cells evolved into latewood cells, as found in the last date analyzed (**Figures 4B** and **6B,F**; **Table 1**). From mid-September until the end of the year, only latewood cells were produced. It is worth highlighting that if IADFs formed in 2014 were classified by visual analysis of the complete tree ring without considering the timing of cambial activity, they could have been classified as either E-type IADF (latewood-like cells within earlywood) or L-type IADF (earlywood-like cells within latewood). By adding the information on xylogenesis, as well as temperature and precipitation of the corresponding calendar year, the rings with IADFs would be classified as L-IADFs.

In the case of A. unedo, the formation of the different types of wood showed high variability among plants. It was possible to follow the genesis of different wood structures in six (out of the eight) analyzed plants. More specifically, in the other two plants, consecutive microcores (taken in spiral along the stem) showed very different growth increments. Due to the great variability of ring width around the stem it was not always possible to differentiate unequivocally between false and very narrow "normal" tree-rings. These plants were also characterized by very diverse twin cores and corresponding tree-ring series, which could not be cross-dated. We therefore excluded them from the analysis. Of the remaining six plants, two did not form IADFs. As summarized in **Table 1**, in these plants, the first mature latewood vessels were evident on 12 and 19 June and the production of this cell type continued until the completion of the annual ring (**Figures 4C** and **6C**; **Table 1**). In the other four plants, MC observed until the middle of June were of the earlywood type (**Figure 6C**; **Table 1**). The first mature latewood vessels appeared progressively from 19 June to 17 July in the different plants (**Figures 6H,I**, black arrowhead, **Table 1**). In these four plants, after the formation of the first latewood vessels, new earlywood-like cells were formed starting from the middle of July (**Figures 6D,I** blue arrowhead, **Table 1**). A successive production of very narrow latewood was recorded progressively starting from 24 July. Similar to P. halepensis, also in A. unedo we classified the 2014-IADFs as L-IADFs. From September until the end of the year, only latewood cells were produced and no other IADFs were formed. Only in one plant did we finally record two L-IADFs in the 2014 ring.

not forming IADFs in 2014 were considered together. Mean values and standard errors are shown.

## DISCUSSION

Analysis of cambial activity supported by dendro-anatomical investigations in the softwood P. halepensis and the hardwood A. unedo allowed us to reconstruct the timing of IADF formation and to hypothesize possible reasons for their formation.

Intra-annual density fluctuation identification and classification is still mainly done by visual analysis of cores and/or microsections under the microscope. When analyzing a tree ring for the presence of an IADF, the operator considers the sequence of wood types (earlywood, latewood) along the tree-ring width. A calendar tree-ring contains an IADF when



the sequence of wood types encountered from the beginning of the tree ring toward the cambium is the following: earlywood, latewood (or latewood-like cells), earlywood (or earlywood-like cells) and latewood. The IADF can then be classified into Eor L-type IADF (De Micco et al., 2016a). It is clear that the classification into one type or the other depends on identification of the region of the ring in which the true latewood begins. Analysis of relations between IADFs and climatic data also helps their classification because the two types of IADFs, can occur at different positions within the growth ring and have completely different functional significance. For example, A. unedo growing at a site in Southern Italy has been recently shown to be able to form two types of IADFs, according to water availability at the growing sites (Battipaglia et al., 2010, 2014). More specifically, under xeric conditions, this species forms E-IADFs, meaning that the "anomaly" is the formation of latewood-like cells in a period when earlywood is expected, thus triggered by a particularly severe or unexpected period of drought (Battipaglia et al., 2010). Under more mesic conditions, A. unedo developed L-IADFs, meaning that the "anomaly" was the formation of earlywood-like cells in a period when latewood should have been formed; such L-IADFs were ascribed to rain events occurring during fall after a period of summer drought. This mechanism of L-IADF formation has also been highlighted as typical of Pinus species (Zalloni et al., 2016). Such a seasonal dynamics of wood development follows the bimodal pattern described for P. halepensis from arid and semi-arid ecosystems in Spain (de Luis et al., 2011a), and also reported recently by Camarero et al. (2010) in Juniperus thurifera from sub-humid and semi-arid Mediterranean continental sites in Spain, and in the co-occurring P. halepensis only in the more xeric site. Such a pattern has been also reported for P. pinaster growing on sand dunes in Portugal by Vieira et al. (2015). The bimodal pattern of wood growth is due to cambial reactivation triggered by spring and fall precipitations, which control cell enlargement and cell wall deposition after winter low-temperatures and summer drought stresses (Camarero et al., 2010; Pacheco et al., 2016). In our study, to classify IADFs objectively in the two analyzed species and to evaluate the period of their formation, xylogenesis proved to be a valuable tool because it allowed identification of the starting moment for latewood formation, as well as observation of the progressive formation of IADFs while they were "under construction" (de Luis et al., 2011b; Novak et al., 2016). Considering the timing of IADF formation and keeping in mind the precipitation pattern observed in 2014, we confidently classified the 2014 IADFs as L-type, following a bimodal pattern of xylogenesis. However, in our case, the second growth flash, leading to the second earlywood formation during the calendar year, already occurred in summer and not in fall as hypothesized for Mediterranean species (De Micco et al., 2016a). We could speculate that, in this specific case, fall precipitations can lead to more than one IADF per ring. This is sometimes observed in A. unedo where such additional IADFs would be better classified L <sup>+</sup>-IADFs according to Campelo et al. (2007b, 2013). The high variability of behavior of different plants in various environments strengthen the need for a common and unambiguous classification of IADFs, also considering

their position within tree rings, to achieve a correct functional interpretation of them (De Micco et al., 2016a).

In the two analyzed species, the formation of IADFs appeared irregularly in time and different individuals showed different predispositions to IADF development in different years, in agreement with other studies (Cherubini et al., 2003). This phenomenon indicates the capacity of these two species to switch from a unimodal pattern of xylogenesis toward a bimodal wood growth, and vice versa, from one year to another. The two species responded with the same strategy (i.e., the formation of L-IADFs) to fluctuating environmental conditions, but their sensitivity was different. The onset of latewood formation and the appearance of IADFs was shifted a couple of weeks earlier in A. unedo than in P. halepensis. However, the occurrence of mature latewood cells in A. unedo was still delayed compared with the few studied broadleaved deciduous species such as chestnut and beech (Cufar ˇ et al., 2008, 2011). A. unedo appeared to be more sensitive than P. halepensis also because of a higher frequency of IADFs and because several growth rings were often formed per year. This would indicate a better and quicker ability to induce dormancy and activation of the cambium after fluctuating environmental conditions. This different sensitivity is not surprising because of the different size, age, and growth and reproductive strategies of the two species which dictate different hydraulic constraints and resource use/allocation. For instance the younger age of A. unedo plants compared with P. halepensis can partly explain their higher tendency to form IADFs. Indeed, IADF frequency is generally higher in younger than in older trees, probably not only because young plants have shallow root systems and may thus be less able to access deep soil layers but also because young plants show increased tree-ring width, which is positively correlated to IADF frequency (Rigling et al., 2001, 2002; Cherubini et al., 2003; Bogino and Bravo, 2009; Hoffer and Tardif, 2009; Rozas et al., 2011; Novak et al., 2013b; Campelo et al., 2015; Pacheco et al., 2016).

The appearance of the first mature latewood conduits in the two analyzed species occurred well after the onset of the summer drought period (May). Indeed, wood formation is the result of the integration of complex cellular processes, in which cell-wall thickening and lignification lags behind cell enlargement by as much as a month or more, according to the recent model by Cuny et al. (2015). The time-lag observed between the increase in xylem size (linked to cell enlargement) and accumulation of woody biomass (due to cell wall thickening and lignification) shows differences between earlywood and latewood, and it has been quantified in a range of 27–49 days in different climates, with maximum values in the Mediterranean region (Cuny et al., 2014, 2015). The application of this time-lag principle to our samples would indicate that latewood conduits are produced by cambium activity from the beginning of the drought period but, given that cell-wall deposition and maturation requires several weeks (up to ≈ 7) (Prislan et al., 2009, 2013a; Cuny et al., 2014), latewood thick-walled cells become evident only later. Similarly, in tree rings showing IADFs, a new production of earlywood may have been primed by rain pulse events during July (which was characterized by almost twice as much precipitation as June). Increased cell enlargement in July thus possibly triggered the formation of earlywood-like cells, which appeared mature only at the end of July and August, based on the time required for cell wall thickening (up to≈ 4 weeks) as estimated by Prislan et al. (2009, 2013b) and Cuny et al. (2014).

The two analyzed species showed wood production during the summer drought period, which is considered limiting for growth, also perhaps inducing a halt in cambial productivity in some species (Liang et al., 2006; Camarero et al., 2010; Pacheco et al., 2016). However, this could be explained by temporary favorable conditions, specific resource use efficiency and strategies, or might have another basis, phylogenetic or biogeographical, given that many Mediterranean species perform costly metabolic processes precisely under unfavorable summer drought periods (Aronne and Wilcock, 1994).

From a methodological viewpoint, the micro-coring technique and analysis of xylogenesis were easily applied to tree rings showing IADFs of P. halepensis, whereas it was more difficult to apply them in A. unedo because of the very high variability of IADFs among and within plants in different years. However, cross-dating was a helpful tool for identifying the beginning of the last growth increment (2014) and for verifying wood growth variability in the same stem. We thus suggest that, when studying xylogenesis in Mediterranean plants forming a high frequency of IADFs and more than one IADF per calendar year, it is useful to perform a preliminary tree-ring analysis in order to predict the applicability of micro-coring in single plants, thus excluding from the micro-coring experiment those trees having high variability of wood growth around the stem and those having clear anomalies.

To conclude, with the support of dendroecological analysis, micro-coring allowed the identification of the period of IADF formation in the two Mediterranean species confirming the hypothesis of the occurrence of a bimodal pattern of cambial activity. Both species were prone to form IADFs that were classified as L-type, indicating a period of growth flash due to favorable environmental conditions for growth occurring during summer, and not in fall as reported for other Mediterranean species (e.g., Camarero et al., 2010; de Luis et al., 2011b; Campelo et al., 2013; Carvalho et al., 2015; Zalloni et al., 2016). A possible explanation, still to be verified is that different kinds of L-IADFs exist (L- and L <sup>+</sup>-type); they would be triggered by temporary favorable conditions occurring during summer and fall, respectively.

The formation of L-IADFs can be considered a way of improving the hydraulic conductivity of wood (Sperry et al., 2006) when water is unexpectedly available after a period of severe drought. As a consequence, species showing high plasticity in cambial productivity, thus prone to form L-IADFs, promptly after a positive climatic event (e.g., unexpected summer rain pulse) following a period of severe drought (e.g., dry periods at the end of spring), should have an advantage under fluctuating environmental conditions over those not able to form IADFs.

#### AUTHOR CONTRIBUTIONS

VDM, AB and GB made a substantial contribution to the conception and design of the study. AB performed sample

collection. AB and MM performed sample preparation. VDM, AB, JG, KC and GB made a substantial contribution to the analysis of tree-ring series and anatomical signals in microsections and in data analysis. VDM, AB, KC and GA contributed to the interpretation of the overall data. VDM, KC, MM, GA and GB contributed to the analysis tools. VDM wrote the main part of the manuscript. AB and GB contributed to writing the text. All authors contributed to final revisions of the manuscript and read and approved the submitted version of the manuscript.

#### FUNDING

The work of AB was partially funded by COST Action STREeSS (COST-FP1106) through a Short-Term Scientific

#### REFERENCES


Mission (STSM). The work of KC, MM, and JG was funded by the Slovenian Research Agency (ARRS), programs P4-0015 and P4-0107, and by the Ministry of Agriculture, Forestry and Food and ARRS, project V4-1419.

#### ACKNOWLEDGMENTS

This article is based on work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology). We thank Luka Krže and Primož Habjan for great help with laboratory work, Peter Prislan for valuable instructions with wood formation analyses, and Enrica Zalloni for helping in the identification of IADFs. We thank Martin Cregeen for editing the English text.




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 De Micco, Balzano, Cufar, Aronne, Gri ˇ ˇcar, Merela and Battipaglia. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Missing Rings in Pinus halepensis – The Missing Link to Relate the Tree-Ring Record to Extreme Climatic Events

Klemen Novak1,2 \*, Martin de Luis<sup>1</sup> , Miguel A. Saz<sup>1</sup> , Luis A. Longares<sup>1</sup> , Roberto Serrano-Notivoli<sup>1</sup> , Josep Raventós<sup>2</sup> , Katarina Cufar ˇ <sup>3</sup> , Jožica Gricar ˇ 4 , Alfredo Di Filippo<sup>5</sup> , Gianluca Piovesan<sup>5</sup> , Cyrille B. K. Rathgeber<sup>6</sup> , Andreas Papadopoulos<sup>7</sup> and Kevin T. Smith<sup>8</sup>

#### Edited by:

Achim Braeuning, University of Erlangen-Nuremberg, Germany

#### Reviewed by:

Bao Yang, Chinesse Academy of Sciences, China Ze-Xin Fan, Chinese Academy of Sciences, China

> \*Correspondence: Klemen Novak knovak@unizar.es

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 23 February 2016 Accepted: 11 May 2016 Published: 31 May 2016

#### Citation:

Novak K, de Luis M, Saz MA, Longares LA, Serrano-Notivoli R, Raventós J, Cufar K, Gri ˇ car J, Di ˇ Filippo A, Piovesan G, Rathgeber CBK, Papadopoulos A and Smith KT (2016) Missing Rings in Pinus halepensis – The Missing Link to Relate the Tree-Ring Record to Extreme Climatic Events. Front. Plant Sci. 7:727. doi: 10.3389/fpls.2016.00727 <sup>1</sup> Department of Geography and Regional Planning – Instituto de Investigación en Ciencias Ambientales, University of Zaragoza, Zaragoza, Spain, <sup>2</sup> Department of Ecology, University of Alicante, Alicante, Spain, <sup>3</sup> Department of Wood Science and Technology, Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia, <sup>4</sup> Slovenian Forestry Institute, Ljubljana, Slovenia, <sup>5</sup> Dendrology Lab, Department of Agriculture and Forestry Science (DAFNE), University of Tuscia, Viterbo, Italy, <sup>6</sup> LERFoB, INRA, AgroParisTech, Nancy, France, <sup>7</sup> Department of Forestry and Natural Environmental Management, T.E.I. Stereas Elladas, Karpenissi, Greece, <sup>8</sup> USDA Forest Service, Northern Research Station, Durham, NH, USA

Climate predictions for the Mediterranean Basin include increased temperatures, decreased precipitation, and increased frequency of extreme climatic events (ECE). These conditions are associated with decreased tree growth and increased vulnerability to pests and diseases. The anatomy of tree rings responds to these environmental conditions. Quantitatively, the width of a tree ring is largely determined by the rate and duration of cell division by the vascular cambium. In the Mediterranean climate, this division may occur throughout almost the entire year. Alternatively, cell division may cease during relatively cool and dry winters, only to resume in the same calendar year with milder temperatures and increased availability of water. Under particularly adverse conditions, no xylem may be produced in parts of the stem, resulting in a missing ring (MR). A dendrochronological network of Pinus halepensis was used to determine the relationship of MR to ECE. The network consisted of 113 sites, 1,509 trees, 2,593 cores, and 225,428 tree rings throughout the distribution range of the species. A total of 4,150 MR were identified. Binomial logistic regression analysis determined that MR frequency increased with increased cambial age. Spatial analysis indicated that the geographic areas of south-eastern Spain and northern Algeria contained the greatest frequency of MR. Dendroclimatic regression analysis indicated a non-linear relationship of MR to total monthly precipitation and mean temperature. MR are strongly associated with the combination of monthly mean temperature from previous October till current February and total precipitation from previous September till current May. They are likely to occur with total precipitation lower than 50 mm and temperatures higher than 5 ◦C. This conclusion is global and can be applied to every site across the distribution area. Rather than simply being a complication for dendrochronology, MR formation is a fundamental response of trees to adverse environmental conditions. The demonstrated relationship of MR formation to ECE across this dendrochronological network in the Mediterranean basin shows the potential of MR analysis to reconstruct the history of past climatic extremes and to predict future forest dynamics in a changing climate.

Keywords: Aleppo pine, tree rings, climate–growth relationship, climate, extreme growth event, Mediterranean

#### INTRODUCTION

fpls-07-00727 May 28, 2016 Time: 15:54 # 2

The reports of the Intergovernmental Panel on Climate Change (IPCC, 2013) and the European Environmental Agency (Füssel, 2012) indicate substantial warming and increased frequency and intensity of drought, heat waves, and uncertainty of regional and seasonal climatic variability across most of the Mediterranean area (FAO and Plan Bleu, 2013). The impacts of climate change are strongly related to the increase in frequency and severity of extreme climatic events (ECE; e.g., IPCC, 2013; Panayotov et al., 2013). Under such conditions, trees may decline in annual growth (Haavik et al., 2015), become more vulnerable to secondary damage from attacks by insect pests (Esper et al., 2007a; Sangüesa-Barreda et al., 2014; Robson et al., 2015) and fungal diseases (Cherubini et al., 2002), and experience higher rates of mortality (Camarero et al., 2015).

Extreme events are difficult to define, because they can vary depending on discipline (e.g., hydrology, climatology, agriculture, forestry), type of events (e.g., heat waves, drought, precipitation, strong winds), and/or climate area (e.g., Mediterranean, Continental, Alpine). Sarewitz and Pielke (2001) defined an extreme event as "an occurrence that, with respect to some class of occurrences, is notable, rare, unique, profound or otherwise significant in terms of its impacts, effects or outcome." Smith (2011) suggested the need to define extreme events synthetically, from both the "driver" (occurrence) and "response" (effect) perspectives.

Similarly, the impacts of ECE on forests are also difficult to evaluate, as they are low in frequency and generally local in occurrence. Dendrochronology is a useful tool to analyze their impacts because it can operate on wide spatial and temporal scale. It is based on analyses of annual tree-ring widths, and its characteristics which vary due to combined influence of various ecological factors and climatic conditions affecting tree growth, and therefore they can be considered as natural archives of past events with high (i.e., annual) resolution (Fritts, 2001). Treering widths and density variations have been widely used to reconstruct past temperature and precipitation (Hughes, 2002).

The anatomical structure of tree rings often contain indications of abrupt change in temperature, precipitation regime, or due to natural disturbances (Panayotov et al., 2013). Such structures in tree rings include intra-annual density fluctuations (Novak et al., 2013a; Campelo et al., 2015), resin canals (Rigling et al., 2003; Novak et al., 2013b), and proportions of earlywood and latewood (Lebourgeois, 2000; Novak et al., 2013b). Anatomical features provide a promising approach to better understand the influence of climate on tree rings (Fonti et al., 2010) and can complement traditional analysis of the climatic signal obtained from tree-ring width (Lebourgeois, 2000; Novak et al., 2013b).

Missing rings (MR) are detected through comparison of crossdated series of tree rings contained in analyzed samples. Rarely is a tree ring missing from all woody parts of the plant (Novak et al., 2011; Wilmking et al., 2012; Liang et al., 2014).

The occurrence of MR is related to the annual pattern of cambial activity (cell division) and cell differentiation which varies across the species' distribution. In Mediterranean areas the cambium generally stops dividing in summer as a consequence of drought (e.g., de Luis et al., 2011a) and in winter at the temperature limited sites (e.g., Liphschitz et al., 1984; Liphschitz and Lev-Yadun, 1986). Under favorable growing conditions, the cambium may be active almost throughout the entire year (de Luis et al., 2007, 2011a,b). In warm Mediterranean sites, prolonged dry winter periods can stop cambial activity, which resumes with increasing water availability (Camarero et al., 2010). Under particularly adverse conditions through the entire growing season, no xylem may be produced in parts of the stem, resulting in a MR (Novak et al., 2011, 2016).

Missing rings are usually considered as a "problem" that hampers the correct dating of tree rings (Grundmann et al., 2008; Rutherford and Mann, 2014). Conceptually, analysis of MR poses a challenge in that they are absent from direct observation. To our knowledge, MR have not been used as a "proxy" or temporal marker to analyze environmental processes.

Missing rings indicate an absence of wood production by the vascular cambium for one or more particular years (and parts of a tree), due to different stresses, such as unfavorable climatic events, competition (Lorimer et al., 1999; Parent et al., 2002), pests (Sangüesa-Barreda et al., 2014), or diseases (Cherubini et al., 2002). They are common in different species and in different environments (Jonsson et al., 2002; Wilmking et al., 2012; Dulamsuren et al., 2013; Liang et al., 2014). MR frequently occur in conifers and particularly in trees growing in the Mediterranean basin, with an increased frequency of occurrence in recent years noted for Pinus halepensis (Raventós et al., 2001; Novak et al., 2011, 2013b).

Missing rings are like extreme events, relatively rare in space and time. A typical dendroclimatic analysis may involve the collection of two increment core samples from each of 15 trees per site. The results of observations from 30 samples is likely not sufficient for the robust estimation of the frequency of occurrence of MR at the sampling site. The combined effect on MR occurrence due to the age and size of trees, known as the biological trend, complicates determination of the influence of climate on tree-ring growth. The binary character of the occurrence of a MR (presence, absence) is a challenge for standardization of the biological trend in MR frequency as well

as to use MR series to identify climatic signals using standard dendrochronological techniques. Different statistical approaches have been proposed so far (Campelo et al., 2007, 2015; Novak et al., 2013b; Zalloni et al., 2016), but the most appropriate tests for this type of data are not yet confirmed.

Pinus halepensis is the most widespread Mediterranean pine tree species. Its spatial distribution has the potential to produce a useful network of sites with different climatic conditions. MR in Pinus halepensis can be evaluated across wide spatial and temporal gradients and related to ECE. Therefore, to determine the potential of MR as markers of ECE in Mediterranean forests, the objectives of this study are:


## MATERIALS AND METHODS

## Dendrochronological Network for Pinus halepensis

Pinus halepensis grows throughout the entire Mediterranean area as shown on the distribution map (**Figure 1**). The network of sites selected for dendrochronological sampling consisted of newly collected and archived tree-ring series from 113 sites (Supplementary Table S1), covering the area extending from 32.23◦ to 45.67◦N latitude, 1.41◦W to 36.17◦E longitude, and altitudes from 15 to 1676 m a. s. l.

Monthly total precipitation and monthly mean temperature for 1,068 spatial grid points within the distribution range of Pinus halepensis for the 1901–2014 period were obtained from the Climatic Research Unit of the University of East Anglia. We used the CRU 3.22 dataset of 0.5◦ grid resolutions (Harris et al., 2014), and assigned the climatic data from the nearest grid point to each sampling site.

The tree-ring dataset was derived from a total of 2,593 increment cores collected from 1,509 trees of the dendrochronological network. Tree selection, core sampling, and processing were performed using standard dendrochronological techniques (Cook and Kairiukstis, 1990; Speer, 2010). Treering widths were measured under a stereo microscope with an accuracy of 0.01 mm, using the TSAP-Win program and LINTABTM 5 measuring device (Rinntech <sup>R</sup> , Heidelberg, Germany<sup>1</sup> ). Globally, a total of 225,428 tree rings were dated

<sup>1</sup>www.rinntech.com

and measured. Tree-ring series were visually and statistically crossdated and compared with each other by calculating the t-value after Baillie and Pilcher (1973) using TSAP-Win. The quality of crossdating was verified using the dendrochronology program library in R (dplR; Bunn, 2008). A MR was determined for each crossdated ring-width series by scoring every year as either being missing (value of 1) or present (value of 0). This process accommodated analysis of a ring missing from a series.

## Age Effect and Standardizing Procedures

The term "cambial age" refers to the number of years that the vascular cambium has produced annual rings prior to and including the ring being examined. For example, the first tree ring produced outside of the pith has a cambial age of 1; the tenth ring has an age of 10, and so on. Consequently, rings produced in a single calendar year by trees of different ages will differ in cambial age. The effect of cambial age on the occurrence of MR was estimated by binomial logistic regression modeling of the complete collection of MR series. To construct this model, the dependent variable was the series of all individual MR values (0 or 1) and the independent variable was their cambial age. The resulting regression line provided the estimated cambial age-related trend in MR frequency. Then, each MR series was standardized by calculating the ratio between the observed MR value and the value predicted by the trend model at that cambial age. The mean MR series for each site was calculated from the individual standardized MR values.

#### Climate Conditions Promoting MR Occurrence

Climatic conditions associated to each individual tree ring formed through the 1901–2014 period were obtained from CRU 3.22 dataset. For each individual tree ring, 16 pairs of monthly climate variables were calculated (sum of precipitation and mean temperatures from the previous September to current December). In addition, for each monthly climate variable, the mean of each preceding 2-months, 3-months and so on up to 60 months window were calculated to test the persistence of climatic effects on MR frequency over longer periods of time. Then, a total of 960 different values of precipitation and 960 of temperature (16 months\*60 months period) were calculated and associated to each individual tree ring.

Logistic regression models (LRM) were used to determine the relationship between MR frequency (dependent variable) and total precipitation, mean temperature, and their interaction (independent variables). A set of LRM was constructed to identify the combination of climatic factors which explain better the observed frequency of MR across the network. Globally a total of 921,600 LRM were constructed to consider all paired combinations of different variables of precipitation and temperature. Goodness-of-fit were compared by the coefficient of determination (r 2 ) of the various models.

For each calculation of LRM hierarchical cluster analysis was used to identify tree rings formed under similar climate

conditions. Thus, all individual tree rings were classified into 200 classes or clusters of similar climate conditions applied to the rank of the selected precipitation and temperature values using the K-means clustering algorithm as described by Hartigan and Wong (1979). Then, the mean MR frequency and the means of precipitation and temperature were calculated for each cluster. Using this procedure, a robust estimation of the frequency of MR (based on more than 1,000 tree rings for each cluster) was calculated for different ranges of climate conditions. Finally, LRM was calculated using, the mean MR frequencies calculated for each cluster as the dependent variable and the mean temperature, the mean precipitation and their interaction as independent variables.

#### Estimation of MR Frequency across the Range Distribution of Pinus halepensis

To estimate the predicted frequencies of MR across the distribution range of P. halepensis, the previously constructed LMR based with the highest explained variance (r 2 ) was selected. The selected model was then applied annually from 1902 to 2013 to gridded climatic data across the species distribution area to obtain annually predicted frequencies of MR. Finally, the averages of annual maps were used as a global predicted frequency of MR during the instrumental period (1902– 2013).

#### RESULTS

#### Dendrochronological Network and MR

The extensive dendrochronological network consisted of 113 different sites (Supplementary Table S1) where in total 1,509 trees were selected, ranging from at least form 5 to 35 trees per site. Altogether 2,953 cores were sampled, ranging from at least 6 to 70 samples per site. In total 225,428 tree rings were counted and measured, ranging from at least 330 to 5,363 per site. From the total number of tree rings, 4,150 MR were identified, yielding a global percentage of 1.84. The proportion of MR at sites within the network ranged from 0 to 11.89%.

Cambial age of tree-ring series ranged from 1 to 301 years, with an overall mean of 44 years. Site mean cambial age ranged from 10 to 98 years. Further analysis was conducted on rings with cambial ages from 1 to 169 years to maintain adequate replication. Older rings were represented by fewer than 50 tree ring series and occurred at fewer than 25 of the study sites.

#### Age Effect on MR Frequency

The frequency of MR significantly increased with increasing cambial age (**Figure 2**). In tree rings younger than 15 years the proportion of MR was 0.05% with progressive increases until reaching 6.5% at the age of 169 years.

## Observed Frequencies of MR across Dendrochronological Network

For each sampling site the average standardized observed MR frequency was calculated (**Figure 3**). The value of 0 indicates the sites with no MR; the value of 1 indicates the sites where the observed frequency of MR is equal to the global average of MR frequency of the whole study area. Values of 2, 3, 4, 5, 6, and 7 indicate the sites where the average standardized observed MR frequency is 2, 3, 4, 5, 6 or 7 times higher than the average.

Lower frequencies of MR were found in the northern and eastern portions of the Pinus halepensis distribution range in the Mediterranean basin. Higher observed frequencies of MR were located in Spain and northern Africa, with the highest values in south-eastern Spain.

Missing ring frequencies as shown in **Figure 3** are partially comparable among the study sites, because the differences in the frequencies of MR among the populations due to different age structure of populations have been removed. However, a direct comparison may be still biased since obtained frequencies of MR are calculated for different periods at each study site, depending on the length of the available local dataset.

## Climate Conditions Promoting the Occurrence of MR

Goodness-of-fit (r 2 ) of the 921,600 constructed LRM varies from 0.12 to 0.927 in dependence of selected combination of precipitation and temperature conditions at different time scales.

The highest explained variance of MR occurrence across dendrochronological network (r <sup>2</sup> = 0.927) is obtained when model includes the sum of precipitation from previous September to current May, the mean temperature from previous October to current February and their interaction as independent variables.

The relationship of MR frequency to climate was non-linear and exponentially increased as the summed total precipitation decreased from the previous September to the current May (**Figure 4A**) and with elevated mean temperature from the previous October to the current February (**Figure 4B**). However, MR was more strongly dependent on the interaction of temperature and precipitation than on either climatic factor taken individually. The frequency of MR for each cluster (**Figure 4C**), the agreement between observed and predicted frequencies of MR (**Figure 4D**), and the predicted frequencies of MR (**Figure 4E**) are shown.

## Estimated Occurrence of MR across the Distribution Range of Pinus halepensis

Logistic regression models constructed using summed precipitation from the previous September to current May, mean temperature from the previous October to current February, and their interaction were robust predictors of MR

frequency over the wide range of climate conditions and beyond those used to construct the models themselves. Consequently, the model is robust in application for the complete range of climatic conditions across the distribution range of P. halepensis (**Figure 5**). The presence of MR is unlikely (frequency lower than 0.05) to occur when total summed precipitation from previous September to current May exceeds 50 mm and when mean temperatures from previous October to current February are lower than 5◦C. However, the frequency of MR exponentially increases with concurrent low total precipitation and high temperatures. As a consequence, MR frequency higher that 0.2 can be expected when total summed precipitation are lower than 30 mm coincident with warm temperatures (monthly mean greater than 12◦C). MR frequency can be expected to occur in 50% of total tree rings (frequency = 0.5) when precipitation drops below 20 mm and temperatures are greater than 13◦C.

Similarly, to have a general view on MR frequency across the species distribution, the selected LRM model was applied annually from the year 1902 until the year 2013 to the gridded climatic data across the distribution area of P. halepensis. Predicted frequencies of MR were calculated annually across the species distribution (see complete collection of yearly maps in the Supplementary Material, Supplementary Figure S1). The frequency varied across the distribution area and for particular years, being even higher than 75% in a particular year and site. The highest predicted frequency of MR occurred for south-eastern Spain and northern Algeria. The years of greatest frequency of MR across the species distribution area are also identified (for example 2012, 2005, 1996, 1995, and 1982).

Yearly calculations of predicted frequencies of MR were averaged for each climatic grid point to represent an unbiased predicted frequency of MR across the distribution range of P. halepensis during the instrumental period 1902– 2013. The frequency of MR varied across the distribution area with the northern and eastern portion of the range distribution with a predicted frequency of less than 0.01%. The predicted frequency of MR was 15% or higher in the western part of the distribution range in south-eastern Spain and northern Algeria (**Figure 6**). The high predicted values were validated by the high observed frequencies of MR from these locations.

#### DISCUSSION

#### General Considerations

Dendrochronological dating of tree rings in an annual series is possible due to distinguishable growth increments that can be

cross-dated with other series in the sampled population. The crossdated chronologies from this dendrochronological network should provide a reference for future dendrochronological research in the region. By definition, MR cannot be seen and therefore they can confound accurate assignation of calendar dates to rings (e.g., Grundmann et al., 2008). Although not directly countable, the position of MR can be identified through careful crossdating (e.g., Novak et al., 2011). Correctly dated MR can identify the climatic conditions that lead to dormancy of the vascular cambium and the lack of production of an annual ring (Novak et al., 2016). Therefore, MR can be markers for the timing of ECE.

#### The Effect of Age on MR

For dendroclimatological analysis of tree-ring width and climate, the age-related trend of MR occurrence needed to be identified and removed prior to determination of the effect of precipitation and temperature on MR. For ring-width series, the most common approach to remove the age- and size-related variation (growth trend) is to fit a curve to the measured ring-width series and then to either subtract or divide the observed measurement from the fitted curve (Cook et al., 1995).

Rather than a quantitative characteristic such as ring width, MR series are binomial since MR are either present or absent for a particular year. Different methods for detrending binomial series in tree rings have been tested so far (e.g., Campelo et al., 2007, 2015; Novak et al., 2013a). Here, we used a recently developed approach which has been applied to intra-annual density fluctuations (Zalloni et al., 2016), and can be used for other binary anatomical characteristics of tree rings.

As trees within the dendroclimatic network increased in age, tree-ring widths tended to narrow and the frequency of MR increased. Ring width can also be related to the length of winter cambial dormancy (e.g., Barnett, 1971) which is largely under genetic control. The genetic potential may become limited by harsh climatic conditions. Ring widths also tend to narrow or be missing in declining trees as in case of silver fir (e.g., Torelli et al., 1986; Bigler et al., 2004). Consequently, the occurrence of MR may indicate tree or forest decline, but rather than being merely a marker of decline, MR formation may also be part of a tree survival strategy. Under extreme conditions, the cambium may produce phloem while producing no xylem (e.g., Gricar et al., 2016 ˇ ; Novak et al., 2016). Annual production of phloem is essential to maintain the distribution pathways of photosynthate and other biomolecules. Although xylem is essential for tree function, the allometric diversion of resources

color: frequency below 0.01% (0.0001); dark red color: frequency higher than 15% (0.15).

to phloem production likely has survival value under extreme limitations of growth. Therefore, we should consider MR in P. halepensis as part of the biological plasticity of the species to adapt to adverse environmental conditions.

#### Climate Signals Derived from MR

The relationship between MR and climate is difficult to explore, because the replication of MR across the network studied is low and therefore it is not easy to get robust estimation. The principle of replication represents one of the keys in dendrochronology highlighting the need to use more than one stem radius per tree and more than one tree per site to obtain reliable treering chronologies. Different statistics used to analyze tree-ring series are often based on high number of samples, and normally sampling strategies in dendrochronology are often designed to ensure the requested number of samples.

The next question is how to deal with binomial character of MR, because as a special anatomical feature it cannot be measured but just characterized, based on its presence or absence in a specific year. In this sense, the criteria to define appropriate number of samples to obtain reliable representativeness of MR frequency cannot be based on the same procedure used for treering chronologies. Future investigation and advices how to work with MR are necessary, and a well-designed sampling strategy, would be the best solution.

In this study, to determine the appropriate number of samples to work with MR a novel methodology of global analysis is proposed combining all the data available. It is based on the identifications of analogous climatic conditions which permit joining the tree rings to bigger clusters/classes for which we can get robust estimation of the MR frequency. This method has been applied to intra-annual density fluctuations already (Zalloni et al., 2016).

The disadvantage of our method is that the full value accrues only from dendrochronological networks that span the complete distribution range of the species under investigation. Otherwise the applicability of climatic drivers for MR across the range remains unknown or uncertain.

The advantage of our analysis is the global application to the distribution range of P. halepensis. In our case, the global model explained 91.7% of the variability of MR, which stresses that regional rather than local climatic factors are responsible for the occurrence of MR. Periods of extreme drought and warmth were the key factors associated with the occurrence of MR. This finding is supported by previous research on climatic drivers including low moisture availability (Liang et al., 2014) or summer drought (Jonsson et al., 2002; Dulamsuren et al., 2013). The climatic conditions before tree-ring formation can also constrain tree growth (Kuptz et al., 2011; Wettstein et al., 2011; Camarero et al., 2013). In contrast with P. halepensis in the Mediterranean Basin, MR in alpine environments were associated with frost events (e.g., Hantemirov et al., 2004; Panayotov et al., 2013).

The climate conditions that triggered the occurrence of MR were mean temperatures higher than 10◦C from the previous October through the current February and total accumulated precipitation of less than 50 mm from previous September till current May. Therefore, MR occurred as a consequence of drought before the onset of cambial division and growth. Nine month effect of low precipitation indicates that persistent drought triggers MR. The same occur also with high temperatures but on even shorter temporal scale during 5 months before the beginning of growth. High temperatures and prolonged drought may exhaust the energy reserves necessary for growth.

Importantly, the combination of high temperatures and low precipitation was identified as the trigger for MR. The identification of this combined role was shown for P. halepensis by de Luis et al. (2013) and for other tree species (Esper et al., 2007b; Rammig et al., 2015; Seim et al., 2015).

Normally, the climate thresholds are not being explored in dendrochronology. They are, however, essential to evaluate the importance of extremes, (i.e., ECE) on tree growth, and to define the range of climatic conditions associated to the occurrence of MR. The thresholds that trigger MR, therefore may be used to define ECE for the Mediterranean Basin.

Under favorable growing conditions, the vascular cambium of P. halepensis may be active throughout the entire year. For most growing seasons, cell division ceases during the winter due to low temperatures and/or dry conditions, resuming growth with increased availability of moisture (Liphschitz et al., 1984; Liphschitz and Lev-Yadun, 1986; de Luis et al., 2011a). Under extreme limiting conditions, the vascular cambium does not produce new cells, and a MR occurs in the treering series (Novak et al., 2016). As we know so far, this is the first time that MR (which can be interpreted as a consequence of extreme growth events) are related to ECE. We demonstrated that the relationship of MR to climate was not linear. We also presented LRM as a method for analysis.

Moreover, MR seem to be promising tree-ring features linked to the occurrence of extreme events of climate, but also to adverse growth conditions, as the competition for light in different species (e.g, Lorimer et al., 1999; Parent et al., 2002). In some cases more than 50 tree rings can be omitted during suppressed a period, which reduces usefulness of these samples-species for dendrochronological studies. MR due to competition for light are mainly linked to the young phase of shade tolerant species, while in the case of Pinus halepensis they are not in sequences but erratically driven by climatic and so increase with tree age. For this reason an analysis of synchronization of MR will be very informative because it can be used to separate ECE form other extreme environmental events, like volcanic eruptions (D'Arrigo et al., 2013) or severe frost or drought from suppression or dieback phenomena. In the latter case MR are very close one to the other and in sequences within small tree-rings.

#### Distribution of MR

Spatial and temporal distribution of MR frequencies showed differences across the distribution area of P. halepensis in the Mediterranean. MR were more frequent in the western portion of the distribution range including south-eastern Spain and northern Algeria, and less frequent in the northern and eastern portion of the range such as northern

Spain, France, central Italy, Slovenia, and along the coast of Greece and Croatia.

Current predictions of climate change (IPCC, 2013) include a greater increase of winter than summer temperatures and a general decrease of precipitation, both of which are likely to increase the occurrence of MR. MR are relatively frequent in south-eastern Spain but absent from the northern part of distribution of the species. Our findings are supported by St. George et al. (2013) who found that for trees in the Northern Hemisphere (which also includes Mediterranean), MR are most common in trees at sites where growth is limited by moisture availability. As climate continues to change, the relationship of MR to climate may also change. The open question remains as to whether the increasing occurrence of MR in recent years should be attributed to natural climate variability or to global climate change (Allen et al., 2010).

#### CONCLUSION

The presence of MR is significantly related to the tree-ring age. The result showed an increase of MR frequency with increasing cambial age.

Across the distribution range of P. halepensis, MR formation was triggered by the combination of high mean temperatures from previous October till current February and scarce accumulated precipitation from previous September till current May. This is a global conclusion and can be applied to every site across the distribution range.

This research is the first to identify spatial and temporal variation of the frequency of MR in an extensive dendrochronological network. Our method allows extending the results to the entire range of P. halepensis in the Mediterranean Basin. Identification of the frequency and position of MR should facilitate future crossdating and construction of tree-ring chronologies. The occurrence of MR can identify ECE of the past and support predictions of the effects of climate change for the future.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

Conception and design of the study were performed by KN and MD; acquisition of data was realized by KN, MD, MS, LL, RS-N, KC, JG, AD, GP, CR, and AP; analysis of data was performed ˇ by KN, MD, JR, KS, CR; interpretation of data was realized by KN, MD, KC, JG, AD, GP, CR, and KS; drafting and writing the ˇ work was performed by KN, MD, MS, LL, RS-N, and KC; critical ˇ revision of work was performed by KN, MD, JR, KC, JG, AD, GP, ˇ CR, AP, and KS. All the authors discussed and commented on the manuscript, gave final approval to be published, and agreed on the integrity of the work.

## FUNDING

This study was supported by Spanish Ministry of Education and Science co-funded by FEDER program (projects: CGL2012- 31668 and CGL2015-69985-R), by the Slovenian Research Agency (programs P4-0015 and P4-0107), by the LLP ERASMUS bilateral agreement between the University of Ljubljana and the University of Alicante, and the USDA Forest Service. The article is based upon work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology).

#### ACKNOWLEDGMENTS

The English of this manuscript was revised by the co-author, English native speaker, KS. This article forms part of doctoral dissertation of KN.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.00727

gradient in the southernmost Pinus nigra relict forests. Ann. For. Sci. 70, 769–780. doi: 10.1007/s13595-013-0321-9



Fritts, H. C. (2001). Tree Rings and Climate. Caldwell, NJ: The Blackburn Press.



Zalloni, E., de Luis, M., Campelo, F., Novak, K., De Micco, V., Di Filippo, A., et al. (2016). Climatic signal from intra-annual density fluctuation frequency in Mediterranean pines at a regional scale. Front. Plant Sci. 7:579. doi: 10.3389/fpls.2016. 00579

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Novak, de Luis, Saz, Longares, Serrano-Notivoli, Raventós, Cufar, ˇ Griˇcar, Di Filippo, Piovesan, Rathgeber, Papadopoulos and Smith. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Flood-Ring Formation and Root Development in Response to Experimental Flooding of Young Quercus robur Trees

Paul Copini1,2 \*, Jan den Ouden<sup>1</sup> , Elisabeth M. R. Robert3,4, Jacques C. Tardif<sup>5</sup> , Walter A. Loesberg<sup>1</sup> , Leo Goudzwaard<sup>1</sup> and Ute Sass-Klaassen<sup>1</sup>

<sup>1</sup> Forest Ecology and Forest Management Group, Wageningen University and Research Centre, Wageningen, Netherlands, <sup>2</sup> Alterra, Wageningen University and Research Centre, Wageningen, Netherlands, <sup>3</sup> Laboratory of Wood Biology and Xylarium, Royal Museum for Central Africa, Tervuren, Belgium, <sup>4</sup> Laboratory of Plant Biology and Nature Management, Vrije Universiteit Brussel, Brussels, Belgium, <sup>5</sup> Centre for Forest Interdisciplinary Research, Department of Biology, The University of Winnipeg, Winnipeg, Canada

#### Edited by:

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### Reviewed by:

Scott St. George, University of Minnesota, USA Marina V. Bryukhanova, VN Sukachev, Institute of Forest, Siberian Branch of the Russian Academy of Sciences, Russia

> \*Correspondence: Paul Copini paul.copini@wur.nl

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 11 March 2016 Accepted: 17 May 2016 Published: 14 June 2016

#### Citation:

Copini P, den Ouden J, Robert EMR, Tardif JC, Loesberg WA, Goudzwaard L and Sass-Klaassen U (2016) Flood-Ring Formation and Root Development in Response to Experimental Flooding of Young Quercus robur Trees. Front. Plant Sci. 7:775. doi: 10.3389/fpls.2016.00775 Spring flooding in riparian forests can cause significant reductions in earlywood-vessel size in submerged stem parts of ring-porous tree species, leading to the presence of 'flood rings' that can be used as a proxy to reconstruct past flooding events, potentially over millennia. The mechanism of flood-ring formation and the relation with timing and duration of flooding are still to be elucidated. In this study, we experimentally flooded 4-year-old Quercus robur trees at three spring phenophases (late bud dormancy, budswell, and internode expansion) and over different flooding durations (2, 4, and 6 weeks) to a stem height of 50 cm. The effect of flooding on root and vessel development was assessed immediately after the flooding treatment and at the end of the growing season. Ring width and earlywood-vessel size and density were measured at 25- and 75-cm stem height and collapsed vessels were recorded. Stem flooding inhibited earlywood-vessel development in flooded stem parts. In addition, flooding upon budswell and internode expansion led to collapsed earlywood vessels below the water level. At the end of the growing season, mean earlywoodvessel size in the flooded stem parts (upon budswell and internode expansion) was always reduced by approximately 50% compared to non-flooded stem parts and 55% compared to control trees. This reduction was already present 2 weeks after flooding and occurred independent of flooding duration. Stem and root flooding were associated with significant root dieback after 4 and 6 weeks and mean radial growth was always reduced with increasing flooding duration. By comparing stem and root flooding, we conclude that flood rings only occur after stem flooding. As earlywoodvessel development was hampered during flooding, a considerable number of narrow earlywood vessels present later in the season, must have been formed after the actual flooding events. Our study indicates that root dieback, together with strongly reduced hydraulic conductivity due to anomalously narrow earlywood vessels in flooded stem parts, contribute to reduced radial growth after flooding events. Our findings support the value of flood rings to reconstruct spring flooding events that occurred prior to instrumental flood records.

Keywords: flooding, hypoxia, leaf phenology, pedunculate oak, Quercus robur, vessel development, root development

## INTRODUCTION

fpls-07-00775 June 10, 2016 Time: 12:31 # 2

Trees growing in riparian forests must cope with regular flooding events and may survive the anoxic conditions associated with flooding (Kozlowski, 1984; Siebel et al., 1998; Glenz et al., 2006). While flooding during dormancy may not affect trees, flooding during the growing season can severely affect development and growth (Gill, 1970; Kozlowski, 1984; Glenz et al., 2006). Species of oak (Quercus) and ash (Fraxinus) trees frequently occur along river systems in Europe (Q. robur L., F. excelsior L.), the United States of America, and Canada (e.g., Q. macrocarpa Michx., Q. lyrata Walter., F. nigra March., F. pennsylvanica March.). These species are ring porous and form large earlywood vessels in spring, followed by small latewood vessels later on in the growing season and have shown to be able to cope with 50 days of flooding as juveniles or even 100 days as adult trees (Siebel et al., 1998; Kreuzwieser et al., 2004; Glenz et al., 2006). In years with spring flooding events, these trees may alter their wood anatomy and frequently form tree rings with anomalously narrow earlywood vessels – such rings are known as 'flood rings' (Astrade and Bégin, 1997; St. George et al., 2002; Tardif et al., 2010; Ballesteros-Cánovas et al., 2015; Therrell and Bialecki, 2015; Bräuning et al., 2016; Kames et al., 2016). These earlywood vessels may sometimes be accompanied by sickle-shaped, collapsed earlywood vessels (Land, 2014). When flooding occurs during summer, exceptionally large latewood vessels may occur (Yanosky, 1983; Yanosky and Cleaveland, 1998; Land, 2014). As flood rings are not only found in living trees but are also preserved in old timber and in subfossil trees, they can be used as a proxy to reconstruct flooding events with an annual or even intra-annual accuracy over potentially millennia and may shed light on the forcing factors between climate, human impact, and flooding events (Yanosky, 1983; Wertz et al., 2013; Land, 2014; Ballesteros-Cánovas et al., 2015; Kames et al., 2016). However, the application of flood rings as proxy for flooding events is hampered by our limited understanding of their formation, in the absence of experimental evidence (St. George, 2010).

The formation of flood rings is, inter alia, depending on the time window during which developing xylem cells are able to register the flooding signal (Fonti et al., 2010; Sass-Klaassen et al., 2011). Flooding events during winter dormancy are most likely not recorded whereas during the period of radial growth the flooding signal can be directly recorded in the earlywood (St. George and Nielsen, 2002; Wertz et al., 2013) or latewood (Yanosky, 1983; Land, 2014). In ring-porous species radial growth may either start during late bud dormancy or after budswell, while earlywood formation normally ends after the leaves are fully expanded (Zasada and Zahner, 1969; Bréda and Granier, 1996; Sass-Klaassen et al., 2011; Takahashi et al., 2013) and fine roots have developed (Ponti et al., 2004). Radial growth cessation is highly variable among trees and from year-to-year, and may end before leaf abscission (Michelot et al., 2012) or immediately after earlywood formation in spring (Land, 2014). Besides timing, the duration of a flooding event is also of importance (Astrade and Bégin, 1997; St. George and Nielsen, 2002; St. George, 2010). So far, it is known from flooding experiments that 6 weeks of flooding during leaf development can induce the formation of a flood ring in adult pedunculate oak (Quercus robur L.; Stuijfzand et al., 2008). Field studies also showed that flooding events of more than 10 days may induce flood rings in Q. lyrata and Q. macrocarpa (Therrell and Bialecki, 2015). Flooding height is less important, as 20 cm of flooding already induced flood rings in the submerged stem parts of pedunculate oak (Stuijfzand et al., 2008).

The physiology of flood-ring formation is poorly understood. During flooding, hypoxic conditions occur as gas diffusion rates are reduced by ∼10−<sup>4</sup> in water compared to air (Cannon, 1925; Kozlowski, 1984). During the growing season, this may inhibit root growth and cause decay and dieback of roots, especially in non-woody fine roots (Coutts, 1982; Yamamoto and Kozlowski, 1987). The reduction of root biomass negatively influences root/leaf ratio and might be the key factor to explain reduced growth of flooded trees (Schmull and Thomas, 2000). Furthermore, reduced growth in flooded trees might occur as trees shift from aerobic respiration to anaerobic respiration which is much less efficient (Hook, 1984). Increased levels of hormones like ethylene and auxin in flooded stem parts have been related to morphological adjustments to cope with the effects of flooding, i.e., the enlargement of lenticels, formation of aerenchyma tissue and adventitious roots which deal with gas exchange and water uptake (Gomes and Kozlowski, 1980; Yamamoto et al., 1995; Parelle et al., 2006). These hormones could also be related to the formation of flood rings, as increased concentrations are associated with decreases in cross-sectional vessel areas and increases in vessel densities (Tuominen et al., 1995; Junghans et al., 2004; Aloni, 2013).

In this study, we experimentally investigated the formation of flood rings in pedunculate oak in relation to spring leafphenology and flooding durations lasting for 2, 4, and 6 weeks. We hypothesized that (i) ring-porous species that need their current year's earlywood vessels for axial water transport (Cochard and Tyree, 1990; Tyree and Zimmermann, 2002; Copini et al., 2015), start earlywood-vessel development both in flooded and non-flooded stem parts; (ii) spring flooding leads to anomalously narrow earlywood vessels in flooded stem parts when the timing of flooding coincides with earlywood-vessel development within 6 weeks of flooding; (iii) based on the study of Land (2014), developing earlywood vessels collapse in response to flooding; and (iv) flooding leads to reduced radial growth and root dieback when flooding occurs during the growing season.

#### MATERIALS AND METHODS

#### Plant Material

We used 200 four-year-old potted pedunculate oak trees (Quercus robur L.) with a stem height of approximately 180 cm, that were randomly selected out of 600 available trees. All 600 trees were obtained from a tree nursery in March 2009, 1 year before the experiment was conducted, and immediately potted in 17 l containers (diameter 30 cm, height 24 cm) in a sand-loam mixture. The trees were then placed in a 1 m × 1 m grid in an experimental garden in Wageningen, the Netherlands (51.9884◦N, 5.6644◦E). The trees were frequently watered using a semi-automatic fertigation system.

#### Experimental Set-up

fpls-07-00775 June 10, 2016 Time: 12:31 # 3

The flooding experiment was conducted at the Sinderhoeve Research Station, Wageningen University, The Netherlands (51.9983◦N, 5.7523◦E) between March and July 2010. To simulate flooding, we used 1.4 m × 1.2 m × 0.7 m basins (length, width, depth) containing pumps for water re-circulation and maintaining a water level to flood trees to a stem height of 50 cm (**Figure 1**). The water came from a rain fed basin. We installed six two-channel HOBO Pro temperature data loggers (Onset Corporation, Bourne, MA, USA) to record water temperature 25 cm below the water surface (i.e., at 25-cm stem height) of every basin and air temperature 25 cm above the water surface, corresponding to 75-cm stem height of the flooded trees. During the experimental period, the mean water temperature of 15.2 ± 4.6◦C (mean ± SD) was generally higher than the mean air temperature (13.1 ± 6.6◦C) while the daily temperature amplitude (maximum temperature – minimum temperature) in water was much lower compared to the air temperature (Supplementary Figure S1). Below water, dissolved oxygen concentrations (mg/L) were recorded once a week using a portable WTW Oxi 330 meter, equipped with a CellOx 325 electrode. Oxygen concentrations were on average 10.6 ± 3.9 mg/l (mean ± SD). Differences in oxygen concentrations occurred both in time and between different basins (Supplementary Figure S1).

Flooding treatments started at three successive leaf phenophases (**Figure 1**) taking into account the leaf status of individual trees: late dormant trees were flooded on March 19th, trees with swelling buds were flooded between April 23rd and May 5th, and trees with expanding internodes were flooded between May 7th and May 19th, 2010, approximately 2 weeks after budswell (**Figure 1**). At each phenophase, 10 trees were flooded for either 2, 4, or 6 weeks by randomly placing them in one of the 10 basins (90 stem-flooded trees). For the budswell group, an additional treatment was added in which the roots of 10 trees were flooded for either 2, 4, or 6 weeks by leveling the water to the soil surface (30 root flooded trees; **Figure 1**). Control trees, corresponding to all flooding durations were placed next to the basins (**Figure 1**).

To assess the status of earlywood-vessel and root formation of the trees entering the experiment, five trees per phenophase were harvested (**Figure 1**). All control trees were watered twice a week and remained – like all flooded trees – exposed to ambient weather conditions. To study the dynamics of floodring development, we harvested half of the trees immediately after the flooding treatment (five treated trees, five control trees); the remainder was returned to the experimental garden and was harvested after the end of the growing season in November 2010. Stem sections were sampled at 25 cm (i.e., 25 cm below water level in the flooded trees) and 75-cm (i.e., 25 cm above water level) stem height. Samples were stored in 50% ethanol solution at room temperature prior to further processing.

## Wood Sample Preparation and Measurements

For both stem samples collected at flooding cessation and at the end of the growing season, transverse wood sections were cut with a thickness of approximately 20 µm using a G.S.L.- 1 sliding microtome (Gärtner et al., 2014) and stained with a safranin/astra blue solution for 5 min. Following dehydration in graded series of ethanol (50–95–100%), samples were rinsed with xylol, mounted in Canada balsam and dried at 60◦C for 15 h. Pictures were made with a digital camera (DFC 320, Leica, Cambridge, UK) mounted on a microscope (DM 2500, Leica, Cambridge, UK) using Leica imaging software (version 3.6.0).

For all trees harvested, we measured earlywood-vessel size (in µm<sup>2</sup> ) over a tangential width of approximately 1 cm – which equals approximately 20% of the circumference of the 2010 tree ring – using ImageJ software (ver. 1.44<sup>1</sup> ; developed by W. Rasband, National Institutes of Health, Bethesda, MD, USA). In order to get an estimate of vessel development in spring, we distinguished unlignified vessels (completely blue) from lignified vessels (partly or completely red). We then determined mean earlywood-vessel area, and calculated the maximum earlywoodvessel size as the mean of the 20 largest earlywood vessels. In addition, earlywood-vessel density (vessels/mm) was calculated by dividing the number of first-row earlywood vessels formed in 2010 over the tangential width (vessels/mm). The 2010 ring width was measured at two radii at both stem heights in the trees that were harvested in November. Collapsed vessels were visually detected per stem height and rings containing three or more collapsed vessels per stem height were recorded as tree rings containing collapsed vessels.

#### Root Development

We recorded root phenophases of all trees that were harvested immediately after the flooding experiments. To do so, we removed the basal stem part with the roots attached from the container and rinsed it with water to remove the soil. Based on studies by Hendrick and Pregitzer (1992) and Ponti et al. (2004), we defined three root phenophases: (i) dormant roots (ii) white roots elongating, and (iii) white roots maturing, forming many small lateral roots. We recorded whether root formation was affected by root dieback as visible by black discolorations or further decay. In addition, the presence of lenticels and adventitious roots was noted.

#### Statistical Analysis

All statistical analyses were performed in the statistical software package SPSS, version 19 (SPSS Inc., Chicago, IL, USA) applying a significance level of 0.05.

#### Earlywood-Vessel Development

We used the trees that were harvested in spring upon the termination of the flooding treatments, to test for differences between earlywood-vessel development below and above water level as compared to control trees. First, we tested for a possible

<sup>1</sup>http://rsb.info.nih.gov/ij

FIGURE 1 | Experimental set-up of the flooding experiment. (A) Experimental set-up showing the concrete basins with experimentally flooded trees on the left and corresponding control trees on the right. (B) At three leaf phenophases, trees were flooded to a stem height of 50 cm for 2, 4, or 6 weeks (n = 10); control trees were placed next to treated trees. Upon budswell, an additional group of trees had only their roots flooded during 2, 4, or 6 weeks (n = 10). After the flooding period in spring, five trees per treatment were harvested immediately while the remaining trees were returned to the experimental garden to be harvested after the end of the growing season, in autumn. (C) Cambium and root phenophases in relation to the three leaf phenophases upon the onset of flooding. When the buds were dormant on March 19th, 2010, the cambium and roots were dormant as well with no earlywood vessels or white roots present. Upon budswell, between April 23rd and May 5th, earlywood-vessel development had started irregularly around the circumference. Normally, earlywood vessels were unlignified while in some cases lignification had started (arrow). In all trees, newly formed elongating white roots were present (arrow). When the internodes started expanding 14 days after budswell, between May 7th and May 19th, earlywood-vessel development had started in all trees and both lignified (arrow) and unlignified vessels were present and many new elongating white roots had formed.

difference in vessel density at 25-cm and 75-cm stem height between all flooded and control (grouping trees with different flooding durations) using a Mann–Whitney U test (n = 15). Subsequently, we performed three Mann–Whitney U tests, to assess whether significant differences occurred after 2, 4, or 6 weeks of flooding, compared to control trees (n = 5).

#### Earlywood-Vessel Area

fpls-07-00775 June 10, 2016 Time: 12:31 # 5

The trees that were harvested at the end of the growing season were used to test whether spring flooding leads to anomalously narrow earlywood vessels when flooding coincides with earlywood vessel development, i.e., during budswell and to a lesser degree when the internodes start growing within 6 weeks of flooding. First, we used mixed factorial analyses of variance (ANOVA) to test whether mean and maximum earlywoodvessel area and vessel density within each treatment (control, flooded during late bud dormancy, budswell and upon internode expansion) was significantly different in relation to stem height (within subject factor) and flooding duration (between subject factor; n = 5). Second, to test for differences between flooding treatments and their control trees, we used mixed factorial ANOVAs with mean and maximum vessel size and vessel density at 25-cm (below water) and at 75-cm stem height (above water) as within-subject factors and flooding durations and treatment as between-subject factors (n = 5).

#### Vessel Collapse

We tested whether the occurrence of collapsed earlywood vessels was significantly different at 25-cm stem height between flooded and control trees by using the trees that were harvested after flooding and after the end of the growing season. First we used Fisher's Exact tests to evaluate whether there was an significant effect between all flooded and control trees per phenophase (n = 15). Subsequently, we performed three separate Fisher's Exact tests per phenophase to determine whether significant differences between flooded and control trees occurred after 2, 4, or 6 weeks (n = 5).

#### Radial Growth

To test whether flooded trees have significantly smaller ring widths in flooded stem parts, or along the whole stem because of the flooding treatments, we used ring widths of the trees that were harvested at the end of the growing season and used mixed factorial ANOVAs to test whether the ring widths at both 25 and 75 cm (within subject factor) was affected by the flooding treatment and flooding durations (between subject factors). Post hoctests with Bonferroni corrections were conducted to assess the impact of flooding durations (2, 4, or 6 weeks) on radial growth.

#### Root Dieback

For roots, it was tested whether dieback occurred in trees that were flooded (stem or root flooded) using the trees that were harvested immediately after the end of the flooding treatments. First we used Fisher's Exact tests to assess whether significant differences occurred between flooded and control trees per phenophase (n = 15). Subsequently, we performed three separate Fisher's Exact tests per phenophase to determine whether significant differences between flooded and control trees occurred after 2, 4, or 6 weeks (n = 5).

## RESULTS

## Earlywood-Vessel Development during Flooding

#### Late Bud Dormancy

At March 19th, when the first flooding treatment started, none of the trees had formed earlywood vessels (**Figure 1A**). Also, after 2 and 4 weeks of flooding none of the flooded and control trees had started earlywood vessel development (Supplementary Figure S2). The first earlywood vessels were present 6 weeks after the start of the flooding treatment in three flooded trees and in three control trees of which the buds were broken. In contrast to the control trees, that started earlywood-vessel development both at 25-cm and at 75-cm stem height, the flooded trees had started earlywood-vessel development only above the water level at 75-cm stem height (Supplementary Figure S2).

#### Budswell

Upon the second phenophase, between April 23rd and May 5th, 2010, earlywood-vessel development had started in all but one trees (**Figure 1C**). Earlywood vessels were initiated irregularly around the stem circumference, mostly near latewood vessels which were bordering the tree-ring boundary. In one tree, some earlywood vessels were lignified while in the others vessels were still unlignified. The flooding treatments induced significant differences in vessel densities between 25- and 75-cm stem height compared to the control trees (Mann–Whitney U test, U = 24, P < 0.001, n = 15; **Table 1**, **Figures 2A–D**). Two weeks of

TABLE 1 | Mean earlywood-vessel density and earlywood-vessel density differences and standard deviations (SD) between 75- and 25-cm stem height measured directly after the flooding treatments.


The trees that were flooded for 2, 4, and 6 weeks were pooled per phenophase (n = 15). The effect of flooding on the difference of earlywood-vessel density was tested using Mann–Whitney U tests. U-values and their significance are given as: <sup>∗</sup>P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; NT, not tested.

FIGURE 2 | Effects of 6 weeks of flooding after budswell on vessel development and radial growth in pedunculate oak. The white scale bar represents 300 µm. The cambial zone of control (left) and flooded trees (right) at 25 cm (flooded) and 75 cm (non-flooded) immediately after the flooding treatment in Spring 2010 (A–D) or after the growing season in Autumn 2010 (E–H). (A,B) The cambial zone at 75-cm stem height. In both trees many lignified earlywood vessels are present. (C,D) The cambial zone at 25-cm stem height of the same trees as in (A,B). Whereas in the control tree vessel development started and many earlywood vessels are lignified in the flooded tree, 25 cm below water hardly any vessels have been formed and most are not yet lignified. (E,F) The cambial zone at 75-stem height (not submerged). In both the control and flooded tree the earlywood vessels are relatively large, while ring width in the flooded tree is strongly reduced. (G,H) The cambial zone at 25-cm stem height (submerged stem parts in flooding treatments). While the trees of the control group showed relatively large earlywood vessels and wide ring widths, the flooded trees formed on average 70% smaller vessel areas and 59% smaller tree rings.

flooding did not lead to differences in vessel densities, but after 4 or 6 weeks significant differences occurred between flooded and control trees (Mann–Whitney U test, U = 0, p = 0.008, n = 5 and U = 2, p = 0.032, n = 5, respectively). After 6 weeks of flooding the second row of earlywood vessels was already being completed above the water level (**Figure 2B**, while below water vessel development was hampered (**Figure 2D**). In the trees of which the roots were flooded upon budswell, no significant effects of the treatment or of flooding duration were observed.

#### Internode Expansion

During the last phenophase, between May 7th and May 19th, earlywood-vessel development had started in all trees (**Figure 1C**). While some trees had just formed narrow unlignified earlywood vessels, others showed few (in one tree many) lignified vessels (**Figure 1C**). After this late flooding treatment significantly larger differences in vessel densities occurred between 25- and 75-cm stem height compared to the control trees (Mann–Whitney U test, U = 24, P < 0.05, n = 15) as in the flooded stem parts earlywood vessel development was slightly lower (**Table 1**). No significant differences were found in relation to flooding duration. At the end of the 6-week flooding treatment, i.e., 8 weeks after budswell, all earlywood vessels of the control trees were lignified both at 25- and 75-cm stem height. In contrast, in flooded trees only the earlywood vessel above the water level were lignified whereas below the water level many vessels remained unlignified.

#### Earlywood Vessel Development during the Growing Season Late Bud Dormancy

Trees that were flooded upon late bud dormancy had formed slightly narrower mean earlywood vessels at 25-cm compared to 75-cm stem height by the end of the growing season (**Table 2**, **Figure 3A**). This difference was mainly caused by the trees of the 6-week flooding treatment of which two had started with leaf formation after 4 weeks of flooding – within the 6 week flooding treatment – and showed a 38 and 66% reduction in earlywood-vessel size in flooded stem parts at the end of the growing season. Maximum earlywood-vessel size and vessel

TABLE 2 | Effects of flooding per phenophase on mean earlywood-vessel area (Mean vessel area µm<sup>2</sup> ), mean of the 20 largest earlywood vessels (Maximum vessel area, µm<sup>2</sup> ), vessel density (#/mm), and ring width (µm), measured after the end of the growing season following the flooding experiments.


The mean and SD and the difference as a percentage are provided for each variable at 25- and 75-stem height. In addition, the table shows the results of mixed factorial ANOVAs with height as a within subject measure and flooding duration as a between-subject factor. F-values and their significance are provided as: <sup>∗</sup>P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001. The interactions of height and flooding duration were always insignificant (not shown).

density (C: #/mm) and ring width (D: µm) after the end of the growing season following the flooding experiments for the different treatments (n = 5) and control trees (n = 35). Gray bars show the mean values and standard errors of stem sections taken at 25-cm stem height; whereas white bars represent the mean and standard errors of the stem section at 75-cm stem height. The control trees were pooled as no significant differences occurred among them.

density were comparable at both heights (**Table 2**, **Figures 3B,C**, Supplementary Figure S3). Compared to control trees, mean earlywood-vessel area and maximum earlywood-vessel area were significantly reduced at both heights and vessel density was slightly higher, especially in the flooded stem part (**Table 3**, **Figure 3**). The significant interaction between duration and treatment was caused by the trees of which the buds started to develop during the flooding experiment. When the 6-week treatment – in which earlywood-vessel formation may have started – was omitted, there was no effect of the flooding treatments upon earlywood-vessel size or vessel density.

#### Budswell

One tree that was flooded upon budswell died after the 4 week flooding treatment and was excluded from analyses. The remaining trees contained anomalously narrow earlywood vessels in the flooded stem parts at the end of the growing season; mean earlywood-vessel area and maximum earlywood-vessel area were significantly reduced on average by 47 and 32%, respectively, compared to 75-cm stem height, independent of flooding duration (**Table 2**, **Figures 2E–H** and **3A,B**, Supplementary Figure S3). Earlywood-vessel density was comparable between 25- and 75-cm stem height (**Table 2**, **Figure 3C**). Compared to the control trees, the flooded trees contained significantly lower mean or maximum earlywoodvessel areas, and higher vessel densities (**Table 3**, **Figure 3C**). In addition, the highly significant interactions for mean and maximum earlywood-vessel size between height and treatment, show that earlywood-vessel size was significantly reduced in the flooded stem parts (**Table 3**, **Figures 3A,B**).

One tree of the root flooding treatment (2 weeks) died and was excluded from the analyses. The remaining trees did not show any effect of root flooding on mean or maximum earlywood-vessel size, earlywood-vessel density in relation to stem height or flooding duration (**Table 2**, **Figure 3**, Supplementary Figure S3). Compared to the control trees, we found no effect of the root flooding treatment or flooding duration on mean or maximum earlywood-vessel area or vessel density (**Table 3**). Only a slightly significant interaction was observed in maximum vessel area between duration and treatment.

#### Internode Expansion

At the end of the growing season all trees that were flooded upon internode expansion, contained significantly narrower earlywood vessels (mean and maximum earlywood-vessel area) at 25-cm stem height compared to 75-cm stem height (**Table 2**, **Figures 3A,B**, Supplementary Figure S3) independent of flooding duration. On average, earlywood vessels areas were 50% smaller in the flooded stem parts compared to 75-cm stem height (**Table 2**). Vessel density was comparable at both heights (**Table 2**, **Figure 3C**). Compared to the control trees, the flooded trees

TABLE 3 | Effects of flooding measured after the end of the growing season following the flooding experiments, on mean earlywood-vessel area (mean vessel area, µm<sup>2</sup> ), mean of the 20 largest earlywood vessels (maximum vessel area, µm<sup>2</sup> ), vessel density (#/mm) and ring width (µm) compared to control trees.


The table shows the results of mixed factorial ANOVAs in which height was the within-subject measure and flooding treatment (treatment vs. control) and flooding duration were between subject factors. F-values and the P-values are given as: <sup>∗</sup>P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001. <sup>∗</sup>Note that the 6-week group (late bud dormancy) was not analyzed as these trees started with budswell and leaf formation.

contained significantly lower mean and maximum earlywoodvessel areas (**Table 3**, **Figures 3A,B**). In addition, we found highly significant interactions for earlywood-vessel size between height and treatment, indicating that earlywood-vessel size was significantly reduced in the flooded stem parts (**Table 3**, **Figures 3A,B**).

#### Collapsed Vessels

We found collapsed, sickle shaped earlywood vessels (**Figure 4**) in flooded stem parts (25-cm stem height) that were flooded upon the phenophases budswell and internode expansion. The number of trees containing collapsed earlywood vessels significantly differed between control and flooded trees after budswell, both immediately after flooding (Fisher's Exact test, p = 0.006, n = 15) as well as after the growing season (Fisher's Exact test, p = 0.017, 15 control trees, 14 flooded trees), Immediately after the flooding experiments, six out of 15 flooded trees contained collapsed vessels, whereas after the end of the growing season seven out of 15 trees were affected by vessel collapse. In the trees that were flooded after internode expansion, vessel collapse occurred in 11 (out of 15) trees immediately after the flooding experiments and was absent in control trees (Fisher's Exact test, p < 0.001, n = 15) and in 14 (out of 15) after the end of the growing season and did not occur in control trees (Fisher's Exact test, p < 0.001 n = 15). Flooding duration did not affect vessel collapse as highly significant differences (Fisher's Exact test, p < 0.01 n = 5) were always observed after 2, 4, and 6 weeks in trees that were flooded upon bud swell or internode expansion; both immediate after the flooding treatments as at the end of the growing season.

#### Root Development Late Bud Dormancy

When the flooding treatment on dormant trees started, root formation had not yet started. The first white, elongating roots were present in four control trees belonging to the 6-week treatment, whereas in the flooded trees new root formation was absent (**Figure 1C**). One tree formed hypertrophied lenticels just below the water level during late bud dormancy.

#### Budswell

At budswell the formation of white roots in the elongation phase had started in all sampled trees (**Figure 1C**). We found significant differences in root dieback between all control and flooded trees (Fisher's Exact test, p < 0.001). During the course of the experiment, root formation of control trees progressed from root elongation (after 2 and 4 weeks), to roots maturing after 6 weeks (**Figure 5A**) and no root dieback occurred. In contrast, in the flooded trees white roots in the elongation phase were unaffected by dieback after 2 weeks of flooding. However, after 4 weeks all white roots in the elongation phase were dying back (Fisher's Exact test, p = 0.011) and after 6 weeks under water, roots were mostly decayed and easily detached from the main root system (Fisher's Exact test, p = 0.008; **Figure 5B**).

Trees of which the roots were flooded, showed a similar pattern with significant root dieback occurring only in flooded trees (Fisher's Exact test, p < 0.002). Significant differences between flooded trees and control trees occurred after 4 weeks (Fisher's Exact test, p = 0.008) and 6 weeks (Fisher's Exact test, p = 0.008). After 6 weeks most roots of flooded tress were decayed (**Figure 5C**). In contrast to the trees of which the stem was flooded, nine (out of 15) of the root flooded trees formed hypertrophied lenticels (**Figure 5C**).

#### Internode Extension

Trees that were flooded upon the phenophase internode extension, showed a similar pattern in root dieback (Fisher's Exact test, p < 0.001). The roots of control trees progressed from roots in the elongation phase, after 2 weeks, to roots maturing after 4 and 6 weeks whereas their flooded counterparts were all affected by root dieback after 4 (Fisher's Exact test, p = 0.008) and 6 weeks (Fisher's Exact test, p = 0.008).

#### Radial Growth

The mean ring widths were always significantly reduced in flooded trees as compared to control trees, when flooding occurred upon budswell or internode expansion (**Table 3**, **Figure 3D**). The trees that were flooded upon budswell showed

a significant reduction in ring width in relation to flooding duration between the 2- and 6-week treatment (Bonferroni Post hoc test, p = 0.005) corresponding to mean ring widths of 4.06 and 1.88 mm, respectively (**Figure 3D**). Root flooding showed a similar patterns with reduced growth with increasing flooding duration; mean ring width was significantly smaller after 6 weeks of flooding compared to the 2-week treatment (Bonferroni Post hoc test, p = 0.018; **Table 2**, **Figure 3D**). In control trees and trees that were flooded during late bud dormancy, ring width was normally larger at 75-cm compared to 25-cm stem height. In flooded trees upon budswell or internode extension the effect of height was insignificant (**Table 2**, **Figure 3D**).

## DISCUSSION

## Earlywood-Vessel Development is Suppressed in Flooded Stem Parts

In 4-year-old pedunculate oak trees harvested immediately after the flooding treatments earlywood-vessel development was suppressed in submerged stem parts if flooding occurred at budswell or internode expansion. In the two trees that were flooded during leaf dormancy but had started leaf development while flooded, vessel development was totally absent in flooded stem parts. This local impediment of earlywood-vessel development in flooded stem parts has, to the best of our knowledge, never previously been reported and is most likely caused by the hypoxic conditions accompanied with flooding (Kozlowski, 1984). As narrow earlywood vessels were frequently found lignified directly after the end of the flooding treatments, it seems likely that lignification occurs under anoxic conditions during flooding. The fact that earlywood-vessel development virtually stops in flooded stem parts is remarkable as ringporous species need to form new earlywood vessel to replace the dysfunctional earlywood in the previous tree ring before the leaves are fully expanded (Cochard and Tyree, 1990; Tyree and Zimmermann, 2002; Copini et al., 2015). In case when no new earlywood vessels are formed in flooded stem parts, water transport most likely occurs in previous-year latewood vessels that are also thought to be of importance during spring reactivation (e.g., Utsumi et al., 1999; Tyree and Zimmermann, 2002; Copini et al., 2015).

#### Flooding Reduces Earlywood-Vessel Size

We found that stem flooding significantly reduces mean and maximum earlywood-vessel area – on average by 50% – in flooded stem parts, when flooding occurs after budswell or internode expansion. This is independent of flooding duration (2, 4, or 6 weeks). The finding that already a 2-week flooding period strongly reduces earlywood-vessel size is in line with Therrell and Bialecki (2015) who studied flood rings in Q. lyrata and Q. macrocarpa along the Lower Mississippi River and linked flood rings to streamflow data. They reported that flooding events of more than 10 days during spring most likely induce a flood ring. Kames et al. (2016) also reported on the formation of flood rings in 2-year-old Fraxinus pennsylvanica trees, when experimental flooding for 3 weeks occurred during the period of earlywood development. Our results are moreover in accordance with many studies that show that earlywood formation normally only lasts between fourth and eighth weeks during spring (Sass-Klaassen et al., 2011; Gonzalez-Gonzalez et al., 2013).

By comparing stem with root flooding, we showed that stems need to be actually flooded to induce changes in the anatomy of tree rings. This is in line with results gained from a flooding experiment in which mature pedunculate oak trees were flooded

till a stem height of 20 cm with stagnant water, resulting in a flood ring in the flooded stem part (Stuijfzand et al., 2008). In addition it may explain why St. George et al. (2002) found that flood rings were present in bur oak (Q. macrocarpa Michx.) at 45 cm, but were absent sometimes above 1.1 m and in general above 3 m. Also Land (2014) found that flood rings in Q. robur were present in flooded basal stem parts but absent at 4-m stem height.

Based on differences found between earlywood-vessel density recorded directly after flooding and after the end of the growing season (**Tables 1** and **2**), our results indicated that a substantial number of earlywood vessels must have been formed after the actual flooding events. These earlywood vessels did not enlarge to normal size, as in the control trees, even though they were not directly affected by anoxic conditions. We can only speculate on the processes behind the reduction of earlywood-vessel sizes after flooding has ceased. Possibly, earlywood-vessel enlargement after flooding was affected by high concentrations of auxin and ethylene which are known to increase during flooding events (Gomes and Kozlowski, 1980; Tang and Kozlowski, 1984; Aloni, 2013).

We expected that juvenile trees flooded after their internodes were expanding, would already contain many enlarged and lignified earlywood vessels, and that consequently mean earlywood-vessel area would be less reduced compared to flooding after budswell. This was not the case in our experiment. A possible reason could be that temperatures during the first 19 days of May 2010 were far below average and among the coldest measured since weather records of the Royal Netherlands Meteorological Institute (KNMI) began. We suspect that these adverse temperatures during the experiment had strongly slowed down earlywood-vessel development, while having less effect on the ongoing leaf development.

## Collapsed Vessels May Pinpoint Flooding Events

We frequently observed vessel collapse in flooded stem sections of many trees flooded upon the phenophase budswell or internode expansion. Earlywood vessel collapse in response to flooding has been reported in pedunculate oak trees growing along the river Main in Germany as well as in a experimentally flooded juvenile trees (Land, 2014). Also freezing temperatures during the period of vessel formation may induce vessel collapse and consequent formation of 'frost rings' (Stahle, 1990; Leuschner and Schweingruber, 1996; Bräuning et al., 2016). However, in contrast to collapsed vessels in frost rings, vessels observed in this study were not surrounded by callus tissue. As collapsed vessels were absent in control trees, and no callus tissue occurred, we can exclude frost as a triggering factor. However, as most trees in the budswell or internodeexpansion phase had already started vessel formation prior to the flooding treatments, we assume that collapsed vessels are the result of vessel-development interruption during the expansion phase (Zasada and Zahner, 1969; Leuschner and Schweingruber, 1996) when the flooding events occurred. As this phenomenon only occurred in submerged stem sections after budswell and internode expansion, the presence of collapsed vessels in a tree ring can be used as a characteristic feature to pinpoint flooding events to the restricted period in the season when earlywood vessels develop.

## Radial Growth and Root Dieback

We found that flooding up to 50-cm stem height as well as root flooding reduces mean ring width when flooding occurred after budswell or internode expansion. This is in line with many studies on juvenile trees that showed that radial growth can be seriously hampered by spring or summer flooding (Coutts, 1982; Yamamoto and Kozlowski, 1987). The presence of many hypertrophied lenticels upon root flooding, which are permeable to water (Groh et al., 2002) and might play an important role in water supply during flooding events (Parelle et al., 2006), did not affect the tendency of trees to grow less with increasing flooding duration.

In our experiment, root dieback occurred in all trees that were flooded (stem or roots) for 4 or 6 weeks after budswell or internode expansion. Reduced growth after flooding is most likely related to the inhibited root development. This is in line with the general view that low oxygen concentrations may inhibit root initiation and seriously affect root development so that new roots need to be developed after flooding events (e.g., Coutts, 1982; Tang and Kozlowski, 1984; Siebel et al., 1998). In addition, reduced radial growth in flooded trees could be an effect of flooding-induced local reductions in earlywood-vessel size in basal stem parts, which creates a hydraulic bottleneck. Other studies explain reduced radial growth by reduced leaf area, stomata closure and early-leaf senescence occurring during flooding (Kozlowski and Pallardy, 1984; Schmull and Thomas, 2000). Whereas, stomata closure may have influenced growth in our experiment, leaf senescence did not occur in our study. Mature trees can tolerate flooding events better than juvenile trees (Kozlowski, 1984), which may partly explain why tree-ring width in mature riparian trees is not reduced in years with flooding events (Astrade and Bégin, 1997; Stuijfzand et al., 2008; Tardif et al., 2010; Land, 2014).

## Implications for Flood Reconstructions

In this study, we synchronized the timing of the flooding for all trees by initiating the flooding treatment at a specific leafphenological stage. Under natural flooding conditions, trees in forests can be in many different stages of leaf and xylem development. Consequently, a flood might be recorded in a particular tree, while the flooding signal is absent in other trees. This is in line with observations by St. George and Nielsen (2000), who found flood rings in between 6 and 24% of the sampled Q. macrocarpa trees. However, it should be noted that Q. macrocarpa is not a true riparian species, it normally grows on the upper floodplain terraces where trees are less frequently flooded. In contrast, Kames et al. (2016) working with riparian black ash (Fraxinus nigra Marsh.) trees found flood rings to be a common feature among trees. The authors also observed that the reduction in the mean vessel areas in flood rings was positively

associated with flood intensity making continuous vessels area chronologies a proxy for flood duration. In line with studies from St. George et al. (2002), Stuijfzand et al. (2008), and Land (2014), we showed that sampling trees for flood signals should include the stem base as flood rings were observed to occur only in flooded stem parts. Potentially this means that by sampling at different heights, an estimation of the flood level might be retrieved. Further studies on mature trees in riparian forests are necessary to support this statement.

In congruence with reports from previous studies (e.g., Therrell and Bialecki, 2015; Kames et al., 2016), we found that relatively short floods, lasting for 2 weeks, induced the formation of narrow earlywood vessels in flooded stem parts when flooding occurs during earlywood formation. Since the earlywood vessels of ring-porous species are formed within a short time window (e.g., Sass-Klaassen et al., 2011; Gonzalez-Gonzalez et al., 2013), flood rings encode for flooding events that occur during a narrow time window. While the start and end of a flooding event does not seem relevant, the event must take place, or start in, the narrow period of earlywood formation. Collapsed vessels can be used to pinpoint flooding events that had started at the moment of earlywood-vessel development. These results indicate that flood reconstructions using flood rings may only be valid to floods that occurred in spring, i.e., at the onset of vessel formation.

#### CONCLUSION

We conclude that relatively short periods of flooding (2, 4, and 6 weeks) reduces earlywood-vessel size drastically, on average by 50%, in flooded stem parts of juvenile pedunculate oak trees. This flood marker occurs in the growing season when the flooding event takes place, but only if flooding occurs after budswell or internode expansion when earlywood vessels are developing. As during flooding, earlywood-vessel development is hampered, the narrow earlywood vessels in flood rings consist of cells that are mainly formed after the actual flooding events. By comparing stem and root flooding, we demonstrated that flood rings only occur in trees of which the stem is flooded. Our study indicates

#### REFERENCES


that root dieback, together with strongly reduced hydraulic conductivity due to extremely narrow earlywood vessels in flooded stem parts, most likely contributes to reduced radial growth along the whole stem after flooding of juvenile oak trees.

## AUTHOR CONTRIBUTIONS

PC, JdO, WL, and US-K designed the study. PC, WL, and LG conducted the research. PC and WL carried out the anatomical and statistical analyses and wrote the first draft of the manuscript which was intensively edited by all authors.

## FUNDING

This article is based upon work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology). PC was supported by the C.T. de Wit Graduate School for Production Ecology and Resource Conservation (PE&RC, The Netherlands) and EMRR by the Research Foundation – Flanders (FWO, Flanders, Belgium).

#### ACKNOWLEDGMENTS

We thank Yvonne Copini, Mathieu Decuyper, Marieke Gonlag, Paula Goudzwaard, Idde Lijnse, John van der Lippe (Unifarm), and Ellen Wilderink for their support with field or lab work or commenting on the manuscript. We are grateful to Ivo Roesink and Marie Claire Boerwinkel of Alterra for their help with the setup of the experiment at the Sinderhoeve. We thank the reviewers for their constructive remarks.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00775



the Red River, Manitoba, Canada. Geology 28, 899–902. doi: 10.1130/0091-7613(2000)28<899:SOHTFI>2.0.CO;2


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Copini, den Ouden, Robert, Tardif, Loesberg, Goudzwaard and Sass-Klaassen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A 6-Year-Long Manipulation with Soil Warming and Canopy Nitrogen Additions does not Affect Xylem Phenology and Cell Production of Mature Black Spruce

#### *Madjelia C. E. Dao1\*, Sergio Rossi2, Denis Walsh2, Hubert Morin2 and Daniel Houle3,4*

*<sup>1</sup> Département Productions Forestières, Institut de l'Environnement et de Recherches Agricoles, Ouagadougou, Burkina Faso, <sup>2</sup> Département des Sciences Fondamentales, Université du Québec à Chicoutimi, Chicoutimi, QC, Canada, <sup>3</sup> Direction de la Recherche Forestière, Forêt Québec, Ministère des Forêts de la Faune et des Parcs, Québec, QC, Canada, <sup>4</sup> Ouranos, Consortium Sur la Climatologie Régionale et l'Adaptation aux Changements Climatiques, Montréal, QC, Canada*

#### *Edited by:*

*Boris Rewald, University of Natural Resources and Life Sciences, Vienna, Austria*

#### *Reviewed by:*

*Zhenzhu Xu, Chinese Academy of Sciences, China Gerhard Wieser, Bundesforschungs- und Ausbildungszentrum für Wald, Austria*

> *\*Correspondence: Madjelia C. E. Dao dao.ebou@gmail.com*

#### *Specialty section:*

*This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science*

*Received: 31 July 2015 Accepted: 02 October 2015 Published: 13 November 2015*

#### *Citation:*

*Dao MCE, Rossi S, Walsh D, Morin H and Houle D (2015) A 6-Year-Long Manipulation with Soil Warming and Canopy Nitrogen Additions does not Affect Xylem Phenology and Cell Production of Mature Black Spruce. Front. Plant Sci. 6:877. doi: 10.3389/fpls.2015.00877* The predicted climate warming and increased atmospheric inorganic nitrogen deposition are expected to have dramatic impacts on plant growth. However, the extent of these effects and their interactions remains unclear for boreal forest trees. The aim of this experiment was to investigate the effects of increased soil temperature and nitrogen (N) depositions on stem intra-annual growth of two mature stands of black spruce *[Picea mariana* (Mill.) BSP] in Québec, QC, Canada. During 2008–2013, the soil around mature trees was warmed up by 4◦C with heating cables during the growing season and precipitations containing three times the current inorganic N concentration were added by frequent canopy applications. Xylem phenology and cell production were monitored weekly from April to October. The 6-year-long experiment performed in two sites at different altitude showed no substantial effect of warming and N-depositions on xylem phenological phases of cell enlargement, wall thickening and lignification. Cell production, in terms of number of tracheids along the radius, also did not differ significantly and followed the same patterns in control and treated trees. These findings allowed the hypothesis of a medium-term effect of soil warming and N depositions on the growth of mature black spruce to be rejected.

Keywords: boreal forest, cambial activity, climate change, N deposition, intra-annual growth, increased soil temperature, wood anatomy, xylogenesis

## INTRODUCTION

Surface temperature is projected to rise over the 21st century under all assessed emission scenarios (IPCC, 2014). Recent forecasts for the boreal forest of eastern Canada estimate increases of 3◦C in mean annual temperature for the year 2050 (Plummer et al., 2006). Plant phenology, the timings of plant development and growth, is one of the traits sensitive to regional climate warming (Schwartz et al., 2006; Cleland et al., 2007). Apart from higher average temperatures, plants may also have to cope with other effects of global change, especially enhanced atmospheric nitrogen (N) deposition. Anthropogenic N depositions have greatly altered the N cycle and plant nutrition in the last two centuries and are projected to increase in the future (Thomas et al., 2010). In the boreal forest, plant growth is often considered to be limited by low temperatures and the availability of N (Reich et al., 2006). Thus, understanding the combined effects of warming and increased N depositions on xylem phenology, tree growth and the amount of cells produced, is critical for improving the prediction of tree responses to future climate.

The key role of soil and air temperatures in xylem phenology and cell production has recently been demonstrated (Rossi et al., 2008; Dufour and Morin, 2013). Soil temperature *<*6◦C strongly inhibits xylem activity and water uptake in various conifers (Alvarez-Uria and Körner, 2007). Also, observations at the northern treeline showed no xylogenesis activity when soil temperature was *<* 3–5◦C (Körner, 2003). Many studies underline the importance of soil temperature in defining the growing season (Körner, 2003; Alvarez-Uria and Körner, 2007). Kilpeläinen et al. (2007) reported that increased temperature resulted in thicker cell walls and higher wood density in Scots pine (*Pinus sylvestris* L.).

Studies on the effect of N deposition on plant growth revealed increased impact of N deposition on plant growth but decreased wood density and cell wall thickness in conifers (Hättenschwiler et al., 1996; Kostiainen et al., 2004).

The combined effect of warming and N fertilization can also be observed in xylem anatomy (Kostiainen et al., 2004; Kilpeläinen et al., 2007). Zhao and Liu (2009), by combining treatments of infrared warming and N deposition in China, obtained further increased performance of *P. tabulaeformis* seedlings but reduced that of *P. asperata*.

Most experimental investigations of N deposition and warming on plant growth have examined juvenile wood or used unrealistically high rates of N addition, so it is not possible to extrapolate the results to the conditions occurring in natural forest ecosystems. Few studies have used an accurate quantity of the additional N inputs that are expected in boreal forest ecosystems in the future, and together with soil warming or singly, the results indicated limited effects on the N status and growth rate after 3-year studies (Lupi et al., 2012; D'Orangeville et al., 2013).

As many factors affect tree growth patterns, short-term studies might be influenced by the confounding effect of several interacting environmental variables on plant growth (Battipaglia et al., 2015). Although these studies indicated no immediate effects after 3 years, we expected that cumulative effects of soil warming and increased N deposition would produce significant modifications in tree growth in the medium term. Few experiments have been done with concomitant variation of soil temperature and N deposition. Thus, the effects in the medium and long term on these environmental factors on trees, especially on cambium phenology and xylem cell production, remain largely unknown. To make realistic predictions of the comprehensive effects of climate change based on experiments a more complete understanding of the relationships between environmental factors (in terms of N deposition and increased temperature) and plant growth requires more years of manipulations.

In this study, we investigated if and how soil temperature and inorganic N deposition influence xylem phenology. We hypothesized that, in the medium term, increased soil temperature and N addition will enhance xylem production and cell differentiation and, in turn, increase the growth rate in boreal forests. To test this hypothesis, during 2008–2013, we used a unique experimental design in the field where inorganic N was repeatedly applied through artificial rain events on the tree canopy and the soil was warmed by 4◦C with buried heating cables in two sites of the boreal forest of Québec, QC, Canada.

## MATERIALS AND METHODS

## Study Site and Tree Selection

The study took place in two mature even-aged stands of black spruce *[Picea mariana* (Mill.) BSP] located at different latitudes and altitudes in the boreal forest of Quebec, Canada. The more northern site Bernatchez (abbreviated as BER) is located near Lac Bernatchez, in the Monts-Valin (48◦51 N, 70◦20 W, 611 m a.s.l.) while the other Simoncouche (SIM) is in the Laurentides Wildlife Reserve, within the Simoncouche research station (48◦13 N, 71◦15 W, 350 m a.s.l.). Both regions are included in the balsam fir-white birch ecological domain (Saucier et al., 1998), with an understorey vegetation mainly composed of *Kalmia angustifolia* L., *Ledum groenlandicum* Oeder, *Cornus canadensis* L., *Vaccinium myrtilloides* Michx., and soil vegetation of *Sphagnum* sp. and mosses [*Hylocomium splendens* (Hedw.), *Pleurozium schreberi* (Brid.), *Ptilium crista-castrensis* (Hedw.) De Not.]. The soil in both regions is podzol with a mor-type humus (Rossi et al., 2015). The mean annual temperature is 0.3 and 2.0◦C at BER and SIM. From May–September mean annual rainfall is 401.8 and 425.4 mm, at SIM and BER, respectively. SIM derived from a forest fire in 1922, while the forest fire at the origin of the stand in BER has been estimated to have occurred between 1865 and 1870. The stands are growing on gentle slopes (8–17%) and drained glacial tills.

In each site, six co-dominant trees were chosen with upright stem, healthy overall appearance and similar growth patterns. The homogeneity in growth rates was assessed during a preliminary investigation by extracting wood cores and counting the number of tracheids along three previous tree rings (Rossi et al., 2007). The average diameter at breast height and the average height of sampled trees were 17 ± 2 and 21 ± 4 cm, and 15 ± 2 and 14 ± 2 m, at BER and SIM, respectively.

## Experimental Design

In each site, two treatments were combined: an increase in soil temperature (H-treatment) and a canopy application of artificial rain enriched with nitrogen (N-treatment). The combination of the treatments resulted in four experimental groups: heated only trees (H), N-enriched only trees (N), heated, and N-enriched trees (NH) and control trees, for which the soil was not heated and that received no N-enrichment (C). The two treatments were attributed randomly to experimental trees resulting in a random split plot design with three replications.

For the H-treatment, heating cables were installed during autumn 2007 between the organic and mineral soil layers, at about 20 cm depth, where the majority of the root system of black spruce is localized (Ruess et al., 2003), following a spiral pattern at a distance of 90–200 cm from the stem collar. Cables were laid by cutting the soil vertically with a shovel or a knife and manually inserting the cable in the resulting narrow "trench", which was then rapidly reclosed. To account for potential root damage and soil disturbance during cable laying, non-heating cables were also installed around non-heated trees (C and N).

Power was supplied by a diesel generator located at 200 m from the site. H treatment consisted of increasing the soil temperature by 4◦C during the first part of the growing season. This led to an earlier snowmelt and an increase in annual soil temperature in agreement with the estimates for 2050 by the FORSTEM climatic model developed for the boreal forest of eastern Canada (Houle et al., 2012). Heating started on different dates according to year and site, usually with a 2 weeks delay between SIM and BER to reflect the difference in temperature between the two sites (Lupi et al., 2010) and achieve a 1–2 weeks earlier snowmelt in heated plots. Soil temperature was measured between the coils of the cables around three heated and three control trees.

The temperature differential between control and treated trees was maintained during April–July, the period in which most cambial division takes place (Thibeault-Martel et al., 2008), to reproduce an earlier snow melt and a longer snow free period. Soil temperature was measured every 15 min and data were stored as hourly averages in CR1000 dataloggers (Campbell Scientific Corporation, Canada). Volumetric water content of heated and non-heated plots was measured in July 2009 using a CS-616 probe (Campbell Scientific Corporation, Canada) mounted on a portable device to check for differences in soil moisture content. No significant difference was found between heated (H and NH) and non-heated trees (C and N) (Lupi et al., 2012).

Artificial rain was produced by sprinklers installed above the canopy of each tree. Each week, the equivalent of 2 mm rainfall was applied to the canopy, during the frost-safe period (June to September), for a number of weeks varying between 12 and 14. Rain was applied over a circular 3-m radius area centered on the stem of each experimental tree, which allowed the canopy area to be covered. Non-N-enriched trees (C and H) were irrigated with a water solution reproducing the chemical composition of natural rainfall at the studied sites (Duchesne and Houle, 2006, 2008), while for N-enriched trees (N and NH), a threefold increase in ammonium nitrate (NH4NO3) concentration was used (**Table 1**).

It is expected that frequent artificial rain additions directly to the canopy, with relatively low inorganic N concentration, better imitate the way anthropogenic derived N depositions are reaching boreal forest ecosystems than massive soil applications do.

#### Meteorological Data

At each site, a standard weather station was installed in a forest gap to measure air, soil temperature and snow depth. The soil temperature was measured both on the mineral and organic layers, at 20–30 and 5–10 cm depth, respectively, and air TABLE 1 | Ion concentration in the artificial rain.


temperature at a height of 2 m. Snow depth was measured with an acoustic distance sensor that quantifies the elapsed time between emission and return of an ultrasonic pulse and automatically corrects for variations of the speed of sound during the year using the measurements of air temperature. Data were collected every 15 min and stored as hourly averages in CR10X dataloggers (Campbell Scientific Corporation, Logan, UT, USA). Daily mean values were later calculated for further analysis.

### Sample Collection and Preparation

During 2008–2013, microcores (2.5 mm diameter and 25 mm long) were collected weekly from the stem from April to October with a Trephor (Rossi et al., 2006a) following a counterclockwise-elevating spiral centered at breast height. In SIM and BER, samples were collected weekly in early summer (May–June) and every 2 weeks during July–October. Microcores usually contained the previous 5–10 tree rings and the developing annual layer with the cambial zone and adjacent phloem tissues. Wood samples were always taken at 5–10 cm intervals to avoid the formation of resin ducts as a reaction to disturbance. The micro-cores were placed in Eppendorf microtubes containing a water:ethanol solution (1:1). Microcores were dehydrated through successive immersions in ethanol and Histosol and embedded in paraffin (Rossi et al., 2006b). Transverse sections 6–10 mm thick were cut with a rotary microtome, stained with cresyl violet acetate (0.16% in water) and observed within 20–30 min under visible and polarized light at magnifications of 400–500× to differentiate the cambium and developing xylem cells. The cambial zone and cells in radial enlargement showed only a primary wall, which, unlike the secondary wall, did not shine under polarized light (Gricar ˇ et al., 2006). Cambial cells were characterized by thin cell walls and small radial diameters, while enlarging cells had a radial diameter at least twice that of a cambial cell. Cells in wall thickening shone under polarized light and during the maturation process showed a coloration varying from light to deep violet. As lignification advanced, a blue coloration starting from the cell corners spread into the secondary walls. Since lignin deposition may persist after the end of cell wall thickening (Gindl et al., 2000), cells were considered mature when the violet was completely replaced by blue (Rossi et al., 2006b; Rossi et al., 2014).

The number of cells in each phase was counted along three radial rows and the total number of xylem cells was calculated as the sum of cells in radial enlargement and wall thickening and lignification and mature cells. In spring, xylem formation was considered to have begun when the average number of cells in enlarging phase between the three radial rows was more than one. In late summer, when no further cells were observed undergoing wall thickening and lignification, xylem formation was considered complete.

The phenology of xylem development was assessed for each tree. Four phenophases, computed in days of the year (DOY), were considered, including onset and ending of both cell enlargement and wall thickening and lignification. Duration of xylem formation was calculated as the difference between the onset of cell enlargement and the ending of lignification.

#### Statistical Analysis

Analysis of variance for repeated measures (ANOVAR) in a splitplot mixed design was used for all variables of phenology and cell production, with site as block factor, treatments N × H as main plots, and Year as repeated factor. As sampling times were correlated, the selection of the covariance structure was based on the lower Akaike's information criterion (AIC). The firstorder autoregressive [AR(1)] provided the suitable correlation structure (Wolfinger, 1993). ANOVAR was performed with the MIXED procedure of JMP Pro 11.1.1 (SAS Institute, Cary, NC, USA). Normality and homoscedasticity were graphically verified on residual plots of the linear models (Quinn and Keough, 2002).

## RESULTS

#### Temperature and Snow Depth

The sites were characterized by long winters with temperatures close to or below zero from October to April. BER, located at the higher latitude and altitude, was the coldest site in winter with the absolute minimum temperatures being measured in 2013 reaching <sup>−</sup>41.63◦C (**Figure 1**). The average temperature for the period ranging between DOY 122– 273 (May–September) in BER was two degrees lower than in SIM (11.8◦C vs. 13.8◦C). Summers were short, with absolute maximum temperatures reaching 26◦C in 2010 (**Figure 1**).

During 2008–2013, mean temperatures in the organic and mineral layers varied between 3.4◦C and 5.5◦C. During winter, soil temperature remained below 3◦C and was always lower at BER (**Figure 1**). Summer temperatures in the organic and mineral soil reached 10–15◦C, starting to increase only after snowmelt (**Figure 1**). Maximum absolute snow depth varied between 1.0 and 1.5 m, with 2012 being the year with the least snow depth (**Figure 1**). The moment of snowmelt varied between years, but, on average, occurred 15 days later in BER.

## Xylogenesis

The onset of xylogenesis started from late May to mid-June (DOY 146–166) (**Figure 2**). The onset of xylem growth was observed later in BER, the colder site. Xylogenesis lasted 95–120 days on average; SIM clearly had the longest duration of xylogenesis, reaching 121 days. Cell production, corresponding to the number of cells produced along the tree ring, differed between the two sites (**Figure 2**). On average, SIM had the highest cell productions and the longest period of growth, with 39 cells.

#### Differences between Sites

The ANOVAR test detected a significant difference in onset of cell enlargement (*<sup>P</sup> <sup>&</sup>lt;* 0.05) between sites (**Table 2**). The onset occurred between DOY 140 and 170 and between DOY 146 and 160 at BER and SIM, respectively (**Figure 3**). It began 1 week earlier in BER, but the latest onset of cell enlargement and the greatest inter-tree variability during the 6 study years were also observed in BER (**Figure 3**). The dynamics of the onset of cell wall thickening was similar in both sites for all treatments but the model showed a significant effect of site (*P <* 0.001), with trees starting wall thickening 11 days earlier in SIM than in BER (DOY 148 and 159, respectively) (**Figure 3**). The onset of cell maturation was significantly different between sites (*P <* 0.01) (**Table 2**) with SIM being the earliest (**Figure 3**). Even if, on average, cell enlargement ended 10 days earlier at BER (DOY 198) than at SIM (DOY 208), no statistically significant difference was found between sites (*<sup>P</sup> <sup>&</sup>gt;* 0.05) (**Table 2**). The duration of xylogenesis was significant between sites (*<sup>P</sup> <sup>&</sup>lt;* 0.01) (**Table 2**), with the whole process being completed in 35 days (between DOY 82 and 117) in BER and 32 days (between DOY 100 and 132) in SIM (**Figure 3**). There was no significant difference for cell production between sites (*<sup>P</sup> <sup>&</sup>gt;* 0.05) (**Table 2**). On average, trees at the SIM site produced more cells in the tree ring, with the highest and lowest values of 48 and 20 cells observed in 2008 and 2012, respectively (**Figure 3**).

#### Differences between Years

Years had highly significant effects on xylem phenology and cell production (*P <* 0.001) and significant effects of the interactions between site and years were also observed for the onset of wall thickening, the mature cell and cell production (*<sup>P</sup> <sup>&</sup>lt;* 0.05) (**Table 2**). The dynamics of xylem formation remained similar in each xylem phenological phase. No significant effect on phenological phase was found with the N treatment but the interaction of N × year was significant (*P <* 0.01). An exception in the year 2010 was observed in the onset of cell wall thickening, which occurred markedly earlier, in late May (DOY 148) (**Figure 3**). The onset of cells maturation also appeared earlier in 2010 (**Figure 3**).

The ending of cell enlargement was observed between mid-July and late August (DOY 198–242) in all sites excepted in 2013 when it occurred later at SIM (mid-September, DOY 252) (**Figure 3**). The ending of cell thickening occurred in late September for all treatments in both sites during 2008–2013, except in 2009 in SIM, where the highest variations between treatments were observed (40 days between control and the other treatments) (**Figure 3**).

During 2008–2013, important variations occurred in the number of cells produced, with the highest variability and the greatest number of cells produced in BER and SIM in 2012 and 2013 (**Figure 3**). In 2013, xylogenesis started 18 days earlier in BER on DOY 82, and ended 15 days earlier on DOY 117, thus the duration of growth was markedly longer in BER (**Figure 3**).

#### Effects of the Treatments

No significant effect of the treatments was found (**Table 2**). The onset of cell wall thickening occurred between mid- (DOY 170) to late- (DOY 181) June for all treatments, with no significant treatment effects (**Table 2**). The treatments also had no significant effect on the moment at which the first mature cells were formed (**Table 2**). The ending of cell enlargement occurred at the same time for all treatments and ANOVAR showed no significant difference between treatments (**Table 2**). It was observed between mid-July and late August (DOY 198–242) in both sites (**Figure 3**). The last cells in wall lignification, which corresponded to the ending of xylem differentiation, occurred between late August and mid-October, with SIM being the later site to complete differentiation (**Figure 3**). ANOVAR performed on the ending of cell wall thickening and lignification showed no significant effect of the treatments (**Table 2**). It was not possible to find significant


TABLE 2 | *F*-values of the mixed procedure with repeated measurements using Site as block factor, Year as repeated factor for the different phases of xylem phenology and cell production.

*The treatments are reported as N (nitrogen), and H (soil warming), and N* × *H (interaction between N and H).*

∗*p < 0.05;* ∗∗*p < 0.01, and* ∗∗∗*p < 0.001.*

effects of the treatments on the overall period for completing process of xylogenesis (**Table 2**).

#### DISCUSSION

This study conducted in two matured black spruce stands of the boreal forest of Quebec, Canada tested the hypothesis that xylem phenology and cell production were affected by increased soil temperature and inorganic N availability in precipitation. Soil temperature was increased by 4◦C during the first part of the growing season and precipitations containing three times the current inorganic N concentration in ambient precipitation were repeatedly applied during the growing season from 2008 to 2013. The experiment consisted of frequent canopy applications of inorganic N at realistic concentrations with the aim of simulating future rain composition. After a 6-year experiment, our results showed no substantial change between treatments in xylem phenology and cell production, so our hypothesis had to be rejected. However, a potential effect of soil warming and increased N deposition could still occur under a longer period of experimentation.

In previous studies, localized warming of the stem often increased cell productions but only in the zone of treatment application (Gricar et al., 2006, 2007 ˇ ). Soil warming significantly enhanced diameter growth of woody individuals, especially shrubs (Farnsworth et al., 1995). McWhirter (2013) investigated the responses of *Malus coronaria* (Crap apple) seed germination and seedling growth to warming and nitrogen in old temperate forests and in greenhouse and the results suggested direct effects on germination and establishment of seedlings. In northern China, the warming response of plant phenology (including flowering and fruiting date as well as reproductive duration)

FIGURE 3 | Phenological phases of xylem differentiation, duration, and total mature cells of xylogenesis in C (control), H (heated), N (nitrogen), and HN (heated and nitrogen) trees in BER and SIM during 2008–2013.

is larger in earlier than later flowering species in temperate grassland systems, but no interactive effect between warming and N addition was found on any phenological event (Xia and Wan, 2013).

In contrast to our results, Lupi et al. (2012) by combining early season soil warming with canopy applications of water containing N at a concentration three times higher than the ambient precipitation in the same sites, detected in the short term, that soil warming resulted in earlier onset and extended duration of xylogenesis in the root and along the stem. The results of Lupi et al. (2012) were not confirmed on our longer period of observations. Melillo et al. (2011) reported carbon gains in the woody tissues of trees in a 7-year soil warming study in a mixed hardwood forest ecosystem. (Overdieck et al., 2007) also recorded increased stem diameter, stem height and stem mass for beech seedlings grown for 2.5 years at increased air temperature.

The absence of growth stimulation in our study, which used realistic increased concentrations of inorganic N concentration in precipitation, suggests that growth stimulation (due to increasing N availability) is not to be expected in the future for the boreal forest of eastern Canada. However, given the experimental conditions used and the gaps in our understanding of N foliar uptake, it would be premature to definitely conclude an absence of effects. For instance, in a labeling experiment with 15N, less than 5% of the label was recovered in live foliage and wood after 2 years of N addition to the canopy with a helicopter (Dail et al., 2009). The majority of the label was recovered in twigs and branch materials. Thus, most of the N was retained on plant surfaces, branches and main-stem bark, with little being assimilated into foliage that could then be transported to new forming cells in the stem. In good agreement with the latter study, another recent study involving canopy application of inorganic 15N in a coniferous stand (Gélinas-Pouliot, 2013) has shown that small twigs, not needles, was the main sink for the added N.

It is, however, possible that the N scavenged by the twigs may take a certain time to reach the stem where cell divisions occur and that a N effect could be observed with a longer period of experimentation. It has also been suggested that changes in nutrient cycling due to increased N deposition and its potential effect on tree growth, may become significant only in the medium and long term, since trees seem less receptive than other plants and microorganisms to the uptake of inorganic and organic N in the short term (Näsholm et al., 2009). Thus, although a direct effect of earlier snowmelt and higher soil temperature on tree growth does not appear likely based on our results, an effect of N addition could potentially appear with longer term addition.

Our results revealed the impact of time scale on xylem phenology and cell production. As shown in **Figure 3** and **Table 1**, the onset of cell wall thickening and the first mature cell had already started earlier in 2010 at the end of May when snowmelt had just finished and air temperature reached values above zero. This led to the conclusion that the onset of wood formation can be affected by snowmelt and temperatures (Rossi et al., 2011a; Dufour and Morin, 2013). Based on studies of conifers across a wide range of different geographical locations, Rossi et al. (2008) found that air temperature is also a critical factor limiting the differentiation of xylem cells.

During the 6 years of treatments, the maximum amount of 48 cells produced occurred in 2012, the least snow depth and warmest year. Several authors had found that the onset and ending increments are affected by an air temperature threshold (Deslauriers et al., 2008; Dufour and Morin, 2010) as well as snowmelt and soil temperature (Rossi et al., 2011b; Lupi et al., 2012). Kalliokoski et al. (2012) indicated that weather variation induced differences of up to 28 days in the onset of tracheid formation between years for Norway spruce. Mäkinen et al. (2003) also revealed that interannual variations in increment onset can be important. The investigations on montane Mediterranean tree species (*Cedrus libani)* at different altitudes reported differences in onset, duration and end of cambial activity and xylogenesis as well as growth rates with respect to temperature, especially daily means of air and stem temperature (Guney et al., 2015).

In the medium term, despite some observed significant effects, the results showed similar dynamics between sites for xylem phenology for all treatments. The differences often observed between sites may therefore rather indicate that xylem phenology and cell production are controlled by temperature and snowmelt. Thus the differences in timing can be explained by the lower average air temperature at the BER site.

## CONCLUSION

The two boreal forest sites studied showed significant differences for xylem phenology of black spruce, except for the ending of cell enlargement and cell production, which could be attributed to their difference in average annual air temperature.

It was, however, found that a 6-year experiment of soil warming and increased inorganic N additions applied directly to the canopy failed to induce significant differences in xylem phenology and cell production at both sites, which allowed our hypothesis to be rejected. Different results could be expected with a longer-term experiment. For instance, it is possible that the N added to the canopy could be slowly translocated from the twigs to the dividing cells and that could lead to an increase in growth in the longer term.

## FUNDING

This work was supported by the Natural Sciences and Engineering Research Council of Canada, the Ouranos consortium and the "Ministère des Forêts de la Faune et des Parcs du Québec". MD received additional financial support through a scholarship from the Program Canadien des Bourses de la Francophonie (PCBF).

## ACKNOWLEDGMENTS

The authors want to acknowledge all persons who contributed to data collection, installations in the field and laboratory assistance, especially F. Gionest, D. Laprise, M. Perrin, L. Balducci, H. A. Bouzidi, M. Montoro, Jannie Trambley.

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Dao, Rossi, Walsh, Morin and Houle. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Seasonal Patterns of Fine Root Production and Turnover in a Mature Rubber Tree (*Hevea brasiliensis* Müll. Arg.) Stand- Differentiation with Soil Depth and Implications for Soil Carbon Stocks

*Jean-Luc Maeght1,2\*, Santimaitree Gonkhamdee3, Corentin Clément4, Supat Isarangkool Na Ayutthaya3, Alexia Stokes2 and Alain Pierret5*

*<sup>1</sup> Institut de Recherche pour le Développement, UMR 242/iEES – Paris (IRD-UPMC-CNRS-UPEC-UDD-INRA), Bondy, France, <sup>2</sup> INRA, UMR-AMAP, Montpellier, France, <sup>3</sup> Khon Kaen University, Faculty of Agriculture, Khon Kaen, Thailand, <sup>4</sup> International Water Management Institute, Vientiane, Laos, <sup>5</sup> Institut de Recherche Pour le Développement, UMR IEES-Paris – Department of Agricultural Land Management (DALaM), Vientiane, Laos*

#### *Edited by:*

*Sergio Rossi, Université du Québec à Chicoutimi, Canada*

#### *Reviewed by:*

*Petra Fransson, Swedish University of Agricultural Sciences, Sweden Aidan M. Keith, Centre for Ecology and Hydrology, UK*

> *\*Correspondence: Jean-Luc Maeght jean-luc.maeght@ird.fr*

#### *Specialty section:*

*This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science*

*Received: 11 September 2015 Accepted: 05 November 2015 Published: 27 November 2015*

#### *Citation:*

*Maeght J-L, Gonkhamdee S, Clément C, Isarangkool Na Ayutthaya S, Stokes A and Pierret A (2015) Seasonal Patterns of Fine Root Production and Turnover in a Mature Rubber Tree (Hevea brasiliensis Müll. Arg.) Stand- Differentiation with Soil Depth and Implications for Soil Carbon Stocks. Front. Plant Sci. 6:1022. doi: 10.3389/fpls.2015.01022*

Fine root dynamics is a main driver of soil carbon stocks, particularly in tropical forests, yet major uncertainties still surround estimates of fine root production and turnover. This lack of knowledge is largely due to the fact that studying root dynamics *in situ*, particularly deep in the soil, remains highly challenging. We explored the interactions between fine root dynamics, soil depth, and rainfall in mature rubber trees (*Hevea brasiliensis* Müll. Arg.) exposed to sub-optimal edaphic and climatic conditions. A root observation access well was installed in northern Thailand to monitor root dynamics along a 4.5 m deep soil profile. Image-based measurements of root elongation and lifespan of individual roots were carried out at monthly intervals over 3 years. Soil depth was found to have a significant effect on root turnover. Surprisingly, root turnover increased with soil depth and root half-life was 16, 6–8, and only 4 months at 0.5, 1.0, 2.5, and 3.0 m deep, respectively (with the exception of roots at 4.5 m which had a halflife similar to that found between depths of 1.0 and 2.5 m). Within the first two meters of the soil profile, the highest rates of root emergence occurred about 3 months after the onset of the rainy season, while deeper in the soil, root emergence was not linked to the rainfall pattern. Root emergence was limited during leaf flushing (between March and May), particularly within the first two meters of the profile. Between soil depths of 0.5 and 2.0 m, root mortality appeared independent of variations in root emergence, but below 2.0 m, peaks in root emergence and death were synchronized. Shallow parts of the root system were more responsive to rainfall than their deeper counterparts. Increased root emergence in deep soil toward the onset of the dry season could correspond to a drought acclimation mechanism, with the relative importance of deep water capture increasing once rainfall ceased. The considerable soil depth regularly explored by fine roots, even though significantly less than in surface layers in terms of root length density and biomass, will impact strongly the evaluation of soil carbon stocks.

Keywords: deep roots, root phenology, root turnover, soil carbon, root access well, drought

## INTRODUCTION

Fine root production and turnover represent 22% of terrestrial net primary production globally (McCormack et al., 2015). Yet there are still major uncertainties about the mechanisms that control fine root production and turnover. With the growing global demand for food and plant-derived commodities, unraveling such mechanisms is becoming increasingly important, particularly with the concomitantly pressing need to develop more sustainable agro-ecosystems. In recent years, rubber tree (*Hevea brasiliensis* Müll. Arg.) plantations have rapidly expanded, especially in marginal regions where the climate is much drier than in the species' natural range and where seasonal drought occurs (Carr, 2011). Soaring prices of natural rubber in the late 2010s influenced governmental policies regarding the expansion of *H. brasiliensis* cultivation. In Thailand, the world's leading latex producer, where the surface area planted with *H. brasiliensis* was multiplied by a factor of 75 between 1980 and 2008, from 24,000 ha−<sup>1</sup> to 1.8 million ha−<sup>1</sup> (Carr, 2011). Given the stress that tapping, i.e., the process by which the latex is collected, already imposes on *H. brasiliensis*, the sustainability and profitability of latex production in such areas could greatly benefit from adapting tapping modalities by taking into account the physiological response of trees to water availability (Boithias et al., 2011; Junjittakarn, 2012).

As roots are conduits for nutrients and water from the soil to plants, they have a determining role with regard to tree resilience to a range of environmental constraints, especially water stress (Boyce, 2005; Lobet et al., 2013). Fine roots are also an integrative indicator of plant response to environmental factors (Edwards et al., 2004) and we assume that root production or elongation of *H. brasiliensis* is synchronized with rainfall patterns (Green et al., 2005), although there exists evidence of endogenous controls of root growth (Abramoff and Finzi, 2015). Therefore, we expect that fine root growth is arrested during the dry season, but no data exist to support or refute this hypothesis. Fine roots of trees contribute to soil water extraction (Danjon et al., 2013), while a variable (and most often poorly quantified) share of the water demand is supplied by deep roots (Maeght et al., 2013). The measurement of root growth and survival *in situ* along a deep soil profile is an approach that can bring essential information to understanding how trees cope with water-limiting conditions and tree resilience to such constraints.

Quantifying fine root phenology and mortality down a soil profile, and particularly in deep soils, will also impact the evaluation of belowground carbon stocks, an area where data are scarce (Wauters et al., 2008). This proportion of soil carbon stocks could well contribute to the balance between the release and accumulation of carbon fluxes, currently described as the "missing sink" (Esser et al., 2011). However, to observe and analyze roots non-destructively within the soil matrix is still a major scientific challenge (Virginia and Jarrell, 1987), especially in deep soil layers and most studies have focused on the superficial layers of soil (Stone and Kalisz, 1991; Maeght et al., 2013). Our knowledge of fine root lifespan is also limited, particularly at depth (Eissenstat and Yanai, 1997).

We hypothesize that: (i) rooting in general is deeper than commonly assumed, (ii) fine root phenology and mortality are synchronized with annual patterns of precipitation, (iii) fine roots growing deep in the soil contribute to the resilience of *H. brasiliensis* to frequently occurring drought conditions in N. E. Thailand. Therefore, we measured seasonal patterns of fine root production and turnover in a mature stand of *H. brasiliensis*, down a 4.5 m soil profile during a 3-year observation period. We examined the interactions between fine root dynamics, rainfall, and soil depth and estimated the relative contribution of fine and deep roots to soil carbon.

## MATERIALS AND METHODS

### Study Site and Climate

The field site was located at Baan Sila Khu-Muang village, Buriram province in North East Thailand (N 15◦16 23, E 103◦04 51.3, 150 m a.s.l.). This region is part of the nontraditional areas for *H. brasiliensis* cultivation established since the 1990s. The experiment was set up in 2006 in a monoclonal plot of 14 years old trees (RRIM 600 clone developed by the Rubber Research Institute of Malaysia), planted at 2.5 m × 7.0 m spacing (∼570 trees ha−1) that had already been tapped for over 4 years to produce latex. Tapping was performed using a semi-spiral cut 2 days out of 3 and is largely adapted to the local climatic conditions. Tapping is usually discontinued for 3–4 months during the dry season. The maximum leaf area index (LAI), measured using 91 m<sup>1</sup> litter traps during the defoliation period (December–February; Isarangkool Na Ayutthaya et al., 2010), was estimated to be 3.9 ± 0.7 (mean ± standard deviation).

This marginal area for rubber tree cultivation is subjected to the Southeast Asian monsoon, with heavy rainfall between April and October. Local microclimate was monitored automatically with a Minimet weather station (Skye Instruments Ltd, UK) attached to a data logger recording air temperature, relative humidity, incident short wave radiation and rainfall at 30-min intervals. Reference evapotranspiration (ET0) was calculated according to Allen et al. (1998) using the data collected from the weather station.

## Soil Properties

The soil at the study site was a deep loamy sand with limited water retention capacity, developed on fine sand or coarse silt deposits with a homogeneous sandy loam texture throughout the profile. The Ap horizon was a 0.25 m thick remnant from previous sugar cane (*Sacharum officinarum L.*) cultivation (Hartmann et al., 2006). Clay, silt, and sand contents were 100, 100, and 800 g kg<sup>−</sup>1, respectively. The clay content increased with depth: from 150 g kg−<sup>1</sup> in the Bt1 horizon (0.25–0.50 m) to 200 g kg−<sup>1</sup> in the Bt2 (0.50–1.0 m). Silt content was similar in all soil layers throughout the soil profile (100 g kg<sup>−</sup>1) while sand decreased to 700 g kg−<sup>1</sup> at a depth of 1.0 m. From 100 to about 4.0 m, these properties remained fairly stable. The laterite layer was found at a depth of 6.0–7.0 m, as previously observed

in this region (Cawte and Boyd, 2010). The water table was not found within the first 7 m of the profile, even during the rainy season. The soil was acidic with a pH ranging from 5.0 to 5.3. Soil organic carbon content was lower than 10 g kg−<sup>1</sup> in the topsoil (Isarangkool Na Ayutthaya et al., 2011). Typical bulk soil density was 1.55 g cm−<sup>3</sup> to a depth of 3.0 m (Gonkhamdee et al., 2009). Additional soil properties can be found in Hartmann et al. (2006) and Isarangkool Na Ayutthaya et al. (2011).

## Root Growth Monitoring and Rooting Profiles

Soil Coring To quantify carbon stocks associated with fine roots, root samples were collected at depths corresponding to the depths covered by root windows. We extracted undisturbed soil cores using standard soil sample steel rings (diameter 53 mm, height 50 mm and 100 cm<sup>3</sup> internal volume, Eijkelkamp Giesbeek, The Netherlands), in the vicinity of the root access well (*n* = 12 at soil depths 0.25, 0.50, 0.75, and 1.0 m; *n* = 5 at soil depths 1.6, 2.8, and 4.0 m). Root samples were analyzed according to Pierret et al. (2007b). Roots were first washed free of soil from the undisturbed soil cores and then imaged using a flatbed scanner (Epson Perfection V700 Photo scanner; Seiko Epson Corp., Japan) in light transmission mode, at a spatial resolution of 600 dpi (pixel size of 0.0423 mm). Special care was taken to separate every root from each other as much as possible, since overlapping roots are known to impair accurate length recovery. Specific root length (SRL) values, i.e., the length of root per unit dry root biomass, obtained from fine root samples collected within the first meter of the soil profile, were used to estimate the root dry biomass (RDB) distribution along the 4.5 m

profile observed in the well, based on the following equation:

$$\text{RDB} = \text{RLD} \times \text{ [Z]/SRL} \tag{1}$$

where RDB (in Mg ha−1) is the RDB in a soil depth layer of thickness [Z] (m), RLD is the root length density (m m<sup>−</sup>3) calculated from soil cores, in this soil layer and SRL the specific root length (m g<sup>−</sup>1).

#### Root Access Well

Root growth was studied using the access well technique described in Maeght et al. (2013). An access well (0.9 m in diameter and 4.5 m deep) was installed in July 2006 at a distance of 1.35 m from two trees and 0.5 m aside from a tree row. The access well observation technique is an evolution of basic techniques for root observation at transparent interfaces with soil (Smit et al., 2000). A total of nine observation windows were cut through the concrete walls of the well in staggered rows (with 1.0 and 0.5 m horizontal and vertical spacing, respectively). Each root window included a specifically designed metallic frame supporting, on its upper side, a piece of 8 mm thick glass (0.25 m × 0.30 m) pressed against the soil at a 45◦ angle by means of two threaded rod actuators. On the frame's lower side, two guide rails allow the insertion of a standard flatbed scanner. Overall, given the geometrical arrangement of windows, the soil depth increments that were accessible were 0.4–0.6, 0.9–1.1, 1.4– 1.6, 1.9–2.1, 2.4–2.6, 2.9–3.1, 3.4–3.6, 3.9–4.1, and 4.4–4.6 m. For simplicity, these are referred to as 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0, and 4.5 m hereafter. Due to time and financial constraints, it was not possible to build and monitor replicate root access wells within the framework of this field experiment. More details about the set up of the root access well set be found in Gonkhamdee et al. (2009) and Maeght et al. (2013).

Images of roots were taken using a flatbed scanner (HP Scanjet 4370 Photo scanner at 200 DPI – Hewlett-Packard Development Company, California) and custom software which offers a convenient, faster and more accurate record than manual techniques (Zoon and Tienderen, 1990; Kaspar and Ewing, 1997). Root windows were scanned at monthly intervals during 3 years starting in January 2007. This scanning was started 6 months after setting up the access well to avoid recording overproduction of roots at the onset, as often occurs in mini-rhizotron experiments (Yuan and Chen, 2012). Long-term observations are also highly recommended to avoid the risk of overestimating fine root turnover (Strand et al., 2008). Root growth monitoring was conducted following a procedure described in Maeght et al. (2007). Images of the soil and roots in direct contact with each window were used to estimate root length, radius, and time of root appearance/disappearance (from which root turnover was inferred). We used the Gimp freeware package1 to digitize roots and classify them as live or senescent. Senescent roots are often difficult to identify with certainty (Majdi et al., 2005). We considered roots as senescent when they exhibited no elongation

and/or radius expansion for at least two successive observation dates and when their color turned from white to brown. Senescent roots were considered dead when their color changed to dark brown/black or when they completely disappeared from one observation to the next. A total of more than 1500 roots distributed in 300 images were processed.

Root emergence was quantified as the number of roots appearing between two monthly observations divided by the number of root windows from which this number of roots was derived (number of new roots per cm2 and per month). Likewise, root mortality was quantified as the number of roots that disappeared between two monthly observations divided by the number of root windows from which this number of roots was derived (number of senescent/dead roots per root window and per month). 95% confidence intervals were computed as an indicator of the variability of root emergence/mortality across windows.

Assuming that all roots had emerged at the same time, the halflife represents the time after which half of all the roots would have died. Half-life values simplify comparisons between the survival

<sup>1</sup>www*.*gimp*.*org

of roots that emerged at the onset of the observation period and for which the time of disappearance could actually be recorded and that of roots that emerged much later and which died after the end of the observation period (right-censored data).

## Analysis of Root Sample Images

Root length measurements were obtained using IJ\_Rhizo's implementation of the method developed by Kimura et al. (1999). IJ\_Rhizo (Pierret et al., 2013) is a software designed to measure roots washed from soil samples and developed in the ImageJ2 macro language. The approach developed by Kimura et al. (1999) is based on discriminating each pixel of the medial axis (or skeleton) of each digitized root according to its number of orthogonal and diagonal (vertical or horizontal) neighbors. We also used IJ\_Rhizo to compute frequency distributions of root radius classes (i.e., the cumulated root length par root radius class).

### Statistical Analyses

All numerical data processing and statistical analyses were performed within the R environment (version 3.0.2; R Development Core Team, 2013). We first explored our dataset using a principal component analysis (PCA; "ade4" package, version 1.6-2). RLD and root radius values are reported as mean ± 95% confidence interval. We applied analysis of variance with Tukey's HSD *post hoc* tests to determine differences in root radius at different soil depths. We assessed fine root survival at the depth of each root using a Kaplan–Meier survival analysis (Kaplan and Meier, 1958) implemented in the "*survival*" package (version 2.37-4) of the R environment. For each individual root, we recorded the time of emergence and the time to either an event (death) or the end of the study (i.e., roots that were still alive at the time of the last observation were right-censored). Differences in survivorship of roots that emerged at different soil depths (regardless of their actual time of emergence) were assessed by *post hoc* pairwise comparisons using the Mann–Whitney test with a Bonferroni correction. Additionally, a Cox proportional hazards regression model was used to test whether root radius had an influence on root survival. Fine root emergence and mortality determined for each root observation window at monthly intervals are reported as pseudo-medians derived from the Wilcoxon test ± 95% confidence intervals.

A PCA (Supplementary Figure S1) indicated that root emergence was partly explained by rainfall and evapotranspiration, but the two first axes accounted for less than 50% of the variance. Therefore, we resorted here to a more descriptive analysis of fine root dynamics. As (i) roots with lifespans of 30 months and more only occurred between the surface and a depth of 2.0 m, (ii) roots were thicker above 2.0 m (with the exception of the 1.5 m depth increment) and (iii) at depths below 2.0 m, roots emerged at least 12 months later than at depths above 2.0 m, we chose to analyze separately root dynamics above and below the soil depth of 2.0 m.

2http://rsbweb*.*nih*.*gov/ij/

## RESULTS

## Climate Measurements

Total annual rainfall during the 3-year period over which the experiment was conducted, was 965, 1265, and 1002 mm, in 2007, 2008, and 2009, respectively (average: 1077 mm; **Figure 1**). Air temperatures increased during the dry season and decreased following the end of the rainy season (with a range from +8.3 to +40.3◦C). Reference evapotranspiration (ET0) was found to roughly follow the monsoon regime, with a peak toward the end of the dry season and a subsequent decrease throughout the rainy season (**Figure 1**).

## Rooting Profiles

Mean fine RLD derived from soil coring decreased by about one order of magnitude from a depth of 0.05 to 0.5 m, then declined slightly from 0.5 to 1.5 m before further increasing at 2.82 m (*F*7*,*<sup>55</sup> <sup>=</sup> 7.49, *<sup>p</sup> <sup>&</sup>lt;* 0.001; **Figure 2**). A *post hoc* Tukey test showed that mean fine RLD was significantly higher at 0.05 m than at all other depths (*p <* 0.05) and that fine root RLD between 0.25 and 4.0 m were not significantly different from each other.

Mean root radius measured in root windows significantly varied with soil depth (*F*8*,*<sup>89</sup> = 34.15, *p <* 0.001), reaching 0.38 ± 0.03 mm at 0.5, 0.38 ± 0.02 mm at 2.0 m and 0.32 ± 0.02 mm at 1.0 m. At all other soil depths, mean root radius was fairly constant at 0.23–0.27 mm (**Figure 3**). A *post hoc* Tukey HSD test showed that mean root radii at 0.5 and 2.0 m were significantly higher than those at 1.0 m at *p <* 0.05 and that mean root radius at 1.0 m was itself significantly higher than those at all other soil depths (*p <* 0.05).

## Root Emergence and Age Distributions

Roots near the soil surface (0.0–0.5 m) had a significantly longer life span compared to that in deeper soil layers (**Figure 4A**). There was a significant effect of soil depth on root age (*c*<sup>2</sup> <sup>=</sup> 94.93, *<sup>p</sup> <sup>&</sup>lt;* 0.001). Roots with life spans of 30 months and more were only observed between the surface and a depth of 2.0 m.

Roots first emerged in layers close to the soil surface (0.5 and 1.0 m) and at 3.5 m, i.e., within the first 6 months of the observation period. However, at depths of 2.5, 3.0, and 4.0 m, roots did not emerge until 11–17 months after the beginning of the observation period (**Figure 4B**). Root emergence differed significantly depending on soil depth (*F*8*,*<sup>89</sup> = 34.15, *p <* 0.001). Root emergence occurred significantly earlier (*p <* 0.05) in the three top windows (means were: 11, 14, and 17 months at 0.5, 1.0, and 1.5 m, respectively) than in the deeper windows (means were: 21–25 months).

#### Root Survival

Kaplan and Meier (1958) curves showed that, overall, root halflife decreased with soil depth, with the half-life of roots at 0.5 m being in the order of 500 days (*>*16 months; turnover of 0.73 yr<sup>−</sup>1). The half-life of roots between 1.0 and 2.5 m was about 180–250 days (6–8 months; turnover 1.46–2.03 yr<sup>−</sup>1) and that of roots at 3.0 m and below dropped to less than 120 days

(∼4 months; turnover 3.04 yr−1), with the exception of roots at 4.5 m; the latter had a half-life similar to that found between 1.0 and 2.5 m (**Figure 5**). Soil depth significantly affected root halflife values (χ<sup>2</sup> <sup>=</sup> 89.9, *<sup>p</sup> <sup>&</sup>lt;* 0.001) and survival at 0.5 m was longer than that at all other depths (*p <* 0.05) except for 2.5 and 1.5 m, while there was no difference in root survival between depths of 3.5–4.5 m.

Differences in root survival might be related in part, to root branching order, with higher branching order roots (Pregitzer et al., 2002), living longer, i.e., thicker roots observed at 0.5 m (**Figure 3**). This could be putatively associated with slower turnover compared to lower order roots (Yao et al., 2009; Sun et al., 2012). However, a Cox proportional hazards model including root radius as a

covariate of soil depth showed that root survival was clearly influenced by soil depth (*p <* 0.001) and not by root radius (*p* = 0.25).

## Root Dynamics as a Function of Soil Depth and Rainfall

Root emergence between soil depths of 0.5 and 2.0 m ranged from 1.60 to 107.42 × 10−<sup>3</sup> emerging roots cm−<sup>2</sup> month−1, with an average of 7.33 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−<sup>1</sup> (**Figure 6A**). Despite much variability, root emergence tended to be lowest in the first 3–4 months of each observed year, followed by an increase that lasted at least until the month of November (although there was much variability between depth increments and observation years, **Figure 6A**). During the first 2 years, root emergence increased approximately 3 months after the first rainfall, usually in the month of June. Root emergence could occur at relatively high rates during defoliation but was generally low during leaf flushing (**Figure 6A**). The dynamics of root emergence observed below 2.0 m was radically different with root emergence ranging from 1.60 to 128.27 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−1, with a mean of 7.11 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−<sup>1</sup> (**Figure 6B**). There was very limited root growth until the 11th month of the observation period – or 14 months after root windows were installed – (i.e., until the onset of the first dry season and during leaf fall). Beyond that point in time, root emergence subsided until February 2008 and increased again and remained relatively high for 1 year (with a mean of 11.81 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−1). Subsequently, root emergence slowed down and became more stable over time with a mean of 6.57 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−<sup>1</sup> (**Figure 6B**).

The range of root mortality between soil depths of 0.5 and 2.0 m was 1.60 to 40.08 × 10−<sup>3</sup> dead roots cm<sup>2</sup> month, with an average of 6.33 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−<sup>1</sup> (**Figure 7A**). Root mortality at these soil depths was relatively stable over time with the highest mortality rates observed from August to January. Below a depth of 2.0 m, the range of root mortality was 1.60–78.56 <sup>×</sup> <sup>10</sup>−<sup>3</sup> dead roots cm−<sup>2</sup> month−1, with an average of 5.28 × 10−<sup>3</sup> emerging roots cm−<sup>2</sup> month−<sup>1</sup> (**Figure 7B**). Following the initial period of root emergence at these soil depths, root mortality tended to remain at relatively high levels from June 2008 until June 2009 (a mean of 10.04 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−1), beyond which it stabilized at a lower level of 6.65 <sup>×</sup> <sup>10</sup>−<sup>3</sup> emerging roots cm−<sup>2</sup> month−1.

Bivariate plots of monthly root length variations as a function of (1) average monthly rainfall, (2) monthly average of minimum daily soil temperature, and (3) average reference evapotranspiration are given in Supplementary Figure S2. There was a weak yet significant (as indicated by the low R-squared and *p*-values of the regressions) positive relationship between, on the one hand, average monthly rainfall and monthly root length variations (Supplementary Figure S2A) and on the other hand, monthly average of minimum daily soil temperature and monthly root length variations (Supplementary Figure S2C).

## DISCUSSION

#### Fine Root Emergence

We showed that root phenology differed along the soil profile and was not synchronized with rainfall patterns as we had hypothesized, particularly below a depth of 2.0 m. Within the first 2 m of the soil profile, the highest rates of root emergence occurred about 3 months after the onset of the rainy season, while deeper in the soil, root emergence remained low until the 11th month of the observation period and was not correlated with the rainfall pattern. Therefore, the shallow parts of the root system were more responsive to rainfall, as roots near the soil surface capture water from rainfall more readily than deeper roots. Deep roots only emerged once rainfall became scarcer and may reflect the need for trees to use increasingly deeper water resources during the dry season.

Below 2.0 m, the first peak of root emergence rates occurred in November and December 2007, followed by a period of high root emergence that spanned from July 2008 to January 2009 (with the maximum peak in January 2009). Surprisingly, the highest emergence rates below 2.0 m occurred during the period of aerial dormancy, i.e., with no leaves supplying resources for root growth through photosynthesis. Similar results, whereby broadleaf tree root growth occurs significantly during a period of aerial dormancy, were also found in a Mediterranean climate in *Juglans regia* L. (Germon et al., under revision). Abramoff and Finzi (2015) suggest that endogenous cuing (i.e., any factor that affects growth other than climate), and subsequent allocation of stored non-structural carbohydrates (NSC) are dominant drivers of root growth in subtropical and Mediterranean trees. Although the climate in our study was tropical, distinct rainy seasons are present, but soil and air temperatures remain warm, therefore water supply is likely the main limiting climatic factor, particularly in the upper soil horizons where less buffering exists against soil drying. In deeper soils, thermic and hydric buffering should thus allow for more constant rates of root growth throughout the year if endogenous cuing is not the main driver of growth. However, we found that the peak of deep root growth was delayed until late into the dry season. As tree root and stem NSC usually decline during the growing season and reaccumulate during aerial dormancy (Richardson et al., 2013), NSC re-accumulation rates may differ between shallow and deep roots, with a time lag resulting in delayed deep root growth. Nevertheless, as root emergence rates were so different between shallow and deep roots, it may be that the drivers between the two compartments are separate and distinct, with rainfall driving shallow root growth and endogenous cuing driving deep root growth. However, further studies using isotopes would be needed to support this hypothesis (Trumbore et al., 2006).

Between soil depths of 0.5 and 2.0 m, root mortality was relatively disconnected from variations in root emergence, although higher mortality values occurred toward the end of the rainy season, as did the highest emergence values. Below 2.0 m, from 2008 onward, peaks in root emergence and death were largely synchronized, e.g., in June and September 2008 as well as April–May 2009, suggesting the existence of a mechanism for the replacement of recently senesced roots. It is also possible that deep roots that first grew after the installation of the well, began to die, either because they could not be maintained by the tree (too costly in terms of resources) or because the relatively high rates of emergence during the second year were a response to the disturbance caused by the well installation, as often occurs in rhizotron experiments (Strand et al., 2008). Roots growing in the direction of the well may have had reduced access to resources (since the volume occupied by the well was inaccessible) thus suppressing root emergence (feedback response). Observed differences in root emergence could also be related to the presence/absence of roots in the immediate vicinity of observation windows, prior to their installation which increased the probability for early root emergence in windows near pre-existing roots. The total higher and lower rates of root emergence observed in 2008 and 2009, respectively, may also have been influenced by the total annual precipitation, which in 2009 (1002 mm) was only 79% of that in 2008 (1265 mm). The year 2007 was the driest during the observation period with only 965 mm total annual precipitation. However, the second semester of year 2009, was the driest second semester of the monitoring period with only 78% of the rainfall that had been monitored for the same period in the two preceding years. In the first 3 months of the rainy season 2009 (April–June) it rained less than 60% of that for the same period in 2008. However, the reduced emergence within the first 2 m, putatively related to drier conditions during the rainy season, was not counterbalanced by increased emergence at depth. Therefore, while our data support the hypothesis that deep root emergence might correspond to a safety net against water stress during the dry season, they do not point at the existence of a similar mechanism against dry periods occurring during the wet season.

The peak of evapotranspiration that occurs every year around March–April, was highest in 2007, intermediate in 2008 and lowest in 2009 (it did not occur in 2009 as high rainfall occurred as soon as March 2009). We hypothesize that high evaporative demand may be a signal that triggers root growth at the onset of the rainy season, particularly near the soil surface and the low evapotranspiration observed in 2009 may have resulted in a weaker pattern of root emergence in that year.

At all depths, in 2008, root emergence reached its lowest level throughout the period during which trees had already shed old leaves but not yet started to grow new leaves (i.e., February). Although this pattern could not be observed in the previous and following years, it might correspond to a dormancy state prior to the resumption of physiological activity with the first rains of the season.

## Fine Root Turnover

As generally reported in the literature (Chen and Brassard, 2013), we found that soil depth had a significant effect on root turnover, but rather unexpectedly, root turnover increased with soil depth from 0.73 yr−<sup>1</sup> at 0.5 m to 2–3 yr−<sup>1</sup> at greater depths, in sharp contrast with what is generally reported (Wu et al., 2013). However, most studies have been concerned with soil depth ranges that were much shallower than those in our study (Baddeley and Watson, 2005; Chen and Brassard, 2013). Similarly, we did not find any evidence that root radius had an influence on longevity, although this has also been reported in the literature (Baddeley and Watson, 2005; Chen and Brassard, 2013). Furthermore, we did not find a linear increase in root turnover with soil depth over the whole 4.5 m range investigated, which is consistent with the theory that several factors, both intrinsic and extrinsic, control fine root life span (Chen and Brassard, 2013). It is known that environmental parameters (e.g., temperature, water content, N availability, CO2 partial pressure) influence fine root turnover to variable degrees (Vogt et al., 1996). Therefore, we hypothesize that during dry periods, deeper distal roots underwent a physiological pruning process, whereby peripheral organs died, as also observed in shoots of *H. brasiliensis* (Chen and Cao, 2015).

#### Fine Root Biomass and Carbon

Assuming that the RLD values that we obtained from soil coring are homogeneous over large volumes of soil, it can be inferred that fine root biomass below a depth of 1.0 m could account for more than half of the overall fine root biomass of the rubber trees measured [4.8 t ha−<sup>1</sup> between 0.0 and 1.0 m compared to 5.8 t ha−<sup>1</sup> between 1.0 and 4.0 m, with a mean SRL of ∼14 m/g−<sup>1</sup> for roots ≤1 mm in diameter (Pierret et al., 2007a)]. As roots may also be present below a depth of 4 m (the bedrock was found at 7–8 m), total root biomass may be underestimated. Assuming that rubber tree root tissues have a mean organic carbon content of approximately 47% (Wauters et al., 2008), our results show that rubber tree roots ≤0.5 mm in diameter, on average account for about 5 t C ha<sup>−</sup>1. This value is similar, although slightly higher, than the 1.91–3.72 5 t C ha−<sup>1</sup> range reported by Wauters et al. (2008) for coarser roots (2.5–25 mm in diameter) for a range of rubber tree clones from Western Ghana and Brazil. Similarly Cheng et al. (2007) estimated carbon stocks of 16.50 t C ha−<sup>1</sup> for roots of all sizes, in rubber tree plantations at Hainan Island, China and Yuen et al. (2013) calculated total carbon stocks of the order of 4–32 t C ha−<sup>1</sup> for rubber trees at six locations in Southeast Asia. Hence, the presence of fine roots at fairly low length densities over considerable soil depths can have important implications with regard to soil carbon accounting. As recently pointed out by Yuen et al. (2013), more attention should be given to sampling roots at appropriate depths if we are to improve baseline data on belowground carbon stocks. In addition, it should be acknowledged that there are still major uncertainties regarding (1) the reliability of coring versus imaging techniques for quantifying fine root biomass and turnover (Yuan and Chen, 2012) and (2) the way different fine root definitions might influence such quantifications (McCormack et al., 2015).

## CONCLUSION

We explored the interactions between fine root dynamics, the rainfall regime and soil along a 4.5 m profile using a root accesswell. Our results reveal that root growth dynamics in the upper 2 m of soil surface were related mainly to precipitation patterns (Chairungsee et al., 2013). Deeper in the soil, root growth was more independent of rainfall and was likely driven by internal tree carbon allocation. We show that fine root production will impact soil carbon stocks and was higher than commonly reported (e.g., Brunner and Godbold, 2007), particularly at depth. Such an input of fine root related carbon in soils could be all the more significant considering the slow breakdown of fine roots in some sub-tropical tree species (Xiong et al., 2012). One major limitation of this work is that observations are taken from a single location, which means that inference and conclusions cannot be generalized. The results of this study thus advocate in favor of more field studies aimed at assessing precisely the production and fate of fine roots, not only near the soil surface but also very deep in the soil.

## AUTHOR CONTRIBUTIONS

JLM and AP designed the experimental setup, implemented it in the field, analysed the data and wrote the paper.

## REFERENCES


SG and SINA performed data collection in the field. CC, AS and SG have contributed significantly to the data analysis, discussing the results and writing of the paper.

### ACKNOWLEDGMENTS

This research was funded by the French Institute of Research for Development (IRD), France, the French Institute for Natural Rubber (IFC), France, and Michelin/Socfinco/SIPH Rubber Tree Plantations Companies. We would also like to thank all our Thai counterparts from Khon Kaen University (KKU), Land Development Department (LDD), and the owner of the plantation (Mr. Chaipat Sirichaiboonwat) who kindly welcomed us. The authors also wish to thanks Drs. D. Nandris and F. Do (IRD) for their interest in and support of this research.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal*.*frontiersin*.*org/article/10*.*3389/fpls*.* 2015*.*01022


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Maeght, Gonkhamdee, Clément, Isarangkool Na Ayutthaya, Stokes and Pierret. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Differentiated Responses of Apple Tree Floral Phenology to Global Warming in Contrasting Climatic Regions

#### Jean-Michel Legave<sup>1</sup> \*, Yann Guédon<sup>2</sup> , Gustavo Malagi <sup>3</sup> , Adnane El Yaacoubi <sup>4</sup> and Marc Bonhomme<sup>5</sup>

1 INRA, Unité Mixte de Recherche 1334 Amélioration Génétique et Adaptation des Plantes Méditerranéennes et Tropicales, Montpellier, France, <sup>2</sup> CIRAD, Unité Mixte de Recherche 1334 et Inria, Virtual Plants, Montpellier, France, <sup>3</sup> Faculdade de Agronomia, Universidade Federal de Pelotas, Pelotas, Brazil, <sup>4</sup> Faculté des Sciences, Université Moulay Ismail, Meknès, Morocco, <sup>5</sup> Unité Mixte de Recherche 547, INRA et Université Blaise Pascal, PIAF, Clermont-Ferrand, France

#### Edited by:

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### Reviewed by:

Ignacio García-González, Universidade de Santiago de Compostela, Spain Rebecca Darbyshire, University of Melbourne, Australia

> \*Correspondence: Jean-Michel Legave legave@supagro.inra.fr

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 07 July 2015 Accepted: 12 November 2015 Published: 15 December 2015

#### Citation:

Legave J-M, Guédon Y, Malagi G, El Yaacoubi A and Bonhomme M (2015) Differentiated Responses of Apple Tree Floral Phenology to Global Warming in Contrasting Climatic Regions. Front. Plant Sci. 6:1054. doi: 10.3389/fpls.2015.01054 The responses of flowering phenology to temperature increases in temperate fruit trees have rarely been investigated in contrasting climatic regions. This is an appropriate framework for highlighting varying responses to diverse warming contexts, which would potentially combine chill accumulation (CA) declines and heat accumulation (HA) increases. To examine this issue, a data set was constituted in apple tree from flowering dates collected for two phenological stages of three cultivars in seven climate-contrasting temperate regions of Western Europe and in three mild regions, one in Northern Morocco and two in Southern Brazil. Multiple change-point models were applied to flowering date series, as well as to corresponding series of mean temperature during two successive periods, respectively determining for the fulfillment of chill and heat requirements. A new overview in space and time of flowering date changes was provided in apple tree highlighting not only flowering date advances as in previous studies but also stationary flowering date series. At global scale, differentiated flowering time patterns result from varying interactions between contrasting thermal determinisms of flowering dates and contrasting warming contexts. This may explain flowering date advances in most of European regions and in Morocco vs. stationary flowering date series in the Brazilian regions. A notable exception in Europe was found in the French Mediterranean region where the flowering date series was stationary. While the flowering duration series were stationary whatever the region, the flowering durations were far longer in mild regions compared to temperate regions. Our findings suggest a new warming vulnerability in temperate Mediterranean regions, which could shift toward responding more to chill decline and consequently experience late and extended flowering under future warming scenarios.

Keywords: fruit tree, flowering, chill period, heat period, warming vulnerability, multiple change-point models

## INTRODUCTION

Phenological events are highly responsive to temperature (Menzel and Fabian, 1999) and the abundance of information on plant phenology outlined substantial responses to global warming (Rutishauser et al., 2009). Most studies focused on bud phenology in natural vegetation and exhibited flowering advances as a main warming response (Abu-asab et al., 2001). Concerning fruit trees, flowering advances were highlighted in the European warming context for apple, pear and cherry trees (Chmielewski et al., 2004; Guédon and Legave, 2008; Eccel et al., 2009), hazelnut tree (Crepinšek ˇ et al., 2012), and olive tree (Garcia-Mozo et al., 2009). This was also observed in the Northeastern American context for apple tree (Wolfe et al., 2005) and in various parts of Asia for cherry tree (Miller-Rushing et al., 2007), apple tree (Fujisawa and Kobayashi, 2010), chestnut tree (Guo et al., 2013), and citrus species (Fitchett et al., 2014). Similar flowering advances have been scarcely reported in the Southern Hemisphere for apple and pear trees (Grab and Craparo, 2011).

Moreover, several studies dealing with warming responses in various perennial plants grown in temperate conditions found stationary or delayed bud phenology, in spite of temperature increases (Doi and Katano, 2008; Gordo and Sanz, 2009; Schwartz and Hanes, 2010; Yu et al., 2010). In fact, the timing of flowering is controlled by multiple and complex determinisms related to temperature at different periods of the year (Cook et al., 2012; Guo et al., 2013). Most temperate trees, including fruit species, are dormant in autumn and winter. Since the work of Lang et al. (1987), it has been widely accepted that among the different phases of bud dormancy, endodormancy corresponds to the growth suspension of the meristematic activity. The dormant buds require exposure to chill temperatures in order to overcome the endodormancy phase, followed by exposure to heat temperatures to resume growth during an ecodormancy phase and to initiate flowering in spring (Campoy et al., 2011). One likely warming impact during endodormancy is a delay in the fulfillment of chill requirements and consequently a delay in the time at which perennial plants become receptive to heat temperatures (Yu et al., 2010; Luedeling et al., 2013). This may explain unexpected phenological changes like those observed in walnut trees grown in California for which the vegetative buds (high chill requirements) shifted to late leaf-out since 1994 (Pope et al., 2013). Inversely, the flowering advances, that have dominated climate-warming responses thus far, were explained by increasing temperatures during ecodormancy leading to a more rapid fulfillment of heat requirements, as shown for apple trees in Europe (Legave et al., 2013) and in Japan (Fujisawa and Kobayashi, 2010). A comprehensive assessment of divergent responses to warming in temperate perennial plants must thus include the potential impacts on the fulfillment of both chill and heat requirements (Schwartz and Hanes, 2010). The sequential chill-growth model was therefore commonly used for analyzing flowering times in temperate fruit trees (Eccel et al., 2009; Darbyshire et al., 2014). When the fulfillment of chill requirements is inadequate, as is currently the case in mild climates, a typical symptom is the extended duration of the flowering phase (Atkinson et al., 2013). However, less attention has been paid to change in flowering duration in response to climate warming (Miller-Rushing et al., 2007; Legave et al., 2013). Moreover, in the case of fruit trees, nearly all the studies have reported warming responses in only one location or a few locations submitted to similar climatic contexts (Chmielewski et al., 2004; Fujisawa and Kobayashi, 2010; Grab and Craparo, 2011; Crepinšek et al., 2012 ˇ ), whereas it has been demonstrated that a given species can have contrasting responses in different locations (Primack et al., 2009). As an illustration, a large spatially-distributed lilac data set in North America demonstrated that the floral phenology has progressively changed from advances in flowering in northern regions to delays in flowering in southern regions (Zhang et al., 2007). In fact, there is evidence that more field studies are needed to determine the extent to which phenological shifts are occurring on large geographical scales (Primack et al., 2009).

Another key question is the use of appropriate statistical methods for analyzing flowering date and temperature series. The statistical analysis of such series is not standardized and various methods were used including linear regression (Fujisawa and Kobayashi, 2010; Grab and Craparo, 2011), multiple changepoint models (Guédon and Legave, 2008) and segmented regression models (Pope et al., 2013). Compared to our previous study (Guédon and Legave, 2008), we extended in this study the statistical modeling framework in order to test not only piecewise constant models but also piecewise linear models that include simple linear regression models when no change point can be detected. We were thus able in this way to identify both abrupt changes and linear trends in phenological series.

Our objectives here were (i) to propose a statistical modeling framework for analyzing flowering date and temperature series with minimum a priori assumptions (ii) to identify on this basis differentiated flowering changes on a large geographical scale in apple tree and (iii) to understand how changing temperature conditions can lead to differentiated flowering changes. These complementary objectives included changes both of the flowering time and the flowering duration. Apple tree offers a relevant study plant because of its worldwide cultivation and relatively high chill requirements (Hauagge and Cummins, 1991a; Ghariani and Stebbins, 1994) which can result in divergent responses to change in temperature conditions (Schwartz and Hanes, 2010).

## MATERIALS AND METHODS

#### Flowering and Temperature Data Collection of Flowering Date Series

A collaborative international network on apple tree phenology has been established between research institutes in six countries. We selected 10 locations, seven in Western Europe, one in Northern Morocco and two in Southern Brazil (**Table 1**; **Figure 1**). The eight locations in the Northern Hemisphere are located across a large latitudinal range (from 34 to 50◦N) with a corresponding large range of climatic conditions during the dormancy and flowering phases, from a cold continental climate in Europe (Bonn, Gembloux, Conthey, Trento) to a mild climate in Northern Morocco (Ain Taoujdate). This includes


#### TABLE 1 | Description of the collected data.

European locations with intermediate climates such as oceanic (Angers) and Mediterranean (Forli, Nîmes). While situated at high elevation to favor apple cropping, the two Brazilian locations in the Southern Hemisphere are clearly characterized by mild climates during the dormancy and flowering phases (mean temperature up to 11◦C).

Within this extensive geographical area, flowering dates were recorded for the beginning of the bloom phase (∼10% of flowers open) and the full bloom (∼50% of flowers open, first petals may have fallen). These dates correspond to stages 61 and 65 of the international BBCH code, respectively. Experienced observers recorded them using similar observation procedures on adult trees grown in long-term orchards. At each location, the flowering dates were assessed at least twice weekly on several trees of a given cultivar. In the mild conditions of Morocco and Brazil where flowering duration is extended (see Results) and flowering intensity is frequently weak due to floral abortions (Oukabli et al., 2003; Petri and Leite, 2004), the observers were trained to collect accurate data for comparison with those collected in temperate conditions. New trees were observed periodically at all locations as trees aged, whereas new observers were trained by the preceding ones.

To compare long-term flowering series between different locations, we chose cultivars grown worldwide. We therefore collected numerous data for Golden Delicious for which records were available at all 10 locations. In addition, records for Gala and Fuji were collected since these cultivars were frequently grown in Southern Brazil, but also in Europe. These three cultivars were characterized by nearly the same high chill requirements (Hauagge and Cummins, 1991a) and concomitant flowering times both in Southern Brazil and Europe (cross pollination in orchard). The collection of different varietal series in a given location was thus considered as a way to repeat the statistical

in abscissa and ordinate.

analysis to reveal a strong phenological change in the location, and not as a way to study the genotype × location interactions. Our data set consists of 30 flowering date series including series for the stages 61 and 65 (16 of them corresponded to the temperate conditions and 14 to the mild conditions). Each series was defined by a location, a cultivar and a flowering stage, including a total of 1121 measurements. Most series were complete aside from some missing data (not interpolated) in some series. The longest series contains 56 years in Bonn (Golden Delicious, stages 61 and 65) and the shortest contains 25 years in Caçador (Golden Delicious, stages 61 and 65; **Table 1**). The consistency of collected data was assessed by the fact that the flowering dates were consistently related to the geographical characteristics (latitude, elevation) and temperature conditions of the locations. Moreover, the flowering duration between the dates of stages 61 and 65 was assessed at all locations where the two dates were recorded. This included 13 series of flowering durations (six for the temperate conditions and seven for the mild conditions) ranging from 56 years in Bonn (Golden Delicious) to 25 years in Caçador (Golden Delicious).

#### Collection of Temperature Series

For characterizing the relationships between flowering and temperatures, we analyzed series of mean temperatures during two successive periods respectively determining for the fulfillment of chill and heat requirements. Annual chill accumulation (CA) period and subsequent heat accumulation (HA) period have thus been defined. Based on previous results concerning the bud dormancy dynamics (Malagi et al., 2015) and the relationships between flowering and temperatures (Legave et al., 2013; El Yaacoubi et al., 2014) in apple tree, the CA period ranged from October to January for the European and Moroccan locations (Northern Hemisphere) and from April to July for the Brazilian locations (Southern Hemisphere). The HA period ranged from February to April for the European locations and from August to October for the Brazilian locations. We chose a shorter HA period for the Moroccan location (March to mid-April), because previous works using Partial Least Squares regression clearly suggested this period as a major period of heat requirement fulfillment in Morocco (El Yaacoubi et al., 2014).

The mean temperature series were constituted from the average minimum and maximum daily temperatures collected from weather stations located near the orchards where the flowering dates were recorded (no more than 10 km). The daily temperatures were provided for each location and checked by the corresponding research institute. The French partner performed a complementary global check for this study. The few missing data were estimated by linear interpolation. All the series started before the end of the 1980s, the instant at which marked increases of temperature have been frequently recorded at the world scale, particularly in Europe (Jones and Moberg, 2003). When temperature series longer than the corresponding flowering date series were available, we collected the longest possible temperature series; this was the case for Gembloux, Nîmes, Ain Taoujdate, Caçador, and Sao Joaquim (**Table 1**).

#### Statistical Modeling

#### Definition of Piecewise Constant and Piecewise Linear Models

Multiple change-point models were used to delimit segments within a flowering date or temperature series of length T, for which the data characteristics were homogeneous within each segment while markedly differing from one segment to another. We made the assumption of homoscedastic Gaussian multiple change-point models, either piecewise constant or piecewise linear models. In the first case, the slope is assumed to be zero and the only within-segment parameter is the intercept (which is also the segment mean in this case) whereas in the second case, the within-segment parameters are the intercept and the slope. In both cases, the variance is assumed to be common to the segments. This homoscedasticity assumption is justified by the data characteristics but also by the fact that the series were rather short (between 25 and 56 years). The two associated models are denoted by Mconstant (for piecewise constant) and Mlinear (for piecewise linear). Piecewise linear models are somewhat related to the segmented regression models used by Pope et al. (2013). Segmented regression or brokenline models are regression models where the regression function is piecewise linear, i.e., made of straight lines connected at change points (Muggeo, 2003). The regression function is thus continuous, but first derivatives are discontinuous. In our case, the regression function is not constrained to be continuous.

For the Mconstant model, we suppose that some J − 1 instants τ<sup>1</sup> < · · · < τ<sup>J</sup> <sup>−</sup> <sup>1</sup> (with the convention τ<sup>0</sup> = 0 and τ<sup>J</sup> = T) exist such that the mean is constant between two successive change points and the variance is assumed to be constant,

$$\text{if } \mathfrak{r}\_{\mathfrak{f}} \le t < \mathfrak{r}\_{\mathfrak{f}+1}, \quad \begin{cases} \begin{array}{l} E(X\_{\mathfrak{f}}) = \alpha\_{\mathfrak{f}}, \\ \text{Var}(X\_{\mathfrak{f}}) = \sigma^2. \end{array} \end{cases}$$

These two families of models enable to test and combine two assumptions: change point of sufficient amplitude separating two phases and linear trend (within phase or for the whole series in the case of no change point).

We adopted a retrospective or off-line inference approach whose objective was to infer the number of segments J, the instants of the J − 1 change points τ1, . . . , τ<sup>J</sup> <sup>−</sup> <sup>1</sup>, the J withinsegment intercepts α<sup>j</sup> , the global variance σ 2 and the J withinsegment slopes β<sup>j</sup> (for Mlinear model). For the selection of the number of segments J, we used the modified Bayesian information criterion (mBIC) proposed by Zhang and Siegmund (2007) and specifically dedicated to Gaussian homoscedastic multiple change-point models. The principle of this kind of penalized likelihood criterion consists in making a trade-off between an adequate fitting of the model to the data and a reasonable number of parameters to be estimated. Jeffreys' rules of thumb (Kass and Raftery, 1995) suggest that a difference of mBIC of at least 2 log(100) = 9.2 is needed to deem the model with the higher mBIC substantially better. For the optimal segmentation of the series into J segments, we applied the dynamic programming algorithm proposed by Auger and Lawrence (1989). This optimal segmentation defines the optimal change points and relies on the estimation of within-segment and global variance parameters; see details on these statistical methods for multiple change-point models in Supplementary Material, Appendix S1.

#### Comparison Between the Selected Piecewise Constant Model and the Selected Piecewise Linear Model

For many flowering date series, we obtained two models that were not discernible according to mBIC: the 2-segment piecewise constant model and the simple linear model (i.e., 1 segment piecewise linear model). This situation is illustrated by the Forli series (**Figure 2A**) for which the difference of mBIC is <1. This can be explained by the similar orders of magnitude for the change-point amplitude and the global standard deviation in the case of the 2-segment piecewise constant model. We thus extracted the residual series from the linear function and we found that the residual series was not stationary but that a change point can be identified in 1988 (this was the change point of the selected 2-segment piecewise constant model), between two increasing linear trends (for this, we selected the best piecewise linear model for the residual series using mBIC; **Figure 2B**). The Ain Taoujdate series illustrates another situation where the 2-segment piecewise constant model can be identified using piecewise linear models (the selected model in this family was a 2-segment model and the two estimated slopes were not significantly different from 0; **Figure 3**). Finally, the Sao Joaquim series for Golden Delicious illustrates the case of very short segments at one end of the series (**Figure 4**). In this situation, we chose to not consider these very

Segmentation of the residual series deduced from the estimated linear model

short segments that cannot be interpreted in our application context.

#### Assessment of the Segmentation Assumption

It is often of interest to quantify the uncertainty concerning change point instants. Let <sup>L</sup>J(**s**, **<sup>x</sup>**; <sup>θ</sup>ˆ) denote the likelihood of the segmentation **s** in J segments of the observed series **x** where θ denotes the set of within-segment and global variance parameters. In the case of a single change point (J = 2), the posterior probability of entering the second segment at τ<sup>1</sup> is given by:

$$L\_2(\mathbf{s}(\pi\_1), \mathbf{x}; \hat{\theta}) / \sum\_{\mathbf{s}} L\_2(\mathbf{s}, \mathbf{x}; \hat{\theta}),$$

where each segmentation **s** defines a unique change point. In our case of short series, the dynamic programming algorithm for computing the top N most probable segmentations proposed by Guédon (2013) was used to compute the T − 1 possible segmentations and the associated likelihood and then to extract the change-point posterior distribution. In this particular case of a single change point, this posterior distribution therefore summarizes the possible segmentations. In particular, the posterior probability of the optimal segmentation **s** ∗ given by:

FIGURE 3 | Segmentation of the Ain Taoujdate BBCH 61 stage date series using a 2-segment piecewise constant model and a 2-segment piecewise linear model.

using a 2-segment piecewise linear model.

$$P(\mathbf{s}^\* | \mathbf{x}; 2) = L\_2(\mathbf{s}^\*, \mathbf{x}; \hat{\theta}) / \sum\_{\mathbf{s}} L\_2(\mathbf{s}, \mathbf{x}; \hat{\theta}),$$

which is the mode of the change-point posterior distribution, can be used to assess the segmentation assumption.

More generally, the posterior probability of the optimal segmentation given by:

$$P(\mathbf{s}^\* | \mathbf{x}; J) = L\_J(\mathbf{s}^\*, \mathbf{x}; \hat{\theta}) / \sum\_{\mathbf{s}} L\_J(\mathbf{s}, \mathbf{x}; \hat{\theta}),$$

can be computed using the dynamic programming algorithm for computing the top N most probable segmentations in our case of short series segmented into a few segments (up to J = 3). The assessment of multiple change-point models thus relies on two posterior probabilities:


#### RESULTS

In this study, we systematically favored longitudinal analyses of the various series (flowering dates and durations, mean temperatures during the CA and HA periods) in order to identify phenological patterns with minimum a priori assumptions. We also chose to not build simple regression models on the basis of these longitudinal data since this would rely on an oversimplified view of the influence of the temperatures on the flowering process regarding the biological bases and current functional models of bud phenology (Cook et al., 2012; Guo et al., 2013; Darbyshire et al., 2014; Pope et al., 2014).

#### Flowering Time

The dates of stages 61 and 65 appeared highly correlated, meaning that flowering durations fluctuate around quite constant values for a given series and that the changes in dates of stages 61 and 65 are markedly synchronous for a given series (see Section Flowering Duration). We thus focused the analysis on the 15 series of stage 61 dates for which data were collected at nine locations. We also analyzed the 2 series of stage 65 dates in Conthey for which only stage 65 dates were collected over long periods (44 and 39 years for Golden Delicious and Gala, respectively; **Table 1**).

Combining model selection criterion (mBIC) and residual analysis in the case of piecewise linear models, we found that the assumption of a piecewise constant model was better supported than the assumption of a piecewise linear model. In order to ease comparison between locations and cultivars, we chose to focus on 2-segment piecewise constant models. This corresponds to the models selected by mBIC for seven flowering date series: Angers (Golden Delicious), Forli (Golden Delicious), Trento (Golden Delicious), Gembloux (Golden Delicious), Conthey (Golden Delicious and Gala), and Ain Taoujdate (Golden Delicious). This was a well-supported alternative model for four other flowering date series: Nîmes (Gala and Fuji), Caçador (Golden Delicious and Gala) according to the posterior probability of the 2-segment model (**Table 2**). It should be noted that in our context of short series (length between 25 and 56), the number of segments given by mBIC should only be considered as indicative. We chose to discard 2-segment piecewise constant models selected by mBIC for Sao Joaquim (Golden Delicious—**Figure 4**—and Gala) since this corresponds to very short segments at the beginning of the series (2 and 1 years respectively) that cannot be reliably interpreted in our context (**Table 3**). Two-segment piecewise constant models are well-defined if the single change point of sufficient amplitude with respect to the global segment standard deviation separates two sufficiently long segments. It should be noted that the 3-segment model selected by mBIC for Bonn includes a short 4-year segment (between 1958 and 1961) at the beginning of the series. Since this range of years was not represented in other series, it was difficult to interpret this first segment. The 2-segment model retained for comparison of locations (**Table 2**) was simply this optimal 3-segment model where the first two segments were merged (**Table 3**; Figure S1 in Supplementary Material). No change point can be detected for Sao Joaquim (Fuji) and Caçador (Fuji). In the case of 2 segment models the change point is located between 1987 and 1989 for most of the flowering date series, which is consistent with our previous analyzes (Guédon and Legave, 2008), but with the notable exceptions of Ain Taoujdate (change point in 1994; **Table 2**). For the flowering series starting at the beginning of the 1980s with a change point detected at the end of the 1980s, Nîmes (Fuji), Trento (Golden Delicious), Gembloux (Golden Delicious), Caçador (Golden Delicious, Gala), the rather short length of the first segment (between 3 and 6 years) makes the mean estimated for this segment less reliable and, consequently, the change-point amplitude. This explains the difference in change-point amplitude for Nîmes between Fuji and the other two cultivars, as well as the difference between Gembloux (Golden Delicious) and Bonn (Golden Delicious; **Table 2**) for which the climatic conditions were rather similar (**Table 1**).

For each flowering date series, the uncertainty concerning the instant of the change point is low for most locations (**Figure 5**) except for Nîmes (Golden Delicious) for which the posterior probability of the segmentation is the lowest among the segmentation in 2-segments (**Table 2**). Moreover, this series is the only one for which the change-point amplitude is lower than the global standard deviation in the case of a 2-segment piecewise constant model (**Table 2**). Hence, the segmentation in 2-segments is not well defined in this case. This can be illustrated by the segmentation in 2- and 3-segments of this flowering date series where the segmentation in 3-segments highlights a change toward later flowering dates since 2003 (**Figure 6**). It should be noted that in the case of Nîmes, the mBIC favors the constant model (i.e., no change point) regardless of the cultivar (**Tables 2**, **3**).

TABLE 2 | Segmentations of flowering date series (BBCH 61 stage for all locations except Conthey—BBCH 65 stage) using piecewise constant models (2 or 1 segment when the 2-segment model was irrelevant): observation period, change-point instant and amplitude, global standard deviation, optimal segmentation posterior probability, model posterior probability, mBIC model, average flowering duration, correlation coefficient between BBCH 61 and 65 stage dates.


In the model posterior probability column, an asterisk indicates that the model is the one given by mBIC. If this is not the case, the model given by mBIC is indicated in the next column (mBIC model).

TABLE 3 | Optimal segmentations of flowering date series (BBCH 61 stage) using piecewise constant models for series where the optimal J-segment model according to the mBIC was not retained for the comparison in Table 2: observation period, change-point instant and amplitude, global standard deviation, optimal segmentation posterior probability, model posterior probability.


#### Flowering Duration

We did not detect any change point or linear trend in the flowering duration series and we thus analyzed them as simple frequency distributions without considering the year indexing (**Table 2**). We first grouped samples corresponding to the different cultivars in a given location (Golden Delicious, Gala and Fuji for Nîmes, Sao Joaquim and Caçador, respectively) for which the frequency distributions were not significantly different according to the Kruskal Wallis test (ANOVA by ranks for these frequency distributions defined on a small set of values). The cumulative frequency distribution functions of the seven samples (**Figure 7**) highlights a clear order for flowering duration (from the shortest to the longest): (1) Nîmes, 2.82 days on average; (2) Forli, Trento and Bonn, between 3.6 and 4.45 days (these three samples are not significantly different according to the Kruskal Wallis test); (3) Sao Joaquim, 9.18 days; (4) Caçador, 11.82 days; (5) Ain Taoujdate, 14.3 days.

## Temperature During the CA and HA Periods vs. Flowering Date and Duration

The flowering date was not significantly correlated with the mean temperature during the CA period (defined for the Northern Hemisphere) for most of the European locations (**Table 4**). The only exceptions were Trento and Conthey for which we found slightly significant negative correlations. The situation was very different for the Brazilian locations for which we found strongly

FIGURE 5 | Two-segment piecewise constant models estimated on the basis of BBCH 61 stage date series for Angers (Golden Delicious), Nîmes (Golden Delicious, Gala, and Fuji), Forli (Golden Delicious), Trento (Golden Delicious), Gembloux (Golden Delicious), Bonn (Golden Delicious), Ain Taoujdate (Golden Delicious), and Caçador (Golden Delicious and Gala) and BBCH 65 stage date series for Conthey (Golden Delicious and Gala): posterior change-point distributions.

significant positive correlations (except for Caçador, cultivar Fuji) between the flowering date and the mean temperature during the CA period (defined for the Southern Hemisphere), which means that the warmer the austral CA period is, the later the flowering date will be. The Moroccan situation seems to be closer to the Brazilian situations than to the European ones but we cannot be conclusive in this case because the correlation coefficient was not significantly different from 0 (**Table 4**). We thus conducted a longitudinal analysis of the series of mean temperatures during the CA period using the methodology previously applied to the flowering date series. We found various patterns:

• Two stationary segments for Nîmes (1974–2013, change point in 1988; **Figure 8A**) and Forli (1970–2013, change point in 1993), but with a rather small change-point amplitude

with respect to the residual standard deviation in this latter case.


We found strongly significant negative correlations for the European locations between the flowering date and the mean temperature during the HA period (defined for Europe), which means that the warmer the Northern HA period is, the earlier the flowering date will be (**Table 5**). The situation was very different for the Brazilian locations for which the flowering date was not significantly correlated with the mean temperature during the HA period defined in Brazil (**Table 5**). The only exception was Sao Joaquim, cultivar Golden Delicious, for which we found a slightly significant negative correlation. For the Moroccan location we found a significant negative correlation between the flowering date and the mean temperature during the HA period. This relationship appears to be closer to the relationships found for the European locations than to the ones found for the Brazilian locations (**Table 5**). Because of the significant correlations between the flowering date and the mean temperature during the HA period found for the European and Moroccan locations, we conducted a longitudinal analysis of the series of mean temperatures during this period (**Table 5**) using the methodology previously applied to the flowering date series. We found a change point at the end of the 1980s in all the European temperature series when applying the piecewise constant model (see an illustration in **Figure 8B** for Nîmes) and the change-point amplitude was around 1.3◦C for most locations (it should be noted that for Gembloux, the change-point amplitude estimated on the complete series up to 1966 is far more reliable than the one estimated on the series corresponding to the flowering date range of years since, in this latter case, the first segment was very short—four years). We found a change point of similar amplitude but in 1993 in the Moroccan location TABLE 4 | Correlation coefficients between flowering date (BBCH 61 stage for all locations except Conthey—BBCH 65 stage) and mean temperature during the CA period (\*\*significant at 1% level; \*significant at 5% level; n.s., non-significant).


Segmentation of the Nîmes HA period temperature series using a 2-segment piecewise constant model and a 2-segment piecewise linear model.

when applying a piecewise constant model (**Table 5**), although the assumption of a simple linear model was better supported (**Figure 9**). In contrast, the series of mean temperatures during the HA period were stationary for the Brazilian locations (**Table 5**).

We also found non-significant correlations between the flowering durations and the mean temperatures, both during the CA and HA periods at all the seven locations where these correlations were analyzed (especially in Morocco and Brazil; results not shown).

#### DISCUSSION

#### Differentiated Flowering Date Series

The analyses showed clear advances of apple flowering time since the end of the 1980s in most locations of Western Europe and a few years later in Northern Morocco, whereas the flowering dates remained stationary in Southern Brazil. Our study thus provides a new overview in space and time of the flowering time changes in apple tree highlighting contrasting behaviors (advance or stationarity) contrary to previous studies that only reported flowering advances (Chmielewski et al., 2004; Wolfe et al., 2005; Guédon and Legave, 2008; Eccel et al., 2009; Fujisawa and Kobayashi, 2010; Grab and Craparo, 2011; Darbyshire et al., 2013). Nevertheless, the flowering advances have been found to a lesser amplitude in the Southern Hemisphere (Australia, Southern Africa) than in the Northern Hemisphere (Europe, Japan, United States) as outlined by Darbyshire et al. (2013). Such lesser changes in flowering date in the Southern Hemisphere are consistent with the stationary series mainly found in Southern Brazil.

While there was a strong consensus concerning abrupt flowering advances in Europe, a notable exception was the French Mediterranean location (Nîmes) for which the flowering dates were stationary. This was clearly demonstrated for Golden Delicious for which the flowering advance reported by Guédon and Legave (2008) up to 2002 became an alternative not strongly supported when records up to 2013 were included in the series.

When all our findings are taken into account, the overview of flowering date changes reveals that flowering advances or stationary flowering dates may be detected in temperate as well as in mild regions: (i) advances for most European series, the Moroccan series and one Brazilian series; and (ii) stationarity for the French Mediterranean series and most Brazilian series. A simple relationship between the flowering date change pattern and a geographical trait such as the elevation or the localization in the Northern or Southern Hemisphere is no longer valid.

#### Relationships Between Temperatures and Flowering Dates

The correlation analysis gave evidence of contrasting climatic determinisms of flowering time in relation to the localization in either temperate or mild regions. In Western Europe, our analysis showed that the higher the HA is the earlier the flowering date will be, whereas the CA was not related to the flowering time. These results emphasize that flowering time in recent past have been strongly determined by the HA in the temperate conditions of Western Europe. This was similarly showed for apple tree in other temperate conditions like those of Northeastern America and Japan (Wolfe et al., 2005; Fujisawa and Kobayashi, 2010).

In Southern Brazil, a main influence of CA on flowering time was emphasized since the correlation analysis mainly showed that the lower the CA is the later the flowering date will be, as would be expected in mild conditions (Atkinson et al., 2013). Although a similar determinism of flowering time was highlighted in Northern Morocco (Oukabli et al., 2003), this appeared less evident on the basis of our correlation analysis. That may be explained by an additional influence of the HA, which was shown TABLE 5 | Segmentations of series of mean temperatures during the HA period using piecewise constant models (2 or 1 segment when the 2-segment model was irrelevant): recording period, change-point instant and amplitude, global standard deviation, optimal segmentation posterior probability, model posterior probability, mBIC model, correlation coefficients between flowering date (BBCH 61 stage for all locations except Conthey–BBCH 65 stage) and mean temperature during the HA period (\*\*significant at 1% level; \*significant at 5% level; n.s., non-significant).


In the model posterior probability column, an asterisk indicates that the model is the one given by mBIC. If this is not the case, the model given by mBIC is indicated in the next column (mBIC model). We analyzed the mean temperatures during the HA period over the range of years corresponding to the flowering dates and in certain cases (Nîmes, Gembloux, Ain Taoujdate, Sao Joaquim, Caçador) over an extended range of years.

†Gembloux (1964–2013): Posterior probability of 0.29 for 1989 but of 0.27 for 1988.

by the correlation analysis in Northern Morocco. Nevertheless, our results globally showed that flowering time would be mainly determined by the CA in mild conditions of both Southern Brazil and Northern Morocco. Such determinism is consistent with the high genetic heritabilities of the chill requirement trait that were found from apple progenies grown in mild conditions like those of Southern Africa (Labuschagné et al., 2002). In addition severe symptoms of inadequate chill requirements, including delays of floral budbreak and flowering time were observed in high-chill genotypes selected from Golden Delicious progenies. Therefore, it was generally accepted in apple tree that budbreak time in mild regions was an accurate biomarker of the fulfillment of chill requirements (Hauagge and Cummins, 1991b). This supposes a short time between the fulfillment of chill requirements and floral budbreak in mild conditions, which was demonstrated in Southern Brazil by comparison with the temperate conditions of Southern France (Malagi et al., 2015).

The analysis of temperature series accounts for contrasting warming contexts in relation to the localization in the Northern or Southern Hemisphere. In Western Europe, a marked warming during the HA period was detected at the end of the 1980s, similarly to the change-point instant generally found for the flowering date. Such a concomitance was also detected in Northern Morocco at the beginning of the 1990s. Heat increase can thus explain flowering advances in Western Europe where the HA would mainly determine flowering time and in Northern Morocco where the influence of HA would be also involved. In Southern Brazil, the stationarity of temperatures during both the CA and HA periods can explain the stationarity of most flowering date series. The absence of warming in Southern Brazil over the last four decades could be attributed to the relatively high elevation of the studied locations and to the globally lower temperature increases since the 1970s in the Southern Hemisphere compared to the Northern Hemisphere (Jones and Moberg, 2003).

As a consequence of these relationships, flowering date advances and stationary flowering dates result from interactions between contrasting thermal determinisms (temperate vs. mild regions) and warming contexts (Northern vs. Southern Hemisphere). A particularly interesting result is the stationarity of flowering dates both in the Brazilian and the French Mediterranean regions. In Southern Brazil no thermal influences could have been exercised in the absence of warming, whereas the French Mediterranean region would have been progressively submitted to two opposite thermal influences: an increase in HA compensated by a decrease in CA, respectively linked to marked warming during the HA and CA periods. While less marked, the decrease of CA tended to delay the flowering dates. This appears mainly due to certain years since 1988, including 1988, 1996, 2007, and 2012, that were characterized by relatively high temperatures during the CA period (**Figure 8A**). This may explain the absence of correlation up to 2013 between the flowering date and the temperature during the CA period in this Mediterranean region.

Stationary flowering dates resulting from two opposite warming impacts, such as in the French Mediterranean region, may thus cast doubt on the relevance of flowering advances in apple tree as a suitable indicator of recent warming in early spring in Europe (Menzel et al., 2011). This indicator would be even more questionable in the near future in the entire Mediterranean region, as suggested by a similar temperature increase during the CA period since 1993 in the Italian Mediterranean region of Forli near the Adriatic coastline (see Section Temperature During the CA and HA Periods vs. Flowering Date and Duration).

There are still many unknowns concerning the relationships between the flowering dates and temperatures from bud dormancy to flowering time. This requires new researches on the timings of fulfillment of the chill and heat requirements and better knowledge of temperature thresholds on flowering time (Guo et al., 2013; Luedeling et al., 2013; Pope et al., 2014).

#### Specificity of the Flowering Duration

As expected (Atkinson et al., 2013), our analyses highlighted contrasting flowering duration patterns between temperate (short durations) and mild conditions (far longer durations). An additional specificity of flowering duration patterns is the lack of detection of a change point or linear trend at all locations, contrary to the flowering time patterns. This suggests different climatic determinisms for the flowering duration and the flowering time. In temperate conditions, a likely climatic determinism of flowering duration is the HA rate during the blooming phase, which is consistent with a significant increase in the average flowering duration from the Mediterranean location of Nîmes (2.8 days) to the colder continental location of Bonn (4.5 days). In mild conditions inversely, longer durations of the blooming phase are currently attributed to insufficient CAs (Campoy et al., 2011). Nevertheless, our results suggest more complex thermal determinisms to explain extended flowering durations in mild conditions since non-significant correlations were found between the flowering duration and the mean temperature during the CA period as well as during the HA period in both the Moroccan and Brazilian locations (see Section Temperature During the CA and HA Periods vs. Flowering Date and Duration).

#### Strengths of the Statistical Modeling Approach

Homoscedastic 2-segment piecewise constant models, where the variance was assumed to be common to the two segments, played a key role in this study. This homoscedasticity assumption can be assessed by examining the standard deviations empirically estimated for the two segments in the case of sufficiently long flowering date or series of mean temperatures during the HA period (see Supplementary Materials, Tables S1, S2). In our context of potentially short segments, homoscedastic models are more robust than heteroscedastic models where changes in variance can induce artifactual short segments. One shortcoming of homoscedastic models with respect to heteroscedastic models used in our previous study (Guédon and Legave, 2008) was the inability to compute posterior change-point distributions. This is no longer the case since, for a single change point, the posterior change-point distribution can be computed using the dynamic programming algorithm for computing the top N most probable segmentations proposed by Guédon (2013). More generally, the posterior probability of the optimal segmentation can be computed using this dynamic programming algorithm in our case of short series segmented in a few segments. The other main methodological improvement consists in the systematic comparison of homoscedastic piecewise constant models with homoscedastic piecewise linear models. The fact that simple linear models corresponding to the linear trend assumption were systematically compared with piecewise constant models and the ability to identify piecewise constant models using piecewise linear models (see the Ain Taoujdate example in **Figure 3**) strengthen the reliability of our results with respect to our previous study.

#### Implications of Flowering Changes on Climatic Adaptation

Early flowering dates might increase the risk of spring frost damages, as pointed out for a long time by Cannell and Smith (1986) in Great Britain. More recently, spring frosts in southern regions of Great Britain have been seen to be decreasing both in frequency and severity (Sunley et al., 2006). Likewise, investigations into the frost risk for apple tree in Northern Italy (Eccel et al., 2009) showed that the risk was lower than in the past, and suggested that it will remain stable or decrease slightly in the future. Thus, the frost risk during the fruit tree flowering phase in Europe will probably be more open to debate in the future context of climate warming because of regional differences in both the magnitude of flowering advances and the frequency of negative temperatures. In particular, our results suggest that the French Mediterranean region may be rarely subject to frost risk at present time because flowering advance is decreasing. Inversely, this risk might remain a true concern for growers in continental regions because of constant flowering advances and also relatively short flowering durations (as shown in Bonn in Germany).

Mediterranean and oceanic regions in Europe might be more affected in the future by excessive delays in chill fulfillment, particularly the French Mediterranean region where our study showed an increase in temperature from October to January since the end of the 1980s. Declining chill will become a limiting cropping factor (Atkinson et al., 2013) and a new warming vulnerability in many European Mediterranean regions, particularly for the apple tree characterized by high chill

requirements for most commercial cultivars. At a certain point, CA might shift from sub-optimal, just inducing a delayed phenology as illustrated in our study by the Nîmes location, to below the requirement leading to irregular or insufficient yields. At the orchard scale, this could cause in the future phenological disorders similar to those observed in the Mediterranean mild conditions of Northern Morocco (Oukabli et al., 2003), such as insufficient flowering synchronization between non self-fertile cultivars (all apple cultivars), which require cross-pollination, and excessive duration of the fruit maturity phase, that may be consequences of late and extended flowering.

#### CONCLUSION

The collection of flowering date series recorded in contrasting climatic locations and their analysis based on multiple changepoint models proved to be appropriate for identifying differentiated flowering patterns in apple tree. The same methodology applied to the corresponding temperature series provided complementary results, making possible to establish a comprehensive overview of the relationships between the flowering phenology and the warming context at the world scale. On the one hand, we showed that the different patterns of flowering time change are not closely related to the localization of apple trees in temperate or mild regions or in the Northern or Southern Hemisphere. On the other hand, contrasting temperature influences during successive CA and HA periods on flowering time appeared to be mainly in relation with the localization in temperate or mild conditions, while contrasting warming contexts appeared to be in relation with the localization in the Northern or Southern Hemisphere. This overview was completed by information on the relationships between flowering duration and temperature, which appeared to be different and more complex than those concerning the flowering time. Because continuous warming will change

#### REFERENCES


the relationships between phenology and temperature, a new warming vulnerability is expected in the more or less long term in Europe especially in Mediterranean regions where apple tree is a native crop.

#### AUTHOR CONTRIBUTIONS

JL conceived the study, conducted the data collection, contributed to the data analysis and edited the manuscript; YG contributed to the study conception, conducted the statistical analysis and edited the manuscript; GM and AE supplied data for Brazil and Morocco respectively, contributed to the data analysis and manuscript approval; MB contributed to the study conception, the data analysis, and manuscript approval.

#### ACKNOWLEDGMENTS

The authors thank INRA Angers (France), Vincent Mathieu (Ctifl, France), Daniela Giovannini, and Marco Fontanari (CRA-FRF, Italy), Danilo Christen (Agroscope C-W, Switzerland), Michael Blanke (INRES-Horticultural Science, Germany), Robert Oger and Viviane Planchon (CRA-W, Belgium), Ahmed Oukabli (INRA, Morocco), Frederico Denardi, and Gabriel Leite (Epagri, Brazil) for providing the data used in the analyzes. The authors also thank Pierre-Éric Lauri, Evelyne Costes (INRA, France), and Isabelle Farrera (Montpellier SupAgro, France) for their fruitful comments on the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2015. 01054


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Legave, Guédon, Malagi, El Yaacoubi and Bonhomme. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Rhizophoraceae Mangrove Saplings Use Hypocotyl and Leaf Water Storage Capacity to Cope with Soil Water Salinity Changes

Silvia Lechthaler1,2† , Elisabeth M. R. Robert2,3† , Nathalie Tonné2,3, Alena Prusova<sup>4</sup> , Edo Gerkema<sup>4</sup> , Henk Van As<sup>4</sup> , Nico Koedam<sup>2</sup> and Carel W. Windt<sup>5</sup> \*

<sup>1</sup> Department of Territorio e Sistemi Agro-Forestali, University of Padova, Padova, Italy, <sup>2</sup> Laboratory of Plant Biology and Nature Management, Department of Biology, Vrije Universiteit Brussel, Brussels, Belgium, <sup>3</sup> Laboratory of Wood Biology and Xylarium, Department of Wood Biology, Royal Museum for Central Africa, Tervuren, Belgium, <sup>4</sup> Laboratory of Biophysics and Wageningen NMR Centre, Department of Agrotechnology & Food Sciences, Wageningen University, Wageningen,

#### Edited by:

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### Reviewed by:

Jorge López-Portillo, Instituto de Ecología, A.C., Mexico Annie Deslauriers, Université du Québec à Chicoutimi, Canada

> \*Correspondence: Carel W. Windt c.windt@fz-juelich.de

†These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 16 March 2016 Accepted: 07 June 2016 Published: 27 June 2016

#### Citation:

Lechthaler S, Robert EMR, Tonné N, Prusova A, Gerkema E, Van As H, Koedam N and Windt CW (2016) Rhizophoraceae Mangrove Saplings Use Hypocotyl and Leaf Water Storage Capacity to Cope with Soil Water Salinity Changes. Front. Plant Sci. 7:895. doi: 10.3389/fpls.2016.00895 Netherlands, <sup>5</sup> IBG-2: Plant Sciences, Institute for Bio- and Geosciences, Forschungszentrum Jülich, Jülich, Germany Some of the most striking features of Rhizophoraceae mangrove saplings are their voluminous cylinder-shaped hypocotyls and thickened leaves. The hypocotyls are known to serve as floats during seed dispersal (hydrochory) and store nutrients that allow the seedling to root and settle. In this study we investigate to what degree the hypocotyls and leaves can serve as water reservoirs once seedlings have settled, helping the plant to buffer the rapid water potential changes that are typical for the mangrove environment. We exposed saplings of two Rhizophoraceae species to three levels of salinity (15, 30, and 0–5h, in that sequence) while non-invasively monitoring

changes in hypocotyl and leaf water content by means of mobile NMR sensors. As a proxy for water content, changes in hypocotyl diameter and leaf thickness were monitored by means of dendrometers. Hypocotyl diameter variations were also monitored in the field on a Rhizophora species. The saplings were able to buffer rapid rhizosphere salinity changes using water stored in hypocotyls and leaves, but the largest water storage capacity was found in the leaves. We conclude that in Rhizophora and Bruguiera the hypocotyl offers the bulk of water buffering capacity during the dispersal phase and directly after settlement when only few leaves are present. As saplings develop more leaves, the significance of the leaves as a water storage organ becomes larger than that of the hypocotyl.

Keywords: dendrometers, leaf thickness variations, magnetic resonance imaging (MRI), mangrove environment, mobile nuclear magnetic resonance (NMR)

**Abbreviations:** Bg, Bruguiera gymnorrhiza; CPMG, Carr Purcell Meiboom Gill; 1WC, delta water content; FOV, field of view; IDL, Interactive Data Language; LVDTs, linear variable displacement transducers; MRI, magnetic resonance imaging; NMR, nuclear magnetic resonance imaging; PAR, photosynthetically active radiation; RF, radio frequency; Rm, Rhizophora mucronata; SW, spectral width; TE, echo time; TR, repetition time; VPD, vapor pressure deficit; VWC, volumetric water content.

## INTRODUCTION

fpls-07-00895 June 24, 2016 Time: 17:1 # 2

Mangrove trees grow in one of the most challenging environments that vascular plants have been able to colonize. Due to intermittent tidal and fluvial inundation, irregular freshwater input, and salt accumulation by evaporation, mangrove trees do not only have to deal with extreme, but also with highly variable levels of soil water salinity (Krauss and Ball, 2012; Bompy et al., 2014b). In addition, they experience hypoxic soils, periodic partial or entire inundation, large changes in VPD, and high levels of solar radiation (Ball, 1988a; Tomlinson, 1994; Robert et al., 2009). Mangroves are able to exclude salt when taking up water from their substrate (Scholander, 1968; Khan and Aziz, 2001; Parida and Jha, 2010) and/or excrete it from their leaves (Wang et al., 2011). The exclusion of salt, however, is only partial and elevated salt concentrations are known to be tolerated in the cell apoplast (from 1 to 10% seawater salinity). Mangroves have been suggested to take advantage of elevated salt concentrations in the apoplast and symplast to decrease water potential, thus reducing the challenge to primary walls in vessels and to cell membranes (Scholander et al., 1966; Parida and Jha, 2010; Reef and Lovelock, 2015).

To be able to deal with these rapidly changing abiotic conditions, mangrove trees developed various means to store and buffer water. The ability to store water is not unique for salt tolerant plants; in many plants and trees it is associated with resistance to drought. The role of the stem as a water storage organ in trees has received considerable attention (Zweifel et al., 2000; Scholz et al., 2008; Carrasco et al., 2015; Morris et al., 2015; Pfautsch et al., 2015b). Various studies demonstrated that living tissues associated with the xylem are able to store water (paratracheal and apotracheal parenchyma, phloem, cambium, bark) and transport it axially (parenchyma rays), thus allowing the tree to buffer water potential changes in its abiotic environment (Spicer, 2014; Pfautsch et al., 2015a and references therein).

For mangrove trees, which usually have access to water but periodically may have difficulty to take it up because of its salinity, the most important benefit of water storage appears to be that it minimizes the carbon costs that are associated with water uptake, osmoregulation, and the upkeep of tissue salt tolerance. For example, the size of the mangrove root system has been shown to correlate with salt resistance: the higher the salt resistance, the larger the root system (Ball, 1988b; Sherman et al., 2003; Adame et al., 2014). The size of the root system thus is suggested to be the limiting factor for water uptake under elevated salinity (Ball, 1988a). Increasing root size, however, brings with it considerable carbon costs for the plant. First in growing the larger root system (Ball, 2002), but subsequently also in operating that root system. Salt exclusion during uptake requires significant amounts of energy (Ball, 1988a,b), and so does osmotic regulation (Munns and Gilliham, 2015). It has been argued that an increased salt tolerance thus comes at the expense of growth and competitive ability at low salinities (Ball, 2002).

Water storage capacity may help to minimize the direct and indirect costs of water uptake in two ways. Firstly, by spreading water uptake over the day and buffering peak demand. In this way the maximum water uptake per unit time is reduced, and so is the size of the root system that would be required to supply it. Secondly, by enabling the tree to take up water when it is less costly to do so: when salinities are low, at cooler temperatures with higher relative humidity (RH), during rain events, or at night when stomata are closed. Mangrove trees with an enhanced water storage capacity thus will be able to take up water more efficiently than those without, giving them a competitive advantage.

Larger trees have a larger relative water storage capacity and cope better with changes in water potential (Scholz et al., 2011 and references therein). It would then follow that small saplings have an especially difficult time to cope with such changes and could benefit greatly from specialized organs to provide additional water storage capacity. Mangrove saplings are especially likely to be confronted with strong salinity fluctuations, as without deep roots they will have to acquire water in the topsoil region where salinity will be the most variable (Bompy et al., 2014b). In such an environment, mangrove seedlings with an enhanced water storage capacity could have a competitive edge, which in turn could be an important determinant of forest structure and species composition (Smith and Snedaker, 1995; Ball, 2002; Bompy et al., 2014a).

The hypocotyl is the basal enlarged part of the propagule that after settlement is found between the root and the cotyledons, particularly in Rhizophoraceae. Many mangroves are viviparous: the embryo develops without immediate dormancy and the developing propagule detaches from the ovary and disperses by floating (Tomlinson, 1994). Rhizophoraceae produce especially long and large propagules, which after settlement become voluminous cylindrical hypocotyls topped by a smaller epicotyl (**Figure 1A**). After serving its role as a float during hydrochorous dispersal, the hypocotyl supplies nutrients and carbohydrates to facilitate rooting and settlement (Bobda et al., 2014; Robert et al., 2015; Tonné et al., 2016). It typically has a length of 15–70 cm and a diameter of 0.5–3 cm (Tomlinson, 1994). Because of its first function as a buoy the hypocotyl is not expected to have a very high tissue density. However, its large volume, mainly occupied by parenchyma and aerenchyma tissues, does invite the question whether after settlement and expenditure of its stored carbohydrates (Ball, 2002), it could change function and also serve as a water storage organ.

An organ that is also known to be suitable for storing water is the leaf (which do not develop until after settlement). Leaves provide significant water storage capacity (Schulze et al., 1985; Sack and Tyree, 2005; Blackman and Brodribb, 2011), especially when they are succulent (Scholz et al., 2011; Ishii et al., 2014). In mangroves an increased tendency toward succulence and high leaf water content is observed at high salinities, as seen in Avicennia germinans (Suárez and Sobrado, 2000), Laguncularia racemosa (Sobrado, 2005), and Bruguiera parviflora (Parida et al., 2004). A larger leaf mass increases heat capacity, thus reducing the need for evaporative cooling (for example when leaves are transiently exposed to full sunlight), while at the same time increasing water storage capacity (Ball, 1988b; Suárez and Sobrado, 2000; Wang et al., 2011). This may allow leaves to keep their stomata open, even when water influx into the leaf cannot immediately keep up with transpiratory loss. Succulent leaves are

suggested to enable mangrove plants to sequester large amounts of solutes without an adverse increase in cell osmotic pressure and to maintain turgor at low water potential (Suárez and Sobrado, 2000).

In this study we test whether the voluminous hypocotyl and the succulent leaves of Rhizophoraceae mangrove saplings can serve as water reservoirs to buffer sudden changes in soil water salinity. We hypothesize that for water storage tissues to have adaptive value, they should be able to contain and release a volume of water that is large enough to buffer such sudden changes. We distinguish short term water content fluctuations, understood as the volume of water that can be withdrawn from an organ during the day and replenished during the night (Goldstein et al., 1998), and long term water content fluctuations, the amount of water lost or gained due to prolonged exposure to changed abiotic conditions. To estimate the amount of water that the hypocotyls and leaves of Rhizophora mucronata and Bruguiera gymnorrhiza saplings have available to buffer changes in their abiotic environment, we challenged the plants with three step-wise soil salinity treatments lasting 5 days each: 15, 30, and 0–5h NaCl. During this period we monitored the water content of the hypocotyls and leaves by means of non-invasive mobile NMR sensors; changes in hypocotyl diameter and leaf thickness were measured with classical dendrometers. In addition, a field study on Rhizophora mangle saplings was done to test whether the use of hypocotyl water storage capacity could be detected under field conditions.

## MATERIALS AND METHODS

#### Plant Material

Four saplings of two mangrove species, R. mucronata (L.) LAMK (Rm – plants 1 and 2) and B. gymnorrhiza (L.) LAMK. (Bg – plants 3 and 4) were grown from propagules, collected in January 2013 in Mida Creek (Watamu Marine National Reserve, Kenya). After planting the propagules in regular potting soil and sand (1:1 ratio), they were grown at 15h NaCl in a greenhouse in Brussels, Belgium. During autumn and winter 100 µmol m−<sup>2</sup> s <sup>−</sup><sup>1</sup> PAR of additional lighting was provided. Leaves were sprayed daily with freshwater to maintain the RH around 70%.

To elicit a release or uptake of water, R. mucronata and B. gymnorrhiza saplings were challenged with a three stage salinity treatment, 15 (native), 30, and 0–5h, while continuously monitoring changes in the VWC of leaf and hypocotyl (by means of NMR), as well as changes in hypocotyl diameter and leaf thickness (by means of dendrometers).

### Magnetic Resonance Imaging Study

Hypocotyl water content was measured in R. mucronata plants 1 and 2 (**Figure 1B**) by means of nuclear MRI at the Wageningen NMR Centre (Wageningen University, The Netherlands). A 3T imager was employed, consisting of a 50 cm vertical bore magnet (Magnex, Oxford, UK) and a 1 T m−<sup>1</sup> openable gradient coil (Bruker, Karlsruhe, Germany). A RF coil with an inner diameter of 4 cm was placed around the plant's hypocotyl at a height of 28 cm. Inside the scanner the plants were exposed to an air temperature of between 22 and 24◦C, a RH of 40% and a day time light intensity of 120 µmol m−<sup>2</sup> s <sup>−</sup><sup>1</sup> PAR at leaf level. Between measurements the plants were grown in a climate chamber (20◦C night and 26◦C day; RH: 70%; 100 µmol m−<sup>2</sup> s <sup>−</sup><sup>1</sup> PAR; 12 h photoperiod). The plants were sprayed with freshwater.

The CPMG measurements were done using the following settings: FOV: 20 mm × 20 mm; slice thickness: 3 mm; matrix size: 256 × 256 pixels; number of averages: 20; TE: 8.1 ms; number of echos: 128; TR: 2000 ms; total scan time: 2 h 51 m; SW: 50 kHz. Emphasis was placed on maximizing the spatial resolution and the signal to noise ratio of the images, causing a degree of saturation as well as a degree of T<sup>2</sup> weighting in the resulting amplitude map. The inversion recovery measurements were done using the following settings: FOV: 20 mm × 20 mm; slice thickness: 3 mm; matrix size 128 × 128; number of averages: 4; first echo: 4.6 ms; TE: 3.7 ms; inversion steps: 10 (50, 100, 200, 400, 800, 1000, 1500, 2000, 3000, 4000 ms); turbo factor:

8; TR: 4500 ms; SW: 50 KHz; scan time: 48 min. The datasets were fitted on a per pixel basis, using a mono-exponential decay function (van der Weerd et al., 2000), yielding maps of amplitude, T<sup>2</sup> and T<sup>1</sup> (Donker et al., 1997; Edzes et al., 1998). For the inversion recovery measurement, the resulting amplitude map was expressed in terms of VWC for which the 100% value was defined as the mean of the pixels in the center of the reference tubes.

After finishing the salinity experiment, three and a half months after the MRI study, plant 1 (Rm) was harvested and conserved in 50% ethanol. Transverse anatomical sections of the upper part of the hypocotyl were made with a sliding microtome (WSL Lab-Microtome, WSL, Birmensdorf, Switzerland). The obtained sections were placed on albumincovered microscope slides, bleached and dried for 30 min at 70◦C, then stained with Astra Blue (0.5%) and Safranin O (1%) for 3 min. Surplus stain was removed with 50 and 96% ethanol, progressively. The sections were embedded in xylol and Eukitt (BiOptica, Milan, Italy), covered with a cover glass and dried overnight. The cross section of the hypocotyl of plant 1 (Rm) was used as an anatomical reference for the MRI images.

## Stepwise 15-30-0<sup>h</sup> Salinity Treatments

The stepwise salinity experiment was done at the IBG-2: Plant Sciences Institute of the Forschungszentrum Jülich (Jülich, Germany). Prior to the experiment, the plants were grown in a climate chamber for an acclimatization period of 3 weeks (air temperature: 24–26◦C, day–night variation 2◦C; relative air humidity: ca. 75%; light: 145 µmol m−<sup>2</sup> s <sup>−</sup><sup>1</sup> PAR; 12 h photoperiod) and watered with a 15h salinity solution. During the experiment, one plant per species was placed in different climate chambers to avoid possible crosstalk between the NMR sensors [air temperature: 26◦C day and 24◦C night; relative air humidity: ca. 65% (climate chamber 1) and ca. 75% (climate chamber 2); light: 185 µmol m−<sup>2</sup> s <sup>−</sup><sup>1</sup> PAR (climate chamber 1) and 145 µmol m−<sup>2</sup> s <sup>−</sup><sup>1</sup> PAR (climate chamber 2), 12 h photoperiod].

At the start of the experiment, the saplings were approximately 15 months-old. Plant growth was monitored weekly by recording the height of the plant and the hypocotyl diameter at 10 cm above the soil, by counting the number of leaves and by calculating the total leaf area after measuring individual leaf area of all leaves by tracing the outline on paper. The stomatal conductance was measured weekly using a SC1-Porometer (Decagon Devices, Pullman, WA, USA; four leaves per plant, three measurements per leaf).

Each plant (N = 4) was subjected to three salinity treatments: (i) 6–7 days at 15h (salinity at which the plants were grown), (ii) 5 days at 30h, and (iii) 5 days at 0–5h salinity. To carry out these treatments, the pots with plants were placed in a waterfilled bucket. Non-iodized salt was used to increase salinity. The salinity of the interstitial water in the soil was adjusted by repeatedly rinsing with large quantities of saline or freshwater until the salinity remained constant at the desired value. The water and soil water salinity was measured daily using a handheld refractometer (0–100h, Erma, Tokyo, Japan), after pipetting the water or extracting it from the bottom of a plastic tube buried in the soil, and adjusted when needed.

## Mobile NMR Measurements

The water contents of the hypocotyls and leaves of the plants were continuously and non-invasively measured by means of novel mobile NMR sensors (Windt et al., 2011; Windt and Blumler, 2015; **Figure 1D**). Two sensors were used to monitor the hypocotyls of plant 1 (Rm) and 3 (Bg), and a leaf of plant 2 (Rm) and 4 (Bg). The NMR sensors consisted of C-shaped permanent magnets with a field strength of 0.235 T over an air gap of 37 mm. For the hypocotyl the magnet was fitted with a 13 turn, 20 mm (inner diameter, i.d.) solenoidal RF coil, hand-wound around the hypocotyl with the help of a Teflon mold. For leaves it was fitted with a 13 turn, 15 mm (i.d.) coil. In order to insert the leaf it was first rolled up and then inserted into the coil, effectively keeping the whole leaf in the dark for the duration of the experiment. For both plants, a representative leaf was selected for the measurements.

The NMR sensor was driven with a Magritek Kea II spectrometer (Magritek, Wellington, New Zealand) with a standard 100 W internal RF amplifier. To avoid temperature dependent modulations of signal amplification by the spectrometer, the spectrometer was placed in a temperature controlled isolated housing, capable of keeping spectrometer temperature constant to ±1 ◦C.

T<sup>2</sup> relaxometry was performed using a CPMG sequence and the following settings for both hypocotyl and leaf: TR 5 s; repetitions: 128; number of echoes: 4000; TE: 400 µs; SW: 200 kHz; pulse lengths 16 µs (leaf) and 8 µs (hypocotyl), Total scan time: 9 min per time point. Pulse amplitude varied to achieve excitation and refocusing.

To allow for real time, unsupervised data processing, no attempt was made to fit the multi-exponential T<sup>2</sup> relaxation curve. Instead, all echoes between 0 and 25 ms were averaged. This maximizes signal to noise and avoids fitting errors that might otherwise occur when dealing with plant parts that may give rise to complex and changing multi-exponential T<sup>2</sup> relaxation behavior, such as expected in the living hypocotyl and leaf samples measured under changing environmental conditions (Gultekin and Gore, 2005; Van As, 2007). See Appendix S1 for detailed information regarding the method and calibration.

The temperature of the saplings in the NMR sensor was continuously monitored. The slight changes in signal amplitude that would result from temperature – dependent changes in the Boltzmann equilibrium and minor changes in the resistance of the RF assembly were corrected by applying a correction factor of −0.44% per ◦C. The correction factor was determined independently by placing the NMR sensor together with a reference sample in a temperature controlled environment. The corrected NMR amplitude then scales linearly with the amount of liquid water inside the sensitive volume of the RF coil in the NMR probe head. Water content measured by means of the NMR sensor thus is expressed as WCcoil, the amount of water detected in the sensitive volume of the NMR RF coil.

## Volume and Water Content of Hypocotyl and Leaves

During the stepwise salinity experiment, the total water content and changes in VWC were calculated from WCcoil and total organ volume. We thus assume that the water content in the middle of the hypocotyl or in the middle of the leaf closely matches the water content in the rest of the organ. Hypocotyl and total leaf volume were determined on the basis of measurements done at the start of the salinity experiment. Total hypocotyl volume was approximated using the truncated cone as:

$$V = (1/3)\pi (r\_1^2 + r\_1r\_2 + r\_2^2)h \tag{1}$$

where r<sup>1</sup> is the radius of the lowest part of the hypocotyl, r<sup>2</sup> the radius of the highest part of the hypocotyl and h the length of the hypocotyl. Leaf surface measurements, obtained by tracing the leaves outlines on grid paper, were used to estimate the total leaf water content. The average leaf thickness was assumed to be the same for all leaves. At the end of the experiment the total freshand dry weights were measured gravimetrically. Fresh weight was measured directly after harvesting, dry weight after drying at 70◦C for 48 h.

The NMR-based water content measurements were used to determine four further parameters:

(i) 1WClong term maximum, here defined as the largest difference between the night values of two consecutive treatments, (ii) 1WCdiurnal maximum, the difference between the maximum values of two consecutive nights within the same treatment, (iii) 1WCstep, the maximum difference between the night value before and the night after the salinity step and (iv) 1WCday−night, defined as the difference between the maximum night value and minimum day value of the same day (24 h).

#### Dendrometer Measurements

In the absence of additional NMR sensors to measure water content directly, dendrometers were used as a proxy (De Swaef et al., 2015 and references therein). LVDTs (root and aquatic plant dendrometer DRO, Ecomatik, Dachau/Munich, Germany; sensor range 11 mm, resolution 2.6 µm) were placed on the hypocotyl at mid-height (**Figure 1C**).

Leaf thickness variation was measured with a miniature displacement transducer (DF-5.0, Solartron Metrology, Leicester, England; sensor range 5 mm, resolution rated "infinite," limited only by the controlling hardware). The sensor was placed on the adaxial surface of one young- or middle-aged leaf; the dendrometer was held by a custom-built support (**Figure 1D**). To minimize the pressure on the leaf surface, the sensor rod was not pressed onto the leaf by means of a spring, but was simply placed on the leaf surface and held there by gravity alone.

The dendrometer data were logged and stored every 5 min (hypocotyl: HOBO U12 4-Channel External Data Logger – U12- 006, Onset, MA, USA; leaf: CR1000, Campbell Scientific, Logan, UH, USA). The dendrometer data were analyzed using SAS v9.4 for Windows (Statistical Analysis System, SAS Institute, Cary, NC, USA). Hourly measurements were extracted as described by Deslauriers et al. (2011). A smoothing degree of 2 on a scale from 0 to 10 was used to avoid the loss of significant daily variation.

## Field Study

To compare our laboratory findings with observations in the field, we analyzed a dataset of dendrometer measurements on saplings of R. mangle, equally a Rhizophoraceae species with a growth habit and morphology closely matching those of R. mucronata. The study was conducted in the Avalon State Park (North Hutchinson Island, lagoon side, St. Lucie County, FL, USA; Feller et al., 2003, 2007). The dominant mangrove tree species in this forest is Avicennia germinans (L.) L. with scattered Laguncularia racemosa L. trees in the forest interior and individuals of R. mangle L. confined to the forest periphery (Feller et al., 2003). Three R. mangle saplings of similar age and size were selected, having six leaves and mid-height hypocotyl diameters of 7.70, 9.44, and 10.80 mm. The saplings were growing under the mangrove tree canopy in a sandy soil at a spot that was twice a day inundated by the tides so that the soil water salinity at this location was fluctuating around seawater salinity during periods without rainfall. The distance between the saplings was no more than 1.5 meters. On each sapling, a dendrometer (Root and aquatic Plant Dendrometer, DRO, Ecomatik, Dachau/Munich, Germany) was placed at hypocotyl mid-height. Radial hypocotyl changes were logged at a 10-min interval from 8 to 16 January 2014 with a resolution of 2.6 µm. Air temperature, RH and rainfall were registered during the study period (further details: Appendix S1).

## RESULTS

For all plants the total water content of the hypocotyls was lower than the total water content of their leaves (total water content determined after harvest, **Table 1** – Organ water content). The hypocotyls of R. mucronata on average contained 45.3 g of water, versus 59.5 g for the leaves. For B. gymnorrhiza these numbers were 16.3 and 34.8 g, respectively (**Table 1** – Organ water content).

We observed that the vascular tissues had the highest VWC with an average value of 68%, whereas, the epidermis, the cortex and the pith tissues had a VWC of 45–46% (**Figure 2C**). The sclereid bundles could not be fully resolved in the apparent proton density image (**Figure 2B**) nor in the VWC image (**Figure 2C**), but certainly appeared to possess a VWC that was lower than that found for the pith parenchyma. The low water contents in the pith and the sclereid bundles indicate the presence of relatively large amounts of air (**Figure 2A**).

## Stepwise Salinity Treatment

The stomatal conductance varied markedly in all plants in response to the salinity changes. It decreased at 30h salinity relative to the 15h treatment, and it increased at 0–5h salinity (**Table 2**). All plants grew over the duration of the experiment, with increases in height, diameter, and leaf area (**Table 2**). Only plant 3 (Bg) lost a leaf naturally, decreasing the total leaf area.

In the stepwise salinity treatment, we distinguish diurnal and long term responses, the first acting over the diurnal time course (hours), the second over the full duration of a salinity stage (days).

TABLE 1 | Total organ fresh weight, dry weight and water content (WC), 1WC long term maximum (difference between the night values of two consecutive treatments), 1WC diurnal maximum (difference between the maximum values of two consecutive nights within the same treatment), 1WC day–night (difference between the maximum night value and minimum day value of the same day) and 1WC step (difference between the night value before and the night after the salinity step) in hypocotyl and leaf of Rhizophora mucronata and Bruguiera gymnorrhiza saplings during the step-wise salinity treatments.


Total organ fresh weight, dry weight and water contents of the hypocotyl and leaves were calculated on the basis of leaf and hypocotyl samples which were harvested after the experiment and analyzed gravimetrically. The dynamic water content variations were calculated on the basis of the WCcoil values as measured by NMR; – No data could be obtained.

In the next section the short and long term responses will be described in details.

#### Diurnal Response of Hypocotyls and Leaves

The diurnal fluctuations in leaf and hypocotyl VWC were strongly influenced by the salinity treatments (**Table 1**; **Figure 3**). After shifting to 30h salinity, the amplitude of the fluctuations increased by 65–71% in plant 1, 2 (Rm) and 4 (Bg), and by 42% in plant 3 (Bg), relative to the amplitudes at 15h salinity. At 30h salinity, in plants 1 and 2 (Rm) and plant 3 (Bg) the amount of water lost during the day was always larger than the amount replenished at night, causing a net water loss.

After transferral to 0–5h salinity, the diurnal fluctuations in the hypocotyl decreased, reaching values lower than or similar to the ones observed during the first treatment stage (15h), whereas, in the leaf the amplitudes of the diurnal fluctuations were similar to the ones observed at 30h salinity or slightly lower, but larger than at 15h salinity (**Table 1** – <sup>1</sup>WC day– night). In plants 1 and 2 (Rm) and plant 3 (Bg) the water gain

the spatial resolution to more clearly resolve the sclereid bundles and to provide optimal anatomical contrast, albeit at the expense of moderate T<sup>1</sup> weighting resulting from a combination of long ETs and short RTs. T<sup>1</sup> weighting caused a loss of intensity in the pith and the top two reference tubes. Image C represents VWC.

#### TABLE 2 | Plant characteristics.

fpls-07-00895 June 24, 2016 Time: 17:1 # 7


Hypocotyl diameter, plant height, number of leaves and total foliar area before (Pre) and after (Post) the stepwise salinity treatments and stomatal conductance during the three treatments in hypocotyl and leaves. The stomatal conductance values are calculated as the average of all measurements (three measurements per leaf, four leaves per plant).

FIGURE 3 | Variations in hypocotyl and leaf water content (expressed as WCcoil), in hypocotyl diameter and in leaf thickness, measured over 17–20 days under three salinity treatments (15, 30, 0–5<sup>h</sup> NaCl) on (A) plant 1 – Rhizophora mucronata, (B) plant 2 – Rhizophora mucronata, (C) plant 3 – Bruguiera gymnorrhiza, and (D) plant 4 – Bruguiera gymnorrhiza. The salinity treatments are indicated by the bars above the graphs and delimited by the vertical dashed lines. The water content traces were measured by means of NMR sensors; the diameter and leaf thickness by means of dendrometers. The amplitude at the start of the NMR measurement is defined as 100%.

during the night was always higher than the water lost producing a daily net water gain (**Figure 3**).

The average amount of water lost and replenished during the day in the leaves was always larger than in the hypocotyl, regardless of the salinity treatment, for all plants and both species (**Table 1** – 1WC day–night, **Figure 3**).

During all stages of the experiment, the leaves reacted very quickly when the lights were turned on. At 7:00 the lights were turned on, resulting in marked decreases in VWC and leaf shrinkage within 15 to 45 min (**Figure 4**). In the hypocotyl a decrease in VWC was not observed until between 8:30 and 9:00, an average delay of about 80 ± 17 min (Rm) and 72 ± 17 min (Bg) after the leaves showed a first reaction (**Figure 4**). There was some variability in the delay, but it remained constant irrespective of the salinity treatment.

#### Long Term Response of Hypocotyls

Upon changing the salinity of the root medium from 15 to 30h, the relative amount of water lost from the hypocotyl was similar in both species, amounting to 3.3% in plant 1 (Rm) and 3.6% in

plant 3 (Bg). The depletion of water progressed gradually with a maximum daily net water loss of 1.1% in plant 1 (Rm) and 0.9% in plant 3 (Bg). After lowering the soil water salinity from 30 to 0–5h the increase in hypocotyl VWC was much larger than the decrease in VWC upon transfer from 15 to 30h. In plant 1 (Rm), an increase of 8.1% was found, with a daily net maximum water gain of 0.9% whereas, in plant 3 (Bg) these values were 5.6 and 0.6% (**Table 1**; **Figures 3A,C**). While during the transition from the 15h to the 30h treatment the decrease of the VWC was slightly lower than the daily net maximum water loss (0.9%, plant 1, Rm) and 0.7% (plant 3, Bg), the increase in the VWC at the transition from the 30h to the 0–5h treatment was larger than the daily net maximum water gain (2.6%, plant 1, Rm) and 3.4% (plant 3, Bg; **Table 1**; **Figures 3A,C**).

The hypocotyl diameter variations almost perfectly mirrored the changes in VWC as measured by NMR, with the exception that the NMR measurement seemed to detect small changes more easily than did the dendrometers especially in the daily variations (**Figure 3**). At 30h salinity, a gradual decrease of between 100 and 130 µm in plants 1 and 2 (Rm; diameters of 12.60 and 13.45 mm, respectively), and in plant 3 (Bg; diameter 15.70 mm) was observed during the 5 days of treatment. Interestingly, plant 4 (Bg) did not show a gradual response, but instead exhibited a rapid decrease of ca. 200 µm (diameter: 11.60 mm) during the first day of treatment upon transferral from 15 to 30h, and stayed at that level for the entire duration of the treatment. On transferral from 30 to 0–5h, the amplitude of the hypocotyl swelling was similar for plants 1 and 2 (Rm) and plant 3 (Bg), gradually reaching values between 175 and 230 µm after 5 days of treatment in 0–5h, whereas, plant 4 showed a rapid increase, reaching its new diameter value overnight (**Figure 3**).

#### Long Term Response of Leaves

Upon transferral from 15 to 30h salinity, the percentage decrease of leaf VWC was more than double in plant 2 (Rm) than in plant 4 (Bg; **Table 1** – 1WC long term maximum, **Figures 3B,D**) while the transferral to 0–5h salinity induced a similar increase of the % VWC in the leaves of both plants. In plant 2 (Rm) the decrease and increase of VWC induced by the change in salinity was gradual, while in the leaf of plant 4 (Bg) the change in VWC occurred rapidly during the first day of treatment (**Table 1**; **Figures 3B,D**). The increase in leaf VWC observed in the transition day (from 30 to 0–5h) and in the 0–5h treatment, was larger than the decrease observed in the transition day from 15 to 30h and in the 30h salinity treatment of both plants (**Table 1** – 1WC step, **Figures 3A,C**).

The leaf thickness variations closely mirrored the response in VWC as measured by NMR (**Figures 3B,D**). The change in the amplitude tended to be much larger in Rm (plants 1 and 2; 100–230 µm) than in Bg (plants 3 and 4; 17–35 µm), after the transferral from 15 to 30h as well as for the transferral from 30 to 0–5h salinity (**Figure 3**). Plant 4 (Bg) is the only sapling that showed a fast change in the first day of treatments, all other saplings exhibited gradual changes.

In order to be able to place all sensors on plant 4 (Bg), the NMR measurements had to be done on a young growing leaf,

while a fully grown leaf just below was used for the dendrometer measurements. During the first treatment (15h NaCl), the growth of the leaf translated into a constant increase in the VWC (NMR data) that was absent in the dendrometer data (**Figure 3D**), in which the net leaf thickness did not change. This was not observed in plant 2 (Rm), as here it was possible to place both devices on mature leaves of similar age.

#### Field Study

Rhizophora mangle saplings 1 and 3, monitored in the Avalon State Park mangrove forest (FL, USA) exhibited comparable daily patterns of hypocotyl diameter variations during rainless days, i.e., late morning shrinkage and swelling in the early to late afternoon (**Figure 5**). The hypocotyl diameter variations of sapling 2 were less pronounced (**Figure 5**). During a day with continuous rainfall (9 January, 16.49 ml cm−<sup>2</sup> , **Figure 5**), all saplings showed an increase in hypocotyl diameter that continued for the next 2 days. On this day RH was high and air temperature low, causing an especially low VPD. After shorter rainfall events (10 and 14 January, **Figure 5**) a swelling or a reduction of hypocotyl shrinkage was observed as compared to drier days.

## DISCUSSION

To estimate the amount of water that the hypocotyls and leaves of R. mucronata and B. gymnorrhiza saplings have available to buffer water potential changes in their abiotic environment, we submitted the saplings with three stepwise soil salinity treatments. These salinities can be regarded as typical of infrequently flooded area of the forest, where fast and drastic variations in soil water salinity are caused by spring and neap tides, freshwater input and rain events (Ball, 1988a; Tomlinson, 1994; Robert et al., 2009).

There was a reduction of stomatal conductance in the saplings when shifting from 15 to 30h NaCl and a marked increase in both parameters when shifting from 30 to 0–5h NaCl, reaching values higher than previously measured at 15h. These observations are in agreement with prior studies that demonstrated a decrease in growth, stomatal conductance and photosynthesis in response to higher soil water salinities, and an increase in these physiological parameters at lower salinities (Parida et al., 2004; Hoppe-Speer et al., 2011).

## Water Storage Capacity of Hypocotyl and Leaves: Anatomy

To investigate what tissues in the hypocotyl could store water, MRI was done on the hypocotyl of R. mucronata specimens. The results showed evidence of both the original functions of the hypocotyl as a buoy (relatively high air content), and also demonstrated the presence of significant amounts of water. In the R. mucronata hypocotyl the vascular tissues showed a higher water content than the parenchymatous tissues (cortex and pith), but as the latter tissues make up the largest part of the hypocotyl by far, their water storage capacity should not be underestimated. In trees, the parenchyma is commonly associated with water storage (Morris et al., 2015; Pfautsch et al., 2015a). The hypocotyl reflects its function as a float during the dispersal phase of the propagule, but contains sufficient water storage tissues to be involved in water buffering once the young mangrove saplings are settled. The high air content of the hypocotyl could be valuable as an oxygen reserve during periods of inundation.

At approximately 15 months of age, the leaves of all mangrove saplings of both species under study had a higher cumulative water content than the hypocotyl, their VWC was 4–5% in response to changes in soil water salinity. The leaves of both species have a thick hypodermis and thick mesophyll layers, giving them a succulent appearance (Tomlinson, 1994; Das, 1999) and a large water storage capacity (Das, 1999; Suárez and Sobrado, 2000). R. mucronata leaves have a multilayer hypodermis, whereas, in B. gymnorrhiza it consists of a single

layer of cells. This might explain why R. mucronata exhibited larger variations in leaf water content and thickness than B. gymnorrhiza. The differences in leaf thickness variations between the two species might also be influenced by the presence of branched fiber-sclereids in the mesophyll of R. mucronata, a feature absent in B. gymnorrhiza. Branched fiber-sclereids give mechanical support under low turgor conditions (Tomlinson, 1994; Das, 1999) and may thus physically limit the shrinkage of the leaf and help it preserve its structural stability when turgor is low.

#### Long and Short Term Buffering

Shrinkage and swelling were observed in response to stepwise salinity changes. The diameter variations closely mirrored the changes in VWC, validating the use of the dendrometers as a proxy to observe changes in water content in leaf and hypocotyl. The NMR sensor detects small changes compare to the dendrometers.

Short term water content fluctuations, 1WCday−night, were higher in the leaves than in the hypocotyl at all salinities, both in absolute and in relative terms. At 30h salinity, <sup>1</sup>WCday−night increased by at least 50% relative to the volume withdrawn at <sup>15</sup>h salinity. Upon shifting from 30 to 0–5h, <sup>1</sup>WCday−night decreased again in leaves and hypocotyls, but it tended to remain higher in leaves compared to native salinity (15h). The fact that this daily pattern could also be observed at 15h salinity, the condition at which the plants were grown, as well as at 0–5h salinity, suggests that water storage capacity does not only have a role under adverse conditions but that it also helps to deal with mild daily variations.

The short term diurnal water content fluctuations were superimposed on much slower, long term responses of water loss at 30h and refilling at 0–5h. In total, the maximum amount of water lost for the hypocotyl was 6.2% for R. mucronata and 4.8% for B. gymnorrhiza, against 11.7% for the leaves of R. mucronata and 7.3% for B. gymnorrhiza. Only plant 4 (Bg) did not show a gradual release of water during the salinity treatments, possibly due to its small size or to a damaged root. The latter cause might be the most likely one. B. gymnorrhiza has thick, brittle roots that sometimes grow out of the bottom of the pot. It is conceivable that during the experimental treatments one of these roots got damaged, thus temporarily damaging the root barrier and opening the xylem to the root environment. This way the plant would not be able to slowly use up its cumulative water storage pools, but in absence of a semipermeable barrier much more quickly reach equilibrium with the water status of the rhizosphere through direct exchange.

The long term responses indicate that water storage in hypocotyl and leaves does not contribute to short term buffering only. Similar patterns have been found in baobab trees (Adansonia spp. L.), in which stem stored-water usage has a major role during longer-term water deficits (Chapotin et al., 2006). Surprisingly, at 15 months of age, in all specimens of both species and under all treatments the water storage capacity of the leaves was found to be larger than that of the hypocotyl. This indicates not only that the leaves of both species add considerably to their water storage capacity, but also that at younger ages, when only few leaves have developed, the hypocotyl is likely to contain the bulk of the stored water in the sapling. At early developmental stages, the availability of water storage capacity might be even more important than later stages, as suggested by Smith and Snedaker (1995). We thus conclude that the role of the young hypocotyl after settlement is not only restricted to the storage of nutrients and air, which are the same functions as the structure had during hydrochory, but also includes the storage of water after the seedling has settled: a function that at first glance would seem to contradict its previous function as a float.

Under the relatively mild challenges that were imposed in the current study, the absolute amount of water released by the short and long term buffering responses was not very large. It may, however, still be sufficient to give the sapling a competitive edge. Evidence to this effect comes from sapling and mature trees. Studies done on five species of tropical adult trees showed that species with greater storage capacity maintained their maximum rates of transpiration for a substantially longer fraction of the day than trees with a smaller storage capacity, demonstrating how relatively small volumes of water withdrawn from the internal tissues can positively influence the carbon balance of a tree (Goldstein et al., 1998; Zweifel and Häsler, 2001).

While the depletion of the water storage pools of the saplings took several days, the replenishment of water storage pools in leaves and hypocotyls occurred very fast when transferred from 30 to 0–5h. This confirms that, as has been observed for radial growth in adult Avicennia marina trees (Robert et al., 2014; Santini et al., 2015), young mangrove trees take advantage of short spells of favorable conditions to quickly replenish their water reserves. The ability to recover after salt stress and to make the most of brief periods of low salinity may not only be key to species survival (Hale and Orcutt, 2000; Bompy et al., 2014a), may also strongly influence mangrove forest structure as it affects which seedlings survive at what positions along gradients of height, salinity, likelihood of freshwater input, and tidal exposure (Ball, 2002).

#### Time Delayed Responses

A surprising result was that even in small saplings, buffering responses are associated with considerable time lags. In both species and during all salinity treatments, light exposed leaves almost instantly lost water or shrunk as soon as the lights were turned on. The hypocotyl, on the other hand, did not exhibit loss of water or shrinkage until 72 (Bg) to 80 min (Rm) later. A much shorter time lag of 5–15 min was measured between a leaf exposed to the light, and another leaf on the same plant in the dark inside the RF coil of the NMR sensor. Due to its proximity, the leaf inside the RF coil probably provided stored water to the transpiring leaves, an hour before the hypocotyl did.

The occurrence of time lags in diameter changes among different stem heights, or between stem, crown or leaves, has been well-documented in adult trees in response to abiotic stimuli (Goldstein et al., 1998; Zweifel and Häsler, 2001; Zweifel et al., 2001; Sevanto et al., 2002) even when the plants were wellwatered (Zweifel et al., 2000). In Scots pine (Pinus sylvestris L.) Sevanto et al. (2002) measured a time lag between the "total" stem

(meant as xylem and bark) and the xylem (without bark) that varied from 30 to 110 min, proving that the bark tissues serve as a water reserve to replace the water lost during transpiration.

As far as we are aware, such delayed responses, typically associated with storage capacity and buffering, have not been reported for saplings. This signifies that the leaves and hypocotyls do buffer sudden changes in the abiotic environment of the sapling.

#### Field Conditions

The observed rapid swelling of the sapling hypocotyl tissues during the day with extensive rainfall, confirmed that mangrove saplings have the ability to rapidly take benefit of freshwater input. In the absence of the experimental abrupt light on or light off events of the laboratory, the diurnal pattern of swelling and shrinkage was less pronounced than in the laboratory, but closely mirrored the changes in VPD in the forest after the rain had stopped. The observation that also in R. mangle and under natural conditions the hypocotyl responds to changes in water potential and VPD, strengthens the view that stored water in the hypocotyl is mobilized, depleted and replenished as needed even under non-stress conditions in the field.

More field experiments are needed to clarify the exact role and dynamics of the water reserves in the different plant organs under natural conditions, considering periods of extreme change in salinity. It could also be interesting to verify whether foliar rain water uptake can play a role in directly feeding into the storage water capacity of the leaves, analogous to the foliar uptake of water in trees where sheer tree height poses an impedance to the transport of water taken up by the roots (Limm et al., 2009; Simonin et al., 2009; Ishii et al., 2014).

#### AUTHOR CONTRIBUTIONS

SL and ER: conception and design of the study, acquisition of data, analysis and interpretation of the data, drafting and revising of the manuscript. NT: acquisition of data, analysis and interpretation of the data, revising of the manuscript. AP,

#### REFERENCES


EG: acquisition of data, analysis of the data, revising of the manuscript. HVA: analysis and interpretation of the data and revising of the manuscript. NK: analysis and interpretation of the data and revising of the manuscript. CW: conception and design of the study, construction of the NMR sensors, acquisition of data, analysis and interpretation of the data, drafting and revising of the manuscript.

#### FUNDING

The study was financially supported by the Research Foundation – Flanders (FWO, Flanders, Belgium), the European Commission through the Marie Curie Action 'CREC' (EU IRSES #247514) and the IBG-2: Plant Sciences institute at the Forschungszentrum Jülich. This work has been published with support of the University Foundation (Belgium).

#### ACKNOWLEDGMENTS

We thank Mohamed O. S. Mohamed (Kenya Wildlife Service – plant material provision), Martine Claeys (Vrije Universiteit Brussel – greenhouse help), Ilka Feller, Lorae Simpson, Michael Lehmann (Smithsonian Marine Station, Florida – field work help), Onno Muller, Thomas Hombach (Forschungszentrum Jülich – lab measurements), Frank Vergeldt (Wageningen NMR Centre – MRI measurements), Ronny Merken (Vrije Universiteit Brussel), Luís Dias, Nuno Grosso (Faculdade de Ciências da Universidade de Lisboa; suggestions data analysis), Alan Crivellaro (University of Padova – microtomy assistance) and Pierluigi Colangeli (Universität Potsdam, Potsdam – practical assistance). We are grateful to the SAS Institute office (Tervuren, Belgium) for providing their software.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.00895


stem water buffer daily water deficits? Plant Cell Environ. 29, 1021–1032. doi: 10.1111/j.1365-3040.2005.01456.x


curves. Magn. Reson. Imaging 18, 1151–1158. doi: 10.1016/S0730-725X(00) 00200-9


Zweifel, R., Item, H., and Häsler, R. (2001). Link between diurnal stem radius changes and tree water relations. Tree Physiol. 21, 869–877. doi: 10.1093/treephys/21.12-13.869

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer AD and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2016 Lechthaler, Robert, Tonné, Prusova, Gerkema, Van As, Koedam and Windt. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Limited Growth Recovery after Drought-Induced Forest Dieback in Very Defoliated Trees of Two Pine Species

#### Guillermo Guada<sup>1</sup> , J. Julio Camarero<sup>2</sup> \*, Raúl Sánchez-Salguero<sup>3</sup> and Rafael M. Navarro Cerrillo<sup>4</sup>

<sup>1</sup> Departamento de Botánica, Universidade de Santiago de Compostela, Lugo, Spain, <sup>2</sup> Instituto Pirenaico de Ecología (IPE-CSIC), Zaragoza, Spain, <sup>3</sup> Departamento de Sistemas Físicos, Químicos y Naturales, Universidad Pablo de Olavide, Sevilla, Spain, <sup>4</sup> Departamento de Ingeniería Forestal, Universidad de Córdoba, Córdoba, Spain

Mediterranean pine forests display high resilience after extreme climatic events such as severe droughts. However, recent dry spells causing growth decline and triggering forest dieback challenge the capacity of some forests to recover following major disturbances. To describe how resilient the responses of forests to drought can be, we quantified growth dynamics in plantations of two pine species (Scots pine, black pine) located in south-eastern Spain and showing drought-triggered dieback. Radial growth was characterized at inter- (tree-ring width) and intra-annual (xylogenesis) scales in three defoliation levels. It was assumed that the higher defoliation the more negative the impact of drought on tree growth. Tree-ring width chronologies were built and xylogenesis was characterized 3 years after the last severe drought occurred. Annual growth data and the number of tracheids produced in different stages of xylem formation were related to climate data at several time scales. Drought negatively impacted growth of the most defoliated trees in both pine species. In Scots pine, xylem formation started earlier in the non-defoliated than in the most defoliated trees. Defoliated trees presented the shortest duration of the radial-enlargement phase in both species. On average the most defoliated trees formed 60% of the number of mature tracheids formed by the non-defoliated trees in both species. Since radial enlargement is the xylogenesis phase most tightly related to final growth, this explains why the most defoliated trees grew the least due to their altered xylogenesis phases. Our findings indicate a very limited resilience capacity of drought-defoliated Scots and black pines. Moreover, droughts produce legacy effects on xylogenesis of highly defoliated trees which could not recover previous growth rates and are thus more prone to die.

Keywords: dendroecology, die-off, extreme climate event, forest resilience, *Pinus nigra*, *Pinus sylvestris*, xylem, xylogenesis

## INTRODUCTION

Mediterranean forests are able to recover following major disturbances such as droughts by displaying high resilience (e.g., Lloret et al., 2004). However, climate warming is expected to magnify drought stress in the Mediterranean Basin by rising air temperatures and evapotranspiration rates thus amplifying drying trends (Cook et al., 2014). Warmer temperatures,

#### *Edited by:*

Achim Braeuning, University Erlangen-Nuremberg, Germany

#### *Reviewed by:*

Eryuan Liang, Chinese Academy of Sciences, China Teemu Hölttä, University of Helsinki, Finland

> *\*Correspondence:* J. Julio Camarero jjcamarero@ipe.csic.es

#### *Specialty section:*

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

*Received:* 28 January 2016 *Accepted:* 18 March 2016 *Published:* 01 April 2016

#### *Citation:*

Guada G, Camarero JJ, Sánchez-Salguero R and Navarro Cerrillo RM (2016) Limited Growth Recovery after Drought-Induced Forest Dieback in Very Defoliated Trees of Two Pine Species. Front. Plant Sci. 7:418. doi: 10.3389/fpls.2016.00418 when superimposed on episodes of scarce precipitation, result in severe water deficits intensifying drought impact and reducing forest growth and productivity (Williams et al., 2013). Consequently, warmer and drier conditions could lead to growth decline of drought-prone Mediterranean conifer forests (Sarris et al., 2007; Sánchez-Salguero et al., 2012a; Galván et al., 2014). Such reductions in productivity may predispose trees to droughtinduced dieback once growth decline and vigor loss become irreversible (Camarero et al., 2015). Thus, it is compelling to determine if Mediterranean forest growth recovers after successive or severe droughts and if this response may cause a loss of resilience.

Severe droughts are extreme climatic events and therefore they constitute rare and unpredictable drivers of forest dynamics (Gutschick and Bassirirad, 2003). However, to account fully for droughts effects on forest growth, their extremity must be documented not only from the climatic perspective but also from the tree response (Smith, 2011). Dendrochronology facilitates the assessment of drought impacts on radial growth by reconstructing tree-ring variables since the unpredictability of droughts makes their continuous surveillance challenging (Dobbertin, 2005; Eilmann et al., 2013). Growth decline and dieback represent long-lasting impacts of severe droughts on forest productivity (McDowell et al., 2008). Tree-ring width records usually reflect a growth reduction in response to prolonged droughts before crown decline symptoms (needle loss and yellowing) appear (Torelli et al., 1986, 1999; Pedersen, 1998; Bigler et al., 2006). Frequently, conifers also show a high growth responsiveness to water availability previous to drought-triggered needle loss or tree death (Ogle et al., 2000). In addition, summer drought is also associated with more conspicuous symptoms of vigor loss as accelerated defoliation (Solberg, 2004). Furthermore, droughts cause legacy effects on tree growth thus compromising forest resilience (Anderegg et al., 2015).

Multiple dieback episodes in Mediterranean conifer forests subjected to long dry spells confirm that pine species are particularly prone to drought-induced growth decline, needle loss or defoliation and mortality (Martínez-Vilalta and Piñol, 2002; Sarris et al., 2007, 2011; Sánchez-Salguero et al., 2012a; Camarero et al., 2015). The vulnerability of some Mediterranean pines to drought stress can be explained because they are tall species (compared with co-occurring shrubby or small conifers such as junipers), display high leaf areas, show isohydric behavior characterized by a rapid stomatal closure in response to drought and present a high xylem vulnerability to embolism (McDowell and Allen, 2015). In fact, some strictly Mediterranean pine species (e.g., Pinus halepensis) apparently well adapted to withstand drought stress (Klein et al., 2011) can show dieback under extremely dry and warm conditions (Camarero et al., 2015; Dorman et al., 2015). This raises the question on how resilient will be pine species from different biogeographical origins to drought-induced dieback.

Here, we compare the post-drought growth responses of two pine species with Eurosiberian (Scots pine) and Mediterranean (black pine) distributions to provide a measure of resilience in similar drought-prone forests. We capitalize on a droughtinduced dieback caused by severe late-20th century droughts affecting Spain (1994–1995, 1999, 2005) and leading to growth decline and enhanced defoliation in pine plantations located in SE Spain (Sánchez-Salguero et al., 2012a,b). We quantify the post-drought growth responses at inter- (tree-ring width) and intra-annual (xylem development or xylogenesis) scales in three defoliation classes since needle loss is a proxy of postdrought changes in tree vigor (cf. Dobbertin, 2005). Our specific aims are: (i) to quantify the post-dieback growth trends; (ii) to characterize xylogenesis; and (iii) to examine climate-growth associations. We explicitly fulfill these objectives by comparing three defoliation classes of the two pine species. We hypothesize that the most defoliated trees will show the lowest growth rates and the highest sensitivity to water availability, i.e., the lowest post-drought resilience capacity. It is also expected to detect this pattern more clearly in Scots pine than black pine since the former species is more vulnerable to drought-induced xylem embolism (Martínez-Vilalta et al., 2004).

## MATERIALS AND METHODS

#### Study Area and Tree Species

The study area is located in the Sierra de Filabres, Andalusia, SE Spain (37◦ 22′ N, 2◦ 50′ W; see Figure S1). This area was planted with Scots pine (Pinus sylvestris L.) and black pine (Pinus nigra Arn.) in the 1970s, with most stands of each species located at approximate elevations of 1850–2000 m and 1700–1850 m a.s.l., respectively (Sánchez-Salguero et al., 2012b; Herrero et al., 2013). Both Scots pine and black pine stands are among the southernmost planted forests of both species (Figure S1). Therefore, these populations can be considered marginal from both biogeographical (southernmost stands) and climatic (xeric limit) points of view, particularly in the case of Scots pine (Barbéro et al., 1998). The climate is Mediterranean of semi-arid type since the mean annual temperature is 13.4◦C and the annual rainfall ranges between 350 and 450 mm. These data are based on a regional climate series calculated for the period 1970–2008 using daily and monthly climate data (mean maximum and minimum temperature, precipitation) obtained from several local stations (see Figure S2 and Table S1). For these stations we also estimated the potential evapotranspiration (PET) using values of mean temperature and solar radiation (Hargreaves, 1983). Then, we calculated the daily water balance as the difference between precipitation and PET. To characterize drought severity in the study area since the 1970s we obtained monthly values of the Standardised Precipitation-Evapotranspiration Index (SPEI) calculated for 3-, 6-, and 12 month long scales since these are the most important scales for the study pine species (Pasho et al., 2012). Negative and positive SPEI values indicate dry and wet conditions, respectively. The SPEI was calculated for the 0.5◦ grid including the study sites and it was obtained from the webpage http://sac.csic.es/spei/index. html. The topography of the study sites is characterized by steep slopes (>35%). Geological substrates are Paleozoic schist and quartzites leading to regosols soil types.

The study trees are located in plantations managed through selective thinning, which involves harvesting the dominated trees while retaining those within specified size classes for future natural seeding. The current density was similar across the study area, with a mean basal area of 25 m<sup>2</sup> ha−<sup>1</sup> . More details on the study sites are available in Sánchez-Salguero et al. (2012a,b).

#### Field Sampling and Tree Selection

A stratified sampling was followed to select trees showing contrasting defoliation after the severe 2005 drought. Firstly, a systematic forest inventory was performed to select stands whose trees shared similar site conditions (soil, topography) but presented contrasting defoliation degrees (Sánchez-Salguero et al., 2012b). Secondly, 30 trees were selected for each species (10 trees per defoliation class) based on their similar size (dbh, diameter measured at 1.3 m, tree and crown heights) and age (mean ± SE = 32 ± 3 years; see also **Table 1**). For each sampled tree, the proportion of crown cover was estimated to the nearest 5% by comparing every tree with a reference tree with the maximum amount of foliage at each site (Schomaker et al., 2007). Finally, the trees were classified in three defoliation classes: defoliation ≤ 25% of the crown (scarcely or not defoliated trees, henceforth abbreviated as N trees), 25 < defoliation < 75% (trees with intermediate defoliation level, henceforth abbreviated as I trees), and defoliation ≥ 75% (highly defoliated trees, henceforth abbreviated as D trees).

#### Dendrochronological Sampling and Processing

Dendrochronological sampling was carried out in winter 2008. Two cores per tree at breast height were collected at 1.3 m from 30 trees per species by using a Pressler increment borer. The cores were air dried, stuck onto wood guides with glue, and sanded using progressively finer grain papers until the rings were clearly distinct (Fritts, 2001). Then, tree rings were visually cross-dated and measured with a resolution of 0.01 mm using a semi-automatic LINTAB device (F. Rinn, Heidelberg, Germany). Cross-dating was checked using the COFECHA software (Holmes, 1983).

Since basal-area increment (BAI, cm<sup>2</sup> year−<sup>1</sup> ) is assumed to be a meaningful indicator of tree growth because it removes variation in growth attributable to increasing circumference and it is related to the transpiring crown surface (e.g., Linares et al., 2009), we converted tree-ring widths into BAI. We assumed a circular shape of stem cross-sections and used the following formula:

$$\text{BAI} = \pi (R\_t^2 - R\_{t-1}^2) \tag{1}$$

where R is the radius of the tree and t is the year of tree ring formation. Mean series of BAI were obtained for the two species and the three defoliation classes.

To calculate climate-growth relationship at inter-annual scales we transformed tree-ring widths into indices following standard dendrochronological procedures (Fritts, 2001). The individual tree-ring width series were double-detrended using negative linear or exponential functions and cubic smoothing splines with a 50% frequency-response cut-off at 20 years to preserve high-frequency variability. Observed width values were divided by fitted values to obtain ring-width indices. Autoregressive modeling was performed on each detrended ring-width series to remove part of the first-order autocorrelation. Then, these indices were averaged using a biweight robust mean to obtain residual chronologies for each species and defoliation class. All chronologies were built using the program ARSTAN (Cook and Krusic, 2005). Finally, we calculated two dendrochronological statistics (first-order autocorrelation, mean sensitivity) and compared their mean values between defoliation classes to assess growth patterns.

#### Xylogenesis

Tree rings are produced by the cambium which generates tracheids differentiating through developmental stages (radial enlargement, wall thickening) until becoming mature (Mahmood, 1971; Wodzicki, 1971, 2001; Larson, 1994). This process of xylem development (xylogenesis) was monitored by sampling wood micro-cores (2 mm in diameter, 1–2 cm in length) from March until mid-October 2008. Sampling was done biweekly in spring and monthly from August onwards in five trees per defoliation class (half of the trees used for dendrochronological analyses) of the two pine species. Samples were taken around the stems at 1.3 m using a Trephor increment puncher (Rossi et al., 2006). The thick dead outer bark was removed, and sampling positions were arranged along an ascending semi-helical pattern in the stem (Deslauriers et al., 2003). The micro-cores were taken about 1 cm apart from each other to avoid wound reaction. The samples usually contained the preceding 4–5 tree rings and the developing annual layer with the cambial zone and adjacent phloem.


Values are means ± SE. Different letters show significant differences (P < 0.05) between defoliation classes within each species according to S-N-K post-hoc tests.

Micro-cores were placed in Eppendorf tubes containing a mixed solution of formaldeid, acetic acid and ethanol (5:5:90) and stored as soon as possible at 5◦C in order to avoid tissue deterioration. All samples were then processed within a maximum of 1–2 weeks after sampling. Micro-cores were sectioned using a sledge microtome (Anglia Scientific AS 2000, UK) achieving samples 20-µm thick. Sections were mounted on glass slides, stained with 0.5% water solution of cresyl fast violet, fixed with Eukitt <sup>R</sup> and observed at 100–200x magnification under a light microscope (Olympus BH2).

Four different xylogenesis phases were identified as follows (cf. Wodzicki, 1971; Antonova and Stasova, 1993; Wodzicki, 2001; Deslauriers et al., 2003): (1) cambial cells characterized by small radial diameters, thin walls and bone shape; (2) radially enlarging tracheids presenting unlignified cell walls and therefore unstained in blue; (3) wall-thickening and lignified tracheids with a transition coloration from violet to dark; and (4) mature cells with lignified cells walls fully stained in blue. We counted separately earlywood and latewood mature tracheids and distinguished them by their thin cell walls and wide lumens and thick walls but narrow lumens, respectively. The numbers of cells in each of the four different phases were counted along five radial rows to obtain a mean value per ring and sampling date.

To relate xylogenesis with microclimatic conditions measured in situ several climatic variables (air temperature, precipitation, solar radiation, air relative humidity) were recorded hourly and then converted to daily values (mean temperature and radiation, precipitation) using a HOBO microclimate station (Onset, Pocasset, USA) located in each pine stand (see Figure S4).

## Timing of Wood Formation

Tracheid differentiation was considered to have started and to be complete when at least one horizontal row of cells was detected in the enlarging phase and cell wall thickening and lignification were completed, respectively (cf. Gruber et al., 2010). To precisely define and compare xylogenesis between defoliation classes we computed the onset and cessation dates and the duration of three developmental phases (E, radial enlargement; L, cell-wall thickening and lignification; M, tracheid maturation) using the package CAVIAR in R (Rathgeber et al., 2011). The onset and cessation dates were defined when 50% of the radial files were active (onset) or non-active (cessation) in each xylogenesis phase. The durations of each phase were calculated as the time elapsed between the onset and cessation of these phases following Rathgeber et al. (2011). Xylem formation (X phase) was defined as the time elapsed between the onset of enlargement and the end of maturation. Finally, to compare the onset and cessation dates and the duration of the main phases of xylogenesis we used the achieved significance level (ASL), which can be interpreted in the same way as a P significance level since the smaller the value of ASL, the stronger the evidence against a null hypothesis considering no difference between dates or phase durations (Efron and Tibshirani, 1993).

## Statistical Analyses

Growth-climate relationships were quantified by calculating Pearson correlation coefficients between daily climate data (mean maximum and minimum temperature, total precipitation, water balance) and ring-width indices. To detect time-dependent growth responses to climate, daily regional climate data were either averaged (temperature) or summed (precipitation, water balance) at 10-day and 15-day long scales following Sánchez-Salguero et al. (2015).

The associations between climate and xylogenesis data (number of cambium cells or tracheids in different developmental stages) were evaluated at 5-, 10-, and 15-day long time scales since daily dynamics of tracheid radial expansion have been described in Scots pine (Antonova et al., 1995). In this analysis we used local climate data recorded in the field during 2008 (mean temperature, precipitation, radiation, relative humidity, water balance). We used linear-mixed effects models to evaluate the effects of defoliation and climate variables on the (x 0.5-transformed) number of different types of tracheids along time, and checked the predicted values and residuals looking for signals of heteroscedasticity (Zuur et al., 2009). Defoliation was regarded as a fixed factor, whereas tree was considered a random factor. Comparison between mean values of tree features (defoliation, size variables) or dendrochronological statistics were based on applying S-N-K post-hoc tests. We fitted linear mixed-effects models using the nlme library (Pinheiro et al., 2015). All analyses were done using the R statistical program version 3.120 (R Development Core Team, 2015).

## RESULTS

## Post-Drought Growth Patterns

Basal area increment (BAI) dropped in all trees during the dry years 1994–1995, 1999, and 2005 (**Figure 1**; see also Figure S3). These BAI reductions were followed by a relatively rapid recovery after 1994 and 2005, but not after 1999 in the case of highly defoliated (D) trees of both pine species (**Figure 1**). We found that BAI for the 2000−2008 period was significantly lower (P < 0.05) in the case of D trees (Scots pine, mean ± SE = 2.2 ± 0.4 cm<sup>2</sup> ; black pine, 1.6 ± 0.3 cm<sup>2</sup> ) as compared with trees presenting intermediate (I) or low (N) defoliation levels whose mean BAI values did not differ (means for I-N trees: Scots pine, 4.3 ± 0.7 cm<sup>2</sup> ; black pine, 5.1 ± 0.9 cm<sup>2</sup> ). These differences were not associated to tree size which did not differ between defoliation classes (**Table 1**). Note also that in Scots pine the D and I trees already grew less than the rest of trees in the 1980s and early 1990s, which was not observed in black pine.

Differences in growth between defoliation classes could be traced back in time. Only the D trees showed a lower tree-ring width than the other defoliation classes considering the 1985– 2008 period, and this difference was more evident in black pine (**Table 2**). In Scots pine, the non-defoliated N trees showed the highest first-order autocorrelation in tree-ring width but the lowest mean sensitivity, whereas in black pine the most defoliated D trees presented the highest mean sensitivity (**Table 2**).

## Climate-Growth Relationships

We found the highest climate-growth correlations when considering mean maximum temperatures and precipitation, which determine water availability during the growing season.

TABLE 2 | Dendrochronological statistics of tree-ring width series for the studied trees and defoliation classes calculated considering the common 1985–2008 period (values are means ± SE).

the annual number of measured trees (right y axis, sample depth) for each defoliation type (same colors are used for growth and sample depth data).


Statistics refer to raw data excepting mean sensitivity which was calculated considering residual indices. Different letters show significant differences (P<0.05) between defoliation classes within each species according to S-N-K post-hoc tests.

<sup>a</sup>The first-order autocorrelation of raw ring-width data measures how much the ring width in year n is correlated with the width in year n-1; the mean sensitivity of residual tree-ring width series measures the relative year-to-year variability in width of consecutive tree rings.

Therefore, we present only results for these two climatic variables (**Figure 2**). In Scots pine, growth was enhanced by wet January and mid-June conditions, with the strongest effect for the latter variable in the case of D trees and at 15-day long intervals (**Figure 2**). Warm mid-June conditions were associated to low growth in Scots pine, regardless its defoliation level, but high maximum temperatures in early August averaged at 10-day long intervals benefitted growth. In black pine, too warm conditions in early to mid-June were negatively associated to growth of D and I trees, whilst high precipitation values in January and also from May to July were positively associated to growth (**Figure 2**). In the case of D black pine trees, their growth was most strongly enhanced by June precipitation summed at 10-day long intervals, but a similar response was observed in I trees for July rainfall accumulated at 15-day long intervals.

#### Xylogenesis and Tree Defoliation

The D trees produced less tracheids than N trees in the radial enlargement, wall-thickening and lignification, and maturation phases (**Figure 3**). For instance, on average the D trees produced from 12 (black pine) to 15 (Scots pine) tracheids per tree-ring, whereas the N trees produced 26 (black pine) to 37 (Scots pine) tracheids. These observations were confirmed by the linear mixed-effects models which evidence that defoliation intensity was significantly related to a lower production of radially-enlarging and mature tracheids in both tree species (**Table 3**). In black and Scots pine, warmer and drier conditions at 15-day long scales were negatively related to the production of radially enlarging tracheids, whereas radiation was positively related to their production (**Table 3**). However, in Scots pine, the production of cambial cells was positively associated to higher temperatures. Warmer and drier summer conditions enhanced the production of lignifying tracheids.

In both pine species the number of cambial cells of N trees reached the highest value in May, but peaked 1 month later in the case of D trees (**Figure 3**). This means that the onset of xylem formation started earlier in the N than in the D trees. In the case of the radial-enlargement phase, the onset occurred significantly earlier in N than in D trees (**Table 4**). The peak of formation of radially-enlarging tracheids occurred from May to June in Scots pine and around mid-June in black pine. This phase ended before in the D trees than in the other defoliation classes in


both pine species (**Table 4**, **Figure 4**). Consequently, the D trees were characterized by presenting the shortest duration of the radial-enlargement phase, but this difference was only significant in black pine (D, 128 days vs. N trees, 160 days; **Table 4**). The wall-thickening and lignification phase was similar among defoliation classes showing a maximum activity from June to July in both pine species. Lastly, maturation proceeded similarly in the three defoliation classes, albeit latewood formation seemed to start earlier in the case of Scots pine D trees but we could not assess if there were significant differences. Overall, the duration of xylem formation was shorter in D trees than in the other types of trees.

## DISCUSSION

Here we document how drought-induced dieback caused growth decline and affected xylogenesis in the widely distributed Scots pine and the Circum-Mediterranean black pine. The duration of xylem formation was shorter in the defoliated than in the non-defoliated or moderately defoliated trees in both species (**Table 4**), which agrees with the fact that the most defoliated trees grew less and thus produced less tracheids than the other types of trees (**Figures 1**, **3**). Defoliated pines form narrow tree rings as the result of a shorter growing season due to a later onset of xylogenesis or a premature cessation of cambial activity

and wood formation as has been observed by others (Bauch et al., 1979; Eilmann et al., 2011, 2013). Note that the duration of xylogenesis phases as the radial enlargement of tracheids, which depends on an adequate turgor pressure of expanding cells (Abe et al., 2003), drives to a great extent the annual ring width and the final size of tracheids (Cuny et al., 2014). However, we could only find a significantly shorter duration of the radial-enlargement phase in the most defoliated black pine trees as compared with their less defoliated conspecifics (**Table 4**).

The long-term climate-growth associations showed a pronounced sensitivity of growth in defoliated Scots pine trees to changes in precipitation during the growing season (**Figure 2**). This agrees with the fact that Scots pine is more vulnerable to drought-induced xylem embolism than black pine (Martínez-Vilalta et al., 2004). This species was also the most negatively affected by the 1990s and 2000s droughts in the study area (Sánchez-Salguero et al., 2012a). Such sensitivity to water availability also agrees with the highest year-to-year variability in growth presented by the most defoliated Scots

TABLE 4 | Statistical tests (ASL) obtained by comparing the estimated onset and cessation dates and the duration of the main phases of tracheid differentiation for *P. sylvestris* and *P. nigra* trees of different defoliation classes (D, severely defoliated; I, intermediate defoliation; N, scarcely or not defoliated).


Significant (P < 0.05) ASL values, based on 10,000 bootstrapped iterations, appear in bold characters. Note that the smaller the value of ASL, the stronger the evidence to support a significant difference.

pine trees (**Table 2**). From this point of view, Scots pine could be considered less adapted to global-change-type droughts than black pine. However, climate-xylogenesis associations did not indicate greater drought sensitivity in Scots pine as compared with black pine (**Table 3**). Scots pine responds to drought by a fast reduction of transpiration through a rapid stomatal closure (Irvine et al., 1998), but this response varies between trees as a function of their stress level (Hölttä et al., 2012). Such isohydric behavior combined with needle shedding could compensate the alterations in source-sink relationships within the tree (Iqbal et al., 2012). However, growth data indicate that growth decline is irreversible in this species, and also in black pine, for very high defoliation levels and after three severe droughts as those which occurred in 1994–1995, 1999, and 2005 (**Figure 1**). Overall, our findings suggest that defoliated trees regulated their water status after the severe 1990s and 2000s droughts by needle shedding so as to keep a stable ratio between conductive area and transpiring area. The most defoliated trees presented the lowest growth rates prior to the droughts but we do not know if they were those transpiring most actively and therefore losing more water through their stomata. Whatever the cause, such low-growth trees were the most prone to drought-induced alterations in their hydraulic system, defoliation, changes in xylogenesis, and reduced growth after the drought. Drought-induced severe defoliation possibly portends tree death in the most affected trees.

The association between defoliation and a reduced growth rate was observed for all the assessed xylogenesis phases in black pine, and for the radially-enlarging and mature tracheids in Scots pine (**Figure 3**, **Table 3**). This is consistent with the finding that the most defoliated black pine trees were characterized by presenting

defoliation class).

the shortest duration of the radial-enlargement phase, a stage which is tightly related to the growth rate of trees (Horacek et al., 1999). A reduced number of enlarging cells was also found when imposing water deficit on black spruce (Picea mariana) saplings under controlled conditions (Balducci et al., 2013). Notably, the ratio between the numbers of total mature tracheids in the non-defoliated as compared with the most defoliated trees was similar between the two species (ca. 2.5). This could indicate that drought-triggered defoliation caused a similar growth reduction in both species despite the aforementioned differences regarding xylogenesis.

It is also remarkable that observational (Torelli et al., 1986) and empirical (Balducci et al., 2016) studies did not detect clear modifications of transversal tracheid dimensions in silver fir (Abies alba) trees showing dieback or in black spruce saplings experiencing imposed water deficit, respectively. However, the stem wood of defoliated silver fir trees showed a higher susceptibility to decay implying different lignifications processes (Shortle and Ostropsky, 1983). In fact, silver fir trees showing dieback produced narrow rings due to a premature end of wood formation characterized by an earlier differentiation of the latewood cell walls as compared with non-declining trees (Torelli et al., 1999). We could not find significant differences between defoliation classes regarding the cell-wall thickening and lignification phase, despite there was a trend toward an earlier cessation of this phase in the most defoliated trees, particularly in the case of black pine. These results suggest that xylogenesis is more sensitive or plastic to water shortage and defoliation than wood anatomy. However, both responses are not mutually exclusive. Dry conditions can stop cambial activity, shorten the growing period, and also induce the formation of tracheids with wider conduits and thinner cell walls as has been described in Scots pine adult trees (Eilmann et al., 2011).

The plasticity of xylogenesis in response to seasonal or punctual water shortage has been profusely documented in empirical and observational studies. In a warming and drought experiment considering black spruce saplings, water shortage reduced the rates of cell production (Balducci et al., 2016), as previously observed in experiments with Aleppo pine saplings (De Luis et al., 2011). Field studies also reported plastic responses of cambial activity to seasonal water availability (e.g., bimodal behavior) in either Mediterranean species as Aleppo pine (De Luis et al., 2007; Camarero et al., 2010) or Eurosiberian species as Scots pine growing at xeric sites in the Alps (Gruber et al., 2010; Eilmann et al., 2011; Oberhuber et al., 2011). In a rainfall exclusion experiment applied to Scots pine, the radialenlargement phase was shortened in trees subjected to drier conditions compared with control trees although this difference was not significant and depended on tree size (Fernández-De-Uña et al., 2013). Overall, these studies indicate that cambial activity is greatly reduced by drought but can rapidly resume once water availability increases (Eilmann et al., 2011, 2013).

Plastic xylogenesis could represent a strategy to respond to changes in water availability and the reduction of photosynthetic area through leaf shedding. In a Scots pine forest located at a xeric site in the Swiss Alps and subjected to a drought trees with medium to high defoliation grew less and showed a shorter growth period than non-defoliated trees (Eilmann et al., 2013), which fully agrees with our findings. When irrigation occurred in this site all trees responded positively and rapidly showing enhanced radial growth and stopping needle shedding, irrespective of their defoliation degree, which is in accordance with previous studies (Dobbertin, 2005). The reduction of the production rates of tracheids in different xylogenesis phases of the most defoliated trees, particularly when cells are radially enlarging, was not compensated by longer durations of these phases as was suggested in a drought experiment (Balducci et al., 2016). Following this line of reasoning, it has been suggested that drought and subsequent defoliation characterizing dieback episodes could lead to the depletion of carbon stores (Galiano et al., 2011). However, other authors indicate that water shortage is a more relevant and direct constrain of growth than a reduced availability of non-structural carbohydrates (Sala et al., 2012). In fact, defoliation caused by insects such as the pine processionary moth (Thaumetopoea pityocampa), which particularly affects black pine, reduce radial growth but not the concentrations of non-structural carbohydrates (Palacio et al., 2012; Puri et al., 2015). Seasonal changes in sugar concentration within the cambial zone have been linked to xylogenesis and peak when most radial growth is finished, i.e., during the wall-thickening and lignifications phase (Simard et al., 2013). Nonetheless, this does not mean that cambial activity of trees is directly limited by the availability of carbohydrates because drought can lead to the use of soluble sugars for osmoregulation but also reduce cell turgor, expansion, and lignification as well as related cambial dynamics (Deslauriers et al., 2014).

We found the reported differences in xylogenesis in the most defoliated trees 3 years after the severe 2005 drought induced dieback and triggered needle loss (Sánchez-Salguero et al., 2012b). This implies that once crown defoliation reaches a threshold (in this case above 75%) the ability of trees to recover growth could be compromised in some pine species or at xeric sites if water availability improves. Globally, legacy effects of droughts cause lags of 2–4 years for the recovery of forest growth (Anderegg et al., 2015). We show that defoliation could lengthen these recovery periods and compromise the resilience of forests experiencing drought-induced dieback. Lastly, we extract a practical lesson of this study. Xylogenesis studies are highly time consuming. Nevertheless, to reach more general and robust conclusions we suggest sampling 10 trees per defoliation class in further studies. Sampling could be done weekly during the most active growing period (for instance from April to July in our case) and biweekly the rest of the year.

To conclude, drought negatively impacted growth and crown cover in Scots pine and black pine. In the most defoliated trees, the duration of xylem formation and the radial-enlargement phase shortened leading to low growth rates and the formation of narrow rings. In Scots pine, the onset of xylem formation was retarded in the most defoliated trees as compared to nondefoliated trees. Despite the widely reported plastic responses of cambial activity to changing water availability, we found a very limited resilience capacity of Scots and black pines after drought in severely (≥75%) defoliated trees. Moreover, droughts produce legacy effects on xylogenesis of these severely defoliated trees which show irreversible growth decline and are prone to die.

#### AUTHOR CONTRIBUTIONS

GG and RS performed the data collection in the field and contributed significantly to data analysis, discussing the results, and writing of the paper. JC conducted the statistical analysis and the manuscript redaction. RN has contributed to the study design and the manuscript discussion and approval.

### ACKNOWLEDGMENTS

We thank the support of Junta de Andalucía and project GESBOME (P06-RNM-1890) and AEMET (Agencia Estatal de Meteorología) for providing meteorological data. JC acknowledges the support of the Excellence Network "Red de Ecología Terrestre para afrontar los retos del Cambio Global— ECOMETAS" (CGL2014-53840-REDT) of the Spanish Ministry of Economy. RS thanks the support of project CoMo-ReAdapt (CGL2013-48843-C2-1-R) and postdoctoral fellowship of (FEDER-Programa de Fortalecimiento de las capacidades en I+D+i de las Universidades 2014-2015, Junta de Andalucía). RN acknowledges the support of the project LIFE13 ENV/ES/001384 "Development of technical guidelines for carbon sequestration and dynamization of carbon compensation in forests" and QUERCUSAT (CLG2013-40790-R). GG acknowledges the granting of a predoctoral fellowship from FPI program (BES-2011-050172) by the Spanish Ministry of Economy and Competitivity (Project BFU 2010-21451). This work was carried out under the framework of the EU COST Action FP 1106 (Studying Tree Responses to extreme Events: a SyntheSis, STReESS).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00418

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Guada, Camarero, Sánchez-Salguero and Navarro Cerrillo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Biological Basis of Tree-Ring Formation: A Crash Course

#### Cyrille B. K. Rathgeber<sup>1</sup> \*, Henri E. Cuny<sup>2</sup> and Patrick Fonti<sup>2</sup>

<sup>1</sup> LERFoB, INRA, AgroParisTech, Nancy, France, <sup>2</sup> Swiss Federal Institute for Forest, Snow and Landscape Research, Birmensdorf, Switzerland

Wood is of crucial importance for man and biosphere. In this mini review, we present the fundamental processes involved in tree-ring formation and intra-annual dynamics of cambial activity, along with the influences of the environmental factors. During wood formation, new xylem cells produced by the cambium are undergoing profound transformations, passing through successive differentiation stages, which enable them to perform their functions in trees. Xylem cell formation can be divided in five major phases: (1) the division of a cambial mother cell that creates a new cell; (2) the enlargement of this newly formed cell; (3) the deposition of its secondary wall; (4) the lignification of its cell wall; and finally, (5) its programmed cell death. In most regions of the world cambial activity follows a seasonal cycle. At the beginning of the growing season, when temperature increases, the cambium resumes activity, producing new xylem cells. These cells are disposed along radial files, and start their differentiation program according to their birth date, creating typical developmental strips in the forming xylem. The width of these strips smoothly changes along the growing season. Finally, when climatic conditions deteriorate (temperature or water availability in particular), cambial activity stops, soon followed by cell enlargement, and later on by secondary wall deposition. Without a clear understanding of the xylem formation process, it is not possible to comprehend how annual growth rings and typical wood structures are formed, recording normal seasonal variations of the environment as well as extreme climatic events.

#### Edited by:

Ute Sass-Klaassen, Wageningen University, Netherlands

#### Reviewed by:

Dieter Eckstein, University of Hamburg, Germany Pieter Baas, Naturalis Biodiversity Center, Netherlands

#### \*Correspondence:

Cyrille B. K. Rathgeber cyrille.rathgeber@nancy.inra.fr

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 27 January 2016 Accepted: 12 May 2016 Published: 26 May 2016

#### Citation:

Rathgeber CBK, Cuny HE and Fonti P (2016) Biological Basis of Tree-Ring Formation: A Crash Course. Front. Plant Sci. 7:734. doi: 10.3389/fpls.2016.00734 Keywords: cambial activity, tree growth, tree-ring structure, quantitative wood anatomy, climatic factors, xylem, lignin, cellulose

## INTRODUCTION

Forests are the most widely distributed biomes on earth. They cover one third of the emerged lands, host more than 50% of the world's biodiversity, and contain more than 60% of the terrestrial carbon pool (Groombridge and Jenkins, 2002). A large part of this carbon is stored in wood, the most abundant biological compound on earth.

The exchanges of carbon dioxide between the forest ecosystems and the atmosphere are crucial processes that influence the balance and the dynamics of the global carbon cycle. Terrestrial plant's photosynthesis (i.e., atmospheric CO<sup>2</sup> uptake by the leaves) captures about 120 petagrams of carbon per year (Lal, 2008). Plant's growth fixes half of this carbon (i.e., 60 petagrams) in form of biomass, while the other half is released back in the atmosphere by autotrophic respiration. Concomitantly, plant's transpiration releases 40 000 petagrams of water into the atmosphere,

influencing global precipitation and heat flux (Seneviratne et al., 2006). Globally, the forest ecosystem photosynthesis draws more carbon from the atmosphere than auto- and hetero-trophic respiration pump into it. So, at the global scale, world's forests constitute a large and persistent net sink of carbon (Pan et al., 2011). Moreover, the sequestration of carbon into forest's woody biomass partially counterbalances the current increase of anthropogenic emissions, slowing down climate warming (Bonan, 2008).

Up to now, scientists have considered photosynthesis as the main driver of plant's growth and have put much effort in better understanding this process. However, another view, which emerged recently, is claiming that, under normal conditions, it is not the source (i.e., photosynthesis) that limit plant's growth, but the sinks (i.e., the ability of meristems to convert carbon into biomass). In other words, carbon can only be sequestered into wood to the extent cambial activity and environmental conditions permit it (Körner, 2015). However, if after decades of researches on photosynthesis, source activity is very well known and quantified nowadays, cambium functioning is still poorly understood. So, in the context of global warming acceleration, we believe, it is crucial to investigate what is ruling tree-ring formation and wood production, in order to better evaluate how climatic changes are impacting trees, forests, biogeochemical cycles, and ultimately the climate itself.

The process of xylem formation carried out by woody plants is called xylogenesis. The monitoring of the seasonal dynamics of xylogenesis started, more than 50 years ago, with few pioneering works, which aimed to better understand the influence of climate on tree growth, cambium phenology, and wood formation dynamics (Wilson, 1970; Denne and Dodd, 1981). These questions received a renewed attention during the last decade because of the pervasive problem of global changes, leading to a rapid increase in the number of scientific studies involving trees growing in natural or experimental conditions (Griçar et al., 2011). Wood formation monitoring studies are based on repeated (weekly or bi-weekly) cytological observations of the developing xylem all along the growing season. In this mini review, we will briefly present the basics of xylogenesis, along with the current knowledge about the influence of the environmental factors on the cellular processes, the intra-annual dynamics, and the phenology of wood formation, in order to help ecologists to better interpret results from wood formation monitoring studies.

#### WOOD STRUCTURES AND FUNCTIONS

Wood performs four essential functions in trees: (1) supporting and spatially distributing the photosynthetic tissues above ground; (2) conducting the raw sap (i.e., water and nutrients) from the roots up to the leaves; (3) storing carbohydrates, water, and other compounds; and finally (4) protecting the tree from pathogens, by storing and distributing defensive compounds (Kozlowski and Pallardy, 1997).

Wood appeared on earth due to the development of the vascular cambium and the invention of lignin (Rowe and Speck, 2005). The vascular cambium is composed of a thin layer of meristematic cells located between the secondary xylem (i.e., the wood) and the secondary phloem (i.e., the living bark), and forming a continuous envelop all around the stems, branches, and roots of woody plants. The cambium gives rise to xylem inward (i.e., toward the pith), and to phloem outward (i.e., toward the bark).

In gymnosperms (i.e., conifers, or softwoods), xylem is made of a simple and homogeneous tissue, mainly composed of two types of cells: (1) tracheids, which represent more than 90% of the total number of cells, and perform both the mechanical support and the water conduction; and (2) parenchyma cells, which are in charge of the storage and radial transport of various compounds. Tracheids are elongated, spindle-shaped cells of 3– 6 mm in length and 6–60 µm in diameter (Sperry et al., 2006). In angiosperms (i.e., hardwoods), xylem is made of a more complex and heterogeneous tissue, composed of several types of cells. Vessels take care of the water conduction, fibers of the mechanical support, and parenchyma cells of the storage. In diffuse-porous trees, vessel dimensions range from 1 to 30 cm in length, and from 15 to 150 µm in diameter; in ring-porous trees, vessel dimensions range from 1 cm to 10 m in length, and from 15 to 300 µm in diameter (Zimmermann, 1983). In both angiosperms and gymnosperms, almost all xylem cells die off at the end of their development to fulfill their functions, only parenchyma cells stay alive for a couple of years.

### CELLULAR PROCESSES INVOLVED IN WOOD FORMATION

Xylogenesis consists in the production and differentiation of new xylem cells into mature functional wood cells. During their differentiation, xylem cells undergo profound morphological and physiological transformations, which will craft them according to their future functions (Wilson, 1970). The formation of a xylem tracheary element can be divided in five major steps: (1) the periclinal division of a cambial mother cell that creates a new daughter cell; (2) the enlargement of the newly formed xylem cell; (3) the deposition of cellulose and hemi-cellulose to build the secondary cell wall; (4) the impregnation of the cell walls with lignin; and finally, (5) the programmed cell death (**Figure 1**). This sequence is common to both angiosperms and gymnosperms but variations in duration and intensity of the differentiation phases, as well as in molecular components involved, finally result in different cell types and tree-ring structures.

**Cell division** is the elementary process through which the cell number is augmented into a forming tissue. In all the cellular organisms that contain a nucleus (i.e., eukaryotes), dividing cells follow a highly controlled sequence of successive events described as the cell cycle. During this cycle, the meristematic mother cell undergoes several stages of development encompassing cellular growth and DNA synthesis, division of the nucleus, and separation of the cytoplasm, in order to give birth to two daughter cells (Lachaud et al., 1999). The process of cell division is slow in the cambium, with cell cycle duration ranging between 10 and 50 days, depending on tree species, developmental stages, and environmental conditions (Larson, 1994). As a result, the

number of cells per developing radial file can only increase by about one cell per day for the most productive trees under the most favorable conditions. Temperature exerts a direct control on cambial cell division, most probably via the polymerization– depolymerisation of the microtubules, a major element of the cell cytoskeleton (Begum et al., 2012). Temperature also influences the division process via hormonal regulation operated by various hormones such as auxins, cytokinins, and gibberellins (Ursache et al., 2013). These phytohormones act in stimulating the synthesis of key proteins: the cyclin-dependent kinases (CDKs), whose enzymatic activity is essential to trigger the start of the cell cycle, and to guarantee its smooth running (Stals and Inze, 2001).

**Cell enlargement** constitutes the first stage of plant cell differentiation. It consists in an irreversible increase of the cell volume (i.e., cell growth) not followed by any cell division. The enlargement of the cell results from (1) the relaxation of the primary cell wall, which (2) creates a passive inlet of water, which (3) is counter-balanced by an active influx of solutes in order to maintain a high turgor pressure (Cosgrove, 2005). The process also requires (4) the biosynthesis and deposition of building material to restore the integrity of the stretched primary cell walls. This process is particularly important for xylem tracheary elements, since their volume is multiplied by 10–100 during this phase. As turgor is the "engine" of cell enlargement, water shortage occasionally affects cell growth. However, under normal conditions, hormonal regulation is the real "driver" of enlargement, determining the final radial diameter of xylem cells. Several phytohormones (e.g., auxins, cytokinins, gibberellins)

increase primary cell wall extensibility through different control pathways (Perrot-Rechenmann, 2010).

**Secondary cell walls** are remarkable structures in many plant cells, but they are of particular relevance for woody plants, providing mechanical support, water transport, and biological resistance. Moreover, secondary cell walls represent the major constituent of wood, which is the most abundant pool of terrestrial biomass. Secondary walls are thick (2–10 µm), poorly hydrated (∼30%), rigid, and multi-layered. Their principal components are celluloses (40–60% of dry mass), hemicelluloses (10–40%), and lignins (15–35%). Cellulose microfibrils together with hemicellulose form the main load-bearing network, in which lignin is impregnated to form another cross-linked network ensuring hydrophobicity, rigidity, and durability (Zhong and Ye, 2009). Secondary cell walls are commonly composed of three layers: S1, S2, and S3, which present a quite similar composition, but differ in the thickness and orientation of their cellulose microfibrils. The S1 layer is composed of a dense matrix of cellulose and hemicellulose microfibrils. While the microfibrils are oriented transversally in the S1 layer (from 60◦ to 80◦ with regard to the cell axis), they change to a longitudinal orientation in the S2 (from 5◦ to 30◦ with regard to the cell axis), before coming back to a transversal orientation in the S3 (Plomion et al., 2001). When cell enlargement comes to an end, secondary cell wall formation starts with the deposition, between the membrane and the primary wall, of the S1 layer, soon followed by the S2 and S3 layers. The secondary wall is not covering the whole cell surface, but is absent around the pits. Here the modified primary wall (the pit membrane) allows the passage of water and solutes from one cell to the next —making the upward sap flow from the root tips to the leaves possible. The formation of the secondary cell wall is a complex developmental process supported by the expression of genes activating the biosynthesis, transport, deposition, and assembly of the wall constituents (Zhong and Ye, 2009). A cascade of transcription factors regulates the coordinated expression of all these genes. Phytohormones are also involved in the regulation of secondary cell wall formation, with auxins acting as inhibitors and brassinosteroids as inducers.

**Cell wall lignification** starts at the cell corners, in the primary wall, at about the same time as the deposition of the S1 layer. Then it extends along the middle lamella, and the primary wall, before progressing inward into the secondary wall following its deposition (Donaldson, 2001). Lignin is polymerized directly into the cell wall from oxidized elementary constituents synthesized into the cytosol from phenylalanine to form complex cross-linked phenol polymers. Lignin is then deposited inside the spaces left by the microfibrils, where it forms chemical bonds with hemicelluloses, acting like cement that reinforce and waterproof the cell walls. As a consequence of the timing of the lignification process and the structure of the walls, the proportion of lignin decreases from the most external layers of the cell walls (i.e., the middle lamella and the primary wall) to the most internal ones (i.e., the secondary wall). This well designed structure perfectly fits the physiological functions of the xylem cells. The heavy impregnation of the middle lamella and primary wall by the rigid and hydrophobic lignin molecules, allows xylem tracheary elements to form strong, rigid and self-supporting networks of waterproof "pipes," while the lighter impregnation of the secondary wall allows tracheary element lumens to keep essential capillarity properties that can support sap ascent. As for the other components of the cell walls, transcription factors play an essential role in the biosynthesis, transport, and deposition of lignin (Zhong and Ye, 2009). Despite the fact that the hormonal control of the lignification process is unclear, it appears that ethylene induces the synthesis of several enzymes involved in lignin biosynthesis. The lignification process is generally described as sensitive to temperature (Donaldson, 2001).

**Programmed cell death** (also called apoptosis) marks the end of xylem cell differentiation and the advent of mature, fully functional, xylem elements (tracheids for gymnosperms, vessels and fibers for angiosperms). It is a highly coordinated and active process of cellular "suicide," which is widespread in multicellular organisms. However, while most cells perform a specific function until their death, xylem tracheary elements die to become functional. In xylem, only parenchyma cells escape programmed cell death and remain alive for several years. The principal trigger of programmed cell death is a massive influx of calcium ions (CA2+) into the vacuole through plasma membrane channels. Death then manifests rapidly (in few minutes or so) as a sudden break-up of the vacuole and the cessation of the cytoplasmic streaming. Moreover, the vacuole break-up releases hydrolases, which attack and degrade the cell organelles and clean the cell content (Bollhöner et al., 2012). After a couple of days, the cell is finally left as an empty space (the lumen) surrounded by a thick wall pierced of pits. In xylem cells, programmed cell death regulation mechanisms appear inextricably linked to those governing secondary cell wall formation. For example, the brassinosteroids that promote secondary wall formation also initiate programmed cell death. In their seminal work, Groover and Jones (1999) proposed a biological mechanism linking together xylem cell apoptosis and secondary wall formation. This mechanism involves the accumulation of a protease in the extracellular matrix during wall material deposition. When the activity of the protease reaches a critical threshold, it triggers the influx of calcium ions, which, in turn, triggers the process of apoptosis. Programmed cell death is an essential step of xylem cell differentiation, allowing mature xylem cells to perform their specific functions in trees. The cell walls (in particular in tracheids and fibers) endow the function of mechanical support to the wood, while the empty cell lumens and the pits (in particular in tracheids and vessels) offer the necessary pathway for water transport into the plant.

## THE SEASONAL DYNAMICS OF WOOD FORMATION

Cambial activity follows the cycle of the seasons (Denne and Dodd, 1981). In extra-tropical regions, the cambium is dormant during winter and active during summer (Delpierre et al., 2015), while in tropical regions it may rest during the dry season and be active during the wet season (Breitsprecher and Bethel, 1990). Annual growth rings and typical tree-ring structures both result from these periodical changes in cambial activity (Evert, 2006).

During winter, the dormant cambium is composed of a few layers of cells (3–6), presenting thickened primary cell walls (**Figure 1A**). Each spring, when day length increases and temperature rises, cambium resumes activity with the division of mother cells (Prislan et al., 2013). During the growing season, the active cambium is composed of numerous dividing cells (6–18), presenting thin tangential cell walls (**Figure 1B**). A couple of days or weeks after the start of cambial cell divisions, newly created xylem cells appeared in the developing xylem. The enlarging cells still only consist of primary cell walls, but present much wider radial diameters than dividing cells. The appearance of these first enlarging cells marks the onset of xylem radial growth and wood formation. A couple of weeks after their birth, these first cells start to thicken, building their secondary walls. Because secondary walls hold most of the biomass, the appearance of the first thickening cells can be seen as the effective beginning of carbon sequestration into wood. Finally, 1 or 2 months after their birth, differentiating xylem cells reached their final mature state. Mature and fully functional xylem treachery elements are composed of thickened secondary cell walls surrounding empty lumens.

During the growing season, new xylem cells, resulting from cambial cell division, are disposed along radial files, and successively undergo the differentiation program phases according to their identity and their place in the "queue" (i.e., the radial file). At the tissue level, the succession in time of cells belonging to the different stages of differentiation is well coordinated between all the radial files, creating a characteristic spatial pattern composed of strip-like developmental zones (**Figure 1B**). Once established, this organization remains rather stable throughout the growing season (Vaganov et al., 2006). Provided that no environmental stress comes into play, cambial activity and xylem radial growth rate generally peak around the summer solstice, when the photoperiod is maximal (**Figure 2**). This period generally marks the transition between earlywood and latewood (Cuny et al., 2014).

At the end of the growing season, in autumn —or even earlier if water or temperature conditions are not favorable anymore— cambial activity stops, soon followed by cell enlargement (flagging the end of stem radial growth). However, the completion of wood formation (marking the end of carbon

sequestration) only occurs a couple of months later (Cuny et al., 2015). Indeed, lignification is a slow process constrained by temperature, so the last xylem cells need up to 2 months for ending cell wall maturation and reaching maturity (Cuny and Rathgeber, 2016).

#### CONCLUSION

The cambium and the developing xylem form a complex dynamic system that periodically produces wood according to the cycle of the seasons. Without a clear knowledge of the biological processes at play in each component of this system, it is not possible to understand how xylogenesis responds to environmental conditions, and how it creates typical tree-ring structures endowing specific functions to the wood. Furthermore, taking into account the interactions between the environmental drivers, the physiological state of the trees, and the developmental stage of the forming xylem, is required to comprehend the creation of the typical tree-ring structures during normal seasonal cycles, as well as special anatomical wood features formed under exceptional conditions (e.g., extreme climatic events).

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

CR wrote the manuscript and prepared the figures, with the assistance of HC and PF.

#### FUNDING

This work was supported by a grant overseen by the French National Research Agency (ANR) as part of the 'Investissements d'Avenir' program (ANR-11-LABX-0002-01, Lab of Excellence ARBRE), and by the Swiss National Science Foundation (project no. 160077, CLIMWOOD). This research also benefited from the support of the FPS COST Action STReESS (FP1106).

#### ACKNOWLEDGMENTS

We would like to thank A. Andrianantenaina and N. Delpierre for helping designing **Figure 2**, as well as the editor and the two reviewers for their accurate and constructive comments, which helped improving the accuracy and clarity of the manuscript.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Rathgeber, Cuny and Fonti. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Assessing Conifer Ray Parenchyma for Ecological Studies: Pitfalls and Guidelines

#### Georg von Arx <sup>1</sup> \*, Alberto Arzac<sup>2</sup> , José M. Olano<sup>3</sup> and Patrick Fonti <sup>1</sup>

<sup>1</sup> Landscape Dynamics Research Unit, Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Birmensdorf, Switzerland, <sup>2</sup> Departamento de Biología Vegetal y Ecología, Facultad de Ciencia y Tecnología, Universidad del País Vasco, Leioa, Spain, <sup>3</sup> Departamento de Ciencias Agroforestales, Escuela Universitaria de Ingenierías Agrarias, Instituto Universitario de Investigación en Gestión Forestal Sostenible-Universidad de Valladolid, Soria, Spain

Ray parenchyma is an essential tissue for tree functioning and survival. This living tissue plays a major role for storage and transport of water, nutrients, and non-structural carbohydrates (NSC), thus regulating xylem hydraulics and growth. However, despite the importance of rays for tree carbon and water relations, methodological challenges hamper knowledge about ray intra- and inter-tree variability and its ecological meaning. In this study we provide a methodological toolbox for soundly quantifying spatial and temporal variability of different ray features. Anatomical ray features were surveyed in different cutting planes (cross-sectional, tangential, and radial) using quantitative image analysis on stem-wood micro-sections sampled from 41 mature Scots pines (Pinus sylvestris). The percentage of ray surface (PERPAR), a proxy for ray volume, was compared among cutting planes and between early- and latewood to assess measurement-induced variability. Different tangential ray metrics were correlated to assess their similarities. The accuracy of cross-sectional and tangential measurements for PERPAR estimates as a function of number of samples and the measured wood surface was assessed using bootstrapping statistical technique. Tangential sections offered the best 3D insight of ray integration into the xylem and provided the most accurate estimates of PERPAR, with 10 samples of 4 mm<sup>2</sup> showing an estimate within ±6.0% of the true mean PERPAR (relative 95% confidence interval, CI95), and 20 samples of 4 mm<sup>2</sup> showing a CI95 of ±4.3%. Cross-sections were most efficient for establishment of time series, and facilitated comparisons with other widely used xylem anatomical features. Earlywood had significantly lower PERPAR (5.77 vs. 6.18%) and marginally fewer initiating rays than latewood. In comparison to tangential sections, PERPAR was systematically overestimated (6.50 vs. 4.92%) and required approximately twice the sample area for similar accuracy. Radial cuttings provided the least accurate PERPAR estimates. This evaluation of ray parenchyma in conifers and the presented guidelines regarding data accuracy as a function of measured wood surface and number of samples represent an important methodological reference for ray quantification, which will ultimately improve the understanding of the fundamental role of ray parenchyma tissue for the performance and survival of trees growing in stressed environments.

Keywords: cutting plane, measured wood surface, measurement accuracy, non-structural carbohydrates (NSC), number of samples, ray density, ray dimensions, ray volume

#### Edited by:

Achim Braeuning, University of Erlangen-Nuremberg, Germany

#### Reviewed by:

Eryuan Liang, Chinese Academy of Sciences, China Francisco Artigas, Meadowlands Environmental Research Institute, USA

> \*Correspondence: Georg von Arx georg.vonarx@wsl.ch

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 25 August 2015 Accepted: 03 November 2015 Published: 18 November 2015

#### Citation:

von Arx G, Arzac A, Olano JM and Fonti P (2015) Assessing Conifer Ray Parenchyma for Ecological Studies: Pitfalls and Guidelines. Front. Plant Sci. 6:1016. doi: 10.3389/fpls.2015.01016

## INTRODUCTION

Parenchyma in the xylem is a neglected living tissue in ecological and eco-physiological research, despite its essential role for tree carbon and water relations and thus tree survival in response to environmental stochasticity. More specifically, it is important for buffering temporal imbalances between carbon uptake and consumption (McDowell and Sevanto, 2010), and has been principally documented to be involved in the storage and transport of water and nutrients (Witt and Sauter, 1994; Gartner et al., 2000). Additionally, parenchyma provides carbohydrates and water for refilling embolized conduits (Salleo et al., 2009; Brodersen and Mcelrone, 2013; Spicer, 2014), is involved in the formation of heartwood (Bamber, 1976), the defense against pathogens (Hudgins et al., 2006), in wounding processes (Arbellay et al., 2010, 2012), and contributes to the mechanical strength of the wood (Burgert and Eckstein, 2001; Fonti and Frey, 2002).

Due to this important role for tree functioning, pioneer studies on parenchyma mainly focused on its anatomical characterization and quantification among different species (e.g., Myer, 1922; Bannan, 1937; Brown et al., 1949). It has been observed that ray parenchyma represents 3–12% and 5–42% of the overall stem xylem tissue in conifers and angiosperms, respectively (Myer, 1922; Panshin and De Zeeuw, 1980; Brandes and Barros, 2008), making ray parenchyma by far the most important living tissue in the xylem despite substantial contribution of axial parenchyma in some angiosperm species (Spicer, 2014). Parenchyma in the stem sapwood stores a large proportion (25–40%) of the overall non-structural carbohydrates (NSC) reserves of a tree, which makes it a more important NSC reservoir than the phloem or the leaves (Jacquet et al., 2014).

Recently, triggered by the increasing awareness of global change impact on forest ecosystems and tree mortality (Allen et al., 2010; Choat et al., 2012; Lloret et al., 2013; Here¸s et al., 2014), there is a renewed and increased need for a better ecological understanding of the role of parenchyma (mainly rays) for tree performance and survival in response to environmental variability (e.g., Olano et al., 2013; Fonti et al., 2015), and particularly with respect to carbon and water balance in stressed trees (Pruyn et al., 2005; Salleo et al., 2006; Esteban et al., 2012; Barnard et al., 2013; Gruber et al., 2013; Rosell et al., 2014). However, the understanding of the ecological role of this tissue has been hampered, mainly due to the scarce data about how parenchyma varies along ecological gradients and/or during tree life, and due to the diversity of methods used for quantification, making comparisons difficult. A review of the measured parameters used for ecological investigations is summarized in **Table 1**. In general, ray properties have been quantified either in terms of "amount" as proxy for investigating variability in tree vigor, growth conditions and storage capacity; and in terms of "spatial distribution" to explore the integration of rays in the 3D xylem network, which is key for its function as storage and transport tissue connecting the xylem and phloem (Spicer, 2014). These observations seem to reveal that within-tree variability is relatively small compared to the variability observed among individuals (Wimmer and Grabner, 2000; Olano et al., 2013; Fonti et al., 2015). Moreover, it has been observed that ray features co-vary at within-tree level related to organ and age, thus reflecting functional needs and/or allometric scaling in relation to variation in conduit size (Carlquist, 1982; Lev-Yadun and Aloni, 1995; Fonti et al., 2015). An open question in this regard is whether ray characteristics might also show some intra-annual variability. Overall, the observed variations suggest a rather strong intrinsic control of the ray characteristics (Aloni, 2013), leaving less room for identifying plastic responses to external environmental factors. Nevertheless, higher percentage of ray tissue appears to be related to tree performance (**Table 1**); and annual time series of anatomical features such as the number of initiating rays and the percentage of ray tissue to respond to environmental conditions (Eckstein, 2013; Olano et al., 2013).

Disentangling the comparably weak environmental signal from the strong intrinsic and allometric component of ray variability requires both a mechanistic understanding of the processes triggering ray formation, and an appropriate methodological toolbox for a sufficiently accurate quantification. However, the relatively few studies performed so far were using different metrics (e.g., ray area, density, size) quantified on different wood cutting planes (i.e., tangential, cross-sectional, or radial planes; see **Figure 1**) of varying measured wood surface, thus hampering comparisons among studies. Consequently, many questions as to the best practice remain unanswered. The studies involving parenchyma quantification have mostly used tangential sections, in which typically a surface of just about 1 mm<sup>2</sup> per sample was measured (e.g., Pruyn et al., 2005; Esteban et al., 2010, 2012), but sometimes surfaces up to 4 mm<sup>2</sup> were also considered (Fonti et al., 2015). Few ecological studies for practical reasons when building up time-series along tree rings—have also quantified rays in cross-sections from 5-mm increment cores (Olano et al., 2013; Arzac, 2014; Fonti et al., 2015), whereas radial sections have hardly been, if at all, used for ray quantification.

In this study we aimed at clarifying some methodological issues related to the quantification of ray tissue in conifer wood and giving some guidelines for ecological studies. In particular we focused on the influence of the methodological approach on the quantification of ray features by comparing several ray metrics in (i) different cutting planes, (ii) within the annual ring (earlywood vs. latewood), and (iii) as a function of the measured wood surface. These aspects were addressed using Pinus sylvestris L. as a model conifer species.

## MATERIALS AND METHODS

#### Study Material

Pinus sylvestris (Scots Pine) is a sub-boreal evergreen conifer with one of the largest distribution ranging between Scotland and northeast Asia (Nikolov and Helmisaari, 1992) at altitudes between 0 to 2700 m asl. As most conifers, it contains uniseriate rays characterized by a single layer of parenchyma cells (Lev-Yadun and Aloni, 1995; **Figure 1**) facultatively embraced by tracheid cells on the upper and lower extremities, and sometimes enclosing resin ducts (Brown et al., 1949).

#### TABLE 1 | Literature review of measured anatomical ray parameters and the inferred ecological interpretation of their variability.


\*Parameter accessible in: C, cross-section; T, tangential section; R, radial section.

Wood samples for this study were obtained in May 2013 in the xeric Pfynwald forest located in the Swiss Rhone Valley (Valais). The climate at this site is continental with 600–700 mm of annual precipitation and a mean annual temperature of 10.1◦C (data from 1981 to 2010 of the weather station of Sion, at 20 km distance from the site, MeteoSwiss). Cores with a diameter of 10 mm were taken at stem breast height from 40 mature Scots pines of 10–13 m height, 12–30 cm dbh, and aged between 45 and 135 years. Ring widths in the cores were measured using a LinTab device and time series were cross-dated using COFECHA (Grissino-Mayer, 2001) to assign each ring to the correct calendar year. The 20 outmost annual rings from all the 40 cores were considered for crosssectional analyses, whereas a subset of six cores was used for additional tangential and radial analyses at three locations each separated by five to seven annual rings. In addition, a single stem disc of 14 cm diameter (58 years old) was collected from a felled individual, split in three radial bars of 20 mm width and 20 mm height at offsets of 120◦ , and used for tangential and cross-sectional analyses in the sapwood rings 1989, 2002, and 2012.

#### Sample Processing and Measurement of Ray Features

Wood sample where processed according to Schweingruber and Poschlod (2005) for the anatomical characterization and quantification of the rays. Therefore, 10–15-µm permanent thin sections were produced with a sledge microtome (Gärtner et al., 2014), placed on a slide and stained with Alcian blue (1% solution in acetic acid) and safranin (1% solution in ethanol) to differentiate between unlignified (blue) and lignified (red) cells. Afterwards, sections were dehydrated using a series of ethanol solutions of increasing concentrations, washed with xylol, and finally permanently preserved by embedding them into Eukitt glue (Gärtner and Schweingruber, 2013). Overlapping images covering the entire samples were captured with a Nikon D90 digital camera mounted on a Nikon Eclipse 50i optical microscope with 40× magnification and merged using PTGUI

v8.3.10 Pro (New House Internet Services B.V., Rotterdam, the Netherlands). In the (subset) of the 40 cores, the average measured width and surface in cross-sectional, tangential and radial sections were 6.69 mm, 4.87 and 9.39 mm<sup>2</sup> , respectively. In the stem disc the average measured width in cross-sections was 19 mm, while the average measured surface in tangential sections was 28 mm<sup>2</sup> .

The outlines of all rays and tree ring borders were vectorized manually in the images using a tailored version of ROXAS 1.6 (von Arx and Dietz, 2005; von Arx and Carrer, 2014), a specialized image analysis tool for wood cell anatomical measurements based on Image-Pro Plus (Media Cybernetics, Silver Spring, Maryland, USA). As a result ROXAS provided several statistics in a global and annual resolution such as individual ray area (all planes), individual ray length (cross- and radial sections) and ray height (tangential sections), the number and position of initiating (NEWRAY) and disappearing rays within the ring (cross-sections), and the percentage of ray surface in the xylem (PERPAR; all cutting planes).

#### Study Design

Several trials were used to identify reliable quantification methods and assess distinctness of ray metrics based on the subset of the six cores. First, the consistency of the measured percentage of ray surface in the xylem (PERPAR) in the three cutting planes (cross-, tangential, radial plane) was evaluated by comparing values from the same annual rings. Second, to explore the potential of more efficient ray quantification, the relationship of individual ray area and ray length (cross-sections) and ray height (tangential sections) was investigated by linear regression analyses. Third, the similarity of different ray parameters in the tangential sections such as PERPAR, ray density, ray area, ray height, and ray width was assessed by correlation analyses. Ray width was estimated from ray area through division by ray height.

The variability of ray parameters was assessed in two different ways. First, the inter-annual variability in crosssectional PERPAR was determined in the full set of 40 cores and time series of 20 rings each by calculating the coefficient of variation (CV) and the mean sensitivity (Cook and Pederson, 2011), i.e., the average percentage change from one yearly value to the next. The same sample set was used to assess the intraannual variability of cross-sectional PERPAR and the proportion of initiating rays (NEWRAY) between early- and latewood using t-test. Second, the variability of the obtained PERPAR values was assessed as a function of the measured wood surface by bootstrapping. Using tangential sections from each of the three locations in the three radial bars of the stem disc, one thousand 1-mm<sup>2</sup> measurement windows were randomly placed and the PERPAR values were extracted and pooled to a single data pool (n = 9000). Bootstrapped values based on 1000 replications for PERPAR mean and coefficient of variation of its estimation (CV) were obtained by randomly combining an increasing number of individual 1-mm<sup>2</sup> values from the data pool, thus simulating the measurement of increasing wood surface (from 1 to 15 mm<sup>2</sup> ). The same procedure was repeated for the cross-sectional images from the same annual rings as the tangential sections for steps of 1-mm width (from 1 to 15 mm). Since mean ring width was around 1 mm (1.047 mm), a measured width of 1 mm in cross-sections corresponded to 1 mm<sup>2</sup> in tangential sections, thus allowing direct comparisons of the obtained variability between the cutting planes. The calculated CVs were used to calculate the relationship between measured wood width/surface, number of samples (n), and estimated accuracy. These calculations were based on Equation (1):

$$\text{Cl95} = 2 \cdot \text{CV} \cdot n^{-0.5} \tag{1}$$

which expresses the relative 95% confidence interval (CI95; given as a percentage of the mean) as a function of CV and n. In fact

## RESULTS

Results from the six cores indicated that the percentage of ray surface (PERPAR) varied substantially depending on the cutting plane. However, PERPAR values within a cutting plane and core were rather similar (mean CV = 0.12, 0.16, and 0.17 for cross-, tangential and radial sections, respectively; **Figure 2**). Mean PERPAR value for cross-sections was 6.50%, which was significantly higher than for the tangential sections (4.92%; t = 5.606; P < 0.001). PERPAR was largest in the radial sections (16.84%), deviating significantly from the values of both other cutting planes (t = −6.287; P < 0.001 and t = −7.641; P < 0.001 for cross-sectional and tangential planes, respectively). Despite the differences, PERPAR values in cross- and tangential sections were significantly correlated (r = 0.502; P = 0.034).

The analysis of the subset of six cores showed that individual ray area was highly correlated with ray length in the crosssectional (n = 2017; R <sup>2</sup> = 0.926; P < 0.001) plane and ray height in the tangential plane (n = 13.218; R <sup>2</sup> = 0.847; P < 0.001). In tangential sections correlation analyses revealed that PERPAR is significantly correlated with all other parameters except for mean ray width (Pearsons's r = 0.196, P = 0.435; **Figure 3**). At the section level we also observed that mean ray

2002, and 2012) along the radial cores of six mature Scots pine trees. Symbols of the same color represent the PERPAR values of a specific cutting. Dotted lines connect the measurements from the same tree and ring. Symbols of the same cutting plane are jittered along the x-axis for easier interpretation. Left upper data indicates the Pearson correlation coefficient (r) between crossand tangential sections (C-T), cross- and radial sections (C-R), radial and tangential sections (R-T), respectively. \*P ≤ 0.05; · P ≤ 0.1.

percentage of ray surface (PERPAR), ray density, mean ray area, mean ray height and mean ray width based on measurement from six mature Scots pine trees. For each tree, three locations separated by five to seven annual rings were analyzed. \*\*\*P ≤ 0.001; \*\*P ≤ 0.01; \*P ≤ 0.05; · P ≤ 0.1; ns, not significant.

width was highly correlated to mean ray area (r = 0.824; P < 0.001), but was unrelated to mean ray height (r = −0.326, P = 0.186).

The 40 cores showed an overall mean CV in the annual PERPAR of 0.121 (ranging from 0.069 to 0.171 among individuals) and a mean sensitivity (MS) of 10.8% (ranging from 6.0 to 17.2%). Moreover, PERPAR in the cross-sectional plane was significantly smaller in the earlywood (5.77%) than in the latewood (6.18%; t = 6.139; P < 0.001; **Figure 4A**). Similarly, the number of initiating rays (NEWRAY) after correction for unequal tissue contributions to the overall ring area differed marginally between earlywood (21.82 rays/mm) and latewood (23.75 rays/mm; t = − 1.810, P = 0.070; **Figure 4B**).

Bootstrapping analyses indicated that the variability of PERPAR mean estimation decreased with increasing measured wood width (cross-sections)/surface (tangential sections) and number of samples (**Figure 5**). However, the relative 95% confidence interval (CI95) was about twice as large in crossthan in tangential sections for a given measured wood surface. In cross-sections, CI95 curves flattened and got almost linear after a measured wood width of 8 mm, while this already occurred with 4–5 mm<sup>2</sup> in tangential sections. Measurements from 10 cross-sections on 5-mm increment cores resulted in a CI95 of ±10.1% of the true mean PERPAR, doubling the number of samples to 20 reduced CI95 to ±7.0%, while increasing the measured width to 8-mm wide strips provided an accuracy of CI95 ±7.7% with 10 samples. In tangential sections, 10 of the 1-mm<sup>2</sup> surfaces frequently used in previous studies (see Introduction) only provided a CI95 of ±12.7% of the true mean PERPAR. In contrast, measuring 2 × 2 mm areas in 10 samples provided a CI95 of ±6.0%, while it was reduced to even ±4.3% with twenty 2 × 2 mm samples.

#### DISCUSSION

### Estimates of Percentage of Ray Surface Depend on the Cutting Plane

In this study we evaluated several methodologies for quantifying ray features in conifers. Tangential and cross-sections proved to be suitable, but with strengths and limitations in terms of accuracy and investment, and with differences in the information they register. In contrast radial sections were generally unsuitable for ray quantification. The large differences in percentage of ray surface (PERPAR) estimation (up to three times) among cutting planes likely reflects sampling artifacts linked to the sheet-like orientation and fusiform shape of uniseriate rays (see **Figures 1D,E**). In fact, in a radial section, the position and contour of the ray outline may change substantially from the lower to the upper side of the section. Because of the transparency of the tissue, the perceived ray outline is the maximum ray projection through the entire thickness of the section, which leads to a systematic overestimation ("radial overestimation artifact"). This effect–although much weaker—also occurs in cross-sections due to the tapering toward the upper and lower ray extremities, supposedly explaining the overestimated values (**Figure 2**). In contrast, the tangential sections are robust in this respect and therefore likely produced most accurate estimates of PERPAR, and therefore relative ray volume (Myer, 1922). The significant correlation between tangential and cross-sectional data confirms the systematic nature of the larger cross-sectional values. With respect to the quantification of the area of individual rays, the strong allometric relationship allows accurate estimation of ray area as a function of ray length in cross-sections (Olano et al., 2013) and ray height in tangential sections. Using such relationships could significantly increase the efficiency of the measurement procedure.

## PERPAR and Ray Width are the Most Complementary Tangential Ray Parameters

Among the different tangential ray parameters, PERPAR positively correlated with all other ray metrics but mean ray width, which suggests PERPAR incorporates the information from most other metrics (**Figure 4**). Notably, the significant correlations with mean ray area and height are in line with previously observed consistent patterns of these metrics in relation to tree vigor and growth conditions (e.g., Bannan, 1954, 1965; White and Robards, 1966; Fonti et al., 2015). Moreover, the missing relationship between mean ray height and width indicates that their ratio mainly varied among the analyzed images (i.e., trees and/or annual rings), since within the images individual ray height well scaled to the width (mean Pearson's r = 0.475; P < 0.001). This variability is mainly due to the variability in ray width: supplementary analyses revealed that the width of individual rays is less strongly correlated with ray area (mean Pearson's r = 0.750; P < 0.001) than ray height (mean Pearson's r = 0.899; P < 0.001). Mean ray width also showed a larger variability among trees and/or annual rings than mean ray height (CV = 0.169 vs. 0.094). Together with its independence from

PERPAR, mean ray width might therefore be a more promising parameter for ecological studies than mean ray height.

#### Spatial Variability of Ray Features within Tree Rings

The higher PERPAR and marginally larger number of initiating rays (NEWRAY) observed in the latewood compared to earlywood was unexpected when considering the radial orientation of the rays. In fact, once initiated, rays grow and extend to keep the connection with the cambium and phloem (Fischer and Höll, 1992; Spicer, 2014). Such small intra-annual differences could indicate slightly larger parenchyma cells (DeSmidt, 1922) and/or a higher ray initiation rate in latewood than in earlywood. They potentially evidence an advantage of having a larger storage and transport capacity close to the cambium to support the onset of cambial activity in the following growing season. However, the small intra-annual difference in PERPAR values also indicates that it should suffice to only roughly balance early- and latewood in a proportional way when analyzing PERPAR in tangential sections. Additionally, if the section was not taken fully parallel to the orientation of the rays, the number of NEWRAY in cross-sections may be overestimated (see **Figure 1B**). The extent of overestimation in ray initiation is probably directly related to the number of disappearing rays ("ending ray artifact"; Eckstein, 2013), because they are both directly linked to the orientation of the section (see **Figure 1C**). Nevertheless, time series of the number of initiating rays from samples with an orientation problem still represent valid data if standardized before statistical analysis.

#### Accuracy of PERPAR Values Depending on Measurement Width and Area

The accuracy of PERPAR greatly changed depending on the measured wood width (cross-sections), surface (tangential


sections), and the number of samples (**Figure 5**). Yet, our results suggest that common measurement strategies—10 cross-sections of 5-mm increment cores or 10 tangential planes of 1 mm2 provide rather inaccurate estimates of PERPAR. They may thus be often inappropriate to extract an ecological signal considering that year-to-year variability (MS) in PERPAR is about 20% (Olano et al., 2013; Fonti et al., 2015), or even only about 11% as observed in the 40 cores analyzed here. This study estimates how an increased number of samples [since, CI95 decreases with the square root of the number of samples (cf. Equation 1)], or expansion of the measured wood width/area will increase the accuracy. Although our assessment was based on a very detailed analysis of a single stem disc, it cannot be excluded that the CI95 curves may be shifted for Scots pines from other populations or different species. However, we speculate here, that a CI95 in the range between ±5 and ±10% (dotted horizontal lines in **Figure 5**), corresponding to approximately half the CV or MS, could represent a reasonable balance between data accuracy and measurement efficiency.

## CONCLUSIONS

Our results suggest that both cross- and tangential sections are suitable for quantitative approaches, whereas radial sections are generally unsuitable due to strong sampling artifacts. **Table 2** summarizes the major potential and pitfalls recognized in this study. Our main conclusions are that the quantification of rays in cross-sections is generally very efficient and allows establishing annual time series of (consistently overestimated) ray volume and number of initiating rays that can be compared to other anatomical traits such as tracheid dimensions, cell wall thickness and resin ducts, and related to time series of environmental conditions. Tangential sections seem more suitable to accurately estimate ray volume and investigate the spatial integration of rays in the xylem such as the connectivity of conduits to rays, or, more generally, research into structure-function relationships. In addition, tangential ray width registers information independent from PERPAR, which makes it a promising complementary parameter for ecological studies. The choice between tangential and cross-sections will therefore depend on the specific study question and on the available lab capacities. In this context, this study presents for the first time a very concrete guidelines (**Figure 5**) for estimating data accuracy depending on the size of the measured wood width (cross-sections), surface (tangential sections), and number of samples, helping to define a suitable sampling strategy, although the latter also depends on the known or expected responsiveness and variability of the target ray features.

Since, most conifer species display a similar ray architecture and a relatively narrow range of ray volume, we are confident that most of our results are representative for other conifer species as well. A similar assessment might be applied to identify the best methodology in angiosperm species. We are convinced that the methodological guidelines presented here are necessary to foster the establishment of robust quantifications, which will ultimately improve the understanding of the fundamental role of ray parenchyma tissue for the performance and survival of trees growing in stressed environments.

## ACKNOWLEDGMENTS

Evaluations were based on data from the long-term irrigation experiment Pfynwald, which is part of the Swiss Long-term Forest Ecosystem Research Programme LWF (www.lwf.ch). We are indebted to L. Mateju, L. Schneider, and G. Juste ˇ for the preparation of anatomical samples, and to S. Meier and J. Helfenstein for assistance in the field. This study was supported by a grant from the Swiss State Secretariat for Education, Research and Innovation SERI (SBFI C12.0100) and from the Spanish Ministry of Economy and Competitivity with UE FEDER funds (CGL2012-34209). A. Arzac was supported by a FPI-EHU predoctoral grant. The project profited from discussions within the framework of the COST Action STReESS (COST-FP1106).

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 von Arx, Arzac, Olano and Fonti. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Quantitative Wood Anatomy—Practical Guidelines

Georg von Arx <sup>1</sup> \*, Alan Crivellaro<sup>2</sup> , Angela L. Prendin<sup>2</sup> , Katarina Cufar ˇ <sup>3</sup> and Marco Carrer <sup>2</sup>

<sup>1</sup> Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Birmensdorf, Switzerland, <sup>2</sup> Dipartimento Territorio e Sistemi Agro Forestali, Università degli Studi di Padova, Padua, Italy, <sup>3</sup> Department of Wood Science and Technology, Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia

Quantitative wood anatomy analyzes the variability of xylem anatomical features in trees, shrubs, and herbaceous species to address research questions related to plant functioning, growth, and environment. Among the more frequently considered anatomical features are lumen dimensions and wall thickness of conducting cells, fibers, and several ray properties. The structural properties of each xylem anatomical feature are mostly fixed once they are formed, and define to a large extent its functionality, including transport and storage of water, nutrients, sugars, and hormones, and providing mechanical support. The anatomical features can often be localized within an annual growth ring, which allows to establish intra-annual past and present structure-function relationships and its sensitivity to environmental variability. However, there are many methodological challenges to handle when aiming at producing (large) data sets of xylem anatomical data. Here we describe the different steps from wood sample collection to xylem anatomical data, provide guidance and identify pitfalls, and present different image-analysis tools for the quantification of anatomical features, in particular conducting cells. We show that each data production step from sample collection in the field, microslide preparation in the lab, image capturing through an optical microscope and image analysis with specific tools can readily introduce measurement errors between 5 and 30% and more, whereby the magnitude usually increases the smaller the anatomical features. Such measurement errors—if not avoided or corrected—may make it impossible to extract meaningful xylem anatomical data in light of the rather small range of variability in many anatomical features as observed, for example, within time series of individual plants. Following a rigid protocol and quality control as proposed in this paper is thus mandatory to use quantitative data of xylem anatomical features as a powerful source for many research topics.

Keywords: anatomical sample preparation, dendroanatomy, microscopic imaging, microtome sectioning, quantitative image analysis, QWA, tree-ring anatomy, wood sample collection

## INTRODUCTION

Quantitative wood anatomy as meant here investigates quantitatively how the variability in xylem anatomical features of trees, shrubs, and herbaceous species is related to plant functioning, growth, and environment, and often explores how these relationships change over time. Xylem performs a wide range of functions that are essential for plants to grow and survive. The xylem transports water, nutrients, sugars, and hormones; buffers water uptake and loss; supports the mass of the

#### Edited by:

Andreas Bolte, Johann Heinrich von Thünen-Institute, Germany

#### Reviewed by:

Steven Jansen, Ulm University, Germany Jia Hu, Montana State University, USA

> \*Correspondence: Georg von Arx georg.vonarx@wsl.ch

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 08 April 2016 Accepted: 20 May 2016 Published: 03 June 2016

#### Citation:

von Arx G, Crivellaro A, Prendin AL, Cufar K and Carrer M (2016) ˇ Quantitative Wood Anatomy—Practical Guidelines. Front. Plant Sci. 7:781. doi: 10.3389/fpls.2016.00781 canopy plus loads from wind, snow, ice, fruits, and epiphytes; displays foliage and flowers to resources like light and pollinators. Many different ways have evolved to perform these functions, and as a consequence, there is an enormous diversity of xylem anatomies that can be spotted through a microscope. Moreover, wood anatomical features represent a natural archive for growthenvironment relationships and plant functioning with intraannual resolution (Fonti et al., 2010). In fact, xylem cells can be localized at a certain position within a specific annual growth ring (e.g., earlywood or latewood), which is linked to the time of their formation. The xylem anatomical structure is influenced during its development by internal and external factors (e.g., Fonti et al., 2010, 2013a; von Arx et al., 2012; Aloni, 2013; Carrer et al., 2015), and mal-adjusted xylem structure may even be responsible for tree mortality (e.g., Here¸s et al., 2014; Pellizzari et al., 2016). Quantitative wood anatomy capitalizes on the xylem anatomical structures mostly fixed in the stems once the cells are mature, and often focuses on a small number of cell types such as conduits (vessels and tracheids), parenchyma (axial and radial), and fibers.

Xylem anatomical features in plants are numerous, and sometime concern very small and delicate details (IAWA Committee, 1989, 2004; Crivellaro and Schweingruber, 2015). This necessitates careful processing and high accuracy during quantification, but also analyzing a sufficiently large and representative subset of the wood sample (Arbellay et al., 2012; Scholz et al., 2013; Seo et al., 2014; von Arx et al., 2015a). In other words, quantitative wood anatomy requires high-quality, high-resolution, and often large images of properly collected and prepared anatomical samples. Improved sample preparation protocols for these needs have lately been developed (Gärtner and Schweingruber, 2013; Yeung et al., 2015). Furthermore, recent improvements in computer performance, automated image-analysis systems (von Arx and Dietz, 2005; Fonti et al., 2009; von Arx et al., 2013; von Arx and Carrer, 2014) and processing and interpretation of anatomical data (Carrer et al., 2015) nowadays allow to significantly increase the number of measured anatomical features. Together, these advancements are providing the basis to create unprecedented datasets in terms of size and quality, thus also allowing to use quantitative wood anatomy for an increasing number of different research topics such as climate-growth interactions (Olano et al., 2013; Castagneri et al., 2015; Rita et al., 2015), stress responses (Fonti et al., 2013b), tree functioning (Petit et al., 2011; Olson et al., 2014; Guet et al., 2015; Pfautsch et al., 2016), functional anatomical properties to identify tree provenances most resistant to climate change impacts (Eilmann et al., 2014), and wood formation (Cuny et al., 2014; Pacheco et al., 2015) and production (Cuny et al., 2015) processes. However, the production of data meeting high quality requirements necessitates following a strict multi-step procedure, to avoid artifacts and mistakes that can significantly influence the measurements. This is critical considering the relatively small range of variability of many anatomical features, in time series often between 5 and 20% from year to year (Fonti et al., 2007, 2015; Olano et al., 2013; von Arx et al., 2015a) as compared to even several fold in ring width.

This paper shows all sequential steps from sample collection to anatomical sample preparation and highquality data production, and presents guidance and pitfalls of quantifying anatomical features. As such, it is intended to reflect the current state of the art for quantitative wood anatomy, particularly for the quantification of the most commonly investigated water-conducting xylem cells (conduits), but we anticipate that many aspects will be similar in other anatomical features of the xylem and even the phloem.

## FROM SAMPLE TO ANATOMICAL DATA: GUIDANCE AND PITFALLS

## Step 1: Collecting Samples in the Field

Quantitative wood anatomy aims to extract information from anatomical structures of stems, shoots, branches, roots, rhizomes, and even needles and leaf petioles of monocots and dicots. In many cases samples used for quantitative wood anatomy are taken with an increment borer. This tool was originally developed to collect samples for forest mensuration and dendrochronological investigations. When collecting increment cores for anatomical analyses, it is even more crucial than for other purposes to check the sharpness of the cutting edge of the borer's tip to avoid macro- and micro-cracks in the samples. This can be tested by punching out paper circles from a newspaper. Furthermore, it is very important to core in an exact radial direction, from the bark toward the pith, perpendicular to the axial direction of xylem cells, and keeping the borer in a fixed position while drilling. The use of a pusher is recommended when collecting cores for anatomical analyses. Cores of 10– 12 mm in diameter are preferable compared to the standard 5 mm or smaller, to have more material to work with and to minimize the risk of fractures and twisting. Wood samples can also be extracted from stem discs obtained with a chainsaw, whereas in branches and smaller plant stems and/or root collars the entire samples can be processed. For the storage of wood samples we refer to literature such as Gärtner and Schweingruber (2013). Collection of herbs requires to excavate the root collar, e.g., with common garden tools. When cutting small branches, twigs, and small stems from a plant with pruners, the first (squeezed) part of the sample needs to be removed with a small-jagged saw (in hard samples) or a razor blade (in soft samples) before preparing microsections to avoid cracks and fragmentation.

#### Step 2: Preparing Microsections 2.1 General Procedure

Typically, sample preparation involves producing microsections of 10–20 µm thickness with a sledge or rotary microtome, staining of the pallid cell walls with an agent as safranin, astrablue, toluidine blue, cresyl violet acetate, and their combinations to increase contrast in an anatomical slide (Gärtner and Schweingruber, 2013; Yeung et al., 2015). Boiling or just soaking the samples in water, embedding in paraffin, or using corn starch solution often helps to avoid damage to cell structures when cutting (Schneider and Gärtner, 2013; Yeung et al., 2015). For samples with very narrow cell lumina rice starch gives better results than corn starch because of the smaller grain size. When analyzing relatively large cells as the earlywood vessels in ringporous species, it is usually sufficient and more efficient to smooth the wood surface by sanding or cutting (for instance with a core microtome, Gärtner and Nievergelt, 2010), removing sawdust and tyloses using high-pressure air or water blast, and increasing contrast of the wood surface with chalk powder and black marker (Fonti et al., 2009; Gärtner and Schweingruber, 2013).

#### 2.2 Microtome Blades

Microtome blades must be sharp and without defects to avoid disrupting the delicate anatomical structures. Damages due to dull blades are usually more pronounced in thinner sections (**Figure 1**). Frequent replacement or use of a previously unused part of the blade (often after cutting one sample, or after an even surface of the sample was prepared) can avoid this problem. Furthermore, using high-quality blades can significantly reduce cutting artifacts (**Figure 2**). For both conifer and angiosperm samples, good results were reported when using Leica DB80 LX and Leica 819 low-profile blades (Leica Biosystems, Wetzlar, Germany), and Feather N35HR and N35 blades (Feather Safety Razor Co., Ltd., Osaka, Japan; e.g., Prislan et al., 2013; Gricar et al., 2014; Pache ˇ co et al., 2015; Pellizzari et al., 2016), however the optimal blade depends on the microtome model and the sample properties (e.g., density of the material, part(s) of the stem, moisture

content) and therefore requires lab-specific testing. Generally, for cutting xylem, blade types designed for hard tissues should be used.

#### 2.3 Sample Orientation While Cutting Sections

When analyzing cross-sections, the wood samples should be cut perpendicular to the axially oriented xylem cells to avoid over- and underestimation of the measured anatomical features (**Figure 3**). When cutting longitudinal (i.e., radial and tangential) sections wood samples should be cut parallel to the axially oriented xylem cells. This is important when analyzing, for instance, rays in tangential sections. Measurement errors due to improper sample orientation increase with cutting thickness.

#### 2.4 Section Thickness

A cutting thickness between 10 and 20 µm is usually optimal. Analyzing thick sections usually results in over- and underestimation of anatomical features such as cell wall thickness and cell lumen area (**Figure 4**). Thick sections also often appear out of focus. On the other hand, sections should not be too thin, since the tissue staining might be too weak to obtain target structures of sufficient contrast. Weak staining can be improved to a certain extent by prolonging the duration of the staining process or slightly increasing the concentration of the stain. In addition, sections from different species and even individuals can differ in staining intensity. However, as the example in **Figure 4** shows, even in the optimal range the measured values can be influenced by different cutting thicknesses. It is therefore important to standardize cutting thickness for all samples of the same project. A good practice is also to record the thickness of each section, if not fully constant for all samples, thus allowing to relate any outliers to potential cutting-thickness effects during data analysis. It is also important to bear in mind that comparing absolute values among different projects could be biased if different cutting thicknesses were used.

#### 2.5 Making Permanent Slides

Permanent slide preparation is recommended to make specimens last over a long time. The procedure requires to dehydrate sections after staining, and a mounting medium (e.g., Canada balsam, Euparal, Eukitt) to permanently fix the sections between two glass slides (Gärtner and Schweingruber, 2013). To avoid buckling of the section, which impairs a uniform focus when capturing an image, the slide with the cover slip is sandwiched between PVC strips with a small magnet placed on the top of the slide on a metal plate to keep the sections flat and air bubbles out during drying. Canada balsam and Euparal require drying in the oven at 60◦C for 12 h. Permanent slides, once prepared, can be used over and over again and can be stored for longer time periods than non-permanent slides.

## Step 3: Microslide Digitizing

#### 3.1 Cleaning Slides and Cover Glasses

Pollution hampers automatic detection of anatomical features during image analysis and increases manual editing effort needed to obtain accurate data. Microslides should be cleaned carefully before capturing images to avoid obscured and lowcontrast image parts (**Figure 5**). Frequent sources of pollution are, for instance excessive mounting medium (Gärtner and Schweingruber, 2013), fingerprints and dust particles. After drying, any hard mounting media on top of the cover slip can be scraped off with razor blades.

#### 3.2 Magnification

High-resolution digital images of anatomical sections are most commonly captured with a camera mounted on a optical

weaker in (A) thinner than in (B) thicker sections as revealed after analyzing the entire images (c. 2500 cells; only subset images shown here) with the image-analysis tool ROXAS (cf. Table 1): mean cell lumen area in (B) was 43% smaller and mean tangential cell wall thickness 46% larger than in (A). Scale bar = 100 µm.

thickest cell walls (CWT99, belonging to the smallest cells) the cutting-thickness error was up to 40%. Note that the quantification of the measurement errors is based on the shown example only. To a certain extent some of the cutting-thickness errors can be alleviated by adjusting the settings of the image analyses, particularly the segmentation threshold (see Section Image Segmentation and Figure 10). Scale bar = 100 µm.

microscope. Cameras integrated in the microscope system or standard cameras mounted with an appropriate adapter can be used. To observe and analyze conifers 10× objectives are usually recommended, which, depending on the camera, can give a resolution of 1.7–2.5 pixels·µm−<sup>1</sup> . In angiosperms the 4× objectives giving a resolution of 0.7–1.0 pixels·µm−<sup>1</sup> are usually sufficient, especially for analyzing larger cells as vessels in trees, whereas smaller cells such as fibers also often require 10× objectives.

#### 3.3 Contrast and Illumination Settings

Insufficient staining (due to too short staining time and/or old staining solutions) as well as wrong illumination, improper white balance and over-illumination lead to poor image contrast (**Figure 6**). Poor image contrast can significantly hamper the accurate automatic detection of anatomical structures during image analysis.

The quality and accuracy of the image critically depend on proper microscope settings. In this respect, the Köhler illumination method represents a major step to improve image quality (McCrone, 1980) and should be applied as a standard.

#### 3.4 Focusing

Careful focusing avoids obtaining blurred structures that can lead to measurement errors (**Figure 7**). Some systems offer automatic or semi-automatic focusing which contributes to consistently high image sharpness. When focusing manually, one should be aware that the live view on the computer screen is often of reduced size; therefore one should use a 100% zooming window for focusing, if available. When not all regions within an image frame can be in focus because of buckling, z- or focus stacking techniques, i.e., the combination of the focused image information from multiple images taken at different focal planes is a solution provided by some systems. Otherwise, the best and first solution would be to retry preparing a better microslide. In some wood samples this problem cannot be resolved even with careful microslide preparation. Then, excluding poorly focused regions from analysis is the best way to avoid measurement errors. Since the impact of poor focus

FIGURE 5 | Image of a slide with some pollution as indicated by yellow arrows (A) before and (B) after cleaning (Pinus sylvestris). Scale bar = 100 µm.

depends on the size of the anatomical features (**Figure 7**), focusing the smaller target features (e.g., latewood lumina) is better than focusing larger target features (e.g., earlywood lumina).

#### 3.5 Scanning

For analyzing relatively large anatomical features such as the earlywood vessels in ring-porous species, it is possible to capture an image directly from the prepared wood surface with a flatbed scanner using an optical resolution of 1500–2500 dpi (Fonti et al., 2009). For permanent anatomical slides, slide scanners are an efficient alternative to optical microscopes, because they can produce high-resolution (e.g., 2.0 pixels·µm−<sup>1</sup> ) images of entire anatomical samples, which avoids time-consuming image capturing and stitching (see next paragraph).

There are also several modifications of the aforementioned basic image capturing approaches, e.g., capturing images directly from the prepared wood surface with a dissecting microscope, thus combining efficient wood preparation with a higher optical resolution compared to flatbed scanners.

#### 3.6 Stitching Composite Images

Quantification of anatomical structures requires high-resolution images in order to obtain accurate data. However, higher magnification goes along with smaller field of view. This means that the anatomical sample often does not fit into a single image frame captured with an optical microscope, particularly when working with larger samples as the ones used, for example, to build time series of anatomical features (tree-ring anatomy or dendroanatomy). If no slide scanner is available (see above), this

FIGURE 8 | (A) Overlapping high-resolution images stitched together using PTGui and (B) the obtained high-resolution image of an entire Verbascum thapsus root cross-section. The used overlap with neighboring images is visualized for one of the images with yellow dashed lines in (A). The input images contained distortions introduced by the used optical system, which were successfully removed by PTGui (verified by creating a composite image of a stage micrometer and measuring the distances between tick marks, which yielded constant values throughout the image). Five randomly selected vessels along a transect (see labels in B) having an lumen area between 100 and 3500 µm<sup>2</sup> were subsequently measured using ROXAS (Table 1) using always the same settings in images stitched with the software PTGui, AutoStitch, Microsoft Image Composite Editor and Photoshop (Automatic and Reposition settings). Panel (C) shows the percentage deviation of the obtained values compared to the PTGui reference values. The values in all used stitching tools and settings deviate from the PTGui reference, thus indicating distortions. In addition, the magnitude of the deviations varied along the transect often changing from over- to under-estimation. Note that Photoshop Reposition setting also produces distortion-free images if input images are already distortion-free, while AutoStitch still introduces distortions. Scale bar = 1 mm.

dilemma can be resolved by capturing several overlapping images and stitch them together (**Figure 8**).

For image stitching, overlapping images are produced using a microscope stage and systematically moving through the sample while capturing images. Re-focusing should be performed after every single or every few images. The overlap between individual images in angiosperm samples should be about 20% (**Figure 8A**), while in conifers we recommend about 30–40% to facilitate the stitching process. Overlapping images of a sample are then merged to an overall composite or panorama image using stitching software (**Figure 8B**). We recommend using specialized tools such as PTGui (New House Internet Services B.V., Rotterdam, NL) and AutoPano Pro (Kolor SAS, Francin, F) since they offer full control and reproducibility while producing distortion-free composite images. In contrast, some of the widely used stitching systems can produce distortions and artifacts which would lead to inaccurate results. With sufficient overlap and focused images PTGui and AutoPano Pro are usually able to create the composite image automatically. If not, both software allow to manually add control points, i.e., identical structures in the overlapping image parts. If the software are configured correctly, they are even able to correct any image distortions introduced by the optical system (**Figure 8C**; see von Arx et al., 2015b), e.g., when not using the recommended distortion-free "plan" type lenses.

## Step 4: Quantifying Anatomical Features in Anatomical Images

#### 4.1 Image Analysis Tools

Once the image is produced, image-analysis tools are used to quantify the anatomical features. While target structures can be outlined and measured manually, automated image analysis allows to quantify a larger number of anatomical features in a much shorter time, and in an objective and reproducible way. Several image-analysis tools are used for quantitative wood anatomy. They differ considerably in functionality, ranging from rather general image analysis software such as ImageJ (Rasband, 1997–2016) to very specialized tools such as WinCELL (Regent Instruments Inc., Québec, Canada) and ROXAS (von Arx, www.wsl.ch/roxas; **Table 1**). The choice of the most appropriate tool depends on the specific needs. For a general characterization of xylem anatomical features in rather small samples a general tool is sufficient. However, if the sample depth in terms of number of trees, years, and anatomical features measured, but also the requirements in terms of specific and comprehensive output is important for the subsequent inferences, we recommend using specialized tools.

Despite the diversity of tools offering different levels of automation, specialization and usability, the way they are used to quantify anatomical features follows the same basic steps that are explained in the following.

#### 4.2 Determining the Spatial Image Resolution

To obtain the measurements in metric units the pixel-tomicrometer resolution needs to be determined first. Some microscopic imaging systems provide this information directly, or add a spatial scale bar to the image that can be used as a reference. Where such information is missing, the best way to obtain the spatial resolution is to take a microscopic image of a stage micrometer or graticule (slide with an engraved high-accuracy micrometer scale) in the target magnification and measure several times the distance between two tick marks in pixels using a line tool. The obtained line length in pixels is then divided by the known line length in micrometers to receive the pixel-to-micrometer resolution. Selecting distant and different tick marks in each line measurement increases the robustness. In images from a flatbed scanner, the same information can be derived from the known resolution in dpi:

$$\frac{\text{x}}{25,400} \tag{1}$$

Where x is the resolution of the scanned image in dpi. A resolution of 1500 dpi, for instance, corresponds to 0.059055 pixels·µm−<sup>1</sup> .

#### 4.3 Image Processing

In images showing deficiencies, the next step is image processing, which helps to increase contrast and enhance edges of target anatomical structures. Some specialized image analysis tools do this automatically. The example in **Figure 9** shows how an unremoved dust particle on a permanent slide (cf. **Figure 5**) is removed by contrast homogenization, thus resulting in a more complete recognition of tracheid lumina. In general, image processing should be used conservatively as it can change the dimensions of anatomical features in the image. Generally, the better the quality of the anatomical sample and image the less image processing is required to detect and quantify the targeted anatomical structures.

#### 4.4 Image Segmentation

The original or processed image usually needs to be converted into a black-and-white (binary) image that allows discrimination between target and non-target structures (**Figure 10**). In this step called "segmentation" or "thresholding" a color or intensity value that optimizes this separation is—depending on the image-analysis tool—manually or automatically defined. Inhomogeneous image brightness and contrast due to inappropriate light source, uneven sample flatness or thickness and sample pollutions (cf. **Figure 9**) make it difficult or impossible to find a segmentation threshold that accurately discriminates between target and non-target structures in the entire image; such artifacts should therefore be avoided or corrected. The incorrect selection of a segmentation threshold can easily influence the data by 5–10%, particularly when the anatomical features in the image are not well defined because of poor contrast and focus. The segmented image is the basis for quantifying the anatomical features.

#### 4.5 Detecting and Measuring Anatomical Features

The segmented (binary) image is the basis for detecting and measuring anatomical features. Most image-analysis tools represent the anatomical features as vector instead of pixel objects (**Table 1**), which is usually better because irregularities can be corrected more easily (**Figure 11**), and the results are given in sub-pixel resolution.

#### 4.6 Improving Score and Accuracy of Anatomical Feature Detection Using Filters

Most image-analysis tools include size filters to automatically exclude objects that are too small or too large. Moreover, specialized tools offer automatic filters based on color and shape (**Table 1**). Some specialized tools such as ROXAS also include shape corrections, e.g., to correct for particles and ripped-off cell walls that protrude into the cell lumen (**Figure 12**), and contextbased filters that allow, for example, to filter out cells that strongly deviate from the closest neighboring cells.

#### 4.7 Manual Editing

To obtain quality results and deal with image deficiencies, final manual editing is often necessary after automated detection and filtering of anatomical features. Specialized image-analysis

tools offer efficient editing options for deleting, adjusting and adding anatomical features. However,—and this is a pivotal information—it is generally several times more efficient to invest time into high-quality anatomical slides and images rather than to manually improve a suboptimal automated feature detection.

#### 4.8 Xylem Anatomical Metrics and Data Storing

Specialized image-analysis tools automatically extract many metrics from the visual output and save them into data files, others offer manual export functions. Examples of primary, but also several derived anatomical metrics that are used to address many distinct research questions can be seen in the instruction film by von Arx et al. (2015b).

Among the primary measurements are:


Among the many derived metrics calculated manually or automatically by some image analysis tools are:


FIGURE 11 | Defining the anatomical features in a (A) sub-optimal image of Quercus petraea (surface scan, 2400 dpi) as (B) vector instead of (C) pixel objects allows to correct some sample artifacts, e.g., by applying a convex outline filter. Panel (D) compares the percent deviation of vessel lumen area when representing the identical vessels in the selected image as pixels vs. vectors after analyzing the entire sample (>2500 vessels) with the image-analysis tool ROXAS. 20.2% of the measured values deviate by ≥5% from the supposedly more accurate vector object value, and 4.3% by ≥10%. While underestimation of lumen area in the pixel representation can be very strong due to artifacts as highlighted by the yellow arrows in (A–C), pixel representation also resulted in slight overestimation (<5%) of 21.6% of all vessels because of pixel rounding effects. Note that some of these deviations can be significantly reduced by manual editing. Scale bar = 1 mm.

FIGURE 12 | (A) Cross-section of a Pinus sylvestris wood piece showing ripped-off cell walls. (B) Same sample with overlay of detected lumen outlines (cyan) without any correction, resulting in measurement errors. (C) A convex outline filter can correct such artifacts, but may also cut off true concavities in the lumen outlines, e.g., due to pit inflections (see examples highlighted by yellow arrows), while (D) a more powerful "protrusion filter" (as implemented in the image-analysis tool ROXAS) better discriminates between artifacts and true concavities. Scale bar = 100 µm.


#### 4.9 Quality Control

How much manual editing is needed? We recommend to define this by comparing the output of the target anatomical parameters after no, moderate and perfect manual editing for one to a few representative subset images (e.g., including 1000–2000 cells from both early- and late-wood). If all previous steps were done properly the output with no or moderate editing will not deviate from the (near to perfect) output obtained after heavy-editing by more than 1–2%; this is an accuracy we deem sufficient for most purposes.

#### CONCLUSIONS

In this paper we provided some practical guidance and identified several pitfalls to successfully use quantitative wood anatomy in research. Producing xylem anatomical data is a challenging multi-step approach from sample collection to image analysis. As we showed with a few examples, potential measurement errors in many steps are between 5 and 20 or even 30%, which is in the same range as the variability of the anatomical metrics of interest, at least when excluding partly much stronger interspecific and ontogenetic variability. This is exacerbated by the fact that deficiencies in one step propagate to the next step, sometimes scaling up. The neglect of following a rigid and standardized procedure in terms of cutting thickness, staining, and illumination settings can therefore introduce considerable measurement errors and reduce the quality of the xylem anatomical dataset. While the specific measurement errors due to sample and image deficiencies can differ significantly within the smallest and the largest anatomical features, sometimes even changing from over- to underestimation, they are usually strongest in the smaller features such as latewood cell lumina and cell wall thickness. This is of particular relevance if the research goals are oriented towards, for example, intra-annual density profiles including maximum latewood density, or mechanical strength of cells. Although during image analysis the presented measurement errors can be reduced by defining specific settings for each image and manual editing, this is subjective, often very time-consuming, and generally still produces less accurate data than minimizing problems beforehand. The importance of producing high-quality anatomical slides and images can therefore not be stressed too much in terms of efficiency and accuracy. Then, quantitative wood anatomy is a very powerful tool that can give novel and mechanistic insights into the relationships between tree growth and environment over decades and even centuries.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

All authors planned and designed the research. GvA and Mc prepared the anatomical images. GvA performed the quantitative analyses. GvA wrote the first draft of the manuscript, which was finalized with contributions from all authors.

#### ACKNOWLEDGMENTS

We thank L. Schneider for sharing his experience in sample preparation, and L. Schneider and P. Züst for preparing some of anatomical microslides and images. GvA was supported by a grant from the Swiss State Secretariat for Education, Research and Innovation SERI (SBFI C14.0104). This work profited from discussions within the framework of the COST Action STReESS (COST-FP1106).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 von Arx, Crivellaro, Prendin, Cufar and Carrer. This is an open- ˇ access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Allometric Trajectories and "Stress": A Quantitative Approach

Tommaso Anfodillo<sup>1</sup> , Giai Petit<sup>1</sup> \*, Frank Sterck<sup>2</sup> , Silvia Lechthaler<sup>1</sup> and Mark E. Olson<sup>3</sup>

<sup>1</sup> Dipartimento Territorio e Sistemi Agro-Forestali, Università degli Studi di Padova, Legnaro, Italy, <sup>2</sup> Forest Ecology and Forest Management Group, Wageningen University, Wageningen, Netherlands, <sup>3</sup> Instituto de Biologia, Universidad Nacional Autonoma de Mexico, México, México

The term "stress" is an important but vague term in plant biology. We show situations in which thinking in terms of "stress" is profitably replaced by quantifying distance from functionally optimal scaling relationships between plant parts. These relationships include, for example, the often-cited one between leaf area and sapwood area, which presumably reflects mutual dependence between sources and sink tissues and which scales positively within individuals and across species. These relationships seem to be so basic to plant functioning that they are favored by selection across nearly all plant lineages. Within a species or population, individuals that are far from the common scaling patterns are thus expected to perform negatively. For instance, "too little" leaf area (e.g., due to herbivory or disease) per unit of active stem mass would be expected to incur to low carbon income per respiratory cost and thus lead to lower growth. We present a framework that allows quantitative study of phenomena traditionally assigned to "stress," without need for recourse to this term. Our approach contrasts with traditional approaches for studying "stress," e.g., revealing that small "stressed" plants likely are in fact well suited to local conditions. We thus offer a quantitative perspective to the study of phenomena often referred to under such terms as "stress," plasticity, adaptation, and acclimation.

#### Edited by:

Achim Braeuning, University of Erlangen-Nuremberg, Germany

#### Reviewed by:

Andria Dawson, University of California, Berkeley, USA Katarina Cufar, University of Ljubljana, Slovenia

> \*Correspondence: Giai Petit giai.petit@unipd.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 14 March 2016 Accepted: 25 October 2016 Published: 09 November 2016

#### Citation:

Anfodillo T, Petit G, Sterck F, Lechthaler S and Olson ME (2016) Allometric Trajectories and "Stress": A Quantitative Approach. Front. Plant Sci. 7:1681. doi: 10.3389/fpls.2016.01681 Keywords: fitness, scaling, morphospace, operationalization, plasticity

## INTRODUCTION

Like many terms in plant ecology, the term "stress" is both very important and vague. Authors have debated its definition in various contexts for decades (e.g., Levitt, 1972) and this debate continues (e.g., Körner, 2003; Lortie et al., 2004). Because the term is so important, it would be useful to operationalize it, to make it readily accessible to empirical study. One of the pitfalls of such operationalization is that definitions can represent simply artificial categories rather than true natural phenomena (see for example discussions of efforts to operationalize "adaptive radiation" Olson and Arroyo-Santos, 2009). We focus on certain situations that are traditionally discussed in terms of "stress" but in which the term is not only unnecessary but might actually be hiding important adaptive phenomena.

Our approach builds on the observation that many plant attributes covary with one another in highly predictable ways, that is, plants grow allometrically (or isometrically). Plant ecologists document webs of trait covariation that seem to involve most crucial plant traits (Weiner, 2004; Westoby and Wright, 2006). For example, wood traits such as density, branch and stem dimensions, and mechanical resistance to bending are tightly correlated

(Sterck et al., 2006a; Rosell et al., 2012; Castorena et al., 2015). Leaf traits such as leaf lifespan, leaf mass, photosynthetic capacity and respiration are also closely coupled as described by the leaf economic spectrum (e.g., Reich et al., 1997; Sterck et al., 2006b). At the whole plant level, the area and mass scaling relations between organs such as leaves and stem and their tissues (e.g., xylem, phloem) are under strong selection (e.g., Cannell and Dewar, 1994; Zhang et al., 2016). Many of these patterns of covariation seem to reflect evolutionarily optimal relationships, i.e., not global optima for any one trait but the "least bad" combination possible given their conflicting demands (e.g., Niklas, 1994; West et al., 1999; Niklas and Enquist, 2001). These relationships, manifest in stable allometric trajectories, are largely thought to be maintained by natural selection. In other words, some combinations are possible but usually not favored by selection. For example, plants with dense wood usually bear small leaves but plants with low-density wood bear large ones (Olson et al., 2009). Presumably the combination of high density wood and very large leaves is one not generally favored. These two observations—that plant traits frequently covary, and that these relationships can vary to some degree—motivate our proposed means of studying phenomena traditionally referred to as "stress."

The central prediction of our proposal is that distance from allometric scaling lines should be associated with differences in performance or fitness. There is no need for recourse to the term "stress" at all in making this formulation. Performance here is understood as any index that should affect fitness, e.g., photosynthetic efficiency, mechanical support, hydraulic resistance, etc. Fitness is understood to comprise its three components, survivorship, mating success, and fecundity. If an allometric relationship, e.g., leaf mass vs. stem volume, is maintained within a species by selection, then a drastic displacement from the relationship is expected to result in lower performance. For example, sustained defoliation markedly reduces fitness (Anderegg and Callaway, 2012). Drastic removal of sapwood tissue can have a similar effect. Both of these disturbances result in marked movement into spaces distant from the common allometric scaling slope. We propose that distance from the line should be associated with quantifiable differences in performance, and this quantifiability obviates the need for categorizing a given individual as "stressed" or not. We will also show how this approach can reveal situations in which responses to "stress" are in fact adaptive. A key element in generating predictions and interpreting patterns under our approach is a theoretical understanding of allometric relationships.

## THEORETICAL UNDERSTANDING OF ALLOMETRIC TRAJECTORIES

Numerous theoretical considerations underpin the understanding of allometric relationships, as reflected in evolutionary optimality models (e.g., Banavar et al., 1999; West et al., 1999; Enquist and Niklas, 2002; Sterck and Schieving, 2007; Dewar et al., 2009). These models have as central tenets that organisms have three fundamental components: a volume of metabolically active cells, resource distribution networks, and metabolite exchange surfaces (Banavar et al., 2014). In other words, allometric relationships between traits reflect evolutionary convergence on the "best" combination of investment in the three components. One of the most studied allometric scaling patterns is the one between metabolic activity (B) and body mass (M), the exponent of which is clearly 3/4 in animals (Kleiber, 1932), whereas for plants it is generally between 1 and 3/4, depending on how much dead tissue (heartwood) makes up the "body mass" (Reich et al., 2006; Mori et al., 2010). Based on these fundamental relationships, other predictions can be generated. For example, in the simplest case of a tree whose crown shape remains constant as the tree grows larger, M should scale vs. tree height (h) as M ∝ h 4 , B ∝ h 3 and leaf area ∝ h 3 (e.g., Simini et al., 2010). These examples illustrate that these relationships are widespread and span many species. They also help identify situations of interest when plants deviate from predicted relationships.

These theoretical considerations motivate the fundamental prediction of our approach, which is that distance from the allometric slope should be associated with variation in performance or fitness (**Figure 1**). Many studies have shown that variants that fall far from common allometric scaling slopes have lower performance or fitness than individuals that fall close to the line (cf. Sinervo and Huey, 1990; Sinervo and Licht, 1991; Bertram et al., 2009). For example, in Raphanus raphinastrum, corolla tube-stamen length poportions, which are constant across most Brassicaceae, was readily altered in just a few generations of artificial selection (Conner et al., 2011). Similar results are found in animal studies as well: butterflies with relatively large fore- or hind- wings had much lower reproductive success than conspecifics with wild-type hind- and fore- wing proportionality (Frankino et al., 2007). These results show that variants corresponding to "empty" morphospace (gray areas in **Figure 1**) can be readily produced. That they are only rarely observed in nature strongly suggests that they are eliminated by selection, favoring instead those in the white band in **Figure 1**). We will show how empirical allometric relations can help to examine phenomena traditionally referred to as "stress" in terms of departure from common allometric scaling relationships. We provide examples of two situations commonly discussed in the context of "stress," plants affected by defoliation (which when sustained leads to lowered fitness) and plants exposed to different environmental growth conditions (in which, on the contrary, plants show adaptive responses, i.e., maximal fitness in that environment).

#### REVERSIBLE DEFOLIATION

Our first example is one in which "stress" results in reversible, quantitative deviations from local optima. For example, leaf and stem mass covary in a highly predictable fashion in all leafbearing species studied so far (Enquist and Niklas, 2002). Here, we present leaf mass vs. stem mass data from young shoots of coppiced individuals of the tropical tree Moringa oleifera. Note in what follows how these patterns can be discussed with

FIGURE 1 | How allometric trajectories can be used for quantitatively studying phenomena traditionally discussed in the context of "stress." The simplest case is to consider a single species in a specific environment. Most individuals of a tree species have the typical proportionalities between traits (e.g., log Y = leaf mass and log X = mass of the sapwood), with some variation about this line frequently observed (white area), e.g., as heritable variants in natural populations or responses to differences in local conditions. Outside of this area, a wide "empty morphospace" (in gray) (sensu Olson, 2012) is potentially available for different morphotypes with allocation patterns that deviate from the commonly observed variants. Individuals in these areas are expected to have lower performance or fitness relative to those within the white zone. The prediction that distance from the white zone broadly correlates negatively with performance is readily testable. A tree at point A would be expected to have lower performance/fitness than a tree within the common morphospace (white band). For example, higher respiratory costs correlated with a larger body biomass per unit of leaf would decrease individual performance. Therefore, plants would be expected to recover the optimal trait combination. At some threshold level of damage, they presumably cannot recover (irreversible decline) and die. Selection is therefore expected not to favor variants that lie in empty morphospace, such as a tree in points A or B. Allometric trajectories with different intercepts within the white area likely represent different trait proportionalities favored in different environmental conditions, e.g., lower intercepts in resource-rich sites, indicating that a unit of leaf area supports a higher amount of consuming/supporting tissues because annual carbon gain (i.e., assimilation) is higher (see Figure 2B and text for further explanations).

reference to the specific selection pressures with no need to refer to generic and vague notions of "stress." In intact shoots, leaf mass scales with stem mass in a highly predictable way (**Figure 2A**; the exponent of the allometric relationship is 0.87). These trees are located in a public area in a village, where local people regularly harvest the leaves for consumption. Sometimes they strip all of the leaves off of a shoot, leaving only the stem, or most of the leaves, leaving only the oldest leaves, and sometimes they remove only the tenderest terminal leaves. The leaf mass vs. stem mass relationships of these "defoliated" shoots are also shown in **Figure 2A**. Over time, the plants produce new leaves both from axial as well as terminal buds and recuperate the leaf mass vs. stem mass relation of undamaged shoots (arrows in **Figure 2A**).

This baseline leaf mass vs. stem mass relationship is thought to be one favored by natural selection, and this expectation leads to testable predictions regarding "stress" understood as deviation from the baseline allometric relationship. The leaf mass vs. stem mass relationship is thought to be driven by the mutual metabolic relationship between leaves, which produce photosynthates, and stems, which consume photosynthates and mechanically support leaves and supply them with water. Defoliation moves trees away from this relationship, and potentially decreases their performance. In a similar way, shoots with substantial amounts of stem tissue removed would lose significant amounts of water conducting and nutrient storage volume. Although defoliation experiments show a variety of short term responses to tissue removal (Ferraro and Oesterheld, 2002), both loss of stem as well as loss of leaves are expected on average and in the long run to result in lower net photosynthesis and lowered fitness components (Anderegg and Callaway, 2012) such as seed production (i.e., fecundity). Irreversible damage presumably marks the amount of damage to one or both variables that leads to death. This view allows empirical investigation of the degree to which distance from allometric trajectories is associated with quantifiable differences in performance. Reference to "stress" would provide absolutely no empirical advantage or theoretical insight. We now turn to another situation in which the word "stress" is commonly used, and again we show that our allometric alternative provides a much more constructive perspective, helping highlight that plants can adjust their structure in different conditions of resource availability, maximizing the fitness possibilities of each environment.

## ALLOMETRY AND GROWTH IN "STRESSFUL ENVIRONMENTS"

A very common use of the word "stress" in plant ecology is to refer to environments that limit growth and are therefore "stressful." These examples highlight notions of "stress" as lowered productivity, obvious inheritance from agricultural settings, where lowered productivity is unwelcome. Such valueladen terminology has no place in science, as our examples will illustrate. Our first example comes from trees growing at treeline, which are traditionally regarded as "limited" or "stressed" because of their slower growth and irregular crown morphologies, a pointless and value-laden classification. From an allometric perspective, however, stone pine (Pinus cembra) trees growing at high elevation (above 1800 m a.s.l.) have similar needle mass vs. body mass scaling slopes but different Y intercepts as compared to trees from lower elevation (**Figure 2B**). Similar slope means that the crucial trait relationships are maintained in spite of different climate conditions and crown shapes. Higher leaf area for a given body mass in treeline trees might be interpreted as a compensation for the lower annual assimilation per needle mass due to the shorter growing season. Thus a higher leaf area is needed to sustain the respiratory C-losses that likely scale isometrically with body mass at any elevation (Reich et al., 2006; Mori et al., 2010) confirming that different allocation strategies (i.e., amount of

shoots have a highly predictable relationship between stem mass and leaf mass (circles). Leaf harvesting temporarily diverts this relationship (diamonds), but plants stripped of their leaves sooner or later recover the pre-damage leaf mass vs. stem mass relationship (arrows). (B–D) Examples of "stressful" environments. (B) Possible variants at different elevations: in Pinus cembra from below (circles) vs. above (triangles) 1800 m a.s.l., the scaling of leaf mass vs. total body mass (roots included) follows the same exponent (∼0.85), but leaf mass per unit of body mass (i.e., Y intercept) is higher in high elevation trees. (C) Boxplot of annual shoot growth and needle length between wet ("favorable") and dry ("stressful") sites in Pinus sylvestris (dashed and solid lines are mean and median values respectively). This approach seems to show categorical differences between trees in sites that could be arbitrarily classified as stressed and unstressed. (D) However, when the same samples of Figure 2C are plotted as part of an allometric series, it is clear that the scaling of leaf mass vs. shoot mass (the last three growing years) converges on the same trajectory in both wet (circles) and dry (triangles) conditions. This result highlights that the species is able to build similar allometries of the distal parts of the plant in spite of different environments.

leaves per unit of body mass) are possible under the same scaling relationship (i.e., the relative co-variation of both traits). This example shows that our approach highlights important biological questions masked by the traditional categorical approach.

Our second example comes from Scots pine (Pinus sylvestris) trees (**Figures 2C,D**), which grow tall on deep soils but are short and thin, on shallow, rocky soils. In terms of traditional value laden terminology, the "stunted" trees on rocky soils are often described as growing in "stressful" conditions, as reflected by their much lower annual length growth increments and shorter needles as compared to taller trees on moist, deep soils (**Figure 2C**). Our allometric approach highlights that the often value-laden terminology of "stress" in fact hides much valuable biological insight, even leading to very different conclusions. Plotting nearly the same variables against one another (**Figure 2D**) shows that, rather than two distinct categories, the plants considered "stressed" and "unstressed" are in fact indistinguishable with regard to their patterns of trait covariation. In this example, needle mass scales with stem biomass in exactly the same way in "stressed" and "unstressed" plants. This result highlights an entirely different set of biological issues as opposed to the traditional categorical, value-laden approach. Whereas the categorical approach highlights limitations of growth on dry sites as opposed to imaginary optima, the allometric approach instead shows that the plants in both situations are constructing the distal part of the stem (that bears the needles) along essentially identical allometric scaling relationships, though of different sizes.

These examples illustrate how, from an allometric perspective, the notion of "stress" is largely an inheritance from forestry and agriculture, in which any factor that reduces yield is described with negative terminology (Körner, 2003). However, from an evolutionary point of view, it is hard to see how the small trees of dry sites can be classified as "suboptimal," "limited," or "stressed." Instead, their small stature likely represents an adaptive response to prevailing conditions. That they scale similarly to their larger conspecifics on deep, moist soil in their trait relationships gives no reason to consider them as "stressed," an observation that the traditional categorical approach conceals.

#### CONCLUSION

Our quantitative approach does not require arbitrary categorizations of "stress," because it involves testing the prediction that distance from the general allometric slope should be associated with differences in performance (**Figure 1**). From this point of view, the dividing area between adaptive differences, which should maximize performance in the relevant environment, and those that push individuals beyond their zones of optimal performance, should be explorable (cf. Ellison and Jasienska, 2007). This exploration is not helped, and indeed is often hindered by, use of the term "stress." For plant ecologists and evolutionary biologists interested in discovering how the plant form-function relationship has evolved, valueladen conceptions of "stress" can be replaced by biologically rich methodological approaches such as those shown in **Figure 2B**. Clearly, the plants in "stressed" (high altitude) habitats grow slowly. Study of vital proportionalities between parts from the point of view that we suggest, however, reveals that these

#### REFERENCES


plants are probably well acclimated to the local conditions (**Figure 2C**). Use of the term "stress" only masks this adaptive adjustment. Whatever the causes of variation in allometric slopes or intercepts, the allometric perspective we describe here offers a means for thinking about "stress" in quantitative terms. This framework will allow exploring simultaneously adaptation, acclimation, and "stress" in plants in a quantitative way beyond artificial categorization of these concepts (see Lortie et al., 2004), and will thus serve as a basis for testing a wide array of hypotheses regarding plant performance and fitness.

#### AUTHOR CONTRIBUTIONS

TA, GP, and MO developed the idea, provided the main experimental data, wrote the first draft of the manuscript and revised the text, SL provided additional experimental data, FS and SL interpreted the data and intensively discussed and revised the text.

#### ACKNOWLEDGMENTS

The Authors warmly thank Claudio Fior for providing the data of allometry in Pinus cembra (**Figure 2B**). This work was promoted and supported by the EU COST Action FP1106 (STReSS). The Authors are profoundly indebted with Ute Sass Klaassen who has energetically encouraged the discussion among the COST delegates. Collection of data from Moringa was possible thanks to project IT200515 of the Programa de Apoyo a Proyectos de Investigación e Innovación Tecnológica, UNAM.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Anfodillo, Petit, Sterck, Lechthaler and Olson. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Meteorological Drivers of Extremes in Daily Stem Radius Variations of Beech, Oak, and Pine in Northeastern Germany: An Event Coincidence Analysis

Jonatan F. Siegmund1, 2 \*, Tanja G. M. Sanders <sup>3</sup> , Ingo Heinrich<sup>4</sup> , Ernst van der Maaten<sup>5</sup> , Sonia Simard<sup>4</sup> , Gerhard Helle<sup>4</sup> and Reik V. Donner <sup>1</sup>

#### Edited by:

Cristina Nabais, University of Coimbra, Portugal

#### Reviewed by:

Walter Oberhuber, University of Innsbruck, Austria Alicia Forner Sales, Museo Nacional de Ciencias Naturales-CSIC, Spain

> \*Correspondence: Jonatan F. Siegmund siegmund@pik-potsdam.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 27 February 2016 Accepted: 12 May 2016 Published: 03 June 2016

#### Citation:

Siegmund JF, Sanders TGM, Heinrich I, van der Maaten E, Simard S, Helle G and Donner RV (2016) Meteorological Drivers of Extremes in Daily Stem Radius Variations of Beech, Oak, and Pine in Northeastern Germany: An Event Coincidence Analysis. Front. Plant Sci. 7:733. doi: 10.3389/fpls.2016.00733 <sup>1</sup> Research Domain IV—Transdisciplinary Concepts and Methods, Potsdam Institute for Climate Impact Research, Potsdam, Germany, <sup>2</sup> Institute of Earth and Environmental Science, University of Potsdam, Potsdam, Germany, <sup>3</sup> Thünen Institute of Forest Ecosystems, Eberswalde, Germany, <sup>4</sup> Department 5 Geoarchives, Helmholtz Centre Potsdam, GFZ German Research Centre for Geosciences, Potsdam, Germany, <sup>5</sup> Institute of Botany and Landscape Ecology, University of Greifswald, Greifswald, Germany

Observed recent and expected future increases in frequency and intensity of climatic extremes in central Europe may pose critical challenges for domestic tree species. Continuous dendrometer recordings provide a valuable source of information on tree stem radius variations, offering the possibility to study a tree's response to environmental influences at a high temporal resolution. In this study, we analyze stem radius variations (SRV) of three domestic tree species (beech, oak, and pine) from 2012 to 2014. We use the novel statistical approach of event coincidence analysis (ECA) to investigate the simultaneous occurrence of extreme daily weather conditions and extreme SRVs, where extremes are defined with respect to the common values at a given phase of the annual growth period. Besides defining extreme events based on individual meteorological variables, we additionally introduce conditional and joint ECA as new multivariate extensions of the original methodology and apply them for testing 105 different combinations of variables regarding their impact on SRV extremes. Our results reveal a strong susceptibility of all three species to the extremes of several meteorological variables. Yet, the inter-species differences regarding their response to the meteorological extremes are comparatively low. The obtained results provide a thorough extension of previous correlation-based studies by emphasizing on the timings of climatic extremes only. We suggest that the employed methodological approach should be further promoted in forest research regarding the investigation of tree responses to changing environmental conditions.

Keywords: dendrometer measurements, event coincidence analysis, climate extremes, growth response

## 1. INTRODUCTION

During the past 15 years the systematic installation and operation of dendrometers and analysis of the obtained data has received increasing interest in forestry sciences. While the first attempts (Friedrichs, 1897) to use dendrometer data to analyze tree response to environmental conditions were clearly limited by the technical conditions of early instruments, recent developments in the production of modern automated highprecision dendrometers offer the ability to generate dendrometer time series at very high temporal and spatial resolution (Drew and Downes, 2009). The detailed representations of activity in the tree stem—shrinkage, recovery, and swelling cycles—allow for a high-temporal investigation of the tree stem as well as long-term morphological and short-term physiological dynamics (Zweifel and Häsler, 2001; Deslauriers et al., 2003; McLaughlin et al., 2003; Bouriaud et al., 2005; Daudet et al., 2005; Vieira et al., 2013). Where additional environmental information is available, dendrometer data can provide information on the tree stems response to external factors, especially meteorological conditions (McLaughlin et al., 2003; Denneler et al., 2010; Miralles-Crespo et al., 2010; Oberhuber and Gruber, 2010; Jezik et al., 2011; Butt et al., 2014). Investigations such as these are also important in order to better understand the diurnal cycle of sap flow and leaf water potential (Drew and Downes, 2009).

Beyond the fundamental understanding of tree functioning, dendrometer data can also indirectly provide important information on the carbon cycle at the local, regional or global level. Even though stem radius does not allow estimates of total cell numbers, it is an important proxy for a forest's above ground biomass (Schulte-Bisping et al., 2012), because it can help to determine the wood volume available for the fixation of carbon (Cuny et al., 2015).

To the authors' best knowledge, almost all above-mentioned studies have investigated dendrometer data using classical statistical tools like linear correlation analysis or linear multiple regression. These powerful methodological approaches have led to an understanding of the relationship between stem size changes and various environmental parameters. Yet, correlationbased approaches generally take all parts of the distributions of two variables of interest into account and therefore describe the joint behavior of these variables. A crucial question only sporadically addressed so far is how tree stem radius variations (SRVs) are linked to extreme weather conditions. This question gains special importance, since recent climate projections suggest a rising frequency and severity of meteorological extreme events for many parts of the world (Barriopedro et al., 2011; IPCC, 2013; Petoukhov et al., 2013). Consequently, analyzing the impact of such extremes on tree SRVs, on the one hand can help to better understand an event's impact on tree functionality and carbon cycle, and on the other hand, because different tree species may be impacted differently, to generate recommendations for future forest management (Spathelf et al., 2014). New findings concerning the response of stem radius to extreme meteorological conditions would also help to improve existing climate-growth-models like TREERING (Fritts et al., 1999) or CAMBIUM (Drew, 2007).

An important study addressing the response of stem-size fluctuations and tree radius growth to climatic extremes using a large number of dendrometer data sets was recently published by Butt et al. (2014). However, this study did not report explicit results regarding the response of the tree stem to extreme meteorological conditions, due to the fact that they have only used ordinary linear regression to analyze the data.

In this study, we employ the novel methodological approach of event coincidence analysis (ECA) to quantify possible simultaneities between extraordinary daily stem variations and extraordinary meteorological conditions. Here, the commonly used term extreme is replaced by extraordinary, referring to the upper and lower tails of the empirical distributions of the variables of interest. Due to the comparatively short investigation period of 3 years (and only 8 years of climatological data), "real" extreme events are difficult to identify or define and therefore the investigation of extraordinary conditions shall represent a tree's reaction to the tails of the empirical distribution of weather events, which may well be exceeded by future developments under climate change. Therefore, conclusions from this study's results on trees' reaction on weather extremes should be made unter consideration of the used definition of "extraordinary events."

Taking into account the existing literature on SRVs and their relation to meteorological conditions we expect that extraordinary climatic events, specifically temperature events, and extraordinary dendrometer variations should occur simultaneously. Additionally, it can be expected that there are clear inter-species differences concerning the reaction to extraordinary meteorological events.

## 2. MATERIALS AND METHODS

#### 2.1. Study Area and Data Sampling 2.1.1. Study Area

The study site was close to Lake Hinnensee (53.33◦N, 13.19◦E) in the northeastern part of Germany. The site is located within the Müritz National Park. Large parts of this protected area have been classified as UNESCO World Natural Heritage in 2011. The park is characterized by 200–300 years old mixed beech, pine and oak stands. The climate is semi-continental, typical for central Europe, with a mean annual temperature of about 8◦C and an annual precipitation between 550 and 650 mm. The soil at the study site as well as at the meteorological station (see Section 2.2.2) is a brunic arenosol on sand of outwash plains, characterized by strong hydraulic permeability Müller (2014).

#### 2.1.2. Data Sampling

The dendrometer data were collected for three tree species: European beech (Fagus sylvatica), Scots pine (Pinus sylvestris), and Sessile oak (Quercus petraea). The distances between the individual trees range from ca. 4 to 20 m and the selection of trees was based on the canopy status of the individual trees (i.e., only dominant trees or co-dominant trees were equipped with dendrometers). This study focuses on the species' response to meteorological conditions, hence the dendrometer data are not differentiated according to the relative positions of the individual trees in landscape (see Section 2.3), but trees were selected along a transect from the lake shore to a terrace ∼15 m above Lake Hinnensee, in order to sample a possibly large variation of individual local stand variations. The equipped trees have an average height of 26 m with diameters between 50 and 70 cm at breast height.

For each tree species, 10 individuals were equipped with Ecomatik DR point radius dendrometers (Ecomatik GmbH, 2015) installed at 1.2 m height. The sensors were installed at the north face of the trunks in order to avoid direct irradiation. They have a temperature coefficient <0.1m/K. Bark was mostly removed from pine and oak trees prior to setup. SRVs were measured at a temporal resolution of 30 min over a 3-years period between 2012 and 2014.

## 2.2. Data Preprocessing

#### 2.2.1. Dendrometer Data

After a comprehensive quality check, the raw dendrometer data were pre-processed using the following three steps:


during which only a few strong positive precipitation events were observed.

#### 2.2.2. Meteorological Data

In order to define days with extraordinary meteorological conditions, data from a nearby meteorological station in Serrahn (at a distance of less than 2 km from the study site) were used. The soil conditions at the dendrometer site and the meteorological station site are comparable, but not identical. Systematically differing soil temperatures, for example, can not be excluded. In addition to air temperature and precipitation, the station provides information on (relative air) humidity, soil temperature, air pressure and incoming solar radiation. The data set is available starting January 2006 at a temporal resolution of 1 h. Similar to the daily SRVs, the meteorological data were pre-processed in order to identify events of extraordinary daily meteorological conditions:


The following meteorological variables were used: air temperature at 2 m (Tmin, Tmean, and Tmax), land surface air temperature at 5 cm (LSTmin, LSTmean, and LSTmax), soil temperature in 20 cm depth (STmin, STmean, STmax), relative humidity (rHmin, rHmean, rHmax), total precipitation (Ptot) and incoming solar radiation (RADmean and RADmax). Many of these variables are highly correlated among each other. Hence, for a study using classical statistical approaches a principal component analysis (as performed by, e.g., Beck et al., 2013) would be appropriate in order to reduce the number of meteorological observables (i.e., to eliminate collinear variables). However, in this study the novel statistical approach of ECA (see Section 2.3) is applied, where the reduction of dimensions based on their common mean behavior (correlation) would not be useful, since the information of interest (timing of extraordinary events) could eventually get lost. Additionally, this study is particularly focused on the different variations of air, surface and soil temperatures, for example. As an alternative, a dimensionality reduction based on event coincidence rates (see Section 2.3) replacing classical linear correlations as similarity measure could be performed as a preparatory step. The utilization of such a novel approach is, however, beyond the scope of the present study.

#### 2.3. Statistical Analysis

#### 2.3.1. Bivariate Event Coincidence Analysis

In order to investigate the simultaneity of events in meteorological variables and SRV, we apply event coincidence analysis (ECA) (Donges et al., 2016), a novel yet conceptually simple statistical concept. In its basic setting, ECA considers two sequences of events of different types (A and B). As the hypothesis to be tested, events of type B are considered to causally influence the timing of events of type A. To cope with realistic scenarios, ECA allows to not only trivially quantify the number of exactly simultaneous occurrences of events of both types, but to consider also lagged as well as time-uncertain responses. For the latter purpose, a time lag parameter τ as well a temporal tolerance window 1T can be additionally taken into account. Then, ECA counts how often events of types A and B occur with a mutual delay τ in both sequences within a certain temporal tolerance 1T. The resulting number of "event coincidences," divided by the total number of events in one of the series is called the coincidence rate r.

Since the statistical analysis described above is not symmetric, ECA defines two distinct types of coincidence rates, r<sup>p</sup> (precursor coincidence rate) and r<sup>t</sup> (trigger coincidence rate). Here, r<sup>p</sup> is defined as the number of event coincidences divided by the number N<sup>A</sup> of events of type A, describing the fraction of events of type A that have been preceded by at least one event of type B. In turn, r<sup>t</sup> is defined as the number of event coincidences divided by the number N<sup>B</sup> of events of type B, thereby describing the fraction of events of type B that have been followed by (and, thus, potentially triggered) at least one event of type A. When using τ 6= 0, this differentiation is essential. A schematic illustration of the two different types of coincidence rates can be found in **Figure 1**.

In addition to the simple calculation of coincidence rates, the R package CoinCalc used in this work for performing the corresponding analyses provides different possibilities to test whether the empirically found coincidence rates are significantly different from what could result from two independent random event sequences (Siegmund et al., 2016). In this work, we will exclusively utilize an analytical significance test based on the assumption of Poissonian event statistics (Donges et al., 2011, 2016; Siegmund et al., 2015).

#### 2.3.2. Conditional and Joint Event Coincidence Analysis

As a thorough extension of the basic ECA method for two event sequences, in this work, we introduce new multivariate generalizations of ECA termed conditional event coincidence analysis (CECA) and joint event coincidence analysis (JECA). While the above formulation of bivariate ECA is sufficient for many applications (e.g., in Donges et al., 2011), in order to analyze the reaction of ecological variables to extraordinary meteorological conditions, it may be important to additionally consider the conditioning of events on specific situations governed by a third observable, i.e., the case that events of type B appear simultaneously with events in series A if and only if also, one or more events of a third type C take place. This conditioning could, for instance, be relevant to account for the interplay between temperature and moisture or between moisture and radiation.

The conditioning of events of type B by events of type C can be described by the precursor coincidence between B and C. Therefore, the conditional precursor coincidence rate rcp and the conditional trigger coincidence rate rct between A and B can be defined (in analogy to r<sup>p</sup> and r<sup>t</sup> as mathematically defined by Equations (3) and (4) in Donges et al., 2016) as:

$$\begin{aligned} r\_{cp}(\Delta T, \mathfrak{r}, \Delta T\_{cond}, \mathfrak{r}\_{cond}) &= \\ \frac{1}{N\_A} \sum\_{i=1}^{N\_A} \Theta \left[ \sum\_{j=1}^{N\_B} \Theta \left[ \sum\_{k=1}^{N\_C} \mathbbm{1}\_{[0, \Delta T\_{cond}]} (\mathfrak{t}\_j^B - \mathfrak{r}\_{cond}) - \mathfrak{t}\_k^C \right] \right] \\ \mathbbm{1}\_{[0, \Delta T]}(\{\mathfrak{t}\_i^A - \mathfrak{r}\} - \mathfrak{t}\_j^B) \end{aligned} \tag{1}$$

and

$$\begin{aligned} r\_{\text{cf}}(\Delta T, \text{r}, \Delta T\_{cond}, \text{r}\_{cond}) &= \\ \frac{1}{N\_{B,cond}} \sum\_{j=1}^{N\_B} \Theta \left[ \sum\_{i=1}^{N\_A} \Theta \left[ \sum\_{k=1}^{N\_C} \mathbbm{1}\_{[0, \Delta T\_{cond}]} (\{t\_j^B - \tau\_{cond}\} - t\_k^C) \right] \right] \\ \mathbbm{1}\_{[0, \Delta T]}(\{t\_i^A - \tau\} - t\_j^B) \Big], &\tag{2} \end{aligned}$$

respectively. Here, {t A i }, {t B j } and {t C k } are the timings of the events of types A, B and C, respectively, N<sup>C</sup> is the number of events of type C, 1Tcond is an additional tolerance window for the condition, τcond a time lag parameter for the condition, and NB,cond is the number of conditional events of type B, i.e., the number of events of type B that show a precursor coincidence with at least one event of type C. 2(·) denotes the Heaviside function (i.e., takes a value of one whenever the argument is nonnegative, and zero otherwise) and 1<sup>I</sup> the indicator function of the interval I (i.e., takes a value of one whenever the argument is within I, and zero otherwise), respectively. In order to visualize the basic idea of the corresponding CECA, **Figure 1** illustrates the approach in a conceptional way.

Using the definitions in Equations (1) and (2), the conditional precursor coincidence rate describes the fraction of events in series A, that appear simultaneously with C-conditioned events of type B, and the conditional trigger coincidence rate is the fraction of C-conditioned events of type B that are followed by at least one event in series A. In the special case of a simultaneous occurrence of events of types B and C (i.e., τcond = 0), we obtain a setting referred to as JECA.

#### 2.3.3. Methodological Setting in the Present Study

For the application of ECA and CECA/JECA, we dissect the 1095 days period from 2012 to 2014 by sliding windows. For the (bivariate) ECA, the window length is chosen as 61 days with a step size of 5 days, resulting in 75 windows per growing season (1 April to 30 September), where each window contains six events

that are preceded by at least one event of type C. This conditioning is expressed by a precursor coincidence between events of type B and type C. While 1T and τ denote the tolerance window and time lag parameter for counting coincidences between events of types A and B, <sup>1</sup>Tcond and <sup>τ</sup>cond are the respective parameters for the conditioning of events of type B on events of type C.

on average. The window length of 61 days is a compromise between a desired high temporal resolution and a possible large window size necessary to produce robust statistics. The step size of 5 days was selected in order to mimimize the computational demand. For the (multivariate) CECA/JECA, the window length is extended to 91 days including nine events on average, since due to the additional conditioning, the number of events in the meteorological variables decreases markedly. In order to cope with the high computational demand of CECA/JECA for 15 variables, the window step size was increased to 10 days, resulting in 15 windows per season.

As a first step, using the sliding window approach, ECA was performed between each of the 15 individual meteorological variables and each tree's SRV series across each window separately. For every window, the fraction of trees with a significant number of coincidences was taken as a proxy describing the reaction of the species to the considered meteorological events. Subsequently, JECA was performed

between the dendrometer data and all pairs of meteorological variables. As a consequence, the chosen analysis setting results in 15 × 14/2 = 105 different variable combinations. Note that for τcond 6= 0, i.e., mutually shifted occurrences of events in the two considered meteorological variables, the number of combinations to be considered in an actual CECA would be twice as large. Therefore, we do not consider mutually conditioned events in this pilot study, but leave a corresponding detailed investigation as a subject of future work. Besides this, we do not consider the possible extension of multivariate conditions (involving different meteorological variables), which would be straightforward yet lead to an even larger combinatorial variety of different cases to be studied.

In all analyses discussed in the remainder of this work only the (conditional) precursor coincidence rates are considered unless stated otherwise.

#### 2.3.4. Cluster Analysis

In order to analyze the simultaneity of event timings between the individual trees, we additionally use the well established approach of hierarchical cluster analysis with complete linkage. Core of the concept of cluster analysis (in this case for dendrometer time series) is a similarity measure, calculated between all possible combinations of individual time series. This similarity measure is, classically, a correlation coefficient. In this study, we additionally introduce the application of event coincidence rate as similarity measure for the cluster analysis. The calculation of the event coincidence rate between each pair of dendrometer event time series follows along the above mentioned approach, using τ = 1T = 0. Due to the above described data preprocessing, there is no difference between precursor and trigger event coincidence rate.

#### 3. RESULTS

#### 3.1. Event Coincidence Analysis with Individual Meteorological Variables

**Figure 2** shows the fraction of trees with significant precursor coincidence rates between extraordinary positive/negative SRVs and positive/negative events in each of the 15 meteorological variables at the same day (1T = τ = 0).

For positive SRV events (**Figure 2**, left panel), five main observations are made: (i) Tmin and Tmax events have an opposite effect in almost all years and for all tree species. Extraordinary positive SRV events mainly coincide with Tmin values above the 90th percentile and Tmax values below the 10th percentile. The same observation is also clearly visible for the land surface temperature. (ii) Extraordinary soil temperature generally has a much lower impact on positive SRV events than extraordinary air temperature. Except for 2013 for beech, STmean and STmax only rarely show coincidences with positive or negative SRV events, while at least extraordinary positive minimum soil temperatures often coincide with positive SRV events. (iii) Extraordinary high values of relative humidity coincide with positive SRV events in all years and for all species. The fraction of trees showing significant coincidences has been slightly reduced in 2014 in comparison with the other 2 years. (iv) Extraordinary high precipitation values do almost continuously coincide with positive SRV events. An important exception is the summer of 2013, where neither tree species showed a corresponding significant relationship. (v) Extraordinary low values of mean and maximum radiation, generally show pronounced coincidences with SRVs above the 90th percentile.

In comparison to positive SRV events, negative events clearly show fewer significant coincidences with extraordinary meteorological conditions (**Figure 2**, right panel). However, two features can be highlighted: (i) For beech, values above the 90th percentile of the maximum temperature strongly coincide with strong negative dendrometer anomalies in 2012, less distinct in 2013, and hardly ever in 2014. In turn, negative oak and pine SRV events do not coincide significantly with maximum temperature events. (ii) Extraordinary low values of relative humidity very often coincide with negative SRV events for all species. This feature is variably expressed during the 3 years of observations but particularly evident in 2012. In contrast, during 2014 maximum relative humidity events (very wet days) significantly coincide with negative beech SRV events.

In addition to the consideration of exactly simultaneous coincidences as described above, **Figure 3** shows the results of precursor ECA using a tolerance window spanning the previous 2 days (1T = τ = 1 day), i.e., this kind of analysis takes into account responses with a time lag of 1 and 2 days. While in this case, positive dendrometer anomalies only show few coincidences with extraordinary meteorological conditions, on the contrary extraordinary negative stem size changes exhibit three major patterns: (i) Maximum land surface and maximum air temperature events show coincidences of their highest values with negative SRV events, which is especially distinct for beech in 2013. (ii) Extraordinary minimum and mean relative humidity values above the 90th percentile show clear coincidences with negative SRV events in 2012 and 2013 for all tree species. (iii) Extraordinary low mean and maximum radiation values coincide with negative SRV events as well. The two last features are mainly visible for beech and pine and are most distinct in 2012 and 2013. Notably, these results are similar to the previous analysis for positive SRV events when using 1T = τ = 0 (see **Figure 2**).

#### 3.2. Joint Event Coincidence Analysis for Paired Meteorological Variables

Due to the large number of possible combinations between meteorological variables, the figures showing the results of JECA are provided in the supplementary material. The following analysis concerns equally directed events if the values of both meteorological variables under study either both exceed their respective upper threshold value or both fall below their respective lower threshold value used for the definition of events. In the other case where one variable takes extraordinarily high values and the other extraordinarily low ones or vice versa, we will speak of oppositely directed events.

#### 3.2.1. Beech

Figure S1 shows the results of JECA between beech SRV events and equally as well as oppositely directed events of each pair

of meteorological variables, respectively. Six main observations are to be highlighted (which are also evident for the two other tree species but to different degrees): (i) In 2012, extraordinary high minimum temperatures in combination with extraordinary high relative humidity strongly coincide with positive SRV events. (ii) Various combinations of all temperature variables coincide with positive SRV events, but for beech almost only in 2014. (iii) Extraordinary low maximum land surface as well as low maximum air temperature events in combination with extraordinary low mean and maximum radiation coincide with positive SRV events, mainly in 2012 and 2013. (iv) Extreme precipitation plays a rather minor role for beech SRV events when applying JECA for equally directed meteorological extremes. In turn, the combinations of extraordinary low maximum air or land surface temperature with extraordinary high precipitation or extraordinary high air humidity strongly coincide with positive beech SRV events. In addition, (v) extraordinary high minimum temperature values together with extraordinary low radiation, and (vi) extraordinary humid conditions (in terms of strong precipitation or high air humidity) again together with

extraordinary low radiation also coincide with positive beech SRV events.

The investigation of negative beech SRV events using JECA shows hardly any significant coincidences (see Figure S2). Whenever evident, the behavior is simply opposite to the effects of positive change anomalies and shall therefore not be further detailed here.

#### 3.2.2. Oak

The results of JECA for oak SRV events are provided in Figure S3. The left panel shows again that (i) during a period in early summer 2012 extraordinary high minimum temperatures in combination with extraordinary high air humidity coincide with positive SRV events, and (ii) various combinations of extraordinary high temperatures coincide with positive oak SRV events as well. Unlike beech, oak stem variations also show this feature in 2013. (iii) Additionally, extraordinary low maximum air and land surface temperatures together with extraordinary low radiation coincide with positive SRV events. (iv) Extraordinary high minimum and mean temperatures together with extraordinary precipitation events appear simultaneously with positive SRV events mainly in 2014 and partly in 2013. In contrast to this, similar to beech SRV events, negative maximum temperatures events with co-occuring extraordinary humid conditions coincide with positive oak SRV events in all 3 years. The features (v) and (vi) are very similar to what has been illustrated in Figure S1 for beech.

For joint coincidences between negative oak SRV events and pairs of meteorological variables (Figure S4), the results are hardly significant in general. However, three specific observations can be made regarding conditions in late summer 2013: (i) Negative SRV events coincide with very dry conditions indicated by the various combinations of extraordinary low humidity as well as (ii) positive maximum temperature events together with negative humidity events and (iii) negative humidity with positive mean and/or maximum radiation events.

#### 3.2.3. Pine

The features (i), (ii), and (iii) indicated in Figure S5 are very similar to the corresponding features in Figure S1. Yet, feature (i) for positive pine SRV events is additionally visible in 2013. The positive impact of very humid conditions (iv) in terms of combinations of extraordinary high air humidity and extraordinary strong precipitation as well as of low maximum air and land surface temperatures and strong precipitation or high air humidity is clearly visible in all 3 years of observations. The features (v) and (vi) reported above are also clearly visible and in general more distinct than for beech and oak.

Negative pine SRV events (Figure S6) coincide with the same extraordinary meteorological conditions that were observed for oak.

#### 3.3. Positive SRVs and Time Lagged Negative SRVs

As already mentioned in Section 3.1, when comparing the results of ECA without time lag and tolerance window with the results using 1T = τ = 1 day, one very important feature is that relative humidity and radiation show coincidences for both, positive (1T = τ = 0) and negative (1T = τ = 1 day) SRV events. A very similar finding was also reported by van der Maaten et al. (2013), based on correlation analysis. The interpretation of this observation leads to the hypothesis that in many cases, after a positive SRV event a negative event occurs during one of the two following days. In order to further test this hypothesis, we also performed ECA between negative SRV events and positive SRV events for previous days. For this purpose, we used negative events (see Section 2) as event series A and positive events as series B with 1T = 2 days and τ = 1 day. In this analysis a precursor coincidence is found, if a negative event is preceded by a positive event at one of the three previous days, whereas a trigger coincidence is observed if a positive event is followed by a negative event during one of the following 3 days. This analysis was performed for all tree species and for each year separately. **Table 1** summarizes the results which indicate very high rates of both trigger and precursor coincidence. Notably, a very high fraction of positive SRV events (up to 64%) precede negative events at one of the three consecutive days. In all these cases, the observed positive SRV events very likely do not correspond to irreversible growth, but rather reversible swelling. On the other hand, the remaining positive events (40–50%) have not been followed by negative events, i.e., either not followed by stem radius decrease at all or by a gradual decrease that is not identified as extraordinary with the employed event definition.

To further investigate this question, we additionally used JECA with the same setup, where the positive SRV events (series B) have been observed in parallel with extraordinarily high rH values (series C). **Table 2** summarizes the results of this analysis. Notably, the observed joint trigger coincidence rates are clearly higher than the trigger coincidence rates in **Table 1** which implies that positive SRV events induced by high air humidity are more likely to be followed by negative SRV events than other positive SRV events. A possible scenario consistent with this finding would be thunderstorms during relatively dry and/or hot periods, where wet conditions induce positive SRVs and rapid hydrological processes return to dry soil conditions very quickly again. Moreover, we find that the joint trigger coincidence rates are distinctively higher than the joint precursor coincidence rates. This indicates that most of the humidity-induced positive SRV events have triggered negative events, but only a smaller fraction of negative events have been preceded by humidityinduced positive SRV events. This finding suggests, that there are different types of positive SRV events (followed or not followed by negative SRV events events) and different types of negative SRV events events (preceded or not preceded by positive SRV events).

### 3.4. Coincidences of the Timings of SRV Events between the Individual Trees

When comparing the results of both bi- and multivariate ECA between the three tree species, the differences are relatively small. Altogether, oak seems to not favor wet conditions as strongly as beech and pine, but systematic inter-species differences appear to be absent. In turn, for the mean behavior (of daily as well as

TABLE 1 | Mean precursor and trigger coincidence rates (10 trees per species) between negative (event type A) and positive SRV events (event type B), using 1T = 2 days and τ = 1 day (i.e., a negative event following a positive events).


TABLE 2 | Mean joint precursor and joint trigger coincidence rates (10 trees per species) between negative SRV events (series A), positive SRV events (series B) and extraordinarily high rHmean events (series C), using <sup>1</sup><sup>T</sup> <sup>=</sup> <sup>2</sup> days, <sup>τ</sup> <sup>=</sup> <sup>1</sup> day, <sup>1</sup>Tcond <sup>=</sup> <sup>0</sup> and <sup>τ</sup>cond <sup>=</sup> <sup>0</sup>.


subdaily features), the growth characteristics are widely known to differ markedly between different tree species (Drew and Downes, 2009; Miralles-Crespo et al., 2010; Köecher et al., 2012; Butt et al., 2014). The question, whether there are differences or commonalities regarding the tree species' upper or lower parts of the distribution of SRVs has not been addressed to far. In order to investigate this issue for the three species of this study, we performed a hierarchical cluster analysis as described in Section 3. **Figure 4** shows the results of this analysis and reveals that although the mutual correlations between the individual trees are quite high, the coincidence rates between days with positive SRV events are comparatively low. Based on correlation, we additionally find that the tree individuals are quite well clustered, while when using coincidence rates as similarity measure, the clusters following the individual species are completely lost. This means that the highest SRV variations of the individual trees vary strongly in their timing (at the daily scale), and that this timing is not generally differing by tree species. Hence, a clear systematic difference of the results between the different species (Section 3) cannot be found.

## 4. DISCUSSION

#### 4.1. Bivariate Event Coincidence Analysis

The two broadleaved species show a positive response to temperature as well as LST while little positive response is found for pine. This positive growth response is likely due to the sufficient supply of water; possibly by reaching groundwater reservoirs. Similar relationships have been found earlier for the mean values (i.e., using correlation-based analyses) by van der Maaten et al. (2013) and others. In contrast to this, the fact that extraordinary negative Tmax and/or LSTmax events also coincide with positive SRV events (for all species) delivers important complementary information to the positive correlation between stem size variations and Tmax found by van der Maaten et al. (2013) and Deslauriers et al. (2003). Due to the fundamentally different nature of these two statistical approaches, two time series can be positively correlated and simultaneously show significant coincidences between negative and positive events. This is not contradictory, but complementary information. The latter finding is further strengthened by the strong coincidence between positive Tmax events and negative beech SRV events. Since both van der Maaten et al. (2013) and Deslauriers et al. (2003) performed their analysis for the entire (not subdivided) growing season, it is not possible to compare the seasonal timings of these contradicting behaviors.

Only few coincidences were found between SRV events and soil temperature extremes. It is likely that this observation is due to the location of the meteorological station 2 km from the dendrometer site. Due to the variability in soil type and ground cover throughout in the study area, actual soil temperatures beneath the sampled trees may systematically differ from the values measured at the station.

A positive instantaneous (lag zero) correlation between air humidity and SRV has been observed in previous studies (Downes et al., 1999; Deslauriers et al., 2003; Köecher et al., 2012; van der Maaten et al., 2013). In the present work, it was confirmed that this relationship is also evident for the upper (positive events) and lower (negative events) tails of the distribution of SRVs. As may be expected, a similar positive SRV coincidence is recorded with precipitation. However, the absence of a notable heavy precipitation impact during summer 2013 in all tree species is a result of a 60-days dry period except for one single day (35 mm). This suggests that one single heavy rainfall event is not sufficient to result in a significant coincidence rate even if it coincides with a marked positive or negative dendrometer anomaly (which was the case for almost all 30 trees).

The significant coincidences between low radiation values and positive SRV events can interpreted in a twofold way: One the one hand, low radiation decreases transpiration leading to water replenishment. On the other hand, low radiation days commonly correspond to cloudy and foggy conditions and are therefore often characterized by high relative humidity as well. A general negative dependency in terms of negative correlations between radiation and stem radius variability was reported earlier by Downes et al. (1999) and Köecher et al. (2012).

Our analysis revealed some counter-intuitive significant coincidences between days with extraordinary high air humidity and negative SRV events in beech stems during 2014 (**Figure 2**). One possible cause for this may be the result of high air associated with low soil moisture conditions typical of foggy days during spring. Further support of this theory is provided by Figure S2, where in the upper right panel joint coincidences between low Tmin, high rHmax and negative beech SRV events are evident. In order to understand these joint coincidences in more detail, **Figure 5** illustrates the temperature and air humidity development of 4 days in spring and early summer 2014, where the above mentioned coincidences appeared. Specifically, the figure shows very high air humidity values of up to 99% around 9–10 a.m. which abruptly decrease simultaneously with increasing temperature. This behavior is a common indication of foggy conditions during morning hours—caused by inverted atmospheric stratification—that are relieved by the rapidly increasing temperatures on a cloud-free day. The negative SRV events of these days are caused by the extraordinary high temperature and low humidity values of the mid-day and are not linked with the high air humidity values of the morning hours. Applying a classical linear correlation analysis between daily maximum relative humidity and daily SRVs, these days would produce strong residuals, clearly deviating from the well-known positive statistical interrelationship between these two variables. Therefore, these specific days provide a very good example of how to explain singular large residuals in classically assumed interrelationships between weather conditions and dendrometer variations as concluded from correlation analysis on a daily basis.

The results of bivariate (**Figures 2**, **3**), as well as multivariate (Figures S1–S6) event coincidence analysis show that for a number of variables, coincidences with SRV events are not equally evident during all the 3 years of the investigation period. One example was mentioned in the previous paragraph. Another example is that positive Tmax events only coincide with negative beech SRV events in 2012 and 2013, but not in 2014 (**Figure 2**, upper right panel). In this case, the reason is, that the positive Tmax events in 2014 are, in absolute values, lower than the

positive Tmax events of the previous 2 years. Even though the events were defined by percentiles over the entire investigation period, a high number of positive Tmax events in 2012 and 2013 led to this non-uniform distribution of events. In other words: in 2014, the highest Tmax events were not warm enough to trigger negative SRV events. The same explanation is also valid for the absence of coincidences between positive SRV events (for all species) with LSTmin, STmin and STmean in 2012. While in these cases the heterogeneous distribution of coincidences among the years can easily be explained, in other cases the reasons are less obvious. One possible factor leading to interyear differences in the relation between weather conditions and SRVs could be the effect of the previous year's weather conditions on the current tree growth. Corresponding impacts on earlywood production (especially for oak, but also for the other two species) have been reported by a variety of studies (Lebourgeois, 2000; Rubino and McCarthy, 2000; Lebourgeois et al., 2004, 2005; Drobyshev et al., 2008; Michelot et al., 2012; Latte et al., 2015). Yet, all previous analyses that the authors are aware of, addressed integrated/cumulative stem size rates in terms of total earlywood production. So far, no statistical evidence or physiological explanation has been reported, why and how previous year's conditions should influence the day-to-day SRVs and their reaction to weather extremes.

## 4.2. Extraordinary Positive vs. Negative Stem Size Variation Events

The counter-intuitive negative beech SRV events found in Figure 2 and discussed in Section 4.1 could also be explained by a statistical artifact due to the co-occurrence between negative SRV events and positive SRV events of the previous day. Such a phenomenon of contradicting climatic signals in tree rings has also been reported for high resolution tree-ring isotope data (Schollän et al., 2013, 2014).

The various cycles of swelling and shrinkage shown by dendrometer data have been addressed in several recent studies (Downes et al., 1999; Köecher et al., 2012; van der Maaten et al., 2013; Vieira et al., 2013). From previous studies like Bouriaud et al. (2005), it is known that stems can shrink over several consecutive days, likely induced by shrinkage of the bark and

relative sap-flow reduction of the stem some time after rain events. Our study refined these findings by pointing out that there are two kinds of strongly positive stem size variations: (1) some that are followed by negative SRV events and (2) some that are not followed by negative stem size changes during consecutive days. For future studies it will be important to investigate how to disentangle the four possible combinations of strong stem size changes defined in this study: There are strong positive SRV events that are followed (i.e., neutralized) by negative SRV events vs. positive SRV events that permanently increase the stem radius, as well as negative SRV events that simply originate from strong positive SRV events during the previous days vs. negative SRV events that have been forced by adverse weather conditions. A first attempt to disentangle these distinct phenomena has been recently published by Chan et al. (2016). The mentioned classical approach do define growth and shrinkage in dendrometer data (Deslauriers et al., 2003; Bouriaud et al., 2005; Köecher et al., 2012; Vieira et al., 2013) does not solve this problem, since (as for the procedures used by van der Maaten et al. (2013) and also in this study) both shrinkage and growth are solely defined based upon the preceding evolution and do not take into account the (short or long-term) following development of the stem radius.

#### 4.3. Joint Event Coincidence Analysis

The JECA revealed six main findings common to all three investigated tree species. (i) The combination of high minimum temperature with high relative humidity events coinciding with positive SRV events describes situations of warm nights followed by moist days. This feature was most clearly visible for pine which is to be explained by pine having the highest potential for water storage due to its larger amount of xylem Pfautsch et al. (2015). (ii) The positive impact of various temperature combinations on stem radius is a logical continuation of the results of bivariate ECA as discussed above. (iii) The combination between low radiation and low maximum temperatures describes very cloudy days. Such days are often also characterized by high air humidity, and this combination (high humidity and low radiation) is highlighted by finding (vi). (iv) Heavy precipitation as an additional event contributing to the aforementioned cloudy and moderately cool days also favors strongly positive SRVs. Such days are often also characterized by high night temperatures (high Tmin) corresponding to situation (v) highlighted in Section 3.2.

## 4.4. Differences in the Timing of Stem Size Variation Events between the Species

The observations found in Section 3.4 are in fact not trivial, since very different reactions of the analyzed species to environmental conditions have been well documented by, e.g., Gonzalez-Munoz et al. (2014), Lévesque et al. (2013), Garcia-Suarez et al. (2009), or Kwiaton and Wand (2015). The difference of our analysis to these studies is, that our dendrometer analysis specifically takes into account the timings of SRVs and meteorological extremes on a daily basis, whereas previous studies analyzed relationships between tree stem growth and weather conditions on a seasonal time scale. Therefore, the finding that no clear species-to-species differences are evident in this study does not contradict studies on seasonal scales. In turn, our results suggest, that the species-specific relations on weather conditions are more clearly expressed on longer rather than on shorter time scales. Nonetheless, our study indicates, that the different species do not differ markedly in their susceptibility to climate extremes on a daily scale. General statements or even suggestions to forest management concerning the species' eligibility in the context of ongoing and future climate change should not yet be drawn from this first case study.

## 5. CONCLUSIONS

We have used high-resolution dendrometer data to investigate tree species-specific responses to extraordinary meteorological conditions. For the first time joint event coincidence analysis as well as a hierarchical clustering analysis based on coincidence rates have been used. This new approach allowed a detailed analysis of the timing of observations falling in the upper and lower tails of the empirical distributions of daily SRVs. This opens new possibilities for interpreting tree-specific responses to meteorological extremes. Our method is able to provide relevant complementary information beyond what has been known from previous correlation-based analyses. Further potential applications of this method include the investigation of dendrochronological data or intra-annual density fluctuations (IADF).

For future investigations, it will be crucial to put additional efforts into disentangling tree stem radius growth from stem swelling, using novel data analysis approaches. Additionally, integrated studies including dendrometer and wood density measurements, as well as an up-scaling across a larger area will be necessary to draw reliable conclusions on tree or forest carbon storage dynamics in relation to meteorological extreme events.

#### AUTHOR CONTRIBUTIONS

JS Study's Design, Data Analysis, Figures, Writing. TS Study's Design, Proofreading. IH Data Production and Preprocessing, Proofreading. EV Proofreading. SS Data Production and Preprocessing. GH Study's Design, Proofreading. RD Study's Design, Writing, Proofreading.

#### ACKNOWLEDGMENTS

This study was conducted within the framework of the Young Investigators Group "CoSy-CC<sup>2</sup> : Complex Systems Approaches to Understanding Causes and Consequences of Past, Present and Future Climate Change" (grant no. 01LN1306A) funded by the German Federal Ministry for Education and Research (BMBF), the COST Action FP1106 STReESS supported by COST (European Cooperation in Science and Technology), the Virtual Institute of Integrated Climate and Landscape Evolution Analysis—ICLEA—(grant no.

#### REFERENCES


VH-VI-415), and the Terrestrial Environmental Observatories project—TERENO—of the Helmholtz Association. Jonatan Siegmund acknowledges financial support by the Evangelisches Studienwerk Villigst e.V. Ingo Heinrich received support from the Deutsche Forschungs-Gemeinschaft (DFG project number He 7220/1-1). The authors wish to thank Jonathan Bauermann and Jonathan Donges for fruitful discussions on the notation of CECA. Thanks to Daniel Balanzategui for proof-reading the manuscript. All presented analyses have been performed using the R package CoinCalc, available at https:// github.com/JonatanSiegmund/CoinCalc. The publication of this article was funded by the Open Access fund of the Leibniz Association.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00733


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Siegmund, Sanders, Heinrich, van der Maaten, Simard, Helle and Donner. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Structure and Function of Intra–Annual Density Fluctuations: Mind the Gaps

Giovanna Battipaglia1, 2, 3 \*, Filipe Campelo<sup>4</sup> , Joana Vieira<sup>4</sup> , Michael Grabner <sup>5</sup> , Veronica De Micco<sup>6</sup> , Cristina Nabais <sup>4</sup> , Paolo Cherubini <sup>7</sup> , Marco Carrer <sup>8</sup> , Achim Bräuning<sup>9</sup> , Katarina Cufar ˇ <sup>10</sup>, Alfredo Di Filippo<sup>11</sup>, Ignacio García-González <sup>12</sup>, Marcin Koprowski <sup>13</sup> , Marcin Klisz <sup>14</sup>, Alexander V. Kirdyanov 15, 16, Nikolay Zafirov <sup>17</sup> and Martin de Luis <sup>18</sup>

#### Edited by:

Judy Simon, University of Konstanz, Germany

#### Reviewed by:

J. Renee Brooks, United States Environmental Protection Agency, USA Rupert Wimmer, Universitaet für Bodenkultur Wien, Austria

\*Correspondence:

Giovanna Battipaglia giovanna.battipaglia@unina2.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 12 January 2016 Accepted: 18 April 2016 Published: 06 May 2016

#### Citation:

Battipaglia G, Campelo F, Vieira J, Grabner M, De Micco V, Nabais C, Cherubini P, Carrer M, Bräuning A, Cufar K, Di Filippo A, ˇ García-González I, Koprowski M, Klisz M, Kirdyanov AV, Zafirov N and de Luis M (2016) Structure and Function of Intra–Annual Density Fluctuations: Mind the Gaps. Front. Plant Sci. 7:595. doi: 10.3389/fpls.2016.00595 <sup>1</sup> Department of Environmental, Biological and Pharmaceutical Sciences and Technologies, Second University of Naples, Caserta, Italy, <sup>2</sup> Centre for Bio-Archaeology and Ecology, PALECO Ecole Pratique des Hautes Etudes, Institut des Sciences de l'Evolution, University of Montpellier 2, Montpellier, France, <sup>3</sup> Euro-Mediterranean Center on Climate Change, Lecce, Italy, <sup>4</sup> Department of Life Sciences, Centre for Functional Ecology, University of Coimbra, Coimbra, Portugal, <sup>5</sup> Institute of Wood Technology and Renewable Resources, University of Natural Resources and Life Sciences, Vienna, Austria, <sup>6</sup> Department of Agricultural Sciences, University of Naples Federico II, Naples, Italy, <sup>7</sup> Swiss Federal Research Institute WSL, Birmensdorf, Switzerland, <sup>8</sup> Department of Land, Environment, Agriculture and Forestry, University of Padua, Padua, Italy, <sup>9</sup> Department of Geography and Geosciences, Institute of Geography, Friedrich-Alexander-University Erlangen-Nuremberg, Erlangen, Germany, <sup>10</sup> Department of Wood Science and Technology, Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia, <sup>11</sup> Department Agricultural and Forestry, Università degli Studi della Tuscia, Viterbo, Italy, <sup>12</sup> Departamento de Botánica, Escuela Politécnica Superior, University of Santiago de Compostela, Lugo, Spain, <sup>13</sup> Faculty of Biology and Environment Protection, Nicolaus Copernicus University, Toruñ, Poland, <sup>14</sup> Department of Silviculture and Genetics, Forest Research Institute in Poland, Budynek, Poland, <sup>15</sup> V.N. Sukachev Institure of Forest SB RAS, Krasnoyarsk, Russia, <sup>16</sup> Department of Math Methods and IT, Siberian Federal University, Krasnoyarsk, Russia, <sup>17</sup> Department of Plant Pathology and Chemistry, University of Forestry, Sofia, Bulgaria, <sup>18</sup> Department of Geography and Regional Planning—IUCA, University of Zaragoza, Zaragoza, Spain

Tree rings are natural archives of climate and environmental information with a yearly resolution. Indeed, wood anatomical, chemical, and other properties of tree rings are a synthesis of several intrinsic and external factors, and their interaction during tree growth. In particular, Intra-Annual Density Fluctuations (IADFs) can be considered as tree-ring anomalies that can be used to better understand tree growth and to reconstruct past climate conditions with intra-annual resolution. However, the ecophysiological processes behind IADF formation, as well as their functional impact, remain unclear. Are IADFs resulting from a prompt adjustment to fluctuations in environmental conditions to avoid stressful conditions and/or to take advantage from favorable conditions? In this paper we discuss: (1) the influence of climatic factors on the formation of IADFs; (2) the occurrence of IADFs in different species and environments; (3) the potential of new approaches to study IADFs and identify their triggering factors. Our final aim is to underscore the advantages offered by network analyses of data and the importance of high-resolution measurements to gain insight into IADFs formation processes and their relations with climatic conditions, including extreme weather events.

Keywords: IADF, tree-ring, wood anatomy, stable isotopes, network analysis, wood formation

### IADF FORMATION AND POSITION

Intra-Annual Density Fluctuations (IADFs) are variations in wood density that are defined by the presence of earlywood-like cells within latewood and (or) by the presence of latewood-like cells within earlywood (Fritts, 2001). Such anatomical structures may hamper cross-dating and any further analyses of treering series (Cherubini et al., 2003). Thus, IADFs have long been considered by dendrochronologists as the "Ugly Duckling" of wood anatomical features, and species forming them have often been discarded for climate reconstructions (Lorimer et al., 1999) and used as indicators of particular events such as flood-regime or air pollution (see Wimmer, 2002 for review). During the 2000s, the "Ugly Duckling" turned into a "Beautiful Swan," when different studies demonstrated the potential of these anatomical features for ecological, environmental, and climatological interpretations (Wimmer et al., 2000; Rigling et al., 2001; De Micco et al., 2007; Campelo et al., 2007a,b; De Luis et al., 2007). Since then, the importance of IADFs has been widely recognized and the number of papers dealing with them has increased significantly (De Micco et al., 2016). It has been underlined that IADFs may provide accurate information at the seasonal level (Glock and Agerter, 1960; Tessier et al., 1997; Bräuning, 1999; Campelo et al., 2007b; Battipaglia et al., 2010, 2014; De Micco et al., 2012), allowing a more detailed climate analysis within the growing season (Wimmer et al., 2000; Novak et al., 2013a,b, 2016).

Variations in IADF features can be used to reconstruct past environmental conditions, and IADF relative position within the ring can be used to estimate when a specific environmental factor occurred (Figure S1). Campelo et al. (2007b) classified a band of latewood-like cells at the end of earlywood in Pinus pinea as an "IADF type E+," corresponding to a gradual transition from early- to latewood (Figure S1). They suggested that those IADFs could be linked to soil water conditions during late spring, hypothesizing that rainfall events in late spring could delay the transition from early- to latewood. Later in the growing season the cambium can reverse latewood production forming again earlywood-like cells. When a band of latewood-like cells is located within earlywood the IADF is labeled as "type E" (Campelo et al., 2007b). This type of IADF (Figure S1) seems to be uncommon in Mediterranean pine species (Vieira et al., 2010; Rozas et al., 2011; Campelo et al., 2013) probably because regular weather conditions during spring can assure continuous growth (Olano et al., 2012), or trees are able to minimize episodic events of water stress during the early growing season (Loustau et al., 1996; Borghetti et al., 1998). In contrast, the formation of IADF type E was found to be rather frequent in other environments: for example, in Pinus nigra sampled in the Vienna Basin, this type of IADF was triggered by a combination of wet April, dry May, and wet June and it was related to the watertable level (Wimmer et al., 2000). In Erica arborea and Arbutus unedo, hardwood species growing in the Mediterranean basin, this type of IADF was also frequent and triggered by summer drought conditions (Battipaglia et al., 2010, 2014; De Micco et al., under revision). In hardwood species the comparison of series of vessel lumen size between tree rings with and without IADFs suggested that: (a) IADF position is related to the period of the season when stressful conditions priming the fluctuation occur, (b) the width of the IADF indicates the duration of conditions triggering its formation (Campelo et al., 2007a; De Micco et al., 2014). Most studies dealing with IADFs found that their frequency increased close to the end of the tree ring (Rigling et al., 2001; Rozas et al., 2011). Two types of latewood IADFs have been classified considering the position within latewood: the first type is characterized by earlywood-like cells within latewood (IADF L; Figure S1) and the other located between latewood and earlywood of the next ring and characterized by intermediate anatomical traits (IADF L+; Figure S1) (Campelo et al., 2007b). In both cases, they were mainly associated with favorable conditions occurring after the summer drought, in early autumn (L) or in late autumn (L+) (Rigling et al., 2001; Masiokas and Villalba, 2004; Campelo et al., 2007b; Battipaglia et al., 2010, 2014). Although, this first classification could be criticized for the fact that IADFs E<sup>+</sup> and L<sup>+</sup> do not correspond to a true fluctuation in wood density, it is important to question the value of the position of IADFs as a proxy for past climate. By using the relative position of IADFs within tree rings it is possible to improve the temporal resolution of tree-ring series (De Micco et al., 2014), especially in areas where the growing season is long, such as in the Mediterranean region (Rozas et al., 2011; De Luis et al., 2011a).

The identification of the environmental conditions triggering IADF formation is based on linear correlations between climatic variables and IADF chronologies, highlighting the importance of water conditions during the growing season in their formation. However, these correlations are not enough to fully understand the process behind IADF formation, namely at the level of cambial activity and cell differentiation processes. Are IADFs the result of cambial reactivation? Are IADF cells already present in the cambial zone undergoing differentiation? Are latewood IADFs caused by changes in the cell enlargement and/or cell wall deposition phase? These are fundamental questions that can only be answered by monitoring xylogenesis at a weekly time scale, and relating it to intra-ring variations of cell features. Studies on cambial dynamics and wood production can help us to understand the physiological mechanisms behind IADFs formation (Camarero et al., 2010; Vieira et al., 2015). There have been recently major developments in this field, leading to a detailed description of the timings of cambial activity, duration of cell production and differentiation phases and response of cambium to environmental conditions in different species and environments (De Luis et al., 2007; Camarero et al., 2010; Cuny et al., 2014). These studies showed that cambial activity in the Mediterranean region presents a high year-to-year variability, strongly dependent on climate (Vieira et al., 2014a). Cambial activity in the Mediterranean, as in other temperate environments, starts in spring in response to warm temperatures and increasing photoperiod (Vieira et al., 2014b), with periclinal cell divisions of the vascular cambium and the production of earlywood tracheids. It reaches a maximum around May and then, when water becomes less abundant, cambial activity slowly decreases, reaching a minimum in the summer months (Camarero et al., 2010), when latewood tracheids are produced (Uggla et al., 2001). Water availability is fundamental for cell division and turgor-driven cell expansion (Kutschera and Niklas, 2013). Expansion only starts once a threshold of turgor pressure is achieved and the pressure applied by the waterfilled vacuole against the cell wall determines the tracheid final size (Oribe et al., 2003). The formation of the latewood cells is expected during summer, while the earlywood-like cells can be formed in autumn, if favorable climatic conditions return. Several studies under Mediterranean climatic conditions suggested that cambial activity could show a bimodal pattern with two main peaks: one in spring and the other in autumn (De Luis et al., 2011a,b; Battipaglia et al., 2014; Vieira et al., 2015). Since cambial reactivation after a dry summer is not always observed, a facultative bimodal pattern is the best way to describe cambial activity in Mediterranean environments. The tracheids differentiated after summer drought differ from those previously formed in latewood, since their cell wall thickness to lumen diameter ratio is lower than in true latewood (Carvalho et al., 2015; Vieira et al., 2015). Thus, tracheids forming IADFs L have larger radial cell and lumen dimensions than true latewood tracheids. Differentiating tracheids can expand beyond the usual radial diameter of latewood tracheids, if water is available. Indeed, the lumen area of a tracheid depends on turgor pressure and duration of cell enlargement (Cuny et al., 2014). These results suggest that the formation of latewood IADFs in the Mediterranean area are defined during the enlargement phase, whereas it is possible that latewood IADFs formed at higher altitudes and latitudes are caused by changes in the cell wall deposition phase. In colder environments, tracheid differentiation must be concluded before the onset of winter (Rossi et al., 2008) and IADF L<sup>+</sup> can be formed if there is not enough time to complete the deposition phase due to a fast drop in air temperatures. However, formation and ontogenesis of this kind of IADFs are still under debate.

#### IMPACT OF IADFs ON TREE HYDRAULICS

One important gap in IADF research is the functional role played by these anatomical structures on tree hydraulics (Wilkinson et al., 2015). It is known that the size of conduits (e.g., tracheids and vessels) is related to the hydraulic conductivity, while protection from drought-induced embolism is a function of the ability to prevent air-seeding and this is strongly related with the number and size of pits, thus indirectly with lumen size (Hacke et al., 2004; Pittermann et al., 2006). Earlywood IADFs, characterized by latewood-like cells within earlywood, potentially represent a fraction of the earlywood with a lower hydraulic conductivity, while the opposite occurs for IADFs located in latewood. Small increases in tracheid lumen can dramatically increase hydraulic conductivity because flow rate is proportional to the fourth power of the tracheid radius (Tyree and Ewers, 1991). Thus, it is important to quantify their impact on tree hydraulics, because currently we only have indirect observations (Campelo et al., 2007b). It can be assumed that all cells forming IADFs are conductive in order to quantify IADFs impact on the total hydraulic conductivity. Afterwards, a more experimental approach is needed to check if IADFs are functional from a hydraulic point of view. It is also important to characterize the cells forming the IADFs, namely their lumen diameter, length, number, and size of pits, as these anatomical characteristics will affect their hydraulic conductivity.

## OBJECTIVE ASSESSMENT OF IADFs

Visual identification of IADFs in conifers is only possible through the analysis of variations in tracheid features (e.g., cell and lumen diameter, and cell-wall thickness). The accuracy of the visual macroscopic identification of IADFs depends on many parameters, such as the quality of wood surface polishing, microscope magnification and criteria used to distinguish IADFs. Since visual identification of IADFs is based on qualitative criteria rather than on quantitative measurements, the subjectivity of the operator can also be one of the major sources of error. Intra-ring variations in tracheid anatomy, and consequently IADFs, can also be identified through quantitative measurements of tracheid features or tracheidograms (Hetzer et al., 2014; Ziaco et al., 2014; Carvalho et al., 2015; Campelo et al., 2016) and image analysis of X-ray densitometry profiles (Cherubini et al., 2013; Gonzalez-Benecke et al., 2015; Wilkinson et al., 2015). Image analysis avoids the long and tedious procedure of visual examination of wood samples and IADF characteristics can be computed automatically (e.g., relative position within the tree ring and IADF-band width). Image analysis also precludes the operator's subjectivity and provides the size distribution of tracheid features (e.g., lumen diameter and cell-wall thickness). However, it is highly recommended that IADFs recognized automatically by algorithms to be compared with those obtained visually by an expert of wood anatomy for an initial calibration to guarantee the correctness of the criteria used for their identification. All general constraints listed for the identification of IADF in softwoods apply also to hardwoods, where IADF analysis is even more complicated due to the occurrence of different cell types and the spatial distribution of vessels, which are usually not arranged in regular rows like softwood tracheids. Further studies are needed because a number of different anatomical functional traits seem to work for IADF identification in hardwoods, but they appear to be species-specific (De Micco et al., 2015). Furthermore, the fact that many features can be involved (e.g., vessel size, fiber wall thickness, spatial arrangement of cells) opens the possibility to define new types of IADFs.

Within this framework, sharpening the focus at the tissue scale and analyzing various xylem histological traits seemingly represents one of the most promising approaches. Recent methodological advances in quantitative wood anatomy (Gärtner and Schweingruber, 2013; Von Arx and Carrer, 2014) allow efficient development of multi-centennial time series of xylem anatomical traits mostly related to the type and number of cells per ring and to cell-lumen and cell-wall dimensions. These advances not only improve the length of the generated timeseries, but most of all the robustness of the measurements. With the currently available computer capacity combined with specific software, accurately tailored to analyze wood anatomical traits, Battipaglia et al. Intra-Annual Density Fluctuation

it is possible to apply a more thorough and unbiased approach considering all cells within each wood anatomical image. This can outperform the previous practice of measuring xylem traits along a few selected radial cell files as it allows information to be collected on hundreds to several thousands of cells per ring (Figure S2). This, together with improved perception of the long-term ontogenetic change in xylem-cell dimension (Carrer et al., 2015), clearly opens the door for a sound statistical analysis not just of IADFs frequency but also on their extension, intensity, position within the ring, or on the relative role of different cell traits to classify different IADFs types that cannot be unambiguously distinguished by visual inspection.

## STABLE ISOTOPES APPROACH

Quantitative wood anatomy has recently been coupled with stable isotope (δ <sup>13</sup>C and δ <sup>18</sup>O) measurements (De Micco et al., 2007; Vaganov et al., 2009; Battipaglia et al., 2010, 2014) to characterize IADFs, offering new perspectives in the interpretation of IADFs in relation to physiological and ecological processes. What is still unsolved is if the stable isotope signals can help us to identify the different types of IADFs within a ring. Battipaglia et al. (2010, 2014) demonstrated with in continuum stable isotope measurements in hardwood species that IADFs have a unique isotopic signature linked to their position, and are completely different from the correspondent well-known earlywood-latewood isotopic range values (Helle and Schleser, 2004; Vaganov et al., 2009). Here, we performed a preliminary study analyzing IADFs of different species at different sites (P. pinea from Italy, Pinus halepensis from Spain and Slovenia; Pinus pinaster from Portugal, Larix decidua from Poland, and L. decidua x kaempferi from Austria), in order to verify a possible common isotopic signal at intra-annual scale for each type of IADF. IADF L was found in all sites and species, whereas IADF L <sup>+</sup> and E<sup>+</sup> were only present in 66% of the sites and type E only in 50%. The carbon and oxygen isotopic signals of the different kinds of fluctuations were consistent between sites and species with differences between IADFs type E, E+, and L and none between L and L+, supporting the hypothesis that the L<sup>+</sup> can be considered as a transitional wood and not as a true fluctuation (Table S1, Figures S3, S4). Although a more complete analysis is required in order to completely understand the link between isotope signals, position, and climatic parameters triggering IADF formation, stable isotopes seem to be a powerful tool not only to increase physiological information on plant responses to climate, but also for the objective identification of each IADF type.

## IADF OCCURRENCE AND NETWORK APPROACH

IADFs have been reported in several species (hardwoods and softwoods), and regions across a wide gradient of temperature and rainfall availability, from tropical to subarctic, to semi-arid and arid environments (De Micco et al., 2016). The majority of studies have been conducted in Mediterranean ecosystems where the highest frequency of IADFs has been reported, particularly in conifers such as Pinus spp., and where several efforts have been made to analyze the characteristics and ecological meaning of IADFs (Bräuning, 1999; Wimmer et al., 2000; Rigling et al., 2002; De Micco et al., 2007; Campelo et al., 2007b, 2013, 2015; De Luis et al., 2007, 2011a,b; Vieira et al., 2009, 2010; Camarero et al., 2010; Rozas et al., 2011; Nabais et al., 2014; Novak et al., 2013a,b). Under boreal or temperate climate, IADFs have been observed in 9% of the tree rings at maximum (Wimmer et al., 2000; Rigling et al., 2001; Copenheaver et al., 2006), while under Mediterranean climate, they have been observed in up to 15–32% of rings (Campelo et al., 2007b; Bogino and Bravo, 2009; Vieira et al., 2009; Novak et al., 2013a).

The role of sex and genetics on the occurrence of IADFs has only recently been investigated. Olano et al. (2015), studying IADF frequency in Juniperus thurifera growing in two sites with contrasting hydrological conditions in Spain, reported that female trees present the highest frequency of IADFs reflecting their opportunistic water use strategy. Treebreeding studies have shown the influence of provenance on tree growth and wood properties (Rozenberg et al., 2002; Hannrup et al., 2004; Klisz et al., under revision). For example, Rozenberg et al. (2002) found that different wood density parameters, including density fluctuations in earlywood and their position within the tree ring showed high heritability. However, the effect of provenances on IADF formation has not been investigated in detail yet, but we can expect that IADF frequency should differ in provenances with different growth rates. As experimental approach, we investigated the influence of provenance on IADF formation by comparing two contrasting provenances (in terms of tree growth) from a long-term trial (1970–2011) of Abies alba (George et al., 2015) grown in eastern Austria. The mean tree-ring width of the fast-growing provenance (Slovakia, P19; 3.3 mm) was twice as wide than the slow-growing provenance (Italy, P45; 1.5 mm). As expected the highest IADF frequency was observed in the fast-growing provenance (**Figure 1**), highlighting the necessity for further investigation of the genetic influence on IADF occurrence.

The consistency of the climatic signal among different pine species and areas suggested that a large-scale network of IADFs in the Mediterranean region could help to study intra-annual climate variability (Zalloni et al., 2016). In the framework of the FPS COST Action FP1106 STReESS (Studying Tree Responses to Extreme Events: a SynthesiS), a catalog and database of IADF occurrence and anatomical and isotopic features have been developed, consisting of data collected on different species across a large geographical range. This unique and novel catalog includes IADF identification and measurements in 10 countries, 14 species, and 108 tree populations with a total of 2199 trees (3670 cores) and 234,262 tree rings. In this perspective we present a first exploratory analysis on IADFs showing a wide range of variability in IADF frequency (**Figure 2**, Table S2), with sites where IADFs are nearly absent (minimum frequency of 0.9% in high-elevation P. nigra on Corsica) and others where IADFs are present in nearly all tree rings (maximum of 93% in P. pinaster in Galicia, Spain). The network approach offers

the mean tree-ring width; the mean tree-ring width in the fast-growing provenance (3.3 mm) was twice as wide than in the slow-growing provenance (1.5 mm).

important advantages since it overcomes limitations due to treeage and tree-size effects (Vieira et al., 2009; Novak et al., 2013a; Campelo et al., 2015) and to local replication (Zalloni et al., 2016). It also provides a unique possibility to interpret the relationship between IADF frequency and the main climate factors promoting their formation at a regional scale as described by Zalloni et al. (2016) for P. halepensis, P. pinea, and P. pinaster across their distribution range.

## CONCLUSIONS

To maximize the extraction of environmental information from IADFs, more researches on IADF formation and data about IADF frequency are needed. There is also a need to classify IADFs more precisely, and to quantify their wood anatomical features. In this context, a network approach could help to identify not only the main climatic drivers of IADF formation, but also to clarify the functional role of IADFs across different environments and species. The catalog presented here will be further explored and new data will be welcome from different environments and species aiming to create a unique network between scientists working with IADFs. This would help us to answer the large number of open questions and to fill the current gaps on IADFs research.

Further, we believe that one urgent issue still under debate is the identification of IADFs using wood quantitative approaches. Until now, each operator has used his own ability (that depends on experience) to recognize IADFs and to assign them to earlywood or latewood. In many tree species, the correct identification of IADFs is more difficult because the transition between earlywood and latewood is not straightforward and unequivocal. Given the subjective nature of IADF identification, the operator must be well trained and experienced. However, an intrinsic error due to the operator's subjectivity will always remain during the process of IADF identification. To overcome this drawback, machine learning based approaches should be specifically developed to recognize IADFs.

## AUTHOR CONTRIBUTIONS

GB and MD gave a substantial contribution to the conception and design of the study. GB, MD, AB, AK, AD, CN, KC, MKl, MK, MG, NZ, IG contributed to the supply of data for the network. MD was in charge of network analyses. GB performed stable isotopes analyses. GB wrote the first draft of the manuscript. MD, FC, JV, VD, CN, MG, MC, PC contributed to writing specific sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted manuscript.

#### ACKNOWLEDGMENTS

This research is linked to activities conducted within the COST FP1106 "STReESS" network. Collection of

### REFERENCES


datasets used for this work was supported by the projects:


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00595


Fritts, H. C. (2001). Tree Rings and Climate. London: The Blackburn Press.

Gärtner, H., and Schweingruber, F. H. (2013). Microscopic Preparation Techniques for Plant Stem Analysis. Remagen: Verlag Dr. Kessel.

George, J. P., Schueler, S., Karanitsch-Ackerl, S., Mayer, K., Klumpp, R. T., and Grabner, M. (2015). Inter- and intra-specific variation in drought sensitivity in Abies spec. and its relation to wood density and growth traits. Agr. For. Meteorol. 214–215, 430–443. doi: 10.1016/j.agrformet.2015. 08.268


Hannrup, B., Cahalan, C., Chantre, C., Grabner, M., Karlsson, B., Le Bayon, I., et al. (2004). Genetic parameters of growth and wood quality traits in Picea abies. Scand. J. For. Res. 19, 14–29 doi: 10.1080/02827580310019536


Ziaco, E., Biondi, F., Rossi, S., and Deslauriers, A. (2014). Intra-annual wood anatomical features of high-elevation conifers in the Great Basin, USA. Dendrochronologia 32, 303–312. doi: 10.1016/j.dendro.2014. 07.006

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Battipaglia, Campelo, Vieira, Grabner, De Micco, Nabais, Cherubini, Carrer, Bräuning, Cufar, Di Filippo, García-González, Koprowski, Klisz, ˇ Kirdyanov, Zafirov and de Luis. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# TreeWatch.net: A Water and Carbon Monitoring and Modeling Network to Assess Instant Tree Hydraulics and Carbon Status

#### Kathy Steppe<sup>1</sup> \*, Jonas S. von der Crone<sup>1</sup> and Dirk J. W. De Pauw1,2

<sup>1</sup> Laboratory of Plant Ecology, Department of Applied Ecology and Environmental Biology, Faculty of Bioscience Engineering, Ghent University, Ghent, Belgium, <sup>2</sup> Phyto-IT, Ghent, Belgium

TreeWatch.net is an initiative that has been developed to watch trees grow and function in real-time. It is a water- and carbon-monitoring and modeling network, in which highquality measurements of sap flow and stem diameter variation are collected on individual trees. Automated data processing using a cloud service enables instant visualization of water movement and radial stem growth. This can be used to demonstrate the sensitivity of trees to changing weather conditions, such as drought, heat waves, or heavy rain showers. But TreeWatch.net's true innovation lies in its use of these highprecision harmonized data to also parameterize process-based tree models in real-time, which makes displaying the much-needed mechanisms underlying tree responses to climate change possible. Continuous simulation of turgor to describe growth processes and long-term time series of hydraulic resistance to assess drought-vulnerability in realtime are only a few of the opportunities our approach offers. TreeWatch.net has been developed with the view to be complementary to existing forest monitoring networks and with the aim to contribute to existing dynamic global vegetation models. It provides high-quality data and real-time simulations in order to advance research on the impact of climate change on the biological response of trees and forests. Besides its application in natural forests to answer climate-change related scientific and political questions, we also envision a broader societal application of TreeWatch.net by selecting trees in nature reserves, public areas, cities, university areas, schoolyards, and parks to teach youngsters and create public awareness on the effects of changing weather conditions on trees and forests in this era of climate change.

Keywords: sap flow, stem diameter variation (dendrometer), process-based modeling, vegetation modeling, turgor, hydraulic failure, plant growth, drought

#### INTRODUCTION

Climate change is impacting forests worldwide, threatening biodiversity, and ecosystem function and services (Anderson-Teixeira et al., 2015). Biological consequences of climate change are already apparent (Hughes, 2000), including increased tree mortality through drought, heat stress, insect infestation, and disease outbreaks (Anderegg et al., 2015; Teskey et al., 2015). Just as internationally coordinated forest monitoring initiatives boosted in the 1970s to respond to

Edited by:

Cristina Nabais, University of Coimbra, Portugal

#### Reviewed by:

Tobias Gebauer, University of Freiburg, Germany Zhenzhu Xu, Institute of Botany, Chinese Academy of Sciences, China

> \*Correspondence: Kathy Steppe kathy.steppe@UGent.be

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 10 March 2016 Accepted: 22 June 2016 Published: 05 July 2016

#### Citation:

Steppe K, von der Crone JS and De Pauw DJW (2016) TreeWatch.net: A Water and Carbon Monitoring and Modeling Network to Assess Instant Tree Hydraulics and Carbon Status. Front. Plant Sci. 7:993. doi: 10.3389/fpls.2016.00993

urgent scientific, political, and societal questions related to forest decline in relation to air pollution (Rautio and Ferretti, 2015), large monitoring networks around the globe (**Table 1**) are now investigating forest ecosystem responses to climate change. In particular, understanding the biological response of forests to climate change remains a great challenge, but is critical to biodiversity conservation, and management of ecosystem function and services (Anderson-Teixeira et al., 2015). Data sets from various monitoring initiatives and model forecasts are two essential components to no only understand forest ecosystem responses to climate change, but are also essential to support forest decision makers (Lindner et al., 2014). Adaptation strategies need a framework that includes monitoring and modeling activities as otherwise the forest manager will be grappling in the dark when making decisions (FAO, 2012).

While data from the larger networks (**Table 1**) add to the abundant evidence that forests globally are changing, it remains difficult to identify the mechanisms underlying such changes (Anderson-Teixeira et al., 2015). Focusing on responses of individual trees could single out the relative contributions of these underlying mechanisms. Indeed, trees have been named the 'living laboratories' for climate change responses (Farrell et al., 2015). A tree, like a human, can be viewed as a complex organism with an array of regulatory mechanisms to keep critical systems operating within appropriate bounds and mechanisms to repair damage that may occur when these bounds are exceeded (Anderegg et al., 2012).

In existing networks (**Table 1**), individual tree measurements are, despite their enormous potential, limited to incremental growth, which is conventionally measured every 3–5 years with a tape measure or with simple band dendrometers to track changes in stem circumference. These tree readings can help to start answering major questions about climate change and the potential uptake of CO<sup>2</sup> emitted by human activity, but how carbon sequestration and the size of the carbon sink will alter with climate change remains highly uncertain (Popkin, 2015). Because radial stem growth is highly dynamic, speciesspecific, and dependent on environmental factors (see Köcher et al., 2012, 2013), it is a good indicator of tree vitality and of tree responses to environmental stress (Dobbertin, 2005). Modern systems connected to data loggers monitor changes in stem radius (or diameter) at high-frequency time resolution (minute scale), and use these as biological drought and growth indicator (e.g., TreeNet<sup>1</sup> , Zweifel, 2016). In addition, a close relationship with net ecosystem productivity, which integrates fluxes over the entire forest ecosystem, has been observed, and although causal explanation for this strong relationship is still fragmentary, it points to a compelling complementarity between both monitoring systems (Zweifel et al., 2010; Gea-Izquierdo et al., 2014).

Continuous time series of stem diameter variations capture diel tree water relations (reflected in reversible shrinkage and swelling) superimposed onto (long-term) irreversible radial stem growth (Steppe et al., 2006; Zweifel et al., 2006; De Swaef et al., 2015). Indeed, changes in diameter result from interactions of water and carbon inside the tree stem (Steppe et al., 2015a). These changes are a great source of tree physiological and ecological information (Zweifel, 2016), which can assist disentangling the mechanisms underlying forest ecosystem responses to climate change. In this paper, we therefore draw on the analogon of an intensive care unit in medical sciences to propose an approach to assess in real-time the tree's trim by considering the tree as a complex organism that transports water from the soil to the atmosphere and sugars from the leaves (sources) to other plant organs (sinks), and whose survival hinges on maintenance of both transport systems. Monitoring equipment on the individual tree level is then used in combination with process-based modeling to translate raw sensor readings into physiological relevant measures, and to simulate key variables, which might be difficult to measure otherwise. We argue that an integrated approach that considers both tree monitoring and process-based modeling is needed to accurately predict forest dynamics in a changing climate.

Smart selection of a certain number of trees in a forest, and of different species in a stand, across different locations and ecosystems over longer time periods, will provide insight into growth, survival and adaptation strategies at the tree and ecosystem scales. The selection of trees and their up-scaling can be inspired by studies that have successfully scaled-up water use from tree to stand level (e.g., Granier et al., 1996; Köstner et al., 1998; Gebauer et al., 2012). Continuous measurements and modeling on the individual tree level are currently lacking in existing networks (**Table 1**) and are largely absent from process-based ecosystem models, while they can provide vital information on internal physiology of tree hydraulic and carbon status complementary to large-scale fluxes measured by eddy covariance and will better inform projections of forest ecosystem responses to climate change.

## THE TREEWATCH.NET INITIATIVE

#### Tree Monitoring

Climate change is expected to drive important changes in tree physiology with manifold but not yet fully understood impacts on forest ecosystem function and services. In this paper, we strongly support the recent calls to focus experimental, observational, and modeling efforts on the tree level to improve our understanding of climate change impacts on forests (Fatichi et al., 2014; Steppe et al., 2015a). We therefore view the tree as a complex organism that needs monitoring equipment to capture a 'heartbeat', which informs us on its actual trim. In contrast to humans, a tree does not have a real heartbeat, but continuous measurements with plant sensors on the tree stem display periodic signals that resemble a human's electrocardiogram (Steppe et al., 2015a). These signals inform us on changes in plant hydraulics and carbon metabolism in xylem and phloem tissues (Steppe et al., 2015a; Zweifel, 2016). From the reviewed methods to quantify real-time water and carbon dynamics within a tree stem (Steppe et al., 2015a), a set of two sensors has currently been selected as basic monitoring equipment: a sap flow sensor and a stem diameter variation sensor (**Figure 1**).

<sup>1</sup>www.treenet.info

#### TABLE 1 | Important international large-scale forest monitoring networks.


Each initiative is briefly defined, and main goals and approaches have been summarized. The year of foundation and current extent are also given. Networks are sorted from oldest to newest.

<sup>a</sup>http://www.iufro.org

<sup>b</sup>http://www.forestgeo.si.edu (Anderson-Teixeira et al., 2015)

<sup>c</sup>http://icp-forests.net

<sup>d</sup>http://icp-forests.net/page/icp-forests-manual

<sup>e</sup>http://fluxnet.ornl.gov

<sup>f</sup>http://www.ilternet.edu

<sup>g</sup>https://www.lternet.edu

<sup>h</sup>http://www.lter-europe.net

<sup>i</sup>https://www.icos-ri.eu

TreeWatch.net, a website built on top of the PhytoSense cloud service, to report instantly on each monitored tree's health status. The unique approach of combining continuous tree measurements with process-based modeling lays the ground for the next-generation global maps displaying direct biological responses of the sampled trees; information which is currently lacking to bridge the gap with meteorology.

When combined with mechanistic modeling (Section "Process-Based Tree Modeling"), these measurements allow revealing the internal tree hydraulics and carbon status. As science evolves, other monitoring equipment, such as acoustic emission sensors (De Roo et al., 2016), may be added or may replace existing ones.

Sap flow is measured with a sap flow sensor, which uses heat to sense water movement in the stem xylem and is typically expressed as sap flow rate (in g h−<sup>1</sup> ; Smith and Allen, 1996; Steppe et al., 2010; Vandegehuchte and Steppe, 2013). Accurate estimates of sap flow are essential in our tree monitoring approach to assess

changes in tree hydraulics, internal water storage dynamics, and to quantify whole-tree water use, but also to estimate the tree's carbon budget and stem respiration based on measurements of xylem CO<sup>2</sup> transport and stem CO<sup>2</sup> efflux (Teskey et al., 2008; Steppe et al., 2015a).

Point dendrometers or linear variable displacement transducers (LVDTs) measure variations in stem diameter (mm) at high temporal resolution (minute scale). The sensor signal simultaneously displays the integrated result of: (i) irreversible radial xylem and phloem growth; (ii) reversible shrinking and swelling of the living stem cells due to changes in internally stored water; (iii) contraction and expansion of dead conducting xylem elements due to the increase and relaxation of internal tensions; and (iv) thermal expansion and contraction of the stem (Daudet et al., 2005; De Swaef et al., 2015). Because of the tight coupling between tree hydraulics and radial stem growth and, hence, carbon metabolism, variations in stem diameter are the second vital component in our tree monitoring approach.

#### Process-Based Tree Modeling

Current spatiotemporal knowledge of climate-forest dynamics is primarily based on simulations by dynamic global vegetation models (DGVMs). Although turgor, or the positive pressure potential in living cells, is the critical component in quantifying growth (Lockhart, 1965; Génard et al., 2001; Steppe et al., 2006), all existing DGVMs simulate long-term tree and forest stand growth using photosynthesized net carbon as source, which is then, according to allometric rules or simplified functional allocation schemes, partitioned among different carbon pools without considering tree hydraulics. Therefore, Fatichi et al. (2014) correctly call for moving from such a carbon source to a more sink-driven vegetation modeling in which water transport and turgor play a key role. As discussed previously (Steppe et al., 2006, 2015a), turgor is not only affecting cell wall expansion or irreversible radial growth (Lockhart, 1965), but also other growth processes, such as cell formation, deposition and assembly of new cell wall material depend on turgor and cell volume (Boyer, 1968; Ray, 1987; Proseus and Boyer, 2006). Because turgor in living tree cells is mainly built-up during night upon refilling of dehydrated tissues, growth processes mainly occur during the night, and are only optimal when tree water status is also optimal (Daudet et al., 2005; Saveyn et al., 2007; Steppe et al., 2015a). If we aspire a better spatiotemporal description of water fluxes together with more realistic scenarios for future climate and the carbon cycle (Friedlingstein et al., 2006; Thornton et al., 2007; Bonan et al., 2011), next generation DGVMs need to include a mechanistic understanding of water and carbon dynamics in xylem and phloem, and their interactions, at the tree level.

During the past few decades, implementation and application of process-based tree models has greatly advanced our knowledge on plant hydraulic functioning and growth (Steppe et al., 2015b). Started from a simple Ohm's law analog model proposed by van den Honert (1948) for steady-state water transport, process-based tree models have greatly improved since, and large efforts have recently being put into the integration of xylem and phloem transport pathways, given the important coupling between hydraulic processes and the transport and allocation of carbohydrates (Höttlä et al., 2009; De Schepper and Steppe, 2010; Hubeau and Steppe, 2015). To further our knowledge of climate change impacts on the forest scale, process-based tree models are likely to become increasingly important.

In our approach, we advocate a combination of processbased tree modeling and continuous measurements at the tree level to better understand impacts of climate change on forests. Given that the current 'heartbeat' tree monitor consists of sap flow and variation in stem diameter, any process-based model that interlinks both processes is a direct candidate for our framework. The history and the current-state-of-the-art of possible candidate process-based models have recently been reviewed (De Swaef et al., 2015). The models we typically consider simulate tree sap flow dynamics, which can be directly linked to variations in stem diameter by using radial flow of water between xylem and phloem (**Figure 1**). The radial water flow causes changes in stem water content via a hydraulic capacitance, which results in changes in turgor, and drives irreversible radial stem growth according to Lockhart's (1965) equation on top of elastic shrinkage and swelling. Steppe et al. (2006) originally developed such a so-called flow and storage model. Of particular interest for our approach is that such models feature essential hydraulic parameters (resistance and capacitance), and enable simulation of vital, but often difficult to measure variables (earlier described turgor, water potential), which all play an important role in hydraulic failure, tree mortality, and, therefore, long-term forest dynamics (Fatichi et al., 2014).

#### Phytosense Cloud Service

Whereas continuous tree measurements, including sap flow and stem diameter variation, have been recognized as promising technology for monitoring tree hydraulics and carbon status (Anderegg et al., 2012; Köcher et al., 2013; De Swaef et al., 2015; Steppe et al., 2015a; Zweifel, 2016), and the use of processbased tree models is expected to get a boost given the recent recommendations on next-generation DGVMs (McDowell et al., 2013; Fatichi et al., 2014), no existing framework combines continuous tree readings with mechanistic modeling in realtime. This is exactly what our approach is aiming at: instant information on tree hydraulics and carbon status using continuous measurements and process-based model simulations.

To optimally combine the continuous tree measurements with the process-based simulation models, and to ensure real-time visualization of the tree's hydraulic function and carbon status, the PhytoSense cloud service is used in our approach (**Figure 1**). This cloud service is the 'brain' behind the commercial plant monitoring system PhytoSense<sup>2</sup> and is responsible for real-time data storage, data analysis, data processing, running model simulations and calibrations, and sending out notifications.

All processing on the cloud service is performed automatically so that little or no user interaction is required. A powerful

<sup>2</sup>Developed by Phyto-IT (Mariakerke, Belgium): http://www.phytosense.net

system based on 'transformations' has been developed for real-time conversion of raw into processed data. A wide range of transformations is available: averaging, cumulating, summing, integrating, filtering, minimum/maximum, and the ability to apply any arbitrary equation to the data. More advanced transformations are also available to calculate sap flow rates in real-time, and to automatically remove disturbances from diameter variation signals. Once defined, transformations are automatically applied each time new data is received. Besides transformations, PhytoSense also allows to run dynamic simulation models in real-time. Although not required, models are typically first implemented in the plant modeling software PhytoSim<sup>3</sup> and then converted into optimized code, which can run on the PhytoSense platform. These models can be any set of algebraic and (first order) differential equations [see for instance Steppe et al. (2006) or De Swaef et al. (2015)], and can be automatically recalibrated at certain time intervals using a moving window calibration procedure [see also Steppe et al. (2008)]. This lays the ground for novel stress detection approaches and ecophysiological warning systems, because daily estimates of the calibrated model parameters can now be displayed as time series in real-time from which important tree physiological behavior can be derived. Finally, notifications can be generated when measured or simulated data is below or above a threshold value for a specified amount of time, when a sensor is offline for a specified amount of time or when a model parameter exceeds the appropriate bounds.

PhytoSense provides a flexible API (Application Programming Interface) that allows any internet-connected device to connect to it. Any online data logger can use the API to send data to PhytoSense and custom-build applications or websites can use the API to visualize the available data. This makes the data from the TreeWatch.net trees readily available, which fits the 'Internet of Things' vision of this era.

#### TreeWatch.net and the Way Forward

TreeWatch.net<sup>4</sup> is a website built on top of the PhytoSense cloud service (Section "Phytosense Cloud Service"; **Figure 1**). TreeWatch.net originated at the Laboratory of Plant Ecology, Ghent University, Belgium, to show how the combination of tree monitoring and process-based modeling can significantly contribute to instant assessment of stress impacts on tree hydraulics and carbon status. The unparalleled enthusiasm and interest of the COST STReESS<sup>5</sup> community in such a measuring/modeling framework gave the impetus to develop TreeWatch.net further as a global water and carbon monitoring and modeling network for advanced research on the dynamic interplay between trees and the regional climate.

Currently, TreeWatch.net monitors beech (Fagus sylvatica L.) and oak (Quercus robur L.) trees in the experimental forest Aelmoeseneie of Ghent University, Belgium. Sap flow is measured with custom-built Sapflow+ sensors (Vandegehuchte and Steppe, 2012) and stem diameter variations are recorded with a point dendrometer (model ZN11-T-WP, Natkon, Switzerland). We plan to gradually extend the network by adding trees across north-south trajectories in different populations in Europe, and in other continents, to profit from a wide climatic gradient going from low temperatures in the northern sites to warm and dry conditions in the southern sites, where tree responses are expected to be temperature- and drought limited, respectively. Trees will be sampled according to a stringent protocol taking into account various tree characteristics (e.g., tree status, tree height, stem diameter, and leaf area), and will be equipped with standardized plant sensors to avoid variability in the collected data due to different sensor types. Sensors connected to data loggers with wireless data transfer and remote control accessibility are used to send the data to the PhytoSense cloud service. The harmonized data offered by TreeWatch.net will be used in an innovative way to parameterize real-time process-based tree models (Section "Process-Based Tree Modeling"), and to run the models to understand tree response to climate change and growth differences across trajectories from underlying water and carbon relationships. Modeling will enable us to put the continuous measurements in a larger context by helping us understand the more general concepts underlying growth and tree hydraulic functioning.

Continuous real-time model simulations of the muchneeded turgor when aspiring growth modeling, but also dynamics in model parameters, including hydraulic resistance and capacitance, are only a few of the opportunities that will be at hand to perform an integrated survey of tree responses to changes in the regional climate. These modeled features should be validated with ground-based data from fieldwork to increase confidence in the model, or to further improve it when discrepancies between modeled and measured data are observed. By visualizing hydraulic features, like hydraulic resistance, we will be the first to show changes in tree hydraulics and vulnerability to drought stress in real-time. The real-time aspect is a much-needed feature because now science relies on off-line, destructively collected vulnerability curves (Choat et al., 2012), which makes assessment indirect and therefore less reliable. In addition, TreeWatch.net aims at displaying pioneering maps with biological tree response to temperature and drought (**Figure 1**), which will be used to bridge the gap between tree functioning and meteorology (weather formation). Weather station data and soil moisture status at the sites, which are needed to interlink tree responses and regional climate, can be either additionally measured or accessed from existing networks (**Table 1**).

The results from TreeWatch.net are expected to spur discussion regarding long-standing assumptions for relationships between fluxes observed at the ecosystem level and the mechanisms responsible. Especially in DGVMs, the use of coarse scale observations and potentially incorrect mechanisms could mislead mitigation and adaptation plans of the future (Hanson and Gunderson, 2009). TreeWatch.net is, therefore, developed to be complementary to the data from these larger networks

<sup>3</sup>Developed by Phyto-IT (Mariakerke, Belgium): http://www.phyto-it.com/Phyto Sim.shtml

<sup>4</sup>http://treewatch.net

<sup>5</sup>COST Action FP1106 – Studying Tree Responses to extreme Events: a SynthesiS: http://www.streess-cost.eu

(**Table 1**) and may help to identify the much-needed mechanisms underlying changes in forests. But the use of TreeWatch.net is not limited to research in natural forest ecosystems only, as it can also be used in cities and urban plantings where trees are known to grow in a 'future climate' and can be used as 'living laboratories' to study plant responses to climate change (Farrell et al., 2015). At present, a maple (Acer pseudoplatanus L.) tree monitored at the Faculty of Bioscience Engineering<sup>6</sup> , Ghent University, serves this purpose.

TreeWatch.net is not only powerful in science, but is also able to serve an important educational role by teaching youngsters about the role of trees as regulators of the environment through so-called 'talking forests'<sup>7</sup> . In May 2015, the experimental forest of Ghent University was opened as such a 'talking forest', for which a dedicated website<sup>8</sup> has been designed using the flexible API of the PhytoSense cloud service. Assisted by experienced nature guides, students from both primary and secondary schools can now be invited to 'listen' to the trees and can find the real-time measurements on their phones or tablets to get insight into the interaction between climate and the forest.

#### CONCLUSION

TreeWatch.net primarily aims at addressing forest and environmental issues that are of concern for our society, and takes the challenge to provide answers to urgent scientific and political climate-change related questions. But because of its intrinsic educational power, one of the long-term dissemination perspectives of TreeWatch.net is that trees are also selected in nature reserves, MAB<sup>9</sup> -sites, public areas, cities, university areas, schoolyards, and parks to create global awareness on the effects

#### REFERENCES


of (extreme) weather conditions on trees growing across all continents.

#### AUTHOR CONTRIBUTIONS

KS initiated TreeWatch.net and designed the outline of the initiative. She directed the network toward simultaneous application of continuous tree measurements and process-based modeling. She established the first network in Belgium and added a city tree to demonstrate its potential. She supervises the practical work, the analysis and data interpretation, and she coordinates the modeling activities within the TreeWatch.net framework. DD developed the PhytoSense cloud service and supervises data acquisition, data processing and visualization. He also assists in modeling activities. JvdC assists in daily data inspection, data analysis and data interpretation. He also designed the website. KS wrote the manuscript, JvdC reviewed the existing networks, and JvdC and DD contributed by critically revising the draft. All authors approve the final version to be published.

#### ACKNOWLEDGMENTS

We thank Ghent University for the Science & Society project funding in support of the development of TreeWatch.net, and the Faculty of Bioscience Engineering for the opportunity to equip a faculty tree for UGent's Tipping Po!nt event. We are grateful to our technicians Philip Deman, Geert Favyts and Erik Moerman for their fantastic support with the installation of every tree in the network. We also sincerely thank Ute Sass-Klaassen, as chair of the COST Action FP1106 network STReESS (Studying Tree Responses to extreme Events: a SynthesiS) and the steering committee, for their trust and strong belief in the TreeWatch.net initiative. Our thanks also go to the COST STReESS community for their enthusiastic support and interest in TreeWatch.net. Special thanks goes to Janne Van Camp for the initial cartoons on TreeWatch.net.


<sup>6</sup> http://treewatch.net/faculty-of-bioscience-engineering/

<sup>7</sup> http://www.flanderstoday.eu/innovation/talking-forest-teaches-youngstersabout-regulating-environment

<sup>8</sup> http://www.aelmoeseneiebos.ugent.be/?p=pratendbos

<sup>9</sup> Man and the Biosphere


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Steppe, von der Crone and De Pauw. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Repeated Summer Drought and Re-watering during the First Growing Year of Oak (Quercus petraea) Delay Autumn Senescence and Bud Burst in the Following Spring

Kristine Vander Mijnsbrugge<sup>1</sup> \*, Arion Turcsán1,2,3, Jorne Maes<sup>4</sup> , Nils Duchêne<sup>4</sup> , Steven Meeus<sup>4</sup> , Kathy Steppe<sup>5</sup> and Marijke Steenackers<sup>1</sup>

<sup>1</sup> Department of Forest Genetic Resources, Research Institute for Nature and Forest, Geraardsbergen, Belgium, <sup>2</sup> Department of Biometrics and Agricultural Informatics, Corvinus University of Budapest, Budapest, Hungary, <sup>3</sup> Department of Forest Reproductive Material and Plantation Management, Institute of Silviculture and Forest Protection, West-Hungarian University, Sopron, Hungary, <sup>4</sup> Department of Agro- and Biotechnology, School of Technology, Odisee University College, Sint-Niklaas, Belgium, <sup>5</sup> Laboratory of Plant Ecology, Faculty of Bioscience Engineering, Ghent University, Ghent, Belgium

#### Edited by:

Achim Braeuning, University Erlangen-Nuremberg, Germany

#### Reviewed by:

Gerald Moser, Justus-Liebig-University Giessen, Germany Zhenzhu Xu, Institute of Botany – Chinese Academy of Sciences, China

\*Correspondence:

Kristine Vander Mijnsbrugge kristine.vandermijnsbrugge@inbo.be

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 20 January 2016 Accepted: 18 March 2016 Published: 31 March 2016

#### Citation:

Vander Mijnsbrugge K, Turcsán A, Maes J, Duchêne N, Meeus S, Steppe K and Steenackers M (2016) Repeated Summer Drought and Re-watering during the First Growing Year of Oak (Quercus petraea) Delay Autumn Senescence and Bud Burst in the Following Spring. Front. Plant Sci. 7:419. doi: 10.3389/fpls.2016.00419 Climate change predicts harsher summer droughts for mid-latitudes in Europe. To enhance our understanding of the putative impacts on forest regeneration, we studied the response of oak seedlings (Quercus petraea) to water deficit. Potted seedlings originating from three locally sourced provenances were subjected to two successive drought periods during the first growing season each followed by a plentiful rewatering. Here, we describe survival and phenological responses after the second drought treatment, applying general linear mixed modeling. From the 441 drought treated seedlings 189 subsisted with higher chances of survival among smaller plants and among single plants per pot compared to doubles. Remarkably, survival was independent of the provenance, although relatively more plants had died off in two provenances compared to the third one with mean plant height being higher in one provenance and standard deviation of plant height being higher in the other. Timing of leaf senescence was clearly delayed after the severe drought treatment followed by re-watering, with two seedlings per pot showing a lesser retardation compared to single plants. This delay can be interpreted as a compensation time in which plants recover before entering the subsequent developmental process of leaf senescence, although it renders seedlings more vulnerable to early autumn frosts because of the delayed hardening of the shoots. Onset of bud flush in the subsequent spring still showed a significant but small delay in the drought treated group, independent of the number of seedlings per pot, and can be considered as an after effect of the delayed senescence. In both phenological models significant differences among the three provenances were detected independent from the treatment. The only provenance that is believed to be local of origin, displayed the earliest leaf senescence and the latest flushing, suggesting an adaptation to the local maritime climate. This provenance also displayed the highest standard deviation of plant height, which can be interpreted as an adaptation to variable and unpredictable weather conditions, favoring smaller plants in drought-prone summers and higher plants in more normal growing seasons.

Keywords: drought, re-watering, oak provenance, seedling, survival, leaf senescence, bud burst, general linear mixed models

## INTRODUCTION

fpls-07-00419 March 28, 2016 Time: 16:13 # 2

Predicted climate change in temperate regions raises concerns about the ability of forest ecosystems to cope with longer and more severe summer drought periods. The intergovernmental panel on climate change (IPCC, 2014) expects not only a raise in mean temperature but also a higher frequency of extreme weather events. Forest vitality will be challenged and forests will become more vulnerable not only in Europe (Lindner et al., 2010), but all over the globe (Choat et al., 2012). For Europe, and more specifically for Belgium, climate projections predict increasing temperatures and irregular precipitation patterns in summer, augmenting the number and the intensity of drought periods (Baguis et al., 2010). A large part of the forests in the lower countries grows on sandy soils which are characterized by a relatively low water holding capacity, making them specifically vulnerable to extreme drought events during the growing season (Van der Werf et al., 2007), implying that drought and heat tolerance will become more critical for tree survival, especially considering additional stresses for trees by competition with others for light, water, and nutrient sources. Among important European tree species, oaks are well-known to be tolerant to drought (Leuschner et al., 2001), having a xeromorphic leaf structure and an adapted root structure that can cope with temporal and spatial variability in soil water and nutrient availability, and displaying an ability to rapidly resume assimilation after periods of water deficiency (Kubiske and Abrams, 1993; Galle et al., 2007; Kuster et al., 2013a,b). For this reason, oaks are put forward as promising candidate tree species to replace drought sensitive species such as beech (Fagus sylvatica) or spruce (Picea abies) on warm and dry sites in Europe (Leuschner et al., 2001). Apart from interspecific differences in degrees of drought tolerance, also among oak species, intraspecific variation may occur among provenances originating from varying growth sites displaying varying responses to drought. Provenances from xeric sites have been suggested to be better adapted to enhanced temperatures and lower water availability than provenances from more humid sites (Bruschi, 2010; Jensen and Hansen, 2010) although Arend et al. (2011) did not detect a correlation between climate at the site of origin and response to drought in different provenances of three European oak species. A reduced aboveground growth pattern with a diminished biomass production, together with a shift toward below-ground root growth, are wellestablished responses to drought in oak species (Broadmeadow and Jackson, 2000; Thomas and Gausling, 2000; Arend et al., 2011; Spiess et al., 2012; Kuster et al., 2013b) whereas the effects on phenology are less thoroughly examined and mainly described as an earlier stop of height growth under dry growing conditions (Jensen and Hansen, 2010; Spiess et al., 2012), also visible in an earlier cessation in secondary (radial) growth (Pflug et al., 2015) and which may show an after effect in the subsequent spring by an advanced bud burst (Kuster et al., 2014).

Plants have adapted the timing of their seasonal developmental processes to environmentally favorable periods of the year. Many aspects of the different phenophases that characterize growth in perennials, such as bud burst, growth cessation, senescence, bud set, and release from dormancy, are regulated by local climate (Rohde et al., 2011). In temperate regions, deciduous trees are prepared to the harshness of the winter by autumn senescing of the leaves, in which nutrients are efficiently remobilised before leaves are shed, and by hardening of the shoots to protect them against frost damage (Keskitalo et al., 2005). Autumn senescence in most trees is triggered by the photoperiod, a stable environmental cue that is considered to be more reliable than temperature as a harbinger of the first frosts (Keskitalo et al., 2005; Jackson, 2009). Still, concurrent with photoperiod, temperature has been shown to play a pivotal role in timing of the phenophases that determine the end of the growing season's length in trees. Translocation of poplar clones (clonally replicated material) to different latitudinal growth sites revealed the effect of temperature, in conjunction with photoperiod, on timing and duration of the bud set process (Rohde et al., 2011). Apart from photoperiod and climatic factors, to which trees are evolutionary adapted, stress factors may influence the timing of growth stop, senescence and autumnal bud set. Plants may induce leaf senescence upon drought stress, allowing a reallocation of nutrients within the plant and thus enhancing its chances on survival, an adaptive mechanism that is well-studied among Mediterranean plant species (Munne-Bosch and Alegre, 2004).

The seedling stage of a forest tree is known to be the most vulnerable phase in its life cycle and therefore, understanding the stress responses of seedlings is crucial for predicting forest tree growth and survival (Niinemets, 2010; Psidova et al., 2015). Oak forests in Belgium are mainly small and fragmented, and additionally originated largely from plantations with forest reproductive material, although in recent decades natural rejuvenation has been emphasized and promoted for forest regeneration. In the context of climate change, information regarding sessile oak provenances tolerant to drought becomes more important, particularly for forest management and future (re)forestation. The purpose of this paper is to examine the impact of severe summer droughts on the survival of first year seedlings of three different sessile oak provenances, sourced in the same region in Belgium but with a variable origin, and on the phenology of the surviving seedlings in this vulnerable phase of their life. We specifically aimed to determine the survival rate and the effect on leaf senescence in the first growing season and on bud burst in the subsequent spring for surviving seedlings, all in relation to the origin and the size of the seedlings, the degree of water deficit and the number of seedlings per pot. We tested the hypothesis that sessile oak which is believed to be local of origin is more drought–tolerant compared to putative non-local origins.

## MATERIALS AND METHODS

#### Source Material

Three provenances of Quercus petraea were chosen in Flanders, the northern part of Belgium, with deviating stand structure and history: Klaverberg (KLA), Voeren (VOE), and Borgloon (BOR;

**Figure 1**). Acorns were collected per mother tree at the end of October 2013. KLA is a small relict of abandoned oak coppice wood growing on inland sand dunes within a former heath land. The oaks here are most probably local of origin (Vander Mijnsbrugge et al., 2003). Coppice wood in the former heath land was exploited by local poor farmers that would not buy planting stock when local sources were available. Acorns were collected from 13 visually older abandoned coppice stools. As the oaks are mostly growing widely spaced on the sand dunes, the chance on mixing acorns from different mother trees was negligible. VOE is a classical planted forest stand, even aged and approximately 80 years-old, growing on a loamy soil type. As is the case for the vast majority of such planted oak stands in Flanders, the origin of the original planting stock is unknown. BOR is a similarly planted forest and approximately 100 years-old, of which the origin of the planted material is also unknown. Here, the oaks grow on sandy soil. Acorns were collected underneath 14 dominant trees in VOE and three dominant trees in BOR, which all showed a well-developed crown. Collection was performed only close to the stem, minimizing the chance on mixing acorns between different mother trees.

#### Germination of the Acorns

In November 2013, the collected seeds were sown in forestry trays with two seeds per cell, using standard nursery potting soil. During winter, the trays were watered manually keeping the soil moist. The experiment was located in a greenhouse with automatic temperature regulation, keeping the greenhouse frost-free in wintertime, but without additional heating. An automatic internal gray shade cloth system operates in the greenhouse, protecting the plants from high levels of insulation. In total, 1015 seeds germinated, 486 from KLA, 431 from VOE and 145 from BOR. All germinating plants were given water at regular times according to the visual needs as judged by experienced greenhouse workers. Seedlings were transferred in April 2014 to 1-lt pots (12 cm × 11 cm × 11 cm) using

standard nursery potting soil [organic matter 20%, pH 5.0– 6.5, Electrical Conductivity (EC): 450 µS/cm, dry matter 25%, fertilization: 1.5 kg/m<sup>3</sup> powdered compound fertilizer NPK 12 + 14 + 24]. As not all seeds had germinated, both cells with only one seedling and cells with two seedlings were present. While transferring the seedlings, double plants in one tray cell were kept together. After the transfer, the seedlings were not additionally fertilized. For our experiment we choose to use seedlings in pots, rather than working in a field experiment outdoor, as this allowed to impose a drought period on a subset of plants while both treated and control plants could be subjected to very similar other growth conditions (light, temperature, nutrient availability, . . .). Furthermore, it allowed monitoring the reduction in water availability by weighing of the individual pots.

#### Drought Treatment, Measurements, and Scoring

The pots were divided in two groups: a control and a treatment group. In both groups the three provenances were individually mingled at random (completely randomized). On 15th May and 6th August 2014 respectively the two groups of plants were soaked overnight to a fully water saturated condition in a basin with the water level up to two cm above the bottom of the pots. Up to 1st July and 17th October 2014 respectively the drought-treated group was not watered anymore, whereas the control group was further watered according to the visual needs of the plants. All plants were re-watered on 2nd July and 18th October 2014 respectively by soaking the two groups of plants in the same basin in the same way. After these drought periods and the re-watering, both groups were kept in wellwatered conditions according to the visual needs of the plants. The first drought period lasted until stress was detected in stomatal conductance as measured by a porometer on a subset of plants (**Figure 2**). For this, 30 pots with relative high plants were randomly chosen from the control group as well as 30 from the drought-treated group to monitor the treatment effect. Leaf stomatal aperture in terms of leaf resistance to water vapor was measured weekly with a diffusion porometer (Model AP4, Delta-T Devices, Cambridge, UK) during the entire first drought period. As stomata are sensitive to drought stress, high resistance values represent a closing reaction and declining stomatal conductance and leaf assimilation rate. The porometer measurements were conducted during daytime between 10 a.m. and 3 p.m. In the period directly following re-watering after the first drought period, an extra growth flush was detected mainly among the stressed plants (Turcsán et al., 2016). The second drought period lasted until a large amount of plants showed visual signs of stress (wilting and/or curling of the leaves) and started dying off. During this second drought period, the wilting and/or curling of the leaves in the treatment group of plants was monitored visually on a weekly basis on 100 random plants. The number of plants with clear visual stress symptoms was counted, as indicated in **Figure 2**. The data of the second drought period are presented here. Therefore, reference to a drought period in this paper concerns the second drought

period. After the second re-watering, all plants were kept wellwatered according to visual needs, also during the winter and the following spring.

During the drought period all pots were weighed nearly weekly to measure the water loss (**Figure 2**). The initial weight at the beginning of the treatment period was measured after the pots had drained the excess of water. At the end of the treatment period drought symptoms had become clearly visible and plants had started dying off (**Figure 2**). As an approximation for the level of water deficit experienced by the seedlings, the weight loss of the individual pots at the end of the drought period, just before re-watering, was calculated relative to the initial weight at the beginning of the treatment: relative weight loss = (weight doy 218 – weight doy 281)/weight doy 218. This approximation is related to the soil relative water content (SRWC) which is the ratio between present soil moisture and field capacity (Xu et al., 2009), and is applicable to a large amount of potted plants.

During the drought period a large amount of plants died off. Therefore, survival was monitored as a separate binary variable. The height of the seedlings was measured with a ruler at the end of the first drought treatment (on 1st July 2014), at full recovery of the plants (on 4th September 2014) and at the end of the growing season (on 1st of December 2014). Two phenophases were scored on the plants: leaf senescence in autumn 2014 and bud burst in 2015. Leaf senescence on the surviving plants after the second drought treatment and the control group was scored following a eight level scoring protocol (**Table 1**) on 1st and 8th December (doy 335 and 343). Here, all the leaves of a seedlings were observed together and a visual mean of color change was made. Bud burst and leaf unfolding in the apical bud in the following spring was scored according to a six levels scoring protocol (**Table 1**) on 22th and 30th April, and 8th May (doy 112, 120 and 128).

#### Data Analysis

All statistical analyses were performed in the open source software R 3.1.2 (R Core Team, 2014). Three response variables were modeled using generalized linear mixed models: survival, leaf senescence and bud burst. The number of seedlings used in each model are indicated in **Table 2**. Survival was examined using logistic regression (generalized linear mixed models) in the package lme4 (Bates et al., 2015), whereas the phenological response variables were modeled using cumulative logistic regression in the package ordinal (Christensen, 2013). The command clmm in the package ordinal models the chance to maximally have reached a given level of the ordinal response variable. We ordered the score levels of leaf senescence and bud

TABLE 1 | Description of the score levels of the two phenological response variables leaf senescence and bud burst in oak seedlings (Quercus petraea).


burst in decreasing order, so that the chance to have maximally reached, e.g., leaf senescence score 4 equalled the chance to have reached maximally score 4, which included scores 8, 7, 6, 5 and 4, which can be interpreted as having reached at least score 4.

In the fixed part of the models several covariates were examined for significant explanatory power: the plant height before the second drought treatment (continuous variable), the provenance of the seedlings (factor variable) and the number of seedlings per pot (factor variable). As the plants did not show a height increment during or after the second drought period, the plant height before the second drought period was the final height growth of the growing season 2014. The two phenological models leaf senescence and bud burst got an additional covariate, day of observation, as for these response variables repeated observations per plant were available. All three covariates height, provenance and seedlings per pot were first included in each of the three models, survival, leaf senescence and bud burst, with an interaction term with weight loss of the pots relative to the fully water saturated condition during the second drought period. In this way the influence of the drought treatment was examined. In all models the mother plant from which acorns were collected was in the random part (random intercept). For the phenological models, an additional unique plant identity variable was added in the random part of the models (random intercept) to account for the repeated measurements on the same plants. Using drop 1 (a likelihood ratio test) the fixed part of all three models was reduced up to only significant terms. With a significant interaction term, the corresponding covariates (main effects) remained in the model.

The chance (p) of survival was calculated following a logistic regression:

$$\log(\mathbf{p}/(1-\mathbf{p})) = \alpha + \beta\_\mathbf{P}\mathbf{P} + \beta\_\mathbf{H}\mathbf{H} + \beta\_\mathbf{S}\mathbf{S} + \beta\_\mathbf{W}\mathbf{W} + \beta\_\mathbf{P}\mathbf{P}\mathbf{W} + \mathbf{P}$$

$$\beta\_\mathbf{H}\mathbf{H}\mathbf{W} + \beta\_\mathbf{S}\mathbf{S}\mathbf{W}$$

with α as the estimated intercept and the β's as the estimated parameters of the fitted model. Shown here is the full model with all covariates and interaction terms. P is the provenance (KLA, VOE, and BOR), W is the relative weight loss of the pots accounting for the stressed condition, H is the plant height and S is the number of seedlings per pot (1 or 2). The model was reduced up to only significant terms.

For leaf senescence and bud burst, the chance (p) to have reached at least a given phenological score level on a given day was calculated following a cumulative logistic regression:

$$\begin{array}{c} \log(\text{p/(1-p)}) = \alpha\_{\text{T}} - \beta\_{\text{D}} \text{D} - \beta\_{\text{P}} \text{P} - \beta\_{\text{H}} \text{H} - \beta\_{\text{S}} \text{S} - \beta\_{\text{W}} \text{W} - \beta\_{\text{S}} \text{W} \\ \beta\_{\text{P}} \text{P} \text{W} - \beta\_{\text{H}} \text{H} \text{W} - \beta\_{\text{S}} \text{S} \text{W} \end{array}$$

where the β's are the estimated parameters of the fitted model and with α<sup>T</sup> as an estimated threshold value for the passing on from one level of the phenological variable to the next. D is the day of observation. Based on the formulas of the reduced models with only significant terms in the fixed part, time lags were calculated between control and stressed condition, or between the different provenances or between single and double plants per pot.

## RESULTS

#### Survival

The binary response variable survival indicating whether or not a seedling survived the drought period, was modeled using generalized linear mixed models. Influencing factors were the height of the seedlings and the number of seedlings per pot, both depending on the amount of relative weight loss (significant interaction terms, **Table 3**). Interestingly, provenance was not significant in this model, indicating that the provenance of the seedlings did not affect survival rate. Seedlings that shared a pot displayed a lower probability of survival in the drought stressed condition (**Table 3**; **Figure 3**). In addition, the higher the plants, the greater the probability to die off in the stressed condition (**Table 3**; **Figure 3**).

#### Leaf Senescence

Decolouration and finally shedding of the leaves was scored in the autumn of 2014, following a severe drought and rewatering period. In the phenological model, provenance (without interaction term) and the interaction between number of seedlings per pot and relative weight loss appeared significant (**Table 4**). With a more severe drought stress, as expressed by a high relative weight loss, decolouration of the leaves was clearly retarded (**Figures 4** and **5**). Both in the control group as in the drought treated group (thus independent of the drought treatment), leaf senescence appeared first in KLA, than in VOE and finally in BOR (**Figures 4** and **5**). The significant interaction term with seedling per pot was visualized by differing steepness of the modeled S curves between single and double plants per pot (**Figure 5**) indicating that among the double plants per pot in stressed conditions senescence of the leaves was less retarded compared to single plants. According to the fitted model, the control and stressed group of plants differed 18.8 days in timing of leaf senescence for single plants per pot. In the control group double plants per pot senesced 1.5 days later compared to the


TABLE 2 | Number of oak seedlings (Q. petraea) in this study (n) for the different modeled response variables survival (nsu), leaf senescence (nls) and bud burst (nbb), and according to treatment, provenance and number of seedlings per pot.

For the response variable survival, the number of surviving plants are indicated (na), also expressed in % (na%).

TABLE 3 | Model statistics for the general linear mixed model of the binary response variable survival.


W: relative weight loss (continuous variable), H: height of the plant (continuous variable), S: number of seedlings per pot (factor variable with single plant per pot as standard level). Significant results are in bold: ∗∗P < 0.01; ∗∗∗P < 0.001.

single plants, whereas they senesced 6.4 days earlier than the singles in the stressed condition. For the different provenances, a time lag was observed between KLA and VOE of 6.7 days and between KLA and BOR of 14.7 days (independent of the drought treatment).

#### Bud Burst

Bursting of the buds and leaf unfolding was observed in the spring following the growing season with drought treatments. Modeling this ordinal phenological variable revealed a significant influence of relative weight loss, indicating that the drought treated group of plants burst buds later compared to the control group (**Table 4**; **Figure 6**). Provenance and plant height were covariates in the model with a significant explanatory power, both independent of the treatment (no significant interaction terms with relative weight loss). The difference in timing of bud burst between the control and stressed group of plants was 1.5 days. The provenances VOE and BOR flushed earlier than KLA (**Figure 6** with VOE and KLA differing 4 days) and smaller plants burst buds earlier than larger plants (**Figure 7**), both independent of the relative weight loss of the pots (no significant interaction term with relative weight loss).

FIGURE 3 | Modeled probability of survival depending on treatment and on number of seedlings per pot. To calculate the probabilities, the mean relative weight loss for the control and stressed group of plants was applied.

## DISCUSSION

As climate change is predicted to augment the length and severity of drought periods during the growing season in Belgium, we studied the impact of water deficit in the first year on survival of three provenances of sessile oak seedlings (Q. petraea), and on the phenology of surviving plants. Our central aim was to contribute to a better understanding of possible impacts of summer droughts on forest regeneration. Two major results can be stressed. Firstly, there was no significant difference in survival after a severe summer drought period between the three local provenances. Secondly, the drought period followed by a plentiful re-watering induced a retarded leaf senescence with a delayed bud burst, likely as an after affect, in the subsequent spring.

#### TABLE 4 | Model statistics for the general linear mixed model of the ordinal phenological response variables.


D: day of observation (continuous variable), VOE and BOR: oak provenances that are compared to the standard provenance KLA (factor variable), W: relative weight loss (continuous variable), S: number of seedlings per pot (factor variable with single plant per pot as standard level), H: height of the plant (continuous variable). Significant results are in bold: <sup>∗</sup>P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001.

plants, depending on the provenance and on the number of seedlings per pot. To calculate the probabilities, the mean relative weight loss of the pots for the control and stressed group of plants was applied. doy: day of the year.

#### Impact of Provenance

Survival of seedlings after severe water deficit appeared independent of the provenance. Still, a higher number of plants of KLA and BOR had died off compared to VOE (**Table 2**), because the model accounted for the height of the plants with smaller plants having a higher chance of survival. BOR was characterized by a higher mean height of plants and KLA did not show a higher mean height but a higher standard deviation compared to VOE, implying a relative larger number of high plants being more prone to starvation by drought (**Figure 8**). KLA is believed to be local of origin (Vander Mijnsbrugge et al., 2005). A cpDNA analysis revealed a uniform haplotype that fits in the reconstructed postglacial migration routes (Vander

Mijnsbrugge et al., 2003). In addition, the mother trees are abandoned remnants of coppice wood in former heath land where no tradition existed among the relatively poor farmers of introducing foreign provenances. In a provenance trial, located in Belgium, KLA flushes later compared to several commercial provenances from western Germany (data not shown) indicating an adaptation to the less predictable weather (maritime climate) caused by oceanic influences with a smaller contrast between summer and winter and less predictable transitions. Other Belgian commercial provenances in this trial are typical planted forest stands, comparable to VOE and BOR, displaying an earlier flushing in spring, likely indicating a non-local origin of the mother trees. Similarly, in the here presented experiment leaves senesce earlier in KLA compared to VOE and BOR and, in addition, buds flush later, most probably indicating a shorter growing season as an adaptation to a more unpredictable temperate maritime climate for KLA and a non-local origin for VOE and BOR. The latter is likely as forest history in the northern part of Belgium is characterized by successive phases of deforestation and afforestation in this densely populated region (Vandekerkhove et al., 2011) resulting in a scattered and fragmented landscape, and in mostly relatively small oak forests of which a vast majority has a plantation origin, largely with an unknown origin of the planting stock. German origins of planting stock may have been planted as war reparations after the two world wars. The result stresses the importance of local provenances such as KLA not only because of the local adaptation to current climate as expressed by the phenological responses, but also because of putative larger variability in quantitative traits, as is the case for height growth in KLA. The latter probably also indicates for the planted stand VOE a stronger kinship and lesser diversity among the original planting stock (it may have been sourced on a limited amount of mother trees). As BOR was less sampled (only three mother trees) it is more difficult to make assumptions in this respect. The small variability in the quantitative trait height in the putative nonlocal provenance VOE can be related to the findings of Vranckx

et al. (2014b). Based on a molecular genetic analysis (SSR), a stronger relatedness among the pedigrees of typical planted oak stands in the northern part of Belgium compared to the kinship among the mother trees was shown, which was attributed to forest fragmentation, which negatively influences genetic diversity (Vranckx et al., 2012), and is possibly strengthened by the plantation history of the mother trees (Vranckx et al., 2014a).

## Phenological Responses to Water Deficit Followed by Re-watering

In general, leaf senescence follows a highly complex genetic program that is tightly controlled by multiple layers of regulation at the level of transcription, post-transcription, translation and post-translation (Woo et al., 2013). Based on reported results of phenological responses of oak seedlings upon drought treatment (Jensen and Hansen, 2010; Spiess et al., 2012; Kuster et al., 2014), an acceleration of the onset of leaf senescence was expected due to the drought treatment in our experiment, but we observed a clear delay. This can be ascribed to the plentiful re-watering after the drought treatment. It is assumed that enhanced plant growth upon re-watering after a drought period is a compensation for the loss of net primary production due to the preceding water deficit and has been observed in herbaceous plants such as perennial grass (Xu et al., 2009) and in woody species such as oak (Spiess et al., 2012; Turcsán et al., 2016). In perennial grass, the magnitude of this pre-drought stimulation of relative growth rate was found relative to the severity of the drought (Xu et al., 2009). The observed delayed senescence in our experiment can be explained as a compensation time in which plants did not produce an extra shoot but may have taken time to recover before entering the next developmental phenophase of leaf senescence. This recovery may include restoration of the antioxidant metabolism. Similar to other environmental stresses, drought induces a rapid accumulation of reactive oxygen species (ROS), which is reversed upon rewatering (Sofo et al., 2005; Xu et al., 2010). The ROS scavenging mechanisms induced during the recovery phase may interfere with the developmentally regulated autumnal leaf senescence, which includes the degradation of cellular structures in an orderly manner, programmed cell death and the generation of ROS (Woo et al., 2013). Additionally, abscisic acid (ABA) may play a role in the delayed leaf senescence. ABA is a stress hormone wellknown to signal drought in plants showing elevated levels during drought periods which are lowered again upon re-watering (Chaves et al., 2003; Xu et al., 2010). As ABA mediates the developmentally regulated autumnal leaf senescence (Woo et al., 2013), drought induced ABA signals may similarly interfere with this process causing the observed delay. This observed compensation time is diminished for seedlings sharing a pot compared to single plants, indicating that a putative competition for resources attenuated the drought and re-watering induced delay in onset of leaf senescence. The higher drought stress experienced by double plants per pot may have weakened their ability of full recover by reducing physiological resilience and may thus have shortened the recovery period and allowed an earlier onset of leaf senescence. It has been shown that severe drought can cause photodamage in Mediterranean tree seedlings that in turn may impair physiological repair and recovery upon re-watering (Benigno et al., 2014). We observed bud burst as the succeeding phenophase following leaf senescence. It has already been shown that a phenological shift in perennials may show an after effect in the subsequent phenophases (Fu et al., 2014). Similarly, we observed a slight delay in bud burst among the drought treated group of plants in the subsequent spring. In this phenological model, the number of plants per pot was not significant, and plant height was significant but independent of the treatment (interaction term with relative weight loss was not significant). Together, our results emphasize the capacity of oaks to resume growth after periods of water deficit. Still, a delayed senescence in autumn will concur with a delayed autumnal hardening and may render shoots vulnerable to early frosts. In addition, a delayed flushing may diminish the competitive capacity of seedlings in relation to competing plants.

## CONCLUSION

Last decades a strong tendency emerged in the northern part of Belgium to promote natural rejuvenation in a sustainable and nature-oriented forest management. Our results demonstrate that provenance may play an important role in forest regeneration, in the face of the predicted climate change. Although, we found survival rate after a severe water deficit being independent of the provenance, still it was dependent on plant height. A large variability among the seedlings can therefore be advantageous, with smaller plants surviving harsh summer conditions and higher plants profiting from a competitive gain in normal years. Commercial planting stock used in (re)forestations is traditionally graded (Vander Mijnsbrugge et al., 2010), reducing variability and highlighting the value of natural rejuvenation with an option of additional stocking especially in oak stands with a plantation history of unknown origin.

## AUTHOR CONTRIBUTIONS

KVM, AT, SM, KS, and MS concepted and supervised the here presented experiment, including the organization of the seed collection, while AT, KVM, JM, and ND conducted the measurements and observations on the seedlings. KVM, AT, JM, and ND performed the statistical analysis: AT was responsible for the weight loss and height measurements, AT, JM, and ND for the leaf senescence observations and KV for the bud burst observations. All authors (KVM, AT, JM, NL, SM, MS, KS) contributed substantially to the manuscript preparation.

## ACKNOWLEDGMENTS

In the first place, we like to thank Stefaan Moreels for the elaborate and meticulous nursery work and indispensable help during the measurements and observations. We also thank Hans

Beeckman for his help during the manuscript preparation. We thank the Agency for Nature and Forest for admitting and facilitating the seed collections. We also thank the COST FP1106

#### REFERENCES


network Studying Tree Responses to Extreme Events: A Synthesis (STReESS) for the short-term scientific mission granted to AT, and the ERASMUS exchange program for funding.



Xu, Z., Zhou, G., and Shimizu, H. (2010). Plant responses to drought and rewatering. Plant Signal. Behav. 5, 649–654. doi: 10.4161/psb.5.6.11398

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Vander Mijnsbrugge, Turcsán, Maes, Duchêne, Meeus, Steppe and Steenackers. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Early Summer Drought Stress During the First Growing Year Stimulates Extra Shoot Growth in Oak Seedlings (Quercus petraea)

Arion Turcsán1,2,3 \*, Kathy Steppe<sup>4</sup> , Edit Sárközi<sup>5</sup> , Éva Erdélyi<sup>6</sup> , Marc Missoorten<sup>7</sup> , Ghislain Mees<sup>7</sup> and Kristine V. Mijnsbrugge<sup>1</sup>

<sup>1</sup> Research Institute for Nature and Forest, Geraardsbergen, Belgium, <sup>2</sup> Department of Biometrics and Agricultural Informatics, Corvinus University of Budapest, Budapest, Hungary, <sup>3</sup> Department of Forest Reproductive Material and Plantation Management, Institute of Silviculture and Forest Protection, University of West Hungary, Sopron, Hungary, <sup>4</sup> Laboratory of Plant Ecology, Department of Applied Ecology and Environmental Biology, Ghent University, Ghent, Belgium, <sup>5</sup> Department of Soil Science and Water Management, Corvinus University of Budapest, Budapest, Hungary, <sup>6</sup> College of Commerce, Catering and Tourism, Budapest Business School, Budapest, Hungary, <sup>7</sup> Agency for Nature and Forest, Brussels, Belgium

#### Edited by:

Achim Braeuning, University of Erlangen-Nuremberg, Germany

#### Reviewed by:

Jürgen Kreuzwieser, University of Freiburg, Germany Zhenzhu Xu, Institute of Botany, Chinese Academy of Sciences, China

> \*Correspondence: Arion Turcsán raup25@gmail.com

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 28 October 2015 Accepted: 04 February 2016 Published: 23 February 2016

#### Citation:

Turcsán A, Steppe K, Sárközi E, Erdélyi É, Missoorten M, Mees G and Mijnsbrugge KV (2016) Early Summer Drought Stress During the First Growing Year Stimulates Extra Shoot Growth in Oak Seedlings (Quercus petraea). Front. Plant Sci. 7:193. doi: 10.3389/fpls.2016.00193 More severe summer droughts are predicted for mid-latitudes in Europe. To evaluate the impact on forest ecosystems and more specifically on forest regeneration, we studied the response to summer drought in oak seedlings (Quercus petraea). Acorns were collected from different mother trees in three stands in Belgium, sown in pots and grown in non-heated greenhouse conditions. We imposed drought on the seedlings in early summer by first watering the pots to saturation and then stopping any watering. Weight of the pots and stomatal conductance were regularly measured. Re-watering followed this drought period of 5 weeks. Height of the seedlings and apical bud development were observed. Stomatal resistance increased toward the end of the experiment in the drought-treated group and was restored after re-watering. The seedlings from the drought treatment displayed a higher probability to produce additional shoot growth after re-watering (p ≤ 0.05). A higher competition for water (two plants per pot) increased this chance. Although this chance was also higher for smaller seedlings, the actual length of the extra growth after re-watering was higher for larger seedlings (p ≤ 0.01). Both in the drought-treated and in the control group the autochthonous provenance growing on a xeric site produced less extra shoots compared to the two other provenances. Finally, stressed plants showed less developed apical buds compared to the control group after re-watering, suggesting a phenological effect on the growth cycle of oaks (p ≤ 0.0001). The higher chance for an extra shoot growth after the drought period can be considered as a compensation for the induced growth arrest during the drought period.

Keywords: drought, oak seedling, apical bud, shoot growth, re-watering, general linear mixed models

## INTRODUCTION

fpls-07-00193 February 19, 2016 Time: 17:50 # 2

Several periods of dry, wet, cold, or warm climate were recorded during the past centuries in Europe (Masson-Delmotte et al., 2005). Furthermore, the number of precipitation anomalies increased during the past century (Zhang et al., 2007). The predicted climate change indicates even more extreme weather events such as longer dry periods, swifts in precipitation and rain intensity. Forest vitality will be challenged by such changes and forests will become more vulnerable not only in Europe (Lindner et al., 2010), but all over the globe (Choat et al., 2012). For Europe, climate projections predict increasing temperatures and irregular precipitation patterns in summer, augmenting the number and the intensity of drought events (IPCC, 2013). For Belgium, drier conditions due to climate change are predicted for the end of the century especially in summer periods (Baguis et al., 2010).

A large part of the forests in the lower countries grows on sandy soils which are characterized by a relatively low water holding capacity, making them especially vulnerable to extreme drought events during the growing season (Van der Werf et al., 2007). Soil water shortage results in leaf stomatal closure, which limits leaf cooling and is quickly followed by leaf damage (Bréda et al., 2006). Subsequently, drought stress disrupts water transfer in the xylem tissue through cavitation of vessels resulting in dying off of twigs and roots (Barigah et al., 2013). Stomatal closure also hampers carbon assimilation in the tree (McDowell, 2011). These physiological effects of drought stress restrict in the first place biomass production, and additionally increase susceptibility and vulnerability toward secondary stresses such as frost or fungal and insect attacks (Bréda et al., 2006; Lindner et al., 2010).

Sessile oak (Quercus petraea), pedunculate oak (Quercus robur) and beech (Fagus sylvatica) take great stake in European forest ecosystems (Mátyás, 1996). Oak species have a lower competition capacity compared to beech under natural conditions, but they adapt better to poor weather or soil conditions (Thomas and Gausling, 2000) and are well known for being relatively drought tolerant, because of their deep roots system and effective water transport (Landolt et al., 2010; Kuster et al., 2011). During the past centuries dry conditions in early summer restricted the growth of beech and oak trees (Pilcher and Gray, 1982; Dittmar et al., 2003; Lebourgeois et al., 2005), although oak species show a comparably lower sensitivity to drought at the more humid sites in Europe (Scharnweber et al., 2011). Stem diameter growth of adult beech trees is stronger reduced by dry summer periods compared to adult oak trees (Leuschner et al., 2001). In the Netherlands, oak and beech display similar growth patterns, suggesting similar influential environmental factors (Van der Werf et al., 2007). A relative strong correlation between intra-annual growth pattern and precipitation in this region indicates the importance of the latter for proper growth in both species (Van der Werf et al., 2007). Arend et al. (2011) pointed to the provenance-specific growth responses to drought of Quercus sp. with shoot height growth being more negatively affected by drought in northern provenances compared to more southern provenances within Switzerland.

Quercus species are characterized by a cyclic growth. The buds flush several times within the growing season, each time followed by distinctive rest periods, with a trade-off between root and shoot growth (Reich et al., 1980; Harmer, 1990). The auxin/cytokinin ratio plays an important role in the induction of bud dormancy and bud burst (Cline and Harrington, 2006; Su et al., 2011; Vanstraelen and Benková, 2012). Under uniform growing conditions buds flush, and thus shoots and leaves grow, and subsequent rest periods occur very synchronous among oak seedlings (Reich et al., 1980). Also, oak seedlings commonly display multiple flushing in spring when environmental conditions are not limiting (Reich et al., 1980).

Broadmeadow and Jackson (2000) observed an overall reduction in plant biomass induced by drought stress, but at the same time the root-shoot relation shifted with the oak (Q. petraea) saplings producing more roots compared to shoots under drought conditions. Furthermore, increase of fine root biomass in oak was found under water deficit, attempting to reach water supplies at lower soil layers (Thomas and Gausling, 2000). However, biomass (leaves and shoots) decreased in the long term under repeated drought stress (Broadmeadow and Jackson, 2000; Thomas, 2000) with diminished shoot growth being less severe in sessile oak (Q. petraea) compared to pedunculate oak (Q. robur; Fonti et al., 2013). Saplings from an oak provenance growing at a drier site showed higher biomass loss at water limiting conditions compared to a provenance from a more humid site, questioning the suitability of xeric provenances in mitigating predicted climate change (Kuster et al., 2013). Spieß et al. (2012) observed "compensated growth" induced by re-watering after one or two drought periods within the growing season. In their study, Spieß et al. (2012) reduced soil water conditions by 20– 25% compared to the control groups, and observed reduced shoot growth during the drought treatment on 2–3 years old oak saplings from one genotype and a fourth shoot on some of the saplings after re-watering.

In this study, we investigated the immediate effects of a short drought stress in early summer on potted, 1-year old oak (Q. petraea) seedlings from three different provenances. As oak is characterized by a cyclic growth pattern, we focused on shoot growth and bud development. Competition for water between plants generally occurs when availability is reduced (Craine and Dybzinski, 2013). Intraspecific competition has impact on seedling performance and growing pattern (Shainsky and Radosevich, 1992). Seedlings were single or double in the pots in our experiment, adding an extra competition effect for the latter.

#### MATERIALS AND METHODS

#### Source Material

Acorns were collected per mother tree at three different locations in Flanders (northern part of Belgium): Klaverberg (KLA, 51◦ 0 0 57.855600N 5◦ 310 57.038400E), Voeren (VOE, 50◦ 450 31.561200N 5◦ 450 39.934800E) and Borgloon (BOR, 50◦ 480 22.068000N 5◦ 200 34.87200E) at the end of October 2013. The three provenances differ in soil type and/or in stand history.

Klaverberg is a small relict of oak coppice wood growing on inland sand dunes within a former heath land. The oaks here are most probably local of origin. The stand is characterized by a large structure diversity. Acorns were collected from the visually older coppice stools. As the oaks are mostly growing widely spaced, the chance on mixture of acorns from different mother trees was negligible. Voeren is a classical planted forest stand, even aged and approximately 80 years old, growing on a loamy soil type. The origin of the planted material is unknown. Borgloon is also a planted forest and approximately 100 years old. The forest stand grows on sandy soil and the origin of the planted material is also unknown. Acorns were collected underneath 14 mother trees from Voeren, 13 mother trees from Klaverberg, and three mother trees from Borgloon, which all showed a well-developed crown (dominant trees). Collection was performed only close to the stem, minimizing the chance on mixing acorns between different mother trees. Before sowing, a water-swimming test was used to assess the vitality of the seeds and the unhealthy seeds were removed. The seeds were further visually controlled. The checks resulted in 664 acorns from Klaverberg, 744 acorns from Voeren, and 154 from Borgloon.

#### Experimental Design

In November 2013, the collected seeds were sown in forestry trays (two seeds per cell) using standard nursery potting soil. During winter, the trays were watered manually keeping the soil moist. The experiment was located in a greenhouse with automatic temperature regulation, keeping the greenhouse frost-free in wintertime, but without additional heating. The germinated seedlings were transferred in April 2014 to 1-l pots (12 cm × 11 cm × 11 cm) using standard nursery potting soil (Organic matter concentration 20%, pH 5.0–6.5, Electrical Conductivity (EC): 450 µS/cm, dry matter 25%, Fertilization: 1.5 kg/m<sup>3</sup> powdered compound fertilizer NPK 12 + 14 + 24). Non-germinated seeds were removed, whereas double plants in one tray cell were kept together. The seedlings were kept without any additional fertilization during the experiment.

We choose the usage of seedlings in pots in our experiment, rather than working in a field experiment outdoor, as this allows to impose a drought period on a subset of plants while both treated and control plants can be subjected to very similar other growth conditions (light, temperature, nutrient availability). Furthermore, it allows monitoring indirectly the reduction in water availability by weighing of the pots. The set-up consisted of two main groups: the control group with 148 pots (66 single seedling and 164 double seedling per pot) with seedlings from KLA, 116 pots (52 single seedling and 128 double seedling per pot) from VOE and 37 pots (12 single seedling and 50 double seedling per pot) with seedlings from BOR, and the droughttreated group with 137 pots (57 single seedling and 190 double seedling per pot) containing seedlings from KLA, 124 (66 single seedling and 116 double seedling per pot) from VOE and 43 (10 single seedling and 66 double seedling per pot) from BOR. All germinating plants were given water at regular times according to the visual needs of the pots as judged by experienced greenhouse workers. In both groups, the three provenances were individually mingled at random (completely randomized).

On May 15, 2014 the two groups of plants were soaked overnight in a water basin with a water level up to two cm at the bottom of the pots. In this way all the pots were fully saturated with water. Up to July 1, 2014 the drought-treated group was not watered anymore, whereas the control group was further watered according to the visual needs of the plants. All plants were rewatered on July 2, 2014 by soaking the two groups of plants in the same water basin in the same way. After this, both groups were kept in well-watered conditions according to the visual needs of the plants.

#### Measurements and Scoring

During the drought treatment all pots were weighed every week to measure the water loss following the first water saturation treatment. As a proxy for the level of drought stress, the weight loss of the individual pots at the end of the treatment period was calculated relative to the initial weight at fully saturated condition. In the statistical models, drought stress was expressed as the weight loss of the last weighing of the individual pots at the end of the treatment, just before re-watering, divided by the weight of the individual pots at full saturation at the beginning of the treatment.

Thirty pots with relative high plants were randomly chosen from the control group as well as 30 from the drought-treated group to monitor the treatment effect. Leaf stomatal aperture in terms of leaf resistance to water vapor was measured weekly with a diffusion porometer (Model AP4, Delta-T Devices, Burwell, Cambridge, UK) during the entire drought period. As stomata are sensitive to drought stress, high resistance values represent a closing reaction (Schulze et al., 1972), and declining stomatal conductance and leaf assimilation rate (Farquhar and Sharkey, 1982). The porometer measurements were conducted during daytime between 10 a.m. and 3 p.m.

The height of the seedlings was measured with a ruler at the end of the drought treatment and at full recovery of the plants (on September 4, 2014).

The apical bud of the highest plant per pot was scored on August 28, 2014, following a binary scoring system with buds well developed and colored brown, as opposed to any other stage of bud development (bursting, absent or small and green).

## Data Analysis

The open source software R 3.1.2 (R development Core Team, Vienna, Austria) was used for all statistical analyses. Three response variables were modeled using (generalized) linear mixed models.

As a larger part of the plants did not show any height growth between July 1, 2014 and September 4, 2014, a first binary response variable was deduced from the height data indicating no growth or growth between the two time points. From all plants that had grown between the two height measurements, the height increment was calculated as a second continuous variable. Finally the apical bud score was the third binary response variable. The first and third response variables were modeled using logistic regression (generalized linear mixed models) in the package lme4 (Bates et al., 2014), whereas the continuous response variable was examined using linear regression (linear mixed model) in


TABLE 1 | Number of seedlings used in the models for extra shoot growth, height increment and bud development.

The type of response variable is indicated between brackets. CON: control group, STR: drought-treated group, Sp: number of seedlings per pot. The provenances are VOE, BOR, and KLA. The bolded values are the sum values of each line per model.

the package nlme (Pinheiro et al., 2011). In all three models, the same covariates were checked for significant explanatory power: the height immediately after the drought treatment (continuous variable), the provenance of the seedlings (factor variable) and the number of seedlings per pot (factor variable). All three covariates were first included in each model with an interaction term with weight loss of the pots relative to the fully water saturated condition. Using drop 1, a likelihood ratio test (and a maximum likelihood estimation for the linear model of the continuous response variable height increment), the fixed part of the models was reduced up to only significant terms. In all models the mother plant from which acorns were collected was in the random part (random intercept). In addition, the linear model of height increment showed a better fit using a log transformation of the response variable. The predict command in lme4 and nmle was applied for drawing the regression curves. The number of seedlings used in the calculation are indicated in **Table 1**.

## RESULTS

## Stress Symptoms

In general, oak seedlings respond efficiently to drought stress by closing stomata, allowing the leaf water potential to remain above a critical threshold value at which cavitation damage occurs (Vivin et al., 1993; Cochard et al., 1996). To monitor the stress symptoms of the seedlings, stomatal resistance was measured during the drought period (**Figure 1**) in a sample of both the control and the drought-treated group of seedlings. The average value of the drought-treated group strongly increased from June 10, 2014 onward indicating a response to drought. At the same time, the average weight loss of the pots increased (**Figure 1**) confirming the drying process. Among the treated seedlings a small group of plants (23%) showed visual "wilting or curling of leaves" compared to the control group.

## Height Growth

Boxplots of the first height measurement and of the height increment between the first and the second measurement are shown in **Figures 2** and **3**, respectively. The binary response variable indicating whether or not a seedling showed an extra height growth after re-watering, was modeled using generalized linear mixed models. Significant influencing factors were the provenance (no significant interaction term with weight loss), the number of seedlings per pot, depending on the amount of weight loss (significant interaction term) and the height of the seedlings at the end of the drought treatment, also depending on the amount of weight loss (**Table 2**). Both in stressed and control conditions the provenance VOE produced more extra shoot growth compared to KLA, with BOR in an intermediate position (**Table 2**, **Figure 4**). Especially seedlings that shared a pot displayed a higher probability for extra shoot growth in more stressed conditions (**Table 2**, **Figure 5**).

In addition, the higher the height at the end of the drought treatment, the lower the probability on extra shoot growth (**Table 2**, **Figure 4**).

For the seedlings showing extra shoot growth after rewatering, the actual length of the height increment was modeled using linear mixed models (**Table 3**, **Figure 6**.). The length of the extra growth was found to be only dependent on the plant height

TABLE 2 | Estimated coefficients for the fixed part of the logistic regression models with binary response variables extra shoot (Model I) and apical bud development (Model III).


The provenances VOE and BOR are compared to the standard provenance KLA. H: initial height of the plants before the start of the drought treatment; Wl: weight loss of plant pots relative to fully water saturated condition; Sp: number of seedlings per pot. Significance codes: ∗∗∗P ≤ 0.0001, ∗∗0.0001 < P ≤ 0.01, <sup>∗</sup>0.01 < P ≤ 0.05.

at the end of the drought treatment, with higher plants producing a larger extra shoot.

### Bud Development

Presence or absence of a fully developed apical bud in the fully recovered seedlings after the drought stress was monitored and the binary variable was modeled using generalized linear mixed models (**Table 2**, **Figure 7**). The drought-treated seedlings significantly showed less well-developed apical buds in higher plants (significant interaction term between initial height and relative weight loss). Both in droughttreated and control seedlings (no significant interaction term), VOE had a lower probability on a fully developed apical bud on the measurement day compared to the other provenances.

FIGURE 5 | Modeled probability of extra shoot growth after drought stress, depending on the relative amount of weight loss during the drought treatment (relative to the fully water saturated condition), the provenance of the oak seedlings and the number of seedlings per pot. Modeled probabilities are based on an average height of 10 cm at the end of the drought period.

## DISCUSSION

We studied the effect of a late spring/early summer drought stress on sessile oak (Q. petraea) seedlings originating from Belgium during the first growing year. In our experiment, potted oak seedlings responded with a higher probability on extra shoot formation after re-watering, especially for those seedlings that experienced direct competition for water (two plants per pot). Plant competition for water is less studied and not well understood compared to light and nutrient sources (Craine and Dybzinski, 2013). At the same time, the higher seedlings in the drought-treated group showed less well developed apical buds compared to the control group. Oak seedlings seem to translate the drought signal in an increased chance on extra shoot formation after re-watering. This reaction can be interpreted as a tendency to compensate fairly quickly for a retarded growth

TABLE 3 | Estimated coefficients for the fixed part of the linear model of height increment after the drought treatment.


H, height at the end of the drought treatment. Significance codes: ∗∗∗P ≤ 0.0001, ∗∗0.0001 < P ≤ 0.01.

during the stress period, resulting in our experiment in the surprising and contra-intuitive higher average height increment in the drought-treated group of plants. Although only seedling height was measured, this result seems to contradict many descriptions of diminished biomass production of oaks after recovery from drought stress (Broadmeadow and Jackson, 2000; Thomas, 2000; Thomas and Gausling, 2000; Arend et al., 2011; Kuster, 2012), but it is in line with drought stress responses that led to the developments of extra shoots in oak (Spieß et al., 2012) and Douglas-fir (Kaya et al., 1994) after re-watering. Growth in Douglas-fir is characterized by a similar cyclic growth pattern as in oak (Kaya et al., 1994). Kaya et al. (1994) describe extra shoot production of Douglas-fir seedlings caused by re-watering after two drought periods, which took place 4 weeks long during the first growing season and 8 weeks long during the second growing season. Reich et al. (1980) describes conditions that favor multiple flushing of oak within one growing season. The observation that seedlings show more flushing per growing season compared to adult trees is related to the fact that an additional flush in adult trees would imply a relatively higher photosynthetic demand compared to seedlings. On the other hand, a higher root biomass in comparison to aboveground tissues, which has been studied in resprouting of coppiced stems, may allow a higher number of flushes within one growing season compared to seedlings. Also defoliation within the growing season has been reported to lead to extra flushing in oaks (Borchert, 1975). Even though the oak seedlings in our experiment only showed relatively minor leaf damage at the end of the drought treatment, and the majority of leaves fully recovered after re-watering, a stress signal indicating potential leaf loss may have triggered an extra flushing in a significantly higher number of seedlings compared to the control group, resulting in a larger mean height growth as side effect. Also, the extra shoot growth after re-watering can be considered as a compensation for the growth arrest experienced during the drought period (Spieß et al., 2012).

Our observed responses in primary height growth are contradictory to the observations of stress responses in secondary growth of adult oak trees as expressed in annual radial wood formation, which is dependent on cambial activity. Van der Werf et al. (2007) reported a retarded cambial activity during drought stress, which was not resumed after normalization of the conditions in the same growing season, which underpins the fundamental differences in regulation of primary and secondary growth in trees.

Two opposite response types of drought stress can therefore be suggested depending on the severity of the stress. Mild drought stress (no effect on stomatal conductance) resulted in less second

FIGURE 6 | Fitted height increment of the extra shoots grown after the drought period.

flushes of Q. petraea (Thomas and Gausling, 2000) whereas our results indicate that short and more severe drought stress in early summer augmented the probability on a higher number of flushes, which is in line with the findings of Kaya et al. (1994).

In our experiment, the length of the extra shoot growth showed no difference between control and drought-treated plants, but was found to be larger for higher plants compared to smaller plants. Although the drought signal triggers a higher chance on extra shoot formation after re-watering, it does not influence shoot length as such. As the auxin/cytokinin ratio is one of the main regulators of bud dormancy and bud burst during the year (Cline and Harrington, 2006; Leyser, 2009; Müller and Leyser, 2011; Su et al., 2011; Vanstraelen and Benková, 2012),

it can be hypothesized that drought stress signals act along this pathway.

In our experiment, drought-treated plants showed less fully developed and thus dormant apical buds among the higher seedlings 9 weeks after finishing the drought treatment compared to the control group, suggesting a height-dependent retardation in bud development. This effect seems to contrast with the higher chance on an extra shoot growth mainly among smaller plants, suggesting deviating signal pathways. Although independent of the level of drought stress, we observed significant differences between the studied provenances for extra shoot production and bud development. Compared to KLA seedlings, VOE and to a lesser extent BOR produced on average more shoots during our experiment independent from the treatment. More extra shoot growth later in the growing season may lead to a shortened hardening process in autumn, which increases the sensitivity to early frosts. Furthermore, compared to KLA, VOE seedlings showed less developed apical buds 9 weeks after re-watering, which may additionally increase the vulnerability of the seedlings during autumn and winter (Svejgaard Jensen and Douglas Deans, 2004). KLA seedlings represent a likely local provenance in the study region. Less shoot growth later in the growing season and quicker apical bud dormancy indicate a better adaptation to local climate, minimizing risks on frost damage both in well-watered and dry conditions.

#### CONCLUSION

After an early summer drought event re-watering augmented significantly the probability to form an extra shoot in oak seedlings, with the highest probability in the smaller individuals. Simultaneously, drought retarded the apical bud development significantly in larger seedlings. Competition for water experienced by seedlings that grow together in a pot

#### REFERENCES


further increased the chance of extra shoot formation. As the number of extreme weather events will increase in the future due to climate change, it is important to study the behavior of seedlings subjected to more severe drought stress, because such experiments will greatly assist us in understanding the impact of drought on forest regeneration.

#### AUTHOR CONTRIBUTIONS

AT conducted the experiment with KM and wrote the introduction and parts of the discussion and conclusion. KM was the main supervisor during the project AT, KM, MM, and GM conducted the statistical analyses, created the graphs, and integrated them in the article. MM worked on graph 2. GM worked on graph 3. KS established the experimental design, monitored the observations, reviewed the manuscript, and wrote parts of the discussion. ES and ÉE worked on the seedling evaluation systems, filtered the data, and evaluated it. ES and ÉE wrote the material and methods parts. All authors participated in the finalization of the article.

## ACKNOWLEDGMENTS

We thank Stefaan Moreels and Kim van Neck, affiliated to the Research Institute for Nature and Forest, for seed collections, plant growth and seedlings management, and Michiel Hubeau from the Laboratory of Plant Ecology, Ghent University for the help with the porometer measurements. We also thank the COST FP1106 network Studying Tree Responses to Extreme Events: A Synthesis (STReESS) for the short-term scientific mission granted to AT to visit the lab of KS, and the ERASMUS exchange program for funding.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Turcsán, Steppe, Sárközi, Erdélyi, Missoorten, Mees and Mijnsbrugge. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Variation in Ecophysiological Traits and Drought Tolerance of Beech (Fagus sylvatica L.) Seedlings from Different Populations

Claudia Cocozza<sup>1</sup> \*, Marina de Miguel <sup>2</sup> , Eva Pšidová<sup>3</sup> , L'ubica Ditmarová<sup>3</sup> , Stefano Marino<sup>4</sup> , Lucia Maiuro<sup>4</sup> , Arturo Alvino<sup>4</sup> , Tomasz Czajkowski <sup>5</sup> , Andreas Bolte<sup>5</sup> and Roberto Tognetti 6, 7

1 Istituto per la Protezione Sostenibile delle Piante, Consiglio Nazionale delle Ricerche, Sesto Fiorentino, Italy, <sup>2</sup> BIOGECO, INRA, Univ. Bordeaux, Cestas, France , <sup>3</sup> Institute of Forest Ecology, Slovak Academy of Science, Zvolen, Slovak Republic, <sup>4</sup> Dipartimento Agricoltura, Ambiente e Alimenti, Università degli Studi del Molise, Campobasso, Italy, <sup>5</sup> Johann Heinrich von Thünen Institute, Institute of Forest Ecosystems, Eberswalde, Germany, <sup>6</sup> Dipartimento di Bioscienze e Territorio, Università degli Studi del Molise, Pesche, Italy, <sup>7</sup> The EFI Project Centre on Mountain Forests (MOUNTFOR), Edmund Mach Foundation, San Michele all'Adige, Italy

#### Edited by:

Boris Rewald, University of Natural Resources and Life Sciences, Vienna (BOKU), Austria

#### Reviewed by:

Tobias Gebauer, Albert Ludwig University of Freiburg, Germany Florian Netzer, University of Freiburg, Germany Aikaterini Dounavi, Forest Research Institute of Baden-Wuerttemberg, Germany

> \*Correspondence: Claudia Cocozza claudia.cocozza@ipsp.cnr.it

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 28 January 2016 Accepted: 06 June 2016 Published: 22 June 2016

#### Citation:

Cocozza C, de Miguel M, Pšidová E, Ditmarová L, Marino S, Maiuro L, Alvino A, Czajkowski T, Bolte A and Tognetti R (2016) Variation in Ecophysiological Traits and Drought Tolerance of Beech (Fagus sylvatica L.) Seedlings from Different Populations. Front. Plant Sci. 7:886. doi: 10.3389/fpls.2016.00886 Frequency and intensity of heat waves and drought events are expected to increase in Europe due to climate change. European beech (Fagus sylvatica L.) is one of the most important native tree species in Europe. Beech populations originating throughout its native range were selected for common-garden experiments with the aim to determine whether there are functional variations in drought stress responses among different populations. One-year old seedlings from four to seven beech populations were grown and drought-treated in a greenhouse, replicating the experiment at two contrasting sites, in Italy (Mediterranean mountains) and Germany (Central Europe). Experimental findings indicated that: (1) drought (water stress) mainly affected gas exchange describing a critical threshold of drought response between 30 and 26% SWA for photosynthetic rate and Ci/Ca, respectively; (2) the C<sup>i</sup> to C<sup>a</sup> ratio increased substantially with severe water stress suggesting a stable instantaneous water use efficiency and an efficient regulation capacity of water balance achieved by a tight stomatal control; (3) there was a different response to water stress among the considered beech populations, differently combining traits, although there was not a well-defined variability in drought tolerance. A combined analysis of functional and structural traits for detecting stress signals in beech seedlings is suggested to assess plant performance under limiting moisture conditions and, consequently, to estimate evolutionary potential of beech under a changing environmental scenario.

Keywords: water stress, ecophysiology, gas exchange, chlorophyll a fluorescence, European beech

## INTRODUCTION

Warming-induced drought is threatening forest ecosystems worldwide, increasing water stress and mortality risk for trees (Allen et al., 2010). The vulnerability of plants to drought varies in dependence of stress severity, its duration, and the combination with other stresses (Niinemets, 2010). Intraspecific variation of tree response to drought, recently, has received increasing attention in the case of important forest species, such as Fagus sylvatica L. (beech; e.g., Borghetti et al., 1993; Tognetti et al., 1995; García-Plazaola and Becerril, 2000; Peuke et al., 2002; Aranda et al., 2015; Knutzen et al., 2015; Pšidová et al., 2015), in order to inform forest managers on adaptive capacities of populations for stress tolerance and decrease tree vulnerability to climate change.

Acclimation of trees to water deficit is the result of adaptive changes in plant development and ecophysiological processes, such as gas exchange, growth rate, and water relations (Sala et al., 2010). Drought-induced hydraulic limitation on carbohydrate use may prolong survival in plants under stress. However, if drought persists, reduced photosynthetic carbon assimilation due to stomatal closure (isohydric behavior) may promote carbon starvation, as carbohydrate demand continues for maintenance of osmoregulation, and plants fail to maintain hydraulic integrity (McDowell, 2011). If plants maintain their stomata open during drought (anisohydric behavior), hydraulic failure may occur, thus leading to mortality. Tree mortality may occur when drought has caused >50% loss of stem hydraulic conductivity, corresponding to −4.5 MPa in beech (Barigah et al., 2013). The capacity for adaptive changes to the environment may ultimately be critical in determining tree species survival under climate change (Aitken et al., 2008). Physiological responses, including adaptation and evolution to environmental changes, define phenotypic plasticity that can be assumed as the dominant underlying process with consequences on ecosystem functions (Hovenden and Vander Schoor, 2003; Thomas, 2011). A better understanding of geographic pattern and genetic variation in functional and structural traits of important tree species is essential for implementing adaptive forest management strategies to mitigate anticipated impacts of climate change on plant growth and drought tolerance.

Beech is a naturally dominant tree species in many European forests and sensitive to water deficit (Tognetti et al., 1995; Backes and Leuschner, 2000; Czajkowski et al., 2005; Bolte et al., 2007; Rose et al., 2009). The distribution of beech in Europe is characterized by high genetic diversity, resulting in high potential to adapt to changing environmental conditions (Dounavi et al., 2016). Acclimation to drought and heat stress in beech may occur after increasing levels of proline amino acid that plays as osmo-protectants to raise the osmotic pressure and thus maintain membrane integrity and stabilize proteins (Rennenberg et al., 2006). Beech can also respond to water stress through decrease in photosynthetic efficiency and light sensitivity of the photosynthetic apparatus (Tognetti et al., 1995, 1997; Peuke et al., 2002). In southern Europe, the recent decline in basal area increment of beech has been linked to decreasing water availability (Jump et al., 2006; Piovesan et al., 2008), which can affect carbon dynamics and sequestration potentials at the southern limit of this species distribution (Tognetti et al., 2014). However, this is not a general response and positive growth (tree-ring width) in beech at Mediterranean latitudes has been observed (Tegel et al., 2013). In central Europe, the extreme 2003 drought has not been found to induce dramatic growth reduction in beech (Leuzinger et al., 2005; Van der Werf et al., 2007), while recording large reductions in gross primary production (GPP) in temperate forests (Ciais et al., 2005; Granier et al., 2007). Nevertheless, the radial growth of beech has been reported to be reduced in the context of more frequent and intensive drought extremes (Beck and Heußner, 2012). Although beech is considered vulnerable to climate change (Ohlemuller et al., 2006; Geßler et al., 2007; Scherrer et al., 2011), its distribution has been modulated, by forest conversion from coniferous forests to mixed stands (Fischer et al., 2013). Beech populations have been able to adapt to environmental changes depending on the level and distribution of genetic variation within and between these populations and their phenotypic plasticity (Bresson et al., 2011; Vitasse et al., 2014).

Plasticity of functional traits plays a determinant role in plant response to different environments, providing relevant information in resistance and resilience of beech to global warming (Stojnic et al., 2015 ´ ). Köcher et al. (2009) found that beech adjusted the vitality and productivity as function of the drought sensitivity of stem growth and leaf and root production, and the success of rejuvenation under a drier climate. Furthermore, changes in photosynthetic characteristics induced by water stress were similar between beech seedlings originated from different sites, but the intensity of change was dependent on the origin (Pšidová et al., 2015). However, phenotypic differences in populations are related to the different directional selection defined by environmental conditions and evolutionary processes (Poorter et al., 2010). Disturbances may also lead to higher genotypic variability at the population level (Borghetti et al., 1993). Beech may exhibit intraspecific variation in drought resistance strategies characterized by varying degree of anisohydricity or isohydricity. A prompt stomatal reaction in beech populations from drier sites may be expected, vs. low stomatal regulation in the plants from mesic conditions (Aranda et al., 2015). Differences in the sensitivity of photosystem II (PSII) to drought among populations may also occur, because those with a drought-avoiding strategy limit carbon assimilation more than drought-tolerant seed sources.

Within this framework, provenance trials represent a valuable tool for assessing adaptive potential to a changing environment (Stojnic et al., 2015 ´ ). Common garden investigations are carried out to test hypotheses on the geographic variation of adaptive traits in tree species, and study the relationships between trait variability and seed source environments (i.e., Sork et al., 2013). In the present study, beech from a wide range of European populations was selected throughout its native range, considering the role of local adaptation and phenotypic plasticity of this species (according to Bolte et al., 2016). We hypothesized that divergence in the plasticity of the response to environmental conditions should have occurred in beech populations originating from different localities. Specific hypothesis were: (1) population differentiation of traits will be moderate to high based on beech distribution; (2) population traits will match seed source climate and will reflect local adaptation (higher water use efficiency in populations from drier sites); (3) populations will distribute in European sites based on morphological and physiological characteristics. Two common garden experiments were carried out in Germany and Italy to quantify geographic variation of adaptive traits in beech populations and investigate relationships between trait variability and seed source, determining the effects of water stress on assimilation rate, predawn water potential, chlorophyll a fluorescence and chlorophyll content, (**Experiment 1**) and the assimilation potential (e.g., gas exchanges and leaf traits) (**Experiment 2**).

#### MATERIALS AND METHODS

#### Plant Material

Beech populations were collected through the pan-European EU Cost STReESS network, from seven sites in six European countries. Seeds were collected in the fall of 2013 from numerous maternal families in forest stands throughout the native range of the species, representing a gradient in mean air temperature and rainfall (**Table 1**). Uniformly sized seeds of each population were surface-sterilized by soaking in 3% sodium hypochlorite for 5 min and rinsing with deionized water. Therefore, a stratification procedure was performed: (1) the seed moisture content was reduced to about 8% of the fresh seeds' moisture content (e.g., by storing them about 1 week at a dry and cool place); (2) seeds were preserved in plastic bags in a freezer at −5 ◦C until the mid of February 2014 (stratification by frost); (3) the seed moisture was increased (using a water sprayer) at a temperature of 3–5◦C; (4) as soon as the first little sprout was visible, the seedlings were transplanted in pots. Seedlings obtained from seeds were planted in pots.

A different level of success of seed germination was observed, thus different populations were considered in each experiment. Seven populations ("Denmark" PV1, "Germany" PV2, "France-Crecy" PV3, "France-Montagne Noir" PV4, "Romania" PV5, "Bosnia" PV 6, and "Spain" PV 7) were tested in Experiment 1 in the North-East of Germany. Four populations (PV1, PV3, PV4, and PV5) were used in Experiment 2 in Central Italy.

## Experimental Set-Up

Two common garden studies were carried out accurately to estimate differentiation under diverse environmental conditions determined by different latitude and elevation between sites (Central Europe, Germany; Mediterranean mountains, Italy). The two experiments took place in greenhouses; plants were subjected to the ambient light without additional illumination; plants were grown in cylindrical PVC pots (1.4 L) filled with a mixture of 70% silty-sand soil (grain size 0–2 mm) and 30% peat-based substrate amended with 2 kg m−<sup>3</sup> Osmocote (NPK 14:13:13+7SO3, plus micro elements).

Twenty replicate seedlings for each population per treatment were designated, by selecting homogenous "matched pairs" based on visual comparison. Well-watered (control) seedlings were regularly watered with tap water to field capacity throughout the experiment. Seedlings subjected to drought treatment were submitted to progressive drought by withdrawing irrigation, when full leaf unfolding had occurred. Initial soil water content was estimated by weighing the samples before and after oven drying at 105◦C for 48 h. Pots were initially watered to saturation. After excess water has drained away field capacity (FC, Blume et al., 2016) was reached at −0.06 MPa soil water potential (pF 1.8). Subsequent changes in pot weight were attributed to changes in soil water content.

A group of 20 seedlings (control) was maintained near to FC by daily watering; whereas, water supply was suspended for those seedlings (20 individuals) subjected to drought treatments. Since watering interruption, each pot was weighed three times per week, in order to estimate the plant available water content (θAWC), using the following Equation (1, cf. Veihmeyer and Hendrickson, 1927):

$$
\theta\_{\rm AWC} = \theta\_{\rm FC} - \theta\_{\rm PWP} \tag{1}
$$

where θ is the soil water content (g) at FC (pF 1.8 ≈ −0.06 MPa soil water potential) and at the visual loss of leaf turgor (commonly defined permanent wilting point in agricultural plants; PWP, pF 4.2 ≈ −1.5 MPa soil water potential). The permanent wilting point (PWP = −1.5 MPa) is used as a conventional threshold for determining the soil water availability (SWA) within the entire effective rooting depth according to the concept of Reid et al. (1984). The PWP was derived for agricultural plants, and young trees like European beech may deplete water resources near to the fine roots at lower soil water potentials than the PWP.

The residual soil water availability (SWA; %) is defined as the proportion of available soil water content (θAWC) during drought treatment referred to the initial value under condition of field capacity.

During the drought experiment the pots were placed at random position and re-randomized after every watering event. Each treatment group was ideally represented by 20 individuals corresponding to replicate seedlings for each population per drought treatment. The pots were allowed to reach different thresholds of drought (50, 40, and 30% SWA) at different time.

Plant responses to extreme drought were studied focusing on gas exchange and water relations, and anatomical traits. We put specific emphasis on the assimilation ability and leaf anatomical traits determining gas exchange and transpiration processes.

Leaf gas exchange was measured on one leaf of five seedlings per treatment once per week, using a portable gas exchange system (GFS-3000, Heinz Walz GmbH Germany) in Germany, and using a LI-COR photosynthesis system (LI-6400; LI-COR Inc., Lincoln, NE, USA) in Italy. The adjusted values were: leaf/cuvettes temperature 25◦C, (Air-to-leaf), Vapor Pressure Deficit 13.8 −16.1 (Pa/kPa), CO<sup>2</sup> concentrations 360 ppm, flow 750 µmol s−<sup>1</sup> , and the light intensity (LED Light Source 3040-L) 1500 µmol m−<sup>2</sup> s −1 . Measurements (assimilation rates and water vapor conductance) were taken between 09:30 and 14:30 (UT) on fully expanded un-shaded leaves and calculated according to von Caemmerer and Farquhar (1981). Instantaneous photosynthetic rate and corresponding stomatal conductance, ratio of intercellular (Ci) to ambient (Ca) CO<sup>2</sup> concentration (Ci/Ca), and ratio of photosynthetic rate (A) to stomatal conductance (A/gs), namely leaf intrinsic water use


TABLE 1 | Location, average of mean annual air temperature (◦C), and total annual rainfall (mm) at the seven sites where the seedlings originated.

efficiency (WUE), were measured at light intensity of 1000 µmol m−<sup>2</sup> s −1 (LED source, red blue 6400-02B) and with a controlled flow of 400 µmol s−<sup>1</sup> of ambient air with CO<sup>2</sup> concentration fixed at 400 µmol mol−<sup>1</sup> . Air temperature and relative humidity were maintained close to ambient values.

For specific insights in plant water status variation during treatment predawn (1 h before the start of the daily light regime) leaf water potential (9pd; MPa) was measured in three seedlings per treatment per provenance to obtain the mean soil water potential next to the roots, closely correlated to the relative transpiration rate (Améglio et al., 1999). Predawn leaf water potential was measured between 1:00 and 5:00 (UT). Exact measurement times varied over the course of the experiment, as the predawn time changed during the experiment. One leaf per five seedlings were detached and rapidly enclosed in a Scholander-type pressure chamber (SKPM1400, Skye Instruments, Llandrindod Wells, UK).

#### Experiment 1

The Experiment 1 was carried out in late summer 2014 by the Thünen Institute of Forest Ecosystems—University of Sustainable Development (HNE) in Eberswalde (52◦ 49′ 28′′ N 13◦ 47′ 29′′ E, 30 m a.s.l.). Within the treatment period relative air humidity averaged 69%, with a minimum of 30% and a maximum of 88%. Air temperature ranged between 11◦C (minimum during night) and 31◦C (maximum during day), and attained a mean of 19.0◦C.

Seedlings reached the ontogenic stage of four fully expanded leaves in 06 August 2014, when the water irrigation was interrupted in mid-summer (06 August 2014) until desiccation of seedlings (06 October 2014), which defined the end of experiment (the duration of experiment was 61 days). Root collar diameter, plant height, leaf area and leaf number were measured at the beginning of the experiment in control seedlings.

Fast chlorophyll a fluorescence kinetics were determined using the chlorophyll fluorimeter Handy PEA (Hansatech, Instruments Ltd, UK). The following parameters were measured: minimal (F0), maximal (Fm), and variable (F<sup>v</sup> = F<sup>m</sup> − F0) fluorescence of dark-adapted leaves; the dark-adapted leaf is illuminated with weak modulated measuring light to give the background fluorescence level in the dark (F0); application of a saturation pulse allows measurement of the maximum fluorescence level in the dark (Fm). The maximal efficiency of PSII photochemistry was calculated as Fv/Fm. The performance index, PI, indicator of sample vitality reflecting the energy cascades, was also determined. Relevant fluorescence parameters were measured using the portable chlorophyll fluorometer MINI-PAM (Walz, Heinz Walz GmbH, Effeltrich, Germany). Rapid light curves (RLCs) were recorded after each chlorophyll fluorescence kinetics measurement: yield (8PSII), the photochemical yield of photosystem II; ETR, the relative rate of electron transport; qP, the coefficient of photochemical quenching; qN, the coefficient of non-photochemical quenching; NPQ, the non-photochemical quenching. Relative chlorophyll content of the leaves was estimated by portable chlorophyll meter (CL-01, Hansatech, Instruments Ltd., Kings Lynn, UK). The results were expressed as chlorophyll index (Chl index; Cassol et al., 2008). Values of Chl index were estimated by device on the basis of the absorbance at 620 and 940 nm.

#### Experiment 2

The Experiment 2 took place during mid-summer 2014 at the University of Molise in Pesche (41◦ 37′ 00′′ N 14◦ 17′ 00′′ E, 732 m a.s.l.). During the experiment, relative humidity ranged between 60 and 70% and air temperature between 15◦C (night) and 30◦C (day), with a mean of 22.5◦C. The light intensity at the plant level never exceeded 1000 µmol photons m−<sup>2</sup> s −1 in sunny days (>80% of the days during which the experiment was carried out), whereas the vapor pressure deficit (VPD) rarely exceeded 2 kPa. In both experiments, the plants were subjected to the ambient light without additional illumination.

Due to the different climatic conditions, the seasonal drought is stronger, occurs earlier and lasts longer in Italy than in Germany. Therefore, seedlings reached the ontogenic stage of four fully expanded leaves in 26 June 2014 in Italy, when water irrigation was interrupted in early summer until desiccation of seedlings in 26 July 2014 (the duration of experiment was 30 days).

Light response curves (A/Q) and intercellular CO<sup>2</sup> response curves (A/Ci) were measured on full expanded leaves from five seedlings grown in control conditions at the end of experiment using a portable photosynthesis system (LI-6400, Li-Cor, Lincoln, NE; according to Tognetti et al., 2004). Light response curves were obtained by measurements at Q-values of 2000, 1500, 1000, 500, 200, 100, 50, 20, and 0 µmol m−<sup>2</sup> s −1 (LED source, red blue 6400-02B). Measurements were recorded automatically at each set point when A had equilibrated; irradiance was changed at intervals of 120–200 s. The CO<sup>2</sup> entering the cuvette was adjusted to maintain a chamber CO<sup>2</sup> concentration ([CO2]) of 400 µmol mol−<sup>1</sup> . The response of leaf A to Q was modeled by a non-rectangular hyperbola where the initial slope was apparent quantum efficiency (ϕ), light compensation point (Ŵ<sup>l</sup> ), and apparent dark respiration (Rd) were estimated from axis intercepts, and the light-saturated maximum photosynthetic rate (Amax) was the upper asymptote. All parameters were determined by fitting data to the model function (Prioul and Chartier, 1977). See Tognetti et al. (2004) for further details on curve fitting and equations.

The CO<sup>2</sup> response curves were obtained by changing the [CO2] entering the cuvette from 50 to 800 µmol mol−<sup>1</sup> with an external CO<sup>2</sup> cartridge mounted on the LI-6400 console and automatically controlled by a CO<sup>2</sup> injector. The CO<sup>2</sup> assimilation rate was first measured by setting the reference [CO2] near ambient (400 µmol mol−<sup>1</sup> ) and then at 300, 200, 100, 50, 400, 400, 600, and 800 µmol mol−<sup>1</sup> . Gas exchange was determined at each step after exposure of the leaf to the new [CO2], waiting for A to reach equilibrium, which was typically <3 min; Q was maintained at 1000 µmol m−<sup>2</sup> s −1 . The response of leaf A to Ci was analyzed according to the mechanistic model of CO<sup>2</sup> assimilation proposed by Farquhar et al. (1980) and subsequently modified by Sharkey (1985). A non-linear regression technique was used to estimate Rday, the maximum rate of carboxylation (Vcmax), the light-saturated rate of electron transport (Jmax), and the rate of triose phosphate utilization for sucrose and starch synthesis (TPU; Sharkey, 1985; Wullschleger, 1993). Description of parameter estimates can be found in Tognetti et al. (2004). The entire leaf area entered the cuvette, limiting errors associated with area determination and the occurrence of a patchy distribution of stomata.

Leaf thickness was measured (lower and upper epidermis, palisade, and spongy parenchyma) on transversal sections (four measurements in each of five plants were averaged). Stomatal density (number of stomata per mm<sup>2</sup> of leaf area), and the average of polar (length) and equatorial (width) stomatal size, and distance between guard cells were determined. Three regions were measured in the median leaf blade, summarizing 30 observations per population. Structural leaf traits were measured on three leaves of three plants per population on all plants at the beginning of the experiment using the uppermost fully expanded leaves. Leaf portions (1–2 cm in diameter) were fixed with 3% glutaraldehyde (v/v) in 0.1-M phosphate buffer (pH 7.2) for 6–8 h under 4◦C, post-fixed in 1% osmium tetroxide for 1 h, and immersed in 0.1-M phosphate buffer (pH 7.2) for 1–2 h. The samples were then dehydrated in a graded ethanol series (50, 60, 70, 80, 90, 95, and 100%) with a last wash in acetone for a better CO<sup>2</sup> substitution during the dehydration procedure at a pressure of 1200 bars. Dry tissue samples were coated with gold in a sputter coater and observed in a scanning electron microscope (SEM Zeiss DSM 940A, Oberkochen, Germany) operated at 10 keV (according to Hultine and Marshall, 2001).

#### Data Analysis

Because all seedlings in each experiment shared the same substrate conditions (nutrient availability, soil texture, etc.), we assumed a significant main effect of imposed dry-down conditions on phenotypic plasticity of physiological traits. Data were examined for assumptions of the homogeneity of variance and normality and were found to conform to model requirements. Variations in leaf gas exchange parameters during the experiment were evaluated by twoway repeated measures ANOVA with "population," as random factor, and "soil water availability," as fixed factor. One-way ANOVA was performed to analyze the effect of population on A/Ci and A/Q curve parameters, root collar diameter, plant height, leaf area and number, and microscopic leaf traits. Statistical analysis was conducted with OriginPro version 8.5.1 (OriginLab, Northampton, MA). Statistical analyses were separately performed considering the different experimental period between Germany and Italy.

## RESULTS

## Experiment 1

Beech seedlings differed in the root collar diameter, plant height, and leaf number (p < 0.0001) between populations in control conditions (**Table 2**). Root collar diameter was higher in German (PV2), French (PV3 and PV4), and Bosnian (PV6) populations; plant height in French (PV3 and PV4), Romanian (PV5), and Bosnian (PV6) populations; leaf number in German (PV2), French (PV3), Romanian (PV5), and Bosnian (PV6) populations (**Table 2**).

The parameters derived from rapid light curves decreased with decreasing SWA (**Figure 1**). Yield and qP showed the same decreasing pattern in all populations. After 48 days of drought treatment, when 20% of SWA was reached, ETR, NPQ, and qN were reduced in comparison with control; however, PV2 showed higher values and PV1 and PV5 than other populations (**Figure 1**). The reduction of SWA showed an increasing trend for F0, and decreasing pattern for PI and Chl index; whereas, Fv/F<sup>m</sup> was not substantial affected by water stress (**Figure 2**).

Assimilation rate decreased as predawn leaf potential became more negative with a steep drop at leaf water potential close to −2 MPa (**Figure 3**).

#### Experiment 2

Leaf traits were significantly different among populations in control conditions (**Table 3**). However, leaf traits did not show a common pattern among populations, suggesting complex phenotypic effects driven by prevailing environmental conditions at the geographical origin (**Table 3**). The lowest values of stomatal density and spongy mesophyll thickness were found in PV4 and PV5, of mesophyll thickness in PV4, of stomatal length (polar size) in PV3, and of stomatal width (equatorial size) in PV1.

Photosynthetic response curves to internal CO<sup>2</sup> and irradiance (analyzed in control plants) did not show differences among populations (**Table 4**).

Photosynthetic rate, Ci/C<sup>a</sup> and the ratio photosynthesis and stomatal conductance (A/gs) significantly differed among populations (P < 0.001) (**Table 5**). Decreasing SWA impacted


TABLE 2 | Root collar diameter, plant height, leaf area, and number for beech seedlings at the beginning of the water-stress experiment.

The values for the mean (± standard error), and the statistical analysis of the populations (ANOVA test) (F-value, P-level) are shown (experiment 1).

all considered ecophysiological traits (P < 0.001), as well as the interaction population x treatment (P < 0.001).

Gas exchange was related to SWA through a sigmoidal pattern, describing a threshold-type transition from one phase to the next (**Figure 4**). The inflection point of the function was observed between 30 and 26% SWA (29.45, 29.42, and 26.15% SWA, for photosynthetic rate, stomatal conductance and Ci/Ca, respectively).

#### DISCUSSION

#### Differences among Populations

Beech seedlings from seeds collected across a climatic gradient were monitored in two common garden experiments conducted in Germany and Italy to quantify the geographic structure of variation in ecophysiological traits and the adaptation potential of populations to drought stress. We hypothesized that differentiation of tolerance/avoidance traits in beech progenies was constitutively linked to the seed source, and tested whether population-specific functional and structural responses to drought stress were revealed when seedlings were exposed to decreasing soil water availability.

The study highlighted intraspecific differentiation in plant performance when seedlings were not exposed to constraints (**Tables 2**–**4**), which was related to phenotypic variation among populations (e.g., Thiel et al., 2014). Beech has a wide area of distribution and the gradual change in ecophysiological traits and environmental gradients exhibited by different populations can shape local adaptation. Netzer et al. (2016) found that Greek populations were more tolerant to drought stress than German ones, while a Bulgarian population was the most sensitive. These results confirm once again the relevance of seed source for assessing the response of beech to drought stress. The French population (PV3) showed a distinct behavior in morphological traits; whereas, Danish (PV1) and Spanish (PV7) populations showed the lowest plant height, collar diameter, and leaf number (**Table 2**). Well-watered conditions resulted in higher plant height in French, Romanian, and Bosnian populations, and higher leaf number in German, French, Romanian, and Bosnian populations, suggesting higher plant growth potential in French, Romanian, and Bosnian populations, but failing to identify a clear population-related trait. Contradictory results on growth differentiation between beech populations with different native precipitation regimes are reported in literature (e.g., Tognetti et al., 1995; Knutzen et al., 2015). Indeed, genetic differentiation between beech stands was found to be located mainly within stands and not between stands within the distribution range of this species (Buiteveld et al., 2007; Meier and Leuschner, 2008). Even though, clear genetic differentiation also among stands based on adaptive genetic markers for populations coming from different climatic conditions was found by Sander et al. (2000) and Dounavi et al. (2016). Results of photosynthetic response curve to internal CO<sup>2</sup> concentration and irradiance did not show differences among populations, highlighting common assimilation potential and photosynthetic capacity across populations (**Table 4**). These results pointed to the lack of intraspecific differences in these traits among populations related to drier or moister native conditions, as observed by Knutzen et al. (2015). The impact of the last glacial period on the areal distribution of beech and the preferred selection of seed sources by forest managers may have concurred in reducing differences among populations in certain functional traits.

Structural leaf traits did not show a common pattern across populations (**Table 3**). Variations in tissue density, mesophyll thickness, and stomatal features have been related to the influence exerted by environmental conditions on leaf economics via modulation of hydraulic conductance and carbon budget (Brodribb et al., 2013; Sack et al., 2013; Blonder et al., 2015). Thicker leaves (and mesophyll) have long been recognized as structural adaptive trait in dry environments (Vogel, 2009), generally resulting in reductions in light interception and carbon gain expressed on a tissue volume basis. However, thicker palisade layer is a common trait for efficient light interception in mesomorphic sun leaves (Marchi et al., 2008). As efficient light harvesting is expensive in terms of biomass investment in support tissues, the efficiency with which foliage can be supported is a major factor determining species ability to grow under constrained environments (e.g., Niinemets et al., 1998). Intrinsic limitations associated with too high costs for foliar support may constrain more those populations in environments with high irradiance (southern Europe), where carbon for foliage construction is relatively cheaper.

As a major effect of experimental conditions on plant traits was expected to determine the phenotypic plasticity of corresponding traits, a significant interaction between beech populations and greenhouse conditions would indicate that responses to drought depended on the origin of seed sources. Hence, it is important to notice that we considered plasticity of a given trait at the population level, as an average value of the

FIGURE 2 | Fluorescence traits (F<sup>0</sup> ), the minimal Chl fluorescence; (Fv/Fm), the maximal efficiency of PSII photochemistry; (PI), the performance index; (Chl index), chlorophyll index, following the soil water availability during the experiment (experiment 1).

trait across individuals from each population, rather than at the genotype level.

#### Drought Stress Impacts

This experiment and the companion study (Bolte et al., 2016) demonstrated the high correlation between physiological responses and water availability in these beech populations. This is consistent with findings in other common garden experiments conducted on beech, which showed considerable differences among populations in adaptation to environmental conditions (Kreyling et al., 2012, 2014; Thiel et al., 2014). The effect of drought treatment did not elicit marked differences among populations as expected (see Peuke et al., 2002).

When beech seedlings were exposed to water constraints, they adopted stress-compensating mechanisms (e.g., reduction of leaf water potential, regulation of gas exchange, control of plant growth), which differed among populations (Knutzen et al., 2015). Water constraints significantly decreased photosynthetic rate and stomatal conductance to prevent plant dehydration plant (**Table 2**) (Peuke et al., 2002; Aranda et al., 2015). However, according to Tognetti et al. (1995), there was no clear relationship between plant response and population range, suggesting a variable sensitivity to drought. The limitation of gas diffusion through stomata, before any changes occur in plant water status, could suggest a control of leaf water loss toward isohydric behavior (Attia et al., 2015), thus preserving leaf hydration. It is also possible that plants with highly sensitive stomata (isohydric), closing at relatively high water potentials, but with even more drought-sensitive hydraulic system could show anisohydric behavior. In the present study, Ci/Ca-values increased with the reduction of water availability, indicating regulation capacity of water balance achieved by a slow stomatal reaction in response to a decrease in leaf water potential (Lawlor and Tezara, 2009). Yet, a reduction in leaf intrinsic WUE (A/gs) was observed. Drought sensitivity increased in the sequence France Montagne Noir (PV4) < Denmark (PV1) < Romania (PV5) < France Crecy (PV3), with 21, 30, 79, and 81.5% of reduction, respectively, in comparison with control plants. Photosynthesis was substantially reduced and eventually stopped under severe drought, which may affect metabolic patterns and the allocation of carbon to structural traits to varying degrees, ultimately translating into plant death (McDowell et al., 2011). In this framework, the performance of beech seedlings was probably determined by plant size, defining the attitude of a plant to photosynthesize, store nutrients, and mobilize resources, and to growth (e.g., Villar-Salvador et al., 2012). Inhibition of metabolism, with maintenance of light respiration, probably increased C<sup>i</sup> , defining a greater sensitivity of photosynthetic rate than light respiration to water deficit (Cornic and Fresneau, 2002). Plants that display isohydric characteristics have tight and continuous water potential homeostasis through stomatal control. They constantly regulate their water loss within a certain range to avoid damaging water deficits due to hydraulic failure (Buckley, 2005), though this may inhibit CO<sup>2</sup> diffusion into leaves and increase the risk of carbon starvation. The identification of population-specific performance under decreasing SWA, in this as well as in other recent experiments (e.g., Aranda et al., 2015; Knutzen et al., 2015; Dounavi et al., 2016), helps defining the risk of or resilience

drought led to heterogeneous and variable response patterns in these populations.

to local mortality across the native range of beech. However,

FIGURE 3 | Relationships between assimilation rate (AR) and shoot predawn water potential (9pd). Non-linear regression model (exponential function) was used for curve-fitting analysis (experiment 1).

Beech populations variably combined functional traits in response to drought, although there was not a welldefined gradient in drought tolerance among populations. The phenotypic behavior showed a straight reduction of functional processes in response to decreasing SWA. The decrease of leaf water potential with the consequent reduction of assimilation rate under water stress showed a conservative control of water loss in these beech populations (**Figure 3**), highlighting the capacity of the plant hydraulic system to regulate the supply of water to leaves (Attia et al., 2015). The shape parameter of this relationship described accurately the response dynamics of seedlings to changing SWA (**Figure 4**). Furthermore, the inflection point of the function allowed the detection of a SWA threshold, where the SWA-value that induced a sharp drop of gas exchange was 28.34% (calculated as the mean of SWA-values, **Figure 4**). The reduction of SWA induced also the reduction of PI and Chl index and the increase of F0; whereas, the Fv/F<sup>m</sup> was not substantially affected by water stress. Overall, these parameters showed relatively high values, illustrating the resilience of the photochemical apparatus with decreasing SWA in beech (but see García-Plazaola and Becerril, 2000). However, when seedlings reached leaf predawn water potential of −2 MPa, gas exchange approached zero, indicating high sensitivity of stomatal behavior in response to extreme drought (in agreement with 9pd-values as low as −1.76 MPa—Köcher et al., 2009). A parallel reduction in the activity of PSII reaction center, efficiency of light capture, and rate of electron transfer was observed in all populations, which may avoid permanent damage from photoinhibition if stress conditions were quickly released. PV2 maintained the highest values of ETR and NPQ, utilizing more absorbed light energy for photochemistry and allocating more light energy to NPQ pathways than the other populations. The reversible down-regulation of PSII photochemistry would contribute to an enhanced photo-protection in severely stressed beech seedlings (Gallé and Feller, 2007).

Genetic variability and phenotypic plasticity, at morphological, physiological, and phenological level, are the key factors for the adaptation to environmental constraints, and are supposed to be wide in species, such as beech, with ample distribution range (Bussotti et al., 2015). The issue of the genetic background on the enhanced tolerance to drought in beech has been discussed mostly because of the probable evolutionary



Values are means ± standard error for population (n = 5 seedlings). Significance (F-value; P-level) of population effects is shown (experiment 2).


#### TABLE 4 | Populations effects on photosynthesis (A) response curves to internal [CO<sup>2</sup> ] (C<sup>i</sup> ) (A/C<sup>i</sup> ) and to irradiance (Q) (A/Q).

Measurements were made on fully expanded leaves of control plants. Values are the means ± standard error for population (n = 5 seedlings). Significance (F-value; p-level) of population effects for A/Ci and A/Q curve parameters are presented. Rday , daytime respiration (µmol CO<sup>2</sup> m−<sup>2</sup> s −1 ); Vcmax , maximum rate of carboxylation (µmol CO<sup>2</sup> m−<sup>2</sup> s −1 ); Jmax , light-saturated rate of electron transport (µmol electrons m−<sup>2</sup> s −1 ); TPU, rate of triose phosphate utilization (µmol CO<sup>2</sup> m−<sup>2</sup> s −1 ); Amax , maximum photosynthetic rate (µmol CO<sup>2</sup> m−<sup>2</sup> s −1 ); ϕ, apparent quantum efficiency (µmol CO<sup>2</sup> µmol−<sup>1</sup> photons); Rd, apparent dark respiration (µmol CO<sup>2</sup> m−<sup>2</sup> s −1 ); and Ŵ<sup>l</sup> , light compensation point (µmol photons m−<sup>2</sup> s −1 ) (experiment 2).

TABLE 5 | Daytime gas exchange under saturating light conditions of seedlings for each population grown in different drought thresholds (treatment).


Photosynthetic rate (A), stomatal conductance (gs), ratio of intercellular (C<sup>i</sup> ) to ambient (Ca) [CO2] (Ci/Ca), and leaf intrinsic WUE (A/gs) were measured on the first fully expanded leaf (LPI 5) (measurements made at 1000–1200 h). Values are means (± standard error) for population (n = 5 seedlings). Significance of treatment effect (drought treatment expressed at SWA level; ANOVA) for each variable and their interaction is indicated by F-value and P-level (experiment 2).

adaptation of local population by selection processes, which allow only drought-tolerant individuals to inherit genetic and phenotypic traits to next generations (Hampe and Petit, 2005). This is reputed to exhibit the highest resistance to the environmental stress (Thiel et al., 2014; Aranda et al., 2015; Knutzen et al., 2015). It must be pointed out that differences in moisture availability at the native sites may not be high enough to elicit distinct seedling performance among populations, and that factors other than differences in precipitation amount at the origin may be equally important, thus hindering the detection of clear population differences in common garden experiments.

hill slope; (C) is the inflection point; (D) is the maximum asymptote; global goodness of fit is defined by R-squared, and significance of the regression by p-level (experiment 2).

## CONCLUDING REMARKS

A common trend in responses of beech originated from different sites was established in relation to a decreasing of soil water availability. Phenotypic changes of plant traits in relation to SWA thresholds highlighted that decreasing SWA thresholds may favor ecophysiological traits that optimize carbon assimilation to counterbalance the reduced growing period. Functional traits define the contribution of genetic variability and phenotypic plasticity for understanding the adaptive potential to drought stress of beech across its large geographical range of distribution. In this sense, our study did not include populations spanning the entire environmental gradient of beech, which warrants further studies to better elaborate the relationships between ecophysiological acclimation to drought and seed source climate. A deeper understanding of the mechanism of drought tolerance in beech is needed to support strategies of forest management toward assisted selection and seed transfer. Planting of more resistant genotypes into droughtimpacted forest stands may help implementing adaptive forest management options to ameliorate the impact to anticipated climate change, but parallel garden experiments are needed and should include more seed sources and ecophysiological traits.

## AUTHOR CONTRIBUTIONS

Substantial contributions to the conception and design of the work: CC, TC, AB, RT. Acquisition, analysis, or interpretation of data for the work: CC, MM, EP, LD, SM, LM, AA, TC, AB, RT. Drafting the work or revising it critically for important intellectual content: CC, MM, EP, LD, SM, LM, AA, TC, AB, RT. Final approval of the version to be published: CC, MM, EP, LD, SM, LM, AA, TC, AB, RT. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved: CC, MM, EP, LD, SM, LM, AA, TC, AB, RT.

## ACKNOWLEDGMENTS

We are grateful to Dinca Lucian (Marin Dracea National Forest Research-Development Institute, Bucharest, Romania), Anders Ræbild (University of Copenhagen, Dept. of Geosciences and Natural Resource Management, Frederiksberg C, Denmark), Martin de Luis [Grupo de Clima, Agua, Cambio Global y Sistemas Naturales, Departamento de Geografía y Ordenación del Territorio, Facultad de Filosofía y Letras, Instituto de Investigación en Ciencias Ambientales (IUCA), Universidad de Zaragoza, Zaragoza, España], Branislav Cvjetkovic (University of Banja Luka, Faculty of Forestry, Banja Luka, Bosnia and Herzegovina) that provided the seeds of some of the studied populations. We are grateful to Dr. Mirko Liesebach and Rainer Ebbinghaus from the Thünen Institute of Forest Genetics (Groß-Hansdorf, Germany) for the cultivation of the beech seedlings used in the Experiment 1 (Germany). This article is based upon work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology).

## REFERENCES


change. Environ. Exp. Bot. 111, 91–113. doi: 10.1016/j.envexpbot.2014. 11.006


ecophysiological conceptual model of plant survival. New For. 43, 755–770. doi: 10.1007/s11056-012-9328-6


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Cocozza, de Miguel, Pšidová, Ditmarová, Marino, Maiuro, Alvino, Czajkowski, Bolte and Tognetti. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Above-Ground Dimensions and Acclimation Explain Variation in Drought Mortality of Scots Pine Seedlings from Various Provenances

#### Hannes Seidel<sup>1</sup> 1,2 \* and Annette Menzel

<sup>1</sup> Professorship of Ecoclimatology, Department of Ecology and Ecosystem Management, TUM School of Life Sciences Weihenstephan, Technische Universität München, Freising, Germany, <sup>2</sup> Institute for Advanced Study, Technische Universität München, Garching, Germany

#### Edited by:

Cristina Nabais, University of Coimbra, Portugal

#### Reviewed by:

Gerald Moser, Justus Liebig University Giessen, Germany Ana-Maria Heres, National Museum of Natural History, Spanish National Research Council, Spain

> \*Correspondence: Hannes Seidel hseidel@wzw.tum.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 29 January 2016 Accepted: 27 June 2016 Published: 07 July 2016

#### Citation:

Seidel H and Menzel A (2016) Above-Ground Dimensions and Acclimation Explain Variation in Drought Mortality of Scots Pine Seedlings from Various Provenances. Front. Plant Sci. 7:1014. doi: 10.3389/fpls.2016.01014 Seedling establishment is a critical part of the life cycle, thus seedling survival might be even more important for forest persistence under recent and future climate change. Scots pine forests have been disproportionally more affected by climate change triggered forest-dieback. Nevertheless, some Scots pine provenances might prove resilient to future drought events because of the species' large distributional range, genetic diversity, and adaptation potential. However, there is a lack of knowledge on provenance-specific survival under severe drought events and on how acclimation alters survival rates in Scots pine seedlings. We therefore conducted two droughtinduced mortality experiments with potted Scots pine seedlings in a greenhouse. In the first experiment, 760 three-year-old seedlings from 12 different provenances of the south-western distribution range were subjected to the same treatment followed by the mortality experiment in 2014. In the second experiment, we addressed the question of whether acclimation to re-occurring drought stress events and to elevated temperature might decrease mortality rates. Thus, 139 four-year-old seedlings from France, Germany, and Poland were subjected to different temperature regimes (2012– 2014) and drought treatments (2013–2014) before the mortality experiment in 2015. Provenances clearly differed in their hazard of drought-induced mortality, which was only partly related to the climate of their origin. Drought acclimation decreased the hazard of drought-induced mortality. Above-ground dry weight and height were the main determinants for the hazard of mortality, i.e., heavier and taller seedlings were more prone to mortality. Consequently, Scots pine seedlings exhibit a considerable provenance-specific acclimation potential against drought mortality and the selection of suitable provenances might thus facilitate seedling establishment and the persistence of Scots pine forest.

Keywords: die-back, climate change, experiment, drought, warming, acclimation, above-ground biomass, height

## INTRODUCTION

fpls-07-01014 July 5, 2016 Time: 15:11 # 2

Numerous forest-diebacks have been observed in the last few decades triggered by recent climate change-related drought stress and heat spells (Allen et al., 2010) covering a wide range of climate zones from boreal to tropical regions. Risk of forest mortality may increase in the future as climate models project an increase of temperature and a decrease in precipitation. Thus, drought severity might be exacerbated by rising temperatures (Kirtman et al., 2013). Drought-induced tree mortality may have adverse effects on forest structure and ecological communities, ecosystem function and services as well as biosphere–atmosphere interactions (reviewed, e.g., in Anderegg et al., 2012).

Acclimation (adjustment to environmental changes) is a key process to resist re-occurring stress events (Kozlowski and Pallardy, 2002). Mechanisms involved in drought and warming acclimation comprise molecular, physiological, and structural adjustments. Molecular adjustment against drought typically enhances gene expression pathways for production of molecules, such as abscisic acid (ABA), proline, and soluble sugars, linked to the maintenance of turgor and cell integrity (Peñuelas et al., 2013). ABA additionally inhibits cell expansion and thus reduces growth (Lambers et al., 2008). Warming acclimation on the molecular level increases the synthesis of heat shock and antistress proteins (Peñuelas et al., 2013). Physiological drought acclimation comprises changes in resource allocation (Callaway et al., 1994; Peña-Rojas et al., 2005; Poorter et al., 2012), a decrease of photosynthetic activity (Rennenberg et al., 2006) and an increase of water use efficiency (Brodribb and Hill, 1998; Lloret et al., 2004a). Structural acclimation processes toward water stress are the reduction of total leaf area by leaf size reduction (Kubiske and Abrams, 1992; Martínez-Vilalta et al., 2009; Günthardt-Goerg et al., 2013) or leaf shedding (Ogasa et al., 2013), the adjustment of the hydraulic system by altering xylem conduit dimensions (Eilmann et al., 2009; Bryukhanova and Fonti, 2012), leaf/sapwood area ratio and leaf-specific hydraulic conductivity (Martínez-Vilalta et al., 2009). Furthermore, reduction of aboveground biomass might be linked to increased drought resistance (Alía et al., 2001; Valladares et al., 2007).

The seedling stage is an ontogenetic phase that is very drought sensitive and prone to mortality (Hanson et al., 2001; Lloret et al., 2004b) likely due to lower carbohydrate reserves (Niinemets, 2010), smaller rooting volume and lower rooting depth than mature trees (Cavender-Bares and Bazzaz, 2000). In general, proposed mechanisms of drought-induced tree mortality are carbon starvation caused by depletion of carbohydrate pools (Galiano et al., 2011; Adams et al., 2013; Mitchell et al., 2013), hydraulic failure of xylem conduits impairing water transport (Anderegg and Anderegg, 2013; Mitchell et al., 2013; Salmon et al., 2015), and failure of phloem transport affecting carbon translocation (Sala et al., 2010; McDowell et al., 2011; Adams et al., 2013). Survival rates of seedlings under drought were found to depend on various traits. The survival of several Mediterranean tree species is negatively correlated with total plant dry mass, total leaf area and positively correlated to leaf area ratio and shoot–root ratio across species (Valladares and Sánchez-Gómez, 2006). Thus, trait characteristics indicating a higher transpiring surface are linked to higher mortality rates. Tree size in general increases the hazard of droughtinduced mortality (Bennett et al., 2015) which might be explained by an increased risk of hydraulic dysfunction with increasing tree height and greater leaf area (McDowell and Allen, 2015).

Successful and adequate seedling establishment is essential for sustainable forest cover and production. Decreasing seedling survival under climate change may hamper forest regeneration and thus result in alternative forest communities or even nonforest ecosystems (Anderson-Teixeira et al., 2013). Bussotti et al. (2015) proposed three mechanisms of forest adaption to future environmental conditions: persistence by acclimation and phenotypic plasticity, evolution or local adaptation, and migration or substitution of tree species. If natural migration cannot keep pace with the rate of climate change, assisted migration of more adapted alternative tree species, but also suitable provenances within the same species, might support timely adaptation. Several studies have demonstrated provenance differences in sensitivity and response to reduced water availability, indicated, e.g., by differences in shoot length, diameter increment, and stomatal conductance of Fagus sylvatica (Rose et al., 2009; Thiel et al., 2014; Knutzen et al., 2015), in height and dry weight of Picea abies (Modrzynski and Eriksson, 2002 ´ ), and in stomatal conductance, transpiration rates, leaf hydraulic conductance, osmotic potential, loss of hydraulic conductivity, and water-use efficiency of Pinus halepensis (Tognetti et al., 1997; Klein et al., 2013). However, studies on provenancespecific drought mortality are rare. Mortality rates were evaluated for central and marginal provenances of F. sylvatica (Thiel et al., 2012), provenances of Quercus pubescens (Wellstein and Cianfaglione, 2014), Pinus ponderosa provenances from origins differing in summer drought (Cregg, 1994) and central provenances of Pinus sylvestris (Cregg and Zhang, 2001), but only P. ponderosa provenances differed in drought survival, although not significantly.

More than one-third of the forest diebacks reported by Allen et al. (2010) are linked to Scots pine (P. sylvestris L.) forests although this species is considered to be drought resistant (Ellenberg, 1988). Scots pine has an extensive latitudinal (Spain to Scandinavia) and longitudinal (Spain to the far east of Russia) distribution range covering various climate zones from Mediterranean to boreal habitats (Boratynski, 1991). The huge distribution of this species favors local adaptation of provenances to contrasting environmental conditions (Boratynski, 1991; Reich and Oleksyn, 2008). The southernmost rear age of the distribution range in Spain and Italy is composed of post-glacial relict populations (Boratynski, 1991) which might be of special importance for provenance-based assisted migration due to high levels of genetic differentiation (Hampe and Petit, 2005) or even adaptations to heat and drought (Alía et al., 2001). Scots pine is an isohydric species which aims at minimizing water loss by tight stomatal control (Irvine et al., 1998) as well as by adjustment of leaf/sapwood area ratio, leaf-specific hydraulic conductivity, total leaf area, and conduit size (Sterck et al., 2008; Martínez-Vilalta et al., 2009). This drought avoiding strategy by closing stomata is implemented at the cost of reduced photosynthetic

carbon gain (Mitchell et al., 2013), thus the risk of droughtinduced mortality might be especially observed under moderate but long lasting droughts. Regarding local adaptations, Scots pine provenances show differing drought responses in seedling establishment (Richter et al., 2012), in shoot diameter and height increment (Taeger et al., 2013a, 2015), and drought resistance (Taeger et al., 2013b).

However, there is a lack of knowledge how drought acclimation and provenance effects (adaptation) may impact mortality rates of P. sylvestris which would allow optimizing the provenance choice for assisted migration. In order to address this issue, we conducted two drought mortality experiments with potted Scots pine seedlings. The first mortality experiment in 2014 was based on 12 different Scots pine provenances from its south-western distribution in Europe. The second mortality experiment was conducted in 2015 and investigated seedlings of three provenances that had been acclimated to different seasonal drought events and temperature regimes before. Drought acclimation included a spring and a summer drought in 2013 and a spring drought in 2014. All drought acclimation treatments were applied within two temperature regimes, namely ambient air temperature and passively elevated temperature in a greenhouse. We hypothesized that (1) the hazard of mortality induced by a severe drought treatment differs across provenances, (2) acclimation by previous drought events and elevated temperature decreases the hazard of droughtinduced mortality, and (3) the hazard rate depends on the climate at the provenances' origin.

## MATERIALS AND METHODS

#### Plant Material

Scots pine seedlings originated from 12 provenances distributed along their south-western distribution. Climatic conditions at the origin of the seeds range between ∼3 and 11◦C annual mean temperature and ∼600 and 1100 mm of annual precipitation comprising western Mediterranean (Spain, France, and Italy) and continental (Switzerland, Germany, Poland, Hungary, and Bulgaria) sites (**Table 1**). Plants were grown from seeds in a nursery in south-eastern Germany in 2011. All seeds were collected from autochthonous populations, except for Alpenkiefer from Germany and Plantage Pornoapati from Hungary which were obtained from seed orchards. In 2012, seedlings were brought to the Gewächshauslaborzentrum (GHL) near Freising, Germany, where they were potted into 3 l pots containing peat substrate. Seedling treatments/acclimation and provenances studied differed between the first mortality experiment conducted in 2014 and the second one conducted in 2015 as explained in the next chapters.

#### Mortality Experiment 2014

Seedlings used in the first mortality experiment in 2014 comprised all 12 provenances listed in **Table 1**. In contrast to the second mortality experiment (see Mortality Experiment 2015 and Acclimation Treatments), they had not experienced any prior acclimation treatment and had been exclusively grown in the greenhouse. In total, 760 individuals were examined, but numbers of individuals per provenance varied between 31 and 187 individuals per provenance (**Table 2**). They were randomly arranged on three tables next to each other in a greenhouse (**Figure 1A**). All pots got well watered manually until the lethal drought treatment started by withholding irrigation (March 26 to July 1, 2014).

## Mortality Experiment 2015 and Acclimation Treatments

In total, 139 seedlings were studied in the second mortality experiment in 2015. They comprised three provenances (Alpenkiefer, D; Mont Ventoux, F; Suprasl, PL) with 43, 44, and 52 individuals, respectively (**Table 2**). These seedlings were remnants of an extensive seasonal drought and warming experiment on more than 1000 individuals from 10 European Scots pine provenances. Out of these, three provenances were selected in order to provide sufficient sample sizes and to cover contrasting climatic conditions/seasonal precipitation


Annual mean temperature (T), annual sum of precipitation (PPT) are shown for the period 1950–2000 (WorldClim data base; Hijmans et al., 2005). All seeds except the provenances HU14 and D7 (seed orchards) are from autochthonous stands. Provenances are arranged in decreasing order of latitude.



Height was not measured in 2014. In the mortality experiment 2015, the seedlings were acclimated in different drought and temperature regimes during 2013 and 2014. The drought treatment in 2013 was applied in two phases with a spring and/or a summer drought from March 22 to June 14 and July 10 to August 21, respectively. The drought treatment in spring 2014 lasted from March 23 to June 23. Seedlings were acclimated in two temperature regimes, in a vegetation hall (ambient temperature) and a greenhouse (elevated temperature by passive warming). Provenances are arranged in decreasing order of latitude. Mean values for above-ground dry weight and height are given with respective standard deviation (SD). Group differences were calculated comparing contrasts with the Tukey's range test. Values sharing the same letter are not significantly different.

patterns (**Table 1**, Supplementary Figure S1). Prior to this second mortality experiment seedlings were subjected to two different temperature acclimation treatments, and within each they were exposed to drought acclimation treatments in 2013 and 2014 in order to study the influence of carry-over effects on mortality (see description of acclimation treatments below). In March 2014, seedlings were replanted from 3 l pots into 20 l pots containing peat substrate of identical composition as before. All 139 plants were put together in the greenhouse in December 2014 (**Figure 1B**), were randomly distributed on three tables and got well watered till the start of the mortality experiment in May 2015. During the lethal mortality experiment, irrigation was totally intermitted from May 22 to August 31, 2015.

#### Temperature Acclimation

Seedlings were grown under two temperature regimes from mid of 2012 till end of 2014 (Supplementary Figure S2). During this period 65 of the 139 seedlings were placed in a vegetation hall (glass-roofed building with open sidewalls), and 74 seedlings were placed in the greenhouse. Temperatures in the vegetation hall were similar to ambient conditions, but passive warming of the greenhouse increased temperatures. Within each building, air temperature and relative humidity (RH) were recorded in 10-min intervals with a temperature/RH data logger (HOBO U23 Pro v2, Hobo <sup>R</sup> , Onset Computer Corporation, Bourne, MA, USA). The vapor pressure deficit (VPD) was calculated using air temperature and RH as input variables after Allen et al. (1998) as a measure of atmospheric dryness. The mean temperature difference during 2013 and 2014 between the vegetation hall and the greenhouse was 3.0◦C (Supplementary Figure S2B); however, the difference was more pronounced in frost periods (defined as periods with days having temperatures below 0◦C), i.e., 5.7◦C (January 1 to April 8, 2013), 4.0◦C (November 12, 2013 to April 17, 2014) and 5.8◦C (December 8–11, 2014) than in frost-free summer periods (1.3◦C in 2013 and 2.3◦C in 2014). The daily means of VPD were in general

higher in the greenhouse than in the vegetation hall with some exceptions during the whole acclimation period (Supplementary Figure S2D). The overall mean VPD during 2013 and 2014 was 0.16 kPa higher in the greenhouse than in the vegetation hall.

#### Drought Acclimation in 2013 and 2014

The drought acclimation comprised three drought periods in total (two in 2013, one in 2014), identical within each temperature regime, and accompanied by respective well-watered control groups (Supplementary Figure S3). In 2013, the drought was applied in two separated periods with a spring and/or a summer drought from March 22 to June 14 and July 10 to August 21, respectively. In both periods, automated irrigation was stopped and thereafter, soil moisture was adjusted to oscillate around the permanent wilting point by adding small amounts of water when necessary. The permanent wilting point (pF 4.2) corresponded to 12 Vol% soil moisture derived from water retention curves following the pressure plate method by Richards (1941). Thus, drought treated seedlings had to survive near the limit (permanent wilting point) for around 5 weeks in spring and for around 4 weeks in summer 2013 and got well-watered in between. The control and the three drought treatment groups (spring drought, summer drought, spring and summer drought 2013) comprised 32–37 individuals (**Table 2**).

These individuals of 2013 were split to a spring drought and well-watered control group in 2014, all within their respective temperature regime. This drought acclimation treatment lasted from March 23 to June 23 by totally withholding irrigation. Soil moisture fell below the permanent wilting point for around 5 weeks. Fifty-eight individuals belonged to the watered control group and 81 individuals were part of the drought acclimation group.

Soil moisture was monitored twice a week in the afore mentioned overarching experiment using a hand-held soil moisture sensor (UMP1, Umwelt-Geräte-Technik GmbH, Müncheberg, Germany) on 240 pots equally spread across provenances and treatments (Supplementary Figure S3).

## Mortality, Above-Ground Dry Weight and Height Assessment and Meteorological Conditions during the Mortality Experiments

Mortality assessment was done with a knife carefully scratching the bark. The seedlings were classified as alive if the cambium tissue underneath the bark was green and classified as dead when this tissue was brownish. Evaluation of mortality was done on six dates in 2014 from April 28 to July 1 (day 33, 54, 62, 76, 82, and 97 after the lethal drought started) and almost once a week between June 2 and August 31 in 2015 (on day 12, 19, 26, 32, 40, 47, 54, 61, 68, 75, 82, 88, and 102 after initiation of the lethal drought treatment). In 2015, the height of seedlings was measured at the beginning of the experiment from the substrate surface to the terminal tip using a folding rule. After each mortality experiment, above-ground biomass was harvested and oven dried at 60◦C for 48 h to assess total above-ground dry mass.

Meteorological conditions were variable between the mortality experiments conducted in 2014 (March 26 to July 1) and in 2015 (May 22 to August 31). During the study periods mean temperature, mean RH, and mean VPD were 16.6 and 24.5◦C; 62.5 and 54.7%; and 0.9 and 1.6 kPa in 2014 and 2015, respectively. Variables ranged from 2.9 to 39.1◦C, 21.9 to 93%, and 0.1 to 5.3 kPa in 2014; and from 11.2 to 41.6◦C, 18.8 to 88.5%, and 0.2 to 6.3 kPa in 2015, clearly indicating that the lethal drought stress in summer 2015 was higher than in spring 2014.

#### Statistics

#### Above-Ground Dimensions (Seedling Height and Above-Ground Dry Weight)

The effect of provenances and acclimation treatments on seedlings' above-ground dry weight and height was evaluated separately using linear models (stats, R Core Team, 2015) in R 3.2.2. In the case of analyzing above-ground biomass in the mortality experiment in 2014, just provenance (12 different provenances) served as an explanatory variable. For analyzing above-ground biomass and height in the mortality experiment in 2015, explanatory variables were provenance (three different

ones), drought acclimation treatment 2013 (only spring drought, only summer drought, spring and summer drought, and control), drought acclimation treatment 2014 (spring drought, control) and building (vegetation hall, greenhouse). Provenance, drought acclimation treatment and temperature acclimation treatment (building) were incorporated in the models as factorial dummy variables; more precisely, as the presence or absence of drought and the affiliation to vegetation hall (ambient temperatures) or greenhouse (elevated temperatures).

#### Survival Analysis

Survival analysis was conducted using a Cox proportional hazards regression model (survival package in R; Therneau and Grambsch, 2000). This type of model calculates the hazard ratio (HR) that is the probability of a death event in the treatment group in relation to the probability in the reference group. The covariates provenance and above-ground dry weight as well as their two-way interaction were added to analyze mortality of Scots pine seedlings in the mortality experiment 2014.

Two different models were constructed to analyze mortality and the effect of covariates on the drought hazard in the 2015 experiment. The first model contained the covariates provenance, drought treatments in 2013, drought treatment in 2014, temperature regime and the two-way interactions of provenance with drought treatments in 2013, drought treatment in 2014 and temperature regime (so called factorial model). The second model to analyze the 2015 mortality experiment included, in addition to the covariates and twoway interactions of the first model, seedling height and aboveground dry weight (so called continuous model). Provenance, drought acclimation treatment and temperature acclimation treatment were incorporated in the models as factorial dummy variables as explained above. Since above-ground biomass and height varied significantly between levels of provenances and hardening treatments (**Table 2**, see Supplementary Tables S1 and S2 for the summary output of linear models), weight and height were centered on respective group means to avoid confounding effects of covariates with tree dimension effects.

All survival models were simplified by stepwise excluding covariates/interactions that did not improve the models explanatory power with the Anova function (car, Fox and Weisberg, 2011). The proportional hazard assumption was checked by examining diagnostic plots and with the cox.zph function (survival; Therneau and Grambsch, 2000).

Pairwise comparisons in the linear and Cox proportional hazard models were done using the glht function (multcomp, Hothorn et al., 2008) comparing contrasts with the Tukey's range test and the false discovery rate method was applied to correct p-values for multiple comparisons.

#### RESULTS

#### Seedling Dimensions

Seedling dimensions (above-ground dry weight and height) varied across provenances and acclimation treatments (**Table 2**; TABLE 3 | Analysis-of-variance table evaluating the set of covariates included in final Cox proportional hazards regression models using partial-likelihood ratio test.


Bold values indicate significance at a level of 0.05.

see also Supplementary Tables S1 and S2 for detailed summary output of models). In the mortality experiment in 2014, the provenance with the largest mean above-ground dry biomass was the Alpenkiefer from Germany with 34.82 g, being significantly heavier than the provenances Wallis, Hauptsmoorwald, Alto Ebro, Montes Universales, Mont Ventoux, Prealpes du Sud, and Suprasl. The provenance from Wallis had the lowest above-ground dry weight (22.39 g) and was significantly different from all other provenances except Alto Ebro, Montes Universales, and Prealpes du Sud. Significant differences across provenances with intermediate weight could be also found (**Table 2**). Both provenances from Spain (Alto Ebro, Montes Universales) had a lower above-ground biomass than Garmen, Plantage Pornoapati, Mittel-/Ostdt. Tiefland, and Emilia Romagna. Additionally, Montes Universales was lighter than Mont Ventoux and Suprasl, and Plantage Pornoapati was heavier than Prealpes du Sud.

Provenances used in the experiment in 2015 differed significantly in mean above-ground dry weight and mean height. The seedlings of Alpenkiefer (182.23 g) were heavier than the pines of Suprasl (153.88 g) and Mont Ventoux (133.93 g). The smallest mean height was scored for the provenance Mont Ventoux (734.84 mm) being significantly smaller than Alpenkiefer (956.19 mm) and Suprasl (1030.85 mm). Regarding the acclimation treatments (**Table 2**) no effect of the seasonal drought events in 2013 could be detected neither on mean above-ground biomass at the end of the experiment in August 2015 nor on seedling height. The drought in 2014 significantly affected mean above-ground dry biomass (190.65 g vs. 131.49 g) and mean seedling height (991.07 vs. 858.9 mm). Pines grown in the vegetation hall from 2012 to 2014 were significantly heavier (173.10 vs. 141.87 g) and taller (1014.23 vs. 826.05 mm) than those grown in the greenhouse.

## Mortality 2014

of death).

#### Mortality Rates and Influence of Above-Ground Dry Weight

The first mortality events of Scots pine seedling were observed on the 54th day (second observation day) of the 97-day drought treatment in 2014 and most diebacks (415 out of 760) were detected on the 62nd day (third observation day) after the drought started (**Figure 2A**, Supplementary Figure S4A). Total mortality rates of the 12 provenances used in the experiment (**Table 2**) ranged between 81% (Emilia Romagna, I) and 98% (Mont Ventoux, F; Alpenkiefer, D; Plantage Pornoapati, HU; and Wallis, CH). Variation in mortality rates of seedlings in 2014 could be explained by above-ground dry weight and provenances (**Table 3**, see Supplementary Table S3 for a detailed model summary). Above-ground dry weight significantly increased the HR by 1.7%/g (p < 0.001, Supplementary Table S4, **Figure 2B**), thus the hazard of the individual with the highest weight (57.86 g) had an almost 96% higher hazard of mortality than the individual with the lowest weight (1.49 g).

#### Provenance Differences in Mortality Hazard Rates

Provenances differed significantly in their hazard rate for mortality due to the severe drought event in the experimental setup (**Figures 2A** and **3**, Supplementary Table S4). The provenance from Emilia Romagna (Italy) showed significantly lower hazard rates than all other provenances (p < 0.05) with the exception of the provenances Prealpes de Sud (France) and Garmen (Bulgaria), which were the second and third most drought resistant provenances in terms of mortality, respectively. Almost each provenance showed higher hazards than those three, except Mont Ventoux (France) and Alto Ebro (Spain) that did not indicate a higher mortality risk than the provenance Garmen (Bulgaria). Additionally, the provenance Mittel-/Ostdt. Tiefland (Germany) was not at higher risk than the provenances Prealpes de Sud (France) and Garmen (Bulgaria). The most vulnerable provenance in respect to the mortality hazard caused by severe drought was Alpenkiefer (Germany) that was at greater risk than Prealpes de Sud, Mont Ventoux (both France), Alto Ebro, Montes Universales (both Spain), Emilia Romagna (Italy), Garmen (Bulgaria), and Mittel-/Ostdt. Tiefland (Germany).

#### Mortality 2015 Mortality Rates and Influence of Above-Ground Dimensions

The drought mortality experiment in 2015 lasted 102 days. The first dead individual was observed on day 26 after drought initiation and most mortality events were observed on the 54th day of the drought treatment when 69 out of 139 individuals were recorded as dead (Supplementary Figure S4B). Total mortality rates (**Table 2**) were almost 100% in all three provenances (Alpenkiefer, D; Mont Ventoux, F; and Suprasl, PL) and for all acclimation groups (drought treatment 2013, drought treatment 2014 and building). Differences in the drought mortality hazard were explained by acclimation induced by the drought treatment in 2014 (p < 0.001, **Table 3**) as well as the interaction between provenance and building induced by growing seedlings in the two different buildings, thus temperature regimes, in the so called factorial model (p < 0.05, **Table 3**, see Supplementary Table S5 for a detailed model summary). Adding the covariates aboveground dry weight and height in the so-called continuous model increased model concordance from 0.74 to 0.81 (Supplementary Tables S5 and S6, respectively). Both covariates exhibited a significant influence on the HR (p < 0.01, **Table 3**, see Supplementary Table S6 for a detailed model summary). The HR increased by 0.8%/g above-ground dry weight and by 0.2%/ml height (**Table 4**, **Figures 4C,D**), thus resulting in a doubling of mortality hazard (increase by 199.5%) between the lightest (49.28 g) and the heaviest individual (298.64 g) and in a more than doubling of mortality hazard (increase by 237.4%) between the smallest (305 mm) and the tallest individual (1492 mm).

#### Variation in Mortality Hazard Rates with Acclimation and Provenance

Although drought mortality was almost 100% at the end of the lethal drought experiment in 2015 there were still differences in the mortality hazard among acclimation groups and provenances since the time span till maximum mortality was observed varied

between 75 and 102 days (**Figure 4**). The drought treatment in 2014 reduced the mortality hazard by 57.4% in the factorial model and by 59.6% in continuous model compared to the control (**Table 4**, **Figure 4A**). Variation in mortality hazard between provenances depended on the building (temperature regime) in which they were grown till December 2014. Whereas HR did not differ between provenances in the vegetation hall, they did in the greenhouse (**Table 4**, **Figure 4B**). The mortality hazard for Alpenkiefer (Germany) is 61 and 49.6% significantly higher (p < 0.05) than for Mont Ventoux (France) and Suprasl (Poland), as estimated with the factorial model. Mont Ventoux (France) and Suprasl (Poland) were not different in their mortality hazard. Adding the covariates above-ground dry mass and seedling height increased the difference in hazard between Alpenkiefer and Mont Ventoux to 67.6% and diminished the hazard between Alpenkiefer and Suprasl to 46.6%, that was not significant anymore at the 5% level (p = 0.081). Growing in the vegetation hall till December 2014 increased the hazard of mortality in Mont Ventoux and Suprasl (p < 0.01), but not in Alpenkiefer compared to the hazard experienced in the greenhouse (**Table 4**, **Figure 4B**). Hazard in the vegetation hall was 188.1 and 277.7% higher for Mont Ventoux compared to the greenhouse as indicated by the factorial and continuous model, respectively. Suprasl showed a 177.6 and 138.9% higher mortality hazard in the vegetation hall compared to the greenhouse.

#### DISCUSSION

Drought mortality varied fundamentally across provenances which was to our knowledge not reported in literature before. Above-ground dimensions had a significant impact on the hazard of drought-induced mortality. Taller and heavier individuals died earlier than individuals which were smaller and had lower weights. Thus, there seems to be a trade-off between growth and drought survival (Alía et al., 2001; Bennett et al., 2015). Acclimating seedlings by drought in the year before the actual drought mortality experiment considerably lowered the mortality hazard. In the further paragraphs, we discuss the link between provenance and above-ground dimensions as well as the provenance and acclimation effects on the mortality hazards.

## Variation of Seedling Dimensions among Provenances and Acclimation Treatments

Seedling dimensions (above-ground dry weight and seedling height) showed considerable variation among provenances and acclimation treatments. Differences in dry weight were consistent for provenances that were monitored in both mortality experiments in 2014 and 2015. Variation in biomass and height across provenances of P. sylvestris is well documented in literature. Oleksyn et al. (1992, 1998, 1999, 2000) showed that above-ground biomass and height of seedlings and adult trees had a hump-shaped relationship with the latitude of their origin, central provenances (45–55◦N) having higher aboveground biomass and height than the southern (<45◦N) and northern provenances (>55◦N). This pattern is similar for the latitudinal range (40–53◦N) in our study at least for aboveground dry weight, however, it has to be taken into account that the provenance Alpenkiefer is from a seed orchard. Oleksyn et al. (1999) attributed their findings to reduced day length and climate transfer distance in northern provenances and to


TABLE 4 | Estimated main and interaction effects of the survival model including factorial covariates only (factorial model) and factorial with continuous covariates (continuous model) on hazard of mortality in the experiment 2015.

The hazard ratio (HR) shows the probability of a death event in the treatment group in relation to the probability in the reference group or the change of hazard in relation to the change in continuous covariates per one unit (height, weight). Provenance abbreviations are D7 (Alpenkiefer, D), PL9 (Suprasl, PL), and F12 (Mont Ventoux, F). Group differences were calculated comparing contrasts with the Tukey's range test.

Bold values indicate significance at a level of 0.05.

genetic adaptation to warm and arid environments in southern provenances.

Two of the acclimation treatments (drought treatment 2014, warming treatment building) had a significant impact on seedling dimensions in the mortality experiment 2015. Spring drought in 2014 reduced the seedlings' above-ground dimensions. Water stress is known to exert a negative influence on the rate of cell wall division and cell expansion (Hsiao and Acevedo, 1974). This is a direct effect of reduced turgor pressure (Hsiao and Acevedo, 1974) or an indirect effect of the suppression by growth regulators such as ABA (Peñuelas et al., 2013). Furthermore, Scots pine increases root–shoot ratio under dry conditions (Richter et al., 2012; Taeger et al., 2015), which might be caused by lower investment in above-ground structures (Alía et al., 2001; Taeger et al., 2013b, 2015).

Seedling above-ground dimensions were smaller when grown in the greenhouse under elevated temperatures from 2012 to 2015. Reich and Oleksyn (2008) reported that height growth of southern Scots pine provenances was negatively correlated with the absolute temperature difference between planting site and origin. Since all the provenances in our study belong to the geographical group which Reich and Oleksyn (2008) classified as southern provenances (≤53◦N) this may explain the finding of decreased seedling height at elevated temperature in the greenhouse and might also contribute to the lower above-ground biomass. Additionally, decreasing radial growth of P. sylvestris with increasing temperature (Martínez-Vilalta et al., 2008; Michelot et al., 2012) could also contribute to lower aboveground biomass in the greenhouse compared to the vegetation hall.

The seasonal drought treatments (spring and/or summer drought) in 2013 did not affect dimensions measured in 2015. Most likely, effects induced by the drought treatment in 2013 were diluted since potting seedlings into much larger pots at the beginning of March 2014 caused an increase in overall mean above-ground dry weight by a factor of almost 6 (27.0 g in 2014 experiment, 156.4 g in 2015 experiment).

Nevertheless, the reduction of above-ground biomass might be an adaptation linked to drought tolerance, either across provenances or caused by acclimation (temperature, water availability).The reduction of above-ground biomass is linked to a decrease of foliage (Xiao and Ceulemans, 2004; Jagodzinski and ´ Kałucka, 2008) which is reducing evaporative water loss (DeLucia et al., 2000). Future work should also consider molecular and physiological acclimation pathways which might be associated with provenance.

#### Mortality Hazard Rates

Meteorological conditions during the 2015 mortality experiment were more severe than in the 2014 experiment, thus considerable first die-backs were observed after ∼60 and ∼40 days of completely withholding water, respectively.

It is difficult to disentangle acclimation effects (drought treatment 2014 and growing pines under different temperature

regimes) as well as provenance effects on mortality hazard rates from pure dimension effects (above-ground biomass and height) because both acclimation treatments and provenance influenced above-ground dimensions (**Table 2**, Supplementary Tables S1 and S2). We performed all survival modeling twice, with absolute dimensions (not reported in the paper) and with dimensions centered on respective means of acclimation treatments and provenances. In the latter case dimension effects were merely effects on the deviations from any provenance effect, and thus independent thereof. Since both attempts led to similar results, we conclude that (1) besides provenance and acclimation sheer above-ground dry weight (2014 experiment) and weight and height (2015 experiment) variations within provenances and treatments influenced mortality hazard (as presented in the Section "Result") and that (2) the provenance induced dimension differences were not the only provenance-treatment effects, as even by accounting for absolute differences in dimensions, there was still an additional provenance/acclimation effect, likely linked to physiological, wood anatomical differences, and/or adjustments.

#### Provenance Effects on Mortality Hazard Rates

Drought mortality was fundamentally different across provenances in both experiments, thus fully supporting our first hypothesis. This finding is novel since Cregg and Zhang (2001) could not find any variation in mortality rates of P. sylvestris seedlings from several Eastern European and Central Asian origins.

Opposite to our expectations, we could not identify any clear continuous relationship between provenance survival and the climate at their origin, such as mean annual temperature and mean annual precipitation sum (**Table 1**, data of correlation analysis not shown). However, the three provenances withstanding drought mortality the most (Emilia Romagna, Prealpes du Sud, Garmen) originate from locations with a pronounced precipitation minimum in summer, whereas the three provenances which were at the highest risk of drought mortality (Alpenkiefer, Hauptsmoorwald, Wallis) experience relatively high precipitation at their origin throughout the year (Supplementary Figure S1). F. sylvatica provenances originating from locations with summer drought had lower mortality rates under experimental induced extreme drought than provenances from wetter locations (Thiel et al., 2014). In contrast to their study, all the Scots pine provenances used in our study are from its south-western distribution range, which might blur an obvious drought survival– climate relationship. Mortality rates in P. ponderosa could not been related to climatic variables either (Cregg, 1994).

Rather to show a distinct pattern between mortality and climate, Ponderosa pine seems to have an optimum root– shoot ratio balancing water accessibility and photosynthetic capacity.

#### Trade-Off between Growth and Drought Survival

The hazard of mortality significantly varied with above-ground dimensions, a fact that was not explicitly mentioned in our hypotheses but was subsumed in the variable provenance. The mortality hazard increased with above ground dimensions suggesting a trade-off between growth and drought survival. In dry years, Alía et al. (2001) observed a lower branch dieoff on slow growing provenances than on faster growing ones. On a global scale, larger trees have higher mortality rates than smaller ones (Bennett et al., 2015). Since taller trees face greater challenges to lift water along height against gravity and conduit resistance, they are at higher risk of hydraulic dysfunction than smaller trees (McDowell and Allen, 2015). In neotropical savannas, larger trees lose 50% of hydraulic conductivity at higher values of water potential than smaller tress (Zhang et al., 2009). Taller trees counteract this hydraulic limitation by reducing stomatal conductance and leaf specific hydraulic conductance (Ryan et al., 2006) and since needle area is positively correlated with tree height and above-ground biomass (Xiao and Ceulemans, 2004; Jagodzinski and Kałucka, 2008 ´ ), this increases the hazard for hydraulic failure of larger trees (McDowell and Allen, 2015). Additionally, individuals with a larger transpiring needle area might exploit water resources faster than smaller individuals and thus die earlier under severe drought conditions.

#### Acclimation Reduces the Hazard of Drought Induced Mortality

Although we did not measure acclimation directly we suppose that exposure time to the treatments was sufficient since thermal acclimation can occur within hours to few days (Mäkelä et al., 2004; Lee et al., 2005). In addition, we could observe a change of above-ground dimensions caused by temperature and drought acclimation treatments in our study.

The drought acclimation treatment in 2014 significantly decreased mortality hazard rates as suggested in our second hypothesis. However, acclimation effects of 2013, two years prior to the mortality treatment, could not be identified any more, most likely due to the re-potting, significantly increasing seedlings' size. It might be worth to redo such a multiple acclimation mortality experiment without this confounding effect. Our finding of an acclimation effect in drought-induced mortality is in accordance to drought preconditioned Ponderosa pine seedlings which survived 2 weeks longer during drought than untreated seedlings and might be related to reduced gas exchange caused by stomatal control of water loss (Cregg, 1994).

It is known from literature that drought reduces needle area, decreases leaf/sapwood area ratio, and/or leads to smaller conduit sizes in Scots pine (Sterck et al., 2008; Martínez-Vilalta et al., 2009). Adjustments like these, might have favored higher survival rates under severe drought conditions in our study since they support avoidance of critical water potentials: Smaller leaf area results in lower transpiration rates reducing soil water exploitation (Verbeeck et al., 2007), decrease of leaf/sapwood area ratio results in total leaf area to be supported by a relatively larger area of water conducting tracheids (Mencuccini and Grace, 1995), and smaller conduit sizes increase resistance to cavitation (Sterck et al., 2012). Lower above-ground biomass and height of drought acclimated individuals in our study might also be linked to lower total needle area and thus preserves water transport and soil water resources by decreasing the risk of hydraulic failure and lower transpiration rates respectively. A drought induced decrease in leaf/sapwood area might be reflected by lower aboveground biomass in our study possible driven by lower needle biomass resulting in smaller total needle area.

The provenance-specific temperature effect on the hazard of mortality might be driven by changes in dimensions since variations in above-ground biomass follow the same pattern as hazard of mortality between provenances and buildings (data not shown) when only three provenances were considered in the mortality experiment 2015. As discussed above, elevated temperatures decrease height and diameter growth in Scots pine (Martínez-Vilalta et al., 2008; Reich and Oleksyn, 2008; Michelot et al., 2012). Additionally, high temperatures increase VPD, which was equally observed in our study. An increase of VPD alters biomass allocation reducing relative investment in leaf biomass (DeLucia et al., 2000), which in turn might decrease the hazard of mortality by a reduction of the transpiring surface.

However, it is also reported in the literature that exposure of seedlings to high temperatures can increase thermal tolerance caused by expression of heat shock proteins (Colombo and Timmer, 1992). Unfortunately little is known about their "lifetime" and thus it is unclear whether pines grown in the greenhouse and the vegetation hall from 2012 to 2014 still expressed different levels of heat shock proteins after growing exclusively in the greenhouse from December 11, 2014 till the start of the mortality experiment on May 22, 2015. The fact that in our study provenance effects were only apparent in the high temperature regime might suggest some association with heat shock proteins, however, this needs to be thoroughly tested in a follow-up study.

## CONCLUSION

Above-ground dimensions were the main determinants of seedling mortality in Scots pine. Conversely, the pattern of provenances' specific hazards for drought mortality did not follow the pattern of above-ground biomass when 12 provenances were studied. Thus, also adaptations to local climate and genetic specific control of water relations (e.g., transpiration and stomatal conductance), which was observed in P. ponderosa and Pinus taeda (Seiler and Johnson, 1988; Monson and Grant, 1989), might play a dominating role in resistance to drought mortality. Our study revealed a clear acclimation potential of Scots Pine seedlings, since drought episodes and warmer temperatures increased their survival time

under repeated stresses. This feature might facilitate Scots pine forest persistence under future climate change in the long run.

#### AUTHOR CONTRIBUTIONS

HS collected data, contributed to the experimental design, analyzed and interpreted the data, and wrote the paper. AM contributed to the conception of the work and experimental design, interpreted the data, and wrote the paper.

#### FUNDING

This work was financed by the European Research Council under the European Union Seventh Framework Programme (FP7/2007–2013/ERC grant agreement No. 282250).

#### REFERENCES


#### ACKNOWLEDGMENTS

We thank Ricardo Acevedo-Cabra for discussion, Charles Mellert for provision of WorldClim data, and especially Michael Matiu for discussion and comments on the manuscript, Gerda Benner, Philipp Falk, Elena Walter, and Stefanie Weindler for help with mortality assessment and Steffen Taeger, Andreas Ludwig (BaySF) and the ASP/Teisendorf for providing plant and seed material. We further thank the team of the GHL Dürnast for handling plant material.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.01014



stress interactions, tolerance and acclimation. For. Ecol. Manage. 260, 1623– 1639. doi: 10.1016/j.foreco.2010.07.054



interspecific trends in eleven species. Plant Biol. 8, 688–697. doi: 10.1055/s-2006-924107


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Seidel and Menzel. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Genetic Structure of a Naturally Regenerating Post-Fire Seedling Population: Pinus halepensis As a Case Study

#### Anna Gershberg<sup>1</sup> , Gidi Ne'eman<sup>2</sup> and Rachel Ben-Shlomo<sup>2</sup> \*

<sup>1</sup> Department of Evolutionary and Environmental Biology, Faculty of Natural Sciences, University of Haifa, Haifa, Israel, <sup>2</sup> Department of Biology and Environment, Faculty of Natural Sciences, University of Haifa-Oranim, Tivon, Israel

To study the effects of wildfire on population genetics of a wind pollinated and wind dispersed tree, we have analyzed the genetic structure of a post-fire, naturally regenerating seedling population of Pinus halepensis Miller, on Mt. Carmel, Israel. We tested the existence of spatial genetic structure, which is expected due to the special spatial demographic structure of the post-fire seedling and sapling populations of this species. Explicitly, we asked whether or not seedlings that germinated under large, burned, dead pine trees are also their offspring. The results revealed that the post-fire seedling population is polymorphic, diverse, and reflects the pre-fire random mating system. In contrast to our prediction, we found no division of the post-fire seedling population to distinct sub-populations. Furthermore, as a result of post-fire seed dispersal to longer range than the average pre-fire inter-tree distance, seedlings found under individual burned trees were not necessarily their sole offspring. Although the population as a whole showed a Hardy-Weinberg equilibrium, significant excess of heterozygotes was found within each tallest seedlings group growing under single, large, burned pine trees. Our finding indicates the possible existence of intense natural selection for the most vigorous heterozygous genotypes that are best adapted to the special post-fire regeneration niche, which is the thick ash bed under large, dead, pine trees.

#### Edited by:

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### Reviewed by:

Eryuan Liang, Chinese Academy of Sciences, China Meltem Alper, Aksaray University, Turkey

> \*Correspondence: Rachel Ben-Shlomo ekly@research.haifa.ac.il

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 27 October 2015 Accepted: 11 April 2016 Published: 27 April 2016

#### Citation:

Gershberg A, Ne'eman G and Ben-Shlomo R (2016) Genetic Structure of a Naturally Regenerating Post-Fire Seedling Population: Pinus halepensis As a Case Study. Front. Plant Sci. 7:549. doi: 10.3389/fpls.2016.00549 Keywords: disturbance, fire, natural selection, pine, spatial demographic structure, spatial genetic structure

## INTRODUCTION

Wildfires are a recurring disturbance in the Mediterranean basin, and a major factor affecting the ecology and evolution of plants in Mediterranean-type ecosystems (Naveh, 1994; Keeley et al., 2011). More than 10<sup>6</sup> ha of forests were consumed by fires in the southern EU during 1980–2000 (European Commission, 2001). In Israel, about 13,600 ha of natural maquis and planted pine forests were consumed by fire during 1980–1991 (Kleiot and Keider, 1992); in December 2010, one fire consumed about 2500 ha covered mainly by pine forests (Perevolotsky et al., 2011).

Pinus halepensis Miller (Aleppo pine) is a main constituent of Mediterranean lowland forests (Barbéro et al., 1998; Quézel, 2000). It is primarily a western Mediterranean tree (Mirov, 1967), with some small, native and disjunct populations in Israel (Barbéro et al., 1998). The Israeli natural populations of P. halepensis differ in their genetic composition from all the other native populations, and were defined as the eastern variety (Korol et al., 2002).

Aleppo pine is a highly flammable tree that is killed by fires, consequently it is an obligate seeder whose post-fire populations depend solely upon seed germination (Ne'eman and Trabaud, 2000; Pausas, 2015). Post-fire germination of pine seeds consists mainly of seeds originating from serotinous cones that comprise a canopy stored seed bank, which release their seeds primarily after crown fires (Lamont et al., 1991; Nathan et al., 1999). Young P. halepensis trees have larger proportion of serotinous cones than older trees, which reduced the risk of being burned before establishing a large enough canopy stored seed bank (juvenility risk); and post-fire established populations have also higher degree of serotiny than those established in the absence of fire (Ne'eman et al., 2004). Aleppo pine populations that face repeated fire episodes showed finer-scale spatial aggregation of serotiny relative to those residing lower fire recurrences areas (Hernandez-Serrano et al., 2013). Seeds from serotinous cones are better adapted for post-fire germination than those released from regular cones (Goubitz et al., 2003).

The fine scale spatial distribution of large Aleppo pine trees that are burned by canopy fires determine the spatial structure of the post-fire pine seedlings generation. The extremely low pH of the thick ash layer, deposited during the fire under burned pine canopies, inhibits the germination of many herbaceous and woody species but less so of Aleppo pine seeds (Henig-Sever et al., 1996; Ne'eman and Izhaki, 1999; Eshel et al., 2000; Izhaki et al., 2000). Consequently, pine seedlings that grow under large burned pine canopies experience lower interspecific competition, grow faster, and have a higher probability to comprise the post-fire pine forest generation than those growing elsewhere. Therefore, the spatial demographic pattern of post-fire forest is similar to the pre-fire forest (Izhaki et al., 1992; Ne'eman et al., 1995; Ne'eman and Izhaki, 1998; Nathan and Ne'eman, 2004).

Most (97%) of P. halepensis seeds are dispersed by wind over relatively short distances of up to 20 m from their mother trees, and 72% fall under their canopies (Nathan, 1999; Nathan et al., 2000; Nathan and Ne'eman, 2000). The pre- and post-fire demographic patterns are comparable. In the absence of data regarding post-fire seed dispersal, and under the assumption that the pattern is similar to dispersal with no fire, we can hypothesize that most of the post-fire dispersed seeds find their preferred regeneration niche under the canopy of their mother trees.

Pines reproduce only sexually via seed germination. Conifers including pines have not been reported for self-incompatibility (Hagman, 1975). Facultative selfing was found in pines (Reviewed by Ledig, 1998): medium inbreeding levels were reported for P. pinaster (De-Lucas et al., 2009), low levels for P. strobus (Marquardt and Epperson, 2004) and no inbreeding was found in P. brutia (Panetsos et al., 1998).

Fine scale spatial genetic structure (FSSGS) is the nonrandom spatial distribution of genotypes and alleles, which commonly results from fine-scale aggregation of siblings within a population (Wells and Young, 2002; Vekemans and Hardy, 2004). FSSGS of reproductive P. halepensis individual trees was empirically studied over time in an expanding native population; in early stages the genotypes were randomly distributed in space, but over time, FSSGS developed through increased genetic clustering with increasing density (Troupin et al., 2006). FSSGS was explained by fine-scale environmental heterogeneity and possibly by microenvironmental selection, inbreeding, and individual variation in the reproductive success of trees (Troupin et al., 2006).

Shohami and Nathan (2014), studying recently Mt. Carmel P. halepensis population (the same area as in this study), reported a limited pre-fire pollen dispersal and consequently, significant genetic structure and high kinship. Hence, the post-fire open landscape permits pollen gene flow over greater distances.

Disentangling the effects of fire on spatial genetic structure of post-fire regenerating seedling populations is an important challenge that has not yet been addressed. In this research we aimed to study the effects of fire on the genetic composition, variability and spatial genetic structure of the first naturally regenerating post-fire seedling population of P. halepensis, as a model for trees with wind dispersed seeds and pollen. Specifically we: (1) examined the genetic structure and variability of the first regenerating post-fire pine seedling population and compared it to near non-burned adult pine stands; (2) checked whether there was a spatial genetic structure in this regenerating seedling population, and (3) determined whether the tallest seedlings growing under large burned trees are also their descendant siblings. We hypothesized that: (1) The genetic structure and variability of the first post-fire seedling population is similar to that of unburned populations in the same geographical region (Mt. Carmel, Israel). (2) There is a spatial genetic structure in the regenerating post-fire seedlings population; namely, the variation within the seedlings groups growing under a dead tree will be smaller than that among the seedling groups growing under different dead trees. (3) The seedlings growing under a burned tree are also its descendant siblings.

## METHODS

## Study Site

The study site was located within a native P. halepensis forest area in Lubim on Mt. Carmel, Israel (E35◦ 00′ N32◦ 44′ ). This site was an abandoned field that has been colonized by native pines about 60 years ago. The climate is typical east Mediterranean one with mean annual rainfall of 700 mm that falls mainly from December to February, and a long, hot and dry summer from May to October. The forest was completely burned (i.e., all the large mature trees did not survive the fire) in autumn 2006 and was left for natural regeneration. Prior to the fire in 2006, the multiages stand (ca 140 ha), was comprised of about 150 large trees; in a distance of more than 1Km from any other pine population. The site has not been under fire prior to 2006.

## Sampling Method

After the 2006 fire, within the 140 ha site, we have selected, numbered and mapped 20 large dead trees with canopy diameter of 5 to 10 m. The canopies of the chosen trees were at least 6 m apart (up to 18 m between any two adjacent trees, average 13 m between neighboring trees; **Figure 1**). In the spring 2008, one and a half years after the fire, we sampled the 10 tallest seedlings (out of tens to several hundreds) growing up to 2 m radius around the trunks of all selected burned trees. This non-random selection

was done as we wanted to sample the seedlings that had the highest probability to replace the burned tree. These groups of 10 seedlings under any given tree will be referred hereafter as a seedling "group"—the number of each group is that of the burned tree. Ten to fifteen fresh needles were sampled from each seedling in all groups and were kept separately in an icebox until transferred to the lab and stored at −20◦C.

## Molecular Analysis

Frozen needles (20–80 mg) were crushed to powder using the TissueLyser device (QIAGEN). Lysis buffer AP1/RTL (QIAGEN) was added to the samples. DNA was extracted from the lysate using Biosprint 15 (QIAGEN; following BS15 DNA PLANT protocol). Nine known nuclear microsatellite loci were amplified by PCR (Table 1S; Keys et al., 2000; Mariette et al., 2001; González-Martínez et al., 2004; Guevara et al., 2005; Steinitz et al., 2011, 2012a). Each 15 µl PCR reaction contained template DNA (∼20 ng), 0.2µM of each of the primers and 7.5µl enzyme Taq Master Mix Purple 2x (Lamda Biotech, St. Louis, Missouri, USA). The forward primer of each microsatellite was labeled with florescent dye (6-Fam, Vic, Ned or Pet). To validate the integrity of amplification, we repeated genotyping for 10–20% of the samples for each locus.

Amplification products were separated using ABI 3130xl Florescence-Reader (Applied Biosystems). Manual scoring of PCR product size with reference to a 500-Liz standard marker (Applied Biosystems) was made using peak scanner version 1.0 (Applied Biosystems) and GeneMapper software version 4 (Applied Biosystems).

## Data Analysis

Unfortunately, we failed to amplify DNA from the burned trees, and thus could not genetically type them. Therefore, to estimate the minimal possible numbers of mothers for each seedling group, we compared the genetic structure of seedlings within each group, using the locus ITPH4516, which displayed 15 different alleles within the studied population. If a seedling group shared a mutual heterozygote mother tree, each seedling within the group should show either one of the two possible alleles in the ITPH4516 locus (e.g., if a mother is heterozygote A1A<sup>2</sup> in a locus, all her offspring should comprise of either A<sup>1</sup> or A2). If a group of seedlings presented more than such two mutual alleles, they should share more than one founder. Additional alleles can be contributed by various fathers. This test cannot confirm our "each group one mother" hypothesis, but it can negate it, if many seedling groups show a minimum of more than one mother that had contributed to their genetic structure.

To determine whether the sampling, which was not random, represents a homogenous population, we performed three virtual random sampling sessions of small data sets (N = 20) from the full data set (N = 181). We used GenAlEx 6.41 (Peakall and Smouse, 2006) to calculate genetic distance (D; Nei, 1978), as well as the genetic differentiation (FST based AMOVA) of the three sets and the whole data (altogether 4 sets). The four data sets did not differ from each other (D = 0.0; FST = 0; p = 0.4). Therefore, we considered our data set as representing the whole Lubim population. To substantiate that the Lubim population truly represents the Carmel native pine populations, we compared gene diversity parameters of Lubim to three additional populations (Mitla, burned in 1983, Antenna and Beit Oren (Figure 1S) with no evidence of previous fires). In each population we sampled minimum 30 large trees, at least 5 m apart. These four populations demonstrated similar gene diversity levels (Table 2S).

We scored the level of observed heterozygosity (Ho) and calculated the level of unbiased expected heterozygosity (He), inbreeding coefficient (FIS), Hardy-Weinberg equilibrium (HWE), and genetic distance (D), for the whole Lubim seedling population and separately for each seedling group. Data were analyzed by Tools for Population Genetic Analyses (TFPGA) V 1.3 (Miller, 1997) and by GenAlEx V 6.41 (Peakall and Smouse, 2006) softwares.

The significance level of seedling groups' differentiation (pairwise analysis of all groups; Exact tests—Raymond and Rousset, 1995) based on genetic distance between the groups (Nei, 1978), was determined after 1000 dememorization steps and 10 batches of 2000 permutations per batch, using TFPGA software version 1.3 (Miller, 1997). Bonferroni correction for significance level for that test was extremely strict (p < 0.0003). Considering the controversial use of Bonferroni correction (see Field, 2005), we also used a less stringent value for multiple test of p < 0.005. When using this significance level, less than 1 of 190 comparisons will show significant genetic difference by chance alone.

The correlation between genetic and geographical distances was tested by Mantel test, using GenAlEx6.41 software (Peakall and Smouse, 2006). The degree of the genetic difference between the seedlings within each group and among the groups was also tested by molecular analysis of variance (AMOVA) based on FST values (Wright, 1965) following Michalakis and Excoffier (1996; 999 permutations).

We also applied a Bayesian clustering method (STRUCTURE) that divided the samples into possible homogenous groups (sub populations) according to their degree of similarity (Pritchard et al., 2000, 2010). We checked clustering possibilities for K values from 1 to 20 (STRUCTURE 2.3.4 admixture model; LOCPRIOR option; burn-in of 100,000 steps and 100,000 iterations). The inference of the probable number of clusters was extracted by the log likelihood for each putative number of populations (K; four replicates for each K), Ln P(D) = L(K), and by the delta K method (Evanno et al., 2005), using the program Structure Harvester (Earl and vonHoldt, 2012).

To further compare genetic similarity of seedlings within and between groups, we selected the individuals that had genotypic data for minimum 8 loci (78 seedlings in total) for supplementary cluster analysis (UPGMA). The UPGMA cluster analysis assumed equal evolutionary rate along all branches which considered suitable for comparing regenerating seedlings groups within a population. Each of the 20 seedling groups had at least one individual in this analysis, and 17 groups had at least two seedlings.

## RESULTS

## The Genetic Structure of the Seedling Groups

All the analyzed microsatellite loci were polymorphic, ranging from two to 15 alleles, with an average (± SE) of 2.5 ± 0.1 alleles per locus per seedling group (**Table 1**). The observed heterozygosity (Ho) of the population (±SE) was 0.450 ± 0.080, and the unbiased expected heterozygosity (He) was 0.444 ± 0.0079. Fixation indices indicated that while inbreeding coefficient within individuals relative to the group (FIS) were generally negative (i.e., excess of heterozygotes); genotypic frequency within individuals relative to the population (FIT) was positive and close to 0 (0.017 ± 0.043; Table 3S). All nine loci exhibited Hardy-Weinberg equilibrium (HWE; Table 4S). Nonetheless, most of the post-fire seedling groups (16 out of 20) indicated within group negative FIS (i.e., excess of heterozygotes (**Table 1**); this situation is significantly different from random (p = 0.0118, two-tail Sign test). Heterozygote excess was observed for all microsatellites within each group (108 out of 147; p < 0.0001, two-tail Sign test; Table 5S).

The locus ITPH4516 exhibited the highest number of alleles (15), of which the most frequent were of 158 bp (29.6%) and 140 bp (26.2%). Eighteen of the 20 seedling groups presented four or more alleles in this locus, allowing assessment of maternal contribution. At least 11 seedlings groups, each growing under different burned pine tree, were comprised of seedlings that originated from two or more different maternal trees (Table 6S).

#### Genetic Differentiation among Groups

The genetic distances among the seedling groups were small ranging from 0.000 to 0.243 (**Table 2**). No significant pairwise differences between groups were detected by Exact test using strict Bonferroni correction of α < 0.0003; and only three significant pairwise differences between groups when less strict α < 0.005 (**Table 2**; lower-left triangle). Similarly genetic differentiation among the tested group was relatively low (FST = 0.144 ± 0.032, and the number of migrants between group, Nm = 1.786 ± 0.243; Table 7S). Mantel test indicated no correlation between geographic and genetic distances (rxy = 0.075; p = 0.32; R <sup>2</sup> = 0.0056; y = 0.0001x + 0.0625). Molecular analysis of variance (AMOVA) indicated that most of the variance (95%) was found within groups and only 5% originated from differences among groups (**Table 3**).

The Bayesian assignment test, STRUCTURE, showed a uniform distribution of the sampled seedling (Figure 2S); this distribution was consistent for all tested number of sub-populations (K) ranging from 2 to 20. Similarly, cluster analysis considering only individual seedlings that generated genotype data for at least 8 microsatellites (n = 78), also showed no indication of spatial genetic structure, or a tendency of individuals from the same group to be clustered together (UPGMA analysis, **Figure 2**), as would have been expected from genetically related individuals.



N, Sample size; Na, Average number of alleles per locus (±SE); P, polymorphism (% polymorphic loci); Ho, Observed heterozygosity (±SE); He, Unbiased expected heterozygosity (±SE); FIS, inbreeding coefficient.

## DISCUSSION

## Genetic Structure of the Post-Fire Pine Seedling Population

The effect of fire on the genetic structure of new post-fire populations are multifaceted, vary among ecosystems, and are associated with the ecology, mating system and post-fire demography of the focus species (reviewed by Steinitz et al., 2012b). In this study we analyzed the genetic structure of postfire seedling population of P. halepensis as a model for an obligate seeding tree with wind pollination and seed dispersal systems. The fire related traits of this species, post fire regeneration niche and demography are relatively well known (Ne'eman and Trabaud, 2000). The spatial pattern of the post-fire seedling, sapling and adult pines is largely determined by the spatial pattern of the large trees of pre-fire forest (Ne'eman and Izhaki, 1998). Because the distance for seed dispersal is limited and gene exchange is probably little, the post-fire genetic structure and diversity of the post-fire seedling generation is also expected to be similar to that of the pre-fire pine generation. Indeed, supporting our first hypothesis, the observed heterozygosity level (Ho = 0.45) of the post-fire seedling population at Lubim was similar to that found in other three nearby populations on Mt. Carmel, Israel (Table 2S). Using comparable microsatellite analyses, our studies revealed that the genetic diversity of the P. halepensis populations found on Mt. Carmel was higher than that found in other pine forests in Israel, where heterozygosity levels of 0.35 were found in natural unburned pine forests (Steinitz, 2010), and 0.2–0.25 in a naturally expanding P. halepensis population (Troupin et al., 2006).

The post-fire seedling population in Lubim exhibited Hardy-Weinberg equilibrium, indicating a random mating system in the pre-fire generation, and an inbreeding coefficient (FIS) that was not significantly different from zero, as was also found for other unburned populations in the Carmel (Table 2S). These results differ from previous findings in which several natural populations of P. halepensis in Israel indicated a certain level of inbreeding (i.e., positive FIS values; Troupin et al., 2006; Steinitz, 2010; Steinitz et al., 2012a). Moreover, a considerable selfing rate of 26–43% was recently found for the P. halepensis population on Mt. Carmel (using the same set of microsatellite; Shohami and Nathan, 2014).

Genetic differentiation between native populations at the adult-tree stage indicated a high level of differentiation (FST = 0.32) (Troupin et al., 2006; Steinitz et al., 2012a). Spatial genetic structure was detected in three post-fire (>30 years after fire) populations of P. clausa, but not in neighboring older populations (Parker et al., 2001). To the best of our knowledge spatial genetic structure was never examined before in post-fire regenerating seedling populations (i.e., 1–2 years after the fire). In contrast to our second hypothesis, we found no spatial structure within the post-fire tallest seedling population; no correlation was found between genetic and geographic distances (Mantel test). AMOVA results indicated that most of the variation in the genetic composition was due to genetic differences within the groups and not among them.



TABLE 3 | Partition of the molecular variance (AMOVA) among and within pine seedling groups.


Probability, P(rand ≥ data), for FST , FIS and FIT is based on permutation across the full data set.

In the absence of fire, P. brutia populations in Greece exhibited high within population genetic diversity compared to that of among populations (Panetsos et al., 1998), and similar partition of genetic diversity was found also in other tree species using allozyme diversity (Hamrick, 2004).

The distances between the seedlings in our study site ranged from few centimeters to about 170 m. Effective seed dispersal distances (without fire) of P. halepensis leading to germination and establishment of progeny is 20–70 m from the seed source tree, where most of the offspring did not germinate next to their mothers (Troupin et al., 2006). Parental analyses of a naturally expanding population of P. halepensis in the Judean Mountains, Israel, revealed longer effective seed dispersal distance of 6–492 m with an average of 42 m from the mother trees (Steinitz et al., 2011). Such effective seed dispersal distances can eliminate FSSGS formation in a scale of tens of meters.

Seed dispersal distance after fire should not be shorter than that in the absence of fire. The destruction of dense living pine canopies by fire creates two factors that have a positive effect on post-fire dispersal distances in pine samaras: wind speed in the burned forest at canopy height is higher than in the unburned forest, and the absence of the physical obstruction of a dense canopy increases the distance of seed dispersal (Shohami and Nathan, 2014). Once this distance equals the average distance between neighboring trees, it increases random distribution of seed genotypes on the burned forest floor and of the first postfire seedling generation. In such a case no FSSGS is expected, and this is probably also the explanation of our results.

#### The Genetic Structure of the Post-Fire Seedling Groups

Our results implied that most post-fire seedling groups, growing under the same burned tree, originated from more than one mother tree, and in contrast to our third hypothesis, we found no evidence that the dominant seedlings in the pine's favored regeneration niche are siblings, or have a higher rate of genetic similarity. Moreover, although the inbreeding coefficients of the entire post-fire seedling population were close to zero, an excess of heterozygotes and negative inbreeding coefficients in all tested microsatellite (i.e., Ho>He) were found in most of the postfire seedling groups, implying a general genomic phenomenon. Selective pressure that favors heterozygote individuals at the germination or establishment stages (Hufford and Hamrick,

eight individuals belonging to group #10.

2003) can explain the apparent excess of heterozygotes among the tallest dominant seedlings.

Our seedlings were not randomly sampled; we have collected the tallest saplings growing under each burned tree, assuming that they reflect advantageous and have high probability to replace the large burned tree. Our results suggest a heterozygote advantage of the post-fire seedling generation in the regeneration niche under the burned canopies of large pine trees. The high pH of the ash and its underlying soil create harsh conditions for the germination of seeds in their regeneration niche (Henig-Sever et al., 1996), which may cause natural selection to favor more vigorous heterozygous individuals. Alternatively, competition between siblings can be stronger than the competition between non-siblings while competing over the same regeneration niche. In such a case, even if most of the seeds that reach the ground under any given large burned pine tree are its offspring, the few that are not can have an advantage, and consequently become the largest saplings in that micro-site.

The wide-ranging excess of heterozygotes found in this research was of noncoding genetic markers (i.e., microsatellites), thus highlighting the general phenomenon of the entire genome of the post-fire seedling group. The strict criteria for outlining balancing selection or heterozygote advantage selection put forward by Hedrick (2012) cannot be applied to our findings, as we cannot specify the fitness of any specific allele or gene. Nonetheless, the significant excess of heterozygotes and their possible adaptive nature cannot be overlooked.

To conclude, the effects of fire on population genetic structure are complex. Our results indicate that the seedling groups growing in their favored regeneration niche under large burned trees are not entirely their sole offspring. The effective seed dispersal distances are probably larger than expected in the post-fire environment, overruling the expected spatial genetic structure. It is likely that intense natural selection pressure in the post-fire regeneration niche causes excess of heterozygous and more vigorous individuals. The mode of the selection is not yet clear, and further studies are needed for revealing its significance. Additional research is required also to further analyze the spatial genetic structure of the developing seedling generation.

#### REFERENCES


#### DATA ARCHIVING STATEMENT

Data available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.6r725

#### ACKNOWLEDGMENTS

We thank ISF for funding (grant number 150/07 to GN); The MOFET Institute and The Carmel Research Center of Haifa University for financial support and Avi Bar-Massada, Na'ama Tesler, Ofer Steinitz, and Yuval Eitan for their help.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00549


Nathan, R. (1999). Spatiotemporal Dynamics of Seed Dispersal by Wind in Aleppo Pine. Ph.D. thesis, The Hebrew University of Jerusalem, Israel.

Mirov, N. (1967). The Genus Pinus. New York, NY: The Ronald Press Company.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Gershberg, Ne'eman and Ben-Shlomo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Indirect Evidence for Genetic Differentiation in Vulnerability to Embolism in Pinus halepensis

Rakefet David-Schwartz<sup>1</sup> \*, Indira Paudel<sup>2</sup> , Maayan Mizrachi<sup>1</sup> , Sylvain Delzon<sup>3</sup> , Hervé Cochard<sup>4</sup> , Victor Lukyanov<sup>2</sup> , Eric Badel<sup>4</sup> , Gaelle Capdeville<sup>3</sup> , Galina Shklar<sup>1</sup> and Shabtai Cohen<sup>2</sup>

1 Institute of Plant Sciences, Volcani Center, Agricultural Research Organization, Rishon LeZion, Israel, <sup>2</sup> Institute of Soil, Water and Environmental Sciences, Volcani Center, Agricultural Research Organization, Rishon LeZion, Israel, <sup>3</sup> BIOGECO, INRA, Université de Bordeaux, Cestas, France, <sup>4</sup> PIAF, INRA, Université Clermont Auvergne, Clermont-Ferrand, France

Climate change is increasing mean temperatures and in the eastern Mediterranean is

#### Edited by:

Andreas Bolte, Johann Heinrich von Thünen-Institut, Germany

#### Reviewed by:

Christiane Werner, University of Freiburg, Germany Rodica Pena, Georg-August-Universität Göttingen, Germany

> \*Correspondence: Rakefet David-Schwartz rakefetd@agri.gov.il

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 23 February 2016 Accepted: 17 May 2016 Published: 02 June 2016

#### Citation:

David-Schwartz R, Paudel I, Mizrachi M, Delzon S, Cochard H, Lukyanov V, Badel E, Capdeville G, Shklar G and Cohen S (2016) Indirect Evidence for Genetic Differentiation in Vulnerability to Embolism in Pinus halepensis. Front. Plant Sci. 7:768. doi: 10.3389/fpls.2016.00768 expected to decrease annual precipitation. The resulting increase in aridity may be too rapid for adaptation of tree species unless their gene pool already possesses variation in drought resistance. Vulnerability to embolism, estimated by the pressure inducing 50% loss of xylem hydraulic conductivity (P50), is strongly associated with drought stress resistance in trees. Yet, previous studies on various tree species reported low intraspecific genetic variation for this trait, and therefore limited adaptive capacities to increasing aridity. Here we quantified differences in hydraulic efficiency (xylem hydraulic conductance) and safety (resistance to embolism) in four contrasting provenances of Pinus halepensis (Aleppo pine) in a provenance trial, which is indirect evidence for genetic differences. Results obtained with three techniques (bench dehydration, centrifugation and X-ray micro-CT) evidenced significant differentiation with similar ranking between provenances. Inter-provenance variation in P<sup>50</sup> correlated with pit anatomical properties (torus overlap and pit aperture size). These results suggest that adaptation of P. halepensis to xeric habitats has been accompanied by modifications of bordered pit function driven by variation in pit aperture. This study thus provides evidence that appropriate exploitation of provenance differences will allow continued forestry with P. halepensis in future climates of the Eastern Mediterranean.

Keywords: embolism, xylem hydraulics, provenance trial, genetic variation, border pit, torus-margo, water potential, xylem conductivity

## INTRODUCTION

Climate change, which is leading to increased mean temperatures and, in the Eastern Mediterranean is expected to decrease annual precipitation (Giorgi and Lionello, 2008), may be too rapid to allow adaptation of long lived forest trees, leading to changes in biomes in the near future (Seneviratne, 2012). In order to adapt to climate change, long lived forest tree populations will need genetic variability and/or phenotypic plasticity to survive and reproduce allowing the population to adapt to the new climate conditions. This statement is particularly important for the ability to withstand one to multi-year extreme events, which are already testing our forest species,

leading to forest dieback in many regions around the globe (Allen et al., 2010; Choat et al., 2012). It was previously suggested that a rapid climate change requires fast adaptation which relies on existing natural variability rather than on selection of new mutations (Savolainen, 2011). The above considerations have led forestry organizations to consider in situ selection of forest trees based on their ability to withstand drought and thrive in environments whose aridity matches that predicted for coming generations (Joyce and Rehfeldt, 2013).

Pinus halepensis is widespread in the Mediterranean basin and is one of the most drought-tolerant pine species (Ne'Eman and Trabaud, 2000; Maseyk et al., 2008; Klein et al., 2011, 2014a; Chambel et al., 2013). For that reason, it was selected as the main species for afforestation in semi-arid regions of Israel (Liphschitz and Biger, 2001), which now has the southernmost pine forest in the Mediterranean basin (Schiller, 2000; Rotenberg and Yakir, 2010). Recent increases in tree mortality following two drought periods suggest that the P. halepensis plantations are not fully adapted to withstand increasing aridity in the local climate (Dorman et al., 2013).

The fact that P. halepensis is spread over various subtropical dry summer to semi-arid climatic zones of the Mediterranean basin suggests that genetic differences exist between local populations (e.g., Schiller et al., 1986; Grivet et al., 2009), and there is considerable interest in finding the best genetic source to use in future plantations. To this end, provenance trials have been carried out at selected sites where seeds from various locations are sown together. These are essential in finding populations harboring desirable traits (White et al., 2007; Chambel et al., 2013). In the trials it is assumed that plant populations that are locally adapted will demonstrate genetic differences in fitnessrelated traits. Studies on drought resistance through provenance trials have been reported previously for several species including Pinus spp. where various parameters have been analyzed in order to determine adaptation to drought stress (Atzmon et al., 2004; Voltas et al., 2008; Eilmann et al., 2013; Gaspar et al., 2013).

Drought resistance is a complex polygenic trait that involves multiple mechanisms at different levels of tissue structure and function, and various tree species utilize different strategies to cope with water shortage (McDowell et al., 2008; Meinzer and McCulloh, 2013; Klein et al., 2014b; Delzon, 2015). Nevertheless, accumulating evidence suggests that drought resistance, in many cases, is well explained by resistance of xylem to embolism (Brodribb et al., 2010; Choat et al., 2012; Barigah et al., 2013; Urli et al., 2013; Delzon and Cochard, 2014). A recent study emphasized the crucial role of embolism resistance in those coniferous species that do not rely on abscisic acid to close stomata (Brodribb et al., 2014).

Conifer xylem consists of overlapping files of elongated narrow tracheids interconnected laterally by bordered pits. The pit and its torus-margo membrane allow efficient water flow, while preventing the spread of emboli by sealing the pit with the torus (i.e., pit aspiration). Early study on bordered pit structure (Sperry and Tyree, 1990) showed the correlation between embolism resistance and bordered pit structure. That study argued that torus flexibility, which was related to the pressure at which pits close (due to torus aspiration into the bordered pit), determines embolism resistance. Later studies suggested that both torus thickness and depth of the pit chamber correlate with greater vulnerability to embolism (Hacke and Jansen, 2009). Recently published articles support the hypothesis that xylem resistance to embolism is a major component of drought resistance in conifers, and suggest that the torus to pit aperture overlap is mechanically related to embolism resistance (Delzon et al., 2010; Bouche et al., 2014).

Due to their importance for drought adaptation, embolism resistance traits are natural candidates in genetic variation studies. It was previously hypothesized that populations from xeric environments would possess greater resistance to embolism than other populations within a species. Using technically advanced methods to measure vulnerability to embolism, it was found that P. sylvestris as well as the Mediterranean P. pinaster possess low inter-population genetic variation in resistance to embolism (Martínez-Vilalta et al., 2009; Corcuera et al., 2011; Lamy et al., 2011, 2014). A similar study that evaluated Mexican populations of P. hartwegii also demonstrated the lack of genetic variability in the embolism resistance trait (Sáenz-Romero et al., 2013). The only pine species that has shown a significant degree of among population genetic variability in embolism resistance so far is P. canariensis (López et al., 2013).

A recent study on P. halepensis provenances from Israel, Greece, Italy, and Algeria, in three provenance trials reported significant variation in branch hydraulic conductivity and native embolism (Klein et al., 2013). That study supported a previous study that showed higher survival rates of Greek and Israel as compared to Italian and Algerian provenances in semiarid field trials (Schiller and Atzmon, 2009). In the current study we hypothesized that provenances are genetically different in their hydraulic traits and that these differences are driven by variation in xylem structure. To test these hypotheses, we analyzed hydraulic and anatomical traits of P. halepensis in a local provenance trial.

#### MATERIALS AND METHODS

#### Plant Material

Pinus halepensis provenances used in this study were grown in a provenance trial at Bet Dagan, in the center of Israel, since 1991. Trees sampled were fully grown and 10–15 m tall. Bet Dagan, whose climate is Thermo-Mediterranean, is located in the coastal plain 20 km east of the Mediterranean Sea shore (31◦ 590N 34◦ 480E). The site is part of the UN FAO seed collection provenance program (SCM/CRFM/4 bis project<sup>1</sup> ). Four provenances were selected for the current research. These included Elea from Greece, Elkosh from Israel, Otricoli from Italy and Senalba from Algeria. Climatic conditions at the native location of the four provenances are indicated in **Table 1** (based on Klein et al., 2013). Mean annual and summer precipitation and approximate potential evapotranspiration at the provenance trial at Bet Dagan were 524, 0, and 1300 mm respectively. Monthly total precipitation and daily pan evaporation covering

<sup>1</sup>http://www.fao.org/docrep/006/k1203e/K1203E08.htm

TABLE 1 | Climate data for the four seed source provenances used in this study (based on Klein et al., 2013 and references therein).


P (mean annual precipitation), PET (mean annual potential evapo-transpiration) and Ps (mean summer precipitation) are in mm. Aridity index (P/PET).

the sampling period from September 2012 through May 2015 are shown in Supplementary Figure S1.

#### Specific Hydraulic Conductivity

Light-exposed lower branches containing regular stem sections (i.e., 'twigs,' ∼20 cm long, 5–8 mm diameter with 5–6 annual rings) were sampled in the morning for hydraulic measurements and several twigs with needle cohorts for measuring leaf water potential (9). 9 samples were immediately bagged and kept in a cooler during transport to the lab where water potential was measured with a pressure chamber (ARIMAD, MRC Ltd., Holon, Israel). Hydraulic samples were immediately put in an ice bath in the field and remained so during transport to the lab. Resin production was prevented by chilling in the ice bath for between 40 min and 1 h. That led to high conductivity values similar to those obtained by Klein et al. (2013), who boiled sample ends. Twigs were allowed to 'relax' for at least an hour before measurement. In the lab more than 2 cm was re-cut from each side of the twig under water and final twig length was about 10 cm. Since pine tracheid length is less than 1 cm this assured that tracheids that cavitated during cutting in the field were not included in the measurements.

Native specific hydraulic conductivity (Ks) and maximum specific conductivity (Ksmax) were measured under low pressure (7 KPa) generated by a 70 cm water head before and after overnight perfusion of the xylem tissue with a vacuum at a higher negative pressure of ∼0.06 MPa that drew degassed fluid into the samples from a closed container. The vacuum procedure was selected because perfusion at pressures greater than those of the vacuum led to reductions in conductivity, presumably due to pit aspiration. Since a large amount of degassed water was drawn through the stem overnight, we assume that embolisms were refilled, and in fact conductivity after perfusion was much greater. All measurements, including perfusion, were with 0.2 mM KCl solution which was degassed and filtered through Whatman no. 50 (retention of particle size > 2.7 µm) filter paper before use. Hydraulic measurements were made by connecting samples to 25 ml burettes with 0.05 ml resolution and accuracy, allowing measurements with a number of samples in parallel. Readings of the water volume entering the stems from the upstream burette were taken every 20 min to a half hour during which time the water level dropped by less than 4 cm, which we accounted for in the calculations. Water level was readjusted to 70 cm above the water entry point (i.e., the burettes were refilled using a syringe with a long needle) after each reading. Measurement continued for about 2 h until flow rates leveled off. Our protocol has been written up and submitted to the Prometheus Wiki website (not available yet). Mean sample stem diameter without bark, and length were measured and specific hydraulic conductivity K<sup>s</sup> and maximum specific conductivity Ksmax (kg m−<sup>1</sup> MPa−<sup>1</sup> s −1 ) were calculated after Sperry and Tyree (1988) assuming that all of the stem was conductive. Measurements of K<sup>s</sup> and Ksmax were further used to determine percent loss of conductivity (PLC) that can be attributed to xylem embolism according to Sperry and Tyree (1988). During bench dehydration at very low 9 (< −6 MPa), in some cases, Ksmax values were low due to insufficient perfusion. In these cases Ksmax measured at the beginning of measurement series when 9 was higher (> −1.5 MPa) was used.

## PLC Curves

Percent loss of conductivity curves were determined by three methods: bench dehydration (six individual trees within provenance), Cavitron technique (10 individual trees within provenance), and micro-Computed Tomography (micro-CT, five individuals from the Elkosh provenance). With the bench dehydration method (Tyree et al., 1992) using the burette protocol (see above), measurements were made in January, 2014, about a month after a large rainstorm (>100 mm) which saturated the soil. Branches were cut from the trees and these were allowed to dry in the lab or outdoors until they reached the desired needle cohort 9. Twigs were then cut from the branches as described above and their specific conductivity was measured. To obtain very low 9's, branches were left outside in the sun for several days. For bench dehydration with the micro-CT, samples were taken in May 2015 and samples were dried on a bench in the lab.

The Cavitron technique (Cochard et al., 2005) was used at the high-throughput phenotyping platform for hydraulic traits (Cavit\_Place, INRA-University of Bordeaux, Pessac, France). Branches from 10 individuals per provenance were sent in overnight mail to the above and used for vulnerability curve measurements. P<sup>50</sup> (MPa) was defined as the pressure corresponding to 50% PLC (Lamy et al., 2014). Slope (S), which corresponds to the speed of embolism spread, was defined as the slope (% MPa −1) of a tangent at the inflection point (P50) as previously described (Lamy et al., 2014).

The X-ray microtomography (micro-CT) technique is a noninvasive observation technique that allows the embolism to be directly visualized (Cochard et al., 2015). Samples were placed in an X-ray microtomograph (Nanotom 180 XS, GE, Wunstorf, Germany) at the PIAF laboratory of the Institut National de la Recherche Agronomique (INRA, Clermont-Ferrand, France) in order to visualize the hydric status of the xylem in different conditions. For one set of measurements, branches similar to the above were measured in the Cavitron and then scanned in the micro-CT system after each centrifugation step. The X-ray settings were adjusted in order to observe the whole cross-section of the middle of the samples with the best spatial resolution. Each scan provided 3D images from which we virtually extracted the central cross-section of the sample with a spatial resolution of 3.75 µm. The rate of embolism was measured by image analysis

using ImageJ software<sup>2</sup> . In a second set large branches, including a number of leaf cohorts, were dried by bench dehydration (see above). At different levels of dehydration 9 was measured in the pressure chamber and branches from the same branch were imaged with the micro-CT.

## Determination of Pit Closing Pressure with a High Pressure Flow Meter (HPFM)

At high water pressures tori of bordered pits are aspirated into the pit borders, thereby sealing the pit aperture and blocking water flow. This behavior has been documented previously for other conifers (Pappenheim, 1889; Sperry and Tyree, 1990). Using a HPFM (Tyree et al., 1995) and small branch sections we found that at low pressures flow reached a steady state and when pressure was increased gradually, at some point the resistance increased steeply, indicating pit closure. Reversing the direction of flow and applying low pressure resulted in a return to the original resistance, indicating that the torus moved out of the pit aperture and the pit opened, which confirms the previous statements. Utilizing the HPFM in this manner, we determined the torus aspiration pressure (or pit closure pressure), i.e., the pressure at which resistance begins to rapidly increase.

Branch sections were connected under water and the HPFM was operated in the steady state mode at a series of increasing low pressures, approximately 10 min per pressure, at intervals of about 0.01 MPa. The pressure at which resistance increased exponentially was taken as the pressure of pit closure (Pappenheim, 1889). In each case 6 samples (from six individual trees) were measured, each 10 cm long and 8–10 mm in diameter. We note that it was difficult to control the HPFM at these low pressures, and we broke a needle valve in the process.

#### Tracheid Width Measurement

For anatomical analyses, branch tissue used for hydraulic measurements was fixed in 70% ethanol before sectioning. Sections of 15 µm were prepared with a sliding microtome (Reichert Wien, Shandon, Scientific Company, London) and stained with Safranin O. Sectioned material was viewed under a Leica IM1000 microscope and digital images were taken using a CCD camera (model DC2000, Leica, Germany). Images were later analyzed to determine lumen width of early wood of the preceding year using ImageJ software. A microscopic ruler was used for size calibration. Tracheid width measurements were repeated on independent branches from the same trees used for hydraulic measurements. Two branches per tree and five trees from each provenance were sampled. Approximately 200 tracheids were measured for each tree so that for each provenance about 1000 tracheids were sampled.

#### Tracheid Length Measurement

Small segments (toothpick sized) of wood (the outer most second ring) were incubated in maceration solution composed of 1:4:5 of 30% hydrogen peroxide: distilled water: glacial acetic acid, for 3 days and washed five times in distilled water (Peterson et al., 2008). Tracheids were stained with Safranin O and mechanically dispersed before observed and photographed using the microscope. Length was measured using ImageJ software. A microscopic ruler was used for size calibration. Two branches per tree and five trees from each provenance were sampled. Approximately 200 tracheids were measured for each tree so that for each provenance about 1000 tracheids were sampled.

## Scanning Electron Microscopy (SEM)

Branch samples, with 5–6 annual rings, were collected from all provenances and incubated in 70% ethanol. The samples were split in half and small and thin longitudinal sections were cut with a razor blade. These were then oven dried overnight at 65◦C. Sections were mounted on aluminum stubs using double sided adhesive and coated with gold-palladium for 90 s at 20 mA using a sputter coater (SC7620 mini sputter coater, Quorum). All samples were observed with a field emission scanning electron microscope (SEM JCM-6000 benchtop scanning electron microscope, JEOL) with an accelerating voltage of 15 kV. Early wood inter-tracheid pit membranes where the pit aperture underneath the torus is clearly visible were photographed. The photos were analyzed to determine torus diameter and pit aperture area using ImageJ software. A minimum 24 pits per provenance were analyzed for torusaperture overlap [(torus diameter – pit aperture)/torus diameter] following Delzon et al. (2010).

#### Statistical Analyses

Results in this study were analyzed using JMP software (SAS Institutes, Inc., Cary, NC, USA). Variations among provenances in water conductivity, PLC curve parameters and xylem anatomical measurements were evaluated using a one-way analysis of variance (ANOVA) followed by Tukey's Honest Significant Difference (Tukey–Kramer HSD) test. Assessment of phenotypic variability for tracheid width and length was done with a nested ANOVA using the residual maximum likelihood (REML) method. In the nested ANOVA provenances were considered fixed effects and individual trees were nested within provenances as a random effect. Correlations between P50b values and anatomical features were tested with the Pearson correlation coefficient (r).

#### RESULTS

#### Differentiation in Hydraulic Traits

Hydraulic conductivity measurements were made in two seasons; at the end of the dry summer season (October 8, 2013) after 170 days with no precipitation and in the middle of the rainy season (January 15, 2014) after rain saturated the soil (Supplementary Figure S1). Native K<sup>s</sup> was lower at the end of the dry season than it was in the rainy season in all provenances (**Table 2**). In both seasons, Elea had the highest native K<sup>s</sup> , which was significantly higher than that of the Otricoli and Senalba provenances. Elkosh had intermediate native conductivity which did not differ significantly from the others (**Table 2**). No significant differences in maximum conductivity (Ksmax) were

<sup>2</sup>http://rsb.info.nih.gov/ij/

found among provenances at the end of the dry season, albeit a tendency for higher conductivity was evident in Elea and Elkosh as compared to Otricoli and Senalba (**Table 2**). Similar results of Ksmax were obtained with the Cavitron at the beginning of the rainy season of 2014 (with less than 20 mm precipitation, Supplementary Figure S1), after 145 days with no precipitation. A significant difference was found in the rainy season between the high K<sup>s</sup> max of Elea and the low K<sup>s</sup> max of Otricoli (**Table 2**).

Elea and Elkosh provenances had low percent loss of conductivity in both seasons (**Figure 1**). Elea had significantly lower PLC (16.1 ± 6.1 and 3.5 ± 1.2 % at the end of the dry and in the rainy seasons, respectively) than Senalba (37.7 ± 3.8 and 14.9 ± 2.5%) and Otricoli (38.2 ± 2.9 and 11.1 ± 2.5%). Elkosh had low PLC (27.2 ± 5.8 and 4.1 ± 1.0%), which was similar to Elea, but its PLC was not statistically different from Otricoli and Senalba in the dry season and from Otricoli in the rainy season (**Figure 1**).

Vulnerability curves were measured with the bench dehydration and Cavitron methods (Supplementary Figure S2). All methods and provenances showed similar shapes of vulnerability curves and data points were fit to a sigmoidal model (Pammenter and Van der Willigen, 1998). Notable in the curve fit for bench drying (Supplementary Figure S2, upper panel) is that for the lowest water potentials, −8 MPa, some conductivity remained, and extrapolated values for 100% loss of conductivity are very low, close to −10 MPa, which was the lower limit of the pressure chamber used for 9 measurements. For bench drying, Senalba and Otricoli provenances had the highest P50, −3.6 ± 0.04 and −3.7 ± 0.1 MPa, respectively, and Elkosh and Elea were lower, −4.2 ± 0.1 and −4.5 ± 0.1, respectively, indicating higher embolism resistance in Elkosh and Elea (**Table 3**).

Vulnerability curves measured by the Cavitron technique suggested a similar tendency of variation with more negative values. P<sup>50</sup> of Elkosh (−5.51 ± 0.39) was significantly lower than that of Senalba (−5.04 ± 0.24) and Otricoli (−5.08 ± 0.28) but not of Elea (−5.27 ± 0.34). P<sup>50</sup> of Elea was not significantly lower than that of Senalba or Otricoli (**Table 3**). Substantially lower P<sup>88</sup> was observed in Elkosh (−6.7 ± 0.6) as compared to the other three provenances, while differences in P<sup>12</sup> were small and not significant. Consequently, the slope of the vulnerability curves was significantly lower for Elkosh than for the other provenances (Supplementary Figure S2; **Table 3**).

Differences between the curves measured with the bench drying and Cavitron methods were large and significant. On average P12, P50, and P<sup>88</sup> values were 3.0, 1.2, and 0.5 MPa higher, respectively, for the bench drying method, and slopes for the Cavitron were 40%/MPa higher. Thus the largest discrepancy between the methods is in their estimate of the onset of the loss of conductivity at high xylem pressure.

Bench drying and the centrifuge technique were each used to bring branches from the Elkosh provenance to a given xylem pressure and then they were directly visualized via micro-CT technology (**Figure 3**; Supplementary Figure S3). Empty tracheids do not absorb x-rays and appear as dark spots on x-ray images, while water in the fully saturated tracheids appears gray. Thus, an image segmentation allowed to distinguish the embolized area from the conductive areas and to compute the rate of embolism based on hydraulic calculations using tracheid dimensions (Cochard et al., 2015). Results show that from 0 to −3.6 MPa only 10–20% of the conduits were empty and remained very close to the native embolism. For bench drying most conduits cavitated abruptly at about −3.9 MPa. For the centrifuge technique, embolism was more gradual, began at about −4 MPa and reached P<sup>50</sup> at a value around −5 MPa (**Figures 2** and **3**). The results for P<sup>50</sup> (from conductivity) are in agreement with the other bench drying and Cavitron sets, which gave values of −4.2 and −5.5 MPa, respectively (**Table 3**). The slopes from the micro-CT set as well as the lack of change in embolism from 0 to −4 MPa are closer to the results obtained with the Cavitron technique.

**Figure 4** shows pit closure pressures plotted against the conductivity obtained before pit closure. The results show that Elkosh and Elea had higher closure pressure as compared to Senalba and Otricoli, which may also indicate of an adaptation to aridity (Sperry and Tyree, 1990).

#### Differentiation in Xylem Anatomy

No differences in both lumen width and tracheid length were found among provenances when within-population variation was taken into account. However, Elea had the widest tracheids (18.2 ± 0.1 µm; **Table 2**), and tracheids of Elea (1.82 ± 0.02 mm)



Measurements were made at the end of the dry season (2013) and in the middle of the winter rainy season (2014) using the bench dehydration method (n = 5), and at the end of the dry summer season of 2014 with the Cavitron technique (n = 10). Significant differences between means are indicated by different lowercase letters.

were longer than those of Elkosh (1.71 ± 0.02 mm) and Senalba (1.67 ± 0.01 mm) but not of Otricoli (1.76 ± 0.02 mm, **Table 2**).

Measurements of torus-aperture overlap, shown in **Figure 5**, indicate that Elea and Elkosh had similar torus-aperture overlaps (0.68 ± 0.01 and 0.66 ± 0.01, respectively), which were significantly higher than those of Otricoli and Senalba (0.59 ± 0.01 and 0.56 ± 0.01, respectively). These differences were due to smaller pit apertures in Elea and Elkosh, while torus area was similar in all provenances (**Figure 5**). Torus-aperture overlap results were in agreement with pit closure pressures.

Pearson's correlation coefficients for the relationships between P50b values and anatomical features for the four ecotypes were significant for pit aperture area and torus overlap (p < 0.05), but not for torus area.

## DISCUSSION

This study evaluated differences and relationships between xylem hydraulic traits and anatomy in four P. halepensis provenances in a provenance trial. We found differences in both native and maximum (saturated) xylem hydraulic conductivity that were season-dependent and did not correlate with tracheid dimension. A significant correlation was found between resistance to embolism and bordered pit structure.

## Differences in Hydraulic Traits and Tracheid Size

Klein et al. (2013) reported differences in native PLC in the provenances studied here. Our results are in general agreement


different lowercase letters.


## PLC Curves Measured with the Two Methods

The PLC curve results obtained here by the bench dehydration method (Supplementary Figure S2) gave P<sup>50</sup> values between −3.6 and −4.5 MPa for the four provenances, which is lower than the value reported for P. halepensis by Oliveras et al. (2003) using the air injection method, −3.1 MPa, but substantially higher than the values obtained with the Cavitron (Supplementary Figure S2),

fpls-07-00768 May 31, 2016 Time: 12:58 # 7

with theirs, but we found different PLC values and lower variation between measurements (i.e., better repeatability), probably due to

i.e., −5.9 to −6.7 MPa, that were similar to previous Cavitron P. halepensis values (Delzon et al., 2010). Curves relating P<sup>50</sup> values to minimum 9 (which define safety margins) can also be used to estimate an expected P<sup>50</sup> value (Meinzer et al., 2009). Based on the observations of minimum 9's of between −3 and −4 MPa in the summer in the arid Yatir forest (Klein et al., 2011), and using a relationship based on data from other conifers, the expected value of P<sup>50</sup> is less than −6 MPa (Meinzer et al., 2009), which lends support to the Cavitron measurements. However, it might be argued that the values analyzed by Meinzer et al. (2009) were measured with centrifuge-based methods, similar to the Cavitron technique.

The discrepancy between centrifuge based measurements, as represented here by the Cavitron technique, and the bench dehydration and air injection methods is too large to smooth over. As noted recently, a number of conflicting results from different methodologies used in plant hydraulics need attention (Jansen et al., 2015). Our case is one of them, and putting the measurement methods together, i.e., bench dehydration, Cavitron technique and validation with micro-CT can add some important insight. One surprise is that the micro-CT observations confirmed that the P<sup>50</sup> determinations for both methods of cavitating branches do, in fact, correspond to approximately 50% embolism, even though the values of 9 differed, on average, by 1.2 MPa. One possible explanation is that the water potential determined in the two methods is not equivalent, i.e., that obtained from rotor speed is not the same as that obtained in the pressure chamber. It is important to note that for centrifuge method the pressures in the branch are not equal at different positions along the branch. Equivalence of centrifuge and pressure chamber measurements has been demonstrated before for broadleaved stems (Holbrook et al., 1995), but perhaps some peculiarity of P. halepensis, e.g., it's very short tracheids, causes differences. Other explanations are possible, and our results call for more experimentation and analysis. As demonstrated here, the Micro-CT method provides

FIGURE 3 | Micro-CT images of Elkosh provenance at different tensions as induced by centrifugation with the Cavitron (upper row) and by the bench dehydration method (lower row). Each horizontal panel represents one tree for which segments of 28 cm long with 1 cm diameter were scanned at the middle of the sample.

an opportunity for direct validation of some of the results and we expect that further exploitation of this method will bring us closer to the 'truth.'

One important implication of the different results is with respect to evidence for seasonal differences in PLC, as found here (**Figure 1**). Actual soil water potential at which stomata close is reflected by summer needle 9, which has been shown to range from −2.4 to −3.7 MPa in P. halepensis, close to other measurements of leaf 9 in summer (Schiller and Cohen, 1998; Klein et al., 2011), and those values should be close to the minimum 9 in branches. Pb,<sup>12</sup> values (**Table 3**) from bench drying measured by the pressure chamber suggest that embolism can be well above 12% even when 9 is above −4 MPa (**Table 3**), in agreement with the PLC values in **Figure 1**. On the other hand, if the onset of embolism is significantly lower than −4 MPa, as indicated by the Pc,<sup>12</sup> values (**Table 3**), it is hard to explain why PLC was much higher in the summer in our conditions (**Figure 1**).

Significant differences in the slope of PLC curves were found in the provenances, and both methods found that Elkosh was significantly lower than Senalba (Supplementary Figure S2; **Table 3**). Similar results were reported for P. canariensis, P<sup>12</sup> being similar in all populations studied, whereas P50, P<sup>88</sup> and slope showed statistically significant differences (López et al., 2013). A less steep slope suggests that embolism occurs gradually over a larger 9 range, resulting not only in less vulnerability but in a greater safety margin between stomatal closure and catastrophic embolism.

#### The Correlation of Pit Aperture Area with Embolism Resistance

P<sup>50</sup> values of Elea and Elkosh indicate that they are more resistant to embolism than Otricoli and Senalba (**Table 3**). The variations in P<sup>50</sup> correlated with variations in torus to pit ratio that were associated with variations in pit aperture area and not with torus area (**Figure 5**). The dominant effect of the pit aperture size rather than the torus membrane size was also demonstrated in two broad surveys of coniferous species, which suggested that embolism resistance parameters are strongly correlated with pit apertures, whereas only a weak correlation was found with torus diameter (Delzon et al., 2010; Bouche et al., 2014). Interestingly, modifications in pit aperture have been observed in tracheids along the 85 m trunk of Douglas-fir where pit aperture decreases with height, whereas torus diameter remains relatively constant (Domec et al., 2008). That modification along the trunk demonstrates a tradeoff between xylem safety and water conducting efficiency ensuring maximum height in Douglas-fir trees (Domec et al., 2008). Variation in pit aperture size and not in torus size suggests that independent developmental mechanisms control the size of these two elements and that genetic differentiation in border pit function are driven by variation in pit aperture size. Differences in pit closure pressures (**Figure 4**), which showed that the more resistant provenances closed at higher pressure, are an additional indication of functional anatomical adaptations.

The results of the current study are supported by the high survival rates of Elea and Elkosh in comparison to Otricoli and Senalba provenances that failed to survive in a more arid provenance trial in Yatir forest (Atzmon et al., 2004; Schiller and Atzmon, 2009). It seems that Elea and Elkosh are better adapted to drought conditions as they allow better growth performance, and also ensure embolism resistance by high torus to pit overlap. All together, these results imply the possibility to predict provenance performance under drought conditions from their structure and performance under more optimal conditions.

#### Genetic Differences in Resistance to Embolism

In the current study we have tested only four provenances of P. halepensis that were divided into two groups by means of embolism resistance. It has been demonstrated that in general as well as for gymnosperms specifically, genetic diversity is higher in the eastern than in the western Mediterranean (Fady and Conord, 2010). Using chloroplast simple sequence repeats (SSR) markers, Grivet et al. (2009) demonstrated higher genetic diversity in eastern populations of P. halepensis in comparison to more western ones. Particularly, it has been shown that Elea, the Greek population, is genetically different from the other populations in that study, and that Shaharia, the Israeli population (represented here by the Elkosh provenance) is significantly different from those of Algeria and Morocco (Grivet et al., 2009). The alignment of the two eastern provenances Elkosh and Elea as being more embolism resistant might relate to their high genetic diversity in comparison to Otricoli and Senalba provenances, which represent the Western populations and demonstrate low embolism resistance. Still, in order to test the relationship between the embolism resistance traits on the genetic diversity level, it is necessary to analyze more populations of P. halepensis from various regions for their ability to resist embolism and for their level of genetic variation.

The other Pinus species that had been shown to possess intraspecific variation of embolism resistance so far is P. canariensis (López et al., 2013). Both P. halepensis and

P. canariensis are the southernmost pine species of the northern hemisphere, colonizing a wide range of climates, and are thus considered the most drought-resistance pines. It appears that both species possess populations that experience extreme suboptimal climatic conditions.

In contrast to P. halepensis and P. canariensis, no or limited intraspecific variation in embolism resistance was evident in P. sylvestris (Martínez-Vilalta et al., 2009). The wide distribution of P. sylvestris mostly comprises boreal regions, but it also includes dry areas such as sites in Turkey. However, the reported limited genetic variation in embolism resistance did not include xeric sites in that study (Martínez-Vilalta et al., 2009). Limited intraspecific variation in embolism resistance was also reported in P. pinaster (Corcuera et al., 2011; Lamy et al., 2011, 2014). Similar to P. halepensis, P. pinaster is considered a Mediterranean pine, although its distribution is restricted to the western part of the Mediterranean Basin and its habitat also includes the Atlantic coast (Bucci et al., 2007). It was suggested that the two species represent contrasting biogeographic and demographic histories that probably had strong effects on variation in drought resistant traits (Gómez et al., 2005). Lamy et al. (2014) suggested that embolism resistance in P. pinaster is a canalized trait, meaning that the trait stays stable under various environmental conditions. The ability of an organism to canalize a trait

depends on internal genetic factors, and that ability is developed by natural selection (Waddington, 1942). Therefore, it might be that different populations would express greater or lesser extent of canalization to a certain trait, depending on time of separation and on various genetic and environmental factors. The significant but conservative difference in P<sup>50</sup> (less than 1 MPa) between populations in the current study might imply a degree of canalization strength in the embolism resistance trait. Similar moderate differences in P<sup>50</sup> were also demonstrated in P. canariensis (López et al., 2013). It is probable that natural developmental variation in pit aperture was triggered by extreme climatic events that are more present in semi-arid areas with a long dry season, as is the situation in some habitats of P. halepensis and P. canariensis (López et al., 2013; Dorman et al., 2015).

#### CONCLUSION

Our results showed significant differentiation in embolism resistance among P. halepensis in a provenance trial. This observation was consistently found using three different methods, i.e., bench drying, Cavitron technique and micro-CT. These differences were supported by anatomic analysis suggesting that pit aperture size is a key feature in determining embolism resistance. Although moderate, the observed natural variation in the embolism resistance trait in P. halepensis might be sufficient to promote adaptation to climate change. In the light of our results we speculate that species that are subjected to a wide range of climates, including extreme dry environment would express a lesser extent of canalization in embolism resistance traits than species that grow in a more moderate climate range. We therefore suggest including populations that grow at sites

#### REFERENCES


with sub-optimal climate conditions in future studies in order to detect genetic variation in the embolism resistance trait.

#### AUTHOR CONTRIBUTIONS

RD-S and SC designed the research. RD-S analyzed the anatomical measurements and wrote the manuscript. IP and VL performed the bench dehydration hydraulic measurements and IP analyzed the data. MM and GS performed the anatomical measurements. SD and GC performed and analyzed the Cavitron measurements. HC and EB performed and analyzed the micro-CT measurements. RD-S, SC, SD and HC revised the manuscript. All authors carefully read and approved the final manuscript.

#### ACKNOWLEDGMENTS

We thank Simcha Lev-Yadun, Tamir Klein and James K. Wheeler for fruitful discussion on xylem anatomy and technical aspects of the measurements, Hanita Zemach for help in SEM analysis, Eugene David Ungar and Giora Ben-Ari for help in statistical analysis of anatomical results and Aryeh Feinberg from McGill University and ETHZ for his help with the HPFM measurements as part of an undergraduate summer program at the Weizmann Institute. This article is based upon work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.00768

Pinus pinaster Ait. revealed by chloroplast microsatellite markers. Mol. Ecol. 16, 2137–2153. doi: 10.1111/j.1365-294X.2007.03275.x




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 David-Schwartz, Paudel, Mizrachi, Delzon, Cochard, Lukyanov, Badel, Capdeville, Shklar and Cohen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Intraspecific Variation in Wood Anatomical, Hydraulic, and Foliar Traits in Ten European Beech Provenances Differing in Growth Yield

#### Peter Hajek <sup>1</sup> , Daniel Kurjak <sup>2</sup> , Georg von Wühlisch<sup>3</sup> , Sylvain Delzon<sup>4</sup> and Bernhard Schuldt <sup>1</sup> \*

<sup>1</sup> Plant Ecology, Albrecht von Haller Institute for Plant Sciences, University of Göttingen, Göttingen, Germany, <sup>2</sup> Faculty of Forestry, Technical University in Zvolen, Zvolen, Slovakia, <sup>3</sup> Federal Research Institute for Rural Areas, Forestry and Fisheries, Thuenen Institute for Forest Genetics, Großhansdorf, Germany, <sup>4</sup> UMR BIOGECO Institut National de la Recherche Agronomique-UB, University of Bordeaux, Talence, France

#### *Edited by:*

Sergio Rossi, Université du Québec à Chicoutimi, Canada

#### *Reviewed by:*

Tommaso Anfodillo, University of Padova, Italy Jordi Martínez-Vilalta, Autonomous University of Barcelona, Spain

*\*Correspondence:*

Bernhard Schuldt bernhard.schuldt@plant-ecology.de

#### *Specialty section:*

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

*Received:* 28 February 2016 *Accepted:* 22 May 2016 *Published:* 15 June 2016

#### *Citation:*

Hajek P, Kurjak D, von Wühlisch G, Delzon S and Schuldt B (2016) Intraspecific Variation in Wood Anatomical, Hydraulic, and Foliar Traits in Ten European Beech Provenances Differing in Growth Yield. Front. Plant Sci. 7:791. doi: 10.3389/fpls.2016.00791 In angiosperms, many studies have described the inter-specific variability of hydraulic-related traits and little is known at the intra-specific level. This information is however mandatory to assess the adaptive capacities of tree populations in the context of increasing drought frequency and severity. Ten 20-year old European beech (Fagus sylvatica L.) provenances representing the entire distribution range throughout Europe and differing significantly in aboveground biomass increment (ABI) by a factor of up to four were investigated for branch wood anatomical, hydraulic, and foliar traits in a provenance trial located in Northern Europe. We quantified to which extend xylem hydraulic and leaf traits are under genetic control and tested whether the xylem hydraulic properties (hydraulic efficiency and safety) trades off with yield and wood anatomical and leaf traits. Our results showed that only three out of 22 investigated ecophysiological traits showed significant genetic differentiations between provenances, namely vessel density (VD), the xylem pressure causing 88% loss of hydraulic conductance and mean leaf size. Depending of the ecophysiological traits measured, genetic differentiation between populations explained 0–14% of total phenotypic variation, while intra-population variability was higher than inter-population variability. Most wood anatomical traits and some foliar traits were additionally related to the climate of provenance origin. The lumen to sapwood area ratio, vessel diameter, theoretical specific conductivity and theoretical leaf-specific conductivity as well as the C:N-ratio increased with climatic aridity at the place of origin while the carbon isotope signature ( <sup>13</sup> δ C) decreased. Contrary to our assumption, none of the wood anatomical traits were related to embolism resistance but were strong determinants of hydraulic efficiency. Although ABI was associated with both VD and <sup>13</sup> δ C, both hydraulic efficiency and embolism resistance were unrelated, disproving the assumed trade-off between hydraulic efficiency and safety. European beech seems to compensate increasing water stress with growing size mainly by adjusting vessel number and not vessel diameter. In conclusion, European beech has a high potential capacity to cope with climate change due to the high degree of intra-population genetic variability.

Keywords: adaptive capacity, *Fagus sylvatica* L., genetic variability, hydraulic conductivity, leaf morphology, phenotypic plasticity, provenance trial, vulnerability to cavitation

## INTRODUCTION

European beech (Fagus sylvatica L.) dominated the natural vegetation types of forests in Central Europe for centuries, forming large stands of resilient forest ecosystems (Ellenberg and Leuschner, 2010). Despite the competitive superiority of European beech for tree populations in temperate forests, this species is more vulnerable to drought-induced stem growth reductions than other temperate broad-leaved trees (Leuschner et al., 2001; Zimmermann et al., 2015). Consequently, the Central European beech populations are expected to suffer high mortality rates probably altering the distribution range at its dry distributional limit as a consequence of increased physiological stress due to a higher risk of drought exposure associated with recent climate change.

According to its large geographic distributional range, European beech is expected to exhibit substantial genetic diversity (Bolte et al., 2007). The inherent capability of this species to survive and reproduce successfully across such a wide range of habitats is either maintained by long-term adaptation (i.e., genotypic variation) or short-term acclimation (i.e., phenotypic plasticity; Lindner et al., 2010; Kremer et al., 2014). Consequently, evolutionary adaptation may have caused the selection of ecotypes adapted to the regional climatic conditions, which is manifested in phenotypic variation of plant functional traits across various provenances (Hamrick, 2004). Despite an increasing amount of studies on functional traits which are known to be related to drought resistance in beech, e.g., leaf stomatal features (Stojnic et al., 2015 ´ ), wood structure (Eilmann et al., 2014), resistance to xylem cavitation (Herbette et al., 2010, Wortemann et al., 2011), or hydraulic architecture (Aranda et al., 2015; Schuldt et al., 2016), still very little information is available on the relevance of phenotypic plasticity or genetic variation of traits on the adaptation of populations.

The high trait plasticity and remarkable regeneration potential of beech trees after disturbance leads to the assumption that beech is able to bear water shortage to a certain degree (Kahle, 2006; van der Werf et al., 2007). However, other studies report on drought susceptibility or dramatic yield loss and regeneration periods of decades subsequent to drought events and there is still an ongoing debate on the response of beech populations to drought exposure (Leuschner et al., 2001; Peuke et al., 2002; Leuzinger et al., 2005; Bréda et al., 2006; Jump et al., 2006). The mechanisms behind the adaptation and adaptability of beech to drought is a main challenge for a better understanding of the effects of climate change on this economically and ecologically important tree species. Because productivity is closely coupled with hydraulic efficiency (Hajek et al., 2014; Hoeber et al., 2014; Kotowska et al., 2015) and may only be achieved at the cost of hydraulic safety (Cochard et al., 2007), genetic differentiation in wood anatomical and hydraulic traits should be most pronounced between provenances differing in yield. However, the way how different provenances of the same tree species cope with such trade-offs still remains poorly investigated.

The main objective of this study was to investigate the genetic differentiation in vulnerability to xylem embolism and other related hydraulic and foliar properties across ten European beech provenances differing in growth performance. We further aimed to investigate the relationships and potential trade-offs between hydraulic, wood anatomical and leaf traits within species. We hypothesized that (i) drought-related traits are under genetic control and therefore populations are locally adapted to their environment, (ii) hydraulic efficiency and embolism resistance are related to anatomical traits, and (iii) a high hydraulic efficiency leads to high growth rates at the cost of xylem safety.

#### MATERIALS AND METHODS

## Experimental Site, Plant Material, and Microclimatic Conditions at the Place of Origin

The field trial studied is part of the International Beech Provenance Trial Series 1993/1995 under EU funding (AIR3- CT94-2091), which investigates the role of the genetic variation of beech for adaptability, productivity and selected ecosystem functions considering risks of global climate change (von Wuehlisch et al., 1998). For the present study conducted in August 2014, the common-garden field trial with 100 different European beech (Fagus sylvatica L.) provenances established in 1995 in Northern Germany (Schleswig-Holstein) with 2 year old saplings near the Schädtbek Experimental Farm (54◦ 18′N, 10◦ 16′E, 40 m a.s.l.) was used. The climate at the site is oceanic, moderately cold with a mean annual temperature (MAT) of 8.3◦C, a mean annual precipitation (MAP) of 742 mm and a mean early growing season precipitation from April to June (MSP) of 149 mm (data obtained from the German Meteorological Service). The common garden trial consists of 10 × 10 m plots planted with 50 beech saplings per provenance in a 3-times replicated randomized block design. The beech trees are arranged in a rectangular grid with a planting distance of 2 m between and 1 m within rows. The trial is surrounded by a single bordering tree row serving as buffer zone to avoid edge effects. As all provenances were grown in a single environment (common garden), we were able to assess the genetic differentiations between provenances for several functional traits. The common garden experienced no management operations since plantation establishment in 1995, resulting in competitive selection among tree individuals leading to heterogeneity in plot characteristics.

From the 100 available provenances, ten provenances differing in aboveground growth increment (Figure S1) were selected in order to cover not only a gradient in growth yield but also a climatic gradient at the place of origin (Figure S2). By this selection, we covered provenances from different geographic regions and climates throughout Europe with a broad range of MAT (3.4–15.3◦C) and MAP (575–1080 mm; **Table 1**). Mean annual climate data from 1950 to 2000 at the place of origin for each of the ten provenances were obtained from the WorldClim database with 30 arc-seconds resolution (Hijmans et al., 2005). In order to estimate climate dryness at the place of origin, we used the WorldClim database to access the global aridity index (GAI, Zomer et al., 2008) and calculated the forest aridity index (FAI) according to Fuehrer et al. (2011) as FAI = 100 × TJul−Aug/(PMai−Jul + PJul−Aug), where T is the temperature and P the precipitation of the associated interval. Since the atmospheric evaporative demand in the growing season is highest in midsummer (July and August), the July precipitation was weighted by a factor of two in the denominator. We further calculated Ellenberg's climate quotient (EQ, Ellenberg and Leuschner, 2010) as EQ = (TJul/Pannual) × 1000. All measures of water availability and climatic aridity at the places of provenance origin were highly interrelated (data not shown). We therefore decided to use the forest aridity index (FAI) as integrate variable for all subsequent analyses.

Provenances were selected according to differences in aboveground growth increment based on a tree inventory in January 2013 (Figure S1). Within a given provenance, ten tree individuals of comparable size close to the population average (diameter at breast height and tree height) were selected assuming uniform growth conditions (site conditions and intraspecific competition) and consequently representing the average growth increment of a given provenance, yielding 100 processed tree individuals in total. A 2- to 4-year-old branch segment (mean ± SE : 2.25 ± 0.05 year) was collected in August 2014 from the uppermost canopy of each selected tree individual with a long-reaching telescope pruner and recut to approximately 50 cm length on the ground. Selected segments were defoliated and immediately transferred to plastic tubes containing deionized water and Micropur (Katadyn, Wallisellen, Switzerland) to prevent microbial activity and stored at 4◦C until further processing within 4 weeks. All leaves of the respective segments were stored separately for foliar analyses. A list of all measured traits, their symbols and units are given in **Table 2**.

## Timber Volume, Above-Ground Biomass, and Basal Area Increment

Aboveground growth performance (i.e., stem increment and height growth) of the respective genotypes under the local environment was evaluated from diameter at breast height (DBH, cm) and tree height (m). Aboveground productivity expressed as aboveground biomass increment (ABI, kg yr−<sup>1</sup> ) was calculated for the entire growth period (1995–2014), and basal area increment (BAI, cm<sup>2</sup> yr−<sup>1</sup> ) from two inventories in January 2013 and August 2014, respectively. For calculating the volume of stems and branches of at least 7 cm in diameter (standing volume of timber), we used the allometric equation V = π × ((D/100)/2)<sup>2</sup> × H × f, where V is the volume of timber (m<sup>3</sup> ), D the diameter at breast height (cm), H tree height (m), and f an empirically derived form factor for beech trees (Bergel, 1973) with f = 0.4039 + (0.0017335 × H) + (1.1267/H)–(118.188/D 3 ) + (4.2 × 10<sup>6</sup> × D 2 ). Above-ground biomass was estimated from an empirical equation given by Wutzler et al. (2008) as AGB = 0.00523 × D 2.12 × H0.655 .

#### Leaf-Related Measurements

From each branch segment, all leaves were removed from the basipetal segment upwards to determine mean leaf size (Aleaf, cm<sup>2</sup> ) and cumulative leaf area (AL,m<sup>2</sup> ) using a flatbed scanner and the WinFOLIA software (Régent Instruments, Quebec, Canada). Per branch segment, 19 to 194 leaves were scanned, yielding 6008 leaves in total. Specific leaf area (SLA, cm<sup>2</sup> g −1 ) was

TABLE 1 | Climatic data at the place of origin of the ten *Fagus sylvatica* provenances including elevation, mean annual temperature (MAT), mean annual precipitation (MAP), mean early growing season precipitation from April to June (MSP), Ellenberg's climate quotient (EQ), the global aridity index (GAI) and the forest aridity index (FAI).


Mean annual climate data from 1950 to 2000 were obtained from the WorldClim database with 30 arc-seconds resolution (Hijmans et al., 2005).



calculated by dividing the total leaf area by the leaf dry weight (70◦C, 48 h). The Huber value, i.e., sapwood to leaf area ratio (A<sup>S</sup> : AL, 10−<sup>4</sup> m<sup>2</sup> m−<sup>2</sup> ), was calculated by dividing maximal sapwood area by AL. Subsequently, leaf samples were ground and the leaf dry mass analyzed for foliar C and N concentrations as well as the carbon isotope signature (δ <sup>13</sup>C, ‰) by using a Delta Plus Isotope mass ratio spectrometer (Finnigan MAT, Bremen, Germany), a Conflo III interface (Thermo Electron Corporation, Bremen, Germany) and a NA2500 elemental analyser (CE-Instruments, Rodano, Milano, Italy) using standard δ notion: δ = (Rsample/Rstandard–1) × 1000 (‰) at the Centre for Stable Isotope Research and Analysis (KOSI), University of Göttingen. The foliar concentrations of Ca, K, Mg, and P were measured by ICP analysis (Optima 5300 DV, PerkinElmer Inc., USA).

#### Branch Xylem Anatomy, Theoretical Conductivity, and Growth Rate

Measurements of anatomical parameters were carried out on 9–10 branch segments from the basipetal end of the samples used for the hydraulic measurements of each provenance, yielding 98 samples in total. Prior to cutting semi-thin (10– 20 µm) transverse sections with a sliding microtome (G.S.L.1, Schenkung Dapples, Zürich, Switzerland), the ethanol-stored (70%) segments were completely stained with a safranin solution (1 in 50% ethanol, Merck, Darmstadt, Germany) and subsequently embedded in Euparal medium. For digitalization, a stereo-microscope equipped with an automatic stage was used (SteREOV20, Carl Zeiss MicroImaging GmbH, Jena, Germany; Software: AxioVision v4.8.2, Carl Zeiss MicroImaging GmbH, Jena, Germany), enabling a time-efficient digitalization of the complete cross-section at 100x magnification. Image analysis was performed using the software Adobe Photoshop CS2 (Version 9.0, Adobe Systems Incorporated, USA) and ImageJ (v1.44p, http://rsb.info.nih.gov/ij) applying the particle analysis function. For all subsequent calculations, the complete xylem cross-section without pith and bark was analyzed, yielding 888,358 analyzed vessels in total. The following parameters were calculated: idealized vessel diameter (D, µm) as obtained from major (a) and minor (b) vessel radii according to Lewis and Boose (1995) as D = ((32 × (a × b) 3 )/(a <sup>2</sup> + b 2 ))¼, vessel density (VD, n mm−<sup>2</sup> ) and cumulative vessel lumen area. The lumen to sapwood area ratio (Alumen : Axylem, %), was obtained by dividing cumulative vessel lumen area by the corresponding sapwood area. The diameter of individual vessels was used to calculate hydraulically-weighted vessel diameter (Dh, µm) according to Sperry et al. (1994) as D<sup>h</sup> = 6D 5 /6D 4 . Theoretical specific conductivity (K theo S , kg m−<sup>1</sup> MPa−<sup>1</sup> s −1 ) was calculated according to the Hagen-Poiseuille equation as K theo <sup>S</sup> <sup>=</sup> (((<sup>π</sup> <sup>×</sup> <sup>6</sup><sup>D</sup> 4 )/128 η) × ρ)/Axylem, where η is the viscosity of water (1.002 10−<sup>9</sup> MPa s), ρ the density of water (998.2 kg m−<sup>3</sup> ), both at 20◦C, and Axylem (m<sup>2</sup> ) the corresponding xylem area without pith and bark. Branch growth rate (Agrowth, mm<sup>2</sup> yr−<sup>1</sup> ) was calculated by dividing Axylem by the number of growth rings (i.e., branch age, BA). In addition to K theo S , theoretical leaf-specific conductivity was calculated by division by A<sup>L</sup> as K theo <sup>L</sup> <sup>=</sup> (((<sup>π</sup> <sup>×</sup> <sup>6</sup><sup>D</sup> 4 )/128 η) × ρ)/AL.

#### Hydraulic Conductivity Measurement

Hydraulic traits were measured in ten branch segments (mean diameter ± SE: 6.83 ± 0.06 mm) per provenance using the Xyl'em apparatus (Bronkhorst, Montigny-les-Cormeilles, France). In the laboratory, all lateral branches were cut off and the scares sealed with quick-drying superglue (Loctite 431, Henkel, Düsseldorf, Germany) applicable to wet surfaces. Subsequently, the segments were shortened to a length of 290.5 ± 0.8 mm (mean ± SE). For the determination of maximal hydraulic conductivity (Kh, kg m MPa−<sup>1</sup> s −1 ) at 6 kPa, demineralized filtered (0.22 µm) and degassed water (10 mM KCl and 1 mM CaCO3) was used, interrupted by three 10-min flushes at 120 kPa to assure removal off all potential emboli. The diameter of each segment was measured twice at the basipetal and distal end, and at four positions along the segment. The following regression coefficients were used to calculate sapwood area without pith and bark for a given beech branch segment diameter according to Schuldt et al. (2016): Axylem = −3.715 + 0.770 Across . Subsequently, empirical specific conductivity (K emp S , kg m−<sup>1</sup> MPa−<sup>1</sup> s −1 ) was calculated by dividing K<sup>h</sup> by the maximal basipetal, and not average, sapwood area (Hajek et al., 2014; Hoeber et al., 2014; Schuldt et al., 2016). K<sup>h</sup> was further used to calculate empirical leaf-specific conductivity (K emp L , kg m−<sup>1</sup> MPa−<sup>1</sup> s −1 ) by division by AL.

#### Xylem Resistance to Cavitation

Vulnerability to xylem cavitation was determined on 8 to 10 branch samples (replicated trees) per provenance using the Cavitron technique (Cochard et al., 2005), yielding 94 samples in total. Segments with a standardized length of 27.5 cm were mounted in a custom-built honeycomb rotor chamber of the Cavitron, which uses a commercially available centrifuge as basis (Sorvall RC-5C, Thermo Fisher Scientific, Waltham, MA, USA), and spun at defined velocities recorded with the software CaviSoft (version 4.0.1.3, University of Bordeaux, France). Conductivity measurements started at 1.0 MPa and were stepwise repeated at intervals of 0.2 to 0.3 MPa until the percent loss of conductivity (PLC) reached at least 90%. All vulnerability curves of the present study measured on 27.5 cm long segments were s-shaped, indicating that no open vessels were present. For each branch segment, a sigmoid function (Willigen and Pammenter, 1998) was fitted to describe the relationship between PLC and xylem pressure using the expression PLC = 100/(1 + exp(s/25 × (P<sup>i</sup> – P50))), where P<sup>50</sup> (MPa) is the xylem tension causing 50% loss of hydraulic conductivity and s (% MPa−<sup>1</sup> ) is the slope of the curve at the inflexion point. The xylem pressures causing 12% (P12) and 88% (P88) loss of conductivity were calculated as well.

#### Statistical Analysis

In order to assess the significance of differentiation between provenances for all functional traits, we used linear mixed effect models (LME) starting with a random effect model without any fixed effect but with "provenance" and "block" added as a random effects using the "lme" function of the R package "nlme" according to the following model [1]: Yijk = µ + P<sup>i</sup> + b<sup>j</sup> + εijk, where Yijk is the observation of tree individual k for one of the analyzed characters from provenance i and block j, µ is the overall mean, and P<sup>i</sup> the error term for provenance, b<sup>j</sup> for block and εijk for the residual variation. During the analysis, normal distribution of the residuals and homogeneity of variance were assessed visually using residual diagnostics and quantile-quantile plots; some data (K emp L , K theo L , A<sup>S</sup> : AL) had to be log-transformed in order to achieve normal distribution. In order to test whether the variance component accounting for genetic differentiation is significantly different from zero, a likelihood ratio test (LRT) was performed against a reduced model without a random effect for provenance using the restricted maximum likelihood (REML) method. As under those conditions the LRT is performed on the boundary of the parameter space, the resulting P-values had to be corrected by multiplying them by 0.5 (Verbeke and Molenberghs, 2009).

To quantify the influence of climate at the place of origin, FAI was added to the LME as fixed variable according to the following model [2]: Yijk = α + β FAI<sup>i</sup> + P<sup>i</sup> + b<sup>j</sup> + εijk, where α is the intercept and β the slope of the linear model with FAI<sup>i</sup> , which all three together determine µ of model [1]. A significant influence of climate at provenance origin was determined by a LRT using the maximum likelihood (ML) method.

We further calculated coefficients of variation for withinprovenance variation (CVintra) and between-provenance variation (CVinter) for each trait in order to allocate total measured trait variation to a genetic component (CVinter) and a predominantly phenotypic component (CVintra); the between-provenance variability (CVinter) was calculated from the between-provenance standard deviation (SD) and the overall mean value. Additionally, the ratio of provenance variance component to total variance was calculated using the R package "varComp" according to a variance component analyses with the program "lme" to calculate the proportion of total variance (σ 2 total) explained by the variability between provenances (σ 2 inter), replicated randomly distributed "blocks" (σ 2 block) and residual variance within provenances (σ 2 intra). Inter-population variance component (VCinter) was calculated according to VCinter = (σ 2 inter/(σ 2 inter + σ 2 block <sup>+</sup> <sup>σ</sup> 2 intra)) × 100, variance component between replicated plots as VCblock = (σ 2 block/(σ 2 inter + σ 2 block + σ 2 intra)) × 100, and intra-population variance component (VCintra) as VCintra = (σ 2 intra/(σ 2 intra + σ 2 block <sup>+</sup> <sup>σ</sup> 2 inter)) × 100, all in percentage.

Pearson correlation analysis was used to test for interrelationships between different branch traits of the trees and for detecting relationships between traits based on data pooled across all provenances. Applying correlations to raw measurements for such exploratory analysis may mix several sources of variation and result in inflated degrees of freedom and potentially overestimation of significance. Statistical analyses were performed with the software R (version 3.1.3, R Development Core Team, 2011), and all linear and non-linear regression analyses were carried out with the software Xact 8.03 (SciLab, Hamburg, Germany).

## RESULTS

## Genetic Differentiation between Provenances

The estimate of aboveground growth potential as inferred from the basal area increment (BAI) from January 2013 to August 2014 and aboveground biomass increment (ABI) for the entire length of the experiment from August 1995 to August 2014 revealed large significant differences between the ten selected beech provenances (**Table 3**). The provenances with highest ABI from Slovakia and Ukraine (SK: 3.6; UA 4.0 kg yr−<sup>1</sup> , P < 0.001; Table S1) showed four times higher growth rates than the provenances with lowest growth rates from Slovenia and Sweden (SL: 1.02 and SE: 1.69 kg yr−<sup>1</sup> ; Table S1). Both provenances


TABLE 3 | Results of a random effects model on the genetic differentiation between provenances and the coefficient of variation for all traits measured for the variability between provenances (CVinter) and within provenances (CVintra).

Given is the ratio of inter-population (VCinter) variance component between replicated plots within a given provenance (VCplot) and intra-population (VCintra) variance component to total variance, all in percentage. The results of a likelihood ratio test of a linear mixed effect model against the null model assuming that genetic differentiation is significantly different from zero is given; these are the delta Akaike information criterion (1<sup>i</sup> ), the likelihood ratio (LR) and probability of error (P) for the hypothesis that SDprov 6= 0. Significant correlations (P < 0.05) are printed in bold, marginally significant correlations (P < 0.10) in italic bold letters.

had the highest radial growth rates (SK: 33.36 and UA: 23.20 cm<sup>2</sup> yr−<sup>1</sup> ) and grew more than seven-fold compared to the two provenances with lowest basal area increment (ES: 4.34 and SL: 5.68 cm<sup>2</sup> yr−<sup>1</sup> ; Table S1). The other six provenances reached intermediate growth rates (13.05–22.55 cm<sup>2</sup> yr−<sup>1</sup> ). ABI and BAI scaled highly positive with tree size (DBH, r <sup>2</sup> = 0.98, P < 0.001 and r <sup>2</sup> = 0.73, P < 0.001; height, r <sup>2</sup> = 0.66, P < 0.01 and r 2 = 0.31, P < 0.001; **Table 5**).

Significant genetic differentiations between provenances were found in three out of 22 ecophysiological traits only, namely the xylem pressure inducing 88% loss of hydraulic conductance (P88), vessel density (VD) and mean leaf size (Aleaf; **Table 3**). The P<sup>88</sup> values ranged from −3.60 MPa for the most vulnerable provenance from Germany (DE-BB) to −4.21 MPa for the most resistant provenance from Slovenia (SL), i.e., by 15% (Table S1). The differences in embolism resistance between provenances were, however, not mirrored in xylem anatomical adjustments. VD was significantly higher in the branch xylem of the mostproductive provenances from Slovakia (SK) and the Ukraine (UA) while the largest leaves were found for the provenance adapted to the local climate of Schleswig Holstein (DE-SH; Table S1). Other provenances from Sweden (SE) and Czech Republic (CZ) produced on average 30% smaller leaves, which was not mirrored in adjustments of specific leaf area (SLA) or the Huber

value (A<sup>S</sup> : AL). In general, the leaf chemistry was uniform across the provenances.

Significant genetic differentiation between provenances (**Table 3**) was mirrored in high values of inter-population variance component (VCinter) for these three ecophysiological traits (P88, VD, Aleaf) ranging from 8.38 to 14.02% (**Table 3**). Interestingly, all three measures of embolism resistance showed a higher variance between blocks (VCblock : ∼25%) than between populations (VCinter : ∼10%). The remaining measured hydraulic traits, empirical specific conductivity (K emp S ) and empirical leaf-specific conductivity (K emp L ), likewise differed by ∼10% between blocks but showed no variance between populations. Contrary to these hydraulic traits, all wood anatomical and calculated theoretical specific conductivity (K theo S ) revealed no variance between blocks. For all traits measured excluding the growth-related variables, variability within provenances (CVintra) was about two times higher than variability between provenances (CVinter; **Table 3**).


TABLE 4 | Results of a linear mixed effect model examining the influence of the forest aridity index (FAI) at the place of origin on 27 measured parameters 19 years after planting of the provenance trial in Northern Germany.

Given are intercept α and slope β indicating the direction of the relationship for FAI with their corresponding standard errors (SE), standard deviation between provenances (SDprov) as an expression of the variability between provenances, standard deviation of block from the provenance mean showing the design variability, and the remaining unexplained variance of the residuals (SDresid). Further, the results from a likelihood ratio test of the LME against the null model are given; these are the delta Akaike information criterion (1<sup>i</sup> ), the likelihood ratio (LR) and probability of error (P) for the hypothesis that FAI 6= 0. Asterisks indicate traits that have been log transformed, for abbreviations see *Table 2*. Significant correlations of FAI on any given trait (P < 0.05) are printed in bold, marginally significant correlations (P < 0.10) in italic bold letters.

## Trait-Relatedness to Climate at Origin

We observed several significant linear relationships between functional traits and the forest aridity index (FAI) as a measure of the climatic conditions at the place of provenance origin (**Figure 1**). These simple linear regression analyses were supported by linear mixed effect models confirming that FAI had a significant influence on seven of the 27 measured functional traits (**Table 4**). With increasing FAI, vessel diameter (D), lumen to sapwood area ratio (Alumen : Axylem), theoretical specific conductivity (K theo S ), theoretical leaf-specific conductivity (K theo L ) and the C:N ratio significantly increased while the carbon isotope signature (δ <sup>13</sup>C) declined (**Figures 1A–F**; **Table 4**). Surprisingly, the three traits showing genetic differentiation between provenances (P88, VD and Aleaf) were not significantly affected by FAI. However, our results show that provenances originating from dry habitats with high FAI values (e.g., BG, ES, DE-BB) form particularly wide vessels and have high lumen to sapwood area ratios (**Figures 1A,B**) resulting in a high K theo S (**Figure 1C**) and K theo L (**Figure 1D**) compared to the other provenances when grown at this humid site with high precipitation rates received during the entire growing season. The C:N ratio was also significantly positively related to FAI, while a strong significant negative relationship between FAI and δ <sup>13</sup>C was observed (**Figure 1F**).

(*A*growth, B), and tree height (C). Given values are means per provenance, for symbol definition see Figure 1.

## Determinants of Aboveground Growth Performance

Among the xylem anatomical traits, only VD was positively related to aboveground biomass increment (ABI; r <sup>2</sup> = 0.66, P < 0.005; **Figure 2A**) and negatively to branch growth rate (Agrowth; r <sup>2</sup> = 0.35, P < 0.05; **Figure 2B**), which is a simplified measure of the annually produced branch sapwood, at the provenance level (n = 10). At the tree level (n = 100), only the negative relation between VD and Agrowth could be confirmed (r <sup>2</sup> = 0.26, P < 0.001; **Table 5**). Moreover, VD strongly increased with increasing tree height (r <sup>2</sup> = 0.83, P < 0.001; **Figure 2C**). The remaining branch wood anatomical and hydraulic properties, however, varied independently from aboveground growth performance, and no correlations were found between ABI, embolism resistance and hydraulic efficiency, respectively (**Table 5**). However, a weak though significant negative relation was found for basal area increment (BAI) and P<sup>50</sup> at the tree level (r 2 = 0.08, P < 0.01; **Table 5**). Although we only observed weak or no relations between ABI and the wood anatomical and hydraulic traits, several of them were related to Agrowth (**Table 5**). Hence, the expected trade-off between hydraulic efficiency and growth could be confirmed at branch level, but not at the tree level.

## Inter-Relationships between Functional Traits

We found that hydraulic efficiency in beech depends on the xylem properties. The vessel lumen to sapwood area ratio (Alumen : Axylem), annual branch growth rate (Agrowth), vessel diameter (D) and hydraulically-weighed vessel diameter (Dh) were closely positively related to empirical specific conductivity (K emp S ) and empirical leaf-specific conductivity (K emp L ). In contrast, correlations between vessel density (VD) and the hydraulic traits K emp S or K theo <sup>S</sup> were not found (**Table 5**). Furthermore, none of the wood anatomical traits were related to embolism resistance and the expected trade-off between hydraulic conductivity and embolism resistance was absent. In addition, we found significant links between functional leaf traits such as specific leaf area (SLA) and empirical as well as theoretical leaf-specific conductivity (K emp L , K theo L ). The carbon isotope signature (δ <sup>13</sup>C) was positively correlated with ABI (fast growth was associated with a frequent stomatal closure) but negatively with the P<sup>50</sup> value (**Figures 3A,B**). The provenances with the highest growth rates (UA, SK, and BG) exhibited highest carbon isotope signature (r <sup>2</sup> = 0.29, P < 0.05; **Figure 3A**). The foliar C:N ratio was positively related to vessel diameter (D) and the theoretical specific conductivity (K theo S ; **Figures 4A,B**), indicating that higher growth rates (associated with a high C:N ratio) were indirectly related to xylem hydraulic properties of the branch wood.

## DISCUSSION

## Genetic Differentiation

Only recently, a growing number of studies have investigated the intraspecific genetic differentiation in ecophysiological traits between populations (e.g., Bresson et al., 2011; Lamy et al., 2011; Eilmann et al., 2014; Aranda et al., 2015; Schreiber et al., 2015). Despite this positive trend, common-garden studies that have reported anatomical or hydraulic traits are still scarce (Anderegg and Meinzer, 2015). Our study on ten European beech provenances native to different localities in Europe revealed little intraspecific variation of the ecophysiological traits covered. In agreement with former common-garden experiments we observed no significant genetic differences between populations for the xylem pressure causing 50% loss of


for symbol definition see Figure 1.

conductivity. The same pattern has been described for conifers (Sáenz-Romero et al., 2013; Lamy et al., 2014) as well as for beech (Wortemann et al., 2011, but see Aranda et al., 2015).

FIGURE 3 | Aboveground biomass increment (ABI, A) and the xylem pressure causing 50% loss of hydraulic conductivity (*P*50, B) in relation to the carbon isotope signature. Given values are means per provenance,

Contrary to these studies, however, we observed significant genetic differentiation for the xylem pressure inducing 88% loss of hydraulic conductance (P88), which corresponds to the threshold for catastrophic hydraulic failure leading to irreversible drought-induced dysfunction in angiosperms (Urli et al., 2013; Li et al., 2015). In our study, genetic differentiation was further found for the xylem anatomical trait vessel density and the foliar trait mean leaf size. Our first hypothesis postulating that wood hydraulic properties are under genetic control due to local adaptation is further supported by the observation that most wood anatomical and derived hydraulic traits as well as some foliar traits were closely related to the climate at provenance origin. These results may indicate that wood anatomical and hydraulic traits of beech reflect genetic differentiation between provenances although no significant intraspecific differences became evident. This is in line with the observation that several wood properties related to vessel size are predominantly under genetic predisposition (Eilmann et al., 2014).

However, the high intra-population variability in all ecophysiological traits covered, which was on average two to three times higher than inter-population variability, is also a strong evidence for the high genetic diversity within provenances. For example, Aranda et al. (2015) stated that beech exhibits a high degree of intra-population genetic variability for embolism resistance. This indicates that European beech might have a high potential capacity to adapt to climate change. In agreement hereon, both hydraulic efficiency and embolism resistance were unrelated to climatic aridity at the place of provenance origin and showed the highest variance within a given provenance (∼65%), followed by the second highest variance between replicated blocks (∼25%) due to microenvironmental effects including within-canopy variability, and only a comparatively low variance between populations (∼10%). This likewise confirms that beech exhibits a high degree of intra-population genetic variability for embolism resistance. Surprisingly, provenances originating from drought prone habitats (e.g., ES, BG, and DE-BB) grown at our site do not

represent the most drought tolerant provenances in terms of functional trait adaptation as expected (see Eilmann et al., 2014). This finding is in agreement with former studies quantifying the high phenotypic plasticity of embolism resistance across environmental gradients (Herbette et al., 2010; Wortemann et al., 2011; Schuldt et al., 2016). Moreover, the vessel density of the branch xylem seems to play an outstanding role in the adaptation of the branch hydraulic system to different climates. Likewise with embolism resistance, vessel density varied independently of climatic conditions at the place of origin but was significantly different among provenances presumably as a consequence of differences in branch height. European beech trees may accordingly hold the capacity to adapt to local climate conditions by modifying their branch hydraulic traits primarily through adjustments in the relative abundance of vessels. This assumption is further supported by the close relation between vessel density and productivity as tree growth performance is hypothesized to scale with hydraulic efficiency (Tyree, 2003), which contrary to our results has been confirmed for both temperate and tropical tree species (Hajek et al., 2014; Hoeber et al., 2014; Kotowska et al., 2015).

## Functional Trade-Offs

It has been suggested that embolism resistance decreases with growth rate due to conflicting carbon allocation either to the construction of thicker cell walls, or to the building of foliar and axial tissues destined to increase canopy carbon gain (Cochard et al., 2007). However, empirical data from different species or genotypes do not unequivocally support this trade-off. For example, Fichot et al. (2010) reported on embolism-resistant genotypes of poplar which grew faster than more vulnerable genotypes. We observed no relation between P<sup>50</sup> in branches and aboveground biomass increment across the provenances contradicting the hypothesized relation between embolism resistance and growth, which should trade-off with hydraulic efficiency. Several recent studies also failed to detect this relationship between embolism resistance and growth rate (Sterck et al., 2012; Hajek et al., 2014; Guet et al., 2015), indicating that embolism resistance is partly decoupled from hydraulic efficiency and biomass production (Fichot et al., 2015). In agreement hereon and in line with our results, hydraulic safety seems mostly decoupled from hydraulic efficiency, both at the inter-specific (Gleason et al., 2016) and intra-specific level (Schuldt et al., 2016; but see Hajek et al., 2014). In addition to these anticipated trade-offs it is necessary to include the foliage functionality for a holistic understanding of growthrelated processes (Carlquist, 2012). In our sample, the P<sup>50</sup> and P<sup>88</sup> values decreased in parallel with increasing carbon isotope signature. The most embolism resistant provenances thus presumably had to close their stomata more frequently at the expense of significantly reduced carbon assimilation. This observation is contradicting observations of the functional coordination between vulnerability to xylem cavitation and the regulation of stomatal conductance (Sparks and Black, 1999; Brodribb et al., 2002). Either a sensitive stomatal regulation or simply a lower water demand as a consequence of smaller stature of drought prone provenances may explain this trait interrelation.

Moreover, the P<sup>50</sup> and P<sup>88</sup> values were not related to xylem anatomical traits like vessel diameter or vessel density as evidenced in several studies among and across species (Domec et al., 2010; Hajek et al., 2014). In contrast, the range of variation was small in our study and might have hampered a significant relationship as likewise described by other authors who also failed to detect a relation between vessel diameter and embolism resistance in closely related genotypes or different hybrids of poplar (Cochard et al., 2007; Fichot et al., 2010). These findings support the growing evidence that variation in P<sup>50</sup> is mainly determined by other wood structural properties such as the topology of the xylem network (Loepfe et al., 2007; Martínez-Vilalta et al., 2012) as well as the pit membrane morphology (Plavcova et al., 2013; Li et al., 2016). The relation with vessel size in studies covering a sufficiently steep range in diameters is thus rather indirect as already speculated by Tyree and Sperry (1989).

In our study, provenances originating from drought-prone habitats (e.g., Spain) were most vulnerable to xylem cavitation. This contradicts the general observation of a less vulnerable structure for provenances originating from drought-prone sites (Rose et al., 2009; Robson et al., 2012). Surprisingly, the Spanish provenance developed the largest vessels of all ten provenances under these favorable environmental conditions contrary to our expectation. Due to the genetic control of this trait, beech provenances adapted to drier climates seem to over-develop their vascular system when grown in more humid environments. However, this did not translate into a better growth performance of these provenances.

## CONCLUSION

Our study on ten beech provenances from all over Europe revealed that most wood anatomical and derived hydraulic traits and some foliar traits are under genetic predisposition according to significant differentiation between provenances or the relation with climatic aridity at the place of provenance origin. Although a certain degree of genetic differentiation was observed for embolism resistance, the high ratio of intra- vs. inter-population variance suggests that the genotypic variability of this trait is high between genotypes of a given provenance. In European beech, this high adaptive capacity in xylem function seems predominantly to be a consequence of adjusting vessel number but not necessarily vessel size. Nevertheless, we could not confirm the anticipated trade-off between hydraulic efficiency, xylem safety and growth in agreement with several recent studies. It further remains unclear if the observed high adaptive capacity of young European beech tree individuals can be extrapolated to old-growth forest trees, potentially enabling them to withstand increased drought by a flexible hydraulic response.

## AUTHOR CONTRIBUTIONS

BS and GvW designed the study, PH, DK, and BS collected the field samples, DK and PH performed the hydraulic, wood anatomical and leaf morphological measurements and PH, BS, and SD analyzed the data and performed the statistical analyses. PH and BS wrote the first version of the manuscript, which was intensively discussed and revised by all authors.

#### ACKNOWLEDGMENTS

We thank Roman Link for statistical advice, Ana Sapoznikova and all other assistants for their invaluable contributions to the experiment and the two reviewers, who provided helpful suggestions for improving the manuscript. The

#### REFERENCES


work has been financially supported by the University of Göttingen granted to BS. We further acknowledge the support by the Open Access Publication Funds of the University of Göttingen.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016. 00791


cavitation resistance in a Mediterranean pine. New Phytol. 201, 874–886. doi: 10.1111/nph.12556


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Hajek, Kurjak, von Wühlisch, Delzon and Schuldt. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Desiccation and Mortality Dynamics in Seedlings of Different European Beech (Fagus sylvatica L.) Populations under Extreme Drought Conditions

Andreas Bolte<sup>1</sup> \*, Tomasz Czajkowski <sup>1</sup> , Claudia Cocozza<sup>2</sup> , Roberto Tognetti 3, 4 , Marina de Miguel <sup>5</sup> , Eva Pšidová<sup>6</sup> , Lubica Ditmarová ´ 6 , Lucian Dinca<sup>7</sup> , Sylvain Delzon<sup>5</sup> , Hervè Cochard<sup>8</sup> , Anders Ræbild<sup>9</sup> , Martin de Luis <sup>10</sup>, Branislav Cvjetkovic<sup>11</sup>, Caroline Heiri <sup>12</sup> and Jürgen Müller <sup>1</sup>

#### Edited by:

Boris Rewald, University of Natural Resources and Life Sciences, Vienna, Austria

#### Reviewed by:

Christiane Werner, University of Freiburg, Germany Frédéric Holzwarth, Universität Leipzig, Germany

#### \*Correspondence:

Andreas Bolte andreas.bolte@thuenen.de

#### Specialty section:

This article was submitted to Functional Plant Ecology, a section of the journal Frontiers in Plant Science

Received: 28 February 2016 Accepted: 17 May 2016 Published: 14 June 2016

#### Citation:

Bolte A, Czajkowski T, Cocozza C, Tognetti R, de Miguel M, Pšidová E, Ditmarová L, Dinca L, Delzon S, ´ Cochard H, Ræbild A, de Luis M, Cvjetkovic B, Heiri C and Müller J (2016) Desiccation and Mortality Dynamics in Seedlings of Different European Beech (Fagus sylvatica L.) Populations under Extreme Drought Conditions. Front. Plant Sci. 7:751. doi: 10.3389/fpls.2016.00751 <sup>1</sup> Thünen Institute of Forest Ecosystems, Eberswalde, Germany, <sup>2</sup> Instituto per la Protezione Sostenibile delle Piante (IPSP), Consiglio Nazionale delle Ricerche, Sesto Fiorentino, Italy, <sup>3</sup> Dipartimento di Bioscienze e Territorio, Università del Molise, Pesche, Italy, <sup>4</sup> EFI Project Centre on Mountain Forests (MOUNTFOR), Edmund Mach Foundation, San Michele all'Adige, Italy, <sup>5</sup> BIOGECO, INRA, Université de Bordeaux, Cestas, France, <sup>6</sup> Institute of Forest Ecology, Slovak Academy of Science, Zvolen, Slovakia, <sup>7</sup> Marin Dracea National Forest Research-Development Institute, Bucharest, Romania, <sup>8</sup> PIAF, INRA, Université Clermont Auvergne, Clermont-Ferrand, France, <sup>9</sup> Department of Geosciences and Natural Resource Management,University of Copenhagen, Frederiksberg C, Denmark, <sup>10</sup> Grupo de Clima, Agua, Cambio Global y Sistemas Naturales, Departamento de Geografía y Ordenación del Territorio, Facultad de Filosofía y Letras, Instituto de Investigación en Ciencias Ambientales, Universidad de Zaragoza, Zaragoza, Spain, <sup>11</sup> Faculty of Forestry, University of Banja Luka, Banja Luka, Bosnia and Herzegovina, <sup>12</sup> Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Birmensdorf, Switzerland

European beech (Fagus sylvatica L., hereafter beech), one of the major native tree species in Europe, is known to be drought sensitive. Thus, the identification of critical thresholds of drought impact intensity and duration are of high interest for assessing the adaptive potential of European beech to climate change in its native range. In a common garden experiment with one-year-old seedlings originating from central and marginal origins in six European countries (Denmark, Germany, France, Romania, Bosnia-Herzegovina, and Spain), we applied extreme drought stress and observed desiccation and mortality processes among the different populations and related them to plant water status (predawn water potential, 9PD) and soil hydraulic traits. For the lethal drought assessment, we used a critical threshold of soil water availability that is reached when 50% mortality in seedling populations occurs (LD50SWA). We found significant population differences in LD50SWA (10.5–17.8%), and mortality dynamics that suggest a genetic difference in drought resistance between populations. The LD50SWA values correlate significantly with the mean growing season precipitation at population origins, but not with the geographic margins of beech range. Thus, beech range marginality may be more due to climatic conditions than to geographic range. The outcome of this study suggests the genetic variation has a major influence on the varying adaptive potential of the investigated populations.

Keywords: Fagus sylvatica, drought, desiccation, mortality, LD50SWA, soil water availability, genetic variation, pre-dawn water potential

## INTRODUCTION

There is much evidence that ongoing climate change is warming the global climate system given the average temperature rise of 0.85◦C for the combined land and ocean surface over the period from 1880 to 2012. And there is strong evidence that an increased frequency of extreme weather events like heat waves and precipitation extremes is linked to global warming (Coumou and Rahmstorf, 2012). Projections of further warming in the Twenty-first century are linked to a likely increase in, and intensification of, heat waves and drought periods, in particular toward the end of the century (IPCC, 2013). For Europe it has been found, that the severity, duration, and frequency of drought events increased from 1950 to 2012 in Mediterranean regions, but moderately also in parts of Central Europe (Spinoni et al., 2015). Similar tendencies are also projected for the future (IPCC, 2012; Stagge et al., 2015). European forests have already responded to more intensive drought impacts with increased mortality (Allen et al., 2010).

The natural vegetation in Central Europe and higher elevated areas in southern Europe is dominated largely by European beech (Fagus sylvatica L.; Bohn et al., 2004; Bréda et al., 2006). Besides being of high economic value, beech is also of ecological importance, since it is the dominant tree species in many forest ecosystems (Leuschner et al., 2006). Beech can grow well on a wide variety of sites except on extremely dry soils with low water storage capacity, stagnic soils, or soils prone to flooding and high ground water table (Ellenberg and Leuschner, 2010). Beech is dominant in many deciduous forests in Europe under maritime and temperate climate conditions with mild winters and moist summer conditions; the pronounced cold, dry, and continental climate limits its distribution (Bolte et al., 2007). As a distinct shade-tolerant tree species, beech itself reduces below-canopy irradiance often below 5% of the open field irradiance (Emborg, 1998; Collet et al., 2001), giving it the competitive advantage over other tree species (Ellenberg and Leuschner, 2010).

However, beech is generally reputed to be sensitive to drought (e.g., Aranda et al., 2000; Gessler et al., 2007) and could lose its competitive advantage to less drought-sensitive species like sessile oak [Quercus petraea (Matt.) Liebl.] under water-limited conditions (Scharnweber et al., 2011). Looking on direct drought impact, beech's vulnerability to cavitation seems to make it extremely sensitive to singular extreme water deficit, and hence to drought (Barigah et al., 2013; Urli et al., 2013). A critical internal water status in beech seedlings is reached at shoot water potential of –1.9 MPa (Hacke and Sauter, 1995) upon which a loss of hydraulic conductivity may eventuate. In case of continuing drought, 50% loss of hydraulic conductivity (P50) can occur between –2.0 MPa and –3.0 MPa (Cochard et al., 1999; Cruiziat et al., 2002). A critical loss of hydraulic conductivity (P88) was found at –4.2 MPa (Urli et al., 2013), and Barigah et al. (2013) reported 50% mortality among beech seedlings at –4.5 MPa plant water potential (xylem pressure). During the extreme drought year 2003, Granier et al. (2007) identified 40 and 20% of relative available soil water content as thresholds below which gross primary production, and total ecosystem respiration decreased respectively. However, there is still no coherent approach to link quantitatively the environmental drought impact, e.g., assessed as soil water deficit, to the desiccation and mortality of tree seedlings. Bréda et al. (1995) and Czajkowski et al. (2009) demonstrated that plant water status can be linked to soil matrix potential at the lower end of the effective rooting zone (ERD). Accordingly, a simultaneous study of soil hydraulic traits and desiccation dynamics may link plant mortality to soil water deficit, not at an individual, but also at a mean population level. Such an indicator can be applied in regional assessments and projections on soil water availability and critical drought risk (e.g., Bolte, 2015).

In Central Europe, beech exhibits high genetic diversity within populations (Vornam et al., 2004), but genetic differentiation between populations is also evident at continental scale (Magri et al., 2006; Dounavi et al., 2016). Accordingly, several studies on young beech seedlings revealed remarkable differences in the adaptive potential of different beech populations to drought: beech populations from the xeric sites and/or range margins seem to have a higher drought tolerance than those from mesic sites and/or central ranges (Italy: Tognetti et al., 1995; Bosnia and Herzegovina: Ivojevic et al., 2012 ´ ; Slovakia: Pšidová et al., 2015; Germany: Schraml and Rennenberg, 2002; Peuke et al., 2002; Poland and Germany: Czajkowski and Bolte, 2006a; Rose et al., 2009; Spain, Bulgaria and Germany: Thiel et al., 2014; Germany, Balkan peninsula, Bulgaria and Greece: Dounavi et al., 2016). This response could be due to population dynamic processes at the southern and eastern margins of the beech distribution range including local, evolutionary adaptation to increasing drought stress on xeric sites (Hampe and Petit, 2005).

Except for the regional study by Ivojevic et al. (2012) ´ , the previous experimental studies on population level focused on seedling growth performance, hydraulic traits, and/or water status under moderate or severe drought, but did not systematically apply severe drought, which induces mortality among the seedlings. Thus, a continental study of beech seedling mortality induced by extreme drought events and variation in mortality among populations level is lacking. Using the pan-European EU Cost STReESS network, we collected seeds from seven sites in six European countries throughout the native beech range and conducted a common garden experiment in Germany to (1) derive a desiccation and mortality indicator at the population level that can be related to soil water availability (SWA), (2) derive critical limits of soil water availability (SWA) for the studied beech populations, and (3) reveal possible population variation in extreme drought response and desiccation.

#### MATERIALS AND METHODS

#### Plant Material

For the experiments, we collected at least 1 kg of fresh beech seeds from four different autochthonous, old-growth beech stands [location see **Table 1**; population Stenderup Midskov (Denmark), Nevesinje (Bosnia), Valea Boronului (Romania), and Erro (Spain)]. The collected seeds originated from at least ten different old-growth beech individuals. Seeds from three other Bolte et al. Desiccation Dynamics in Beech Seedlings

populations originated from commercial seedbanks (Crecy and Montagne Noir, France) and Sellhorn (Germany), which were also collected in single stands. The stand locations cover a large variety of environments within the natural beech distribution range, in particular including geographically marginal sites (**Figure 1**). For the climatic characterization of the population origins, we used temperature and precipitation parameters and the Ellenberg Climate Quotient EQ (Ellenberg, 1988, Equation 1):

$$EQ = \left(\frac{T\_{\text{max.}}}{P\_{\text{year}}}\right) \cdot 1000 \tag{1}$$

where Tmax. is the mean temperature in the warmest month (◦C) and Pyear the total annual precipitation (mm).

Moreover, we applied De Martonne (1926) Aridity index Am (Equation 2):

$$Am = \left(\frac{P\_{year}}{T\_{year} + 10}\right) \tag{2}$$

with the annual mean temperature Tyear ( ◦C). The found ranges of climatic parameters (**Table 1**) cover quite well the climatic range limits of European beech reported by Fang and Lechovicz (2006) with e.g., Tyear ( ◦C) ranging from 7.2 to 13.5 and EQ from 16.8 to 29.0, but not reaching the absolute xeric extremes. However, our exceedance of EQ limits on higher elevated sites (PV6, PV7) may indicate the problem to adequately characterize both lowland and mountainous climatic limits with indices mainly based on annual means, only.

The seeds were collected in autumn 2013, stored and transported in cool, dry environments to the Thünen Institute of Forest Genetics in Groß-Hansdorf (Germany). Uniformly sized seeds of each population were surface-sterilized by soaking in 3% sodium hypochlorite for 5 min and rinsing with deionized water. Thereafter, a stratification procedure was performed: (1) the seed moisture content was reduced to about 8% of the fresh seeds' moisture content (e.g., by storing them ≈1 week in a cool, dry place), (2) seeds were preserved in plastic bags in a freezer at −5 ◦C until mid-February (stratification by frost), (3) the seed moisture was increased at a temperature of 3–5◦C (using a water sprayer); (4) as soon as the first little sprout was visible, the seedlings were transplanted into pots. With this procedure at least 200 individual seedlings per population were available for the drought experiments.

After the success of seed germination was recorded, plants were cultivated in cylindrical PVC pots (1.4 liters) filled with 70% silty sand (grain size 0–2 mm), 30% peat-based substrate mixed with with 2 kg m−<sup>3</sup> Osmocote (NPK 14:13:13+7SO3, plus micro elements). Plants grew under slightly reduced open field light conditions (≈ 70% rel. open field irradiance) in a greenhouse environment.

After transferring the seedlings to the Thünen Institute of Forest Ecosystems in summer 2014, a drought simulation was carried out in late summer 2014 in a greenhouse at the University of Sustainable Development (HNE) in Eberswalde (52◦ 49′ 28 " N 13◦ 47′ 29" E, 30 m a.s.l.). Within the treatment period relative air humidity averaged 69%, with a minimum of 30% and a maximum of 88%. Air temperature ranged between 11◦C (minimum during night) and 31◦C (maximum during day), and attained a mean of 19.0◦C. The plants grew under ambient light conditions during the experiment without any additional illumination. During the experiment the light intensity never exceeded 1000µmol photons m−<sup>2</sup> s <sup>−</sup><sup>1</sup> under sunny conditions.

Plant traits before the commencement of the drought experiment (**Table 2**) show some variation in root collar diameter, plant height and leaf number among the populations, but a common pattern across populations was not observed. No significant differences were found for total leaf area.

#### Experimental Set-up

For the experimental drought simulation, 100 seedlings per population were arranged in the two variants: "control" (C) in 20 pots and "drought treatment" (D) in 80 pots. The seedlings of the different populations were kept together in groups on trolleys in the greenhouse, but population groups were randomly moved and thus spatially re-arranged every 3 days. The group of "control" seedlings was maintained close to field capacity (FC) by frequent watering; whereas, water supply was suspended for those seedlings subjected to drought treatments. Before beginning the drought experiment, initial soil water content and soil dry weight was assessed by weighing samples of the used soil substrate before and after oven drying at 105◦C for 48 h. Pots then were watered to saturation. After excess water had drained away, field capacity (FC, Blume et al., 2016) was reached at around –0.06 MPa soil water potential (pF 1.8), and the initial field capacity (FC) pot weight was derived. By subtracting the soil dry weight from FC pot weight we derived the initial soil water content at field capacity. Subsequent changes in pot weight were attributed to changes in soil water content.

The available water capacity (θAWC) of the soil was derived using following Equation (3, cf. Veihmeyer and Hendrickson, 1927).

$$
\theta\_{\rm AWC} = \theta\_{\rm FC} - \theta\_{\rm PWP}, \tag{3}
$$

where θ is the soil water content [g] at field capacity (FC, pF 1.8 ≈ −0.06 MPa soil water potential) and at the permanent wilting point (PWP, pF 4.2 ≈ −1,5 MPa soil water potential). θPWP was derived from a soil water characteristic (pF) curve established for the used soil substrate. With this definition we follow the concept of Reid et al. (1984) who induced the term available soil water for laboratory assessments in contrast to extractable soil water for field estimates (Ritchie, 1981).

The residual soil water availability (SWA) [%] (Equation 4) is defined as the actual soil water content (θt) [g] during drought treatment expressed as a percentage of the initial available soil water capacity (θAWC) [g], and corresponds to the relative extractable soil water (REW) in field studies, Granier et al., 2007):

$$SWA = \frac{\Theta\_l}{\Theta\_{AWC}}\tag{4}$$

To assess SWA, each pot was weighed three times per week after watering was stopped. The treatment started in mid-summer (06/08/2014) and lasted for 8 weeks until all seedlings were considerably desiccated.


TABLE 1 | Temperature (T) and precipitation (P) [year, growing season from April (4) to September (9)] at the origin of the seedling populations, derived from WorldClim grid data (Source: http://www.worldclim.org/current, period 1950–2000, ESRI grid, resolution 30 s, ca. 1 km2).

Altitude values represent the means of the 30 s grid cell.

<sup>a</sup>EQ: Ellenberg Climate Quotient.

<sup>b</sup>Am: Aridity index of De Martonne.

FIGURE 1 | Location of the origins of the investigated populations (circles), and continuous distribution range of European beech (gray area) based on the distribution map of Bolte et al. (2007).

#### Desiccation and Mortality Assessments

During the drought treatment, the advanced plant desiccation process was monitored by measuring individual pre-dawn leaf water potentials (9PD) with the Scholander chamber technique (Scholander et al., 1964, using the Plant Moisture Vessel Skye SKPM 1400, Skye Instruments, Llandrindod Wells, UK). 9PD was measured between 0:00 and 5:00 (UT). Seedlings with first optical signs of wilting were measured during the desiccation process. They were regarded as dead when signs of complete wilting occurred with yellow-brown discoloration of the entire leaf surface. To control the status of complete cavitation (>88% loss of hydraulic conductivity at MPa < –6 MPa) we measured pre-dawn water potential of the wilted plants.

The completely wilted plants were separated from the treatment group and re-watered. The date of obvious mortality was recorded. This mortality definition neglects the possibility of wilted beech seedling resprouting after re-watering that were assessed in the following spring 2015. However, the majority of the few found resprouted beech plants died in the days and weeks later due to unspecific reasons which made the viability re-assessment unreliable.

#### Derivation of Critical Drought LD50

For comparing the mortality dynamics of the different beech seedling populations, we adopted the approach for drought impact analyses by Kursar et al. (2009). Due to this, the median lethal desiccation (LD50) describes the drought impact that leads


TABLE 2 | Means (± standard error) of plant traits for the beech seedlings before the drought stress experiment.

Means followed by different letters are significantly different at p < 0.05 (ANOVA, test of population differences, comparison downwards), means of leaf area are not significantly different.

to 50% mortality in the seedling population in comparison to the control treatment (cf. also Ivojevic et al., 2012 ´ ). In our study, LD50SWA defines the drought impact as the residual soil water availability (SWA [%]), which is linked to a 50% mortality rate in the population according to previously reported mortality definition.

The critical soil water availability (LD50SWA) per seedling population (drought treatment) was derived from a doseresponse analysis of mortality rate M (Equation 5) and survival rate S (Equation 3) as a function of soil water availability (SWA) depletion over time:

$$M\_{SWA} = \frac{\sum\_{SWA\_i}^{SWA} m\_a}{n\_a},\tag{5}$$

where m<sup>a</sup> is the number of dead plants m in population a, n<sup>a</sup> is the number of total plants per population a in the drought experiment (na) and period between initial soil water availability SWA<sup>i</sup> and current soil water availability SWA.

The survival rate S (Equation 6) was then calculated from the mortality rate M:

$$S\_{SWA} = 1 - M\_{SWA} \tag{6}$$

The survival rate S (range 0–1) was fitted by a non-linear regression analysis applying the software package SAS JMP 11.0 (SAS Institute Inc, 2014). For this we used a two-parameter logistic model (2PL) of the following form to derive the survival function s (Equation 7) related to soil water availability (SWA) depletion over time:

$$\mathcal{S}\_{SWA} = \frac{1}{1 + e^{\left[-\mathcal{S}\_0\left(\mathcal{S}WA - \mathcal{S}\_1\right)\right]}},\tag{7}$$

where two empirical parameters describe the growth rate (ß0) and the inflection point (ß1).

For the symmetric 2PL model used, the LD50SWA values of the different provenances equate with inflection points (ß1) at SSWA = 0.5 (Gregorczyk, 1991; SAS Institute Inc, 2014).

We tested the fitted models between the different populations for parallelism using a F-Test. The test compares the error sumsof-squares for a full and a reduced model. The full model gives each group different parameters. The reduced model forces the groups to share every parameter except for the inflection point. Moreover, the equality of model parameters across the levels of the populations, used as a grouping variable, was considered. With a comparison of parameter estimates (CPE), including an Analysis on Means (ANOM), the population means are tested against the overall mean.

The effect of decreasing soil water availability (SWA) on the plant internal water status, indicated by the predawn water potential, is indicative for the loss of water conductivity and cavitation, finally leading to hydraulic failure (e.g., Urli et al., 2013). Thus, besides relationships between LD50SWA and climate variables at population origin also correlations between soil water availability (SWA) and mean predawn water potentials (9PD) of the seedlings were analyzed by single linear regression analyses and F-test. Before the regression analysis (SWA vs. 9PD) we multiplied the 9PD values by −1 to derive positive values and then log-transformed both parameters. A linear model was fitted, and values and model were then re-transformed [log (SWA, –9PD)] resulting in a non-linear power function as a nonlinear regression model. A bias correction was not applied. We tested the equality of the model across the populations using the already above mentioned tests on parallelism, CPE and ANOM. For all statistical analyses described and modeling purposes, p < 0.05 was considered significant.

#### RESULTS

#### Soil Water Availability and Seedling Mortality

During the drought treatment, initial mortality of the seedlings was observed between 34 and 43 days from the commencement of the experiment. At the end of the experiment, mortality ranged between 33% (PV4, Montagne Noir, France) and 71% (PV1, Stenderup Midtskov, DK). A considerable increase in seedling mortality occurred when soil water availability (SWA) fell below values of 30–20% (**Figure 2**). However, differing responses between populations were found with respect to seedlings mortality dynamics under soil water depletion. The largest differences were found between PV1 (Stenderup Midtskov, DK) and PV3 (Crecy, FR). PV1 mortality started late (22% SWA), but had the strongest increase in mortality (growth rate ß<sup>0</sup> ≈ 0.85, **Table 3**) overtaking all other populations in final mortality (0.62). In contrast, PV3 mortality began already at 27% SWA, followed by a retarded progress in mortality (growth rate ß<sup>0</sup> ≈ 0.24,

**Table 3**), not reaching 50% mortality at the end of the drought simulation. The mortality dynamics in terms of growth rate of the other populations were within this range. Correspondingly, growth rate (ß0) varies significantly from the overall mean parameter for PV1 (Stenderup Midtskov, DK) by exceeding the upper limit (UPL) and, for PV3 (Crecy, FR), by undershooting the lower limit (LWL) according to the comparison of parameter estimates (CPE, **Table 3**). The different shape of the fitted models (ß0) was also significant according to a parallelism F-test (F value 6.063, p < 0.0001).

The LD50SWA values corresponded to the inflection point of the model (ß1, **Table 3**). High LD50SWA values were found for the populations PV2 (Sellhorn, DE), PV5 (Valea Boronului) and PV7 (Erro, ES), indicating high drought sensitivity (**Figure 2**). Low LD50SWA values were found for PV3 (Crecy, FR) and also PV1 (Stenderup Midtskov, DK). LD50SWA of all populations differed significantly from an overall mean except for PV4 (Montagne Noir, FR) looking on CPE results (**Table 3**).

The analyses revealed that seedlings' mortality dynamics and the critical threshold for drought impact indicated by LD50SWA differ significantly among the selected populations. The most drought tolerant population in our experiment was PV3 (Crecy, FR) whereas the populations from higher elevations (PV5, PV6, and PV7) and northern origin (PV2) were drought sensitive. The most northern population (PV1, Stenderup Midtskov, DK) exhibited a remarkably strong drop in seedling survival that revealed sudden drought mortality risk for low SWA. A considerable extrapolation of 50% mortality is visible when applying the model to the two French populations (PV3, PV4), and thus the LD50SWA values for both populations have to be considered with care. However, the extrapolated LD50SWA values are supported by the clearly retarded mortality dynamics below 20% remaining SWA and the lower (negative) growth rate (ß0) of the regression model for both French populations compared to the other ones.

#### Relationships between LD50SWA and Climate Variables

The critical soil water availability (LD50SWA) correlated significantly (p < 0.05, r = 0.73) with the mean growing season precipitation (Prec. 4–9, **Figure 3**, middle below). This relationship did not correspond to the geographical North-South gradient of the population origin, but is more influenced by the altitudinal precipitation gradient. No statistical relationships were found for temperature parameters (Ty, T4−9, T. max, latter not shown). Some tendencies are visible for mean annual precipitation sum (Py) and the climate indices used, which combined temperature and precipitation parameters (EQ, Am), but here the correlations between the climate parameter and the LD50SWA values were not significant.

### Soil Water Availability and Internal Water Status

The soil water availability (SWA) was closely correlated to the internal water status of the beech seedlings considered by the predawn water potentials (9PD, **Figure 4A**). Due to observed heteroscedasticity the estimates are not unbiased. The figure shows population means of SWA and pre-dawn potentials of selected plants with signs of desiccation (treatment) or irrigated control plants (control) without drought stress. Plants without desiccation or indication of visible wilting during the drought treatment were not included. Mean values refer to nine dates during the experiment between the 33th and 61st day after its start. A distinct change in 9PDvalues was visible when SWA dropped below 20%, corresponding to 9PD of −2MPa. Small decreases in SWA below this threshold led to a strong drop in 9PD values in wilting plants, which correspond to the mean mortality dynamics shown in **Figure 2**. In contrast to mean mortality dynamics, the tests on parallelism and equality of the model parameters gave no significant indication of variation across the populations (p < 0.05). Thus we used the general regression model to estimate the mean predawn water potentials


TABLE 3 | Non-linear regression model parameters (growth rate ß<sup>0</sup> , inflection point ß<sup>1</sup> ) and their standard error (SE) for predicting survival of beech seedlings from soil water availability (SWA), see Equation (4, 2PL) and Figure 2.

Parameter estimates are compared against the overall mean (CPE) with an Analysis of Means (ANOM), upper (UPL) and lower decision limit (LWL) is shown (α = 0.05). Bold parameter values (ß0, ß1) deviate significantly from the overall mean. Overall goodness of fit measures: Akaike information criterion AIC<sup>C</sup> = −295.87, SSE = 0.059, MSE = 0.00093, r<sup>2</sup> = 0.97.

right: climate indices with Ellenberg Climate Quotient (EQ, above) and Aridity index of De Martonne (Am, below). The linear regression line displays a significant predictor effect of precipitation during the growing season (P4−9 ) on LDSOswA (seep values).

(9PD) when 50% mortality was reached (LD50SWA) for the different populations (**Figure 4B**, inlayed figure). The large variation of 9PD values from nearly –5 MPa for PV3 (Crecy, FR) to –2.3 MPa for PV7 (Erro, ES) is induced by variation in LD50SWA below 20% SWA.

#### DISCUSSION

## LD50 as a Critical Threshold of Drought Impact

The outcome of our study demonstrates that the derived LD50SWA indicator is useful for analyzing the drought sensitivity of young trees. L50 was developed and first applied as a lethal dose or concentration indicator referring to 50% mortality of organism populations (Cavalli-Sforza, 1972) for dose-response analysis in the field of toxicology. In plant ecology, it was quite commonly used for lethal temperature (frost) impact on plants, including also trees (LT50, e.g., Zhang and Willison, 1987; Barranco et al., 2005; Kreyling et al., 2014; Hofmann et al., 2015). Some examples for the use of L50 approaches to indicate drought impact (LD50) considered exposure time only (Ivojevic et al., ´ 2012; Granda et al., 2015). Results of those studies, however, are only valid for the specific experimental environments used (e.g., pot size, soil substrate and plant material) and cannot be generalized or transferred to other environments. Kursar et al. (2009) presented an alternative approach of using leaf water status (relative leaf water content RWC, leaf water potentials 9) as a quantitative plant-related parameter for lethal drought assessment (LD50RWC,9) providing more general results for tree species. Our LD50SWA follows this idea in general, but uses soil water availability (SWA), which can be consistently assessed for different soil substrates in relation to different absolute available soil water amounts (cf. Meir et al., 2015). This provides new

possibilities in soil water modeling for plant-related drought risk approaches (Bolte, 2015). However, a reference soil depth has to be defined describing the soil-root interface for water uptake, generally defined as effective rooting depth (ERD, cf. Czajkowski et al., 2009). This concept is supported by the simultaneous study of soil water and plant water status along the rooting gradient in mature oak stands in France, which reveals substantial water depletion dynamics down to the lower end of rooting zone corresponding to ERD (Bréda et al., 1995). In our experiment the entire pot depth that was completely rooted at the end of the experiment was regarded as ERD.

potentials when the different beech populations reached LDSOSWA using the

relationship described in Figure 4A.

Our LD50SWA indicator is a simplifying statistical indicator for drought impact at the population level, complementing and not replacing functional assessments and theories of extreme drought impact and plant mortality at an individual level (in particular hydraulic failure theory, Sperry et al., 1998; Brodribb and Cochard, 2009; Barigah et al., 2013). It also has to be considered in the context of other parameters like the mortality dynamics with decreasing SWA (**Table 3**, slope of regression model ßo). However, the L50SWA result range of about 10–18% lethal soil water availability shown fits well to reported threshold of 20% available soil water told to induce strong effects, but not automatically mortality in mature trees and stands like the drop of whole tree hydraulic conductance (Domec et al., 2015) and the decrease in total ecosystem respiration TER (Granier et al., 2007). Thus, we regard LD50SWA as a valid indicator that links plant-internal water status to soil hydraulics and by this provide novel possibilities for climate—soil water modeling and regionalisation of drought risk from plant to landscape and regional level. Recently, this approach was used for modeling the recent and future risk of lethal drought impact on beech regeneration by assessing period length below the LD50SWA value under the canopy of mature stands of Norway spruce, Scots pine, and European beech on the national scale in Germany (Bolte, 2015).

### Different Drought Response of Populations

Both our significant genetic differentiation of in LD50SWA values (≈ 10–18%, **Figure 2**, **Table 3**) and the varying mortality dynamics (ß0, **Table 3**) among the different populations support the idea of local adaptation of populations within the European beech range. This is in line with many other studies on (1) leaf phenology (Wuehlisch et al., 1995; Chmura and Rozkowski, 2002; Nielsen and Jørgensen, 2003; Cufar et ˇ al., 2012; Robson et al., 2013), (2) cambium, xylem and phloem phenology (Prislan et al., 2013; Martínez del Castillo et al., 2016), (3) frost tolerance (Visnjic and Dohrenbusch, 2004; Czajkowski and Bolte, 2006b; Kreyling et al., 2014), and (4) drought response (Tognetti et al., 1995; García-Plazaola and Becerril, 2000; Peuke et al., 2002; Schraml and Rennenberg, 2002; Czajkowski and Bolte, 2006a; Rose et al., 2009; Ivojevic et al., 2012; Eilmann et al., 2014; ´ Thiel et al., 2014; Pšidová et al., 2015; Dounavi et al., 2016). Some studies, however, found indifferent or even contradicting results (Baudis et al., 2015; Hofmann et al., 2015) after comparing populations along a smaller geographic and climatic gradient within the continuous beech range (cf. Knutzen et al., 2015). Also Wortemann et al. (2011) found no evidence for genetic differentiation across beech populations for vulnerability to embolism by comparing European populations originating from the continuous distribution range of beech, only.

The adaptive potential of European beech, and other plant organisms, to drought and other climatic extreme events is triggered by two main processes: (1) genetic variation and/or (2) phenotypic plasticity (Meier and Leuschner, 2008; Lindner et al., 2010; Aranda et al., 2015). Genetic diversity of beech is mainly shaped by its phylogeographic history during the Pleistocene and Holocene (Harter et al., 2015). The isolated location of Pleistocene refuge areas and re-colonization pathways were indicative for large-scale genetic differentiation in Central European and Mediterranean distributions (Magri et al., 2006). Isolation during the highly variable interglacial climate conditions in the Pleistocene played a major role in increasing the genetic complexity of extant refuge populations, only partly preserved during the post-Pleistocene re-colonization toward north ("southern complexity" and "northern purity" paradigm, de Lafontaine et al., 2013). However, this interferes with recent evolutionary adaptation processes at the local level, occurring over only one or a few generations (Hamrick, 2004), when extreme weather events like droughts induce directed selection processes (Aitken et al., 2008; Spathelf et al., 2015). In particular for beech, marginal populations at the xeric distribution boundary are reputed to be the focus of local adaptation to drought, reducing genetic variation of local populations (Hampe and Petit, 2005) that exist in heterogeneous environments (Pluess et al., 2016).

The close correlation found between precipitation during the growing season (Prec. 4–9) at the population origins and the critical drought thresholds (LD50SWA) of the populations (**Figure 3**) suggests for local adaptation brought about mainly by recent evolutionary adaptation. This would explain also the fact that the actual precipitation conditions are indicative for the drought tolerance found and not the southern origin of the population near or even in Pleistocene refuge areas. In this sense, the distribution margin of beech, and thus the location of marginal beech populations, needs to be interpreted more in an ecological sense as beech occurrence near to its xeric limits rather than geographically by southern or eastern marginal location (cf. Hampe and Petit, 2005). This would mean that "ecologically" marginal populations due to local or regional xeric conditions may also occur within the continuous distribution range.

## Extreme Drought Adaptation, Desiccation Tolerance, and Mortality of Beech

Our findings underline the importance of assessing the adaptation of beech to drought at the intraspecific level. Hydraulic trait variations are seen as a major reason for different drought responses of tree populations within the species distribution range (Lamy et al., 2011; Balducci et al., 2015). Ecophysiological measurements (gas exchange, chlorophyll fluorescence) conducted alongside our drought experiment (Cocozza et al., personal communication) revealed differences in functional traits among the beech populations, but found no clear gradient in relation to location and climatic conditions at population origins. This addressed, however, mainly the drought response phase until complete stomata closure and considerable loss of hydraulic conductivity (at around 20% SWA and 9PD ≈ −2 MPa, **Figure 4A**, cf. Hacke and Sauter, 1995; Cochard et al., 1999; Cruiziat et al., 2002). This is, however, decoupled from later desiccation and mortality dynamics (Delzon and Cochard, 2014). More than 90% loss of hydraulic conductivity of beech seedlings and young stands is reputed to be reached between −2.2 MPa (Magnani and Borghetti, 1995) and −4.0 MPa (Cochard et al., 1999). Advanced mortality in young beech was found at mean xylem water potentials of −4.5 MPa (Barigah et al., 2013, and this study). Furthermore, a recent study of mature beech in Germany revealed 88% of conductivity loss (P88) at xylem pressure means between −4.0 and −4.5 MPa (Schuldt et al., 2015). These findings fit well to our estimated variation of −2.3 MPa and nearly −5.0 MPa 9PD at LD50SWA when 50% mortality have occurred (**Figure 4B**). The outcome also strongly supports the idea of Delzon and Cochard (2014) that 50% mortality is linked to the almost complete loss of hydraulic conductivity (P88) in angiosperm trees like beech. Thus, LD50SWA and P88 seem to represent corresponding indicators for lethal drought in beech and probably other angiosperms.

For plant survival under extreme drought, the ability to prolong the desiccation process and keep hydraulic integrity as long as possible seems to be a key adaptive issue (Bréda et al., 2006). In general, desiccation tolerance in plants involves the capacity to avoid deleterious effects of water shortage on the cellular membranes and maintain the bilayer structure in a xeric environment (e.g., Leprince et al., 1993). However, for taller vascular plants such as trees with complex hydraulic architecture, the resistance to cavitation and xylem embolism is by far the most important feature for desiccation tolerance (Lüttge et al., 2011). Our results suggest that there should be intra-specific variation in (1) morphological traits avoiding uncontrolled leaf water losses and/or (2) resistance to cavitation and hydraulic failure. Genetic variability in cavitation resistance is not clear for European beech, yet (Wortemann et al., 2011), but has been described in combination with morphological adaptation for Holm oak ecotypes (Quercus ilex, Peguero-Pina et al., 2014). Moreover, for beech populations significant differences in xylem anatomy (vessel size and vessel density) were found by Eilmann et al. (2014), which clearly point to higher drought resistance of a southern Bulgarian population from more xeric environments compared to those from mesic environments.

## CONCLUDING REMARKS

Our study demonstrates that the introduced LD50SWA indicator is a feasible indicator for critical soil water availability (SWA) in relation to plant desiccation and mortality. Thus, a residual SWA of 20% represents a critical limit (Granier et al., 2007), below which the risk of beech seedling mortality increases drastically. Also the correspondence of our LD50SWA indicator with the P88 indicator found to describe a lethal water status in angiosperms (Delzon and Cochard, 2014) enables novel links for coupling ecophysiological and statistical mortality assessments. These insights provide new possibilities for local and regional modeling of drought risks based on soil water balance modeling. The significant intraspecific variation in survival under extreme drought (LD50SWA and mortality dynamics) found can be used for the pre-selection of beech populations identified as especially apt for coping with the future climate. Further testing of these populations would be needed as well as more research on how this knowledge could apply in forest management aiming to increase our forests resistance to climate change. The differences revealed between "geographically" marginal and "climatically" marginal beech populations should be a matter of further research since common ideas of adaptive marginal populations may be biased, in particular due to the varying high-altitudinal location of southern population. Further research gaps include (1) the morphological and physiological background of genetic variation of adaptation and (2) the contribution of genetic variability and phenotypic plasticity to adaptive potentials of European beech.

## AUTHOR CONTRIBUTIONS

All authors (AB, TC, CC, RT, MM, EP, LubD, LucD, SD, HC, AR, ML, BC, CH and JM) contributed substantially to the writing of the manuscript. AB, TC, CC, RT, MM, EP, LubD, and JM drafted the conceptual design with the help of the author group and conducted the study. In addition, LucD, AR, ML, BC collected and delivered seed material for the study.

#### ACKNOWLEDGMENTS

We are grateful to Dr. Mirko Liesebach and Rainer Ebbinghaus from the Thünen Institute of Forest Genetics (Groß-Hansdorf, Germany) for the cultivation of the beech seedlings used in this experiment. Moreover we thank Prof. Dr. Harald

#### REFERENCES


Cavalli-Sforza, L. (1972). Biometrie. Stuttgart: Fischer.


Schill and Dr. Bernhard Götz from the University of Sustainable Development (HNE Eberswalde) for giving the opportunity to use the greenhouse facilities. This article is based upon work from COST Action FP1106 STReESS, supported by COST (European Cooperation in Science and Technology).


vulnerability of European forest ecosystems. For. Ecol. Manage. 259, 698–709. doi: 10.1016/j.foreco.2009.09.023


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Bolte, Czajkowski, Cocozza, Tognetti, de Miguel, Pšidová, Ditmarová, Dinca, Delzon, Cochard, Ræbild, de Luis, Cvjetkovic, Heiri and Müller. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.