# ANTHROPOGENIC IMPACTS ON THE MICROBIAL ECOLOGY AND FUNCTION OF AQUATIC ENVIRONMENTS

EDITED BY: Maurizio Labbate, Justin R. Seymour, Federico Lauro and Mark V. Brown PUBLISHED IN: Frontiers in Microbiology

### *Frontiers Copyright Statement*

*© Copyright 2007-2016 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 978-2-88919-939-6 DOI 10.3389/978-2-88919-939-6

### 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

# **ANTHROPOGENIC IMPACTS ON THE MICROBIAL ECOLOGY AND FUNCTION OF AQUATIC ENVIRONMENTS**

Topic Editors:

**Maurizio Labbate,** University of Technology Sydney, Australia **Justin R. Seymour,** University of Technology Sydney, Australia **Federico Lauro,** Nanyang Technological University, Singapore **Mark V. Brown,** University of New South Wales, Australia

Harmful algal blooms in a large pond in Varanasi, India (Photograph by R. P. Rastogi). Adapted from: Rastogi, R. P., Madamwar, D., and Incharoensakdi, A. (2015). Bloom dynamics of cyanobacteria and their toxins: environmental health impacts and mitigation strategies. Front. Microbiol. 6, article 1254.

Aquatic ecosystems are currently experiencing unprecedented levels of impact from human activities including over-exploitation of resources, habitat destruction, pollution and the influence of climate change. The impacts of these activities on the microbial ecology of aquatic environments are only now beginning to be defined. One of the many implications of environmental degradation and climate change is the geographical expansion of disease- causing microbes such as those from the *Vibrio* genus. Elevating sea surface temperatures correlate with increasing *Vibrio* numbers and disease in marine animals (e.g. corals) and humans. Contamination of aquatic environments with heavy metals and other pollutants affects microbial ecology with downstream effects on biogeochemical cycles and nutrient turnover. Also of importance is the pollution of aquatic environments with antibiotics, resistance genes and the mobile genetic elements that house resistance genes from human and animal waste. Such contaminated environments act as a source of resistance genes long after an antibiotic has ceased being used in the community. Environments contaminated with mobile genetic elements that are adapted to human commensals and pathogens function to capture new resistance genes for potential reintroduction back into clinical environments. This research topic encompasses these diverse topics and describes the affect(s) of human activity on the microbial ecology and function in aquatic environments and, describes methods of restoration and for modelling disturbances.

**Citation:** Labbate, M., Seymour, J. R., Lauro, F., Brown, M. V., eds. (2016). Anthropogenic Impacts on the Microbial Ecology and Function of Aquatic Environments. Lausanne: Frontiers Media. doi: 10.3389/978-2-88919-939-6

# Table of Contents

*07 Editorial: Anthropogenic Impacts on the Microbial Ecology and Function of Aquatic Environments*

Maurizio Labbate, Justin R. Seymour, Federico Lauro and Mark V. Brown

## **Impact of climate change on aquatic microbial systems**

*10 The emergence of* **Vibrio** *pathogens in Europe: ecology, evolution, and pathogenesis (Paris, 11–12th March 2015)*

Frédérique Le Roux, K. Mathias Wegner, Craig Baker-Austin, Luigi Vezzulli, Carlos R. Osorio, Carmen Amaro, Jennifer M. Ritchie, Tom Defoirdt, Delphine Destoumieux-Garzón, Melanie Blokesch, Didier Mazel, Annick Jacq, Felipe Cava, Lone Gram, Carolin C. Wendling, Eckhard Strauch, Alexander Kirschner and Stephan Huehn


Jessica Tout, Nachshon Siboni, Lauren F. Messer, Melissa Garren, Roman Stocker, Nicole S. Webster, Peter J. Ralph and Justin R. Seymour


### **Impact of pollution on aquatic microbial systems**

*59 Environmental and Sanitary Conditions of Guanabara Bay, Rio de Janeiro* Giovana O. Fistarol, Felipe H. Coutinho, Ana Paula B. Moreira, Tainá Venas, Alba Cánovas, Sérgio E. M. de Paula Jr., Ricardo Coutinho, Rodrigo L. de Moura, Jean Louis Valentin, Denise R. Tenenbaum, Rodolfo Paranhos, Rogério de A. B. do Valle, Ana Carolina P. Vicente, Gilberto M. Amado Filho, Renato Crespo Pereira, Ricardo Kruger, Carlos E. Rezende, Cristiane C. Thompson, Paulo S. Salomon and Fabiano L. Thompson

*76 Bacterioplankton Dynamics within a Large Anthropogenically Impacted Urban Estuary*

Thomas C. Jeffries, Maria L. Schmitz Fontes, Daniel P. Harrison, Virginie Van-Dongen-Vogels

*93 Bloom Dynamics of Cyanobacteria and Their Toxins: Environmental Health Impacts and Mitigation Strategies*

Rajesh P. Rastogi, Datta Madamwar and Aran Incharoensakdi

*115 Patterns of benthic bacterial diversity in coastal areas contaminated by heavy metals, polycyclic aromatic hydrocarbons (PAHs) and polychlorinated biphenyls (PCBs)*

Grazia Marina Quero, Daniele Cassin, Margherita Botter, Laura Perini and Gian Marco Luna

# **Pollution of aquatic systems with antimicrobials and antimicrobial resistance genes**


# **Impact of other disturbances on aquatic microbial communities**


Ana Paula B. Moreira, Pedro M. Meirelles, Eidy de O. Santos, Gilberto M. Amado-Filho, Ronaldo B. Francini-Filho, Cristiane C. Thompson and Fabiano L. Thompson


Elisa L.-Y. Tan, Mariana Mayer-Pinto, Emma L. Johnston and Katherine A. Dafforn

# **Methods of restoration and for measuring disturbances in aquatic microbial communities**


Anne E. Bernhard, Courtney Dwyer, Adrian Idrizi, Geoffrey Bender and Rachel Zwick

*221 A network-based approach to disturbance transmission through microbial interactions*

Dana E. Hunt and Christopher S. Ward

*229 Next-generation sequencing (NGS) for assessment of microbial water quality: current progress, challenges, and future opportunities*

BoonFei Tan, Charmaine Ng, Jean Pierre Nshimyimana, Lay Leng Loh, Karina Y.-H. Gin and Janelle R. Thompson

# Editorial: Anthropogenic Impacts on the Microbial Ecology and Function of Aquatic Environments

Maurizio Labbate<sup>1</sup> \*, Justin R. Seymour <sup>2</sup> , Federico Lauro3, 4 and Mark V. Brown<sup>5</sup>

*<sup>1</sup> School of Life Sciences and the ithree Institute, University of Technology Sydney, Sydney, NSW, Australia, <sup>2</sup> Climate Change Cluster, University of Technology Sydney, Sydney, NSW, Australia, <sup>3</sup> Singapore Centre on Environmental Life Sciences Engineering, Nanyang Technological University, Singapore, Singapore, <sup>4</sup> Asian School of the Environment, Nanyang Technological University, Singapore, Singapore, <sup>5</sup> School of Biotechnology and Biomolecular Sciences, University of New South Wales, Sydney, NSW, Australia*

Keywords: pollution and global change, climate change impacts, aquatic microbiology, resistance gene reservoirs, microbial community disturbance

**The Editorial on the Research Topic**

### **Anthropogenic Impacts on the Microbial Ecology and Function of Aquatic Environments**

Since, the beginning of the industrial revolution, anthropogenic pressures on natural environments have steadily escalated due to increased human population size, modification and destruction of habitats, and pollution. This has led to ecosystem degradation and widespread extinction of plant and animal species. In recognition of this unprecedented human impact, a new geological time period, labeled the Anthropocene, has been defined as the time period during which humans have significantly impacted the Earth's geological and ecological systems (Waters et al., 2016). During the last 40 years, a growing awareness of the negative impacts of humankind on the Earth has led to enactment of government legislation and a degree of modulation of human behavior. However, the emphasis of these changes in attitude and practice has typically been on the conservation of plants and animals, while the effect of anthropogenic activity on natural microbial populations has been largely ignored. This is a significant oversight because microbial communities are generally the first responders to environmental perturbation and can either augment or buffer environmental shifts via, often complex, positive, and negative feedback loops.

Microbes are the most abundant organisms in aquatic ecosystems playing key roles in ecosystem productivity and biogeochemistry. The impacts of anthropogenic activity on the ecology and function of aquatic microbial assemblages are multifarious and often largely undefined, but the advent of powerful new tools including next generation sequencing and novel modeling approaches has begun to shed light on this important question (Hunt and Ward; Tan et al.). The publications presented in this Research Topic are diverse with many, perhaps unsurprisingly, that address the impacts of two of the most important problems facing microbes in aquatic systems, climate change, and pollution.

One of the major impacts of climate change on marine and aquatic ecosystems is an increase in sea surface temperature (SST). In their contribution, Tout et al. demonstrated that increasing SST disrupts the Pocillopora damicornis coral microbiome, increasing the occurrence of Vibrio species and in particular the coral pathogen V. corallilyticus. In light of previous evidence that V. corallilyticus can lead to coral bleaching (Ben-Haim and Rosenberg, 2002), these patterns provide evidence for an additional mechanism in the mass bleaching events that occur during elevated SSTs. Similar observations of increasing influence of microbial pathogens due to rising SST are being made in other systems. The Petton et al. contribution demonstrated that increasing SSTs are increasing the threat to an important aquaculture species, the Pacific Oyster (Crassostrea gigas), which was found to be more

### Edited by:

*Jonathan P. Zehr, University of California, Santa Cruz, USA*

### Reviewed by:

*Matthew Church, University of Hawaii at Manoa, USA*

> \*Correspondence: *Maurizio Labbate maurizio.labbate@uts.edu.au*

### Specialty section:

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

Received: *16 May 2016* Accepted: *22 June 2016* Published: *06 July 2016*

### Citation:

*Labbate M, Seymour JR, Lauro F and Brown MV (2016) Editorial: Anthropogenic Impacts on the Microbial Ecology and Function of Aquatic Environments. Front. Microbiol. 7:1044. doi: 10.3389/fmicb.2016.01044* susceptible to the ostreid herpes virus (OsHV-1) at an elevated SST. As with the Tout et al. study, disruption of the C. gigas microbiome and replacement with Vibrio species was revealed to be an important factor in the full expression of disease by OsHV-1 (Petton et al.).

In Europe, warming SST has led to increases in Vibrio infections in both humans and marine animals (summarized in Le Roux et al.), which catalyzed the first European workshop (Paris, 2015) aimed at forming a research network to improve understanding of the factors driving Vibrio disease events and their impact on human and ecosystem health and, food security (Le Roux et al.). Vibrio disease events and pathogen evolution are inextricably linked to how humans interact with their aquatic environment. For example, the dense human population and low-lying geography of the river deltas is hypothesized to have allowed regular intrusion of Vibrio cholerae into human drinking water allowing for adaptation of this deadly pathogen to the human gut (Boucher et al.). Based on the conclusions of the workshop, rising SSTs are likely to result in increased and/or more severe disease outbreaks requiring further research to develop strategies for prevention and mitigation. Continuing with the theme of climate change, a potential partial solution to this problem is CO<sup>2</sup> capture and storage in sub-seabed reservoirs however, Rastelli et al. sounded a warning that CO<sup>2</sup> leakage from such reservoirs could significantly impact benthic microbial communities affecting carbon cycling and nutrient regeneration processes.

In addition to the effects of climate change, anthropogenicallyderived water pollution is another major impact within aquatic ecosystems. Contributions to this special issue indicate that influx of pollutants and nutrients into aquatic systems significantly disrupts the structure and function of natural microbial assemblages, leading to reduced species diversity, increased heterotrophy and rises in the numbers of potentially virulent/toxic microbes (Moreira et al.; Quero et al.; Jeffries et al.). Some of these localized pollution events can also amplify the effects of global-scale climate change, by further stimulating events such as the Vibrio blooms described above or cyanobacterial or algal blooms that lead to mortality of aquatic animals through anoxia or production of toxins (Rastogi et al.).

Beyond nutrient pollution, antibiotics and antibioticresistance genes are emerging as significant pollutants in aquatic ecosystems (Zhang et al.). The worsening antimicrobial resistance crisis is now being acknowledged as a One Health issue where under this concept, the health of humans is recognized as being inextricably linked to animal and environmental health. In an environmental context, aquatic environments are a major reservoir for resistant microbes and resistance genes as a result of excessive use of antimicrobials and untreated effluent streams (Michael et al., 2014). Sub-inhibitory concentrations of antibiotics in aquatic environments promote antibiotic resistance in bacteria and significant changes in DNA sequence (e.g., mutation) and genome architecture such as inversions and deletions in the bacterial chromosome (Chow et al.) and, affect biogeochemical cycles (Balcazar et al.). Furthermore, natural biofilms are reservoirs for resistance genes with the close proximity of bacteria to one another promoting lateral gene transfer processes (Balcazar et al.). More research is required to elucidate the role of antibiotic waste on ecological systems and what impact these resistance reservoirs have in promoting antibiotic resistance in clinical environments.

A key point raised by Jeffries et al. is that we lack benchmark data for many of our urbanized estuaries, limiting our ability to assess the impact of anthropogenic activity on these important and valued ecosystems. We also do not know whether a microbial community can be restored following a disturbance (chronic or acute) and what impact industrial pollutants have on the biogeochemistry of an environment (Quero et al.). Although not a result of pollution, these questions are answered in part by Bernhard et al. who reported that disturbance of a natural community due to tidal restrictions had long-lasting impact on sediment nitrogen-cycling bacterial communities even 30 years after restoration. Tan et al. also addressed the question of how artificial urban structures affect microbial assemblages. Regarding restoration, Aracic et al. reviewed numerous currently available biological approaches that could be applied. However, new methods are required to understand how microbial communities shift when disturbed to better understand the resilience of microbial communities and to assist in identifying the best ways of restoring them. Within this context, Tan et al. examined the utility of next generation sequencing approaches for assessing water quality, while Hunt and Ward presented a promising network modeling approach for predicting how disturbances transmit through microbial interactions. Both of these studies provide crucial direction for future studies in determining how anthropogenic processes disturb microbial communities and how (if possible) do we restore them.

Our aquatic ecosystems are highly valued as sources of food (Lu et al.) and recreation (Jeffries et al.) however, they are experiencing unprecedented levels of impacts from human activities particularly those of pollution and climate change processes. The diverse contributions to this Research Topic have highlighted the extent to which natural communities of aquatic microorganisms are faced with environmental perturbation and degradation as a consequence of human activities. With more research and better understanding combined with better public awareness and government policy (Rastogi et al.), improved management of, and restoration of these environments can proceed. The impetus for such change is not easily found, however, aquatic environments are often, and increasingly, highly valued by the human populations living near to them. With effective public communication, this can be used to drive change, as is happening in the 2016 Olympic city of Rio de Janeiro with its massive restoration of Guanabara Bay (Fistarol et al.).

We thank all of the participating authors for their contributions to this issue, which we believe will be a valuable resource as aquatic ecosystems continue to experience increasing anthropogenic impacts into the future.

# AUTHOR CONTRIBUTIONS

ML, JS, FL, and MB contributed to the writing of this editorial and approved its content.

# FUNDING

MB was supported by an Australian Research Council (ARC) grant DP150102326. JS was supported by an ARC Future Fellowship FT130100218.

# REFERENCES


Waters, C. N., Zalasiewicz, J., Summerhayes, C., Barnosky, A. D., Poirier, C., Gałuszka, A., et al. (2016). The anthropocene is functionally and stratigraphically distinct from the holocene. Science 351:aad2622. doi: 10.1126/science.aad2622

**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 Labbate, Seymour, Lauro and Brown. 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 emergence of** *Vibrio* **pathogens in Europe: ecology, evolution, and pathogenesis (Paris, 11–12th March 2015)**

### *Edited by:*

*Maurizio Labbate, University of Technology, Sydney, Australia*

### *Reviewed by:*

*Daniela Ceccarelli, University of Maryland, USA Yan Boucher, University of Alberta, Canada Fabiano Thompson, Federal University of Rio de Janeiro, Brazil*

### *\*Correspondence:*

*Frédérique Le Roux, Sorbonne Université, UPMC Paris 06, CNRS, UMR 8227, Integrative Biology of Marine Models, Station Biologique de Roscoff, CS 90074, F-29688 Roscoff Cedex, France fleroux@sb-roscoff.fr*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 15 May 2015 Accepted: 28 July 2015 Published: 13 August 2015*

### *Citation:*

*Le Roux F, Wegner KM, Baker-Austin C, Vezzulli L, Osorio CR, Amaro C, Ritchie JM, Defoirdt T, Destoumieux-Garzón D, Blokesch M, Mazel D, Jacq A, Cava F, Gram L, Wendling CC, Strauch E, Kirschner A and Huehn S (2015) The emergence of Vibrio pathogens in Europe: ecology, evolution, and pathogenesis (Paris, 11–12th March 2015). Front. Microbiol. 6:830. doi: 10.3389/fmicb.2015.00830* *Frédérique Le Roux 1,2 \*, K. Mathias Wegner <sup>3</sup> , Craig Baker-Austin <sup>4</sup> , Luigi Vezzulli <sup>5</sup> , Carlos R. Osorio <sup>6</sup> , Carmen Amaro <sup>7</sup> , Jennifer M. Ritchie <sup>8</sup> , Tom Defoirdt <sup>9</sup> , Delphine Destoumieux-Garzón <sup>10</sup>, Melanie Blokesch <sup>11</sup>, Didier Mazel <sup>12</sup>, Annick Jacq <sup>13</sup> , Felipe Cava <sup>14</sup>, Lone Gram <sup>15</sup>, Carolin C. Wendling <sup>16</sup>, Eckhard Strauch <sup>17</sup> , Alexander Kirschner <sup>18</sup> and Stephan Huehn <sup>19</sup>*

*<sup>1</sup> Unié Physiologie Fonctionnelle des Organismes Marins, Ifremer, Plouzané, France, <sup>2</sup> CNRS, UMR 8227, Integrative Biology of Marine Models, Station Biologique de Roscoff, Sorbonne Universités, UPMC Paris 06, Roscoff cedex, France, <sup>3</sup> Coastal Ecology, Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, List, Germany, <sup>4</sup> Cefas, Weymouth, UK, <sup>5</sup> Department of Earth, Environmental and Life Sciences, University of Genoa, Genoa, Italy, <sup>6</sup> Departamento de Microbioloxía e Parasitoloxía, Instituto de Acuicultura, Universidade de Santiago de Compostela, Santiago de Compostela, Spain, <sup>7</sup> Estructura de Investigación Interdisciplinar en Biotecnología y Biomedicina, Department of Microbiology and Ecology, University of Valencia, Valencia, Spain, <sup>8</sup> Faculty of Health and Medical Sciences, University of Surrey, Guildford, UK, <sup>9</sup> UGent Aquaculture R&D Consortium, Ghent University, Ghent, Belgium, <sup>10</sup> Interactions Hôtes-Pathogènes-Environnements, UMR 5244, CNRS, Ifremer, Université de Perpignan Via Domita, Université de Montpellier, Montpellier, France, <sup>11</sup> Laboratory of Molecular Microbiology, Global Health Institute, School of Life Sciences, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, <sup>12</sup> Département Génomes et Génétique, CNRS UMR3525, Unité Plasticité du Génome Bactérien, Institut Pasteur, Paris, France, <sup>13</sup> Institute for Integrative Biology of the Cell, CEA, CNRS, Université Paris-Sud, Orsay, France, <sup>14</sup> Laboratory for Molecular Infection Medicine Sweden, Department of Molecular Biology, Umeå Centre for Microbial Research, Umeå University, Umeå, Sweden, <sup>15</sup> Department of Systems Biology, Technical University of Denmark, Kongens Lyngby, Denmark, <sup>16</sup> Geomar, Helmholtz Centre for Ocean Research Kiel, Kiel, Germany, <sup>17</sup> Federal Institute for Risk Assessment, National Reference Laboratory for Monitoring Bacteriological Contamination of Bivalve Molluscs, Berlin, Germany, <sup>18</sup> Institute for Hygiene and Applied Immunology, Medical University of Vienna, Vienna, Austria, <sup>19</sup> Institute of Food Hygiene, Free University Berlin, Berlin, Germany*

Global change has caused a worldwide increase in reports of Vibrio-associated diseases with ecosystem-wide impacts on humans and marine animals. In Europe, higher prevalence of human infections followed regional climatic trends with outbreaks occurring during episodes of unusually warm weather. Similar patterns were also observed in *Vibrio*associated diseases affecting marine organisms such as fish, bivalves and corals. Basic knowledge is still lacking on the ecology and evolutionary biology of these bacteria as well as on their virulence mechanisms. Current limitations in experimental systems to study infection and the lack of diagnostic tools still prevent a better understanding of *Vibrio* emergence. A major challenge is to foster cooperation between fundamental and applied research in order to investigate the consequences of pathogen emergence in natural *Vibrio* populations and answer federative questions that meet societal needs. Here we report the proceedings of the first European workshop dedicated to these specific goals of the *Vibrio* research community by connecting current knowledge to societal issues related to ocean health and food security.

**Keywords: global warming, human health, aquaculture, interactions, animal model, bacterial disease, genome plasticity, european network**

# **State of the Art and Perspectives of** *Vibrio* **Research in Europe**

According to the European Environment Agency the rise of global sea surface temperature (SST) is one of the major physical impacts of climate change. However, SST in coastal European seas has increased 4–7 times faster over the past few decades than in the global oceans (Reid et al., 2011). This local increase in SST has been linked to outbreaks of *Vibrio*-associated human illness caused by *Vibrio cholerae* non O1-non-O139, *V. parahaemolyticus*, and *V. vulnificus* in several European countries (**Table 1**). However, the lack of mandatory notification systems for *Vibrio*associated illnesses prevents accurate estimates of the number of *Vibrio* infections occurring in Europe. Also mass mortalities of marine animals increase in frequency (**Table 1**), particularly in heavily polluted coastal areas, suggesting human activities as a factor favoring disease epidemics. Prominent examples include several *Vibrio* species associated with the recent great devastation of oyster beds in France. The salmonid farming industry is constantly threatened by *V. salmonicida* and *V. anguillarum*. Moreover, different subspecies of *Photobacterium damselae* are associated with diseases in cultured fish species like sole, sea bass, sea bream and turbot, while *V. vulnificus* causes hemorrhagic septicaemia in eel, derbio, tilapia, trout and shrimps but can also cause septicemia in humans. Finally, evidence has accumulated linking *Vibrio* infections (e.g., *V. coralliilyticus*) to increasing mass mortalities of benthic corals (e.g., *Paramuricea clavata*) in the NW Mediterranean Sea.

To cover the large diversity of infectious vibrios, the development of operational tools to identify and detect emergent pathogens is essential to zoosanitary monitoring of cultivated species as well as on wild animal populations. Yet, compared to human pandemic strains, little is known about the virulence mechanisms of emergent environmental vibrios. This lack of knowledge may be attributed to the high genetic diversity of*Vibrio* isolates and the diversity/plurality of virulence mechanisms. To date pathogenic capacity cannot be inferred by taxonomic affiliation, because virulence factors (e.g., secretion systems, toxins) are rarely species-specific and are often shared between *Vibrio* species by lateral gene transfer. On top of that there are very few animal models to distinguish pathogenic strains and extend our understanding of the mechanisms involved in hostmicrobe interactions. Hence the elucidation of virulence for agent and target is a prerequisite to develop prophylactic methods to fight infectious diseases.

Due to the extent of the environmental, economical, and public health consequences resulting from *Vibrio* infections, a large scientific community is working on these bacteria in Europe. In order to join fundamental and applied research teams and to investigate the emergence of pathogens in natural *Vibrio* populations, we organized the first European workshop dedicated to the research on vibrios in Paris (11–12th March 2015), that provided a forum for experts in *Vibrio* ecology, evolution and pathogenesis to address societal issues involving ocean health and food security.

### *Vibrio* **Spread in Europe linked with climate change**

Vibrios preferentially grow in warm (*>*15°C) saline aquatic environments. Warming of marine and saline inland waters is likely to support larger numbers of *Vibrio* populations and consequently an increased risk of *Vibrio* infections. An increase in the prevalence of human infections caused by


*V. parahaemolyticus*, *V. cholerae* non-O1-non-O139 and *V. vulnificus* has been recorded in Europe even at high latitudes (Baker-Austin et al., 2013). In northern Europe, the increase in reported infections corresponds both in time and space with spikes in domestically-acquired *Vibrio* cases in "heatwave" years. Similarly, samples collected in the last 60 years by the continuous plankton recorder (CPR) survey (Vezzulli et al., 2012) showed that the genus *Vibrio*, including the human pathogen *V. cholerae*, has increased in prevalence in the last 44 years in the coastal North Sea, and that this increase is correlated with warming SST. Elevated water temperatures might also facilitate the successful invasion of pathogenic variants via food trade (Nair et al., 2007), ballast water (Dobbs et al., 2013), travelers (Fillion and Mileno, 2015) or natural animals. For example, migrating birds may act as vectors of intercontinental transport of *V. cholerae* (Vezzulli et al., 2010a). The direct comparison of the population structure of *V. cholerae* from a major bird sanctuary (Lake Neusiedl, Austria), with strains collected from six other European countries revealed that several strains in the lake shared the same alleles with other European strains, consistent with pan-European transport between distant ecosystems via birds (A. Kirschner, unpublished data). As a future challenge, macro-ecological studies on the impact of climate change on *Vibrio* persistence and spread in the aquatic environment combined with studies investigating climate change effects on epidemiologically relevant variables, such as host susceptibility and exposure are needed to significantly improve prediction and mitigation strategies against the future occurrence of *Vibrio* disease outbreaks.

### **Virulence as a Function of Biotic Interactions With Host and Microbiome**

Virulence is a widespread phenomenon across the *Vibrio* phylogeny (Wendling et al., 2014). Its expression critically depends on biotic interactions with the host but also with other resident microbiota. On the host side, spatially-structured cross-infection experiments indicated that virulence of only distantly related *Vibrio* strains was lower when infecting oysters from the same geographic location. This suggests that oyster hosts are locally adapted and have evolved resistance to genetic factors shared within *Vibrio* populations (Wendling and Wegner, 2015). When considering interactions of *Vibrio* with the resident microbiome, the hemolymph microbiome modulates infections but is vulnerable to environmental disturbance (Lokmer and Wegner, 2015). Accordingly, *Vibrio* disease cannot be seen as an isolated event but needs to be considered in the context of the microbiome, which includes other non-virulent *Vibrio.* Indeed, the successive replacement of non-virulent with virulent strains during oyster infections occurs in the natural environment (Lemire et al., 2014) and the amplification of virulence in the presence of non-virulent strains suggests that also non-virulent strains contribute directly or indirectly to the development of disease. Future research on *Vibrio* disease should therefore focus on the higher order biotic interactions between the environment, the host and the pathogenic as well as the non-pathogenic fractions of microbial communities.

A key feature of the interaction between microbes within a community is the production of molecules that determine behavior like antagonism, competition or cooperation. Cellto-cell communication in vibrios coordinates virulence gene expression based on the biotic and abiotic environment (Defoirdt, 2014). For example, the three-channel quorum sensing (QS) system of *V. harveyi* controls the pathogenicity of the bacterium toward different aquatic hosts (Defoirdt and Sorgeloos, 2012; Pande et al., 2013), and our most recent research revealed that another signaling molecule, indole, controls the virulence of *V. anguillarum* toward sea bass larvae (Li et al., 2014b). Another potential signaling mechanism has been described based on the production and release of high concentrations of Damino acids into the extracellular milieu (Lam et al., 2009). First discovered in *V. cholerae*, these D-amino acids are different from those known to be part of the cell wall in bacteria (D-Ala and D-Glu) and were therefore called non-canonical Damino acids (NCDAAs; Cava et al., 2011a). NCDAAs released into the media by producer strains can affect non-producer organisms beneficially or detrimentally in a particular niche (Cava et al., 2011b; Alvarez et al., 2014). The possible implications of NCDAAs in the biological processes of co-inhabitants still remains to be investigated but the enormous energy demand suggests that these molecules should have a great impact in poly-microbial communities. Finally several strains of *Vibrio* have been demonstrated to produce potent antibacterial agents (andrimid and holomycin) or agents that block QS regulated genes (solonamides, ngercheumicin) in human pathogens (Wietz et al., 2010; Mansson et al., 2011; Kjaerulff et al., 2013; Nielsen et al., 2014). Comparative and functional genomics using software like antiSMASH (Medema et al., 2011) could identify the genetic determinants of these secondary metabolites (polyketide synthases and non-ribosomal peptide synthetases). Hence further elucidation of virulence regulatory mechanisms will enable us to better understand *Vibrio*-host interactions and ecology, and to identify targets for the design of novel agents to control disease caused by vibrios.

### **Horizontal Gene Transfer, Genome Plasticity, and Chromosome Partitioning**

Evolution of *Vibrio* species is often driven by mobile genetic elements via horizontal gene transfer (HGT). However, very little is known about HGT in environmental *Vibrio* isolates infecting marine organisms. In 75 marine *Vibrio* spp. isolated from the broad-nosed pipefish, *Syngnathus typhle*, associated prophages were characterized and the virulence of strains carrying different prophages was then assessed by comparing the relative expression of 44 immune genes during controlled infection experiments on juvenile pipefish. Preliminary results suggest that virulence is significantly influenced by the associated prophages, further supporting a role for bacteriophages in manipulating the virulence of environmental *Vibrio* isolates (C. Wendling unpublished data).

Virulence of *V. vulnificus* and *P. damselae* subsp. *damselae* in fish is determined by transferable plasmids (pVvbt2 in *V. vulnificus* and pPHDD1 in *P. damselae*). pVvbt2 contains two highly conserved virulence genes involved in serum resistance (*vep07*) and the ability to grow from eel transferrin (*vep20*) (Pajuelo et al., 2015). Interestingly, pPHDD1 also contains *vep07* and *vep20* homologs suggesting that both genes are involved in resistance to fish innate immunity. pVvbt2 also encodes RtxA13, a toxin belonging to MARTX (multifunctional, autoprocessive, repeat in toxin) family. RtxA1<sup>3</sup> is considered a host-non-specific virulence factor because it is involved in resistance to phagocytosis by murine and human phagocytes as well as in eel death (Lee et al., 2013). The other virulence plasmid, pPHDD1, encodes phospholipase-D damselysin (Dly) and the pore-forming toxin HlyA*pl* (Rivas et al., 2011). A second HlyA (HlyA*ch*) is encoded in chromosome I (Rivas et al., 2013a, 2014) and the three toxins contribute to hemolysis and virulence, and are secreted by a typetwo secretion system (Rivas et al., 2015). While the two HlyA hemolysis produce an additive effect, Dly and any of the two HlyA interact in a synergistic manner, being responsible for maximal virulence for fish and for mice (Rivas et al., 2013a). Due to their host range and their duality as pathogens for both poikilotherm and homeotherm animals, *P. damselae* and *V. vulnificus* constitute valuable biological models to study the role of mobile genetic elements in the rise of novel pathogenic strategies.

Vibrios contain large chromosomal integrons (Cambray et al., 2010) and belong to the group of naturally competent bacteria, which allows them to absorb free DNA from their surrounding environment and recombine it into their genome (Seitz and Blokesch, 2013a). For *V. cholerae*, entry into competence is tightly regulated and requires growth to high cell densities on chitinous surfaces (Meibom et al., 2005; Lo Scrudato and Blokesch, 2012, 2013). Uptake of external DNA is accomplished by a sophisticated DNA-uptake machinery (Seitz and Blokesch, 2013b, 2014; Seitz et al., 2014). As the competence regulon also encompasses the type VI secretion system-encoding gene clusters, HGT is enhanced through deliberate killing of neighboring non-sibling cells followed by the transfer of their DNA (Borgeaud et al., 2015).

The presence of two chromosomes is another characteristic feature of vibrios. While distinctive localization patterns have been described for the two chromosomes, the selective advantages brought by this bipartite architecture are still under debate (Val et al., 2012, 2014). Replication of both chromosomes is tightly coupled so that replication termination is synchronized (Rasmussen et al., 2007). Moreover, the chromosomal position of genes determines the relative copy number during growth thereby impacting the bacteriums physiology (Soler-Bistue et al., 2015). Notably, mechanistic aspects of chromosome organization, architecture, and cell cycle-dependent dynamics are only starting to be deciphered (Yamaichi et al., 2012; Demarre et al., 2014). The elucidation of the mechanisms that coordinate the interplay between chromosomes, accessory replicons, mobile DNA and HGT mechanisms is essential to better apprehend the evolution and niche adaptation of *Vibrio* species.

### **Adaptation of Pathogenic Vibrios to Intracellular Life**

The pathogenic *V. tasmaniensis* strain LGP32, a member of the *V. splendidus* clade (Gay et al., 2004) was found to be a facultative intracellular pathogen of oyster immune cells called hemocytes (Duperthuy et al., 2011). This is a rare example of

*Vibrio* adapted to intracellular life. The virulence of LGP32 in oysters correlated with the ability to enter hemocytes (Duperthuy et al., 2010, 2011). Both cellular invasion and pathogenicity depend on the major outer membrane protein OmpU, which serves as an adhesin to invade host cells. Once inside the phagosome, LGP32 releases outer membrane vesicles (OMVs) that protect the organism against antimicrobial peptides and act as vehicles for the delivery of virulence factors (Destoumieux-Garzón et al., 2014; Vanhove et al., 2015). Moreover, entry into hemocytes and intracellular survival of LGP32 are required for expression of LGP32 cytotoxicity toward hemocytes. This capacity to survive intracellularly relies on potent antioxidant and copper tolerance responses, both of which are highly induced in the hostile environment of the phagosome.

Small regulatory RNAs have been shown to play important roles in regulating virulence gene expression in response to conditions encountered in the host. sRNAs present in multicopies such as Qrrs and CsrBs were found in several *Vibrio* spp. to mediate QS regulation of virulence gene expression (Nguyen and Jacq, 2014). One peculiarity of the *Splendidus* clade seems to be the presence of four highly expressed copies of the CsrB sRNAs in their genome, instead of 2–3 found in other vibrios (Lenz et al., 2005; Toffano-Nioche et al., 2012). CsrB sRNAs are highly transcribed inside oyster hemocytes suggesting a role in adaptation to the intracellular environment (Vanhove et al., submitted). The landscape and phylogeny of putative sRNAs encoded by LGP32 demonstrate rapid vertical evolution, with a vast majority of sRNAs being species/strain specific, and only a small number (28/250) conserved in all *Vibrio* sequenced so far (Toffano-Nioche et al., 2012). Thus, sRNAs contribute to a high diversity between species and provide opportunities for adaptation/colonization of new hosts and virulence emergence, a question that will be tackled by comparative functional studies of conserved *Vibrio* sRNAs.

### **Model Systems to Study Pathogenicity Mechanisms and** *Vibrio***-host Interactions**

Microbiologists are increasingly aware that how organisms behave *in situ* in the "real world" might be distinct from those that occur in laboratory monocultures grown under tightly controlled conditions (Smith, 2000). Thus, model systems, which replicate at least part of the natural processes of infection, are needed in order to examine the relevance and biological impact of *in vitro* findings. *In vivo* models that reproduce the main clinical and pathological signs of disease seen following the consumption of contaminated food or water, are available for toxigenic and nontoxigenic *V. cholerae*, and for *V. parahaemolyticus* (Ritchie et al., 2010, 2012; Shin et al., 2011). In these studies, a combination of microbiological, histological and genetic analysis was used to identify key virulence factors and the pathologic mechanisms associated with the respective strains (e.g., see Zhou et al., 2013, 2014). However, a growing number of *Vibrio*-associated illnesses are associated with a diverse group of strains, some of which lack known virulence factors (Garcia et al., 2009; Jones et al., 2012; Ottaviani et al., 2012). A future challenge will be to examine the pathogenesis of these strains and identify additional virulence markers, which should be used to improve risk assessment tools

Microbial genome annotation and analysis platform (MAGE). A genetic resource

targeted to the different pathogens. Furthermore, a growing number of human *Vibrio* infections in Europe were not foodoften confounded by the presence of the natural microbiota (which usually contains *Vibrio* spp.). Gnotobiotic animals provide researchers with a means to examine host-microbe interactions without interference or influence from unknown microbiota (Gordon and Pesti, 1971). A model based on the use of gnotobiotic 1-day old larvae of brine shrimp (*Artemia franciscana*) has

stimulated by the workshop.

borne, but instead associated with the ability of non-O1-non-O139 *V. cholerae*, *V. parahaemolyticus*, or *V. vulnificus* to cause septicemia via wound infections (**Table 1**). Models to examine this aspect of their pathogenicity are currently lacking and should become a high priority given the poor prognosis of individuals acquiring this type of infection.

Next to models for human pathogens there is also an increasing need for aquatic animal models. However, studies aimed at investigating the pathogenicity mechanisms in aquatic hosts are been recently developed to study *V. campbellii*, *V. harveyi*, or *V. anguillarum* pathogenesis (Defoirdt et al., 2005). An alternative model system for *V. anguillarum* involves the use of gnotobiotic European sea bass (*Dicentrarchus labrax*) larvae, where survival is monitored over 1 week (Li et al., 2014a). Finally, specific-pathogen-free (SPF) juveniles of *C. gigas* (Petton et al., 2013, 2015) have been developed to investigate the diversity and dynamics of microbial populations in an oceanic environment during disease. When combined with methods to monitor gene expression and activity of vibrios during infection (e.g., Ruwandeepika et al., 2011; Defoirdt and Sorgeloos, 2012), a better understanding of the infection process(es) will emerge.

# **Conclusion**

This workshop clearly demonstrated the importance of vibrios to our understanding of emergent diseases in marine and inland aquatic ecosystems as well as their potential impact on society. The rising frequency of disease events not only affects humans directly but also indirectly by reducing food security and ecosystem health. The synergistic investigation of mechanistic and ecological processes contributing to disease is therefore paramount for our understanding of the larger scale consequences of changing *Vibrio* populations. A better understanding of *Vibrio* ecology is pivotal for the development

# **References**


of prevention and mitigation strategies. In addition, the mechanistic knowledge of virulence regulatory mechanisms could ultimately be used to inhibit disease. However, these tasks are complicated by the high diversity present within *Vibrio* populations, and the fact that biotic interactions within and between microbial communities, modify disease expression on different levels. Therefore, we have to consider *Vibrio* disease as an emergent, multi-faceted phenomenon that will require experimental model systems covering molecules to whole organisms. Expertise for most of these crucial challenges already exists and became united at the workshop under the European umbrella of *Vibrio* research thereby fostering a more productive combination of basic and applied research in the future (**Figure 1**).

# **Acknowledgments**

We warmly thank EUROMARINE, EMBRC-France, the Agence Nationale de la Recherche (ANR) for their financial support and EFOR organization to this first workshop.


**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 Le Roux, Wegner, Baker-Austin, Vezzulli, Osorio, Amaro, Ritchie, Defoirdt, Destoumieux-Garzón, Blokesch, Mazel, Jacq, Cava, Gram, Wendling, Strauch, Kirschner and Huehn. 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.*

# *Crassostrea gigas* mortality in France: the usual suspect, a herpes virus, may not be the killer in this polymicrobial opportunistic disease

*Bruno Petton1, Maxime Bruto2,3, Adèle James2,3, Yannick Labreuche2,3, Marianne Alunno-Bruscia1 and Frédérique Le Roux2,3\**

*<sup>1</sup> LEMAR UMR 6539, Ifremer, Argenton-en-Landunvez, France, <sup>2</sup> Unité Physiologie Fonctionnelle des Organismes Marins, Ifremer, Plouzané, France, <sup>3</sup> CNRS, Equipe Génomique des Vibrios, LBI2M, UPMC Paris 06, UMR 8227, Integrative Biology of Marine Models, Sorbonne Universités, Roscoff, France*

### *Edited by:*

*Maurizio Labbate, University of Technology, Sydney, Australia*

### *Reviewed by:*

*Tom O. Delmont, Marine Biological Lab, USA Jesus L. Romalde, Universidad de Santiago de Compostela, Spain*

### *\*Correspondence:*

*Frédérique Le Roux, CNRS, Equipe Génomique des Vibrios, LBI2M, UPMC Paris 06, UMR 8227, Integrative Biology of Marine Models, Sorbonne Universités, Station Biologique de Roscoff, Place Georges Tessier, 29688 Roscoff Cedex, France fleroux@sb-roscoff.fr*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 29 April 2015 Accepted: 22 June 2015 Published: 06 July 2015*

### *Citation:*

*Petton B, Bruto M, James A, Labreuche Y, Alunno-Bruscia M and Le Roux F (2015) Crassostrea gigas mortality in France: the usual suspect, a herpes virus, may not be the killer in this polymicrobial opportunistic disease. Front. Microbiol. 6:686. doi: 10.3389/fmicb.2015.00686* Successive disease outbreaks in oyster (*Crassostrea gigas*) beds in France have resulted in dramatic losses in production, and subsequent decline in the oyster-farming industry. Deaths of juvenile oysters have been associated with the presence of a herpes virus (OsHV-1 μvar) and bacterial populations of the genus *Vibrio*. Although the pathogenicity of OsHV-1 μvar, as well as several strains of *Vibrio* has been demonstrated by experimental infections, our understanding of the complexity of infections occurring in the natural environment remains limited. In the present study, we use specificpathogen-free (SPF) oysters infected in an estuarine environment to study the diversity and dynamics of cultured microbial populations during disease expression. We observe that rapid *Vibrio* colonization followed by viral replication precedes oyster death. No correlation was found between the vibrio concentration and viral load in co-infected animals. We show that the quantity of viral DNA is a predictor of mortality, however, in the absence of bacteria, a high load of herpes virus is not sufficient to induce the full expression of the disease. In addition, we demonstrate that juvenile mortalities can occur in the absence of herpes virus, indicating that the herpes virus appears neither essential nor sufficient to cause juvenile deaths; whereas bacteria are necessary for the disease. Finally, we demonstrate that oysters are a reservoir of putative pathogens, and that the geographic origin, age, and cultivation method of oysters influence disease expression.

Keywords: pacific oysters, summer mortality, herpes virus, vibrio pathogenicity, experimental infection

### Introduction

The oyster aquaculture industry in France has been shaped by various infectious diseases caused by bacteria, viruses, or parasites (Grizel and Héral, 1991; Cochennec et al., 2000; Le Roux et al., 2001). Today, cultivation of the imported species, *Crassostrea gigas* (Pacific oyster), constitutes the vast majority of oyster farming (Rohfritsch et al., 2013), and diseases affecting this species have been steadily rising over the past decades, threatening the long-term survival of farmed and natural stocks. For example, a syndrome known as "summer mortality" has been affecting oyster production on the west coast of France (Samain, 2008). This syndrome has been associated with an ostreid herpes virus, designated OsHV-1, and bacteria of the genus *Vibrio* as infectious agents (Samain, 2008). "Summer mortality" is also thought to be influenced by elevated temperature (*>*19◦C), physiological stress associated with maturation, genetic traits of the host, and aquaculture practices. However, none of these individual factors have been shown to consistently be responsible for the syndrome.

In addition to "summer mortality," since 2008, juvenile oyster (*<*12 months) deaths have increased considerably, and current mortality levels range between 60 and 90% (Martenot et al., 2011). These events are more pervasive because they occur at a lower threshold temperature (16◦C) and geographically extend to all coastal regions in France (Atlantic, Channel, and Mediterranean; Pernet et al., 2012; Petton et al., 2013). This dramatic mortality level coincides with an increase in the detection frequency of herpes virus and with the identification of OsHV-1 variants, the OsHV-1 μvar being the most predominant genotype (Segarra et al., 2010; Martenot et al., 2011, 2013). It has been postulated that the variant OsHV-1 μvar is more virulent than the reference strain (OsHV-1) and could be the etiological agent of juvenile mortalities (Segarra et al., 2010; Martenot et al., 2011, 2013; Schikorski et al., 2011a,b). However, these assumptions are speculative (to the authors' knowledge, no published data has formally demonstrated that the pathogenicity of OsHV-1 μvar is higher than that of OsHV-1) and await further confirmations. In particular, recent technological improvments in the procedures used to detect the viral DNA load in animal tissues now offer an opportunity revisiting these assumptions. The real-time PCRbased assays recently developed and now widely used as detection tools have been shown to be at least two orders of magnitude more sensitive than earlier quantitative PCR assays (Arzul et al., 2002; Pepin et al., 2008; Martenot et al., 2010). Hence, the determination of the viral DNA quantity in archived samples (collected before 2008) using the older and newer protocols should be performed to verify that the increase of mortality correlates with an increase in virus prevalence. Therefore, it is yet to be determined if the increase in juvenile oyster mortalities since 2008 corresponds to a new disease or to a worsening summer mortality syndrome, emphasizing the need to continue investigating the diversity and functioning of all oyster disease agents.

This is particularly important for oyster farming because the only regulation implemented to help the oyster industry to date is based on the hypothesis that juvenile mortalities are due to the emergence of a new disease connected to OsHV-1 μvar. The European Union Regulation (175/2010), implemented in March of 2010, suggests that when the presence of OsHV-1 μvar is detected, disease control measures should be implemented, including the establishment of a containment area to restrict the movement of *C. gigas* oysters. Thus, demonstrating that the herpes virus is a marker of the disease and/or the unique etiological agent requires further investigations.

Previous research efforts have mainly focused on the viral hypothesis and knowledge on the role of bacteria remains limited. We recently investigated the oyster disease ecology of microdiverse *Vibrio* genotypes using a new field-based approach (Lemire et al., 2014). We took advantage of recently developed specific-pathogen-free (SPF) juveniles of *C. gigas,* which become naturally infected when placed in an estuarine environment (Petton et al., 2013). We showed that the onset of disease in oysters is associated with progressive replacement of diverse and benign colonizers by members of a non-clonal but phylogenetically coherent virulent population of *Vibrio crassostreae*. While the detection of putative pathogens (OsHV-1-μvar and *V. crassostreae*) coincides with oyster deaths, the respective role of each infectious agent in the disease remains unresolved. When oysters are co-infected by the herpes virus and vibrios, one can ask if a synergy between these microbes occurs and, if so, is required to the disease. Finally as the European Union regulation implies the establishment of containment areas of infected oysters their role as reservoirs of pathogens needs to be demonstrated.

In the present study, we investigated the respective roles of herpes virus and vibrios in causing oyster mortality. The first issue we address is the potential for co-infection by the herpes virus and *Vibrio* spp. in oysters infected in the field. To this end, we use standardized SPF oysters, descendants of a pool of genitors produced in hatchery under highly controlled conditions, in order to decrease the influence of genetic and environmental parameters that could affect the host sensitivity to the disease. SPF oysters represent a suitable model to investigate the role of each pathogenic agent in the disease process, because the vibrio fraction is low in SPF oyster tissue and herpes virus undetected. Second, we investigate whether the quantity of virus and/or *Vibrio* spp. in infected oysters can be used as a predictor of mortality. Third, we aim at differentiating the roles of *Vibrio* spp. and the herpes virus as causative agents of the disease. Finally, we investigate the role of oysters as a reservoir of putative pathogens.

# Materials and Methods

### Specific-Pathogen-Free Juveniles

Wild oyster spats were collected in Fouras (Marennes-Oléron, France) in 2011 and were moved to grow-up areas located at Paimpol and at Aber Benoît (northern Brittany, France) between 2012 until 2014. These animals were exposed to diseases during the spring 2012 and suffered mortality *>*50% (Petton et al., 2013). In January and in April 2014, 60 individuals were successively transferred to the experimental Ifremer facilities located at Argenton (Brittany, France) and treated by chloramphenicol (8 mg/L) for 5 days prior to maturation conditioning as described previously (Petton et al., 2013). After gamete stripping and fertilization, animals were reared under controlled conditions up to 5 months, or until reaching a mean individual wet mass between 0.5 and 3.0 g. Quantification of the herpes virus was performed by qPCR at the different steps of the SPF production, i.e., from the broodstock at the beginning of their conditioning to the larval and post-larval stages and spats.

### Natural Infection in the Field and Cohabitation Experiments in the Laboratory

SPF oysters were maintained in Bay of Brest (Pointe du Château, 48◦ 20 06.19 N, 4◦ 19 06.37 W) during a disease outbreak (seawater temperature *>* 16◦C). Total mortalities resulting in the field were recorded daily for 57 days (from 7 August 2014 to 3 October 2014). For cohabitation experiments, field-exposed oysters (i.e., "donors") were brought back to the laboratory and cohabited in a tank with SPF oysters (i.e., "recipients") at 21◦C under controlled conditions of water renewal (Petton et al., 2015). When indicated, chloramphenicol (8 mg/L) was added in tanks every 2 days to remove the cultivable microbiota in the oyster tissues. Alive and dead donors and recipients were daily counted, and dead animals were removed from the tanks.

### Bacterial Isolation and Identification

Oysters were sacrificed, dissected to remove the digestive gland, and ground in sterile seawater (10 mL/g of wet tissue). The total cultivable bacteria and *Vibrio* microbiota was quantified (cfu/mg of tissue) using serial dilutions on Marine agar (Difco) and TCBS selective media, respectively. In addition, in two distinct experiments, the hemolymph of 40 field-exposed oysters (8 to 11 July; 25 to 29 July 2014) was non-destructively collected from marked individuals that were held for 10 days in 1-L tanks to identify animals that died or survived. Vibrios isolated from the hemolymph were quantified on TCBS. Randomly selected colonies (∼80/from 4 animals that died after 24 h; ∼80/from 4 animals that survive after 9 days) were re-streaked two times on TCBS, cultivated in Zobell media (4 g/L bactopeptone and 1 g/L yeast extract in artificial sea water, pH 7.6) and stored at -80◦C. For DNA sequencing, purified isolates were grown in Zobell overnight and their DNA was extracted using a DNA extraction kit (Wizard, Promega). The partial *hsp60* gene was amplified for all isolates as described previously (Hunt et al., 2008). The PCR conditions were: 3 min at 95◦C followed by 30 cycles of 30 sec each at 95◦C, 37◦C and 1 min at 72◦C with a final step of 5 min at 72◦C. Genes were sequenced using the reverse primer and sequencing was performed by GATC-biotech (https:// www*.*gatc-biotech*.*com). The partial *hsp60* gene sequences were aligned using Muscle (Edgar, 2004). Phylogenetic trees were built using PhyML applied to maximum-likelihood algorithm and GTR model as parameters (SPR, γ4, invariant site; Guindon et al., 2010). Reliability was assessed by the bootstrap method with 100 replicates. Circular tree figures were drawn using the online iTOL software package (Letunic and Bork, 2011).

### OsHV-1 DNA Quantification

The OsHV-1 DNA was quantified from ground tissues by real time PCR using the SYBR<sup>R</sup> Green chemistry (Labocea Quimper, France) as described previously (Pepin et al., 2008). The TaqMan<sup>R</sup> chemistry, previously demonstrated to be more sensitive (Martenot et al., 2010) was used to quantify the herpes virus from hemolymph samples (Labéo, Laboratoire Frank Duncombe, Caen, France).

### Virulence Studies using Oysters

Bacteria were grown under constant agitation at 20◦C for 24 h in Zobell media. One hundred mL of the pure culture (10<sup>7</sup> cfu) was injected intramuscularly into anesthetized SPF oysters. The bacterial concentration was confirmed by conventional dilution plating on Zobell agar. After injection, the oysters were transferred to aquaria (20 oysters per aquarium of 2.5 L) containing 1 liter of aerated 5 μm-filtered seawater at 20◦C, kept under static conditions for 24 h.

### Statistical Analysis

Statistical analysis was performed using the computing environment R Development Core Team. For all statistical comparisons, we first tested normality of distribution. For normal distributions we compared variables using Student's *t*-test. For non-normal distributions, we compared variables using Wilcoxon rank test and tested correlations using the Spearman correlation coefficient.

### Results

### Rapid *Vibrio* Colonization followed by Viral Replication Precedes Oyster Death

Specific-pathogen-free oysters were maintained in the field during a disease outbreak. Mortalities began at day 5, increased dramatically until day 15, and then stabilized, reaching a final percentage of 36% cumulative mortalities after day 30 (**Figure 1A**). The starting concentration of cultivable bacterial microbiota was about 103 cfu/mg of oyster-wet tissue prior to deployment in the field (**Figure 1B**) and did not change significantly after field exposure. However, the vibrio concentration did increase significantly after the first day in the field (*p* = 0.0001 using Wilcoxon test) and reached a maximum of 10<sup>2</sup> cfu/mg (**Figure 1C**). After day 15, the vibrio concentration was stabilized to 10 cfu/mg. The ratio of vibrio to total cultivable microbiota was 10−<sup>4</sup> in SPF oysters and 10−<sup>2</sup> in animals maintained under field conditions. The OsHV-1 DNA was detected after day 3 and the amount of virus DNA reached a maximum of 10<sup>8</sup> genome units (GUs)/mg between days 5–13 (**Figure 1D**). Surprisingly, at day 15, no viral DNA was detected from the 10 sampled oysters. Finally, from day 20 to the end of the experiment, the viral load was maintained around 10<sup>5</sup> GU/mg. The inter-individual variability in microbial counts was higher during the first 15 days of exposure (**Figures 1B–D**). No correlation was found between the vibrio concentration and viral load. This result was confirmed by a Spearman's test (*p*-value <sup>≥</sup> 0.05; R2 <sup>=</sup> 0.009). Altogether our results show that while diseased oysters are co-infected by vibrios and virus, the presence of the herpes virus does not seem to influence the infection by vibrios and vice versa.

### Oyster Mortalities Correlate with a Higher Quantity of Herpes DNA

We asked whether the quantity of OsHV-1 and vibrios could predict oyster mortalities. In two separate experiments in July 2014, SPF oysters were exposed to natural seawater for 8 days in the field and then returned to the laboratory before the onset of death. The hemolymph was non-destructively collected from marked individuals, and the oysters were kept in individual tanks in the laboratory to observe the onset of the disease. Mortalities started at day 1, reached 30% (first experiment) or 50% (second experiment) at day 5 and then leveled off (Supplementary Figure S1). The hemolymph sampling did not cause any effect on

intervals, ten oysters were sacrificed, the total culturable bacterial (B) and *Vibrio* microbiota (C) were quantified (cfu/mg of tissues, y axis) by serial dilutions on Marine agar and TCBS, respectively. DNA was extracted from ground tissues and qPCR was performed to quantify the OsHV-1 load, expressed in Genome Unit per mg (GU/mg, y axis) of tissues (D). In (B–D), the lines indicate the average of 10 animals and dots the individual value.

the mortality rate and disease kinetics as the sampled infected animals showed the same rate of mortality as the unsampled infected animals (Supplementary Figure S1). The quantity of OsHV-1 DNA detected in the hemolymph of animals that died within the first 24 h (**Figure 2**) was found to be significantly higher (up to 4 log units) than in the hemolymph of animals that were still alive after 9 days (pairwise comparison using Wilcoxon rank sum test; *p* = 0.0008). In contrast, the concentration of vibrios isolated from the hemolymph varied between individuals,

but was not correlated with the mortality occurrence (data not shown).

Previous results have suggested that only a subset of vibrio strains are virulent by experimental infection (Lemire et al., 2014). Here, we investigated the diversity and pathogenicity of vibrios isolated from oysters that died and compared it to the diversity of those that survived. A total of 247 vibrio isolates sampled from the hemolymph of animals that died within the first 24 h and from animals that were still alive after 9 days were characterized by partial sequencing of a protein-coding gene (*hsp60).* Phylogenetic analysis allowed the grouping of 148/247 isolates into 11 clades (designated from A to K) with a bootstrap value *>*70% (Supplementary Figure S2). Some of these clades could be matched with named species using type strains: *V. ichthyoenteri* (A), *V. harveyi* (B), *V. mediterranei* (C), *V. fortis* (D), *V. chagasii* (E), *V. cyclitrophicus* (H), *V. tasmaniensis* (J), *V. breoganii* (K), and *V. crassostreae* (I). We did not observe a genotype signature in oysters that died with the first 24 h.

To address the pathogenic potential of individual strains, we used an injection model of infection, which allows for rapid infection in the laboratory (**Figure 3**). A total of 91 *Vibrio* strains, isolated in the second experiment (29 July), were injected individually into oysters at a concentration of 107 cfu/animal (**Figure 3A**). The percentages of mortalities induced by strains isolated from the hemolymph of animals that died within the first 24 h and strains isolated from hemolymph of oysters that were still alive after 9 days were not found to be significantly different (Student test, *<sup>p</sup>* <sup>=</sup> 0.2513; **Figure 3B**). Half of the strains with *<sup>&</sup>gt;*50% mortality rates belonged to the *hsp60* clade I (**Figure 3A**) and matched to the phylogenetically coherent virulent population described recently as *V. crassostreae* (Lemire et al., 2014). Hence, our results show that viral load is sufficient to predict mortalities.

### In the Absence of Bacteria, a High Load of Herpes Virus is not Sufficient for a Full Expression of the Disease

Antibiotic treatments are typically used to investigate the role of bacteria in a specific disease. SPF oysters were exposed to natural seawater in the field and then returned to the laboratory after 10 days (see Material and Methods). These animals (referred to as "donors") were kept together with SPF oysters (referred to as "recipients") in the presence or absence of chloramphenicol for 14 days. In the absence of the antibiotic, donor mortalities started at day 2 and reached a cumulative value of 50%. In comparison, recipients began to die at day 8 and reached a cumulative mortality of 36% (**Figure 4A**). The antibiotic treatment resulted in a 2- and 4-fold decrease in mortality for donors and recipients, respectively, demonstrating a role of bacteria in the disease transmission and development. While chloramphenicol exposure completely removed the cultivable vibrio microbiota in the oyster tissues (data not shown), no significant effect was observed on the OsHV-1 DNA quantity detected in the donors or the recipients until day 8 (**Figure 4B**).

### Oysters are Reservoirs of Pathogens

The results above demonstrate that oysters can asymptomatically host putative pathogens, raising concern about the role of oysters as pathogen reservoirs. Using a cohabitation experiment, we asked whether oysters carry putative pathogens at temperatures of *<*16◦C (i.e., excluding mortality context). French oysters are cultivated to the size of "spats," the point at which they attach themselves to a substrate. These spats originated from natural settlement zones (wild seeds) or were produced in hatchery (hatched seeds). Wild adult oysters (i.e., not farmed) can also be collected from the environment. Hence, we also determined whether the geographic origin, age, and farming of oysters influence the disease(s) transmission.

The first experiment was performed with juvenile oysters originated from wild seed collected in 6 different geographic areas as donors, and 3-months-old SPF hatch oysters as recipients (**Table 1**). Over the course of the experiment (30 days), the hatched animals did not die on their own (not shown). On the contrary, death occurred in oysters originated from wild seeds collected in four regions (Arcachon, Pertuis Charentais, Bay of Vilaine, and Bay of Brest). In this case, 11/15 batches of juveniles originating from these regions were able to transmit the disease to the recipients. Finally, six batches originating from two regions (Thau and Bay of Bourgneuf) did not show significant mortality rates and did not cause mortality in the recipients. Mortalities always coincided with the detection of OsHV-1 in both donors and recipients. Interestingly, in 7/21 batches of donors, the virus was not detected at the beginning of the experiment but was found at the end of the trial, confirming that a thermal stress in the mesocosm (see Materials and Methods) can be used to highlight viral infection (Petton et al., 2015).

The second experiment was performed with farmed or wild adults (*>*18 months) sampled from four different geographic

areas as donors, and 3-, 6-, and 15-months-old SPF oysters as recipients (**Table 2**). Only 1/10 sets of wild oysters expressed mortalities and transmitted the disease to the recipients. On the contrary, 13/18 (72%) batches of farmed oysters originating from the four regions showed mortalities and caused the disease in the recipients. A Wilcoxon test confirmed that the farmed animals had a significantly higher death rate and transmitted more of the disease than the wild oysters (*p* = 0.0007). In all of these cases, the mortality rate was negatively correlated with the age of recipient animals. Interestingly, OsHV-1 was detected only once in the recipients (not shown), suggesting no significant role for the virus in the disease transmetted by adults (**Table 2**). Altogether, our results demonstrate that oysters are a reservoir of putative pathogens. Only juvenile oysters were shown to carry and transmit a detectable amount of OsHV-1, while adult animals carried infectious agents that are able to induce mortality of juveniles in the absence of the virus.

### Discussion

In this study, we investigated the respective roles of herpes virus and vibrios in the juvenile oyster disease by exposing SPF oysters in the field during a disease outbreak. Prior to field exposure, the cultivable fraction of the SPF oyster microbiota was determined to be 103 cfu/mg tissue, and a low vibrio presence (≤1 cfu/mg) was detected in these animals (**Figure 1**). Hence, these animals represent a suitable model to investigate the role of each pathogenic agent in the disease process, both independently and together.

The limited sensitivity in detecting viral DNA (10<sup>4</sup> GU/mg tissue), prevented the monitoring of initial infection kinetics for the animals. Hence, the detection of viral DNA after 3 days of exposition under field conditions likely resulted from the viral replication culminating at 10<sup>8</sup> GU/mg (**Figure 1**). While the mortalities stabilized after 15 days, the viral load was maintained at 10<sup>4</sup> GU/mg, and can thus be considered as an asymptomatic carriage. Whether this load corresponds to a latent phase of the virus or to a limited shedding of particles remains to be determined, but our results highlight the interest of our model to explore the herpes virus biological cycle (i.e., latent and lytic cycle; cellular tropism). Whereas previous data correlating the virus load with mortalities have been based on moribund animal sampling (Sauvage et al., 2009; Oden et al., 2011), we performed a non-destructive collection of hemolymph and showed that viral DNA quantification (*>*107 GU/mg) can predict the mortalities of juvenile oysters (**Figure 2**). However, when using an antibiotic treatment, we also demonstrated that in the absence of bacteria, a load of herpes virus <sup>≥</sup>108 GU/mg is not sufficient to induce oyster mortalities in a range classically ascribed to this disease (**Figure 4**). These results suggest that viral replication may be a consequence rather than the cause of the disease. Recently, Green and Montagnani (Green et al., 2014) observed that a poly I:C treatment decreases the OsHV-1


TABLE 1 | Cohabitation experiments performed for 30 days with juvenile oysters collected in six different geographic areas as donors and 3 months-old-SPF oysters as recipients (nd, not detected; d, detected).

replication in *C. gigas* infected by injection. Hence developing poly I:C and/or RNA interference approaches (Labreuche and Warr, 2013; Huvet et al., 2015) in oysters would be of prime interest to decipher the direct role of this virus in the disease process.

Having shown that oysters are naturally co-infected by the herpes virus and vibrios, we next ask if a synergy between these microbes is fundamental to the disease. In a first scenario, oysters infected with herpes may be more susceptible to subsequent *Vibrio* infection. However, this hypothesis is contradicted by the fact that we did not observe a correlation between the herpes load and vibrio concentration at the individual level. In a second scenario, oysters infected with *Vibrio* may be more susceptible to herpes replication. However, as in antibiotic treated oysters the viral load is similar to that of untreated animals (**Figure 4**), this scenario seems also to be unlikely. In any case, diseased oysters that are co-infected by several putative pathogens (virus and bacteria) should sustain much greater losses.

Since we aimed at differentiating the roles of *Vibrio* spp. and the herpes virus as causative agents of oyster deaths, we used an antibiotic treatment and highlighted an essential role of bacteria in the disease process (**Figure 4**). As chloramphenicol treatment is not restricted to vibrio, the next step will be to demonstrate that vibrio(s) is (are) the causative agent(s) of the death. We observed that the quantity, genotype and virulence of vibrios (determined by the injection of single strain) cannot predict the mortalities (**Figure 3**). Our results seem to contradict a recent study that describes the microbial communities in oysters challenged by a virulent vibrio strain (Lokmer and Wegner, 2015). Lokmer and Wegner found that dead and moribund oysters displayed signs of community structure disruption characterized by a very low diversity and proliferation of few taxons. However, their analysis used barcoded 16S rRNA 454 amplicon sequencing, the genetic resolution of which is not fine scale enough to analyze the vibrio diversity. Furthermore, the bacterial challenge was based on single strain injection, which does not capture the complexity of infection within the natural environment as described in the present study. We previously hypothesized that *Vibrio* diversity may have significance in the disease onset. In a first study we showed that heightened virulence is observed when combination of virulent strains are used in experimental infections, suggesting that virulence of vibrio results from a combination of factors, each associated with different strains within the population (Gay et al., 2004). More recently, we demonstrated that naturally infected oysters initially contain a large proportion of nonvirulent strains, and that these are progressively replaced by a virulent population that comprise ∼50% of the bacterial isolates at the peak of mortality (Lemire et al., 2014). We suspected that this result reflects a contribution of the non-virulent strains to the development of disease. We further demonstrated that the presence of non-virulent bacteria dramatically increases the virulence of a virulent strain at low doses. Thus, non-virulent strains must have some features that contribute either directly or indirectly to the pathogenicity of virulent isolates. In the future


TABLE 2 | Cohabitation experiments performed for 30 days with farmed or wild adults sampled in 4 different geographic areas as donors and 3-, 6-, and 15-months-old SPF oysters as recipients.

we will investigate the significance of *Vibrio* population assembly in diseased oysters with emphasis on cooperative behavior and weapons sharing that may be an important feature of the disease.

From the data collected in this study, the oyster physiology seems to be a key feature of this disease. We observed that when oysters are naturally co-infected by the herpes virus and pathogenic vibrios, only 30% to 50% of animals died. This suggests that viral replication and/or the vibrio's pathogenic effect depends on oyster susceptibility. This susceptibility may result from environmental factors (acquired susceptibility) and/or genetic traits (innate susceptibility). It may implicate the physiology of the animal itself and its associated microbial communities (Lokmer and Wegner, 2015). It has been demonstrated that resistance toward oyster juvenile disease has a genetic basis (Huvet et al., 2004; Segarra et al., 2014). More recently, Wendling and Wegner suggested that a simple genetic resistance mechanism of the oyster is matched to a common virulence mechanism shared by sympatric *Vibrio* strains (Wendling and Wegner, 2015). In the future, we will investigate the genetic bases of oyster susceptibility to the vibrio isolated in the present study.

Both adult and juvenile oysters were shown to act as reservoirs of pathogens which were able to induce mortalities of young animals. Only the juveniles were demonstrated to transmit the herpes virus. This suggests that the diseases transmitted by the juveniles or by the adults are distinct. However, it also demonstrates that the virus infection is not a prerequisite for juvenile mortality outbreaks. Farmed animals seem to present more risk of disease transmission than wild animals. In addition, the geographic origin of oysters influences the carriage of pathogen. Though our data suggest that some farming zones are less infected than others they also indicate that movements of animals between ecosystems are very noxious in terms of diseasespreading. Beyond the environmental factors that are known to potentially influence mechanisms of disease transmission in marine systems, i.e., hydrodynamics, biomass of infected animals (Petton et al., 2015), identifying other potential reservoirs of pathogens and the source(s) of the disease is a future avenue to qualify infected vs. non-infected areas.

### Acknowledgments

We warmly thank Professor Martin F. Polz and Fatima Aysha Hussain (MIT, Cambridge, MA, USA) for critically reading the manuscript and for their help in editing the English language. We thank Dr. Maryline Houssin for kindly performing qPCR analysis on hemolymph and fruitful discussions. We thank the staff of the station Ifremer Argenton and Bouin, particularly Max Nourry, for technical support. The present study has been supported by the ANR (13-ADAP-0007-01OPOPOP ),

### References


the French Ministry of Ecology, Sustainable Development, Transport and Housing (Convention DPMA 2014 – Ifremer 2013/13/1210868/NYF) and Ifremer (MB fundings).

### Supplementary Material

The Supplementary Material for this article can be found online at: http://journal*.*frontiersin*.*org/article/10*.*3389/fmicb*.* 2015*.*00686


**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 Petton, Bruto, James, Labreuche, Alunno-Bruscia and Le Roux. 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.*

# Increased seawater temperature increases the abundance and alters the structure of natural *Vibrio* populations associated with the coral *Pocillopora damicornis*

*Jessica Tout1\*, Nachshon Siboni1, Lauren F. Messer1, Melissa Garren2, Roman Stocker2, Nicole S. Webster3, Peter J. Ralph1 and Justin R. Seymour1*

*<sup>1</sup> Plant Functional Biology and Climate Change Cluster, University of Technology, Sydney, NSW, Australia, <sup>2</sup> Ralph M. Parsons Laboratory, Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, Cambridge, MA, USA, <sup>3</sup> Australian Institute of Marine Science, Townsville, QLD, Australia*

### *Edited by:*

*Ian Hewson, Cornell University, USA*

### *Reviewed by:*

*Xiu-Lan Chen, Shandong University, China Roberto Bastías, Pontifical Catholic University of Valparaiso, Chile*

### *\*Correspondence:*

*Jessica Tout, Plant Functional Biology and Climate Change Cluster, University of Technology, Sydney, P. O. Box 123 Broadway, NSW 2007, Australia jessica.tout@uts.edu.au*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 05 March 2015 Accepted: 22 April 2015 Published: 18 May 2015*

### *Citation:*

*Tout J, Siboni N, Messer LF, Garren M, Stocker R, Webster NS, Ralph PJ and Seymour JR (2015) Increased seawater temperature increases the abundance and alters the structure of natural Vibrio populations associated with the coral Pocillopora damicornis. Front. Microbiol. 6:432. doi: 10.3389/fmicb.2015.00432* Rising seawater temperature associated with global climate change is a significant threat to coral health and is linked to increasing coral disease and pathogen-related bleaching events. We performed heat stress experiments with the coral *Pocillopora damicornis*, where temperature was increased to 31◦C, consistent with the 2–3◦C predicted increase in summer sea surface maxima. 16S rRNA amplicon sequencing revealed a large shift in the composition of the bacterial community at 31◦C, with a notable increase in *Vibrio*, including known coral pathogens. To investigate the dynamics of the naturally occurring *Vibrio* community, we performed quantitative PCR targeting (i) the whole *Vibrio* community and (ii) the coral pathogen *Vibrio coralliilyticus*. At 31◦C, *Vibrio* abundance increased by 2–3 orders of magnitude and *V. coralliilyticus* abundance increased by four orders of magnitude. Using a *Vibrio*-specific amplicon sequencing assay, we further demonstrated that the community composition shifted dramatically as a consequence of heat stress, with significant increases in the relative abundance of known coral pathogens. Our findings provide quantitative evidence that the abundance of potential coral pathogens increases within natural communities of coral-associated microbes as a consequence of rising seawater temperature and highlight the potential negative impacts of anthropogenic climate change on coral reef ecosystems.

Keywords: *Vibrio*, *Vibrio coralliilyticus*, *Pocillopora damicornis*, corals, heat stress, pathogen

# Introduction

The health and function of coral reefs is profoundly influenced by microorganisms, which often form species-specific associations with corals (Rohwer et al., 2002; Rosenberg et al., 2007; Mouchka et al., 2010). These ecological relationships can be mutualistic, commensal or pathogenic (Rosenberg et al., 2007), and diseases caused by pathogenic microbes have been identified as a key threat to coral reefs globally (Bourne et al., 2009; Burge et al., 2014). Diseases including white syndrome – which causes bleaching and lysis (Kushmaro et al., 1996; Ben-Haim et al., 2003a; Rosenberg and Falkovitz, 2004), white band (Ritchie and Smith, 1998; Aronson and Precht, 2001), white plague (Thompson et al., 2001), white pox (Patterson et al., 2002), black band (Frias-Lopez et al., 2002; Sato et al., 2009), and yellow band (Cervino et al., 2008) have all been attributed to microorganisms and have led to mass mortalities and significant loss of coral cover (Bourne et al., 2009).

There is evidence that the occurrence and severity of coral disease outbreaks is increasing globally (Harvell et al., 2004; Bruno et al., 2007; Mydlarz et al., 2010), potentially due to environmental stressors associated with phenomena such as increases in seawater temperature (Mouchka et al., 2010; Ruiz-Morenol et al., 2012). Heat stress may compromise the health of corals, leading to enhanced susceptibility to disease (Hoegh-Guldberg, 1999; Hoegh-Guldberg and Hoegh-Guldberg, 2004; Jokiel and Brown, 2004), or increase the abundance and/or virulence of pathogens (Vega Thurber et al., 2009; Vezzulli et al., 2010; Kimes et al., 2011). Increases in seawater temperature have been shown to change the composition and functional capacity of coralassociated microbial communities, including shifts to an elevated state of virulence, and pathogenicity (Vega Thurber et al., 2009).

While diverse groups of microbes, including bacteria, fungi, and viruses have been implicated in several coral diseases, one bacterial genus in particular has become a recurrent feature within coral disease research. *Vibrio* are globally distributed marine *Gammaproteobacteria* (Pollock et al., 2010), which harbor a diverse virulence repertoire that enables them to be efficient and widespread pathogens of a wide range of marine species (Santos Ede et al., 2011), including shell-fish (Jeffries, 1982), fish (Austin et al., 2005), algae (Ben-Haim et al., 2003b), mammals (Kaper et al., 1995; Shapiro et al., 1998; Oliver, 2005), and corals (Ben-Haim et al., 2003b). White syndrome in *Montipora* corals is caused by *V. owensii* (Ushijima et al., 2012), white band disease II in *Acropora cervicornis* has been attributed to *V. charchariae* (synonym for *V. harveyi*; Gil-Agudelo et al., 2006; Sweet et al., 2014), and a consortium of *Vibrio* are responsible for yellow band disease (Cervino et al., 2008; Ushijima et al., 2012). Furthermore, *V. shiloi* and *V. coralliilyticus* are the causative agents of bleaching in the coral species *Oculina patagonica* (Kushmaro et al., 1996, 1997, 1998; Toren et al., 1998) and the cauliflower coral *Pocillopora damicornis* (Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b), respectively.

Laboratory experiments using cultured isolates of *V. shiloi* (Kushmaro et al., 1996, 1997) and *V. coralliilyticus* (Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b) have fulfilled Koch's postulates, with each species proven to be the causative agent of coral bleaching. *V. shiloi* causes bleaching in *O. patagonica* by using chemotaxis toward the coral mucus, before adhering to the coral surface and penetrating the epidermis (Banin et al., 2001). After colonization of the coral, cell multiplication occurs followed by production of the Toxin P molecule, which inhibits photosynthesis in the symbiotic zooxanthellae, resulting in coral bleaching, and tissue loss (Rosenberg and Falkovitz, 2004). Similarly, *V. coralliilyticus* causes bleaching, lysis and tissue loss in the coral *P. damicornis* (Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b; Meron et al., 2009; Garren et al., 2014). The mechanism behind *V. coralliilyticus* infection also includes motility and chemotaxis (Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b) and involves the post-colonization production of a potent extracellular metalloproteinase, which causes coral tissue damage (Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b). Another key similarity in the infection and bleaching mechanisms of *V. shiloi* and *V. coralliilyticus* is an increased infection rate under elevated seawater temperatures (Toren et al., 1998; Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b).

Heat stress can enhance coral disease by increasing host susceptibility to infection by pathogens (Bourne et al., 2009; Mouchka et al., 2010) or altering the behavior and virulence of pathogenic bacteria (Kushmaro et al., 1998; Banin et al., 2001; Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003a,b; Koren and Rosenberg, 2006; Bourne et al., 2008; Kimes et al., 2011; Santos Ede et al., 2011). Notably, *V. shiloi* can only be isolated from bleached corals during summer months (Kushmaro et al., 1998) and laboratory experiments have shown that this species causes bleaching at an accelerated rate above 29◦C, yet has negligible effect at 16◦C (Kushmaro et al., 1998). Similarly, tissue loss caused by *V. coralliilyticus* is most rapid at elevated temperatures between 27 and 29◦C (Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b). Seawater temperatures above 27◦C have also been shown to play a direct role in the upregulation of several *V. coralliilyticus* virulence genes, including factors involved in host degradation, secretion, antimicrobial resistance, and motility (Kimes et al., 2011). Up-regulation of motility is particularly notable as both *V. shiloi* and *V. coralliilyticus* exhibit enhanced chemotactic capacity at elevated temperatures (Banin et al., 2001; Garren et al., 2014). Heat-stressed corals also increase the production and release of signaling compounds including dimethylsulfoniopropionate (DMSP) at elevated temperature, further enhancing the ability of pathogens to locate, and colonize heat-stressed corals (Garren et al., 2014).

To date, our understanding of coral-associated *Vibrio* dynamics under elevated seawater temperatures has been solely derived from laboratory-based experiments using cultured isolates (Kushmaro et al., 1998; Toren et al., 1998; Banin et al., 2001; Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003b; Garren et al., 2014). However, there is currently little understanding of how native communities of *Vibrio*, occurring within diverse natural assemblages of bacteria, will respond to elevated seawater temperatures. Understanding the dynamics of *Vibrio* populations within this complex, but also more realistic, scenario is important because it is very probable that Vibrios living in co-habitation with other competing and interacting species, will display different dynamics to those displayed by cultured isolates under laboratory conditions. For instance, inter-species antagonistic interactions among bacteria can strongly influence the growth and proliferation of other *Vibrio* species (Long et al., 2005), and we may expect similar ecological complexities to also occur within the coral holobiont. Here, we examined changes in the *Vibrio* population within a natural, mixed community of bacteria associated with the coral species *P. damicornis* on Heron Island, the Great Barrier Reef, Australia, and demonstrate that heat stress increases the abundance and changes the composition of potentially pathogenic *Vibrio* populations associated with corals.

# Materials and Methods

### Heat Stress Experiment

Three separate colonies (denoted A, B, and C) of the coral species *P. damicornis* were collected from within the Heron Island lagoon, on the Great Barrier Reef, Australia (23◦26 41S, 151◦54 47E), and translocated to the Heron Island Research Station. Colonies were placed into flow-through aquaria, which circulated water pumped from the reef flat to the Heron Island Research Station. The colonies were fragmented into 48 nubbins using bone cutters and acclimated for 8 days across six flow-through experimental tanks. The placement of the nubbins from each colony within each tank and the position of the tanks were randomized. During the experiment, three tanks were maintained at the ambient seawater temperature (22◦C) experienced on the reef flat (control), while the remaining three tanks were exposed to a heat stress treatment, which involved the incremental ramping of seawater temperature by 1.5◦C each day for seven consecutive days using one 25W submersible aquarium heater (Aqua One, Ingleburn, NSW, Australia) per tank, until a final temperature of 31◦C was reached. Water was circulated in the tanks using one 8W maxi 102 Powerhead pump (Aqua One, Ingleburn, NSW, Australia) per tank. This temperature increase is in line with the predicted 2–3◦C increases above current summer average seawater temperature (Hoegh-Guldberg, 1999, 2004; Berkelmans et al., 2004; Hoegh-Guldberg et al., 2007) for Heron Island.

Coral nubbins were sampled using sterile forceps at the start of the experiment (t0) and after 7 days for both the control (tfinal Control) and heat stress treatments (tfinal Heat stress). The nubbins were immediately placed into 15 mL falcon tubes containing 3 mL of RNA*later* (Ambion, Life Technologies, Australia; Vega Thurber et al., 2009), which was a sufficient volume to completely immerse the nubbins. The nubbins were subsequently stored at −80◦C until processing.

### Photosynthetic Health of Corals

Photosynthetic health of the corals was checked using a diving pulse amplitude modulated (PAM) fluorometer (Walz, Germany) in the tfinal Control and tfinal 31◦Ctreatments. Corals were darkadapted for 10 min before their minimum fluorescence in the dark (F*O*) was recorded. Maximum fluorescence (FM) was determined using a saturating pulse of light for 0.8 s. The corals were then illuminated under 616 µmol photon m−<sup>2</sup> s−<sup>1</sup> light for 5 min to test their ability to sustain photosynthetic function under light. Maximum Quantum Yield (FV/FM) was measured on dark-adapted samples and effective quantum yield Y(PSII), regulated non-photochemical quenching Y(NPQ), and non-regulated non-photochemical quenching Y(NO) were measured on light adapted samples. To compare the changes in the FV/FM*,* Y(PSII), Y(NPQ), and Y(NO) measurements in the t0, tfinal Control, and tfinal Heat Stress treatments, a 1-way analysis of variance (ANOVA) was used (treatment) to determine significant differences (*P <* 0.05) between these measurements. Prior to this, data was tested for normality using the Kolmogorov– Smirnov test and Levene's test was used for homogeneity of variance.

### Coral-Bacterial Cell Separation

Coral nubbins were thawed slowly on ice and removed from the RNA-later solution using sterile forceps and kimwipes to remove excess solution (Vega Thurber et al., 2009). Replicate nubbins from the same donor colony (A, B, or C) were pooled and placed into sterile 150 mL conical flasks containing 15 mL sterileautoclaved calcium and magnesium free seawater plus 10 mM EDTA (CMFSWE). The surfaces of the nubbins were airbrushed using 80 psi with a sterile 1 mL barrier tip (fresh tip for each new nubbin) in the conical flasks using sterile forceps to hold the nubbin in place. For each sample, the 15 mL tissue slurry was then filtered through a sterile 100 µm cell strainer (BD 352360) into a sterile 50 mL plastic centrifuge tube to remove host cells. The *<*100 µm filtrate was then filtered through a 3 µm filter (Whatman) and sterile filter tower apparatus (Nalgene) using vacuum pressure to remove any host cells larger than 3 µm. The resultant *<*3 µm filtrate (∼15 mL) was centrifuged at 14462 × *g* to pellet the microbes for 5 min. DNA was extracted from the cell pellet using the MO BIO Ultra Clean Microbial DNA Kit (Carlsbad, CA, USA) according to the manufacturer's instructions. Genomic DNA concentrations were measured using a Qubit 2.0 fluorometer (Invitrogen).

### 16S rRNA Amplicon Sequencing and Analysis

The bacterial community composition in each nubbin was determined using the universal bacterial 16S rRNA gene primers 27F (5 -AGAGTTTGATCMTGGCTCAG-3 ) and 1392R (5 - ACGGGCGGTGTGTRC-3 ; resulting in a 1365bp product) and the HotStarTaq Plus Master Mix Kit (Qiagen, USA). A 30 cycle amplification process was employed, incorporating the following cycling conditions: 94◦C for 3 min, followed by 28 cycles of 94◦C for 30 s, 53◦C for 40 s and 72◦C for 1 min, after which a final elongation step at 72◦C for 5 min was performed. In addition, the composition and diversity of the *Vibrio* community was assessed using the *Vibrio* specific 16S rRNA gene primers VF169 (5 -GGATAACYATTGGAAACGATG-3 ; Yong et al., 2006) and Vib2\_R (5 -GAAATTCTACCCCCCTACAG-3 ; Thompson et al., 2004; Vezzulli et al., 2012), resulting in a 511 bp product. In this instance, MangomixTM (Bioline) Taq polymerase was used and the following cycling conditions were performed: an initial activation step at 95◦C for 120 s, followed by 30 cycles of denaturation at 95◦C for 15 s, annealing at 53◦C for 30 s and extension at 72◦C for 30 s, after which a final elongation step at 72◦C for 10 min was performed. In both cases, PCR products were used to prepare DNA libraries with the Illumina TruSeq DNA library preparation protocol. Sequencing was performed, following an additional amplification step using the 27F-519R primer pair for the 16S rRNA amplicon sequences on an Illumina MiSeq (at Molecular Research LP; Shallowater, TX, USA) following the manufacturer's guidelines.

16S rRNA gene sequences were analyzed using the QIIME pipeline (Caporaso et al., 2010; Kuczynski et al., 2011). De novo Operational Taxonomic Units (OTUs) were defined at 97% sequence identity using UCLUST (Edgar, 2010) and taxonomy was assigned to the Greengenes database (version 13\_8; McDonald et al., 2012) using BLAST (Altschul et al., 1990). Chimeric sequences were detected using ChimeraSlayer (Haas et al., 2011) and filtered from the dataset. Sequences were then rarefied to the same depth to remove the effect of sampling effort upon analysis (Santos et al., 2014) and chao1 diversity estimates were calculated. ANOVA was used (treatment) to determine significant differences (*P <* 0.05) between the diversity estimates in each treatment. Prior to this, data was tested for normality using the Kolmogorov–Smirnov test and Levene's test was used for homogeneity of variance. In cases where these assumptions were not met, log10 transformations were performed. The community composition for each of the treatments t0, tfinal Control, and tfinal Heat Stress was averaged across the three replicates within each treatment.

Multivariate statistical software (PRIMER v6) was used to measure the degree of similarity between the bacterial community composition in each treatment (Clarke and Gorley, 2006). Data was square-root-transformed and the Bray–Curtis similarity was calculated between samples. Similarity percentage (SIMPER) analysis (Clarke, 1993) was used to identify the sequences contributing most to the dissimilarity between the treatments.

For the *Vibrio*-specific assay, the OTUs representing *>*1% of the total sequences were combined with various *Vibrio* species nucelotides taken from Yong et al. (2006) and *V. coralliilyticus* nucleotides taken from Huete-Stauffer et al. (unpublished), Ben-Haim et al. (2003b) and Ushijima et al. (2014) to build a phylogenetic tree. Sequences were first aligned and inspected using MUSCLE (Edgar, 2004) and the tree was constructed after 1,000 bootstrap re-samplings of the maximum-likelihood method using the Tamura-Nei model (Tamura et al., 2007) in MEGA 6.0 (Tamura et al., 2013), where only values *>*50% were displayed on the tree (Felsenstein, 1985). The OTU abundance was represented as a percentage of the overall community composition. OTUs were included on the tree if responsible for driving significant differences between the treatments according to SIMPER analysis and were color coded according to whether the OTU was more abundant in the tfinal Control treatment (blue circle) or the tfinal Heat stress treatment (red circle).

### Quantitative PCR and Analysis

Quantitative PCR analyses targeting a *Vibrio-* specific region of the 16S rRNA gene and the heat shock protein gene (*dnaJ*) specific to *V. coralliilyticus* were conducted on all samples. Standards were created by growing the bacterial isolates *V. parahaemolyticus* (ATCC 17802) and *V. coralliilyticus* (ATCC BAA-450) overnight in Marine Broth (BD, Difco) at 37◦C (150 rpm shaking water bath) and 28◦C (170 rpm in a shaking incubator), respectively. Prior to qPCR analysis, calibration curves for each assay were created using viable counts from dilution series of the isolates. The cultures were homogenized and divided into 4 × 1 mL aliquots, washed three times with sterile phosphatebuffered saline (PBS) and pelleted at 5200 g for 10 min. Three of the washed pellets were used for DNA extraction using the MO BIO Ultra Clean Microbial DNA Kit (Carlsbad, CA, USA), while the remaining washed pellet was resuspended in 1 mL PBS and 10-fold serial dilutions with Phosphate Buffered Saline were prepared in triplicate. Three replicate 100 µL aliquots from each dilution (10−5–10−8) were spread onto marine agar plates and grown at 37◦C (*V. parahaemolyticus*) or 28◦C (*V. coralliilyticus*) over 24–48 h, and resultant colonies were counted.

A 1:5 dilution of DNA: nuclease free water was used for all qPCR assays to reduce pipetting errors. The *Vibrio* population was assessed using 16S rRNA *Vibrio* primers Vib1\_F (5 -GGCGTAAAGCGCATGCAGGT-3 ) and Vib2\_R (5 -GAAATTCTACCCCCCTACAG-3 ; Thompson et al., 2004; Vezzulli et al., 2012) producing a 113 bp product. Power SYBR Select Master Mix (Applied Biosystems) was used, with reaction mixtures comprising 10 µL Master Mix, 5 µL of diluted (1:5) sample, and 0.4 µM of each primer to a final volume of 20 µL. The qPCR was performed using a Step One Plus (Applied Biosystems) and the following optimized cycling conditions: 2 min at 50◦C, then an initial denaturation-hot start of 2 min at 95◦C, followed by 40 cycles of the two-step reaction: 95◦C for 15 s and 60◦C for 1 min. This was followed by a holding stage at 72◦C for 2 min and a melt curve stage.

The relative abundance of *V. coralliilyticus* was measured by targeting the *dnaJ* gene that encodes heat shock protein 40 in this species (Pollock et al., 2010), using the primers: Vc\_dnaJ\_F1 (5 -CGGTTCGYGGTGTTTCAAAA-3 ) and Vc\_dnaJ\_R1 (5 - AACCTGACCATGACCGTGACA-3 ) and a TaqMan probe, Vc\_dna-J\_TMP (5 -6-FAM-CAGTGGCGCGAAG-MGBNFQ-3 ; 6-FAM; Pollock et al., 2010). Reaction mixtures included a 10 µL TaqMan Universal Master Mix II (Applied Biosystems), 5 µL of diluted (1:5) sample, 0.6 µM of each primer and 0.2 µM fluorophore-labeled TaqMan probe in a final total volume of 20 µL. The optimized qPCR cycling conditions were: 2 min at 50◦C, then an initial denaturation-hot start of 10 min at 95◦C, followed by 40 cycles of the following incubation pattern: 95◦C for 15 s and 60◦C for 1 min.

Resultant qPCR data for the *Vibrio*-specific and *V. coralliilyticus* assays were analyzed using Step One Software V2.3 (Applied Biosystems). The concentrations of bacteria were normalized to the coral surface area per cm2, which was calculated by paraffin wax dipping as described in Holmes (2008) and Veal et al. (2010). To compare the abundance of bacteria in the t0, tfinal Control, and tfinal Heat Stress treatments using the qPCR assays, ANOVA was used (treatment) to determine significant differences (*P <* 0.05) between the abundances in each treatment (qPCR). Prior to this, data was tested for normality using the Kolmogorov–Smirnov test and Levene's test was used for homogeneity of variance. In cases where these assumptions were not met, log10 transformations were performed.

### Results

### Effects of Elevated Temperature on Coral Health

No visual signs of stress or bleaching were evident in the Control nubbins over the course of the experiment, yet evidence of bleaching was observed in the Heat Stress nubbins where significant levels of heat stress of the zooxanthellae were detected in the tfinal Heat Stress treatment compared to the tfinal Control nubbins using PAM fluorometry. Heat stressed corals showed a strong decline in zooxanthellae condition (significant decrease in the FV/FM (*P* = 0.002) and Y[PSII] (*P* = 0.003) measurements; Supplementary Information Tables S1 and S2), while simultaneously the zooxanthellae were protecting their cells from further photodamage by significantly increasing the xanthophyll cycle – Y[NPQ] measurements (Supplementary Information Tables S1 and S2).

### Bacterial Community Composition

Differences in bacterial community composition between the tfinal Control and tfinal Heat Stressed corals were identified using 16S rRNA gene amplicon sequencing (**Figure 1**). The community composition of the tfinal Control and tfinal Heat Stress treatments were 42% dissimilar (SIMPER analysis; Supplementary Information Figure S1, Supplementary Information Table S3), while the largest difference (56%) in the community composition was between t0 and the tfinal Heat Stress treatments (Supplementary Information Table S4). Chao1 diversity estimates revealed that the tfinal Heat Stress treatment had significantly (*P <* 0.05) higher diversity (1406 ± 155 SD) compared to the tfinal Control (995 ± 23 SD).

The bacterial community at t0 was dominated by the *Oceanospirillales* (47%), which were primarily composed of members from the *Endozoicomonacea,* followed by *Burkholderiales* (8.5%), *Rickettsiales* (7%), and *Rhodobacterales* (6%) (**Figure 1**). A shift in the community was observed in control corals over the 7 day experiment involving an increase in the relative occurrence of sequences matching the *Rhodobacterales* (21%) and *Flavobacteriales* (8.6%), and a decrease in *Oceanospirillales* sequences (29%). These shifts are indicative of a mild experimental effect (**Figure 1**). However, a dramatic community shift was detected in the tfinal Heat Stress treatment relative to both the t0 and tfinal Control samples, which involved an increase in the relative proportion of *Rhodobacterales* (46.7%), *Flavobacteriales* (17.3%), and *Vibrionales* (10.5%)*.* The occurrence of *Vibrionales* is notable because these organisms were not present in either control treatment (**Figure 1**). SIMPER analysis revealed that the decrease in *Oceanospirillales* abundance and increase in *Vibrionales* abundance were primarily responsible for differences in community composition between the tfinal Control and Heat Stress treatments (Supplementary Information Table S3).

### Quantification of the General *Vibrio* Population and of *V. coralliilyticus* Using Real Time qPCR

To confirm and quantify the increased abundance of *Vibrio* observed in tfinal Heat Stressed corals, we applied a qPCR assay to track changes in the relative abundance of the *Vibrio* community. The *Vibrio* community-specific qPCR assay detected Vibrios in all treatments (**Figure 2**, standard curve: *<sup>R</sup>*<sup>2</sup> <sup>=</sup> 0.99, Efficiency % = 93.1), but abundances were significantly higher in

FIGURE 1 | Bacterial taxa (order) associated with the coral *Pocillopora damicornis* on Heron Island, the Great Barrier Reef at t0 (22**◦**C; A–C are replicates), tfinal Control (22**◦**C; A–C are replicates), and tfinal Heat stress (31**◦**C; A–C are replicate nubbins) conditions using 16S rRNA gene amplicon sequencing. Hits were generated by comparing the sequences with

BLASTn to the Greengenes database in QIIME.

the tfinal Heat Stress treatment, where they reached an average of 2.2 <sup>×</sup> <sup>10</sup><sup>7</sup> (±6.3 <sup>×</sup> <sup>10</sup><sup>6</sup> SD) cells cm−<sup>2</sup> of coral surface (*<sup>P</sup> <sup>&</sup>lt;* 0.01, Supplementary Information Table S5). *Vibrio* abundances in this treatment were two–three orders of magnitude higher than in the tfinal Control [1.4 × 105 (±9.5 × 10<sup>4</sup> SD) cells cm−2] and t0 (2.0 <sup>×</sup> 104 (±1.5 <sup>×</sup> <sup>10</sup><sup>4</sup> SD) cells cm−2) samples, respectively, (**Figure 2**).

FIGURE 2 | Quantitative PCR was performed to quantify the abundance of natural populations of Vibrios associated with the coral *P. damicornis* on Heron Island, the Great Barrier Reef at t0 (22**◦**C), tfinal Control (22**◦**C), and tfinal Heat stress (31**◦**C) conditions. Standard curve: *<sup>R</sup>*<sup>2</sup> <sup>=</sup> 0.99, Eff% <sup>=</sup> 93.1. Abundances are expressed as the number of bacteria per cm2. *<sup>n</sup>* <sup>=</sup> 3.

FIGURE 3 | Quantitative PCR assays were used to quantify the abundance of natural populations of *Vibrio coralliilyticus* associated with the coral *P. damicornis* on Heron Island, the Great Barrier Reef at t0 (22**◦**C), tfinal Control (22**◦**C), and tfinal Heat stress (31**◦**C) conditions, standard curve: *<sup>R</sup>*<sup>2</sup> **<sup>=</sup>** 0.995, Eff% **<sup>=</sup>** 99.9. Abundances are expressed as the number of bacteria per cm2. *<sup>n</sup>* <sup>=</sup> 3.

Variation in the abundance of the coral pathogen *V. coralliilyticus* was also assessed using qPCR*.* In the t0 corals, *V. coralliilyticus* was detected in only one of the three replicate colonies, in very low abundance (17.5 cells cm−<sup>2</sup> of coral surface). Similarly, low concentrations were observed in the tfinal Control samples, with abundances in one replicate below the detection limit and a mean of 81.5 cells cm−<sup>2</sup> observed in the other two replicates. In contrast, *V. coralliilyticus* concentrations in the tfinal Heat Stress treatment 6.3 <sup>×</sup> 104 (±3.4 <sup>×</sup> 104 SD) were significantly higher (*P <* 0.05, Supplementary Information Table S6) and reached up to four orders of magnitude higher than the tfinal Control (**Figure 3**).

### Characterizing Changes in the *Vibrio* Population Induced by Heat Stress

Using a *Vibrio*-specific 16S rRNA amplicon sequencing approach we observed a clear shift in the composition of the coral *Vibrio* community between the tfinal Control and tfinal Heat Stress treatments. Consistent with the results of the qPCR assay, where negligible numbers of *Vibrio* were detected, a small number (*n* = 2024) of *Vibrio* sequences were observed in the t0 treatment. To avoid rarefying to this very low number of sequences, the t0 treatment was subsequently omitted from the data set, as we consider the key comparison to test for the effects of increased seawater temperatures to be the tfinal Control vs. Heat Stress treatments. The *Vibrio* community composition was different between the tfinal Control and Heat Stress treatments. In particular, two OTUs, denoted *P. dam* bact 1 and bact 2, were responsible for driving the largest differences (29 and 25%, respectively, according to SIMPER analysis) between treatments (**Figure 4**, Supplementary Information Table S7). While the *P. dam* bact 1 OTU comprised an average of 38.5% (±6.8%) of the community in the tfinal control treatment (**Figure 4**), it was not present in the tfinal Heat Stress treatment. In contrast, the *P. dam* bact 2 OTU was more abundant in corals from the tfinal Heat Stress treatment, comprising an average of 70.6% (±6.0%) of the total *Vibrio* community (**Figure 4**), while representing only 10.4% (±3.4%) of the community in the tfinal control treatment. Phylogenetic analysis of the two dominant *Vibrio* OTUs (**Figure 5**) revealed that *P. dam* bact 1 appears to be closely related to *V. pomeryoi* (AJ491290), while *P. dam* bact 2 may be related to *V. tubiashii* (KJ094891.1) and *V. coralliilyticus* (KF864214.1; **Figure 5**).

# Discussion

Rising global temperatures, related to anthropogenically driven climate change, are expected to drive the geographical expansion of pathogens and the spread of disease outbreaks (Harvell et al., 1999, 2002; Burge et al., 2014). In marine habitats, a rise in *Vibrio-*induced diseases has been identified as an emerging global issue and has been correlated to rising seawater temperatures (Vezzulli et al., 2012; Baker-Austin et al., 2013). For instance, increasing seawater temperature has been linked to increased *Vibrio* occurrence in the North and Baltic Seas and a concurrent increase in cases of human infections by *Vibrio*

species in this region (Vezzulli et al., 2012; Baker-Austin et al., 2013). Similarly, increasing numbers of human infections by *V. vulnificus* and *V. parahaemolyticus* off the coast of Spain have been linked to higher seawater temperatures (Martinez-Urtaza et al., 2010).

Clear links between elevated seawater temperature and the global decline of corals have also become increasingly apparent (Mydlarz et al., 2009; De'ath et al., 2012). Elevated seawater temperatures have led to (i) increased occurrence of coral bleaching, whereby symbiotic dinoflagellates are expelled from the coral host (Hoegh-Guldberg et al., 2007) and (ii) a situation where many corals are living close to their thermal physiological maximum (Hoegh-Guldberg et al., 2007). In addition to these direct effects on coral physiology and the coral-*Symbiodinium* symbiosis, rising seawater temperatures have also been linked to increased incidence of coral disease and microbial-associated bleaching, or white syndrome (Bruno et al., 2007). In particular, *V. shiloi* and *V. coralliilyticus* have been identified as temperaturedependent pathogens responsible for coral bleaching (Kushmaro et al., 1996, 1997, 1998; Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003a).

To date, the majority of research investigating the roles of *Vibrio* sp. in coral disease has been conducted in the laboratory using cultured isolates obtained from healthy and diseased corals (Kushmaro et al., 1998; Banin et al., 2000; Ben-Haim and Rosenberg, 2002; Ben-Haim et al., 2003a; Koren and Rosenberg, 2006; Vidal-Dupiol et al., 2011; Garren et al., 2014; Rubio-Portillo et al., 2014) with relatively few studies assessing natural populations of coral-associated *Vibrio* during heat

stress or bleaching events (Bourne et al., 2008; Vezzulli et al., 2010). Community finger-printing approaches have previously revealed increases in the relative abundance of *Vibrio* populations during a naturally occurring bleaching event on the GBR (Bourne et al., 2008), while the appearance of *V. coralliilyticus* in diseased specimens of the octocoral *Paramuricea clavata* was also linked to elevated seawater temperature (Vezzulli et al., 2010).

In our study, initial evidence for a temperature induced increase in coral-associated *Vibrio* was provided by 16S rRNA gene amplicon sequencing. Corals from the control treatments were dominated by the *Oceanospirillales*, primarily due to the abundance of *Endozoicomonacea,* a group widely shown to be associated with healthy colonies of diverse coral species (Morrow et al., 2012, 2014; Bayer et al., 2013; Neave et al., 2014) including *P. damicornis*(Bourne and Munn, 2005). In contrast, the bacterial community in tfinal Heat Stressed corals was characterized by significantly higher levels of diversity (Chao1) than the tfinal Control corals. This is consistent with previous studies where diversity increased among white plague affected corals (Sunagawa et al., 2009). The tfinal Heat Stressed corals contained diverse assemblages of copiotrophic and potentially opportunistic microbes including *Rhodobacteriales*, *Flavobacteriales,* and *Vibrionales*. Notably, while *Vibrio* sequences made up 10.5% of sequences in corals from the tfinal Heat Stress treatment, they were completely absent in the t0 and tfinal Control samples. In addition, a substantial decrease in *Oceanospirillales* and a disappearance of *Burkholderiales* was observed in tfinal Heat Stressed corals. The changes observed here are consistent with previous research

indicating that specific bacterial populations, including putative pathogens, emerge, and dominate the coral-associated bacterial community during environmental stress events (Roder et al., 2014). These community shifts may be a direct effect of temperature on the growth of specific members of the microbial community, or alternatively caused by a change in the chemicals released by heat-stressed corals (Garren et al., 2014).

Due to the increased proportion of *Vibrio* sequences in the 16S rRNA amplicon analysis and the potential role of *Vibrio* in coral disease (Vezzulli et al., 2010), we investigated the dynamics of this community further using targeted qPCR and *Vibrio*-specific amplicon sequencing approaches. A clear shift in the composition of the *Vibrio* community was observed in conjunction with the increased *Vibrio* abundance under elevated seawater temperature. Using qPCR, we detected low abundances of total *Vibrio* in the t0 and tfinal Control treatments, consistent with previous observations in healthy corals (Ritchie and Smith, 2004; Raina et al., 2009; Vezzulli et al., 2013) and our 16S rRNA amplicon sequencing results. However, we observed an increase in relative *Vibrio* abundance of two– three orders of magnitude in the tfinal Heat Stressed corals. These patterns support previous reports that *Vibrio* abundance is linked to seawater temperature (Rubio-Portillo et al., 2014). While the increased abundance of *V. coralliilyticus* is part of a broader increase in abundance of total Vibrios, the magnitude of increase was substantially higher in *V. coralliilyticus* (four orders of magnitude compared to 2–3 orders of magnitude). This indicates that the putative coral pathogen *V. coralliilyticus* particularly benefited from the increased seawater temperature during in this study.

The ecological role of the resident *Vibrio* community in the health of corals is likely to vary substantially across species. Some Vibrios appear to form mutualistic relationships with corals by fixing nitrogen in the mucus (Chimetto et al., 2008) whereas others are putative agents of coral disease. However, despite substantial evidence of links between coral disease and *Vibrio* occurrence, in many cases it is unknown whether these organisms are the primary etiological agents or simply opportunistic colonizers that exploit the coral when host health is compromised (Bourne et al., 2008; Raina et al., 2010). While difficulties in assigning *Vibrio* taxonomy using 16S rRNA sequencing approaches are sometimes encountered (Cana-Gomez et al., 2011), our *Vibrio* specific 16S amplicon assay demonstrated clear differences between the *Vibrio* communities in the control and heat-stress samples and identified two key OTUs responsible for driving these differences. In the control corals the *Vibrio* community was dominated by OTUs that matched *V. pomeroyi* (AJ491290), supporting previous research showing *V. pomeryoi* is found year round in healthy corals (Rubio-Portillo et al., 2014). *V. pomeryoi* is not known to be involved in coral disease and is likely a normal resident member of the coral-associated community (Rubio-Portillo et al., 2014). Up to 70% of the *Vibrio* community in tfinal Heat Stressed corals was comprized of a single OTU (OTU *P. dam* bact 2), which our phylogenetic analysis indicates is closely related to the oyster pathogen *V. tubiashii* (KJ094891.1; Hada et al., 1984; Hasegawa et al., 2008; Richards et al., 2015) and the coral pathogen *V. coralliilyticus* (KF864214.1). *V. tubiashii* and *V. coralliilyticus* are highly related species (Ben-Haim et al., 2003b), and whilst the taxonomy of OTU *P. dam* bact 2 remains to be fully resolved, the phylogenetic positioning close to several *V. coralliilyticus* strains indicates that this organism may be *V. coralliilyticus*. This would be consistent with the findings of our *V. coralliilyticus* qPCR analysis, where a four orders of magnitude increase in abundance of *V. coralliilyticus* was observed in corals from the tfinal Heat Stress treatment. These results are consistent with findings of Vezzulli et al. (2010) who only observed *V. coralliilyticus* in diseased coral specimens, as well as Ben-Haim and Rosenberg (2002) who, using cultured isolates

# References


of *V. coralliilyticus,* demonstrated that elevated temperatures are crucial to the infection of *P. damicornis*.

Our findings demonstrate, for the first time, that elevated seawater temperature increases the abundance and alters the composition of an environmental *Vibrio* community occurring among a mixed natural microbial community associated with the ecologically important coral species *P. damicornis*. Importantly, these microbial shifts involve a dramatic rise in the relative abundance of pathogens including *V. coralliilyticus.* Our research builds upon previous studies using cultured isolates, to highlight that natural populations of Vibrios, occurring within mixed natural communities of coral associated microbes may rise to prominence under heat stress conditions. Currently, up to a third of all coral species face extinction (Carpenter et al., 2008), with coral disease recognized as a significant and increasing threat. Our data provide direct quantitative support for the theory that increasing sea surface temperature occurring as a result of climate change, will affect coral reefs by promoting an increase in the abundance of coral pathogens.

# Acknowledgments

This research was funded by an Australian Research Council Grant (DP110103091) to JS and RS, a Human Frontiers in Science Program Award (no. RGY0089) to RS and JS, the Australian Coral Reef Society Terry Walker Prize 2012 to JT, and a post-graduate award to JT from the Department of Environmental Science and Climate Change Cluster at the University of Technology Sydney. JS and NW were funded through Australian Research Council Future Fellowships FT130100218 and FT120100480, respectively. We are grateful to the Great Barrier Reef Marine Park Authority for coral collection permits G09/31733.1 (PJ Ralph, University of Technology Sydney).

# Supplementary Material

The Supplementary Material for this article can be found online at: http://journal*.*frontiersin*.*org/article/10*.*3389/fmicb*.* 2015*.*00432/abstract

latitudes in response to ocean warming. *Nat. Clim. Chang.* 3, 73–77.


and dead coral surfaces. *Appl. Environ. Microbiol.* 68, 2214–2228. doi: 10.1128/AEM.68.5.2214-2228.2002


damicornis to bacterial stress from *Vibrio coralliilyticus*. *J. Exp. Biol.* 214, 1533–1543. doi: 10.1242/jeb.053165

Yong, L., Guanpin, Y., Hualei, W., Jixiang, C., Xianming, S., Guiwei, Z., et al. (2006). Design of *Vibrio* 16S rRNA gene specific primers and their application in the analysis of seawater *Vibrio* community. *J. Ocean Univ. China* 5, 157–164. doi: 10.1007/BF02919216

**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 Tout, Siboni, Messer, Garren, Stocker, Webster, Ralph and Seymour. 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 out-of-the-delta hypothesis: dense human populations in low-lying river deltas served as agents for the evolution of a deadly pathogen**

*Yan Boucher <sup>1</sup> \*, Fabini D. Orata <sup>1</sup> and Munirul Alam <sup>2</sup>*

*<sup>1</sup> Department of Biological Sciences, University of Alberta, Edmonton, AB, Canada, <sup>2</sup> Centre for Communicable Diseases, International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B), Dhaka, Bangladesh*

Cholera is a diarrheal disease that has changed the history of mankind, devastating the world with seven pandemics from 1817 to the present day. Although there is little doubt in the causative agent of these pandemics being *Vibrio cholerae* of the O1 serogroup, where, when, and how this pathogen emerged is not well understood. *V. cholerae* is a ubiquitous coastal species that likely existed for tens of thousands of years. However, the evolution of a strain capable of causing a large-scale epidemic is likely more recent historically. Here, we propose that the unique human and physical geography of low-lying river deltas made it possible for an environmental bacterium to evolve into a deadly human pathogen. Such areas are often densely populated and salt intrusion in drinking water frequent. As *V. cholerae* is most abundant in brackish water, its favored environment, it is likely that coastal inhabitants would regularly ingest the bacterium and release it back in the environment. This creates a continuous selection pressure for *V. cholerae* to adapt to life in the human gut.

### *Edited by:*

*Maurizio Labbate, University of Technology, Sydney, Australia*

### *Reviewed by:*

*Dong W. Kim, Hanyang University, South Korea Michelle Dziejman, University of Rochester School of Medicine and Dentistry, USA*

### *\*Correspondence:*

*Yan Boucher yboucher@ualberta.ca*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 01 July 2015 Accepted: 28 September 2015 Published: 19 October 2015*

### *Citation:*

*Boucher Y, Orata FD and Alam M (2015) The out-of-the-delta hypothesis: dense human populations in low-lying river deltas served as agents for the evolution of a deadly pathogen. Front. Microbiol. 6:1120. doi: 10.3389/fmicb.2015.01120* **Keywords:** *Vibrio cholerae***, cholera, pandemic, epidemic, delta, pathogen, evolution, salt intrusion**

# **INTRODUCTION**

Bacteria that are pathogenic to humans need to be able to survive inside the human body, which is vastly different from any environment outside a eukaryotic host (Martínez, 2013). Many bacteria able to infect humans are adapted to other animal hosts, making the jump to survival in humans small evolutionarily. Such bacteria, which are transmitted from animals to humans, are called zoonoses. This is the case for *Yersinia pestis*, the causative agent of plague, of which there have been three pandemics, all originating from foci where this bacterium was found in rodent populations within areas densely populated by humans (Morelli et al., 2010). Other bacteria pathogenic to humans, such as *Pseudomonas aeruginosa*, are opportunistic and not necessarily associated with an animal vector, but rather found on the skin or in various environments such as air, water, and soil. These opportunistic pathogens are not specifically adapted to infect humans but can nonetheless survive inside the immune-compromised host (Martínez, 2014).

Other pathogenic bacteria infecting humans are neither zoonotic nor opportunistic. *Vibrio cholerae*, the cause of the ancient disease cholera, is an environmental bacterium, living mainly in estuarine brackish water regions, either free-living or associated with the surface of various marine

eukaryotes, such as algae, oysters, crustaceans, fish, or copepods (Colwell, 1996). *V. cholerae* can occasionally infect humans, even if they are healthy, and are therefore not opportunistic pathogens, but rather facultative pathogens, which vary in their pathogenic potential to humans from non-virulent to highly virulent (Haley et al., 2014). In the case of *Y. pestis*, the fact that warm-blooded animals (rodents) are its primary host can help explain its pathogenicity toward humans (Perry and Fetherston, 1997). For opportunistic pathogens such as *P. aeruginosa*, we understand that they do not have a specific adaptation to infecting humans but rather a general ability to do so when the host is compromised. For *V. cholerae*, however, it is not clear how a specifically coastal aquatic microbe would have evolved to become one of the deadliest pathogens in human history, killing millions through seven pandemics. *V. cholerae* is an extremely diverse species, harboring more than 200 different serogroups. Among all this diversity, only a single genetic lineage, the phylocore genome (PG) group, which is mostly composed of strains displaying the O1 serogroup, is responsible for all seven known pandemics (**Figure 1**). Did this transition to an ability to cause pandemics simply occur by chance or was it the result of ongoing natural selection?

# *V. CHOLERAE* **IS GENETICALLY PREDISPOSED TO SURVIVE THE HUMAN GUT**

Although most *V. cholerae* are not pathogenic to humans, any random strain of this species found in a coastal environment has several characteristics that would facilitate its survival in the human gut. The first is its ability to grow from 18 to 41°C, allowing it to proliferate in the human body. This is a rare trait among *Vibrio*, most species being unable to grow above 33°C (Brenner et al., 2005). Another key adaptation is the ability to survive a range of salinities, as it is mainly found in ocean to freshwater transition areas but can be isolated from waters with minimal amounts of salinity (Louis et al., 2003). This is also a rare trait among *Vibrio*, most of which tolerate a smaller range of salinities and have a higher minimal requirement for salt (Brenner et al., 2005). *V. cholerae* has to be able to survive low salt conditions to be ingested by humans drinking fresh or slightly brackish water.

*Vibrio cholerae* as a species also displays a number of other characteristics that likely result in a higher survival rate in the human gut. Members of this species naturally form biofilms in aquatic environments, often on the chitinous surface of various marine animals and zooplanktons (Meibom et al., 2004). Interestingly, this ability of *V. cholerae* to attach to chitin and form biofilms provides it with an indirect adaptation for survival in the human gut. The digestive tract contains abundant mucin, which is an analog of chitin. *V. cholerae* can degrade and bind to this mucin like it does to chitin (Wong et al., 2012). Therefore, the ability of *V. cholerae* to form biofilms on chitinous surfaces in aquatic environments likely facilitates this bacterium attachment and survival in the human gut mucilaginous surfaces. These biofilms have been shown to increase resistance to gastric acids and bile salts found in the intestinal tract, and also protect against immune responses (Hung et al., 2006; Almagro-Moreno et al., 2015; Teschler et al., 2015). *V. cholerae* is often shed from human hosts as biofilms, which enhances their infectivity and their survival in the environment. Indeed, while in biofilms, *V. cholerae* can be found in a hyperinfectious physiological state, and infectious dose for biofilm-derived *V. cholerae* can be orders of magnitude lower than that of free-living cells (Faruque et al., 2006). Biofilms also provide *V. cholerae* with resistance to eukaryotic grazing in the environment, enhancing their survival outside the host after shedding (Matz et al., 2005).

It has also been suggested that the type VI secretion system (T6SS), ubiquitously found in *V. cholerae*, could give it an edge in competing with the normal human gut flora (Fu et al., 2013). This system encodes a syringe-like structure that can pierce cellular envelopes of other bacteria and some eukaryotes, injecting effector proteins that can kill the recipient (Pukatzki and Provenzano, 2013; Unterweger et al., 2014). This could allow *V. cholerae* to kill competing bacteria native to the human gut, which are well adapted to this environment and could outcompete it. This secretion system likely evolved in nature because it allowed *V. cholerae* to compete more effectively with other bacteria or avoid predation from eukaryotes, as it has been shown to be effective against a range of microbes, including eukaryotes such as slime molds (Pukatzki et al., 2006).

Also important for bacteria pathogenic to humans is the regulation of gene expression. The expression of virulence factors only in a context in which they are useful (i.e., in the human gut) is critical to the long-term viability of a pathogen. *V. cholerae* displays quorum sensing control of numerous virulence factors, allowing it to produce toxins only when the population is sufficiently large to have an effect on the host (Zhu et al., 2002). Quorum sensing also controls biofilm formation, which we already mentioned as a virulence factor in *V. cholerae*. Another key part of gene regulation in *V. cholerae* is the ToxR regulon. The latter is a network of genes modulated by transcriptional regulators encoded by *toxRS*, *tcpPH*, and *toxT*. It modulates the expression of a number of virulence factors so that they are produced at the right time in the infection process. Although *tcpPH*, *toxS*, and *toxT* appear in low frequency in environmental *V. cholerae* strains, *toxR* is almost always found (Bina et al., 2003). Several other genes considered to be virulence factors are also very frequently found in *V. cholerae* strains regardless of their origin, including RTX (repeat in toxin), MSHA (mannose-sensitive hemolysin agglutination pilin), *pilE* (pilin), *hlyA*(hemolysin), and *nanH* (sialic acid degradation; O'Shea et al., 2004).

All these genetic traits, found in all or a majority of *V. cholerae* strains, almost undoubtedly evolved outside the context of a human host, as most members of this species are harmless residents of aquatic environments. These traits likely evolved by providing a survival advantage in marine coastal environments, such as protecting against predators, increasing resistance to various environmental conditions, outcompeting other bacteria, or associating with resource-rich zooplankton and phytoplankton (Martínez, 2013). For example, the T6SS is useful to defend against predation by eukaryotic grazers or competition with other microbes in the water column (Pukatzki et al., 2006). Various pilins can serve to adhere to surfaces and facilitate biofilm formation in aquatic environments (Almagro-Moreno et al., 2015). These traits provide a basic genetic background that fortuitously makes survival in a human host more likely, but not sufficient for *V. cholerae* to become a human pathogen.

Other virulence factors, found only in a minority of *V. cholerae* strains, add to this background and enhance their potential to cause various illnesses in humans, ranging from ear infection, to septicemia and gastroenteritis. One such factor is the type III secretion system (T3SS), which allows the delivery of effector proteins to a target cell (Chatterjee et al., 2009; Alam et al., 2011a). *V. cholerae* strains of various serogroups harboring the T3SS and/or some of the other virulence factors mentioned earlier have caused sporadic diarrheal episodes in Thailand (Dalsgaard et al., 1999), India (Mukhopadhyay et al., 1995), Peru (Dalsgaard et al., 1995), Nigeria (Marin et al., 2013), and numerous other countries (Dalsgaard et al., 2001; Chatterjee et al., 2009). These virulence factors alone, however, are not usually sufficient to cause an actual epidemic. Strains causing long-lasting and larger-scale outbreaks display two major virulence factors: the cholera toxin (CT) and toxin co-regulated pilus (TCP). Both of these are relatively rare in most environmental *V. cholerae* populations (O'Shea et al., 2004), but are found in all strains of the PG lineage responsible for pandemics (with a few exceptions, likely due to secondary loss; Chun et al., 2009). The CT is the main virulence factor of *V. cholerae*, causing massive release of electrolytes and water in the intestinal lumen through its activation of adenylate cyclase and consequent increase in intracellular levels of cyclic AMP (Vanden Broeck et al., 2007). The genes encoding this toxin are carried by a lysogenic phage, CTXΦ, inserted in the genome of *V. cholerae* (Waldor and Mekalanos, 1996; Faruque and Mekalanos, 2012). It has been shown that CTXΦ can be lost, gained *de novo*, or recombine if several phages are inserted in the genome of a single *V. cholerae* strain (Kim et al., 2015). This has led to a significant variation in the CT and associated genes of strains in the PG group (Safa et al., 2010). The selective advantage this phage provides or the nature of its association with PG *V. cholerae* strains in nature is such that it has remained almost ubiquitously present in members of this lineage, with occasional loss and regain episodes (Kim et al., 2015). TCP, the second major virulence factor of *V. cholerae*, is a pre-requisite to the presence of CTXΦ, as it is the receptor for that phage (Karaolis et al., 1999). TCP is also a virulence factor on its own, as it facilitates bacterial interaction through direct pilus-pilus contact required for microcolony formation, playing a role in adhesion and biofilm formation (Almagro-Moreno et al., 2015). The capacity to become toxigenic (produce CT) is believed to have evolved through the sequential acquisition of TCP and CTXΦ (Faruque and Mekalanos, 2003). Although the origin of these major virulence factors is unknown, homologs of both can be found in *Aliivibrio fischeri* symbionts of squid, a distant relative of *V. cholerae* (Ruby et al., 2005). The phage found in *A. fischeri* lacks the CT, but is otherwise very similar to CTXΦ. The TCP gene cluster of *A. fischeri* is missing a few genes present in its *V. cholerae* homolog, but is otherwise very similar. This suggests that other bacterial species could have been the source for the major virulence factors of *V. cholerae* from the PG group. The presence of CTXΦ and TCP is not limited to the PG lineage, and strains belonging to unrelated lineages with various serogroups have been found to contain one or both of these genes (Chun et al., 2009). Although these non-PG strains carrying CTXΦ and TCP have caused sporadic diarrheal episodes at various times across the world, the combination of virulence factors they harbor is different from that of PG strains and insufficient for causing actual pandemics of the scale observed historically for cholera. Of all *V. cholerae* strains that have caused outbreaks, only the PG group is known to have spread far and wide and survived through more than a few years as a lineage infecting humans. The main distinctive characteristics likely harbored by the ancestor of this group are the presence of CT, TCP (as part of the *Vibrio* pathogenicity island-1 or VPI-1), VPI-2, and the O1 antigen (Chun et al., 2009). A range of other factors contributing to pandemic potential is also likely present in this lineage, but the identity of many of them is still unknown. A study, which passaged a population of *V. cholerae* O1 transposon mutants through infant rabbits, found 133 genes contributing to survival in the mammalian host (Kamp et al., 2013). This included all the virulence factors already mentioned for PG *V. cholerae* O1, but also genes with identified functions in purine and pyrimidine biosynthesis and amino acid metabolism, as well as genes with putative function of phosphate acquisition, posttranslational modification, fatty acid metabolism, and transport. This highlights that a constellation of genes was assembled to give pandemic potential to harmless aquatic *V. cholerae*. Lateral gene transfer is frequent among *V. cholerae* and most virulence factors are found on mobile elements such as pathogenicity islands, phages, or integrative conjugative elements (Faruque and Mekalanos, 2003). This means that various combinations of virulence genes must be regularly created in environmental reservoirs where *V. cholerae* strains interact. However, since the ability to become pandemic, as opposed to merely infecting humans or causing sporadic outbreaks, is a complex trait likely requiring over a 100 genes, sustained natural selection would be necessary for the incremental build-up of this complexity. What selective pressure made it possible for this combination of genes to be successfully fixed in a population of PG *V. cholerae* and maintained in most of its descendants for centuries? For bacteria linked to historically important epidemics or pandemics such as the plague, the general answer is relatively simple. Such zoonotic bacteria undergo selective pressure for adaptation to life in their warm-blooded animal hosts and are consequently able to survive in humans (Martínez, 2013). For example, plague pandemics resulted from the contact of rodent populations (*Y. pestis* primary host) with dense human populations in China (Morelli et al., 2010; Wagner et al., 2014). However, although *V. cholerae* can be found associated with zooplanktons such as copepods, these animals are invertebrates very different from the human host. This association is unlikely to be sufficient for the evolution of pandemic strains, and another source of long-term selection would be required.

# **THE HISTORY OF PANDEMIC CHOLERA**

How and where pandemic cholera first evolved is a fundamental question that has been elusive for decades. This is due to the difficulty of defining cholera precisely (as it has a broad clinical spectrum) and to distinguish it from other diseases that cause vomiting and diarrhea (Barua, 1992). Only since the advent of serological and molecular methods have we been able to diagnose cholera with reliability and trace the relationship between various epidemics.

Seven pandemics of cholera have been recorded in medical history, the first one starting in India in 1817 and the last in Indonesia in 1961 (Barua, 1992). Phylogenetic analyses based on whole genomes and multi-locus sequence typing suggest a common ancestor for strains of *V. cholerae* belonging to all seven pandemics (**Figure 1**). These pandemic strains and their close relatives have been termed the PG group (Chun et al., 2009), and their common ancestor has been estimated to date back to anywhere from the beginning of sedentary agriculture (10,000 years ago) to 430 years ago (Devault et al., 2014). There is little contention that cholera-like diseases were present in Asia and Europe since ancient times, based on descriptions of patient symptoms closely matching those of cholera found in ancient texts and records (Barua, 1992). The earliest texts describing choleralike symptoms appeared in Greece (Hippocrates, fourth century B.C. and Aretaeus of Cappadocia, first century A.D.) and in India (Sushruta Samhita, fifth century B.C.). Reports of cholera from Arabic scholars Rhazes and Avicenna can be found in the tenth and eleventh centuries A.D. (Barua, 1992). However, these writings describe sporadic cases, not on the epidemic scale cholera is known for. *V. cholerae* strains outside the PG group can cause symptoms similar to pandemic cholera, and could have been the cause of these cases. Numerous reports of cholera can also be found for Europe between the sixteenth and eighteenth century. In Asia, the first written accounts in India after that of Sushruta Samhita started in 1503, shortly after the Portuguese settled in Goa, with some records for the disease being present in Delhi in 1325 and in Merwah in 1428 and again in Goa in 1563 (Barua, 1992). There were reports of the disease in India and neighboring countries such as Bangladesh and Indonesia throughout the seventeenth and eighteenth centuries. The frequent reports of cases matching closely the symptoms of cholera in Europe and Asia from the sixteenth century onwards, sometimes as epidemics, suggest that the disease was widespread before the first pandemic caused by the PG group of *V. cholerae* in 1817. However, *V. cholerae* was not isolated in pure culture until 1884 and historical medical records are hard to interpret, with frequent confusion of cholera with other diarrheal diseases before and even during the pandemics. It is therefore not possible to determine with certainty that these epidemics were caused by *V. cholerae* from the PG lineage.

There is evidence suggesting that pre-pandemic cholera cases were caused by strains genetically related to isolates from the pandemics. The PG group responsible for pandemics, includes only strains of the O1 serogroup (with a few exceptions that are clearly derived traits, such as the O139 serogroup in *V. cholerae* El Tor Bengal and O37 in *V. cholerae* V52 Sudan; Chun et al., 2009). It is divided into two main branches, which are significantly divergent from each other. One branch (PG-2) includes the classical biotype (responsible for the sixth and presumably the earlier pandemics) and the other (PG-1) includes the El Tor biotype (responsible for the current and seventh pandemic). Strains related to both the El Tor and classical biotypes but temporally or geographically unlinked to the pandemics are found in both PG-1 and PG-2 groups. The *V. cholerae* V52 strain isolated from a 1968 epidemic in Sudan, despite having a different serogroup, is clearly related to the ancestor of classical strains that caused the first six pandemics (**Figure 1**). Multiple strains related to the El Tor biotype, but not associated with the seventh pandemic, have also been found. These strains seem to be native to Australia (various rivers in Queensland), the USA (Gulf Coast), Thailand, and Russia, and are all of the O1 serogroup and clearly phylogenetically closely related to the seventh pandemic strains (**Figure 1**). This suggests that *V. cholerae* belonging to the PG-1 group had spread globally before the strain ancestral to the seventh pandemic emerged in Indonesia. The latter likely evolved locally, as epidemics caused by the El Tor biotype started in 1937 in Sulawesi (Indonesia), until they spread globally in 1961, starting the seventh pandemic (Salim et al., 2005). *V. cholerae* related to the ancestor of the classical strains that caused the first six pandemics are also likely to have been dispersed globally. The V52 Sudan strain was found far from the origin of the first pandemic in India, and is too divergent from sixth pandemic strains to have been descended from them, despite being part of the PG-2 group (**Figure 1**). Consequently, it is possible that the ancestor of the PG group had spread widely before the first pandemic, with its descendants locally evolving in India into the classical biotype and in Indonesia into the El Tor biotype (Feng et al., 2008). This would be consistent with cholera cases being widely observed in Europe and Asia well before the pandemics (Barua, 1992).

# **A PROPOSED ORIGIN OF PANDEMIC CHOLERA IN HISTORICALLY DENSELY POPULATED RIVER DELTAS**

We propose that extensive contact between *V. cholerae* and humans was made possible by high population densities in river deltas where humans drink the brackish surface waters where this bacterium lives. *V. cholerae* then circulates via the fecal-oral route in local populations and environmental reservoirs, leading to a selection and enrichment of variants capable of surviving in the human gut and eventually infecting individuals.

Indeed, *V. cholerae* occurs widely in coastal areas as it strongly favors low salinities (between 2 and 14 ppt) and is tolerant of a wide range in temperatures (18–41°C; Louis et al., 2003; Brenner et al., 2005), with highest densities being achieved in warmer temperatures (*>*25°C; Takemura et al., 2014). People living in coastal areas of Bangladesh (the Ganges Delta), for example, ingest water from rivers or shallow wells with salinities averaging between 2.8 and 8.2 ppt and temperatures between 26 and 35°C during the dry season (Khan et al., 2011). Today, of the *∼*400 million people inhabiting the Ganges River Basin (**Figure 2**), there are around 20 million living in coastal areas of Bangladesh who are affected by salt intrusion to varying degrees (Khan et al., 2011). If the proportion of the population in the Ganges Basin exposed to salt intrusion was historically similar to what is observed today, around 6–7 million people would have been regularly drinking brackish water in that area in 1800 A.D. (McEvedy and Jones, 1978). We suggest that such a large number of people constantly ingesting water at an ideal temperature and salinity for *V. cholerae* will be exposed to a significant amount of the bacteria. *V. cholerae*

abundance can reach up to 6 *<sup>×</sup>* <sup>10</sup><sup>3</sup> cells/ml in water (Heidelberg et al., 2002), or higher if they are found as biofilms on the surface or in the gut of zooplanktons, each microscopic animal carrying up to 1 *<sup>×</sup>* <sup>10</sup><sup>4</sup> *V. cholerae* cells (Colwell, 1996). It is not possible to know the historical abundance of *V. cholerae* cells in waters of river deltas such as the Ganges, but it is unlikely that it would be much different from what is observed today in coastal areas across the world. Low-lying river deltas are the places more likely to put humans in contact with *V. cholerae*, as these are traditionally highly populated because of their fertile soils and their close contact with the ocean makes salt intrusion in drinking water likely. Deltas in warm regions, which have been densely populated for long historical periods, are therefore the most likely place for fecal-oral transmission and evolution of *V. cholerae* as a pandemic pathogen. Deltas outside Eurasia are less likely candidates, as their historical population densities are relatively lower (Goldewijk et al., 2010). Asia is home to many of the world's largest and historically most densely populated deltas (the Indus in Pakistan, Ganges/Brahmaputra in India and Bangladesh, Mekong in Vietnam, Huang He and Yangtze in China) and is where all recorded cholera pandemics have started, making it the most likely region for the origin of this disease. The areas surrounding the Huang He, Yangtze, and Ganges rivers have been estimated as some of the most densely populated areas in the world from the beginning of the Holocene period (10,000 B.C. to the present; Goldewijk et al., 2010; **Figure 2**). The population of the Indian subcontinent was already estimated at over 30 million in 200 B.C., 20 million of which lived in the Ganges Basin (McEvedy and Jones, 1978). This population would grow to an estimated 200 million by the time the British took control around 1800 A.D., with an estimated 130 million in the Ganges Basin. China was similarly densely populated, with an estimated 40 million people in 200 B.C., mostly centered around the Huang He and Yangtze, growing to around 300 million in 1800 A.D. (McEvedy and Jones, 1978). These two areas alone represented 40% of the world population in 200 B.C. and over 55% in 1800 A.D. (McEvedy and Jones, 1978).

Although it is not possible to determine exactly from which of the world's large river deltas cholera would have originated, the Ganges Delta is a likely candidate for several reasons. Among all of the world deltas, the only place where cholera has been continually endemic since the start of the pandemics is the Ganges Delta. Despite similar population densities and sanitary conditions, other deltas in Southeast Asia do not maintain cholera transmission, indicating that other local determinants are critical for the maintenance of cholera (Pascual et al., 2002). The Ganges Delta is unique in its pattern of cholera incidence. Most locations with endemic cholera have a single peak of cholera incidence during the warm season of the year, in which water temperatures rise. The Ganges Delta, however, has a dual peak pattern of cholera incidence, which is high in spring and fall. Low flow in the Ganges and Brahmaputra in the spring can help seawater intrusion, which is favorable to the growth of *V. cholerae*. The low flow also helps the movement of *V. cholerae* inland with marine phytoplankton and zooplankton to which it is associated (Akanda et al., 2009). Higher flow volumes of these rivers in monsoon season, which increases the extent of flood-affected areas, are responsible for high fall incidence (Bouma and Pascual, 2001).

Cholera is more prevalent in coastal districts in the spring and in inland regions in the fall (Sack et al., 2003), suggesting that cholera enters the human population from infections initiated on the coast and then spreads inland, facilitated by shedding from infected individuals and spreading from high water levels and floods. Epidemics can also occur outside this regular pattern when abnormally high rainfall and associated flooding occurs (Alam et al., 2011b). This highlights that floodwaters transmit infectious clones of *V. cholerae*, which then circulates via the fecal-oral route, causing an epidemic.

There is also evidence of selective pressure exerted by cholera on the human population of the Ganges Delta. Blood group O is linked to an increased risk of severe cholera symptoms and has a low prevalence in the Ganges Delta area (Nelson et al., 2009). Other genetic factors are also likely to be involved in cholera susceptibility, as biologically related household contact of a patient are three times as likely to contract cholera as an unrelated contact living in the same household. Strong selective pressure has been detected on key pathways of innate immunity among the Bengali residents of Dhaka (Karlsson et al., 2013). Most people in Bangladesh have developed some immunity to cholera by the age of 10–15, vibriocidal antibodies being detectable in their blood.

Given this long-term influence of *V. cholerae* on human populations, it is likely that humans have also influenced the evolution of *V. cholerae*. The seventh pandemic has spread across the world in three waves since its start in 1961, each caused by a specific genetic variant. It has recently been shown that all three of these variants evolved in the Bay of Bengal region (Mutreja et al., 2011). The process by which established pandemic *V. cholerae* strains are successively globally replaced by novel variants is not well understood. Even if it is assumed that travelers, oceanic currents, and ship ballasts can disseminate novel strains worldwide (Ruiz et al., 2000), older established strains still have to be outcompeted by the newcomers. It has been suggested that existing human immunity to older variants in endemic areas could give an advantage to novel strains, but the possibility that environmental factors are involved in this evolution cannot be excluded (Kim et al., 2015). Consistent with the suggestion that humans have influenced the evolution of pandemic *V. cholerae* strains over the last two centuries, our hypothesis proposes that humans are also responsible for their origin. Through their constant ingestion and shedding of *V. cholerae* from environmental populations found in local brackish water sources, humans selected for variants capable of residing in their gut and causing cholera.

# **TRACING THE ORIGINS OF PANDEMIC CHOLERA**

Our hypothesis implies several specific predictions, which could be falsified. It predicts that humans living in coastal areas and drinking brackish water where *V. cholerae* is present should harbor this bacterium in their gut. The residence time of *V. cholerae* in the gut of such brackish water consumers is an open question, but it is likely to be transient, as even pandemic strains rarely reside in the gut for long periods of time (Sack et al., 2004). A portion of strains found in the human gut should be specifically adapted to that environment and be a non-random subsample of the environmental population. There will also likely be an enrichment of such human-adapted genotypes in the environmental population, compared to coastal locations where no humans live. An analogy could be drawn here to *A. fischeri* symbionts of squids. These have specific genes allowing them to find the light organ of squids and survive there (Wier et al., 2010). They are released back in the environment every morning, where they mix with other *A. fischeri*, which are not adapted to colonizing squids.

Although non-pandemic (non-PG) *V. cholerae* have been found either alone or associated with the PG strains in various sporadic diarrhea episodes (Ramamurthy et al., 1993) or epidemics (Hasan et al., 2012), it as yet to be found in the gut of healthy individuals. This is because our focus during epidemics is naturally on patients, and microbes are usually isolated from individuals that are sick, not the water that they drink or food they consume. This results in vast amount of information on the pandemic *V. cholerae* strains, but very little on other strains that could also be present in the gut of healthy humans. Culture-based studies, although sometimes done on samples from household contacts of cholera patients who could be healthy, usually target PG *V. cholerae* specifically (by using antibiotic selection and enrichment cultures) and rarely allow the identification of other strains. Culture-free studies of the human gut suffer from the same sampling bias toward infected patients. Although the gut flora of numerous healthy humans has been analyzed through microbiome studies across the world, we have yet to sample healthy humans regularly drinking coastal brackish water. Furthermore, current molecular analyses of the human gut is limited to the 16S *rRNA* gene, which lacks the resolution to differentiate between various *Vibrio* species that could be found there (Hsiao et al., 2014). It is of interest that one study using the 16S *rRNA* gene to look at the gut microbiota of children with cholera found traces of *Vibrio* a month after patients were admitted, even though the infection was treated successfully in a few days (Monira et al., 2013). This raises the possibility that *V. cholerae* could reside in the gut for prolonged periods of time. It is also known that pandemic *V. cholerae* infections can often be asymptomatic, meaning that the bacteria can be present in the human gut without being detected (King et al., 2008).

A second prediction of our hypothesis is that virulence factors contributing to cholera, such as CT and TCP, should be more abundant in an environment selecting for such factors. There should also be a larger diversity of virulence determinants. For example, a comparison of such genes in the Ganges Delta and a coastal cholera-free location in North America should reveal a higher abundance and variety of virulence factors in the former than the latter. This would be due to an enrichment of those factors in the general *V. cholerae* population because of ongoing selection in the human gut.

If the origin of pandemic PG group *V. cholerae* is not very ancient, it could also be possible to trace it to a particular geographical area. Extent relatives of the PG lineage could be found in a specific delta, similarly to what has been done to trace the origins of plague pandemics inside or near China (Morelli et al., 2010). Biogeographical study of the human pathogen *Vibrio parahaemolyticus* (mostly seafood-borne) shows that it is not panmictic (with all individuals free to move between locations), but that the oceans surrounding Asia have a distinct gene pool (Cui et al., 2015). Since *V. parahaemolyticus* is a marine organism, as opposed to *V. cholerae* which is mostly estuarine, the latter is likely to show even more biogeographical structure in its global distribution. If extent relatives of the PG lineage can be found, it is likely that its geographical origin could be confirmed. This would likely require extensive sampling of *V. cholerae* across the world, but such an effort is not beyond what has been done for several other bacterial pathogens.

It has recently been proposed that dysbiosis, a change in the proportion of various species in the normal microbiota, promotes lateral gene transfer (Stecher et al., 2013). *V. cholerae* infections often cause dysbiosis, as a massive bloom of *V. cholerae* in the intestine completely changes the normal microbiota (Hsiao et al., 2014). In a scenario in which *V. cholerae* progressively adapts to the human gut by regular ingestion, lateral gene transfer from gut microbes is likely, and traces of such transfer might be found inside its genome.

# **CONCLUDING REMARKS**

*Vibrio cholerae* is unusual as a human pathogen, as some representatives of this species can cause pandemics but are not zoonotic bacteria, instead having an environmental reservoir in coastal waters. This makes the evolution of variants with epidemic potential a fascinating mystery. A single phylogenetic lineage is known to have led to all seven pandemics, making evolution of pandemic variants a rare event. As for other human pathogens, extended contact with our species or other warmblooded animals is likely necessary to yield highly virulent lineages. For a coastal bacterium, such contact was possible in human populations regularly drinking warm brackish waters, the favored environment of *V. cholerae*. We could be able to determine where pandemic cholera originated through a biogeographical study of the *V. cholerae* species, if the PG lineage is not panmictic. It might also be possible to understand the process through which it evolved by investigating the gut microbiota of healthy humans consuming water affected by salt intrusion from the ocean. This would represent a local impact of human population on a bacterial species, influencing its evolution with dire consequences.

# **AUTHOR CONTRIBUTIONS**

YB wrote the manuscript, FDO revised the manuscript and created the figure, MA contributed the original hypothesis idea and revised the manuscript.

# **ACKNOWLEDGMENTS**

We are grateful to Jesse Shapiro (Université de Montréal) for the helpful discussions and feedback. YB is funded by the Natural Sciences and Engineering Research Council of Canada and the Canadian Institute for Advanced Research and FDO is funded by the Alberta Innovates – Technology Futures.

# **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 Boucher, Orata and Alam. 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.*

# Impact of CO2 leakage from sub-seabed carbon dioxide capture and storage (CCS) reservoirs on benthic virus–prokaryote interactions and functions

*Eugenio Rastelli1,2, Cinzia Corinaldesi1, Antonio Dell'Anno1, Teresa Amaro3,4, Ana M. Queirós5, Stephen Widdicombe5 and Roberto Danovaro1,2\**

*<sup>1</sup> Department of Environmental and Life Sciences, Polytechnic University of Marche, Ancona, Italy, <sup>2</sup> Stazione Zoologica Anton Dohrn, Naples, Italy, <sup>3</sup> Hellenic Center for Marine Research, Heraklion, Greece, <sup>4</sup> Norwegian Institute for Water Research, Bergen, Norway, <sup>5</sup> Plymouth Marine Laboratory, Plymouth, UK*

### *Edited by:*

*Federico Lauro, University of New South Wales, Australia*

### *Reviewed by:*

*Jennifer F. Biddle, University of Delaware, USA Hélène Montanié, Université de la Rochelle, France*

### *\*Correspondence:*

*Roberto Danovaro, Department of Environmental and Life Sciences, Polytechnic University of Marche, Via Brecce Bianche, 60131 Ancona, Italy r.danovaro@univpm.it*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 15 July 2015 Accepted: 24 August 2015 Published: 07 September 2015*

### *Citation:*

*Rastelli E, Corinaldesi C, Dell'Anno A, Amaro T, Queirós AM, Widdicombe S and Danovaro R (2015) Impact of CO2 leakage from sub-seabed carbon dioxide capture and storage (CCS) reservoirs on benthic virus–prokaryote interactions and functions. Front. Microbiol. 6:935. doi: 10.3389/fmicb.2015.00935* Atmospheric CO<sup>2</sup> emissions are a global concern due to their predicted impact on biodiversity, ecosystems functioning, and human life. Among the proposed mitigation strategies, CO<sup>2</sup> capture and storage, primarily the injection of CO<sup>2</sup> into marine deep geological formations has been suggested as a technically practical option for reducing emissions. However, concerns have been raised that possible leakage from such storage sites, and the associated elevated levels of pCO<sup>2</sup> could locally impact the biodiversity and biogeochemical processes in the sediments above these reservoirs. Whilst a number of impact assessment studies have been conducted, no information is available on the specific responses of viruses and virus–host interactions. In the present study, we tested the impact of a simulated CO<sup>2</sup> leakage on the benthic microbial assemblages, with specific focus on microbial activity and virus-induced prokaryotic mortality (VIPM). We found that exposure to levels of CO<sup>2</sup> in the overlying seawater from 1,000 to 20,000 ppm for a period up to 140 days, resulted in a marked decrease in heterotrophic carbon production and organic matter degradation rates in the sediments, associated with lower rates of VIPM, and a progressive accumulation of sedimentary organic matter with increasing CO<sup>2</sup> concentrations. These results suggest that the increase in seawater pCO<sup>2</sup> levels that may result from CO<sup>2</sup> leakage, can severely reduce the rates of microbial-mediated recycling of the sedimentary organic matter and viral infections, with major consequences on C cycling and nutrient regeneration, and hence on the functioning of benthic ecosystems.

Keywords: viral infection, benthic prokaryotes, biogeochemical cycles, climate change, heterotrophic carbon production, enzymatic activity

### Introduction

Atmospheric CO2 emissions are driving ocean warming and acidification with consequences for marine biodiversity and ecosystem functioning. This is causing increasing concern among researchers, policy makers and the public alike, especially in light of the ongoing increase in atmospheric CO2 concentrations expected under the majority of future emission scenarios (Orr et al., 2005; IPCC, 2014).

Carbon dioxide (CO2) capture and storage (CCS) has been proposed as a technical method to reduce CO2 emissions and thus mitigate the growing climate impacts, by concentrating and injecting the CO2 into specifically dedicated geological reservoirs (Schrag, 2009). Due to issues of accessibility, the majority of these reservoirs are sub-seafloor geological formations in coastal areas, raising concerns over the possibility that such activities could lead to CO2 leakage, especially during the period of active injection (Shaffer, 2010; de Vries et al., 2013; Taylor et al., 2014). The effects of CO2 leaking from subseafloor reservoirs on the marine environment includes localized seawater and sediment acidification as well as a number of changes in other physical–chemical properties (Gehlen et al., 2011; Lichtschlag et al., 2014; Queirós et al., 2014; Rodríguez-Romero et al., 2014), many of which have been shown to alter benthic assemblages and biogeochemical processes (Laverock et al., 2013; Widdicombe et al., 2013, 2015; Tait et al., 2014). These studies have also highlighted a variety of responses amongst different species and taxa, depending on the local hydrodynamic regime, and the duration/extent of the leakage (de Vries et al., 2013; Widdicombe et al., 2013; Blackford et al., 2014). Most studies reported also that the scale of the impact on benthic biota increases with increasing CO2 leakage (Dewar et al., 2013; Phelps et al., 2015; Shitashima et al., 2015), but results are not consistent amongst different storage sites and with changing seasons (Jones et al., 2015). The large variability in the biological response to CO2 is particularly evident in coastal ecosystems where, compared to the deep sea, the lower pressures, higher temperatures and the high variability of the physical–chemical variables can alter the leakage process and the biotic components. A number of uncertainties still surround efforts to predict the impact of leakage from CCS, and this limits our ability to evaluate costs vs. benefits and to make final decisions about the suitability of the CCS approach (de Vries et al., 2013).

The effects of CO2 leakage on benthic microbial processes has received limited attention so far (Jones et al., 2015). Results from short-term *in situ* experiments (benthic chambers) of the impact of elevated CO2 revealed that high CO2 concentrations increased the relative importance of Archaea over Bacteria (Ishida et al., 2013). Whilst two separate sets of chamber experiments conducted on water samples provided contrasting results in terms of the impact of elevated CO2 on bacterial C production (Coffin et al., 2004; Piontek et al., 2013). However, while available information indicates that elevated levels of CO2 can impact upon microbial organisms and processes, no data are currently present to determine the likelihood of impacts on virus–host interactions under such CO2-leakage scenarios. Marine viruses play a key role in biogeochemical processes (Suttle, 2007; Danovaro et al., 2008; Jover et al., 2014; Dell'Anno et al., 2015) and sediments are hot-spots of viral infections with rates up to 1,000 times higher than in the overlying water column (Danovaro and Serresi, 2000; Danovaro et al., 2008). Since all biotic components are potentially affected by viruses, the effect of any environmental stressor (such as CO2) on marine viruses could have important consequences far beyond this specific component of the ecosystem.

In the present study, we used a controlled mesocosm experiment to investigate the impact of a simulated CO2 leakage on benthic microbial assemblages, with focus, for the first time, on virus–host interactions and their implications on organic matter cycling.

### Materials and Methods

### High-CO2 Experimental Setup and Sample Processing

Acidification experiments were conducted over a period of 140 days to simulate the potential effects on benthic ecosystems of the spreading above marine sediments of CO2-enriched seawater plumes originating from a CCS reservoir or pipeline. Sediment samples were collected during August 2012 in the Oslofjord coastal area (59◦49.4788 N, 10◦58.8595E), Norway, at 100-m depth, using a KC Denmark box corer. We retrieved a total of 50 independently collected liners (i.e., cores with 0.1 m × 0.1 m surface area each and average sediment penetration of ∼40 cm), which were then immediately transferred to the benthic mesocosm system at the Norwegian Institute of Water Research, Solbergstrand, Norway. The experimental mesocosms were setup according to Widdicombe et al. (2009). Briefly, all liners were kept in complete immersion during transport to the mesocosm system to prevent desiccation and temperature changes, and then placed in a flow-through holding basin filled with seawater to a depth of 1 m for ∼4 weeks for acclimatization. A pipeline situated at 60 m in the adjacent fjord continuously supplied the holding basin with filtered natural seawater. The liners were then distributed randomly and in equal numbers amongst CO2 treatments, replicating increasing levels of acute acidification in relation to the natural (control) conditions at the collection site. Five concentrations were used to mimic the impact of plumes of acidified seawater emanating from a theoretical point source leak of CO2 from a sub-seabed geological reservoir. The levels chosen were 400 ppm (control) and 1,000, 2,000, 5,000, and 20,000 ppm, according to a hypothesized pCO2 decreasing gradient from the site of leakage (20,000 ppm) to the unaffected area (400 ppm). Each liner received a constant supply of seawater within the mesocosm at a flow rate of 120 mL min−1. The day light regime in the basin was 8:16 h (light–dark). Sampling took place at the beginning of the experiments (0 days), after 14 days and again after 140 days to assess the short-term longterm effects of the CO2 exposure. Seawater acidification was achieved as described by Widdicombe et al. (2009). Briefly, CO2 gas (supplied in the form of very fine bubbles, enabling their rapid solution in the seawater) passed through large (450 L) reservoir tanks filled with natural seawater. The CO2 flux was regulated via an automated feedback relay system (Walchem) to maintain a set pH level, and the reservoir tanks were continuously supplied with natural seawater (pH∼8.1). The required pH levels for each desired pCO2 treatment were calculated using CO2Sys based on the alkalinity values measured from the supply seawater. Seawater temperature, salinity, oxygen concentration and pH were monitored three times a week in each liner and in header tanks, using macro probes. The pH in the sediments was acquired in two replicate liners per treatment using Dual Lifetime Referencing based optical sensors, with sampling and calibration and temperature compensation methods described in Queirós et al. (2014).

Sediment samples from the top 0-1-cm sediment layer were collected by using sterile Plexiglas<sup>R</sup> tubes (Polymethyl methacrylate; 5.5 cm inside diameter) to investigate benthic prokaryotic variables, virus–host interactions and sedimentary organic matter content, composition, and degradation rates. The sediment samples were immediately processed for the analysis of viral and prokaryotic abundance, and for the determination of phytopigments, lipids, carbohydrates, and proteins concentration as detailed below. The analyses of viral production, prokaryotic heterotrophic carbon production and extracellular enzymatic activity (aminopeptidase) were conducted by means of time course experiments at *in situ* temperature as following described. All variables were analyzed in at least three replicates, and all data were normalized to sediment dry weight after desiccation (48 h at 60◦C).

### Phytopigments, Organic Matter Composition, and Extracellular Enzymatic Activities

Phytopigments were extracted in 5 ml of 90% acetone (12 h at 4◦C in the dark) and analyzed fluorometrically for the determination of chlorophyll-*a* concentrations. Samples were then added with 200 µl of 0.1 N HCl and analyzed fluorometrically for the determination of phaeopigments concentrations.

The concentration of proteins, carbohydrates, and lipids was measured spectrophotometrically in the top 1-cm sediment layer (Danovaro, 2010). The sum of the carbohydrate, protein, and lipid concentrations converted into carbon equivalents (by using the conversion factors of 0.40, 0.49, and 0.75 µg C µg<sup>−</sup>1, respectively) was defined as biopolymeric carbon (BPC; Pusceddu et al., 2009; Danovaro, 2010).

The extracellular aminopeptidase activities (used as proxy of potential mobilization and consequent utilization of proteins) were determined by the analysis of the cleavage rates of the artificial fluorogenic substrate L-leucine-4-methylcoumarinyl-7 amide (Leu-MCA, Sigma Chemicals) under saturating substrate concentrations (Danovaro, 2010). Briefly, sediment sub-samples were diluted with 0.02-µm pre-filtered seawater collected at the water-sediment interface from each liner, and incubated in the dark at the *in situ* temperature for 1 h. The fluorescence of the samples was measured fluorometrically (380 nm excitation, 440 nm emission), immediately after the addition of the substrate and after the incubation, and converted into enzymatic activity using standard curves of 7-amino-4-methylcoumarin (Sigma Chemicals). The amount of the artificial fluorogenic substrate hydrolyzed by proteases were converted into protein degradation rates using 72 µg of C per micromole of substrate hydrolyzed.

The cell-specific degradation rates were calculated by the ratio of aminopeptidase activity and the corresponding total prokaryotic abundance in each sample.

The turnover times of proteins in the sediment, used as a proxy of protein cycling efficiency, were calculated as the ratio of the whole protein concentrations and their degradation rates converted into C equivalents.

### Total Prokaryotic Abundance

Total prokaryotic abundance was determined by epifluorescence microscopy as described in Danovaro et al. (2001). Briefly, the sediment samples were treated by ultrasounds (Branson Sonifier 2200, 60W) three times for 1 min after addition of 0.2 µm prefiltered tetrasodium pyrophosphate solution (final concentration, 5 mM), then properly diluted before filtration onto 0.2 µm pore-size Nuclepore black filters (Whatman). Filters were then stained with SYBR Green I (Sigma Chemicals) by adding, on each filter, 20 µl of the stock solution (previously diluted 1:20 with 0.2 µm pre-filtered Milli-Q water), washed twice with 3 ml sterilized Milli-Q water and mounted onto microscope slides. Filters were analyzed using epifluorescence microscopy (Zeiss Axioskop 2MOT, magnification 1,000×). For each filter, at least 20 microscope fields were observed and at least 400 cells counted.

### Viral Abundance, Production, and Virus-Induced Prokaryotic Mortality

Viral abundance was determined by epifluorescence microscopy according to the procedure described in Noble and Fuhrman (1998) and applied to the sediments as described in Danovaro (2010). The sediment samples were sonicated three times (Branson Sonifier 2200, 60W) for 1 min after addition of 0.02 µm pre-filtered tetrasodium pyrophosphate solution in seawater (final, 5 mM). In order to eliminate uncertainties in virus counting due to extracellular DNA interference, subsamples were supplemented with DNase I from bovine pancreas (10 U mL−<sup>1</sup> final concentration) and incubated for 15 min at room temperature. The samples were properly diluted with 0.02 µm pre-filtered seawater, filtered onto 0.02-µm-pore-size Al2O3 filters (Anodisc; diameter 25 mm) and then stained with 100 µl of SYBR Gold 2x (diluting stock solution with 0.02-µm pre-filtered TE buffer (10 mM Tris-HCl, 1 mM EDTA). Filters were incubated in the dark for 20 min, rinsed three times with 3 ml of 0.02 µm pre-filtered Milli-Q water, dried under laminar flow hood and then mounted on glass slides with 20 µl of antifade solution (50% phosphate buffer pH 7.8, 50% glycerol, 0.5% ascorbic acid). Viral counts were obtained by epifluorescence microscopy (Zeiss Axioskop 2MOT, magnification 1,000×) examining at least 20 fields per slide, and at least 400 viral particles per filter.

Viral production rates were determined by time-course experiments using the dilution approach (Dell'Anno et al., 2009). Briefly, replicate samples (*n* = 3) for viral counts were collected immediately after dilution of the sediments and after 1–3, 3– 6, and 12 h of incubation in the dark at *in situ* temperature. Subsamples were then analyzed as reported for the determination of viral abundance. Virus-induced prokaryotic mortality (VIPM) was calculated as follows:

### VIPM =

*(*Viral production*/*Burst Size*)* × 100*/*Prokaryotic Abundance

We assumed a burst size of 45 viruses cell−<sup>1</sup> reported for marine sediments worldwide (Danovaro et al., 2008).

The C released by the viral shunt was calculated by converting the number of killed prokaryotes into C content using 20 fg C cell−<sup>1</sup> as conversion factor (Danovaro, 2010).

### Prokaryotic Heterotrophic C Production

The determination of prokaryotic heterotrophic carbon production was carried out using the method of 3[H]–leucine incorporation, according to the procedure described in van Duyl and Kop (1994) as modified in Danovaro (2010). Briefly, sediment sampl were added with 0.2-µm pre-filtered seawater, containing [3H]-leucine (68 Ci mmol−1; final 0.5–1.0 µM), then incubated in the dark and at the *in situ* temperatures. Time-course experiments over 6 h and concentration-dependent incorporation experiments (from 0.05 to 5.0 µM leucine) were also carried out to define the linearity and the saturation level of the [3H]-leucine incorporation, respectively. Blanks (*n* = 3) for each sediment sample were added with ethanol immediately before 3[H]-leucine addition. After incubation, samples were supplemented with ethanol (80%), centrifuged, washed again two times with ethanol (80%), and the sediment was finally re-suspended in ethanol (80%) and filtered onto polycarbonate filters (0.2 µm pore size; vacuum *<*100 mm Hg). Subsequently, each filter was washed four times with 2 ml of 5% TCA, then transferred into a Pyrex tube containing 2 ml of NaOH (2 M) and incubated for 2 h at 100◦C. After centrifugation at 800×*g*, 1 ml of supernatant fluid was transferred to vials containing the appropriate scintillation liquid. The incorporated radioactivity in the sediment samples was measured with a liquid scintillation counter (PerkinElmer-Packard Tri-Carb 2100 TR).

### Statistical Analyses

To test for differences in the investigated variables between different treatments and exposure time, a two-way analysis of variance was conducted using distance-based permutational multivariate analyses of variance (PERMANOVA; Anderson, 2005), after checking the homogeneity of variance using the Cochran test. Treatment and time were used as fixed factors and *post hoc* comparison was carried out when significant (*p <* 0.05) differences were encountered. Statistical analyses were performed using the PRIMER v.6.1. program and the PERMANOVA+ add-on.

# Results

Temperature, salinity and oxygen concentrations during the experiments are reported in **Table 1**. The values of all of these variables did not change significantly among treatments, indicating that the primary differences between treatments were associated with the CO2 induced changes to the carbonate chemistry system. The injection of CO2 at high concentration caused a significant decrease of pH in all treated mesocosms, both in the water column and within the sediment under leakage-like scenarios (**Table 1**).

### Composition of the Sedimentary Organic Matter

Carbohydrate and lipid concentrations did not show significant changes either with changing CO2 levels or increasing time of exposure (**Figure 1A**). Conversely, protein concentrations showed a progressive increase with increasing time of CO2 exposure and, after 140 days, protein concentrations in all acidified sediments were significantly higher than in the control. Biopolymeric C content in the sediments during the experiment showed an increase after 140 days from 7 to 22% at 1,000 and 20,000 ppm when compared with the controls (**Figure 1A**). The concentration of chlorophyll-*a* and phaeopigments did not show significant changes either with changing CO2 supply or increasing time of exposure (**Figure 1B**).

### Extracellular Enzymatic Activity

Aminopeptidase activities decreased significantly at all CO2 levels after 14 days and were further reduced after 140 days. Such changes were more evident in systems at 5,000 and 20,000 ppm of CO2 (**Figure 2A**). The cell-specific activities generally decreased with increasing CO2 concentrations and increasing time of exposure. With the exception of CO2 treatment at 1,000 ppm, the cell-specific activities in the acidified systems were significantly lower when compared with the controls (**Figure 2B**). As a result, the turnover time of the sedimentary proteins (**Figure 2B**) increased in the acidified systems for up to ca. three times after 14 days of incubation and up to ca. six times after 140 days, with longer turnover times in the most acidified treatment (20,000 ppm).

### Prokaryotic and Viral Abundance and Production

The CO2 treatment resulted in no significant changes in the abundance of prokaryotes either in the short and long term (**Figure 3**). Similarly, viral abundance did not change significantly in the short and long term, nor between treated systems (at the different CO2 levels) and control (400 ppm).

The treatments at different CO2 concentrations resulted in a significant decrease of prokaryotic heterotrophic C production either in the short- and long-term experiments, with values 1.5– 3.0 times lower than the controls (**Figure 4A**). As a consequence, both in the short and long-term exposure, the cell-specific C production rates decreased (**Figure 4B**). After 140 days, the turnover time of the prokaryotic biomass was significantly longer in the acidified systems than in the controls, with the exception of CO2 treatment at 1,000 ppm (**Figure 4B**).

Viral production decreased significantly when compared to the control during both the short and long-term experiment, with the exception of CO2 treatment at 1,000 ppm after 14 days (**Figure 5**). Similar patterns were observed for the (VIPM), which was significantly lower than in the control (**Figure 5**).

# Discussion

In the present study, we investigated the impact of CO2-enriched seawater plumes originated by possible CO2 leakages on benthic prokaryotic metabolism and virus–host interactions according to a putative gradient of CO2 concentration (from an unaffected area to the site of epicenter leakage).

Previous studies conducted on the impact of CO2 leakage on microbial components revealed the presence of significant shifts in the abundance of different microbial components either in the


TABLE 1 | Temperature (T), salinity (S), oxygen concentration (O2), and pH in the seawater (SW) and sediment (Sed) in the different high-CO2 experimental systems (1,000–20,000 ppm) and in the controls (i.e., 400 ppm), reporting mean values and ranges.

*Significant differences between treatments and control are reported in the table with asterisks indicating* ∗∗*p < 0.01 and* ∗∗∗*p < 0.001; n.s., not statistically significant.*

short- or in the long-term exposure (Tait et al., 2014; Watanabe et al., 2014) and using different experimental approaches (Ishida et al., 2013; Blackford et al., 2014).

Conversely, in the present study prokaryotic abundance did not change significantly. Such discrepancies could be related to several factors, including the different experimental set up (e.g., different CO2 exposure levels and leakage scenarios), the effects of acidification on the physico-chemical properties of the sediments (e.g., different levels of mobilization of heavy metals or other toxic compounds, potentially impacting the biota; de Orte et al., 2014), or the changing virus–host interactions depending on the ecosystem investigated. We also found that viral abundance, as observed for the abundance of their hosts, was unaffected by CO2 leakage.

Despite the fact that we did not observe any significant effect on prokaryotic and viral abundances, we provide evidence that the high-CO2 exposure, over a period of 140 days, had a significant impact on the benthic microbial components in functional terms, with no signs of acclimatization to any of the tested CO2 leakage scenarios. In our experiments, heterotrophic C production in the sediments, when compared with control values, was significantly reduced at all CO2 concentrations. In particular, the prolonged (140 days) exposure at 2,000 ppm determined a decrease of heterotrophic C production to values close to those reported for the treatment at 20,000 ppm. Such an effect was even more evident when heterotrophic C production was normalized to the total prokaryotic abundance, suggesting that CCS-induced CO2 leakage reduced significantly the C production efficiency of heterotrophic prokaryotes even at relatively low CO2 concentrations (i.e., 2,000 ppm).

The abatement of benthic microbial metabolism was associated with the increase of replication times of prokaryotic

experimental systems (1,000–20,000 ppm) and in the controls (i.e., 400 ppm), showing mean values and relative SDs. Significant differences between treatments and controls are indicated with asterisks: ∗*p <* 0.05, ∗∗*p <* 0.01, and ∗∗∗*p <* 0.001.

cells. In addition, our results on growth rates indicated that prokaryotic turnover was affected not only by the CO2 concentrations, but also by the duration of the treatments. Since microbial metabolism plays a crucial role in the functioning of benthic ecosystems, the strong abatement of heterotrophic C production and the slowdown of prokaryotic turnover rates due to CCS-induced CO2 leakage, can have important functional implications.

Our experiment provides evidence, for the first time, that viral production and VIPM decreased significantly in the acidified sediments. It is known that there is a strong interconnection between benthic viral replication and host

metabolism (Danovaro et al., 2008; Corinaldesi et al., 2012), thus a decrease in prokaryotic metabolism is expected to determine a decrease in viral production, and consequently in prokaryotic mortality rates. We found, indeed, significant relationships between heterotrophic C production and, either viral production (*n* = 33, *r* = 0.860, *p <* 0.01), or VIPM (*n* = 33, *r* = 0.917, *p <* 0.01). A reduced top–down control exerted by viruses could thus explain why prokaryotic abundance remained largely unvaried despite the significant impact of acidification on prokaryotic metabolism.

A further confirmation of the impact of CCS-induced CO2 leakage on microbial metabolism was highlighted by the concomitant and progressive decrease of aminopeptidase activities under acidified conditions. This finding was consistent for exposures at all CO2 concentrations suggesting that the impact on organic matter degradation rates can be significant even far from the CO2 leak epicenter.

Recent findings provide evidence that aminopeptidase activity in the sediment is the main cause of virus decomposition (Dell'Anno et al., 2015). Since viral abundance depends on the balance between the rates of viral production and decay (Corinaldesi et al., 2010), the strong abatement of aminopeptidase activity in acidified sediments can contribute to explain why viral abundance remained rather constant despite the significant decrease of viral replication rates.

We report here a progressive accumulation of proteins (by 21–34%) in the acidified sediments when compared to the controls. The increased concentrations of sedimentary proteins cannot be explained by an increase of the benthic primary production. Indeed, the lack of changes in the concentrations of phytopigments (either chlorophyll-*a* and phaeopigments) among treatments, and comparing treated sediments and their controls, suggests that the high-CO2 exposure had no effects on photoautotrophs and the photosynthetic production of organic matter. The increase in sediment protein concentrations in acidified conditions can thus be explained by the decrease of proteases, which indicates the reduced ability of the benthic system to degrade detrital organic matter, as also confirmed by the increase of turnover times of sedimentary protein pools.

Viruses, by killing prokaryotic hosts, convert prokaryotic biomass into organic detritus (through the so-called *viral shunt,* Wilhelm and Suttle, 1999), which is utilized to sustain the metabolism of un-infected prokaryotes (Corinaldesi et al., 2012; Weitz et al., 2015). Our findings indicated that the organic C released by viral shunt in the acidified sediments decreased by ca 2–5 times when compared to the controls. Based on the fact that the viral shunt is known to stimulate the prokaryotic metabolism, turnover rates, and the cycling of the sedimentary organic matter (Danovaro et al., 2008), our results suggest that the concomitant reduction in the VIPM rates can exacerbate such effects, by further lowering prokaryotic metabolism under CO2-leakage scenarios. A lower viral "predatory" pressure on prokaryotes indeed could reduce the prokaryotic efficiency in the utilization of organic substrates, and contribute to the effect of accumulation of organic matter in the sediments of acidified systems.

To date the information on the effect of CO2-induced acidification on the benthic bacterial and archaeal diversity is extremely scarce (Blackford et al., 2014; Watanabe et al., 2014), and the effects on the diversity of viruses remain completely unknown (Jones et al., 2015). The lack of significant changes in the abundances of prokaryotes and viruses under different levels of CO2 exposure in our experiments could indicate high resilience of the original microbial assemblages, or compensating shifts in the benthic microbial diversity with species proliferating, which are favored in acidified conditions. Some authors have hypothesized a possible advantage for archaea (in particular for chemoautotrophs) in acidified sediments (Ishida et al., 2013; Tait et al., 2014). Although the effects of pH shifts on marine viruses are difficult to predict (see Danovaro et al., 2011), possible changes in the relative abundance of specific viral taxa and in the overall diversity of viruses can be certainly expected.

Overall, our data indicate that the impacts of potential leakage form CCS storage sites and pipelines may not be limited to the sediments only in the vicinity of the point-source leak. Indeed, the lateral dispersal of plumes of seawater enriched in CO2 over wider areas of marine sediments can significantly reduce the benthic microbial metabolism and organic matter degradation rates, VIPM, and viral shunt, with consequences, thought at local scale, on organic matter cycling and nutrient regeneration.

Our experiment was conducted on sediments collected at ca 100-m depth, but CCS can potentially be located in deepsea sediments (i.e., at depths *>* 200 m). Recent findings provided evidence that the viral control over benthic prokaryotic assemblages increases with increasing water depths and that *>*80% of prokaryotic production is abated by viral infection beneath 1,000-m depth (Danovaro et al., 2008). As a consequence it is possible to hypothesize that the development of CCSs in deep ocean sediments and porous subseabed formations (Ishida et al., 2005; House et al., 2006; Goldberg et al., 2008; Shaffer, 2010; Barry et al., 2013), and the eventual leakages from storage sites, could impact the functioning of deep-sea ecosystems by altering the virus–host interactions and the consequent *equilibria* of the benthic microbial component, with potentially important effects at local scale on biogeochemical processes.

# Acknowledgments

This research was conducted as part of the European Community's Seventh Framework Programme (FP7/2007-2013) grant agreements no. 265847 for the project Sub-seabed CO2 storage: impact on marine ecosystems (ECO2), and no 265847 for the project DEVOTES. Further support was provided by the National project EXPLODIVE (FIRB 2008, contract no. I31J10000060001, PI CC). TA was partially supported by Marie Curie Actions through the project CEFMED (project number 327488). We are grateful to Oddbjoern Petersen, Per Ivar Johannessen, Morten Schaanning and the personnel at the Marine Research Station (Solbergstrand, Norway) of the Norwegian Institute of Water Research (NIVA, Oslo, Norway) for support and advice during the ECO2 mesocosm experiments. Joana Nunes and Sarah Dashfield at Plymouth Marine Laboratory and Kai Sørensen at NIVA are also thanked for support during the organization and analyses for this work.

# References


*on Climate Change,* Core Writing Team, eds R. K. Pachauri and L. A. Meyer (Geneva: IPCC), 151.


assessment for geological carbon storage. *Int. J. Greenh. G. Con.* 38, 3–17. doi: 10.1016/j.ijggc.2014.09.007


infaunal diversity and sediment nutrient fluxes. *Mar. Ecol. Prog. Ser.* 379, 59–75. doi: 10.3354/meps07894


**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 Rastelli, Corinaldesi, Dell'Anno, Amaro, Queirós, Widdicombe and Danovaro. 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.*

# Environmental and Sanitary Conditions of Guanabara Bay, Rio de Janeiro

*Giovana O. Fistarol1,2, Felipe H. Coutinho1,3, Ana Paula B. Moreira1, Tainá Venas1, Alba Cánovas2, Sérgio E. M. de Paula Jr.2, Ricardo Coutinho4, Rodrigo L. de Moura1, Jean Louis Valentin1, Denise R. Tenenbaum1, Rodolfo Paranhos1, Rogério de A. B. do Valle2, Ana Carolina P. Vicente5, Gilberto M. Amado Filho6, Renato Crespo Pereira7, Ricardo Kruger8, Carlos E. Rezende9, Cristiane C. Thompson1, Paulo S. Salomon1,2 and Fabiano L. Thompson1,2\**

### *Edited by:*

*Maurizio Labbate, University of Technology Sydney, Australia*

### *Reviewed by:*

*Asli Aslan, Georgia Southern University, USA Thomas C. Jeffries, University of Western Sydney, Australia*

### *\*Correspondence:*

*Fabiano L. Thompson fabianothompson1@gmail.com*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 04 June 2015 Accepted: 22 October 2015 Published: 20 November 2015*

### *Citation:*

*Fistarol GO, Coutinho FH, Moreira APB, Venas T, Cánovas A, de Paula SEM Jr., Coutinho R, de Moura RL, Valentin JL, Tenenbaum DR, Paranhos R, do Valle RAB, Vicente ACP, Amado Filho GM, Pereira RC, Kruger R, Rezende CE, Thompson CC, Salomon PS and Thompson FL (2015) Environmental and Sanitary Conditions of Guanabara Bay, Rio de Janeiro. Front. Microbiol. 6:1232. doi: 10.3389/fmicb.2015.01232*

*<sup>1</sup> Institute of Biology, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil, <sup>2</sup> Laboratório de Sistemas Avançados de Gestão da Produção, COPPE, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil, <sup>3</sup> Centre for Molecular and Biomolecular Informatics, Radboud Institute for Molecular Life Sciences, Radboud University Medical Centre, Nijmegen, Netherlands, <sup>4</sup> Instituto de Estudos do Mar Almirante Paulo Moreira, Rio de Janeiro, Brazil, <sup>5</sup> Oswaldo Cruz Institute, Fundação Oswaldo Cruz, Rio de Janeiro, Brazil, <sup>6</sup> Instituto de Pesquisas Jardim Botanico do Rio de Janeiro, Rio de Janeiro, Brazil, <sup>7</sup> Laboratory of Marine Chemical Ecology, Insitute of Biology, Federal Fluminense University, Niteroi, Brazil, <sup>8</sup> Laboratory of Enzimology, Institute of Biology, University of Brasília, Brasília, Brazil, <sup>9</sup> Laboratory of Environmental Sciences (LCA-UENF), Campos, Brazil*

Guanabara Bay is the second largest bay in the coast of Brazil, with an area of 384 km2. In its surroundings live circa 16 million inhabitants, out of which 6 million live in Rio de Janeiro city, one of the largest cities of the country, and the host of the 2016 Olympic Games. Anthropogenic interference in Guanabara Bay area started early in the XVI century, but environmental impacts escalated from 1930, when this region underwent an industrialization process. Herein we present an overview of the current environmental and sanitary conditions of Guanabara Bay, a consequence of all these decades of impacts. We will focus on microbial communities, how they may affect higher trophic levels of the aquatic community and also human health. The anthropogenic impacts in the bay are flagged by heavy eutrophication and by the emergence of pathogenic microorganisms that are either carried by domestic and/or hospital waste (e.g., virus, KPC-producing bacteria, and fecal coliforms), or that proliferate in such conditions (e.g., vibrios). Antibiotic resistance genes are commonly found in metagenomes of Guanabara Bay planktonic microorganisms. Furthermore, eutrophication results in recurrent algal blooms, with signs of a shift toward flagellated, mixotrophic groups, including several potentially harmful species. A recent large-scale fish kill episode, and a long trend decrease in fish stocks also reflects the bay's degraded water quality. Although pollution of Guanabara Bay is not a recent problem, the hosting of the 2016 Olympic Games propelled the government to launch a series of plans to restore the bay's water quality. If all plans are fully implemented, the restoration of Guanabara Bay and its shores may be one of the best legacies of the Olympic Games in Rio de Janeiro.

Keywords: Guanabara Bay, anthropogenic impacts, sanitary conditions, bacteria, microalgae, pollution

# INTRODUCTION: GENERAL VIEW OF THE ISSUE AND HISTORICAL BACKGROUND

Guanabara Bay is the second largest bay on the coast of Brazil, and is situated along the northeast coast of Rio de Janeiro city, one of the most populated urban areas of the world1 (26th, according to Demographia, 2014), and one of the most important cities in Brazil. Rio de Janeiro is worldwidely known for its many natural beauties, among which are included Guanabara Bay. This environment harbored a great diversity of organisms with undeniable ecological importance, and has been an important landmark of Brazil, being one of the first settlement sites, and, from the beginning, one of country's most important economic regions. The importance of the bay, however, did not prevent the occurrence of a series of environmental impacts that are currently reflected on its present state (**Figure 1**).

Guanabara Bay drainage basin covers 4081 km2, being, at the same time an important water resource, and the receptor of most of the liquid effluents produced along its drainage basin, which covers totally or partially 16 municipalities (Belford Roxo, Cachoeira de Macacu, Duque de Caxias, Guapimirim, Itaboraí, Magé, Mesquita, Nilópolis, Niterói, Nova Iguaçu, Petrópolis, Rio Bonito, Rio de Janeiro, São Gonçalo, São João de Meriti, and Tanguá) (SEMADS, 2001). From these, the largest and most important is Rio de Janeiro (**Figure 2**).

The current environmental and sanitary conditions of Guanabara Bay results from the cumulative impacts and environmental changes that occurred since the first European settlers got established in the XVI Century. Since then, the region went through a series of economic cycles that started with timber exploitation (especially *Caesalpinia echinata*, known as Pau-Brasil), sugar-cane cultivation, mining, and then cultivation of coffee beans and orange. The impacts started to escalate around 1930, when industrialization of the area began, also bringing urbanization and a sharp population increase (Amador, 1997). Population density is currently high, with ca. 11.8 million inhabitants in the Rio de Janeiro City metropolitan area alone2 (IBGE, 2010). Moreover, this is one of the most industrialized coastal areas of Brazil, harboring more than 16000 industries, gas and oil terminals and two ports<sup>2</sup> (IBGE, 1985; Coelho, 2007). On the other hand, sewage and water treatment is still very limited. Both industrial and urban contaminants are discharged into the bay or in the sea (disposal systems in Ipanema Beach and Barra da Tijuca). Other anthropogenic impacts include destruction of habitats, poor fisheries management, sedimentation, flooding, landslides, and various public health problems.

Attention toward Guanabara Bay pollution increased together with the increase in environmental awareness around the 1990's3 (Rio de Janeiro hosted the UN conference on environment

1http://www*.*demographia*.*com/db-worldua*.*pdf

3https://sustainabledevelopment*.*un*.*org/content/documents/Agenda21*.*pdf

non-degraded areas of Guanabara Bay (f), which still have some fringing mangrove system. There are different urban landscapes surrounding the bay, such as industries (a), slums (b), metropolis (c,d). The bay has social-economic importance and it is used, among others, as harbor (e), for artisanal fisheries (f), and recreational purposes (d). Pictures by: (a) and (c): Michelle Vils; (d): Wanderson F. de Carvalho; (b, e) and (f): Giovana O. Fistarol.

and development, ECO 92, in 1992, which produced the Agenda 21), triggering some governmental efforts to recover the bay's water quality. One of the largest actions was the Program for Remediation of Guanabara Bay (PDBG) that began in 1994, as a cooperation between the Inter-American Development Bank, the government of Rio de Janeiro State, and the Japan Bank for International Cooperation (JBIC). This program planned to implement a large set of sewage treatment plants at strategic locations within the bay's drainage basin. However, several plants were not concluded and others are not fully functional (e.g., some plants are still not connected to sewage collection and disposal systems) (Coelho, 2007). Currently, efforts toward the recovery of Guanabara Bay have gained new inputs, motivated by the fact that the city of Rio de Janeiro is going to host the 2016 Olympic Games. As

<sup>2</sup>www*.*ibge*.*gov*.*br

Guanabara Bay will be the venue for various outdoor aquatic sports during the games, the government had to commit with a series of measures to ensure safe water quality levels for the athletes.

In this article we present an overview of the environmental and sanitary conditions of Guanabara Bay, and its consequences to marine life and human health. We address how the impacts have affected the bay, and then focus on the potentially pathogenic and toxic microorganisms, and on the possible implication of these organisms to higher trophic levels. We summarize the current efforts being made to restore water quality of the bay, and show the importance of implementing a continuous monitoring program to produce consistent datasets and to follow up restoration initiatives. Through the examination of available information about the impacts suffered and by comparing it to what has happened in other coastal environments around the world, we provide an assessment about the understanding of the system, and the information gaps that require further research.

### GUANABARA BAY HYDROGRAPHY

Guanabara Bay is part of a large ecosystem that forms the Guanabara Bay drainage basin. With 4081 km2, the basin is drained by 50 rivers and streams (SEMADS, 2001), six of which are responsible for 85% of the 100 m<sup>3</sup> s−<sup>1</sup> total mean annual freshwater discharge into the bay: Guapimirim (20.8%), Iguaçu (16.7%), Caceribu (13.7%), Estrela (12.7%), Meriti (12.3%), and Sarapuí (9.3%) (Coelho, 2007). The freshwater discharge ranges from 33 m<sup>3</sup> s−<sup>1</sup> in the dry austral winter to 186 m<sup>3</sup> s−<sup>1</sup> in the rainy austral summer (Kjerfve et al., 1997). Sedimentation rates are high, varying from 0.6 cm year−<sup>1</sup> near the mouth to 4.5 cm year−<sup>1</sup> in the inner part of the bay, mostly as a result of deforestation of the drainage basin and channelization of rivers (Amador, 1980). Besides, the surface area of the bay has been reduced by 10% as a result of land reclamation for construction of two commercial airports, roads, bridges, and residential areas (Amador, 1980; Coelho, 2007). Nevertheless, there are still circa 90 km<sup>2</sup> of a fringing mangrove system bordering the inner margins of the bay, which includes the Guapimirim Environmental Protection Area (Pires, 1992). Land-use around the bay changed, with the urban area increasing in 80 km2, pastures increasing in 150 km2, and forests decreasing in nearly 100 km2 (Coelho, 2007). According to the map of land-use from the Guanabara Bay Watershed Committee4 (Comite de Bacia da Baía de Guanabara) the main land uses on the watershed are high and medium density urban areas, pasture, and forest (this last one being found mostly on the mountain areas), then there are some areas used for agriculture and some fringing mangrove system. The proportion of urban and rural population around the bay ranges from a maximum of 34.7% of rural population to 100% of urban population; most municipalities, in fact, have zero to ca. 5% of rural population<sup>2</sup> (IBGE, 2000). Furthermore, a significant pattern is the uncontrolled

<sup>4</sup>http://www*.*comitebaiadeguanabara*.*org*.*br/sig-rhbg/

and disordered occupation of land by slums on the hills in Rio de Janeiro metropolitan area, taking over the remaining forest areas. These vast areas lack sewage collection and disposal systems.

The bay has a surface of 381 km2, with 22 islands (the largest one being Governador Island, with 40.8 km2) (SEMADS, 2001). It has a narrow and relatively deep entrance of 1.6 km, and measures approximately 30 km in its west to east axis, and 28 km in its north to south axis, with a mean water volume of 1.87 billion m3. Most of the bay (84%) has *<*10 m depth, with a maximum depth of 58 m on its central channel (Ruellan, 1944). From the bay's mouth to ca. 7 km inward, it has a sandy bottom that extends from the adjacent continental shelf, with some isolated sand areas at northeast and southwest of Governador Island. Otherwise, the bottom of the bay consists mostly of mud deposits (Amador, 1980).

Like in other coastal bays (e.g., Chesapeake Bay; Shi et al., 2013), the tidal regime is an important component of Guanabara Bay's water circulation (Mayr et al., 1989; Paranhos et al., 1998). Tide in the bay is mixed, but mainly semidiurnal, with a mean tidal range of circa 0.7 m (spring tidal range: 1.1 m, neap tidal range: 0.3 m), without significant spatial variance. The associated peak tidal ebb and flood volume fluxes with the ocean are in the order of 16 <sup>×</sup> <sup>10</sup><sup>3</sup> <sup>m</sup><sup>3</sup> <sup>s</sup>−1. Currents are also semi-diurnal, with flood currents (∼1.25 m s−<sup>1</sup> at the surface and <sup>∼</sup>1ms−<sup>1</sup> near the bottom) being faster than ebb currents (∼1ms−<sup>1</sup> at the surface and <sup>∼</sup>0.55 m s <sup>−</sup><sup>1</sup> near the bottom). Currents also intensify at the entrance of the bay and between the mainland and the Governador Island, because of a narrowing of the channel in these areas. Circulation inside the bay is a composition of gravitational (which presents bidirectional flux: toward the ocean in the surface, and toward the continent near the bottom) and residual tide circulation, which are affected by the prevailing wind. As a result, it takes 11.4 days to renew 50% of the water in the Bay (Kjerfve et al., 1997). This relatively short residence time is one of the main factors explaining why water quality is not worse, considering the amount of untreated sewage discharged into the bay. However, it is important to have in mind that this renewal is not the same in all parts of the bay. In fact the innermost regions of the bay, which are the ones receiving most urban sewage, have lower circulation and a longer residence time, causing accumulation of organic matter and other contaminants, making these the most polluted areas of the bay. The stronger effects of tidal currents near the mouth of coastal bays compared to inner parts of the water bodies is well known. In Chesapeake Bay, water-quality parameters such as total suspended solids and chlorophyll-*a* (Chla) concentration decrease in the lower bay region under high tidal currents, while the tidal effect are small or negligible in the middle and large part of the upper bay region (Shi et al., 2013). Likewise, for Guanabara Bay it was found significant differences in the water quality between tidal stages (the quality being better at high tide, at the maximum tide dilution). It was also found that the influence of tidal stage was twice as high as the influence of rainy/dry season, and the influence of tides on the variation of several parameters, such as salinity, nutrients, Chla, and fecal coliforms concentration (Paranhos et al., 1998).

Salinity in the bay ranges from 13 to 36 (Mayr et al., 1989). However, a decadal trend of decreasing salinity has been detected, especially in the inner parts of the bay that are under high anthropogenic influence (Paranhos et al., 1993). Due to coastal versus riverine contributions to bay waters, vertical gradients can be up to 20◦C and 18 salinity units (Paranhos et al., 1998). Vertical stratification of the water column is most pronounced in the shallow inner parts of bay, near stream discharges, and is correlated to the rainy and dry seasons: during the rainy period (October to April) temperatures are higher and salinities are lower, with the presence of thermo- and haloclines, while during the dry season (May to September) temperatures are lower, salinities are higher and there is no stratification (Paranhos and Mayr, 1993; Paranhos et al., 1993). Interestingly, the deep central channel traps cold salty water (21.1◦C with a salinity of 34.4 at 24 m, see **Table 1**) (Kjerfve et al., 1997). This hydrographic heterogeneity between different parts of the bay shows that it cannot be considered as a homogeneous environment, and sub-regional dissimilarities should always be considered in its assessments and monitoring.

TABLE 1 | Differences in water quality for different areas of Guanabara Bay, and other significant parameters: (A) near the entrance of the Bay, (B) station located between Ilha do Governador and Ilha do Fundão, on the west part of the Bay, close to the continent; (C) at the northwest part, close to the discharge of Rivers Iguaçu and Sarapui; (D) at the central channel at 24 m deep (see also Figure 2 for stations location).


<sup>∗</sup>*Annual mean calculated from data by INEA. When data was available, differences are shown for surface (s) and bottom water (b), and presented as mean* ± *SD. Data were retrieved from several sources: JICA (1994), Kjerfve et al. (1997), Paranhos et al. (1998), Paranhos et al. (2001), Marques et al. (2004), Gregoracci et al. (2012), INEA (2013; 2014a).*

# WATER QUALITY: POLLUTANTS IN THE BAY

# Types of Pollutants

Eutrophication of coastal waters, due to human activities, is a world-wide problem. It has been estimated that the export of P to the oceans has increased threefold, and N had even a higher increase in the last four decades (e.g., N increased more than 10-fold into the rivers entering the North Sea, and by six to eightfold in coastal waters of the northeastern United States generally and to Chesapeake Bay specifically) (Boynton et al., 1995; Howarth, 1998; Smil, 2001; Anderson et al., 2002). In Guanabara Bay there are several sources of contaminant's discharge into the bay, from untreated domestic effluents to industrial waste, which causes inputs of organic matter, nutrients, hydrocarbons, heavy metals, and large amounts of suspended solids. Domestic waste is responsible for discharging organic matter, and potentially pathogenic microorganisms. An average of 50.4% (*SD* = 17.9%) of the urban households are connected with sewage treatment system in the municipalities within the basin. Rio de Janeiro city has 78% of the households connected with sewage system<sup>2</sup> (IBGE, 2000). However, these data should be looked with caution, since it considers only "permanent" households, i.e., the households with a regular legal situation with the municipalities. It does not consider all the households in the slums, which houses a very large population living irregularly and marginally in terms of public services and often not included in the statistics. All the domestic sewage from households that are not connected with the sewage system are discharged directly, untreated, into the bay or in the rivers of the basin. As an example, approximately 14% of households have no bathrooms and tap water in the Duque de Caxias area (Comissão Nacional Sobre Determinantes Sociais da Saúde, 2008). The lack of sanitary conditions and sewage treatment system in the poorest areas around Guanabara Bay, e.g., Duque de Caxias area, is directly reflected on infant mortality, which reaches 23,9%, contrasting with the areas served by sewage system, where the infant mortality is 4% (Coelho, 2007; Comissão Nacional Sobre Determinantes Sociais da Saúde, 2008). According to Coelho (2007), there has always been a deficit between the produced and treated sewage. In 2005, this deficit was 13 m3s−<sup>1</sup> (treated sewage discharge was 7 m<sup>3</sup> s <sup>−</sup>1, while untreated was 20 m3 s <sup>−</sup>1). The most recent estimation is that 18 m3s−<sup>1</sup> of untreated sewage is discharged into the bay, although the data is not yet in the scientific peer reviewed literature5 . Coelho (2007) also draws attention to the problem of stormwater runoff, which has been overlooked by the authorities. Stormwater runoff is collected in the same systems that receive sewage, which are undersized to receive all these effluents. According to official information from the State Government of Rio de Janeiro, the efforts being made toward increasing the capacity of wastewater treatment plants (WWTPs), including the construction of new plants, to reach the goals set before the Olympic Committee have increased the percentage of treated sewage from 17 to 49%6 (O Globo).

The estimated daily organic load into the bay is 470 t of biological oxygen demand (BOD rages from 1-24 mg l−1), and around 150 t of industrial wastewater, which comes from the almost 17,000 industries among pharmaceutical and refineries, besides oil and gas terminals and two ports2 (IBGE, 1985; Coelho, 2007). Furthermore, it is estimated that 18 t day−<sup>1</sup> of petroleum hydrocarbons enter the bay, mostly from urban runoff (Wagener et al., 2012). The industries around the bay are responsible for 20% of the organic load input, and for most of the toxic substances discharged into the bay. It has been estimated that 10,000 t month−<sup>1</sup> of hazardous substances are produced in the Guanabara Bay's Basin (Coelho, 2007). Solid waste is also present and visible in several areas of the bay's margins (**Figure 1b**), including beaches that are commonly used for recreation. The majority of the beaches within the bay are not appropriate for swimming. Coelho (2007) estimated that 813 t day−<sup>1</sup> of solid waste reaches the bay via its effluents. Solid waste affects fisheries, navigation, leisure, tourism, the native fauna, and the landscape's aesthetical value. Furthermore, solid waste also results in slurry production, from which an unknown amounts leaches to the bay.

# Distribution of Pollutants in the Bay

The first indication of the pollution impacts in the bay is the low transparency of the water (as low as 0.7 m) (Mayr et al., 1989; Valentin et al., 1999). In general, the innermost sites are characterized by low salinity (due to fluvial and sewage input), very high nutrient concentrations, and often low levels of dissolved oxygen as a result of the eutrophication process. The outermost waters present lower nutrient concentrations and higher levels of salinity and dissolved oxygen. Between these two extremes, a gradient of decreasing pollution is created toward the Atlantic Ocean.

The gradients of water quality in the bay are mainly controlled by: (i) seasonal changes between the rainy (September to May) and dry periods (June to August), which influence the hydrography of the bay, stormwater runoff, and also the discharge of untreated wastewater into the basin's rivers; (ii) discharge of contaminants (influenced by the seasonal patterns, and a consequence of the population distributional patterns, and implementation or not of sewage treatments), and (iii) circulation and influence of marine water (which have greater influence on the south and southeast parts of the bay, and is mainly controlled by the tides) (Mayr et al., 1989; Paranhos et al., 1998; Valentin et al., 1999). Thus, the areas with the worst water quality are on the northern and northwestern parts of the bay, where two characteristics interact synergistically to aggravate the problem (it is the region receiving the greatest load of wastewater, and the one with the lowest water circulation). In the areas where dilution by seawater is higher as a consequence of tidal mixing (i.e., the central channel and the eastern part

<sup>5</sup>http://g1.globo.com/natureza/blog/nova-etica-social/post/estudo-feito-porfundacao-alema-mostra-que-baia-da-guanabara-recebe-18-mil-litros-de-esgotopor-segundo.html

<sup>6</sup>http://oglobo.globo.com/esportes/pezao-admite-que-sera-dificil-alcancar-os-80 de-despoluicao-da-baia-ate-os-jogos-de-2016-15415948

of the bay), water quality is significantly better (**Table 1** and **Figure 3**).

This pattern of water quality gradient (worse conditions on the N and NW parts, and better conditions on the E and on the central channel) is observed for most of the parameters. Values of total mean nitrogen ranged from 0.6 to 68.3 µM close to the entrance of the bay (Area A, on **Table 1**, for location see **Figure 3**), to 5–346 µM between Governador Island and Fundão Island (Area B, on **Table 1**), on the west part of the bay, close to the mainland. Total phosphorus ranged from 0.05 to 7.4 µM on site A, to 0.2–26.4 <sup>µ</sup>M on site B (Paranhos et al., 2001) (**Table 1**). The same pattern is observed for fecal coliforms (**Table 1**). A State monitoring program (Environmental Institution of Rio de Janeiro State, INEA) shows that surface waters on the central channel and on the eastern part of the bay usually had fecal coliform counts below 1,000 MPN 100 ml−<sup>1</sup> (INEA, 2013, 2014a7*,*<sup>8</sup> ), which is the limit set by Brazilian law for recreational use (it should be pointed that this limit is higher than the one adopted in Europe and North America) (Conselho Nacional do Meio Ambiente, 1992). On the northern and northwestern parts, however, counts may reach values as high as 920 × 10<sup>3</sup> MPN 100 ml−<sup>1</sup> (INEA, 2013). From the data collected by INEA, it is possible to observe that the concentration of coliforms is also correlated with seasonal changes, increasing during the rainy season. Fecal coliforms counts in the area where aquatic sports

<sup>7</sup>http://download.rj.gov.br/documentos/10112/2026700/DLFE-68424.pdf/ GraficoQualidadedasAguasdaBaiadeGuanabara.pdf

<sup>8</sup>http://www.inea.rj.gov.br/cs/groups/public/documents/document/zwew/ mdq3/∼edisp/inea0047160.pdf

of the 2016 Olympic Games will occur also extrapolate the limit of 1,000 MPN 100 ml−<sup>1</sup> mainly from September to May (INEA, 2014a). Also, enterococcus values follow the same trend (INEA, 2014a). Furthermore, the uncontrolled growth of urban areas caused destruction of habitats and deforestation, which increased the runoff, silting up, and impairing even more water circulation in the western part of the bay. Remarkably, several areas that currently present better water quality are under the influence of the cities of Niterói and São Gonçalo, which are under steep population and industrial growth, and therefore under the risk of an increase in domestic and industrial load of wastewater. However, in these areas, there is the opportunity for concomitant implementation of sanitation, preventing the degradation that happened at the other parts of the bay, and which is much more difficult and expensive to restore.

The most concerning fact is that values for fecal coliforms (for the most contaminated areas in the western part of the bay) are above the limit throughout the year (INEA, 2013). This is a reflex of the poor water quality of the rivers on the basin. The water quality index (IQANSF, which takes into account: dissolved oxygen, biochemical oxygen demand, total P, NO3, pH, turbidity, total dissolved solids, temperature, and coliforms) determined by INEA during 2014 on the basin's rivers for 55 stations, showed that, except for six stations for which data was not available, only five stations were classified as having medium IQANSF, all the others were classified as bad or very bad IQASNF9 (INEA, 2014b). Poor water quality of the basin's rivers was also found by Aguiar et al. (2011), who studied four streams in the region of São Gonçalo (on the eastern part of the bay, usually found to be less polluted). They found that these streams were hypereutrophic, with abnormally high phosphate levels (from 4.35 to 130.8 µM, resulting in very low N/P ratios, and limiting primary production), and low oxygen values (from non-detectable to 3.72 ml l<sup>−</sup>1, presenting anoxia and hypoxia, respectively). The high input of nutrients into the bay results in strong hypoxic conditions near the bottom sediment (see OD values, **Table 1**).

Most studies that investigated long-term trends (e.g., Paranhos et al., 1993, 1995; Contador and Paranhos, 1996; Mayr, 1998) found an increase in pollutants load into the bay and a corresponding decrease in water quality. These studies detected increases on Chla *a* nitrogen, phosphorus, and coliforms, as well as a decrease in dissolved O2, concentrations, especially in the northwestern part of the bay, as consequence of an increase in urban areas without adequate sanitation. All these studies, however, point out that it is difficult to make conclusive analyses due to the irregularity of the data over time, because of the lack of a continuous and integrated monitoring program. A more recent study, which used a different approach to investigate water quality trends in Guabanabra Bay, casted new light on temporal trends. Borges et al. (2009) dated the P locked in sediment layers, showing that the bay is indeed being subjected to increasing eutrophication. They found three distinct periods, the first from 1810 to 1870s, a period of low level of P concentration [with total P (TP) concentration of 195 ± 23 µg/g, and a mean inorganic P (IP) of 47 ± 34%]; the second period up to the early 1900s, with a slight P increase (TP 2915 ± 30 µg/g, mean IP 63 ± 3%); and a third period, up to the present, of strong P enrichment (TP 1196 ± 355 µg/g, mean IP that 90 ± 3%). Besides the chronic load of pollutants into the bay, occasional acute loads can cause serious damages to the biota, as was the case of the rupture of an oil-pipe in 2000, which resulted in at least a 1,300 m<sup>3</sup> of fuel spilled along the northern margin of Guanabara Bay. The spill reached the Guapimirim mangrove protection area, and affected marine life and the fisheries (Brito et al., 2009).

# POTENTIALLY PATHOGENIC AND HARMFUL MICROORGANISMS

It has been estimated a low-income population living around Guanabara Bay of ca. 4 million urban inhabitants, which corresponds to 45% of the basin's population. This population lives in areas with no sanitation, i.e., their sewage is discharged untreated either into the bay or into its tributaries. This leads to serious environmental and human health problems: the rate of water-transmitted diseases is quite high (although sanitation has improved in Brazil, still circa 65% of hospitalizations are due to water-transmitted diseases) (Brasil, 2005). Microbes entering the bay also end up affecting other aquatic species and the ecosystem (Paranhos et al., 1995, 2001; Valentin et al., 1999; Guenther and Valentin, 2008; Gregoracci et al., 2012; Coutinho et al., 2014).

### Bacteria and Archaea

Recent metagenomic analyses of the Guanabara Bay water shed new light into how anthropogenic impacts modulate microbial communities (Gregoracci et al., 2012). Overall, Guanabara Bay is a heterotrophic system with lower abundance of microbial genes in the photosynthesis subsystem than in other tropical and temperate bays around the world based on metagenomics analysis (Gregoracci et al., 2012). Less impacted sites that are more influenced by the adjacent ocean, such as the central channel, are dominated by classes *Alphaproteobacteria* and *Flavobacteria*, which commonly dwell in oligotrophic waters. Meanwhile, highly impacted sites have microbial communities enriched by *Betaproteobacteria, Gammaproteobacteria,* and *Actinobacteria* (e.g., Alteromonadales, Pseudomonadales, Enterobacteriales, Oceanospirrilales, Chromatiales, Vibrionales, and Thiotrichales), which harbor several species that thrive in nutrient rich environments (**Figure 4**). Diversity levels inversely correlate with the degree of pollution (Vieira et al., 2008; Gregoracci et al., 2012). In the highly eutrophic sites, microbes metabolize organic matter at high rates, leading to depletion of dissolved oxygen, favoring anaerobes, and facultative anaerobes (e.g., *Firmicutes*) (Vieira et al., 2007, 2008). The pollution gradient is also reflected by the metabolic profile of the microbes. Despite the massive amounts of organic matter it receives daily, this environment is phosphorus limited (as evidenced by high N:P ratios). Guanabara Bay metagenomes are enriched with genes associated with phosphorus metabolism, when compared to other bays' metagenomes. This may result from

<sup>9</sup>http://www.inea.rj.gov.br/cs/groups/public/documents/document/zwew/mdi2/<sup>∼</sup> edisp/inea0026989.pdf

high abundances of phosphorus degrading organisms dwelling in this habitat, or reflect the necessity of this microbiome to optimize its phosphorus harvesting and utilization capacities (Gregoracci et al., 2012).

The highly impacted inner region shows the highest values of bacterial abundance and high nucleic acid content cells throughout the year. Both these variables were shown to be dependent on orthophosphate and ammonia concentrations. The inner portions of the bay are especially rich in methanogenic archaea (Vieira et al., 2007; Turque et al., 2010), a consequence of eutrophication, as methanogenesis occurs in anoxic areas with high concentrations of organic matter (Peng et al., 2008; Angel et al., 2011). Sponge associated archaea, as well as the diversity of ammonia oxidizing genes within them, were shown to adapt according to the gradient of anthropogenic impact to which the sponges are submitted (Turque et al., 2010). Thus, shifts in the microbial community due to eutrophication are likely to play an important role in the health and survivability of their eukaryotic hosts, which will also suffer the impacts brought by sewage contamination. Altogether, these results point to nutrient enrichment leading to a microbial community that has higher concentrations of microbial cells and a more active metabolism, promoting the dissemination of copiotrophic bacteria, such as vibrio. The number of culturable vibrios correlates positively with nutrient (e.g., orthophosphate and nitrite) (Gregoracci et al., 2012).

Vibrios are indicators of water quality because they respond promptly and positively to nutrient enrichment (**Figure 5**) (Gregoracci et al., 2012). Gregoracci et al. (2012) found significant correlation between vibrio abundances and nitrogen and phosphorus concentrations (see **Figure 5**). There is, clearly, a higher abundance of vibrio on the more polluted "site B", while the concentration of vibrios, though also correlated with nutrient concentration, is lower in the less polluted sites "A" and "under Rio-Niteroi Bridge" (**Figure 5**), which have a stronger influence of oceanic waters. Moreover vibrios are well known for harboring causative agents of a variety of zoonoses both to captive and wild aquatic animals, with risk of transmission to humans (Thompson et al., 2004). Pathogenic vibrio species, both to animals and humans, have been isolated from Guanabara Bay waters (Gregoracci et al., 2012.). Two pathogenic vibrio species previously isolated from Guanabara Bay (*Vibrio alginolyticus* and *V. parahaemolyticus*) and also the infective *Photobacterium damselae, V. harveyi*, *V. rotiferianus,* and *V. xuii* were retrieved from fishes (*Brevoortia aurea*, Clupeidae, known as Brazilian menhaden) during a mass mortality event in 2014. Mortality might have been multifactorial since toxic microalgae were also detected during the event, see section about microalgae). In any case, the presence of such pathogenic species are a clear threat to ecosystem and human health.

### Opportunistic Pathogens and Drug Resistance

Water pollution favors species adapted to nutrient rich habitats, among which are included many potentially pathogenic bacterial lineages that are capable of infecting humans and other life forms (Smith and Schindler, 2009). Opportunistic pathogens can either thrive as free living or take up an infectious lifestyle, when in contact with a suitable host. Many genera that include potentially pathogenic bacteria were detected in Guanabara Bay, e.g., V*ibrio, Klebsiella, Pseudomonas, Clostridium,* and *Bacillus* (Vieira et al., 2008; Coutinho et al., 2014). When considering only the culturable fraction of Guanabara Bay's microbiome, members of the orders Vibrionales and Clostridiales are extremely abundant, accounting for more than 50% of the antibiotic resistant culturable organisms (Coutinho et al., 2014). Exploring the diversity of potentially pathogenic organisms occurring at Guanabara Bay, and how it is affected by anthropogenic impacts, is fundamental to assess the potential risks to the local population.

The risk posed by these organisms is enhanced by increased resistance of some bacteria to antibiotics. The use and misuse of antibiotics throughout the decades has created a strong selective pressure that favors antibiotic resistant strains (Martínez, 2008). Sewage is one of the forms by which these organisms can spread from the human organism to the aquatic environment, and wastewater discharges have been shown to contribute to the dissemination of antibiotic resistant organisms and resistance genes (Kristiansson et al., 2011; Czekalski et al., 2012; Tacão et al., 2012). Highly impacted sites of Guanabara Bay (between Fundão Island and the mainland, and just north of Fundão Island, see **Figure 3**) were shown to be rich in antibioticresistant microorganisms. Several strains are classified as superresistant, due to their capacity of tolerating drugs in doses up to 600 times higher than the clinical levels (Coutinho et al., 2014). Among these resistant and super-resistant lineages, several species of human pathogens were identified (e.g., *Vibrio cholerae, Klebsiella pneumonia,* and *Shigella* sp.), some of which are classified also as multi-resistant, capable of tolerating three or more different classes of antibiotics. The presence of resistant

nutrient concentration, the figure shows a clear separation between the less impacted sites "A" and "under Rio-Niteroi Bridge", and the more impacted site "B".

bacteria (KPC-producing bacteria) was recently reported in Guanabara Bay, making the media headlines (e.g., http://g1*.* globo*.*com). The occurrence of resistant bacteria was registered in two sites in a river that discharges into the bay, and in one site in the bay, at Flamengo Beach, which is located close to where the sailing Olympic competitions will occur. Taxa that are typical of the mammalian gut (e.g., *Enterobacteria, Firmicutes,* and *Bacteroidetes*) showing drug resistance were also identified. Altogether these facts indicate that untreated domestic and hospital waste/sewage are relevant sources of microbial contamination to the bay (Coutinho et al., 2014). Prevention of further dissemination of these organisms can only be achieved by providing adequate sewage treatment.

Contamination of Guanabara Bay's waters by potential human pathogens has drawn attention from the international media, which has raised concern about the possibility of athletes participating in aquatic sports during the 2016 Olympic Games to fall ill because of the presence of the pathogens, especially viruses (Associated Press10; Denver Post11; NBC News12; The Guardian13). An independent investigation carried by the Associated Press revealed the presence of high levels of viruses (such as adenovirus, rotavirus, and enterovirus) and bacteria. Although the investigation was carried out by a researcher from the FEEVALE University, southern Brazil, the location, protocols, and exact data of the investigation are not available. There is no recent peer reviewed publication showing an investigation about virus presence in Guanabara Bay. The only report found dates back to 1975 (Homma et al., 1975). They investigated the presence of virus in several sites in the bay and found concentrations ranging from 19 to 1800 particles l<sup>−</sup>1, and transitory cytopathic effects on monkey kidney cells. They could identify the type of virus only for four samples, with all belong to the type ECHO virus (enteric cytopathic human orphan). The presence of pathogenic microorganisms in Guanabara Bay is undeniable. Among these microorganisms, the abundance, diversity and dominance of bacteria has been more thoroughly studied (Paranhos et al., 1998, 2001, Guenther and Valentin, 2008; Gregoracci et al., 2012; Coutinho et al., 2014). On the other hand, the lack of information about virus clearly indicates the necessity of more complete and systematic studies. Only with this kind of data it is possible to infer the treats to human health and determine efficient action plans.

### Microalgae and Harmful Algal Blooms

Environmental impacts in natural water bodies can be detected through changes in species composition, biomass, and in the trophic structure of natural communities. In these sense, aquatic microorganisms are reliable and efficient biosensors, as they have fast growth rates, and reflect environmental changes rapidly. Microalgae (O2-evolving photosynthetic protists and cyanobacteria) can be used to access environmental impacts through changes in Chla concentration (which is often used to indicate eutrophication, since levels of Chla present a good correlation with increase in nutrient loads), by the decrease

<sup>10</sup>http://bigstory.ap.org/article/d92f6af5121f49d982601a657d745e95/apinvestigation-rios-olympic-water-rife-sewage-virus

<sup>11</sup>http://www.denverpost.com/sports/ci\_28558663/ap-investigation-dirty-riowater-threat-at-2016?source = skipframe-feedspot.com

<sup>12</sup>http://www.nbcnews.com/health/health-care/dirty-water-blamed-sick-rowersbrazil-olympic-trial-n407456

<sup>13</sup>http://www.theguardian.com/sport/2015/jul/31/alistair-brownlee-calm-despite despite-rio-2016-pollution-fears-at-olympics

in species number and diversity, and changes in species functional roles, e.g., shifts toward heterotrophy. Furthermore, some microalgae species may form harmful algal blooms (HABs), which cause impacts on the biota due to oxygen depletion or to toxin production. In the latter case, the effects can reach human populations by the consumption of contaminated seafood and inhalation of seawater aerosols (Hallegraeff, 1995).

Besides the pollution indicator parameters mentioned earlier, we also find high values of Chla in Guanabara Bay. Chla concentrations can be as high as 483 mg m−<sup>3</sup> (Paranhos et al., 2001), indicating hypereutrophic conditions (Rast et al., 1989). During 2014, values reaching up to 700 mg m−<sup>3</sup> were recorded during intense algal blooms in some locations. However, because Guanabara Bay is strongly influenced by tides, it is a dynamic environment that presents a large spatial and temporal variability. In general, mean Chla values across the bay detected during decadal monitoring programs ranged from 2.5 mg m−<sup>3</sup> in the less polluted areas, to *>*100 mg m−<sup>3</sup> in its inner parts (Mayr et al., 1989; Paranhos et al., 1998, 2001; Marques et al., 2004; Santos et al., 2007) (see **Table 1** and **Figure 3**). These values of Chla give a good representation of the spatial variability of the system, and demonstrate that they can be used as indicators of water quality. Values of primary productivity (PP) also show the system's dynamics and influence of tides: PP ranged from 23 µgC l <sup>−</sup><sup>1</sup> h−<sup>1</sup> during flood tide to 582 µgC l−<sup>1</sup> h−<sup>1</sup> during ebb tide (mean = 169 ± 152 µgC l−<sup>1</sup> h−1) in the central channel ca. 7 km inward from the bay's entrance (Guenther and Valentin, 2008). A "positive" effect of the eutrophication of the bay is that, because of its high PP, especially in the inner sites, coupled with the intense radiation and thermal stratification during summer, the bay plays a role as a sink of atmospheric CO2. The annual CO2 sink for the inner parts was calculated as <sup>−</sup>19.6 mol Cm<sup>2</sup> yr−1, which matches the values of C found in the sediments) (Cotovicz et al., 2015).

The high values of Chla also highlight potential impacts on the aquatic community since they are usually connected to the dominance of one or few phytoplankton species (i.e., blooms of microalgae), and low species diversity. The consequences of lowered diversity are reflected on higher levels of the trophic web, especially if it is accompanied by changes in species functional role, and with cases of dominance by harmful microalgae species (Fistarol et al., 2003, 2004; Suikkanen et al., 2005; Granéli et al., 2008). Both these facts have been observed in Guanabara Bay. Heterotrophic/mixotrophic species have always been present in the bay. For example, about 25% of the dinoflagellates identified until 2010 were heterotrophic, several of them belonging to the genus *Protoperidinium* (Villac and Tenenbaum, 2010). However, based on the literature, it seems that there is an ongoing shift toward an increase in the frequency of dominance by heterotrophic organisms and toward smaller phytoplankton species. Valentin et al. (1999) reported the dominance of nanoplankton (flagellates and diatoms smaller than 20 µm) and filamentous cyanobacteria. Such dominance by nanoplankton was also reported by Santos et al. (2007), who found densities of 10<sup>8</sup> cells l−<sup>1</sup> of nanoplankton species, representing *>*57% of the phytoplankton community. Recent observations based on a monthly monitoring program (PELD Guanabara) carried out by our research group shows a somewhat permanent bloom of *Tetraselmis* spp. (Prasinophyceae). However, because of the lack of a long-term, uninterrupted monitoring in the bay, if this shift is occurring remains to be confirmed, as also remarked by Santos et al. (2007). The shift from an autotrophic to a heterotrophic community is an indication of water quality deterioration, and the presence of high loads of dissolved and particulate organic matter (Granéli et al., 1999). At the clearer waters at the entrance of the bay, phytoplankton community is dominated by autotrophs (except at depth, where the light decreases) (Guenther et al., 2012). On the other hand, Valentin et al. (1999) draw attention to the increase on phytoplankton biomass in the central channel, which may indicate a decreasing capacity of pollutants dilution by the incoming seawater. The pattern of shifting from phototrophic to smaller heterotrophic organisms linked to water quality deterioration was observed, e.g., in the Black Sea. Heterotrophic organisms may thrive on the increase in organic nutrients (Bodeanu and Ruta, 1998).

Some of the opportunistic microalgal species blooming in the bay may cause impacts on higher trophic levels due to the production of toxins or to oxygen depletion when blooms decline. The dinoflagellate *Scrippsiella trochoidea,* which has been reported in the bay since 1914, may cause fish kills due to anoxia, and is frequently found in densities as high as 10<sup>6</sup> cell l−<sup>1</sup> (Villac and Tenenbaum, 2010). Filamentous cyanobacteria (that may reach summer concentrations of 10<sup>8</sup> l <sup>−</sup>1), and dinoflagellates from the genus *Prorocentrum* (Santos et al., 2007; Villac and Tenenbaum, 2010) are other potential toxin producers. The raphydophycean *Chattonella* spp. are fish-killing species that also have always caused problems in the bay, and it was detected by our group in concentrations of 3.54 · <sup>10</sup><sup>6</sup> cells l−<sup>1</sup> concomitantly to the aforementioned massive fish kill event that took place in October 2014. The presence of the domoic acid (neurotoxin) producing diatom *Pseudo-nitzschia* has also been detected, though in low abundances.

The relationship between increase in HABs and in eutrophication has concerned managers around the world, which try to implement measures to prevent factors that trigger these events (Anderson et al., 2002). As there are numerous examples from different regions around the world (e.g., Chesapeake Bay, U.S Coastal areas, Seto Inland Sea, Black Sea, Chinese coastal waters) showing a correlation between increase in eutrophication and frequency and magnitude of algal blooms, there are also many examples demonstrating that the regions that have implemented nutrient load controls have also witnessed reductions in phytoplankton biomass and harmful bloom events (Anderson et al., 2002). Some examples of regions where there was a decrease in HAB in connection to controlled reduction in nutrient inputs include Lake Washington, Seto Inland Sea, and the Black Sea (Edmondson, 1970; Bodeanu, 1993; Bodeanu and Ruta, 1998; Anderson et al., 2002; Imai et al., 2006; Nishikawa et al., 2014). High phytoplankton biomass cannot be sustained under low nutrient concentration, and there is consistent evidence showing that decrease in nutrient concentration will result in a decrease in biomass, and on the frequency of algal blooms. Edmondson (1970), Imai et al. (2006), and Davidson et al. (2014) showed significant correlation between decrease in Chl*a* concentrations and HABs, following actions to reduce nutrient concentrations in marine coastal ecosystems. These evidences demonstrate that high phytoplankton biomass and HABs are consequences of eutrophication that can be reversed if appropriate measures are taken to decrease nutrient inputs into water bodies. The presence of toxic and potentially harmful phytoplankton species in Guanabara Bay is a problem that may affect not only higher trophic levels of the aquatic food web, but also human populations. Thus, implementation of a continuous monitoring program is essential to provide information about the causes of these events. Such monitoring would allow the implementation of warning systems to the public during bloom events, and, following remediating strategies, to monitor the efficacy of the measures, besides giving important scientific information about the relationship of algal blooms and eutrophication.

Despite the polluted conditions of Guanabara Bay, microalgae richness is considerably high in its waters, with as many as 323 species described: 202 diatoms, 104 dinoflagellates, 9 cyanobacteria, 5 euglenophyceans, 1 chlorophycean, 1 prasinophycean, and 1 silicoflagellate (Villac and Tenenbaum, 2010). However, this high species richness does not reflect the diversity of the bay, since diversity is influenced by the relative abundance of species (i.e., if the environment is dominated by one or few species). Therefore, diversity is a better indicator of the impact level in the system than species richness. Nevertheless, the high number of microalgae species recorded demonstrates the great ecological relevance of this system and the importance of recovering it. The low diversity index found for the western margin of the bay (Shannon-Weaver index: 0.03-1.80 bits.cell<sup>−</sup>1, 50% *<* 1.00 bits.cell<sup>−</sup>1) indicates the instability of the system in that area, compared to the central channel that has a higher diversity index (Shannon-Weaner index: 1.30 -3.30 bits.cell<sup>−</sup>1, 35% *>* 1.00 bits.cell−1), and, consequently, a higher capacity to absorb disturbances (Villac and Tenenbaum, 2010).

# CONSEQUENCES OF GUANABARA BAY POLLUTION TO HIGHER TROPHIC LEVELS AND TO HUMAN POPULATIONS

The impacted state of Guanabara Bay, especially at its most degraded areas, can be observed simply by looking at its shores and the color or transparence of the water. Additionally, severe effects of pollution can impact all levels of the aquatic trophic web. The microbial community of Guanabara Bay, for example, adapts to the pollution gradient, by altering its species composition and metabolic activities according to the local conditions. These changes, in turn, are reflected into all the upper trophic levels. Protozooplankton abundance is higher in the inner hypereutrophic parts of the bay (103–105 cell l−1) with dominance of small heterotrophic dinoflagellates and naked ciliates, while large marine dinoflagellates were only found at the entrance of the bay (Gomes et al., 2007). This confirms the trend of a shift from autotrophic to heterotrophic microorganisms in the bay. Paranhos et al. (2001) suggested that this higher abundance of protozooplankton in the inner sites could exert a top-down control on bacterial population in these areas. Zooplankton also responds promptly to pollution, and a decrease in copepods, appendicularia, cladocerans, and chaetognats numbers has been observed, concomitantly with the disappearance of siphonophores and thaliaceans in the most polluted areas (Valentin et al., 1999). Moreover, higher density of fish eggs and larvae were recorded in the less polluted areas at the entrance of the bay and in the central channel (Valentin et al., 1999).

A singular fact observed during early studies in Guanabara Bay is the different distribution of two dominant copepod species, *Acartia tonsa* and *Paracalanus parvus. P. parvus*, usually occurring in deeper and colder waters, almost disappeared from the inner part of the bay and does not seem to be adapted to highly polluted environments, whereas *A. tonsa*, found in estuarine waters worldwide, persists there in reasonably high proportions and appears capable of adapting and surviving in the unfavorable conditions of the polluted areas of the Guanabara Bay. Following Gomes et al. (2004) different ecological requirements of these two species rule their vertical migration behavior through the stratified water column at the entrance of the bay, with warm, low-salinity water from the inner bay at the surface and cold, high-salinity deep-ocean water below. *A. tonsa* maintains itself in the surface water layer during the night and does not seem to be affected by the presence of a sharp thermocline, whereas *P. parvus* shows a vertical migration limited to deeper waters, below the thermocline, a behavior that helps it to avoid being carried into the inner, polluted part of the bay (Gomes et al., 2004). Another important species of mesozooplankton is the cladocera *Penilia avirostris*. This species is an important component of the microbial loop between bacterioplankton and higher consumers because of its predation on bacterivorous microflagellates (Turner et al., 1988). This species decreases sharply in abundance from the entrance to the inner bay (Valentin et al., 1999). It is an interesting case of adaptation to the changing environmental conditions since this species has a complex reproductive strategy with the shift from asexual (parthenogenetic) to sexual (gamogenetic) phase under unfavorable environmental conditions, like the combined effects of pollution, low salinities and oxygen, leading to changes in trophic web structure (Valentin and Marazzo, 2003).

At the most degraded parts of the inner bay, the food web is highly compromised, fisheries yield have declined to 10% of the levels of three decades ago, mangrove areas have been reduced to 50% of their original size, and many of the beaches are not recommended for recreation (swiming) due to pollution (Marques et al., 2004; Coelho, 2007). Despite the accentuated degradation, there are still a number of families (around 6,000) that depend on the bay's fisheries for their income (being sardines one of the most important catches), and also a number of people that practice recreational fishing (Marques et al., 2004). While fish stocks are stagnated (FAO, 2012), fishing effort has increased, producing a false idea of increase in fisheries yields (FIPERJ, 2011). Moreover, many species of cetaceans that used to be observed in the bay (e.g., *Balaenoptera edeni, Tursiops trucatus, Steno bredanensis*), are now absent. Only the Guiana dolphin (*Sotalia guianensis*) is still found in the bay, usually at the more clear waters of the central channel and around the mangrove's protected areas (Bisi et al., 2013).

A clear example of impacts with consequences to human population was the aforementioned massive fish kill registered on October/November 2014. Although the fish species that died (*Brevoortia aurea*, Clupeidae, known as Brazilian menhaden) was not of significant commercial value, the intensity of the event, when at least 80 tons of dead fish were removed from the bay, caught attention of the media and raised concern from the government and the public. The causes of this particular massive fish kill could not be precisely determined, mostly due to the patchy nature of the phenomenon and lack of adequate monitoring previous to and during the event. Nevertheless, a thorough analysis of the microbiota, both free living in the water and associated to the fish, collected during the event point to some possible explanations. Firstly, in November, 2014, in one site where several fishes were found dead and moribund, we detected an algal bloom that turned the water yellowish and was composed almost exclusively by a gymnodinoid dinoflagellate of ca. 10 µm resembling the fishkiller genus *Karlodinium*, at densities higher than 60 <sup>×</sup> 106 cells l−1. In another less intense fish kill event registered on February, 2015, once again an algal bloom was detected during our routine monitoring in the bay, formed by the ichtiotoxic raphidophyte *Chattonella* sp. in densities up to 3.54 × 106 cells l−1. These two bloom-forming microalgal species are known fish-killer species in coastal waters worldwide (Imai and Yamaguchi, 2012; Place et al., 2012; Nishikawa et al., 2014). Their occurrence in the Guanabara Bay waters indicates a potential threat to fish stocks and a possible explanation to the recent fish kill events observed in the area. This is especially significant considering that the species that died was described to feed on phytoplankton, contrarily to most sardines that feed primarily on zooplankton (Sanchez, 1989). Besides potentially toxic microalgae in the water, several pathogenic vibrio species were isolated from the gills and kidney of moribund and dead fish we collected during the event. Sequencing of ca. 500 base pairs of the pyrH gene from approximately 50 bacterial isolates revealed the presence of *Vibrio harveyi*, *V. parahaemolyticus,* and *V. alginolyticus*, besides *Photobacterium damselae*. These are recognized pathogens of marine organisms, including fish, crustaceans and molluscs (Austin, 2010; Gauthier, 2015). Thus, two groups of microorganisms (toxic algae and pathogenic bacteria) with potential to harm and kill fish occur in the bay. Their role in massive fish-kill events (and also if we want to prevent future events) could only have been precisely assessed if a proper monitoring program was implemented. The duration and intensity of these fish-killing events was relatively new in Guanabara Bay. Thus, they should be taken as an alert and an opportunity to establish an adequate monitoring program. The presence of harmful algal species in Guanabara Bay had been registered (Santos et al., 2007; Villac and Tenenbaum, 2010), and the authorities are aware of that. However, it has not trigged yet a desirable response from them, as it has occurred in other coastal environments around the world that had this problem. In Seto Inland Sea, a region that played important role to the economic growth of Japan in the 1960's, incidents of HABs had dramatically increased in connection to increase in eutrophication in the 1960s and 1970s, until a bloom of the toxic raphidophyte *Chattonella antiqua* in 1972 caused a large fish-killing event which resulted in large economic losses (7.1 billion yen). This bloom trigged the enactment of the "Law Concerning Special Measures for Conservation of the Environemnt of the Seto Inland Sea" and the "Total Pollutant Load Control" (TPLC). This prompt response from authorities decreased the nutrient load and, consequently, the number of HABs, which is maintained under control up to this date (Imai et al., 2006). The Seto Inland Sea case is a good example that, with proper regulation, economic and industrial development can occur without concomitant impact on the environment. Industrial production on that region continues to grow up to this date, but the TPLC guarantees that there is no increase in chemical oxygen demand (COD) on the water, consequently preventing the negative effects of eutrophication (Anderson et al., 2002; Imai et al., 2006).

# CURRENT ACTIONS FOR REMEDIATION OF THE BAY

In 2011, the government of the State of Rio de Janeiro enacted the "Sanitation Pact"14 (Pacto pelo Saneamento), based on the Federal Law 11.445 (January, 2007) that established the national guidelines for sanitation. With this pact, the government intended to expand the access to sanitation for the population of Rio de Janeiro State, which encompassed (i) supply of potable water, (ii) wastewater treatment, (iii) urban cleaning services and solid waste management, and (iv) stormwater runoff drainage system. The pact is divided in three programs (**Figure 6**): (a) "Zero Open Dump" (Lixão Zero), which aims to replace openair landfills by sanitary landfills or solid waste treatment plants; (b) "Rio+Clean" (Rio+Limpo), which aims to collect and treat 80% of the wastewater/sewage of Rio de Janeiro State until 2018; and (c) "Clean Guanabara Plan"(Plano Guanabara Limpa – PGL). This last one, PGL, aims to restore Guanabara Bay having as main goal to reduce in 80% the amount of sewage discharged into the bay. This goal was part of the commitments made before the International Olympic Committee (IOC) as part of Rio's candidacy proposals as host for the 2016 games, which promised to properly treat 80% of the sewage entering the Guanabara Bay until 2016.

The PGL includes a series of actions (**Figure 6**), the main one being the "Program for Sanitation of the municipalities on the border of Guanabara Bay (Programa de Saneamento dos Municipios do Entorno da Baía de Guanabara – PSAM). PSAM received resources (1.5 billion Brazilian Reais, ca. 0.5 billion USD) from the Interamerican Development Bank (BID), and from the Government of Rio de Janeiro State, and intend to use 80% of this resource to implement sewage catchment systems and WWTPs on the municipalities around the bay, reducing the organic load discharged into the bay (SEA, 2011).15 Although

<sup>14</sup>http://www.rj.gov.br/web/sea/exibeconteudo?article-id=330838

<sup>15</sup>http://www.rj.gov.br/web/sea/exibeconteudo?article-id = 1055505

it is not clear if all targets will be reached, some actions are being implemented, such as the expansion of the WWTP Alegria, which serves some of the regions with highest sewage load (such as Rio de Janeiro City Center, and the neighborhoods of Caju, Madureira, Maracanã, Grajaú, Vila Isabel, and others). The implementation of secondary treatment on Alegria WWTP already upgraded from 40 to 95% the reduction on the organic load of the treated sewage16 (Infraestrutura Urbana, 2011). The expectation for Rio de Janeiro State is to increase from the 2000 L s <sup>−</sup><sup>1</sup> (in 2006) to 13000 L s−<sup>1</sup> of treated sewage (currently the state treats 4500 L s−1, which corresponds to a population of 2 million inhabitants). The total treated sewage already increased from 17 to 49%. To reach these marks, other WWTPs are also being expanded (e.g., WWTP Sarapui, and WWTP São Gonçalo) (Infraestrutura Urbana, 2011). However, the Government of Rio de Janeiro State already recognized that these targets may not be reached until 2016, and may demand more time for their conclusions (O Globo17).

The goals of the other actions of PGL follow the premises from the Sanitation Pact and include (**Figure 5**): (i) amplification of the WWTPs, (ii) replacement of open-air landfills by sanitary landfills or solid waste treatment plants, (iii) removal of solid waste from the bay, (iv) recovering the area between Fundão Island and the mainland, (v) recovering the beaches around the bay, (vi) reforestation of the bay's margins, and (vii) depollution of the bay waters (this is the continuation of the program PDBG started in 1994 (see introduction), which did not reach its goals, but has now been reactivated). The Guanabara bay may also play an important recreational role. In the 1970s, several beaches in the inner bay (e.g., Ramos and Caju) were massively used for recreation, particularly by the population living in the north area of Rio de Janeiro city.

Microorganism can be studied to assess the effectiveness of the clean-up strategies. If the targets of the PGL are reached and the WWTP are upgrade and new ones are built, the input of organic nutrient into the bay will decrease considerably. Consequently, a gradual change in the microbial community toward: (i) an increase in *Alphaproteobacteria* and *Flavobacteria* is foreseen, with also (ii) a shift back to phototrophic dominated phytoplankton community (as previous conditions, and as it was observed in the Black Sea, see Bodeanu, 1993), and (iii) a decrease in Chla concentration and on the frequency of algal bloom (Edmondson, 1970; Anderson et al., 2002; Imai et al., 2006). A decrease in the frequency of algal blooms, including HABs, is directly correlated to a decrease in nutrient concentration (mostly P in freshwater environments, and N in marine and estuarine waters) (Anderson et al., 2002), as demonstrated by Nishikawa et al. (2014), Imai et al. (2006), and by Edmondson (1970). Therefore, the completion of the WWTP and the other measures planned to remediate the current conditions of Guanabara Bay would produce a chain of positive results by reducing the nutrient load, the input of pathogens, including human, and decreasing the algal bloom and promoting phytoplankton diversity, which would have effects on the whole trophic web. Remediating strategies applied in other regions (e.g., Chesapeake Bay, Seto Inland Sea, Sydney Harbor) demonstrate the importance of a continuous monitoring program, which allows following the efficacy of the measures applied. Furthermore, it represents an excellent opportunity to study long-term shifts in the microbial community. Although remediating plans are costly, they bring undeniable benefits to the society, from decreasing health risks to providing clean recreational site, and increasing landscape's aesthetical value. In addition, it can generate monetary benefits such as increase in fisheries, tourism, recreational activities,

<sup>16</sup>http://infraestruturaurbana.pini.com.br/solucoes-tecnicas/4/artigo220154- 2.aspx

<sup>17</sup>http://oglobo.globo.com/rio/coi-diz-esperar-que-compromisso-de-despoluirbaia-de-guanabara-seja-cumprido-ate-2016-15438299

and, consequently, economic sectors that provide services to supply/maintain these activities. It also increases the value of waterfront land. Furthermore, when society perceives water quality improvement, they increase their willingness to pay for its improvement, as it was observed in Chesapeake Bay (Bockstael et al., 1989; Leggett and Bockstael, 2000).

# PERSPECTIVES

Pollution in Guanabara Bay reflects the neighboring social conditions of this vast and densely populated hydrographic basin. If we are to recover the bay, besides remediating environmental degradation and treating contamination sources, it is important to implement social projects that address the causes of the problem. This is one of the most densely populated areas in the world, which includes numerous slums that do not have basic sanitation systems. The projects being implemented for remediation of the bay, tackle exactly these points, however, remediation plans on this scale take years to produce the expected effects, and should not, therefore, be dependent on political mandates. Even if these projects are implemented by federal or state government agencies, they would need to be carried and followed by academic research institutions, which are independent, and not limited by mandate's time, and also organizations representing the population, which are the main stakeholders interested in the recovery of the bay. The experiences from the recovery of other bays around the world should be used as examples of what works and what is necessary for an efficient recovery strategy. Chesapeake Bay, for example, has been carrying a program to recover the watershed's water-quality for nearly three decades18*,*<sup>19</sup> (the Chesapeake Bay Program – CBP). The success of the program is based on some specific characteristics: (i) the involvement of all states encompassing the watershed, establishing a total maximum daily load (TMDL) of nutrients into the watershed (which is identified by a watershed implementation plan); (ii) setting 2 years milestones (which allows quick rectifications of plan); (iii) implementing a track and assess progress system, and (iv) federal intervention (Federal Actions) if milestones are not achieved. An important strategy used to guarantee the success of the program is the transparence of the results. Therefore, part of the resources of the program are allocated to provide mechanism to inform the population of the progress, such as the Chesapeake Tracking and Accounting System (BayTAS), an interactive on-line tool that the public uses to track progress of the implementation of TMDL. Only if the stakeholders feel involved and see the results, they will contribute to the recovery. The experience from the recovery of other coastal areas around the world also reveal that remediation programs are not an easy task and some of the results may take time to produce the desired effects. Sydney Harbor, in Nova Scotia, was one of Canada's most contaminated sites. In 2004 the government of Canada and Nova Scotia committed to remediate the area, which began in 2009. The remediating program was followed by a monitoring program to check the recovery of the system. This monitoring revealed that, mostly, the remediation program reduced the concentration of contaminates (e.g., of polycyclic aromatic hydrocarbons – PAHs, and, to some extent, polychlorinated biphenyls – PCBs); however, metals (As, Cd, Cu, Hg, Pb, Zn) showed little spatial-temporal variability (Walker et al., 2013a,b; Walker and MacAskill, 2014). In the Guanabara Bay water circulation is high, pointing out to an optimistic scenario in face of environmental sanitation.

Monitoring programs provide data and scientific publications for informed decision-making, for evaluation of water quality conditions, and to build predictive models (which are used for implementation of management strategies). Monitoring programs need to be implemented and supported by the State. A good example of long-term monitoring is the one carried in the Seto Inland Sea by the Fisheries Technology Institute of Hyogo Prefecture (Imai et al., 2006; Nishikawa et al., 2014). A monthly monitoring sampling is made since 1973, and the data collected allowed to make, for example, correlations between the phytoplankton abundance and nutrient levels, and then, to predict the occurrence of algal blooms based on the nutrient availability (Imai et al., 2006; Nishikawa et al., 2014). To ensure its continuity, the monitoring is carried by a government-funded institute, and the data is publicly available.

# CONCLUSION

Guanabara Bay has an environmental, social, and economic importance for the region around its basin. Its current state endangers wildlife and poses risks to human populations that use this water resource. As highlighted in this review, Guanabara Bay waters harbors opportunistic pathogenic microbes capable of harming humans and several other life forms. The population of Rio de Janeiro is often in contact with the bay's water, be it directly (e.g., by bathing in its waters or in the nearby oceanic beaches) or indirectly (e.g., through the consumption of seafood). Thus, the restoration of Guanabara Bay is not only of ecological, socialcultural and esthetic relevance, but is also a public health issue. Despite enduring decades of severe environmental degradation, Guanabara Bay still present some resilience and, because of its hydrodynamic characteristics, the capacity to recover from these impacts, providing that sanitary measures to recover the bay and prevent further degradation are taken. The current sanitary conditions of the bay are not worse than those of other heavily contaminated bays around the world before they had been successfully recovered. It has been assumed that it may be possible to restore the bay's water quality if ca. 80% of all domestic and industrial sewage are appropriately treated. Therefore, a remediating plan implemented by the government, with the participation of all stakeholders, may produce the desired effects in medium and long run.

An important feature of the bay, provided by the researches that have been carried in the bay, is its heterogeneity, with the inner parts having lower circulation, higher contaminates discharge, and consequently, worse water quality. A continuous

<sup>18</sup>http://www*.*epa*.*gov/reg3wapd/pdf/pdfchesbay/TMDLPoster7-13*.*pdf 19http://www2*.*epa*.*gov/chesapeake-bay-tmdl

monitoring program will be needed to evaluate the restoration plan of Guanabara bay, and the results should be available to the population. If all ongoing plans are implemented, the restoration of Guanabara Bay and its shores may be one of the best legacies of the Olympic Games in Rio de Janeiro.

### AUTHOR CONTRIBUTIONS

GF, CT, PS, and FT conceived and designed the article. AC, SP Jr. RV made all the maps presented in the article. GF wrote the manuscript. FC, AM, wrote the section "Bacteria and Archea" and "Opportunistic Pathogens and Drug Resistance" with contributions of CT, FT, TV, and RK. PS and DT contributed

### REFERENCES


to the writing of the Microalgae and Harmful Algal Blooms section. RP, PS, RC, DT, RP, CR contributed to the section "Water Quality: pollutants in the Bay". JV, RM, and GF contributed to the writing the section "Consequences of Guanabara Bay Pollution to higher Trophic Levels and to Human Populations". All authors provided scientific expertise and all authors contributed to the editing of the manuscript.

### ACKNOWLEDGMENTS

The authors thank CNPq, CAPES, and FAPERJ for financial support. The authors thank Michelle Vils (pictures **Figures 1a,c**) and Wanderson F. de Carvalho (picture **Figure 1d**).


high levels of resistance and gene transfer elements. *PLoS ONE* 6:e17038. doi: 10.1371/journal.pone.0017038


and environmental variables in a tropical marine environment, Rio de Janeiro. *Environ. Microbiol.* 10, 189–199. doi: 10.1111/j.1462-2920.2007.01443.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 © 2015 Fistarol, Coutinho, Moreira, Venas, Cánovas, de Paula, Coutinho, de Moura, Valentin, Tenenbaum, Paranhos, do Valle, Vicente, Amado Filho, Pereira, Kruger, Rezende, Thompson, Salomon and Thompson. 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.*

# Bacterioplankton Dynamics within a Large Anthropogenically Impacted Urban Estuary

Thomas C. Jeffries 1, 2 \* † , Maria L. Schmitz Fontes 1 †, Daniel P. Harrison3, 4 , Virginie Van-Dongen-Vogels <sup>1</sup> , Bradley D. Eyre<sup>5</sup> , Peter J. Ralph<sup>1</sup> and Justin R. Seymour <sup>1</sup>

*<sup>1</sup> Plant Functional Biology and Climate Change Cluster, University of Technology Sydney, Sydney, NSW, Australia, <sup>2</sup> Hawkesbury Institute for the Environment, Western Sydney University, Penrith, NSW, Australia, <sup>3</sup> School of Geosciences, University of Sydney Institute of Marine Science, The University of Sydney, Sydney, NSW, Australia, <sup>4</sup> Sydney Institute of Marine Science, Mosman, NSW, Australia, <sup>5</sup> Centre for Coastal Management, Southern Cross University, Lismore, NSW, Australia*

### Edited by:

*George S. Bullerjahn, Bowling Green State University, USA*

### Reviewed by:

*Ryan J. Newton, University of Wisconsin-Milwaukee, USA Barbara J. Campbell, Clemson University, USA*

> \*Correspondence: *Thomas C. Jeffries t.jeffries@uws.edu.au*

*† These authors have contributed equally to this work.*

### Specialty section:

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

Received: *29 July 2015* Accepted: *02 December 2015* Published: *26 January 2016*

### Citation:

*Jeffries TC, Schmitz Fontes ML, Harrison DP, Van-Dongen-Vogels V, Eyre BD, Ralph PJ and Seymour JR (2016) Bacterioplankton Dynamics within a Large Anthropogenically Impacted Urban Estuary. Front. Microbiol. 6:1438. doi: 10.3389/fmicb.2015.01438* The abundant and diverse microorganisms that inhabit aquatic systems are both determinants and indicators of aquatic health, providing essential ecosystem services such as nutrient cycling but also causing harmful blooms and disease in impacted habitats. Estuaries are among the most urbanized coastal ecosystems and as a consequence experience substantial environmental pressures, providing ideal systems to study the influence of anthropogenic inputs on microbial ecology. Here we use the highly urbanized Sydney Harbor, Australia, as a model system to investigate shifts in microbial community composition and function along natural and anthopogenic physicochemical gradients, driven by stormwater inflows, tidal flushing and the input of contaminants and both naturally and anthropogenically derived nutrients. Using a combination of amplicon sequencing of the 16S rRNA gene and shotgun metagenomics, we observed strong patterns in microbial biogeography across the estuary during two periods: one of high and another of low rainfall. These patterns were driven by shifts in nutrient concentration and dissolved oxygen leading to a partitioning of microbial community composition in different areas of the harbor with different nutrient regimes. Patterns in bacterial composition were related to shifts in the abundance of *Rhodobacteraceae*, *Flavobacteriaceae*, *Microbacteriaceae*, *Halomonadaceae*, *Acidomicrobiales,* and *Synechococcus*, coupled to an enrichment of total microbial metabolic pathways including phosphorus and nitrogen metabolism, sulfate reduction, virulence, and the degradation of hydrocarbons. Additionally, community beta-diversity was partitioned between the two sampling periods. This potentially reflected the influence of shifting allochtonous nutrient inputs on microbial communities and highlighted the temporally dynamic nature of the system. Combined, our results provide insights into the simultaneous influence of natural and anthropogenic drivers on the structure and function of microbial communities within a highly urbanized aquatic ecosystem.

Keywords: microbial ecology, estuarine ecology, metagenomics, anthropogenic impacts, environmental pollutants, environmental microbiology, microbiome, eutrophication

# INTRODUCTION

Estuaries are among the most urbanized coastal ecosystems (Line and White, 2007) and as a consequence experience substantial environmental pressures, including habitat loss, decreased biodiversity, harmful algal blooms, anoxia, and contamination by sewage, pesticides, polycyclic aromatic hydrocarbons, heavy metals and other organic and inorganic pollutants (Birch et al., 1999; Diaz and Rosenberg, 2008). While these anthropogenically derived pressures have negative influences felt throughout entire estuarine food webs, the most prominent impacts often occur among populations of planktonic and sediment-bound microorganisms. Due to their sensitive and rapid responses to chemical perturbations, estuarine microbial communities often simultaneously act as sentinels of environmental impact and contributors to further deterioration of habitat health (Sun et al., 2012). While dramatic and conspicuous changes in estuarine biogeochemistry (e.g., anoxia, hydrogen sulfide evolution) often occur as a consequence of shifts in ecosystem microbiology (Cole, 1999; Crump et al., 2007; Breitburg et al., 2010; Abell et al., 2013, 2014), the microbial ecology underpinning these changes is often not well understood.

Estuaries typically host high microbial diversity and display substantial spatiotemporal heterogeneity in microbial abundance, activity and composition as a consequence of natural gradients of physical (e.g., light, salinity, and temperature) and chemical (e.g., inorganic and organic nutrients) conditions between the watershed and the mouth of the estuary (Crump et al., 2007; Fortunato et al., 2012). In addition to natural variation in environmental parameters, anthropogenic impacts such as increased nutrient input (eutrophication) and contamination by substances such as hydrocarbons and industrial effluent can lead to shifts in microbial communities in urbanized estuaries (Gillan et al., 2005; Vieira et al., 2008; Gregoracci et al., 2012; Sun et al., 2012, 2013) with implications for both ecosystem and human health. Microbial communities are extremely sensitive to rapid changes in the environment and can be used as indicators of stress (Paerl, 2006; Sun et al., 2012) as changes in the relative abundance of specific taxa or functional genes can be indicative of shifts in the physicochemical dynamics within estuaries and coastal systems (Smith et al., 2010; Fortunato et al., 2012, 2013; Gregoracci et al., 2012).

Within estuaries, natural freshwater inputs strongly influence the chemical, physical, and biological characteristics of the ecosystem, and salinity gradients often drive strong shifts in the structure of microbial communities, with discrete assemblages forming in zones of different salinity (Bouvier and Del Giorgio, 2002; Crump et al., 2004; Kirchman et al., 2005; Smith et al., 2010; Fortunato et al., 2012, 2013; Campbell and Kirchman, 2013). Due to the high human population densities that often occur near to estuaries, anthropogenic pressures also create significant spatial and temporal heterogeneity in physical and chemical conditions within estuaries. The dynamics of these anthropogenic inputs can be expressed either as spatial gradients along an estuary, such as in the case of concentrated inputs of agricultural or urban contamination at the top of the estuary, or as localized, and often pulse, inputs, leading to hotspots of contamination. Anthropogenic influences can include the input of organic and inorganic nutrient rich stormwater and sewage run-off (Beck and Birch, 2012), agricultural run-off, which can be rich in either inorganic nutrients from fertilizers or toxins from pesticides (Vieira et al., 2008; Gregoracci et al., 2012), industrial waste enriched in heavy metals and other toxins (McCready et al., 2006) or thermal pollution associated with the cooling water from powerplants and other large industry (Shiah et al., 2006). Microbial responses to these impacts can lead to altered and often unbalanced nutrient cycles, anoxic conditions, blooms of harmful algae (Paerl, 1997, 2006; Anderson et al., 2002) and increases in enteric and endemic pathogen concentrations (Hsieh et al., 2007). On the other hand, microbial populations may also assist in remediating estuarine systems, by rapidly degrading contaminants including hydrocarbons and fertilizers (Head et al., 2006). However, the inherent spatial and temporal variability in microbial community composition and function that occurs in estuaries means that these "positive" and "negative" ecosystem services provided by microbial assemblages will likely show substantial heterogeneity in space and time. Therefore, understanding the microbial ecology of urban estuaries is fundamentally important to monitoring the occurrence, distribution and fate of pollutants.

Here, we used a model-urbanized estuary, Sydney Harbor, to examine microbial dynamics over space and time within an anthropogenically impacted ecosystem. More than 85% of the harbor's catchment area is urbanized, with almost one quarter of the total population of Australia living in the surrounding area (Birch et al., 1999). Historically, the most urbanized areas are located in the western region of Sydney Harbor, where multiple negative impacts have been experienced, including elevated levels of pesticides, polycyclic aromatic hydrocarbons, nutrients, heavy metals, and suspended material as well as regular algal blooms and periods of anoxia (Birch et al., 1999; McCready et al., 2000, 2006; Birch and Rochford, 2010; Beck and Birch, 2012; Hedge et al., 2014).

Precipitation regimes strongly influence the hydrology and biogeochemistry of the Sydney Harbor estuary system, which remains well-mixed under low- or no-rainfall events, but experiences substantial horizontal stratification following highrainfall events (Birch et al., 1999), primarily as a consequence of point-source inputs from localized stormwater runoff points (Beck and Birch, 2012). As a consequence, a qualitative correlation between rainfall events and a range of ephemeral environmental episodes, including algal blooms and periods of anoxia, have been recorded in the estuary (Birch et al., 1999; Beck and Birch, 2012). It is well established that rainfall results in the allochtonous input of both naturally derived nutrients from the catchment area, and nutrients from anthropogenic sources such as fertilizer and sewage (Hedge et al., 2014). Indeed sewage has been responsible for over 50% of the total nitrogen and phosphorous in this system (Birch et al., 2010) and urban runoff and sewage overflows are two major contributor's to nutrient input into the system (Hedge et al., 2014). However, despite the ecological, economic and intrinsic national importance of the Sydney Harbor estuary and an increasing interest in the ecology of this habitat (Hedge et al., 2014) the microbial ecology of this system remains virtually unexplored and we lack any perception of how the diversity and ecological function of Sydney Harbor's microbiota responds to ongoing anthropogenic pressures. Here we address this knowledge gap by applying amplicon sequencing and shotgun metagenomic approaches to investigate the microbial ecology of a model urbanized estuary within the context of spatiotemporal heterogeneity in environmental variables. Therefore, we hypothesize that structure of the microbial community inhabiting the Sydney Harbor estuary will be highly variable in space and time as a consequence of both natural and anthropogenically-driven environmental heterogeneity.

# MATERIALS AND METHODS

### Study Area and Water Collection

The Sydney Harbor estuary has a total area of 480 km<sup>2</sup> , and it is among the most highly urbanized regions of the Australian coast (Birch et al., 2010). The Harbor's average depth is approximately 10 m (maximum 46 m), and maximum tidal range is 2.1 m (Hatje et al., 2003), with flushing times that vary from < 1 day in the mouth up to 225 days in the uppermost regions of the estuary (Das et al., 2000). Land-use surrounding the Sydney Harbor estuary also varies, with more commercial and industrial areas located in the inland, western parts of the estuary (Birch et al., 2010), relative to the more pristine and marine conditions near to the estuary mouth in the east (**Figure 1**).

We examined patterns in the composition and function of microbial assemblages across 30 sites spanning the 30 km length of the estuary (**Figure 1**) from Parramatta weir in the western brackish region of the estuary, where salinity levels are riverine (e.g., 12 ppt during rainfall), to the mouth of the estuary (Sydney Heads), where marine conditions are experienced (e.g., 34 ppt; **Figure 2**). These sites covered six main regions of the estuary, and were defined geographically as the Parramatta River (n = 5), Lane Cove (n = 4), Middle Harbor (n = 5), Western-Harbor (n = 5), Eastern-Harbor (n = 7) and Marine/Harbor Heads (n = 4) regions (**Figure 1**). Sample collection was timed to capture the dynamics of bacterial communities during two contrasting precipitation periods: a moderately high rainfall period in February (average monthly total 165.4 mm; 184 mm in the 2 weeks preceding sampling) and a low rainfall period in September (average monthly total 32.2 mm, 0 mm in the 2 weeks preceding sampling; Australian Bureau of Meteorology). We recognize that these sampling periods provide snap-shots of different precipitation regimes in this environment and are not suitable for predicting the long-term temporal heterogeneity of the estuary. Given the disparate environmental conditions of these time-points (discussed below), they were representative examples of contrasting ecosystem states in the estuary.

# Water Quality and Nutrient Analysis

Physical and biological properties were measured at each site using a multi-parameter water quality probe (YSI-6600, Yellowstone Instruments, USA) fitted with; conductivity, temperature, depth, pH, O2, turbidity, & chlorophyll sensors. For nutrient analysis, 5 ml unfiltered sample was immediately collected for Total Nitrogen (TN) and Total Phosphorus (TP) and placed into a 10 ml polycarbonate vial, 80 ml was filtered immediately through a polycarbonate 0.45µm cellulose acetate syringe filter (Whatman) for dissolved nutrient analysis including mono-nitrogen oxides, nitrite, nitrate, phosphates, ammonium, silicate (NOx, NO<sup>−</sup> 2 , NO<sup>−</sup> 3 , PO3<sup>−</sup> 4 , NH<sup>+</sup> 4 , and Si respectively), Total Dissolved Nitrogen (TDN) and Total Dissolved Phosphorous (TDP). All vials were placed on ice in the dark and frozen (−30◦C) prior to analysis (within 3 months).

Nutrients were analyzed using flow injection analysis on a LaChat 8500 instrument. Total Nitrogen and Total Phosphorus

(TN/TP) and TDN/TDP were prepared jointly using the modified alkaline peroxidisulfate autoclave digestion described by Maher et al. (2002). NOx, NH<sup>+</sup> 4 , PO3<sup>−</sup> 4 , and Si were analyzed using standard methods 4500-N0<sup>−</sup> <sup>3</sup> G., 4500-NH<sup>3</sup> H., 4500- P, and 4500-SiO2, F respectively (APHA, 2005). NO<sup>−</sup> <sup>2</sup> was analyzed using the NO<sup>x</sup> method without cadmium reduction. Total suspended solids (TSS) were measured gravimetrically after drying at 105◦C using a 25 mm GF/F filter (Whatman), Standard Method 2540 D, (APHA, 2005). Further details of nutrient analysis can be found in Eyre (2000).

### DNA Extraction and 16S rRNA Analysis

Two liter near-surface water samples (∼0.5 m depth) water were filtered onto 0.2µm polycarbonate membrane filters (Millipore), and filters were stored at -20◦C until DNA extraction. DNA was extracted using a bead beating and chemical lysis kit (MOBIO PowerWater, Carlsbad, CA, USA), according to the manufacturer's instructions. Genomic DNA concentrations were measured using a Qubit 2.0 fluorometer (Invitrogen, Carlsbad, CA, USA).

16S rRNA amplicon pyrosequencing was used to profile the composition of bacterial communities. Briefly, DNA samples were amplified with the 16S rRNA universal Eubacterial primers 803F (5′ -attagataccctggtagtc-3′ ) and 1392R (5′ -acgggcggtgtgtRc-3 ′ ; Engelbrektson et al., 2010) using the following cycling conditions: 95◦C for 3 min; 25 cycles of 95◦C for 30 s, 55◦C for 45 s and 72◦C for 90 s; followed by a final extension at 72◦C for 10 min. Amplicons were subsequently sequenced on the 454 platform using titanium chemistry (Roche) at the Australian Centre for Ecogenomics (Queensland, Australia). DNA sequences were processed using the Quantitative Insights Into Microbial Ecology (QIIME) pipeline (Caporaso et al., 2010b) as previously described for 454 data (Gibbons et al., 2013). Briefly, DNA sequences were de-multiplexed and reads shorter than 200bp, with a quality score < 25, or containing homopolymers exceeding 6bp were discarded. The 16S rDNA data was rarefied to an equal number of sequences per sample (1563) and thus normalized for differences in sequencing depth. Operational Taxonomic Units were defined at 97% sequence identity using UCLUST (Edgar, 2010) and assigned taxonomy against the Greengenes database (version 13\_5) (McDonald et al., 2012) using BLAST (Altschul et al., 1990). Chimeric sequences were detected using ChimeraSlayer (Haas et al., 2011) and filtered from the dataset. For beta-diversity analyses, representative sequences were aligned using PyNAST (DeSantis et al., 2006; Caporaso et al., 2010a) and the resultant phylogenetic tree, constructed using FastTree (Price et al., 2010), was used to calculate the weighted UniFrac distance between samples (Lozupone and Knight, 2005).

### Shotgun Metagenomes

Complete environmental DNA (metagenome) analyses were carried out on six samples, which were chosen as representative samples from each of the six Sydney Harbor regions described above (MH1, MH3, P3, P6, PJ3, and PJ7) during the February sampling (**Figure 1**), when environmental variables and community composition (as determined by 16S rRNA amplicon sequencing) was observed to be most variable. DNA was sequenced using the Illumina HiSeq 2000 platform (2 × 100 bp; paired-ends, Australia Genome Research Facility Ltd (AGRF) in Victoria, Australia). Sequences were subsequently analyzed using the Meta Genome Rapid Annotation using Subsystems Technology (MG-RAST, version 3.5; Meyer et al., 2008; Glass et al., 2010). Quality control was performed using DRISEE (Duplicate Read Inferred Sequencing Error Estimation) (Keegan et al., 2012) to check for Artificial Duplicate Reads (ADRs), and estimating sequence error (Gomez-Alvarez et al., 2009) within the MG-RAST pipeline using default paramaters. After quality control a total of 1.5–2.3 Gbp were generated per metagenome. Clusters of proteins were based on a 60% identity level, and protein annotation was conducted using BLAT (Kent, 2002) and OpenMP (Wilke et al., 2014). Metabolic assignments were annotated using the SEED subsystems database (Overbeek et al., 2005). Matches with an E-value of 1 × 10−<sup>5</sup> were considered significant using a minimum alignment of 30 bp of pairedend reads. All data were normalized to sequencing effort. The metagenomes can be accessed through MG-RAST under sample numbers 4550933.3 (site PJ3), 4550934.3 (site MH1), 4551435.3 (site MH3), 4551436.3 (site P3), 4551437.3 (site P6), and 4551749.3 (site PJ7).

# Statistical Analyses

Statistical analyses were carried out in PRIMER + PERMANOVA software v.6 (Clarke, 1993; Clarke and Gorley, 2006; Anderson et al., 2008). Homogeneity of variance in our data was tested using PERMDISP (Anderson et al., 2008), and where homogeneity of variances was indicated, the chance of a Type I error was reduced by rejecting the null hypothesis at a probability of 0.01. Environmental data was log+1 transformed prior to analysis if homogeneity of variances was determined using draftsman plots, then standardized by subtracting the mean from each value and dividing by the standard deviation, and finally Euclidian distance was used to form similarity matrices. Biological data was square-root (SQRT) transformed prior to calculating the resemblance matrix using Bray-Curtis similarity. Environmental and biological data were then graphically represented using non-metric multi-dimensional scaling (MDS). A similarity percentage (SIMPER) was used to identify the phylogenetic groups and functional levels contributing mostly to the dissimilarity in each area of the harbor.

We also determined whether patterns seen in bacterial community and environmental data ordinations were similar using the RELATE analysis through a rank correlation value (Rho) and significance levels. Permutational multivariate analysis of variance (PERMANOVA) was used to determine significant dissimilarity within bacterial communities and environmental data, comparing February and September and six regions in the harbor. In order to show the environmental variables that best explained community patterns, we used BEST analysis, and a distance based linear modeling (DistLM)—using a stepwise procedure for adjusted R <sup>2</sup>—that selected the variables that most likely explained patterns in the biological data. This analysis was graphically represented by a distance-based redundancy analysis (dbRDA) plot. MDS, analysis of similarity (ANOSIM), DistLM, dbRDA analyses were also shown for each month separately.

To define the statistical relationships between all environmental variables and taxonomic groups identified in the 16S rRNA amplicon analysis, we used the Maximal Information-based Nonparamteric Exploration (MINE) algorithm (Reshef et al., 2011). MINE calculates the strength of the relationship between each individual variable (MIC score) in addition to descriptors of the relationship such as linearity and regression. Only variables with values for >50% of samples were included and the dataset was filtered to include only significant (p < 0.05) correlations. Results were visualized with Cytoscape V3 (Shannon et al., 2003).

For metagenomic analysis, functional reconstructions generated using MG-RAST were imported into the Statistical Analysis of Metagenomic Profiles (STAMP 2.0.8) package (Parks et al., 2014) to determine statistically significant differences among metagenomes. We conducted a Fisher's exact test (Rivals et al., 2007) with Benjamini FDR multiple correction to identify the significant different functional categories between two samples (Benjamini and Hochberg, 1995). The corrected p-values (q-values) were used and only q-values < 0.05 were reported (Parks and Beiko, 2010). Differences between proportions of two samples were shown within the 95% confidence intervals as positive and negative values for the most different functions using the Newcombe-Wilson method (Parks and Beiko, 2010). It must be noted that the metagenomic data was not replicated within each habitat, constraining the conclusions regarding site-driven differences in function. However, we feel that these samples provide valuable "snap-shots" of discrete communities within a highly heterogeneous system. Fisher's exact test uses a hypergeometric distribution of sequences drawn without replacement from a pair of metagenomic samples to generate a statistical significance value (Parks and Beiko, 2010) and is routinely applied for the pairwise comparison of metagenomes (e.g., Parks and Beiko, 2010; Mendes et al., 2014; Chen et al., 2015; Tout et al., 2015). Our results however should be interpreted as differences between discrete metagenomes/samples rather than ecologically distinct environments within the harbor.

# RESULTS

# Environmental Variables

Environmental parameters in Sydney Harbor were highly heterogeneous in February after a strong rainfall event, as reflected by patterns in salinity and water temperature (**Figures 2A–D**). Salinity levels varied between 34 at the mouth of the estuary to less than 13 at several up-river sites. Alternatively, salinity levels during September were much more homogenous, exceeding 28 at all sites (**Figure 2F**). During the February sampling, temperatures exceeding 26◦C were observed in the western and upper river regions (Middle Harbor, Lane Cove, and Parramatta River), while in the eastern—central and marine regions of the estuary temperatures were between 22 and 24◦C (**Figure 2A**). In September, temperatures were substantially more homogenous and near to 20◦C throughout the entire estuary (**Figure 2E**).

Dissolved oxygen (DO) and pH levels were generally higher across the estuary during February, but were more heterogeneous at this time, with some localized sites where DO levels were below 3 mg L−<sup>1</sup> (e.g., site MH1; **Figures 2C,D**). In September, DO levels remained relatively consistent across the estuary, ranging between 6.6 and 9.5 mg L−<sup>1</sup> (**Figures 2G,H**). Nutrient concentrations (NOx, NH<sup>+</sup> 4 , PO3<sup>−</sup> 4 , and Si) were also more heterogeneous across the estuary in February relative to September. During February, higher nutrient concentrations were generally observed in the western and upper river sites (Parramatta River, Lane Cove, and Middle Harbor; **Figure 3**). Within the Parramata River, Lane Cove River, and Middle Harbor regions of the estuary, the most inland (up-river) sites displayed higher levels of NOx, NH<sup>+</sup> 4 , PO3<sup>−</sup> 4 , and Si (**Figure 3**). While less variable than February, localized nutrient hotspots also occurred in September, often within the same locations as were observed in February. This was particularly true for phosphate, which was consistently elevated in the west and upper river sites. Notably phosphate levels were also elevated in the eastern-central harbor and marine-harbor heads regions in September (**Figures 3C,G**).

Chlorophyll a (Chl a), Total Suspended Solids (TSS), TN, and TP all displayed the same spatial patterns, with hotspots observed in the west and upper river sites (Parramatta River, Western-central Harbor, Lane Cove, and Middle Harbor) during February, while during September, concentrations of suspended solids were elevated in the western region of the estuary relative to the east (**Figure 4**). When all environmental parameters were combined, the resultant ordination demonstrated a clear partitioning of samples between February and September. An exception to this was that during February the isolated and up-river sites at MH1 and MH5 displayed separation from all other sites (**Figure 5A**), suggesting that conditions within these two sites were physicochemically distinct from the rest of the Sydney Harbor ecosystem. These sites displayed reduced DO levels and increased concentrations of NOx and NH<sup>+</sup> 4 and PO3<sup>−</sup> 4 (**Figures 2D**, **3**).

### Patterns in Bacterial Taxonomy

As observed in the ordination analysis for environmental variables, the bacterial community composition also displayed substantial spatiotemporal heterogeneity across the Sydney harbor estuary, with two clear clusters of samples corresponding to the February and September samples (R of RELATE Spearman = 0.779; **Figure 5B**). Spatial patterns corresponding with the different spatial regions of the estuary were also apparent within each cluster, with samples from the marine and eastern and middle harbor regions generally grouping together and those from the western, lane cove and riverine regions showing a high-degree of similarity (**Figure 5B**). Multidimensional scaling conducted for each month individually

FIGURE 3 | Variability in nutrient concentrations in Sydney Harbor during high-rainfall (February 2013) and no rainfall (September 2013). Double circled symbols represent sites for which shotgun metagenomes were analyzed. Nitrate+nitrite (A,E), Ammonium (B,F), Phosphate (C,G), Silicate (D,H).

FIGURE 4 | Variation in total nutrient, chlorophyll, and suspended solid concentrations in Sydney Harbor during high-rainfall (February 2013) and no rainfall (September 2013). Double circled symbols represent sites for which shotgun metagenomes were analyzed. Chlorophyll-a (A,E), Total Suspended Solids (B,F), Total Nitrogen (C,G), Total Phosphorus (TP) (D,H).

FIGURE 5 | MDS Ordination of (A) environmental variables and (B) microbial phylogenetic diversity and abundance in Sydney Harbor. Colors designate geographic regions of the estuary. Circles, February 2013 (High Rainfall); Triangles, September 2013 (Low Rainfall).

highlighted the partitioning of samples by region with ANOSIM analysis demonstrating that this grouping was more significant than the null distribution representing a random structure to community composition (p < 0.05) (Supplementary Material Figures 8, 9)

BEST analysis revealed that temperature, salinity, pH, dissolved oxygen and phosphate were the strongest drivers of differences between bacterial communities (Supplementary Material Table 1). DistML revealed that temperature was the main driver of community shifts over time, explaining 21% of variability. The combined influence of temperature, phosphate, pH, salinity, DO, and silicate explained 47% of the variability (Supplementary Material Table 1). Redundancy analysis revealed that temporal shifts in the bacterial community were driven by temperature and salinity (Supplementary Material Figure 1); and that the key environmental drivers for spatial shifts in the microbial community were silicate, phosphate, DO and pH (Supplementary Material Figure 1), where bacterial communities from western and upper river sites—MH1, MH5, LPR1, LC1, and P1—were positively related to silicate, and phosphate, and negatively to DO. In February, DistLM and RDA plots showed that salinity and pH were the main drivers of community shifts, whereas TDP and NOx were more important in September (Supplementary Material Figure 7).

Rhodobacteraceae were the most abundant bacterial family, followed by the Flavobacteriaceae and Halomonadaceae. However, the relative abundance of these groups shifted in both space and time (**Figure 6**). Common marine bacterial groups such as SAR11 and Synechococcus were also found in the Sydney Harbor estuary, not surprisingly occurring in highest abundance at the mouth of the harbor (MH6, PJ6, PJ7, and LPR14) relative to the western estuarine sites (LPR1, P1, LC1, and LC2), while groups such as the Microbacteriaceae were more relatively abundant in the western and upper-river sites, particularly during February (**Figure 6**).

# Temporal Variability in Community Composition

SIMPER analysis revealed that Actinobacteria, Cyanobacteria and Bacteroidetes were the phyla that contributed the most to differences in bacterial community composition between the February and September sampling periods, and were together responsible for 39% of the temporal dissimilarity. On the other hand, Proteobacteria, the dominant phylum within the entire estuary, contributed to only 7.5% of total dissimilarity (Supplementary Material Table 2). Overall however, the relative abundance of phyla were less variable than at finer levels of taxonomic resolution, with individual families contributing to the dissimilarity between periods and showing larger shifts in relative abundance. For example, Microbacteriaceae, Halomonadaceae, Rhodobacteraceae, Acidimicrobiales OCS155, and Flavobacteriaceae contributed the most to differences in bacterial community composition between February and September, explaining 14.2% of dissimilarity (Supplementary Material Table 3).

FIGURE 6 | Relative abundance of microbial taxa (family level) in Sydney Harbor. Only families representing >0.1% abundance in any sample are shown. *F*, February 2013; S, September 2013.

Jeffries et al. Microbial Ecology of Sydney Harbor

Bacterial communities in the Parramatta River, upper-river and western-central harbor sites (i.e., the sites furthest from the marine conditions at the mouth of the estuary) displayed the strongest temporal shifts. The community shifted from one dominated by Rhodobacteraceae and Microbacteriaceae in February, to a community dominated by Halomonadaceae, Acidimicrobiales OCS155, and Flavobacteriaceae during September. In these three regions, there was also a notable shift in the ratio of Rhodobacteraceae to Flavobacteriaceae between February and September, with a higher relative proportion of Flavobacteriaceae observed during the dry period in September. Notably, these substantial shifts in bacterial community composition differed from those observed in the eastern sites of the harbor, where the bacterial assemblage was less variable between the two sampling periods (**Figure 6**).

# Spatial Variability in Community Composition

Within individual sampling periods, there were clear biogeographic shifts in the relative abundance of individual bacterial families. During both periods Rhodobacteraceae and Flavobacteriaceae showed some fluctuation in abundance but remained the dominant families, and the Halomonadaceae increased in abundance in the eastern regions of the estuary, generally increasing with proximity to the marine conditions at the harbor mouth. In contrast, Microbacteriaceae and Acidimicrobiales OCS155 were often relatively more abundant in the western regions of the estuary (**Figure 6**). Combined, the Microbacteriaceae, Halomonadaceae, Acidimicrobiales OCS155 and Flavobacteriaceae accounted for 13.4% of the spatial dissimilarity between the western, upper river sites and the eastern, marine sites in February, when the estuary was most heterogeneous (Supplementary Material Table 4). In addition to being key drivers of dissimilarity using SIMPER, the differential abundance of these groups between regions of the estuary were supported by two-group statistical analyses using STAMP (Supplementary Information Figure 10).

In addition to the broad spatial and temporal trends in bacterial community composition observed here, localized sites, where hotspots in nutrient concentrations or decreased DO levels occurred (e.g., MH1, LPR1, LC1 and LC2), often hosted taxonomically discrete microbial assemblages. For example, despite being in close proximity, the Lane Cove sites LC1 and LC3, hosted substantially different microbial communities, with LC1 having a much higher relative abundance of Acidomicrobiales C111 and OCS155 compared to LC3. These fine-scale shifts in community characteristics are likely explained by differences in the chemical characteristics of these two sites, with LC1 characterized by lower DO and pH and higher nutrient concentrations than LC3.

We used network analysis to identify statistical links between the relative abundance of specific bacterial taxa and environmental conditions (**Figure 7**). In the overall network, which was filtered to show only significant (P < 0.05) relationships between taxonomic groups and nutrients, there were 115 correlations with the nutrients to which there were the most associations being PO<sup>−</sup> 4 (16 taxa) and TN (15 taxa). The nutrients showing the strongest relationships were TDN and TN (average MIC score = 0.488 and 0.482 respectively). Chl a (indicative of high biomass and nutrient loading) was the most connected node (17 taxa, average MIC = 0.52). Several taxa that were identified as being key drivers of community dissimilarity using the SIMPER analysis were highly correlated to nutrient concentrations. In particular, the Flavobacteriaceae and Halomonodaceae showed negative correlations to seven and nine nutrients respectively, with the strongest being chl a, TN, TDN, and TP. Microbacteriaceae, which increased in abundance in the nutrient enriched western harbor was positively correlated to seven nutrients with the strongest relationships to chl a, TN and TDN.

# Spatial Shifts in Functional Potential (Shotgun Metagenomes)

To investigate how spatial heterogeneity in environmental parameters influenced the functional capacity of microbial assemblages inhabiting the Sydney Harbor estuary, we conducted a metagenomic survey of six representative sites in the Harbor. At the most coarsely defined level of metabolic processes (Level 1 in the SEED hierarchy), core "house-keeping" functions, including genes encoding carbohydrates, protein metabolism, amino acid and derivatives, cofactors, and RNA metabolism were dominant (representing over 60% of sequences) across all metagenomes (Supplementary Material Figure 2). To statistically compare differences in metagenomes between different habitats within Sydney Harbor we compared the functional profile of the representative marine sample (Marine/Harbor Heads) (PJ7) against the west and river metagenomes (Eastern-central Harbor, PJ3, and Parramatta River, P3; **Figure 8**). Metagenomes from a western more industrialized, region of Sydney Harbor (PJ3) displayed an overrepresentation of genes involved in lateral gene transfer, degradation of toxic aromatic compounds, and virulence and disease (**Figure 8A**) in addition to cofactors, vitamins and pigments fatty acids/lipids and sulfur metabolism (q < 0.05). Relative to the marine sample (PJ7), the Parramatta river sample (P3) was over-represented in genes for aromatic compound degradation, virulence, and disease and phosphorous metabolism (q < 0.05) (**Figure 8B**). Generally, the metagenomes located further from the estuary mouth had a higher contribution of genes involved in the sulfur and phosphorus metabolism, virulence, disease and defense, and degradation of aromatic compounds (Supplementary Material Figure 2; **Figure 9**).

Within the context of inorganic nutrient metabolism, there were several differences observed between the different regions within the Sydney Harbor estuary. Relative to the other sites, the western-most Parramatta River metagenome (site P3) was characterized by an over-representation of genes associated with phosphorus metabolism, but interestingly an under-representation of genes involved in nitrogen metabolism (**Figure 9**). The Middle Harbor site (MH1) was characterized by a higher abundance of nitrogen metabolism genes (**Figure 9**) including nitrate and nitrite ammonification genes, as well as denitrification genes and dissimilatory nitrite reductase

(Supplementary Material Figure 3, q < 0.05). The west and upper-river metagenomes (more eutrophic sites) displayed an over-representation of genes involved in phosphorus metabolism generally (**Figure 9**) including "P uptake by cyanobacteria" and "high affinity phosphate transporters" (Supplementary Material Figure 4, q < 0.05). The Middle Harbor metagenome (MH1), which was characterized by low DO levels, displayed an over-representation of genes involved in sulfur metabolism (**Figure 9**), in particular those involved in "sulfate reduction associated complexes," "inorganic sulfur assimilation," and "sulfur oxidation" (Supplementary Material Figure 5, q < 0.05). Additionally, genes associated with "phages and prophages" were higher at Middle Harbor (MH1) and Eastern-Central Harbor (PJ3), while Middle Harbor (MH1 and MH3) revealed a higher number of genes involved in "virulence, disease, and defense," which were largely made up of antibiotic resistance pathways (Supplementary Material Figure 6).

### DISCUSSION

By combining measurements of physicochemical conditions with assessments of patterns in microbial diversity and functional potential, we have provided a first insight into the microbial ecology of the Sydney Harbor estuary. Overall, abiotic variables and nutrient concentrations displayed significant differences between the high-rainfall period in February and the low rainfall period in September. During February we observed a higher occurrence of localized hotspots of nutrient concentrations and decreased DO, particularly within the western and up-river regions of the estuary. These patterns are indicative of localized eutrophication potentially related to point source, storm-water related inputs of nutrients. This localized variability overlayed larger scale spatial gradients in parameters including salinity and nutrients, which are reflective of both the natural environmental gradients expected within estuaries and the higher levels of urbanization and industrialization in the western regions of the estuary. The spatial heterogeneity of nutrient concentrations observed here is consistent with previous observations in Sydney Harbor that described a general trend of higher nutrient concentrations toward the western (inland) end of the harbor (Birch et al., 1999), with an enhanced influence of tidal flushing at the marine end of the system (Hedge et al., 2014).

Increased environmental heterogeneity of this system following periods of heavy rain has previously been observed (Lee et al., 2011), where at moderate rainfall levels nutrients accumulate near to input sources (Birch et al., 2010). This patchiness was evident in the nutrient distributions observed

here, which were heterogeneous during both sampling periods, but in particular during February. Freshwater inflows deliver nutrients, suspended solids and contamination into the Sydney Harbor (Birch and Rochford, 2010) with hydrological models suggesting that sewage contributes over 50% of TN and TP to the estuary (Birch et al., 2010) in addition to nutrients derived from fertilizer use and anthropogenic hydrology modification (Hedge et al., 2014). The spatiotemporal differences in the magnitude of these inputs were a likely driver of many of the shifts in the composition of microbial communities observed in this study. Additionally, the direct input of microorganisms from sources such as sewage and wastewater treatment could directly influence the community composition of the system. High concentrations of microorganisms persist in wastewater systems (Vandewalle et al., 2012; Liu et al., 2015b) and have been shown to influence the function and nutrient cycles of aquatic habitats (Mußmann et al., 2013). Fecal coliforms are known to be present in Sydney Harbor (Hose et al., 2005) and wastewater is thought to be a

direct source of microbial contamination into the estuary (Hedge et al., 2014).

Our data indicated that bacterial assemblages inhabiting the Sydney Harbor estuary exhibit substantial shifts in composition in both space and time, which can be explained by patterns in physical conditions and nutrient concentrations. The clear partitioning of community beta-diversity and shifts in relative abundance of taxa between a period of high-rainfall (February) and low-rainfall (September) are potentially linked to allochtonous inputs driven by increased inflow from rainfall. Temporal shifts and seasonality in microbial assemblages are well established in aquatic habitats (Fuhrman et al., 2006, 2015) and have been linked to increased anthropogenic nutrient loads (Perryman et al., 2011). In Sydney Harbor specifically, increased stormwater runoff during wet periods has been shown to drive variability in coliform and enterococci concentrations, particularly in the less often flushed western reaches of the system (Hose et al., 2005). Here we extend upon these previous culture-dependent observations to demonstrate that broad community-level shifts occur in both space and time.

Temperature, salinity, pH, dissolved DO and phosphate were identified as key drivers of the spatiotemporal dissimilarity between microbial assemblages inhabiting different regions of the Sydney Harbor estuary. In this study we have considered salinity, temperature, and pH as naturally varying parameters, and increasing nutrient concentrations and low DO levels as potentially anthropogenic influences in addition to obvious anthropogenic inputs such as hydrocarbons and chemical toxins. Salinity and temperature are common natural drivers of microbial biogeography in estuaries, with the microbial assemblages inhabiting the typically warmer, low salinity upper regions of estuaries often displaying markedly different community characteristics to those observed in the often cooler, marine conditions near to estuary mouths (Schultz et al., 2003; Crump et al., 2004; Kirchman et al., 2005; Campbell and Kirchman, 2013). Whilst some nutrient levels are the result of natural processes, anthropogenic sources are responsible for significant proportions of nutrient input in this system (Birch et al., 2010; Hedge et al., 2014). Both natural gradients in organic and inorganic nutrients and point source inputs of high concentrations of nutrients associated with stormwater, agricultural run-off and sewage create heterogeneity in nutrient distributions within estuaries (Paerl, 2006; Birch et al., 2010; Liu et al., 2015a). In particular, phosphorous, and phosphates are common contaminants in urban and agricultural runoff and lead to eutrophication in aquatic systems; resulting in community shifts, reduced oxygen levels and potentially blooms of harmful organisms (Correll, 1998; Anderson et al., 2002; Paerl et al., 2003; González-Ortegón and Drake, 2012; Carney et al., 2015). We found that within Sydney Harbor, phosphate was a principle driver of differences among microbial assemblages and found that localized hotspots of phosphate concentration often coincided with the occurrence of microbial assemblages with differing community characteristics to those observed throughout other regions of the harbor. Often increased eutrophication drives decreases in oxygen availability in aquatic systems (Paerl, 2006; González-Ortegón and Drake, 2012) and dissolved oxygen is a major factor in structuring aquatic microbial communities (Crump et al., 2007; Wright et al., 2012; Laas et al., 2015), particularly those inhabiting the freshwater zones of estuaries (Liu et al., 2015a) as was observed here. Thus, shifts in microbial community composition with nutrients may also correspond with anoxic hotspots and peaks in anaerobic bacteria such as the Purple Sulfur Bacteria (Chromatiales), as was observed at several up-river sites in our dataset.

To further elucidate the interaction between specific bacterial taxa and measured environmental parameters we applied a network analysis approach. Generally nutrients were highly correlated to the abundance of specific taxa and both phosphate (PO<sup>−</sup> 4 ) and Total Nitrogen (TN) were the most highly connected nutrients further highlighting their role in structuring the bacterioplankton community. The Microbacteriaceae and a marine clade of Acidimicrobiales, previously shown to exhibit strong temporal dynamics in aquatic systems (Needham et al., 2013), both showed strong positive associations with phosphate and other nutrients. Both taxa were among the top drivers of community dissimilarity identified using SIMPER analysis, which when linked to the correlations observed using network analysis confirm that inorganic nutrient concentrations are a principle driver of microbial community dynamics within the Sydney Harbor estuary. However, other major contributors to community dissimilarity such as Halomonodaceae and Flavobacteriaceae were negatively correlated to nutrient concentrations, as were Pelagibacter and Synechococcus. These taxa are all common in marine habitats (Arahal and Ventosa, 2006; Gómez-Pereira et al., 2010; Brown et al., 2012; Mazard et al., 2012) and showed higher abundances in the eastern seaward regions of the estuary, reflecting the preferences of these organisms for marine salinity and lower nutrient concentrations. Interestingly, the most abundant bacterial family across Sydney Harbor, the Rhodobacteraceae, was not significantly associated with nutrient concentrations, suggesting the distribution of these bacteria is controlled by other factors such as salinity and temperature, or that there is heterogeneity of nutrient acquisition strategies at finer phylogenetic levels within this group. This family belongs to the order Rhodobacterales, which is highly abundant in the marine environment (Gilbert et al., 2010) and has previously been found to increase in abundance at the marine end of estuaries (Campbell and Kirchman, 2013; Liu et al., 2015a).

Spatial shifts in the metabolic potential of the bacterial communities inhabiting the Sydney Harbor estuary also indicated the links between environmental variables and microbial function in the estuary. This is consistent with other metagenomic surveys in aquatic habitats (Rusch et al., 2007; DeLong, 2009; Gilbert et al., 2010) and along salinity gradients (Jeffries et al., 2012). Over-representation of metabolic pathways involved in the degradation of aromatic compounds (hydrocarbons), degradation of toxic compounds, and virulence observed in metagenomes from the more industrialized sites in the western region of the harbor, provide further evidence for the influence of anthropogenic activity on microbial function.

Strong patterns in pathways involved in nutrient cycling corresponded to gradients in nutrient concentration and hotspots of substrates and anoxia. The observation that pathways involved in phosphorous utilization were higher in the metagenomes collected at the more eutrophic inland and river sites is congruent with the high concentrations of phosphate and total phosphorous in these locations. The increase in metabolic pathways associated with phosphorous and nitrogen cycling in specific sites indicated that Sydney Harbor microbial communities exhibit functional responses to allochtonous pulses of these chemicals, potentially influencing overall nutrient flux and aquatic health. Similarly a strong peak in the abundance of genes involved in sulfur utilization occurred in the metagenome from middle harbor (MH1), which also corresponded to a peak in the abundance of the Chromatiales. This site was characterized by relatively stagnant conditions, highly anoxic and high nutrient concentrations. The metagenome from this site also had an overrepresentation of genes involved in sulfate reduction, relative to the other environments, which is also consistent with a low oxygen, substrate rich habitat. This trend of increasing nutrient metabolism gene content in eutrophic conditions may only be relevant for some abundant lineages within taxonomic groups, as in some cases oligotrophic conditions have been shown to increase the diversity and significance of pathways such as those involved in phosphate utilization in the open ocean (Martiny et al., 2009; Temperton et al., 2011).

Taken together these metagenomic results highlight the influence of anthropogenic inputs from effluent, industry, and agriculture on key microbial functions such as hydrocarbon degradation (industry) and nutrient cycling (run-off and agriculture) respectively. In particular, genes involved in the nitrogen and phosphorous cycling are potential influenced by eutrophication as a result of fertilizer use in agriculture and in high nutrient loadings in sewage. Genes related to hydrocarbon degradation and virulence are potentially related to industry and run-off from human waste respectively. Together with the 16S rRNA amplicon data, these results are consistent with studies in other estuarine environments that have linked the distributional dynamics of bacterial communities to nutrient gradients (Crump Jeffries et al. Microbial Ecology of Sydney Harbor

et al., 2004; Jeffries et al., 2012; Fortunato et al., 2013; Liu et al., 2015a) and allochtonous nutrient pulses (Carney et al., 2015). The links between nutrient enrichment, likely derived from stormwater and sewage inputs (Birch et al., 2010; Hedge et al., 2014), and microbial biogeography within Sydney Harbor highlight the influence of anthropogenic forces on defining the microbial ecology of urban estuaries. We acknowledge that it is not always possible to delineate between the mechanism behind physicochemical heterogeneity and that nutrient variability could be simultaneously driven by natural and anthropogenic influences. Based on previous literature however (e.g., Birch et al., 2010; Hedge et al., 2014) we are confident that large amounts of the input of nutrients to this system were derived from anthropogenic sources. Due to the multiple sources of impact and nutrient input in the Sydney Harbor estuary, particularly within the western regions of the ecosystem, it is not possible to discriminate single sources (e.g., industry, agriculture, and sewage over-flow) of the eutrophication underpinning the shifts in the microbial community. Nonetheless, the cumulative impact of a variety of inputs in the western region of the Sydney Harbor estuary have clearly led to elevated nutrient levels and associated shifts in the composition and function of the microbial communities.

# CONCLUSION

By combining physicochemical, taxonomic and metabolic datasets, this study has demonstrated the influence of both natural (e.g., temperature and salinity) and anthropogenically enhanced (e.g., high nutrients and low DO) environmental variability on defining microbial phylogenetic and functional biogeography in an urbanized estuary. As the first detailed survey of microbial diversity in Sydney Harbor, the most heavily populated region of coastline in Australia, this study revealed the highly dynamic spatiotemporal patterns in microbial communities within this habitat, which were linked to environmental heterogeneity driven by the natural physicochemical gradients, the influence of a rainfall event

# REFERENCES


and allochtonous nutrient inputs. The apparent links between potentially anthropogenically derived factors and microbial biogeography highlights the need for future studies which directly correlate individual microbial taxa and functional pathways to anthropogenic sources of input under low and high freshwater flows. This will lead to a better-understanding of the factors which underlie shifts in microbial composition and function in urbanized aquatic systems, and the consequential effects on aquatic and human health.

## ACKNOWLEDGMENTS

This research was supported by a Transfield Foundation Early Career Researcher Grant awarded to TJ, by the Australian Research Council Discovery Grant Scheme (DP110103091 and DP120102764) to JS and an Australian Research Council Future Fellowship (FT130100218) to JS. MS was supported by CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior), a Science Without Borders Program. The authors would also like to thank the Australian Centre for Ecogenomics for assistance with sequencing and Peter Freewater who facilitated the collaboration between organizations involved in this study. The authors appreciate the technical assistance of Iain Alexander, and technical staff at the Centre for Coastal Biogeochemistry at Southern Cross University, Tom Savage at the school of Geosciences, University of Sydney, and Ulysse Bove, Amanda Scholes, and Paul Hallam at the Sydney Institute of Marine Science. Sample collection and physio-chemical analyses were funded by; Greater Sydney Local Land Services as part of the "Sydney Harbor Water Quality Improvement Plan" and conducted at the Sydney Institute of Marine Science, and Southern Cross University.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmicb. 2015.01438


implications. Environ. Monit. Assess. 169, 531–551. doi: 10.1007/s10661-009- 1195-z


metagenomic sequencing data: drisee. PLoS Comput. Biol. 8:e1002541. doi: 10.1371/journal.pcbi.1002541


in the project to annotate 1000 genomes. Nucleic Acids Res. 33, 5691–5702. doi: 10.1093/nar/gki866


**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 Jeffries, Schmitz Fontes, Harrison, Van-Dongen-Vogels, Eyre, Ralph and Seymour. 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.

# Bloom Dynamics of Cyanobacteria and Their Toxins: Environmental Health Impacts and Mitigation Strategies

*Rajesh P. Rastogi1,2, Datta Madamwar1 and Aran Incharoensakdi2\**

*<sup>1</sup> BRD School of Biosciences, Sardar Patel University, Anand, India, <sup>2</sup> Laboratory of Cyanobacterial Biotechnology, Department of Biochemistry, Faculty of Science, Chulalongkorn University, Bangkok, Thailand*

Cyanobacteria are ecologically one of the most prolific groups of phototrophic prokaryotes in both marine and freshwater habitats. Both the beneficial and detrimental aspects of cyanobacteria are of considerable significance. They are important primary producers as well as an immense source of several secondary products, including an array of toxic compounds known as cyanotoxins. Abundant growth of cyanobacteria in freshwater, estuarine, and coastal ecosystems due to increased anthropogenic eutrophication and global climate change has created serious concern toward harmful bloom formation and surface water contamination all over the world. Cyanobacterial blooms and the accumulation of several cyanotoxins in water bodies pose severe ecological consequences with high risk to aquatic organisms and global public health. The proper management for mitigating the worldwide incidence of toxic cyanobacterial blooms is crucial for maintenance and sustainable development of functional ecosystems. Here, we emphasize the emerging information on the cyanobacterial bloom dynamics, toxicology of major groups of cyanotoxins, as well as a perspective and integrative approach to their management.

### *Edited by:*

*Federico Lauro, University of New South Wales, Australia*

### *Reviewed by:*

*Xavier Mayali, Lawrence Livermore National Laboratory, USA Gurjeet Singh Kohli, University of Technology Sydney, Australia*

*\*Correspondence:*

*Aran Incharoensakdi aran.i@chula.ac.th*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 18 June 2015 Accepted: 28 October 2015 Published: 17 November 2015*

### *Citation:*

*Rastogi RP, Madamwar D and Incharoensakdi A (2015) Bloom Dynamics of Cyanobacteria and Their Toxins: Environmental Health Impacts and Mitigation Strategies. Front. Microbiol. 6:1254. doi: 10.3389/fmicb.2015.01254*

Keywords: cyanobacteria, eutrophication, cyanobacterial blooms, cyanotoxins, ecotoxicology, mitigation strategies

# INTRODUCTION

Cyanobacteria are considered the most primitive groups of photosynthetic prokaryotes (Bullerjahn and Post, 2014) and possibly appeared on the Earth about 3.5 billion years ago (Tomitani et al., 2006). They are ubiquitous in nature and thrive in a variety of ecological niches ranging from desert to hot springs and ice-cold water. Most cyanobacteria are an immense source of several secondary natural products with applications in the food, pharmaceuticals, cosmetics, agriculture, and energy sectors (Rastogi and Sinha, 2009). Moreover, some species of cyanobacteria grow vigorously and form a dominant microflora in terms of their biomass and productivity in specific ecosystems. Bloom formations (**Figure 1**) due to excessive growth of certain cyanobacteria followed by the production of toxic compounds have been reported in many eutrophic to hypertrophic lakes, ponds, and rivers throughout the world (Rastogi et al., 2014). A range of toxic secondary compounds, called cyanotoxins, have been reported from cyanobacteria inhabiting freshwater and marine ecosystems. These toxic compounds are highly detrimental for survival of several aquatic

FIGURE 1 | Examples of excessive nutrient enrichment and bloom dynamics in freshwater ponds. (A) Harmful algal blooms in a pond at Chulalongkorn University, Bangkok, Thailand, showing the life of turtles (in red circle) under the toxic blooms condition (B). (C) Harmful algal blooms in a large pond in Varanasi, India (Photograph by R.P. Rastogi).

organisms, wild and/or domestic animals, and humans. Aquatic organisms, including plants and animals, and phyto/zooplanktons inhabiting under toxic bloom rich ecosystems, are directly exposed to the harmful effects of different cyanotoxins. The intoxication occurring in wild and/or domestic animals and humans is either due to direct ingestion of cells of toxin producing cyanobacteria or the consumption of drinking water contaminated with cyanotoxins (Rastogi et al., 2014). The toxicity of different cyanotoxins is directly proportional to the growth of cyanobacteria and the extent of their toxin production. It has been shown that the growth of different cyanobacteria and their toxin biosynthesis is greatly influenced by different abiotic factors such as light intensity, temperature, short wavelength radiations, pH, and nutrients (Neilan et al., 2013; Häder et al., 2014; Rastogi et al., 2014). Global warming and temperature gradients can significantly change species composition and favor blooms of toxic phytoplanktons (El-Shehawy et al., 2012; Häder and Gao, 2015).

It has been assumed that cyanotoxins play an important role in chemical defense mechanisms giving survival advantages to the cyanobacteria over other microbes or deterring predation by higher trophic levels (Vepritskii et al., 1991; Jang et al., 2007; Berry et al., 2008). Cyanotoxins may also take part in chemical signaling. Overall, information regarding the specific role(s) of cyanotoxins in the life of individual cyanobacteria or their ecological and biotechnological operations is still very limited and needs extensive research. In the present review, we summarize the recent advances on bloom dynamics, cyanotoxin production, and mitigation strategies as well as their consequences on environmental health perspectives.

# EUTROPHICATION, GLOBAL CLIMATE CHANGE, AND CYANOBACTERIAL BLOOM DYNAMICS

Occurrence of toxic cyanobacterial blooms (cyanoblooms) is a serious global problem which affects the water quality due to the production and accumulation of different cyanotoxins and other malodorous compounds. These blooms may cause an increase of biological oxygen demand (BOD) and anoxia in the water bodies, and death of aquatic life (Havens, 2008; Brookes and Carey, 2011; Rastogi et al., 2014). The factors contributing to the worldwide occurrence of cyanobacterial blooms are still debatable. Nevertheless, cultural eutrophication from domestic, industrial, and agricultural wastes as well as global climate change can play a major role in the global expansion of harmful algal blooms and toxin production (Kaebernick and Neilan, 2001; Conley et al., 2009; Smith and Schindler, 2009; Paerl and Scott, 2010; Kleinteich et al., 2012; O'Neil et al., 2012; Paerl and Paul, 2012; Neilan et al., 2013; Gehringer and Wannicke, 2014; **Figure 2**). Excessive loads of certain inorganic and/or organic nutrient concentrations are considered as strong risk factors for bloom promotion both in fresh and marine water habitats (Smith, 2003; Heisler et al., 2008; Conley et al., 2009; Ruttenberg and Dyhrman, 2012; Michalak et al., 2013; Davidson et al., 2014; Beversdorf et al., 2015). The anthropogenically mediated change in the N/P ratio has frequently been interrelated to the appearance of cyanobacterial blooms (Glibert et al., 2004). The phosphorus concentration was found as a primary regulating factor for increased cyanobacterial growth and changes of genotypes, both of which were found to be closely related to the water temperature, signifying the role of eutrophication in the occurrence of toxic blooms (Joung et al., 2011). Recently, Molot et al. (2014) presented a novel conceptual model linking anoxia, phosphorus (P), nitrogen (N), iron (Fe), and sulfate to the formation of harmful cyanobacterial blooms across three gradients, i.e., nutrients, salinity, and acidity. Continued transfer of sediments to a water body may block the natural flow of water and enrich the dissolved organic carbon and other compounds leading to potential risk of bloom formation.

Global climate change followed by changes in air/water temperature gradients, as well as increased nutrient precipitation can affect the cyanobacterial bloom formation and production of different cyanotoxins (Kanoshina et al., 2003; Paerl and Huisman, 2009; El-Shehawy et al., 2012; Paerl and Paul, 2012). Several environmental factors related to the dynamics of the abundance of toxic cyanobacterial bloom formation have been verified (Joung et al., 2011; Neilan et al., 2013). Warm and calm weather and low turbulence can enhance the formation of cyanobacterial blooms (Paerl and Huisman, 2008). Increased emission of ozone depleting substances (ODSs), due to huge burning of fossil-fuels and concomitant changes in air temperature, may promote the water cyanobacterial growth. As a result of climate change, the frequent droughts in summer as well as flash-flooding may lead to abandoned nutrient discharges from urban areas to unloading water bodies such as ponds, lakes, ditches, and estuaries with the consequence of the augmentation of toxic blooms and the increase of the BOD of a water reservoir (Whitehead et al., 2009). Nitrogen limitation under drought condition may cause a shift from non-N2-fixing to N2-fixing cyanobacteria leading to an increase in biologically available nitrogen and a subsequent production of cyanotoxins (Posch et al., 2012). The increased salination due to summer droughts, rising sea levels, wind flow, and common practices of the use of freshwater for agricultural irrigation, all have led to the origin and existence of several salt tolerant freshwater toxic cyanobacteria as evidenced by an increased number of blooms in brackish waters (Kanoshina et al., 2003; Paerl and Huisman, 2009). Under increased temperature and low wind mixing, the water column becomes stagnant and a large number of buoyant cyanobacteria move upward at the water surface causing dense surface blooms to fulfill their photosynthetic needs (Huisman et al., 2004; Paerl and Huisman, 2008, 2009; **Figure 2**). It has been established that dense cyanobacterial blooms require excessive CO2 to support their photosynthetic growth (Paerl and Huisman, 2009). Furthermore, global climate change due to anthropogenically released ODS and increased atmospheric CO2 levels can minimize carbon limitation of photosynthetic growth leading to increased algal biomass productions in the water reservoirs (Paerl and Huisman, 2009). Moreover, increased CO2 levels may increase the problems associated with the harmful cyanobacteria in eutrophic lakes (Sandrini et al., 2015). Recently, Verspagen et al. (2014) reported that rising CO2 levels may result in a marked intensification of phytoplankton blooms in eutrophic and hypertrophic waters. Climate change, which is predicted to lead the changes in rainfall patterns along with an increase in temperature may also influence the occurrence and severity of toxic cyanobacterial blooms due to a significant impact on inland water resources (Reichwaldt and Ghadouani, 2012). It has been suggested that UV-B radiation may significantly influence strain composition of cyanobacterial blooms in favor of microcystin (MC) producers (Ding et al., 2013). Several species/strains of bloom forming cyanobacteria produce different toxic peptides and alkaloids (**Table 1**), which are a major threat to the safe drinking water and pose a serious threat to the global environmental and human health (Kaplan et al., 2012; Rastogi et al., 2014). Until now, a number of views have been given for world-wide occurrence of cyanobacterial blooms (Paerl and Huisman, 2009; Paerl and Paul, 2012; Neilan et al., 2013; Rastogi et al., 2014); however, the exact mechanisms and the role of different environmental factors regulating the bloom dynamics are disputable and yet to be understood.

Our understanding of the responses of various environmental factors associated with climate change and their impact on marine/freshwater ecosystems is based on several experimental and/or inferential data. From the above discussions, it is clear that the appearance of a cyanobacterial bloom is the consequence of several coherent signals. It is utmost important

to unravel the specific effects of nutrient enrichment and other global climate change on our aquatic ecosystem, and to establish the facts on how the structure and function of an ecosystem can be maintained. Moreover, if the existing level of anthropogenically induced nutrient loading in the water bodies and environmental warming continues, multiple-fold increase in algal bloom followed by contamination of our aquatic ecosystem by several toxic substances is expected in future. Henceforth, most conceptual and empirical research on the triggers of cyanobacterial blooms is needed to understand the multifarious set of situations that influence the worldwide incidence of toxic cyanoblooms.

# TOXINS FROM CYANOBACTERIA

Cyanobacteria produce a wide range of toxic secondary compounds causing human and domestic/wildlife intoxication. A number of bloom forming cyanobacteria from diverse habitats have been reported to produce different toxins (Rastogi et al., 2014). Chemically, the cyanotoxins are divided into three main groups, i.e., cyclic peptides (MCs and nodularins), alkaloids (anatoxin-a, anatoxin-a(s), saxitoxins, cylindrospermopsin, aplysiatoxin, lyngbiatoxin-a), and lipopolysaccharides (LPSs; Kaebernick and Neilan, 2001). However, based on biological effects, the cyanobacterial toxins can be classified into five functional groups such as hepatotoxins, neurotoxins, cytotoxins, dermatotoxins, and irritant toxins (Sivonen and Jones, 1999; Codd et al., 2005).

# CYCLIC PEPTIDES

Among the different cyanobacterial toxins, MCs are the most frequently occurring cyanotoxins in surface as well as drinking water and widely investigated hepatotoxins. MCs are cyclic heptapeptides (**Figure 3**) produced by several strains of cyanobacteria (Sivonen and Jones, 1999; Krienitz et al., 2002; Izaguirre et al., 2007; Aboal and Puig, 2009; Rastogi et al., 2014; **Table 1**). Currently, more than 90 variants of MCs are known, all with the general structure cyclo-(D-Ala-X-D-MeAsp-Z-Adda- D-Glu- Mdha), X and Z being variable L-amino acids. On the basis of acute toxicity, microcystin-LR (MC-LR) is considered the most potent hepatotoxin (Funari and Testai, 2008).

Microcystin is synthesized non-ribosomally by large multienzyme complexes comprising different modules including non-ribosomal peptide synthetases (NRPSs) as well as polyketide synthases (PKSs), and several tailoring enzymes. The gene cluster responsible for MC biosynthesis has been identified in different cyanobacteria (Tillett et al., 2000; Rouhiainen et al., 2004; Christiansen et al., 2008; Gehringer et al., 2012). In the cyanobacterium *Microcystis aeruginosa* PCC7806, the MC gene clusters spans 55 kb of DNA and is composed of 10 (mcyABCDEFGHIJ) bidirectionally transcribed open reading frames (ORFs) arranged in two divergently transcribed operons, *mcyA-C* and *mcyD-J* (Tillett et al., 2000; **Figure 3**). The assembly of MC begins with the activation of a phenylalanine-derived phenyl propionate starter unit at the NRPS/PKS hybrid enzyme McyG (Hicks et al., 2006). The


### TABLE 1 | Some common cyanotoxins found in different cyanobacteria and their possible toxicity and mode of actions.

gene clusters encoding MC biosynthesis sequence from the *Microcystis* (Tillett et al., 2000), *Planktothrix* (Christiansen et al., 2008), and *Anabaena* (Rouhiainen et al., 2004) species revealed that the arrangements of ORFs in the *mcy* cluster vary among different genera. However, a high sequence similarity between the *mcy* gene clusters of different genera suggests a common ancestor for MC synthesis (Rantala et al., 2004).

Similar to MCs, cyclic pentapeptide toxic compounds, nodularins (NODs; **Figure 3**) represent the second group of hepatotoxins produced by the cyanobacteria *Nodularia* and *Nostoc.* At present, more than seven variants of NOD have been reported. Both hepatotoxins (MCs and NODs) contain a unique hydrophobic amino acid, Adda (2S,3S,8S,9S-3-amino-9-methoxy-2,6,8-trimethyl-10-phenyl-deca-4,6-dienoic acid). Chemically, NODs differ from MCs in terms of the absence of two core amino acids and have *N*-methyldehydrobutyrine (*Mdhb*) instead of *N*-methyldehydroalanine (*Mdha*; Rinehart et al., 1998). Similar to MCs, NODs are also produced nonribosomally from *nda* gene clusters by means of NRPS-PKS enzyme systems (Moffitt and Neilan, 2004; **Figure 3**). In the cyanobacterium *Nodularia spumigena* NSOR10, the locus of *nda* gene clusters (48 kb) consists of nine ORFs (*ndaA–I*) transcribed from a bidirectional regulatory promoter region (Moffitt and Neilan, 2004). Moreover, MCs and NODs show similar biological activity in spite of their different chemical structures. These cyclic peptides inhibit the specific protein serine/threonine phosphatases-1 (PP1) and -2A (PP2A) which are important regulatory enzymes in eukaryotic cells (MacKintosh et al., 1990).

# ALKALOIDS

A number of toxic alkaloids have been found in different cyanobacteria. The alkaloids anatoxin-a (MW = 165 Da) and its homolog homoanatoxin-a (MW = 179 Da) are fast-acting neurotoxins, also known as fast death factors (FDFs). Anatoxin-a (**Figure 4**) was first isolated from *Anabaena flos-aquae* and so far has been found in several cyanobacteria such as *A. circinalis*, *A. planctonica*, *A. spiroides*, *Aphanizomenon*, *Cylindrospermum*, *Planktothrix*, and *M. aeruginosa* (Edwards et al., 1992; Park et al., 1993; **Table 1**). The alkaloid homoanatoxin-a has a methylene group at C-2 instead of the acetyl group (**Figure 4**) and structurally resembles anatoxin-a. Homoanatoxin-a has been isolated from the cyanobacteria *Oscillatoria* (*Planktothrix*) *formosa*, *Phormidium formosum*, *Anabaena*, and *Raphidiopsis mediterranea* (Furey et al., 2003; Namikoshi et al., 2003; Watanabe et al., 2003). Another homolog of anatoxin, anatoxina(s) (MW 252 Da; **Figure 4**), isolated from *A. flos-aquae* and *A. lemmermannii*, is a potent acetylcholinesterase (AChE) inhibitor (Matsunaga et al., 1989) but more lethal than anatoxin-a (Carmichael et al., 1990; Méjean et al., 2014). It is synthesized in the cell from ornithine *via* putrescine catalyzed by the enzyme ornithine decarboxylase. Moreover, the partial genome sequencing demonstrated the presence of putative gene cluster (Méjean et al., 2014) encoding the biosynthetic pathway of anatoxin-a and homoanatoxin-a in cyanobacteria such as *Oscillatoria* PCC 6506 (Méjean et al., 2009) and *Anabaena* strain 37 (Rantala-Ylinen et al., 2011; **Figure 4**).

Saxitoxin and its analogs (e.g., neosaxitoxin; **Figure 5**) are a group of carbamate alkaloid toxins which all are highly

2014; Gene cluster not drawn to scale).

potent neurotoxins. These are tricyclic compounds, consisting of a tetrahydropurine group and two guanidine subunits, commonly called paralytic shellfish poisons (PSPs). Currently, about 27 variants of saxitoxins have been found in different cyanobacteria such as *Aphanizomenon*, *Anabaena flos-aquae*, *Anabaena circinalis*, *Lyngbya wollei*, and *Cylindrospermopsis raciborskii* (**Table 1**). Regulation of saxitoxin biosynthetic pathway and characterization of some enzymes involved are not well-studied (Soto-Liebe et al., 2010). However, it has been postulated that biosynthesis of saxitoxin depends on the multifunctional PKS enzyme, SxtA (Kellmann et al., 2008). The saxitoxin biosynthetic gene cluster (25.7–36 kb) includes 33 genes, reported in cyanobacteria such as *Cylindrospermopsis raciborskii* (strain T3), *Anabaena circinalis* (strain AWQC131C), *Aphanizomenon* strain NH-5, *Lyngbya wollei,* and *Raphidiopsis brookii* (strain D9; Kellmann et al., 2008; Mihali et al., 2009, 2011; Soto-Liebe et al., 2010; Stucken et al., 2010; Neilan et al., 2013; **Figure 5**). The positions of genes encoding biosynthetic enzymes, transporters, and regulatory proteins within the cluster differ among the different cyanobacterial strains dicsussed above. Moreover, the toxic profile expressed in different strains is determined by the position and presence, or absence, of specific genes in the respective clusters.

The cyanotoxin cylindrospermopsin (CYN: MW 415 Da) is a polyketide-alkaloid having a tricyclic guanidine moiety and sulfate groups (**Figure 6**). Presently, some analogs of CYN such as deoxy-cylindrospermopsin, demethoxycylindrospermopsin and 7-epicylindrospermopsin have been identified in the cyanobacteria *C. raciborskii* (Norris et al., 1999) and *Aphanizomenon ovalisporum* (Banker et al., 2000). The CYN variant 7-epicylindrospermopsin differs due to the orientation of the hydroxyl group close to the uracil moiety (Banker et al., 2000), and the other variant deoxy-cylindrospermopsin is characterized by a missing oxygen atom related to the initial hydroxyl group close to uracil moiety. Moreover, a number of cyanobacteria such as *Cylindrospermopsis raciborskii*, *Aphanizomenon ovalisporum*,

*Aphanizomenon flos-aquae*, *Anabaena lapponica*, *Anabaena bergii*, *Lyngbya wollei*, *Umezakia natans*, *Raphidiopsis curvata,* and *Oscillatoria* (*Planktothrix*) have been reported to produce CYN and its analogs (Ohtani et al., 1992; Harada et al., 1994; Banker et al., 1997; Preussel et al., 2006; Spoof et al., 2006; Seifert et al., 2007; Mazmouz et al., 2010). McGregor et al. (2011) reported the presence of the cyanotoxins CYN and deoxy-CYN from the cyanobacterium *Raphidiopsis mediterranea* FSS1-150/1 of a eutrophic reservoir in Queensland, Australia. CYN shows hepatotoxic, nephrotoxic, and cytotoxic effects and is a potential carcinogen owing to the inhibition of glutathione, cytochrome P450 and protein synthesis (Humpage et al., 2000; Froscio et al., 2003; Neumann et al., 2007). The gene cluster (*cyr*) encoding the enzymes of the CYN biosynthesis (**Figure 6**) has been reported to be present in several cyanobacteria such as *C. raciborskii* (Mihali et al., 2008; Stucken et al., 2010; Jiang et al., 2012), *Aphanizomenon* strain 10E6 (Stuken and Jakobsen, 2010), and *Oscillatoria* PCC 6506 (Mazmouz et al., 2010). The arrangements of genes and flanking regions differ across genera; however, all the gene clusters are highly conserved with respect to the nucleotide sequence of orthologous genes (Neilan et al., 2013). In case of the cyanobacterium *C. raciborskii* AWT205, the *cyr* gene cluster (42 kb) encodes 15 ORFs (*cyrA*-O). The biosynthesis of CYN is initiated by an amidinotransferase and completed by NRPS-PKS-type enzymes in combination with tailoring enzymes (Muenchhoff et al., 2010). As stated above, the gene cluster for CYN biosynthesis has been sequenced from several cyanobacteria; however, few studies have been conducted on its transcriptional organization and promoter structure (Stuken and Jakobsen, 2010).

# LIPOPOLYSACCHARIDES

The endotoxins LPSs consist of an internal acylated glycolipid (lipid-A), core domain (an oligosaccharide) and an outer polysaccharide (O-antigen) chain (Raetz and Whitfield, 2002). In general, the fatty acid component (lipid-A) of LPS is responsible

for the toxic actions such as irritant and allergenic responses in human and animal tissues (Mankiewicz et al., 2003). The LPSs present in cyanobacteria differ from those in enteric bacteria by having a larger variety of long chain unsaturated fatty acids and hydroxy fatty acids and the lack of phosphate. Moreover, there is substantial diversity of LPSs composition among the cyanobacteria, although variations are basically related to phylogeny. Different genera of cyanobacteria have distinct LPSs compositions conserved within the particular genus (Sivonen and Jones, 1999). Several cyanobacteria such as *Anacystis nidulans*, *Microcystis*, *Anabaena*, *Spirulina*, and *Oscillatoria* all have been reported to produce LPS toxin (Smith et al., 2008; Bláhová et al., 2013). The structure of the lipid-A subunit in the cyanobacterial LPS molecule has not been clearly identified, and furthermore, the exact mechanism of LPS toxicity produced by cyanobacteria is still unknown.

Besides the above mentioned cyanotoxins, a number of toxins such as aplysiatoxin, kalkitoxin, antillatoxin, lyngbyatoxins, cyanopeptolin, aurilides, and jamaicamides have been reported to be present in different cyanobacteria in fresh and/or marine water habitats (**Figure 7**). A phenolic bislactone alkaloid aplysiatoxin has been reported from several cyanobacteria such as *Lyngbia majuscula*, *Schizothrix calcicola*, *Trichodesmium erythraeum,* and *Oscillatoria nigroviridis*(Mynderse et al., 1977; Gupta et al., 2014). Aplysiatoxin and debromoaplysiatoxin (**Figure 7**) are potent tumor promoters and protein kinase C activators and show signs of several lethal effects. Moreover, an analog of the tumorpromoting aplysiatoxin has been reported as an antineoplastic agent rather than a tumor-promoting substance (Nakagawa et al., 2009). Recently, the analogs of aplysiatoxin debromoaplysiatoxin and anhydrodebromoaplysiatoxin, as well as two new analogs, 3-methoxyaplysiatoxin and 3-methoxydebromoaplysiatoxin have been reported from the marine cyanobacterium *Trichodesmium erythraeum* (Gupta et al., 2014). The alkaloid lyngbyatoxin, a prenylated cyclic dipeptide compound, was isolated from *Lyngbya majuscula* (Taylor et al., 2014) and has several similarities with aplysiatoxin in its mechanism of toxicity and both are potent tumor promoters. Kalkitoxin (**Figure 7**) is a lipopeptide neurotoxin produced by some species of cyanobacteria such as *L. majuscula* (Berman et al., 1999). The antillatoxin is an ichthyotoxic cyclic depsipeptide isolated from *L. majuscula* (Orjala et al., 1995). A number of bioactive peptides such as microviridins, microginins, cyanopeptolides, and <sup>β</sup>-*N*-methylamino-L-alanine (BMAA; **Figure 7**) have also been reported from diverse cyanobacteria, but their toxicological profiles and impacts on the environment as well as human health are not known (Downing et al., 2014). Moreover, the cyanobacterial neurotoxin, BMAA has been suggested to function as a causative agent for certain neurodegenerative diseases (Lobner et al., 2007). The compound curacin-A, isolated from *L. majuscula* (Gerwick et al., 1994), exhibited potent anti-proliferative and cytotoxic activity against colon, renal, and breast cancer derived cell lines (Verdier-Pinard et al., 1998). A cyanobacterial toxin cyanopeptolin (CP1020) produced by *Microcystis* and *Planktothrix* strains was found to cause transcriptional alterations of genes involved in DNA damage and repair (Faltermann et al., 2014). Recently, two new cyanobacterial peptides named micropeptins 1106 and 1120 were reported from cyanobacterial blooms in North Carolina's Cape Fear River. However, their biological activities have not yet been determined (Isaacs et al., 2014). Moreover, several studies indicate the presence of several additional, still unidentified and not characterized biotoxins in cyanobacterial blooms.

# ECOLOGICAL HEALTH IMPACTS OF CYANOTOXINS

The increased incidence of toxic cyanobacterial blooms is posing potential risks to aquatic ecosystem as well as human and animal health. Cyanotoxins may cause several harmful effects on humans or animals either through direct contact or by means of intake of untreated contaminated water and food (Miller et al., 2010; Papadimitriou et al., 2012; Rastogi et al., 2014; Sukenik et al., 2015). Aquatic organisms may be affected either through direct ingestion of toxic cyanobacterial cells or through contact with cyanotoxins. It has been established that intake of contaminated water or food is a key route for cyanotoxin intoxication (Zhang et al., 2009; Miller et al., 2010). Several secondary compounds have been reported to have their toxic effects on different organisms ranging from plant to animals. In the subsequent section we have focused on the adverse toxic effects of some common cyanotoxins on aquatic/wild animals and humans.

The cyanotoxin MCs are well-known for their toxic effects. MCs can affect the cellular system through disorganization of cytoskeleton, cell proliferation, genome damage, inhibition of enzyme activity, imprecise mitotic cell division, loss of membrane integrity, oxidative stress, and lipid peroxidation (Rastogi et al., 2014). To know the detailed mechanisms or mode of action of MCs, readers are referred to the recent review by Rastogi et al. (2014). MCs act by blocking protein PP1 and -2A, causing toxicity at the hepatic level. It has been demonstrated that MC-LR can induce reproductive (Chen et al., 2011; Zhou et al., 2012) as well as cardio-toxicity in animals (Qiu et al., 2009). MC-LR was found to cause normocyte anemia and the bone marrow injury, and also affected the immune system of rabbits (Zhang

et al., 2011; Yuan et al., 2012). Moreover, a number of fatal poisonings of MCs regarding the health risk of domestic and wild animals, birds, fish, and several other aquatic as well as terrestrial organisms have been reported worldwide (Stewart et al., 2008; Rastogi et al., 2014). The mass mortalities of Lesser Flamingos were reported at Lake Bogoria, Kenya due to MC intoxication (Krienitz et al., 2003). A new episode of cyanotoxin (MC-LR, -YR, and -RR) intoxication and mass mortalities of Lesser Flamingos (*Phoeniconaias minor* Geoffroy) have also been reported at Lake Manyara in Tanzania (Nonga et al., 2011). At least 6,000 birds belonging to 47 species, including endangered species such as the marbled teal (*Marmaronetta angustirostris*) and white-headed duck (*Oxyura leucocephala*), died due to MC-LR intoxication at the Doñana National Park, Spain (Lopez-Rodas et al., 2008).

Despite numerous reports of cyanotoxins impact on the aquatic organisms and wild or domestic animals, the epidemiological facts for cyanotoxins intoxication in humans are very limited (Rastogi et al., 2014). Recent studies have established the cytotoxic and genotoxic potentials of various cyanotoxins including MCs (Žegura et al., 2011). The use of untreated water contaminated with cyanobacterial blooms and MCs resulted in normocytic anemia (Pouria et al., 1998), liver failure and several other symptoms such as nausea, vomiting, and acute liver damage leading to human death in a hemodialysis center in Caruaru, Brazil (Pouria et al., 1998; Hilborn et al., 2007). The use of MC-contaminated water can be a potential risk factor for liver and colorectal cancer among humans (Lun et al., 2002; Hernández et al., 2009). Moreover, MCs may cause hepatotoxicity and neurotoxicity, kidney impairment, allergies and eye, ear and skin irritation, and certain gastrointestinal disorders such as nausea/vomiting and diarrhea in humans (Torokne et al., 2001; Pilotto et al., 2004; Codd et al., 2005).

As stated above, the cyanotoxin NODs have chemical structure as well as mechanisms of action similar to those of MCs (Yoshizawa et al., 1990); however, NODs have not been studied as extensively as MCs (Funari and Testai, 2008). NODs are a potent inhibitor of protein phosphatase 1 and 2A (Ohta et al., 1994) and show accumulative toxicity and tumor formation (Ohta et al., 1994; Sueoka et al., 1997; Song et al., 1999). The toxic effects of NODs have also been investigated in fish (Sotton et al., 2015). In the flatfish *Platichthys flesus*, NODs induced oxidative stress as indicated by a decrease of GST and CAT activity resulting in increased vulnerability of the cells to reactive oxygen species (ROS; Persson et al., 2009). NOD can also induce apoptosis and hyperphosphorylation of signaling proteins in cultured rat hepatocytes (Ufelmann and Schrenk, 2015). Nevertheless, not much toxicological data are available for NODs carcinogenicity in humans.

A cytotoxic alkaloid CYN can irreversibly inhibit the biosynthesis of protein and glutathione leading to cell death (Ohtani et al., 1992; Terao et al., 1994; Froscio et al., 2003). A *Cylindrospermopsis* bloom episode was found to cause cattle mortalities and human poisonings in north Queensland (Saker et al., 1999; Griffiths and Saker, 2003). Moreover, a number of disorders such as damage to liver, kidney, thymus, and heart, as well as hepatic and renal toxicity were observed in mice (Terao et al., 1994; Falconer et al., 1999; Bernard et al., 2003; Froscio et al., 2003). CYN may induce DNA strand breaks and possibly disrupt the kinetochore spindle, leading to chromosome loss, specifying its clastogenic and aneugenic action (Humpage et al., 2000). In primary rat hepatocytes, CYN has been shown to inhibit protein and glutathione synthesis and induce apoptosis (López-Alonso et al., 2013). Recently, Huguet et al. (2014) studied the effects of CYN on human intestinal Caco-2 cells and reported that CYN can modulate different biological functions by overexpressing the genes encoding proteins involved in DNA damage repair and transcription including modifications of nucleosomal histones. It has been shown that CYN may cause a decrease in glutathione synthesis (Runnegar et al., 1994) and induce oxidative stress in fish (Guzmán-Guillén et al., 2013a,b). Indeed, CYN can accumulate in various organs of fish, leading to deleterious effects on their normal physiology and biochemistry (Sotton et al., 2015). CYN may interfere with the basic functions of fish phagocytic cells and as a consequence, influence the fish immunity (Sieroslawska and Rymuszka, 2015).

A number of neurotoxic alkaloids from cyanobacteria have been reported, exerting their action on the neuromuscular system by blocking skeletal and respiratory muscles leading to respiratory failure. The cyanotoxins such as anatoxins, saxitoxins, antillatoxin, kalkitoxin, and jamaicamide are major groups of neurotoxic compounds (Aráoz et al., 2010). It has been established that anatoxin-a is a potent depolarizing neuromuscular blocking agent which acts by binding to nicotinic receptors for acetylcholine in the central nervous system (CNS), peripheral nervous system (PNS) and in neuromuscular junctions (Carmichael, 1998). Several studies regarding the mechanisms of anatoxins toxicity were performed in mice using the sub-lethal or lethal doses of anatoxin-a. Anatoxin-a, wellknown as a "Very Fast Death Factor," can cause contraction, muscular paralysis, and respiratory arrest leading to death of mice in a very short time after intraperitoneal injection (i.p. mouse LD50: 250 to 375 μg/kg; Devlin et al., 1977). Anatoxin-a can impair blood pressure, heart rate and gas exchange triggering hypoxia, respiratory arrest and severe acidosis leading to death of the animals (Adeymo and Sirén, 1992). The toxicological properties of homoanatoxin-a are more or less identical to those of anatoxin-a (Namikoshi et al., 2003). The neurotoxic alkaloid saxitoxins are considered the most toxic compounds. The mode of action of all analogs of saxitoxins is more or less similar; however, they differ in toxicity (Funari and Testai, 2008). Saxitoxins may block voltage-gated sodium channels in nerve cells and discontinue the entry of sodium flow, preventing the generation of a proper action potential or electrical transmission in nerves and muscle fibers leading to paralysis of muscles and death by respiratory arrest in mammals (Strichartz et al., 1986; Su et al., 2004; Bricelj et al., 2005). Another neurotoxic cyanotoxin antillatoxin is a novel ichthyotoxic (LC50 = 0.1 μM) cyclic lipopeptide isolated from the marine cyanobacterium *Lyngbya majuscula* (Orjala et al., 1995). Antillatoxin-A prompted a rapid neuronal death in cerebellar granule cell cultures (LC50 = 0.18 μM; Berman et al., 1999). Voltage-gated sodium channels were shown as the main molecular target of antillatoxin (Li et al., 2001). The neurotoxic compound kalkitoxin isolated from *L. majuscula* is a thiazoline-containing lipopeptide compound (Wu et al., 2000). Lyngbyatoxin-A, a cyclic dipeptide found in *L. majuscula,* appears to have been responsible for a severe oral and gastrointestinal inflammation suffered by a person who accidentally ingested this cyanobacterium (Sims and Zandee Van Rillaud, 1981). Kalkitoxin was shown ichthyotoxic to the goldfish *Carassius auratus* and toxic to the aquatic crustacean brine shrimp (*Artemia salina*) with an LC50 700 and 170 nM, respectively (Wu et al., 2000). Kalkitoxin may also block voltage-gated sodium channels (LePage et al., 2005). The neurotoxic amino acid BMAA acts in mammals as a glutamate agonist (Corbel et al., 2014). BMAA increases the intracellular concentration of calcium in neurons and induces neuronal activity by hyperexcitation (Brownson et al., 2002).

The endotoxic LPSs are known to cause fever in mammals and are involved in septic shock syndrome and liver injury (Choi and Kim, 1998). LPS can impair the immune system and also affect the detoxification system of diverse organisms (Wiegand and Pflugmacher, 2005). Until now, very little is known about the LPS intoxication and its toxicity is assumed to be associated with the host-mediated factors (Stewart et al., 2006a,b). More extensive research is needed to clarify a definite toxicity mechanism of LPS. Overall, it is no doubt that the acute effects of several cyanotoxins represent the major concern for ecological health impacts.

# CYANOBLOOMS AND CYANOTOXINS: MITIGATION STRATEGIES

The increased incidence of toxic cyanobacterial blooms (cyanoblooms) worldwide and their potential health risks have generated tremendous concern for dynamic management of toxic cyanoblooms. The economic cost of freshwater blooms in the United States was estimated to be about 2.2–4.6 billion dollars/annum (Dodds et al., 2009). Henceforth, advanced approaches or development of a new technology is needed to terminate or prevent/suppress the harmful cyanobacterial blooms for environmental sustainability and economic vitality (Hudnell, 2008, 2010; Srivastava et al., 2013; Harris et al., 2014; Koreivienë et al., 2014). Several factors boosting the incidence of harmful cyanobacterial blooms, such as nutrient input, wind velocity, sediment deposition, reduced water flow, increased salinity and temperature gradients, global warming and drought can be regulated to a certain extent to eliminate or minimize the bloom incidence. The approaches implemented for bloom suppression should be environmentally sustainable without adversely influencing the aquatic ecosystems. A number of strategies or approaches such as chemical, physical, biological, and other cognizance approaches came into consideration for mitigating the harmful cyanobacterial bloom incidences.

# CHEMICAL APPROACHES

Cyanoblooms can be controlled to a certain extent using some chemicals such as algicides, inhibitors or flocculants; however, use of these chemicals can inevitably recontaminate water bodies (Murray-Gulde et al., 2002; Van Hullebusch et al., 2002; Jancula ˇ and Maršálek, 2011). The use of certain pigments (aquashade) can reduce the amount of light availability, and inhibit the growth of harmful algae; however, this approach may not be effective due to growth inhibition of other beneficial microalgae, thereby undesirably influencing the aquatic ecosystems (Spencer, 1984). The use of some algicides has been reported to decline the bloom formation. The natural product cyanobacterin has been shown to be toxic to most cyanobacteria at a concentration of approximately 5 μM (Gleason and Baxa, 1986). Many biologically derived (but non-antibiotic) bioactive substances are known to inhibit the growth of aquatic bloom-forming cyanobacteria (Shao et al., 2013). Recently, Dai et al. (2012) have shown the fast removal (up to 98.99%) of MC-LR by a low-cytotoxic microgel-Fe(III) complex. Preoxidation with chlorine dioxide followed by flocculation and settling was found effective in removing cyanobacterial blooms and MCs (Bogialli et al., 2013). The use of aluminum salts can be used as algicides for nuisance algae and cyanobacteria control (Lelkova et al., 2008). The use of slaked lime [Ca(OH)2] or calcite (CaCO3) has also been reported to remove the algal communities, including cyanobacteria (Prepas et al., 2001; Zhang et al., 2001). Aluminum compounds can be used to remove the nutrients from industrial and domestic wastewaters (Auvray et al., 2006; Rodriguez et al., 2008; De Julio et al., 2010). Besides aluminum, several other metals such as iron and copper are used to remove the algal blooms. The salt of copper (CuSO4.5H2O) is widely used as an algicide (Murray-Gulde et al., 2002). The herbicide diuron together with copper sulfate as well as other copper-based compounds have been approved by the United States Environmental Protection Agency (USEPA) for use as algicides in fish production ponds (Schrader et al., 2004). Moreover, the use of synthetic compounds for bloom control has their own limitations, and therefore, a range of natural chemicals (e.g., anthraquinone, nostocarboline, and stilbenes) from diverse organisms have been derived as potent substituents of synthetic algicides (Schrader et al., 2003, 2004; Becher et al., 2005; Mizuno et al., 2008; **Table 2**). Recently, Jancula and Maršálek (2011) ˇ reviewed the availability of different chemical compounds for prevention and management of cyanobacterial blooms.

## PHYSICAL APPROACHES

Bloom control by physical methods generally involves mechanical removal techniques or short wavelength radiation treatment to control the incidence of cyanobacteria. The use of new and improved technologies can eliminate industrial/agricultural/ household pollutants to a certain extent to minimize the environmental pollution, including the water pollution by the incidence of harmful algal blooms. Global climate change and rising fresh water demand for multipurpose usage caused a remarkable increase in drought frequency and decreased freshwater flow rates (Paerl and Huisman, 2008; Paul, 2008). However, increasing flow rates and decreasing water residence time can remove fresh water algal blooms of a reservoir even in nutrient-rich conditions (Paerl, 2008). The artificial circulation for increased water flow is reported to suppress the blooms, but it may also cause habitat disturbance (Visser and Ibelings, 1996; Jungo et al., 2001; Huisman et al., 2004; Hudnell et al., 2010). Moreover, a solar powered circulation (SPC) has been designed to create long-distances circulation of the epilimnion (*>*200 m) to suppress freshwater harmful algal blooms (Hudnell et al., 2010). Data obtained from a case study of nutrient-enriched, source-water reservoirs, revealed the role of SPC in reduction of cyanobacterial peak density by about 82 and 95% during the first and second year of SPC deployment, respectively (Hudnell et al., 2010). Intensity of light and temperature play a significant role in bloom incidence as mentioned above. However, the increase in incidence of light and temperature can hardly be controlled in a large water reservoir, where as it is energy intensive in smaller water bodies. Short wavelength ultraviolet radiation can bring about a rapid degradation of the cyanotoxins MCs (Tsuji et al., 1995; Kaya and Sano, 1998). Moreover, it has been concluded that photosensitized processes may play an important role in the photochemical transformation of cyanotoxins (e.g., MC-LR; cylindrospermopsin) in the natural water (Lawton et al., 1999; Song et al., 2007; Wörmer et al., 2010; He et al., 2012). Simulated waterfalls or fountains may also be effective to control the cyanobacterial blooms in smaller water bodies; however, it requires electric-grid power constantly (Clevely and Wooster, 2007). The use of hydraulic jet cavitation may be a good approach to cyanobacterial water-bloom management (Jancula et al., 2014 ˇ ). Moreover, cavitation treatment can disintegrate gas vesicles of cyanobacterial cells, and can remove up to 99% cyanobacteria growing in a lake, ponds or reservoirs (Jancula et al., 2014 ˇ ).

# BIOLOGICAL APPROACHES

Control of cyanoblooms through biological mechanisms such as regulation of nutrient uptake or availability, alteration of normal physiology (such as a decrease in photosynthetic pigment), and/or direct feeding of cyanobacterial biomass by some aquatic organisms may be promising ways of ecological restoration (Bond and Lake, 2003; Qin et al., 2006; Zhang et al., 2008; Zhang et al., 2012, 2014). The gastropod *Radix swinhoei* can ingest cyanobacteria and survive well without loss in fecundity in the water reservoirs with cyanobacterial blooms (Zhang et al., 2012). The combined use of snails (*R. swinhoei*) and a submerged plant (*Potamogeton lucens*) in eutrophic waters can eliminate cyanobacterial bloom by minimizing the eutrophication; however, this method is under the preliminary stage due to the lack of the field study (Zhang et al., 2014). Occurrence and growth of aquatic plants are considered good candidates for limiting algal growth as the aquatic plants directly compete with algae for nutrients, light and space (Qiu et al., 2001; Wang et al., 2009). Some aquatic plants release different allelochemicals that can inhibit the growth of cyanobacteria and other phytoplanktons (Nakai et al., 2000; Körner and Nicklisch, 2002). Biodegradation using different species/strains of bacteria (**Table 3**) and other organisms may be the most efficient process to control the fate of some cyanotoxins in natural waters (Zhang et al., 2008; Manage et al., 2009; Lawton et al., 2011; Rastogi et al., 2014).

# RESEARCH AND MANAGEMENT

Development of wastewater research and management program is highly amenable to prevent or control the worldwide incidence of algal blooms and maintaining the ecological integrity and sustainability. Documentation of different environmental factors responsible for increased incidence of harmful cyanoblooms is crucial toward the development of demarcated management strategies. Moreover, interactive management of anthropogenic over nutrient-enrichment and global climate change is a major task for ensuring the protection and sustainability of aquatic ecosystems (Paerl et al., 2011a,b). The availability of phosphorus plays an important role in the growth of cyanobacteria and other microalgae or phytoplanktons (Schindler et al., 2008); henceforth, controlled input of phosphorus to the water reservoir may be an effective management strategy for

### TABLE 2 | Allelochemicals and their inhibitory effects against some bloom forming cyanobacteria.


*(Continued)*

### TABLE 2 | Continued


∗*Lowest-complete-inhibition concentration.*

restoration of aquatic ecosystems. In order to minimize the bloom boosting organic or inorganic nutrients coming from common practices such as excessive use of fertilizers (e.g., NPK) and detergents, prior wastewater treatment may be needed to reduce the incidence of cyanobacterial blooms (Conley et al., 2009; Paerl et al., 2011a; Jacquet et al., 2014). The modeling of different water bodies at risk of toxic blooms may be a good approach to develop a proactive algal bloom monitoring and management strategies (Tyler et al., 2009; Coad et al., 2014). Moreover, the fundamental research and quantitative ecological awareness toward the bloom incidence can be a supportive tool guiding large-scale water management against harmful bloom incidence.

### PUBLIC AWARENESS APPROACHES

Public environmental awareness (PEA) is a fundamental approach for the attainment of sustainable environment (Xu et al., 2013; Kirkpatrick et al., 2014). PEA about the incidence and harmful effects of toxic cyanoblooms may be a dynamic approach to eradicate and avoid the blooms and their toxic effects. The edifying approaches will allow people to think about their practices in their day-to-day life, such as unregulated disposal of organic/inorganic domestic wastes in the water reservoir, thereby enhancing the risks of bloom formation. As discussed elsewhere, global climate change may potentially impact the success of cyanobloom incidence. An emphasis on social practice to minimize the bloom formation and intoxication can allow intellectuals actualizing the significant development to control the environmental pollution. A change in the PEA levels in response to the increased incidence of environmental pollution is indispensable for ensuring the effective environmental protection and restoration (Xu et al., 2013). An increase in public awareness regarding the environmental sustainability and ecosystem health can inform the policy or decision makers to develop the strategies or to set-up the environmental protection laws against anthropogenic environmental pollution (such as direct disposal of domestic or industrial waste in open water reservoir such as rivers, ponds, lakes, and catchments). Moreover, various means of environmental protection program should be launched worldwide by the concerned government or non-government



organization (NGO) to spread the knowledge about different environmental issues such as harmful cyanobloom incidence (Mikami et al., 1995; Palmer et al., 1998; Li et al., 2008; Várkuti et al., 2008; Xu et al., 2013).

Overall, little is known concerning the formation of cyanoblooms and production of different variants of cyanotoxin in diverse water bodies. Furthermore, each of these strategies mentioned above has their own advantages and limitations, and more extensive collaborative work is needed to control or manage the occurrence of algal blooms worldwide. Since eutrophication is considered as the most immediate environmental consequence of cyanoblooms, the uncontrolled disposal of organic/inorganic nutrients in the water reservoir through agricultural runoff or through industrial and household sewage water must be diminished or even forbidden. The establishment of several dyes or chemical based industries are the source of several blooms forming substances and therefore the government law should strictly be implemented to sanitize unwanted industrial effluents before reaching the water bodies. Furthermore, the eutrophication of water reservoirs must be regularly checked for an increased prevalence of toxin producers mainly in the bloom sensitive areas of subtropical and temperate climates. Severity on global warming is also an important trigger that is likely to create toxic cyanoblooms, therefore a proper environmental management toward increasing global climate change is necessary for sustainability of the pollution-free aquatic ecosystems.

# CONCLUSION AND PERSPECTIVE

Cyanobacterial blooms are an increasing issue in both the wastewater-treatment and drinking water systems. Eutrophication and global climate change is the key factors for the occurrence of cyanoblooms all over the world. Cyanoblooms and production of several cyanotoxins in water bodies may reduce the surface/drinking water quality leading to high health risk to the organisms in aquatic ecosystems as well as wild/domestic animals and humans. A number of cyanotoxins such as MCs, nodularins, cylindrospermopsins, anatoxins, saxitoxins, and LPSs have been recognized as the major environmental contaminants in the immediate aquatic ecosystems. Control of cyanobloom using the chemical approaches can be effective; however, some algicidal/herbicides chemicals can cause secondary pollution of aquatic ecosystems. Several mitigation strategies have been tested and employed at laboratory levels; however, their efficacy to remove the blooms has not been confirmed under field environments. Establishment of effective mitigation strategies such as chemical, biological as well as public cognizance approach toward environmental awareness may be the most realistic measure to overcome the worldwide incidence of algal blooms and the attainment of a sustainable environment. Some natural algicidal compounds are really very effective to control the cyanoblooms; however, their production and availability is still very limited. The costeffective synthesis of these biochemicals would be highly valuable to control the cyanoblooms. Furthermore, several cyanobacteria may become resistant toward certain chemicals. The use of biocides or several different biological processes against cyanoblooms may also affect other non-target aquatic organisms. Hence, common ecotoxicological impacts should be sensibly evaluated in the milieu of the lack of ecological health risk assessment. Moreover, a combined policy should strictly be regulated to diminish the bloom-boosting cause such as massive eutrophication of aquatic ecosystems by anthropogenic sources.

## ACKNOWLEDGMENTS

RR is thankful to the University Grant Commission (UGC), New Delhi, India, for Dr. D. S. Kothari Postdoctoral Research Grant. AI thanks Chulalongkorn University and Thailand Research

# REFERENCES


Fund for financial support through Ratchadaphiseksomphote Endowment Fund (Food and Water cluster) and for research grant (IRG 5780008), respectively. We also thank Professor Peter Lindblad for critical reading and English language editing.

*rubescens*) affecting an Italian drinking water basin: a case study. *Environ. Sci. Technol.* 47, 574–583. doi: 10.1021/es302260p


members of a new class of tumor promoters. *Proc. Natl. Acad. Sci. U.S.A.* 78, 3872–3876. doi: 10.1073/pnas.78.6.3872


is an antineoplastic agent rather than a tumor promoter: development of a synthetically accessible protein kinase C activator with bryostatin-like activity. *J. Am. Chem. Soc.* 131, 7573–7579. doi: 10.1021/ja808447r


*Anabaena* sp. strain 37 and molecular methods to detect potential producers. *Appl. Environ. Microbiol.* 77, 7271–7278. doi: 10.1128/AEM.06022-11


channels. *Ann. N. Y. Acad. Sci.* 479, 96–112. doi: 10.1111/j.1749-6632.1986.tb 15564.x


blooms in eutrophic and hypertrophic lakes. *PLoS ONE* 9:e104325. doi: 10.1371/journal.pone.0104325


with potential risks to human health. *Sci. Total Environ.* 407, 2191–2199. doi: 10.1016/j.scitotenv.2008.12.039


**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 Rastogi, Madamwar and Incharoensakdi. 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.*

# Patterns of benthic bacterial diversity in coastal areas contaminated by heavy metals, polycyclic aromatic hydrocarbons (PAHs) and polychlorinated biphenyls (PCBs)

*Grazia Marina Quero, Daniele Cassin, Margherita Botter, Laura Perini and Gian Marco Luna\**

### *Edited by:*

*Federico Lauro, University of New South Wales, Australia*

### *Reviewed by:*

*Matthew Lee, University of New South Wales, Australia Sara Sjoling, Södertörn University, Sweden*

### *\*Correspondence:*

*Gian Marco Luna, National Research Council-Institute of Marine Sciences (CNR-ISMAR), Castello 2737/f, Arsenale Tesa 104, 30122 Venezia, Italy gianmarco.luna@ve.ismar.cnr.it*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 29 May 2015 Accepted: 14 September 2015 Published: 13 October 2015*

### *Citation:*

*Quero GM, Cassin D, Botter M, Perini L and Luna GM (2015) Patterns of benthic bacterial diversity in coastal areas contaminated by heavy metals, polycyclic aromatic hydrocarbons (PAHs) and polychlorinated biphenyls (PCBs). Front. Microbiol. 6:1053. doi: 10.3389/fmicb.2015.01053* *National Research Council-Institute of Marine Sciences (CNR-ISMAR), Venezia, Italy*

Prokaryotes in coastal sediments are fundamental players in the ecosystem functioning and regulate processes relevant in the global biogeochemical cycles. Nevertheless, knowledge on benthic microbial diversity patterns across spatial scales, or as function to anthropogenic influence, is still limited. We investigated the microbial diversity in two of the most chemically polluted sites along the coast of Italy. One site is the Po River Prodelta (Northern Adriatic Sea), which receives contaminant discharge from one of the largest rivers in Europe. The other site, the Mar Piccolo of Taranto (Ionian Sea), is a chronically polluted area due to steel production plants, oil refineries, and intense maritime traffic. We collected sediments from 30 stations along gradients of contamination, and studied prokaryotic diversity using Illumina sequencing of amplicons of a 16S rDNA gene fragment. The main sediment variables and the concentration of eleven metals, polychlorinated biphenyls (PCBs) and polycyclic aromatic hydrocarbons (PAHs) were measured. Chemical analyses confirmed the high contamination in both sites, with concentrations of PCBs particularly high and often exceeding the sediment guidelines. The analysis of more than 3 millions 16S rDNA sequences showed that richness decreased with higher contamination levels. Multivariate analyses showed that contaminants significantly shaped community composition. Assemblages differed significantly between the two sites, but showed wide within-site variations related with spatial gradients in the chemical contamination, and the presence of a core set of OTUs shared by the two geographically distant sites. A larger importance of PCBdegrading taxa was observed in the Mar Piccolo, suggesting their potential selection in this historically polluted site. Our results indicate that sediment contamination by multiple contaminants significantly alter benthic prokaryotic diversity in coastal areas, and suggests considering the potential contribution of the resident microbes to contaminant bioremediation actions.

Keywords: microbial diversity, chemical pollution, PCBs, marine sediments, next generation sequencing

## Introduction

Coastal marine ecosystems are amongst the most productive and diverse on Earth, providing over US\$ 14 trillion worth of ecosystem goods (Harley et al., 2006). The human impact is altering the coastal sea functioning under the consequence of a plethora of pressures, including chemical pollution and wastewater discharge, eutrophication, hypoxia, utilization of living resources (e.g., over-fishing), habitat destruction, and climate change effects. The diversity and functioning of coastal ecosystems are largely affected from chemical pollution by a plethora of compounds, including those identified as emergent (Elliott and Elliott, 2013). Contaminants accumulate in marine and freshwater environments, and determine the reduction of biodiversity, by means of adverse effects to the resident biota, the removal of sensitive species and the selection of the more tolerant ones (Johnston and Roberts, 2009). The human impact in the coastal ocean is becoming evident also at the microbial level (Paerl et al., 2003), with the more obvious effects in terms of inputs of autochthonous and pathogenic microbes (spreading diseases to human and marine populations; Stewart et al., 2008), shifts in community composition, and impairment of ecological functions (Nogales et al., 2011). Marine microbes are extremely sensitive to environmental changes because of their small size, fast growth rates and genome plasticity. Consequently, the study of microbial community diversity, and their fluctuations over spatial and temporal scales, represents a useful tool to evaluate the consequences of the anthropogenic perturbation on marine ecosystem health (Ager et al., 2010).

Among chemical contaminants which are recovered in the coastal environment, heavy metals, polychlorinated biphenyls (PCBs) and polycyclic aromatic hydrocarbons (PAHs) represent some of the most ubiquitous and spread (Förstner and Wittmann, 1983; Fernandes et al., 1997; Field and Sierra-Alvarez, 2008), and occur primarily as a result of anthropogenic inputs. Heavy metals are highly persistent and may exert toxic effects at all levels of biological organization, from cells to population, and community structure, by altering enzymatic functioning and metabolic pathways (Chapman et al., 1998). PCBs are stable organic molecules that have been largely used, for more than 50 years, for a wide number of industrial applications because of their convenient physical and chemical properties. Once in the environment, PCBs accumulate along the food web and can exert multiple adverse effects for marine life and human health, even under low environmental concentrations (Borja et al., 2005; Carpenter, 2006). Finally, PAHs are aromatic compounds produced mainly during combustion in natural and anthropogenic processes, and are typically found in high concentrations on industrial sites, particularly those associated with the petroleum, gas-production, and wood-preserving industries (Wilson and Jones, 1993). They cause concern as environmental pollutants because some are carcinogens and mutagens, and have a high potential for biomagnification through the aquatic food web due to their lipophilic nature. All these types of contaminants accumulate in coastal marine sediments representing their main depository, where they can exert a significant influence on the benthic biota (Rocchetti et al., 2012).

A number of studies have been performed to shed light on the influence of chemical contaminants on microbial communities in coastal marine sediments. High concentrations of heavy metals have been shown to inhibit benthic bacterial metabolism and turnover (Dell'Anno et al., 2003), as well to influence the diversity of resident microbes and to select for those more tolerant to metals (Gillan et al., 2005). Nonetheless, certain aquatic microbes play a role into the fate and cycling of toxic metals (Ford and Ryan, 1995), as well in the biodegradation of PCBs(Bedard, 2008; Field and Sierra-Alvarez, 2008; Pieper and Seeger, 2008), and a variety of bacteria are capable of transforming and mineralizing PAHs (Kanaly and Harayama, 2000). Studies to investigate relationships between contaminants and bacteria reported lower diversity and richness in polluted sediments, and dominance of few OTUs (Sun et al., 2012), likely resulting from selective advantage of those microbes able to metabolize contaminants. Jonhston and Leff (2015) reported that bacterial community composition was strongly influenced by PAHs in riverbank sediments. The large majority of studies in coastal sediments have been performed with the aim of isolating and characterizing pollutant-degrading bacteria, and to evaluate the potential for sediment remediation. Conversely, only few studies addressed the potentially combined effects of multiple contaminants, such as heavy metals, PAHs and PCBs together, on the diversity and spatial patterns of complex sediment communities (Sun et al., 2012; Brito et al., 2015), and even less studies have used to this purpose the recently developed next generation sequencing (NGS) techniques (Sun et al., 2013; Korlevic et al., 2015 ´ ), which allow detailed characterizations of the rare and dominant taxa within assemblages. Consequently, our study has potential to widen significantly our limited understanding of the bacterial community response to sediment pollution in coastal areas.

The Po River Prodelta (Northern Adriatic Sea) and the Mar Piccolo of Taranto (Ionian Sea) are two contaminated sites, located in the northern and southern coast of Italy, respectively. The Po River Prodelta receives significant contaminant discharge from one of the largest rivers in Europe (Boldrin et al., 2005). The delta system has a daily mean discharge of 1500 m3/s (ranging from 100 m3/s to 11550 m3/s) and produces a freshwater plume able to influence the whole Adriatic Sea (Falcieri et al., 2014), including the functioning of sediment microbes (Manini et al., 2004). The Po typically experiences major floods, during which transport and depositional processes lead large amounts of suspended sediments and associated contaminants, which can be stored or transported offshore to the adjacent areas (Correggiari et al., 2005). Many studies have shown that this site is severely chemically polluted, as the river carries yearly tons of anthropogenic chemicals collected from the entire Po valley and the river tributaries (Viganò et al., 2003). The Mar Piccolo of Taranto (Ionian Sea) is an inner, semi-enclosed basin which communicates with the adjacent Ionian Sea through two channels (Petronio et al., 2012). It is a chronically polluted area due to the presence, since decades, of the largest steel production plant in Europe. It also hosts a variety of other sources of pollutants, among which oil refineries, a large naval base (including a military ship-yard) and intense maritime traffic, and has received in the past considerable amounts of sewage from several pipes discharges (Cardellicchio et al., 1997). This chronical contamination is now evident in terms of PCBs, PAHs, and heavy metal concentrations in the sediments (Cardellicchio et al., 1997; Petronio et al., 2012). Knowledge on the biodiversity of benthic bacteria in the two study sites is today scant or non-existent, being limited only to the Mar Piccolo site where cultivation-based studies have described fecal bacteria and the biodiversity of culturable microbes only (Cavallo et al., 1999; Zaccone et al., 2005). There is currently no information available obtained using the recent NGS techniques, consequently our understanding of the microbial processes and the potential of resident microbes for remediation actions is largely hampered.

In this study, we collected sediments from 30 stations, located along gradients of putative sediment contamination, in the two sites and we described in detail, by NGS of the 16S rRNA gene, the bacterial richness, diversity, and community composition. Diversity was studied as a function of the main environmental variables and the concentration of eleven heavy metals, PCBs and PAHs, to test the hypothesis that the chemical contamination influences the richness and community composition of benthic prokaryotes. This study is among the first to investigate, by mean of a large sequencing effort (consisting of millions of 16S rDNA sequences), the combined effects of multiple contaminants on benthic prokaryotes across different spatial scales. Moreover, hypotheses on the potential contribution of the resident microbes to contaminant bioremediation actions are given.

### Materials and Methods

### Sampling Sites and Activities

Surface sediments were collected in two sites located on the eastern coast of Italy: the Po River Prodelta (Northern Adriatic Sea) and the Mar Piccolo of Taranto (Ionian Sea). Sampling activities were performed in the periods between 10th and 14th June 2013 in the Po River Prodelta, and between 17th and 21th June in the Mar Piccolo of Taranto. The experimental design in the Po River Prodelta site included 19 stations, which were distributed along coast to open sea transects (**Figure 1A**), at depths comprised between 9–21 mt. The sampling transects were placed in front of the outlets (Busa di Tramontana, Busa Dritta, Busa di Scirocco) of the main branches of the river delta (Po di Maistra, Po della Pila, Po di Tolle, Po della Gnocca, Po di Goro). The distribution of stations was chosen in order to follow the possible deposition of material transported in the area following an exceptional flood event occurred in the third week of May 2013 with a maximum of flow rate measured in Pontelagoscuro (FE) equal to 6830 m3/sec. The Mar Piccolo of Taranto (**Figure 1B**) displays a restricted circulation and extends for a total surface area of 20.7 km2. It is structured into two inlets, the "First Inlet" (having the maximum depth of 13 meters) and the "Second Inlet" (maximum depth 8 meters; Cardellicchio et al., 2007). The Mar Piccolo is connected with the adjacent Mar Grande through two channels, termed "Navigabile" and "Porta Napoli". In terms of the hydrographic characteristics, this

site can be compared to a brackish lake. Salinity is influenced by the input of freshwater originating by small tributary rivers and freshwater springs called "Citri" (Cavallo et al., 1999). The sampling design in this area included 11 stations, eight of which located in the First and three in the Second inlet, at depths comprised between 6 and 12 mt. One station (TA1) was located immediately in front of the channel linking the Mar Piccolo with the adjacent open sea. Basing on CTD measurements in the water column (full data are not shown), the Po River Prodelta showed a more marked influence of the river flow regime on the hydrological characteristics, evident in the wide range of temperature (16.4–24.9◦C) and salinity values (6.10–37.56 PSU). Conversely, the Mar Piccolo of Taranto showed more stable hydrological conditions, with temperature and salinity values in the range 21.0–23.5◦C and 36.0–38.4 PSU (respectively), highlighting poor fresh water inputs during the sampling period. Detailed geographical coordinates and sampling depths of all stations in the two sites are reported in the Supplementary Tables S1 and S2. Samples were collected in triplicate using a Van Veen grab sampler (capacity 10 L) onboard small research vessels. Only in the site of Taranto, the sediments for chemical analysis were collected with a gravity corer (model SW-104). Once onboard, the uppermost 0–2 cm layer of each sediment sample was immediately processed according to the specific protocols required for each type of analysis. For the analysis of organic pollutants an aliquot of the sediment was put in a hexane-rinsed aluminum foil and stored at −20◦ C until analysis. For heavy metals, organic matter content and grain size determinations, the sediments were put in a PET jar and stored at 4◦C until analysis. With regard to the sediment cores, they were kept vertical at the temperature of 4◦C until being processed into different sub-samples. For microbiological analyses, the sediment was immediately placed, using sterile spatulas, under sterile containers for their immediate transport at 4◦C to the laboratory, where the samples were then stored at −20◦C until molecular analyses of biodiversity.

### Environmental Variables and Chemical Contaminants

The organic matter content was estimated by the method based on loss on ignition (LOI) at 550◦C for 2 h. The grain-size distribution was measured by means of a laser beam analyser (Microtrac mod. X-100, Leeds and Northrup, USA). For metals analysis, each sediment sample was gently squeezed to break down aggregates, screened through a sieve with a mesh of 1 mm and the dried sediment was ground to power using an agate mortar. The sample (about 0.4 g d.w.) was then digested with 8 ml HNO3 in a microwave oven (Multiwave 3000, Anton Paar, Austria). The digested was left to cool at room temperature and then filtered through a 0.45 μm nitrocellulose membrane filter. The filtered digestates were diluted with distilled deionised water to 40 ml in a volumetric flask (U.S. Environment Protection Agency of United States of America, 1994a). The concentrations of the metals (Al, As, Cd, Cr, Cu, Fe, Mn, Ni, Pb, and Zn) and total P content were determined by inductively coupled plasma atomic emission spectrometry (ICP-AES) (Optima 2100DV, Perkin Elmer, USA; U.S. Environment Protection Agency of United States of America, 1994b). Mercury analyses were carried out by atomic absorption spectrophotometry by cold vapor (Analyst 100, Perkin Elmer, USA; U.S. Environment Protection Agency of United States of America, 1976). The precision of instrumental analysis was checked by control chart of each heavy metal. Data quality was monitored using 10% procedure blanks and 10% sample replicates. Recovery, checked analyzing a certified reference material for heavy metals (BCR-277r estuarine sediment, Community Bureau of Reference). For organic pollutants, samples were air dried in the dark at room temperature for 48 h on hexane-rinsed aluminum foil and then the dry samples were finely ground in a agate mortar. The extraction of about 2 g of sediment was performed using a Microwave Sample Preparation System, in accordance with the EPA recommendation (U.S. Environment Protection Agency of United States of America, 2007) with a 25 ml 1:1 acetone/hexane solvent mixture. The samples were concentrated in a rotating evaporator (Rotavapor-R Buchi, CH), and the sulfur compounds were removed by soaking the extracts with activated copper powder. Purification and fractionation were performed by eluting extracts through chromatography glass columns packed with Silica gel/Alumina/Florisil (4 + 4 + 1 g). The first fraction, containing PCBs, was eluted with 25 ml of n-hexane, whereas the second fraction, containing the PAHs, was eluted with 30 ml of 8:2 n-hexane/methylene chloride solvent mixture (Fossato et al., 1996, 1998). The concentration of 16 USEPA priority pollutant PAHs were analyzed with high performance liquid chromatograph (PE 200, USA), coupled to a programmed fluorescence detector (HP 1046A, USA). The column used was a reverse-phase Supelcosil LC-PAH (*L* = 150 mm, *f* = 3 cm, 5 μm). PCBs (32 congeners) were analyzed by gas chromatography/mass spectrometry (GC/MS). The system consist of an Agilent 7820A GC coupled with an Agilent 5977E Series GC/MS, and the software MassHunter for data analysis. The GC is equipped with a 30 m HP-5MS capillary column (0.25 mm ID, 0.25 μm film). The identification of PAHs and PCBs was based on matching retention time, and the quantification was determined from calibration curves established for each compound by analyzing five external standards. The method detection limits, measured using the calibration curve method, ranged between 0.05 and 0.1 ng g−<sup>1</sup> for PAHs, and 0.05 ng g−<sup>1</sup> for PCBs. Blanks were run for the entire procedure. Validation of the recovery and accuracy was carried out with IAEA-417 and IAEA-159 sediment sample certified reference materials.

### Bacterial Diversity Analyses using Illumina Sequencing

DNA was extracted from 1 g of each sediment sample using the PowerSoil<sup>R</sup> DNA Isolation Kit (MoBio Laboratories Inc., California), according to the manufacturer's instructions with some slight modifications to increase the DNA yield and quality. These modifications included two additional vortexing steps (following the one which is recommended by the manufacturer) at the maximum speed for 2 min, each one being preceded by an incubation at 70◦C for 5 min, and by adding one more washing step with Solution C5 as an additional removal step for contaminants. The concentration of each DNA extract was determined spectrophotometrically, and the DNA was stored at −80◦C until PCR. Illumina Miseq V3 sequencing were carried out on the hypervariable V3 and V4 regions of the 16S rRNA gene by amplifying using the 341F (5 -CCTACGGGNGGCWG CAG-3 ) and 785R (5 -GACTACHVGGGTATCTAATCC-3 ) universal bacterial primers (Eiler et al., 2012). Paired-end reads were quality checked (with default settings and minimum quality score of 20) and analyzed with QIIME v1.8.0 software package (Quantitative Insights Into Microbial Ecology) (Caporaso et al., 2010). Reads were clustered into OTUs by using UCLUST v1.2.22 (Edgar, 2010) with a *>*97% similarity threshold with a openreference OTU picking strategy and default settings. Chimeras were detected by using USEARCH v6.1 (Edgar, 2010). Chimera checking and taxonomy assignment was performed using Greengenes 13.8 as reference database (De Santis et al., 2006). Abundances in each sample were normalized on the number of sequences of sample with the lowest number of reads retained. The sequences have been submitted to the SRA -Sequence Read Archive (accession number SRP061637).

### Data Handling and Statistical Analyses

The distribution maps which show the concentrations of some contaminants were produced as a contour plot based on geographic information system (GIS) technology. The software used was QGIS and the interpolation was carried out with inverse distance weighted method (IDW) with power parameter equal to 2. The Spearman-Rank correlation analysis was performed to test linear relationships between some of the microbiological variables and the concentration of environmental and chemical contaminants. Correlation coefficients (*r*) were considered significant at *p*-values less than 0.05. Differences in the community composition between the two sites were assessed, on the Illumina dataset, using the analysis of similarity (ANOSIM) tool based on a Bray–Curtis similarity matrix. The presence of statistical differences between samples is indicated by a significance level at *p*-values less than 0.05. Similarity was calculated by performing an UPGMA clustering based on unweighted Unifrac distance matrix (Lozupone and Knight, 2005). A PCOa (Principal Coordinates analysis) was performed on the environmental variables and chemical contaminants to explore and visualize similarities among the two sites. This analysis was performed basing on normalized data and using a Euclidean distance matrix. Multivariate, multiple regression analyses were performed to identify drivers of bacterial community composition in the investigated samples. The analysis was performed using the Distance-based linear modeling (DistLM) analysis on the Illumina-based resemblance matrix (Bray–Curtis similarity) at either phylum and OTUs level and including the main environmental variables (total P content, % LOI, and silt) and chemical contaminants (Al, As, Cd, Cr, Cu, Fe, Hg, Mn, Ni, PAHs, Pb, PCBs, and Zn, or a selection of them as later specified) as predictor variables. Prior of each DistLM analysis, the values of environmental variables and chemical contaminants were normalized. We used, as selection procedure, the option "All specified," and AICc as the selection criterion. A distance-based redundancy analysis (dbRDA) plot was prepared using all the tested variables. The ANOSIM, PCOa, DistLM, and dbRDA analyses were performed using the PRIMER 6 + software (http://www*.*primer-e*.*com/).

## Results

### Analysis of Environmental Variables and Contaminants

The concentration of all the environmental variables and chemical contaminants in the two sites is reported in Supplementary Table S1 (Po River Prodelta) and Supplementary Table S2 (Mar Piccolo), while the spatial distribution of selected contaminants in the two sites is shown in **Figures 2A–F**. Furthermore, the concentration of the measured PAHs and PCBs congeners is reported in Supplementary Tables S3 and S4. Surface sediments appeared as mainly constituted by fine materials, but differences were observed between the two sites. In the Po River Prodelta, the percentage of silt, clay, and sand showed an average of 68, 18, and 14%, respectively. The silty fraction showed a very low variability (coefficient of variation CV = 6.2%), while the sandy component showed a wider range (CV = 52%). In this site, a decreasing trend in the sandy percentage from north to south was observed. Conversely, in the Mar Piccolo of Taranto, the sandy fraction was on average twice compared with the other site. Silt and clay accounted for 56 and 15%, respectively. More uniform textural features were evident, with CV of 18, 6, and 16% for sand, silt and clay, respectively. The station TA1 differed from all other stations, with a very high percentage of sand (79%) and a very low silt (17%) and clay (*<*5%) content. The analysis of chemical contaminants highlighted different levels of contamination in the two sites. The Po River Prodelta site showed an overall lower contamination level. Mercury concentration displayed very low concentrations, always below the current Italian regulatory limits (0.3 mg kg−<sup>1</sup> according to DL 260/2010). Chromium and nickel showed high concentrations (94 and 77 mg kg<sup>−</sup>1, respectively). PCBs concentrations always exceeded the same regulatory limits (which are set to 8 ng g<sup>−</sup>1). PAHs showed overall low concentrations (148 ng gr−<sup>1</sup> on average), well below the abovementioned limits (set to 800 ng gr<sup>−</sup>1), with the highest values observed at the stations PO10, PO11, and PO13. Among the measured PAHs congeners, fluoranthene and pyrene were those displaying, as average of all stations, the higher concentration. Among the PCBs congeners, the dominant were 138, 153, and 180. The Mar Piccolo of Taranto site showed much higher concentrations of contaminants, especially in terms of arsenic, copper, lead, zinc, mercury, PAHs, and PCBs. For the last three contaminants, the values resulted to be on average 24, 12, and 25 times higher, respectively, than in the Po River Prodelta. The dominant PAHs congeners were pyrene and benzo(g,h,j)perylene + indeno(1,2,3-cd)pyrene. Chemical analyses demonstrated the presence of a very large number of PCBs congeners. The most contaminated area was the one located in front of the arsenal of the Italian Navy (e.g., TA5), which showed very high concentration of several contaminants and particularly mercury and PCBs (9.0 mg kg−<sup>1</sup> and 1045 ng g<sup>−</sup>1, respectively). Chromium and nickel showed low values, while Fe, Al, Cd, and total P showed values comparable to those found in the other study site.

The Principal Coordinates analysis of the environmental variables and chemical contaminants showed a clear separation between the two sites (Supplementary Figure S1). This separation was evident also considering the only group of Mar Piccolo of Taranto stations, in which two separate subgroups can be identified according to the location of the stations in the first and second inlet. The station TA1 differed completely from all other stations, which reflects its peculiar textural characteristics (very high percentage of sand).

### Bacterial Diversity, Community Composition, and Core Microbiome

A cumulative number of 3,500,760 raw sequences was obtained for the 30 stations from the Illumina sequencing analyses. The average value was 116,692 reads per sample, with a minimum number of 30,119 in the station PO4 and the maximum number of 201,629 in the station PO10. The average length of sequences in the entire dataset was 418.9 ± 1.35 base pairs. After the quality check, the final dataset included 2,953,054 reads, with an average value of 98,453 per sample (minimum number of 25,354 at station PO4 and maximum number of 168,785 at station PO10), with an average length of 418.5 ± 12.41 base pairs.

FIGURE 2 | Spatial patterns of contaminants. Spatial patterns of selected chemical contaminants in the two sites. Shown are mercury (A,D), polychlorinated biphenyls (PCBs) (B,E) and polycyclic aromatic hydrocarbons (PAHs) (C,F) in the Po River Prodelta and the Mar Piccolo of Taranto, respectively.

Bacterial OTU richness in the two study sites ranged from 3,992 (PO1) to 7,640 (PO7) OTUs in the Po River Prodelta, and from 1,103 (TA12) to 6,752 (TA1) OTUs in the Mar Piccolo of Taranto (Supplementary Figure S2). Typically, richness displayed higher values in the Po River Prodelta (on average 5,898 OTUs) than in the Mar Piccolo samples (on average 3,809 OTUs).

Bacterial community composition showed that OTUs were affiliated to 81 known and 2 unknown phyla, and to 284 known and 64 unknown classes. The relative importance of the most abundant phyla and classes is shown in **Figure 3** (right). In all samples, the phylum Proteobacteria represented the most abundant (on average 47.6 and 48.6% in Po River Prodelta and Mar Piccolo, respectively). Within this phylum, the classes Delta- (average 16.9 and 22.6%, respectively) and Gammaproteobacteria (average 14.6 and 17.4%, respectively) were the two most frequently observed. Alphaproteobacteria accounted for 9.1 and 4.4% (respectively), while the other Proteobacteria (including Beta-, Epsilon-, Zeta-, and TA18) accounted for 6.9 and 4.1%, respectively. The second most abundant phylum was Bacteroidetes (average 26% in Po River Prodelta and 9% in Mar Piccolo). Among the other most dominant phyla, Planctomycetes accounted for 2.1–12.3 and 2.5– 9% (Po River Prodelta and Mar Piccolo, respectively), followed in importance by Firmicutes (range 0.9–10% in Po River Prodelta, and 1.5–20.9% in Mar Piccolo), and Chloroflexi (1.6–5.5% in Po River Prodelta, 4.5–7.9% in Mar Piccolo). Other phyla included Verrucomicrobia (average 5.4 and 1.5% in the two sites, respectively), Acidobacteria (average 2.7 and 4.8%, respectively), Actinobacteria (2.6 and 3.3%, respectively) and Spirochaetes (1.6 and 0.6%, respectively). Despite the primers we used are designed to target mostly Bacteria and do not provide a representative picture of the whole archaeal assemblages, it is worth mentioning that we found a low frequency of Euryarchaeota (0.7 and 1.8% respectively), especially in the Mar Piccolo samples. Finally, Tenericutes were detected infrequently (average value 1.3 and 0.04% in Po River Prodelta and Mar Piccolo, respectively). The percentage of "Unassigned and Other sequences" was 2.4% in the Prodelta River Po and 3.3% in the Mar Piccolo. The UPGMA clustering based on unweighted Unifrac distance revealed a clear separation between the two study areas (**Figure 3**, left). ANOSIM confirmed the presence of significant differences in community composition at the phylum level between the two sites (*r* = 0.704, *p <* 0.01).

The analyses at the OTU level indicated that the cumulative number of OTUs observed in the two sites was 56,917 OTUs, with a total number of 43,613 OTUs recorded in the Po River Prodelta, and the total number of 20,235 OTUs in the Mar Piccolo of Taranto. In the Po River Prodelta site, the most abundant OTUs (here defined as those accounting for at least *>*1% across all stations within the each site) were represented by one OTU belonging to the Helicobacteraceae family (class Epsilonproteobacteria) which displayed the average percentage of 3.05% of the total reads. This OTU was followed in importance by two OTUs within the Deltaproteobacteria class, the first belonging to the family Desulfuromonadaceae (average 1.62%) and the second to the family Desulfobulbaceae (1.15%), by a gammaproteobacterial OTUs in the family Piscirickettsiaceae (1.08%) and lastly by another OTU affiliated to the genus Lutimonas within the Bacteroidetes phylum (average 1.06%). In the Mar Piccolo site, the most abundant OTUs were one identified as belonging to the genus Thermoanaerobacter (phylum Firmicutes, average 2.96%), followed by one OTU in the Syntrophobacteraceae family (Deltaproteobacteria, 2.28%), the same OTUs observed in the Po River Prodelta site within the Helicobacteraceae family (1.59%) and the family Desulfobulbaceae (1.05%), and one OTU in the genus

Tissierella/Soehngenia (Firmicutes, 1.20%). These dominant OTUs represented altogether 7.96 and 9.08% of the total OTU abundance in the two sites, respectively. The remaining less abundant OTUs, contributing each for *<*1%, represented the largest fraction of the obtained sequences within each station and site (92.04 and 90.92% of the sequences in the two sites, respectively).

When the sites were compared at the OTU level, we found a core microbiome which comprised 33 OTUs, 31 of which belonged to Bacteria and 2 to Archaea (**Figure 4**). These OTUs were observed consistently at all stations of the two sites with at least one read. This core microbiome comprised some of the same OTUs previously identified as dominant, such as the same OTU within the Helicobacteraceae family observed at all stations, and the OTU belonging to the genus Thermoanaerobacter which was particularly abundant in the Mar Piccolo site. The remaining core microbiome OTUs were spread across 11 bacterial and archaeal classes.

### Relationships between Contaminants and Community Composition

OTU richness typically showed a decreasing pattern with increasing chemical contamination in the sediment. This pattern was reflected in the significant negative correlations observed between richness and the concentration of several contaminants, among which PCBs (*r* = 0.439, *p <* 0.05), PAHs (*r* = 0.439, *p <* 0.05), Hg (*r* = 0.397, *p <* 0.05), Cu (*r* = 0.507, *p <* 0.001), Pb (*r* = 0.487, *p <* 0.01), As (*r* = 0.408, *p <* 0.05), Zn (*r* = 0.616, *p <* 0.001), and Al (*r* = 0.397, *p <* 0.05). DistLM analyses performed using all the environmental and chemical variables, except for sand and clay (which showed redundancy with the "silt" variable and were thus not included in the analyses) showed that, at the phylum level, all the considered variables, with the only exception of Al, Cd, and Fe, were able to significantly explain the differences observed in the community composition at p *<* 0.001 significance (Supplementary Table S5), while total P explained significantly but at lower significance level (p *<* 0.05). A similar output of the distLM was observed at the OTU level (Supplementary Table S6). The cumulative percentage of variance explained by the set of environmental and chemical variables was 73.3 and 72.1% at the phylum and OTU level, respectively. A dbRDA analysis was used for the graphical visualization of the DistLM results. The dbRDA plot showed that, at both phylum and OTU level, the environmental and chemical parameters divided the stations into two separate clusters corresponding to the two study sites (**Figure 5**). The two axes of the phylum- and OTU-dbRDA explained 61.5 and 44.2% of the total variation, respectively.

According to the evident clustering between the two sites, we performed separate DistLM and dbRDA analyses at the OTU level to test the role of environmental variables and chemical contaminants in shaping community composition within each sampling site. In the Po River Prodelta, the DistLM analysis indicated a significant influence of several environmental variables and chemical contaminants, which together explained up to 89.8% of community variation. The dbRDA plot showed that the environmental and chemical parameters structured the stations into several separate clusters (**Figure 6**, left). These included groups of stations located along gradients characterized by different level of contamination, among which one group made up by the stations PO1 and PO2 (which were best correlated to % LOI, *r* = −0.514), a second one containing the stations from PO3 to PO7, which were best structured by Cu (*r* = −0.445), Cd (*r* = 0.472), and PCB (*r* = −0.587), a third group containing the stations PO8, PO16, PO17, and PO18, best related to Cu (*r* = −0.445), Cr (*r* = −0.251), and Al (*r* = −0.123; Supplementary Table S7). The plot showed an outlier station which was mostly related to Fe (PO9, *r* = 0.632), another group including two stations (PO10 and PO12) which were best related to Mn (*r* = 0.167) and silt (*r* = 0.203) and finally a large cluster comprising the remaining five stations, mostly structured by Hg (*r* = 0.453), As (*r* = −0.262), and PAHs (*r* = 0.184).

In the Mar Piccolo site, the DistLM analysis similarly showed a significant influence for certain environmental variables and contaminants. The dbRDA plot divided the stations into three major groups (**Figure 6**, right), one containing the stations TA9, TA10, TA11 and TA12 (which were best correlated to % LOI, *r* = −0.466), another one including TA7, TA8, and TA14 (best correlated with Zn, *r* = 0.301) and a third group of stations including TA4 and TA5 (best related with Hg, *r* = −0.518; Supplementary Table S8). The remaining stations appear to be scattered on the plot and differently influenced by other variables.

### Discussion

Coastal marine sediments represent biogeochemically relevant areas of the global ocean, where prokaryotes are recognized to play a significant role (Luna et al., 2002; Gobet et al., 2012). Chemical pollutants, such as heavy metals and xenobiotics, can accumulate in sediments in highly anthropized areas, and potentially influence prokaryotic communities. However, relatively few studies have investigated bacterial diversity and community composition in highly polluted sediments (Gillan et al., 2005; Paissé et al., 2008; Zhang et al., 2008; Sun et al., 2012), and even less using the NGS techniques (Sun et al., 2013; Korlevic et al., 2015 ´ ). This study was performed to shed

light on the role of chemical pollutants in shaping bacterial diversity in coastal contaminated sediments. The study was carried out in two contrasting areas along the coastline of Italy, characterized by different type and level of chemical contaminants. The chemical analyses confirmed this difference, evident in terms of higher concentration of Ni, Cr, and Mn in the Po River Prodelta site, and in much higher concentrations (up to one hundred times) of certain pollutants, among which the more relevant were Hg, PAHs, and PCBs, in the chronically polluted Mar Piccolo site. These results corroborate previous studies carried out in these areas to investigate spatio-temporal patterns of contaminants (Cardellicchio et al., 2007; Petronio et al., 2012). In the Po River Prodelta site, Cr and Ni showed a high concentration, probably due to the enrichment by the leaching of sedimentary ophiolite complexes, which emerge in the Western Alps and some areas of the Italian Apennine. The elevated Cr and Ni backgrounds are therefore a geogenic character of the Po River alluvial sediments and likely unrelated to anthropogenic contamination (Bianchini et al., 2013). Conversely, the influence of the plume on the enrichment of PCBs in sediments was reflected in the highest values observed in front of the main river mouths. In the Mar Piccolo of Taranto, the most contaminated area was located in front of the arsenal of the Italian Navy, which showed the highest concentrations of several pollutants of concern and, particularly, Hg and PCBs. An additional source of PCBs contamination was found in the vicinity of shipyards in the north of the first shelf of Mar Piccolo. On the other hand, Cr and Ni showed very low values, consistent with the background values reported in the area (Buccolieri et al., 2006).

### Patterns of Richness and Community Composition in Contaminated Coastal Sediments

OTU richness showed a decreasing pattern with increasing chemical contamination in the sediment. A negative correlation between richness and contaminants has been previously reported in soils (Joynt et al., 2006; Goł˛ebiewski et al., 2014), while contrasting patterns have been reported in marine sediments. A pattern similar to our has been observed in South*–*Eastern Australia in metal-and PAH-contaminated sediments (Sun et al., 2012) as well as in certain, despite not at all sites, contaminated sediments in the Northern Adriatic Sea (Korlevic et al., 2015 ´ ). A similar decrease in richness was observed when benthic bacterial communities from a contaminated harbor were exposed to hydrocarbons in laboratory experiments (Yakimov et al., 2005). Conversely, other studies, performed using a community fingerprinting technique, highlighted that OTU richness was unaffected by the sediment contamination (Gillan et al., 2005), while Zhang et al. (2008) reported higher richness in the reference site with respect to the more polluted ones. The lack of consistent patterns in richness can be explained by a multitude of reasons, assuming that the response of complex communities to chemical contamination may vary according to the magnitude-dependent toxic effect of pollutants (Ager et al., 2010; Sun et al., 2012), but also that contaminants may favor, over time scales, the proliferation of microbial consortia of more

tolerant species replacing non-tolerant ones, leading in turn to increases in diversity (Gillan et al., 2005). Also, the use of different microbiological techniques to describe richness (e.g., fingerprinting, cloning and NGS of the 16S rRNA gene) can lead to different results, as fingerprinting methods describe only the richness of dominant species while NGS allows identifying dominant and rare species within the communities (Bent et al., 2007). Our results, among the first based on large sets (millions) of 16S rDNA sequences, indicate that chemical contamination reduce benthic prokaryotic richness in coastal sediments. This pattern agrees with ecological theories predicting that multiple stressors lead to diversity decreases, because of inability of certain individuals to develop tolerance (Vinebrooke et al., 2004).

Despite the negative correlation between richness and contaminants, bacterial community composition showed a very high taxonomic diversity at both sites, with OTUs affiliated to more than 80 phyla and more than 300 classes. This taxonomic composition appears wider than reported by Sun et al. (2013) and Korlevic et al. (2015) ´ , and this can also be related with the use of different sequencing techniques, with our use of the Illumina platform leading to higher number of detected taxa (Wang et al., 2012). We show that community composition in both sites was dominated by Delta- and Gammaproteobacteria, followed in importance by Alphaproteobacteria. This pattern is similar to that described by Sun et al. (2013). Conversely to our findings, Korlevic et al. (2015) ´ reported dominance of Gammaand Alphaproteobacteria, and an overall lower frequency of Deltaproteobacteria, which were mostly observed in the more PAH-contaminated station. The dominance of the classes is not unexpected, as Gamma- and Alphaproteobacteria are widely recognized to utilize aliphatic and aromatic compounds, and to play a key role in oil degradation (Kostka et al., 2011; Acosta-González et al., 2013), and members of Deltaproteobacteria are crucially involved in the anaerobic degradation of organic contaminants and the cycling of sulfur compounds (Gillan et al., 2005; Sun et al., 2013). In our samples, the percentage of Deltaproteobacteria appeared to be significantly and positively correlated to the concentration of PAHs (at all sites, *p <* 0.05, *r* = 0.517) suggesting the presence of a tight link between contaminants and the growth of members of this class.

The second most abundant phylum in our samples was Bacteroidetes. This is in accordance with previous studies in contaminated sediments (Paissé et al., 2008; Sun et al., 2013; Korlevic et al., 2015 ´ ). This class appeared particularly important in the Po River Prodelta, where their percentage is up to three times higher than in previous studies. Members of the Bacteroidetes phylum are known to represent one of the most abundant groups of bacteria in coastal areas, and can be found as free-living or associated with organic particles (Fernández-Gómez et al., 2013). The high relevance observed in contaminated sites, and particularly in the estuarine one, can be explained by considering their potential freshwater origin, and the known ability of members of this phylum to tolerate the toxic effects of certain metals (Oregaard and Sørensen, 2007), or alternatively to grow on some contaminants, such as hydrocarbons, as substrates (Yakimov et al., 2007). Firmicutes, Chloroflexi, and Planctomycetes were also abundant at both sites, followed by less represented phyla such as Verrucomicrobia, Acidobacteria, Actinobacteria, and Spirochaetes. All these phyla have been widely reported as members of contaminated sediments communities (Dell'Anno et al., 2012), and this can be associated to their role as heavy metals-reducers and polyaromatic compounds- and PCB-degraders, or to their ability to survive to toxic effects of metals (Nogales et al., 2001; Petrie et al., 2003; Zhang et al., 2008; Zanaroli et al., 2012).

It is worth noting that all our samples contained, despite at low percentages, members of the class Epsilonproteobacteria. This class included one dominant OTU (belonging to the Helicobacteraceae family) which was consistently identified among the most abundant in all sediment samples. A similar finding was reported by Korlevic et al. (2015) ´ , who showed that this taxon, along with the Lachnospiraceae family, accounted for 60% of the total sequences. At the same time, the dominance of Helicobacteraceae has been observed in sediments collected from the Arctic Mid-Ocean ridge (Jorgensen et al., 2012), where the sulfur cycle was found to be particularly relevant. The dominance of this OTU within Helicobacteraceae in contaminated coastal sediments deserves further investigation, as this class includes members able to survive to high toxic effects (Mitchell et al., 2014) and, given their known involvement in the sulfur cycle, suggests that this process may be particularly relevant in contaminated coastal sites. The other dominant OTUs were affiliated with the sulfurreducing Desulfuromonadaceae family (Deltaproteobacteria), whose members have been previously found in contaminated sediments (Gillan et al., 2005; Sun et al., 2013), as well within the family Desulfobulbaceae, observed in artificial oilspill and harbor sediments (Zhang et al., 2008; Suárez-Suárez et al., 2011). Other dominant OTUs belonged to the sulfur-oxidizing Piscirickettsiaceae family and to the genus Lutimonas. These taxa have been reported among the most abundant during laboratory-scale experiments of dechlorination of contaminated sediments (Zanaroli et al., 2012; Koo et al., 2014), suggesting their possible decontamination role in heavily polluted areas. In the Mar Piccolo site, the dominant OTU belonged to phylum Firmicutes, whose members have been already reported in similar environments worldwide and include several hydrocarbon-degrading species (Zhang et al., 2008; Korlevic et al., 2015 ´ ).

### A Core OTUs Microbiome in Contaminated Sediments

We found a core sediment microbiome comprising 31 OTUs belonging to Bacteria and 2 belonging to Archaea, which were consistently recorded at all stations of the study sites. Sun et al. (2013) reported a core sediment microbiome made up of 13 OTUs in a study carried out in six polluted estuaries in SE Australia. However, these authors reported that the core OTUs were mostly comprised of Gamma-, Delta-, Alphaproteobacteria, and Acidobacteria. Our results, in contrast, show a more diversified microbiome, which also includes OTUs belonging to Beta- and Epsilonbacteria, Clostridia, Cytophagia, Flavobacteria, and to Archaea (within the classes Methanobacteria and Methanomicrobia). These differences can be related to the geographic distance, as well to the different type, level and history of pollution between the sites targeted in the two studies. For

instance, the sites investigated by Sun et al. (2013) showed, on average, three times lower concentration of Cu, six times lower concentration of Ni and two times lower concentration of Pb, as well as differences in the type and concentration of PAHs congeners (such as presence of naphthalene and acenaphthylene, which were absent in our sites). Nonetheless, both studies point out, for the first time, that highly polluted coastal environments, despite being located at very long distances (more than eight hundreds kilometers in our case), does appear to share common microbial players.

Given that non-polluted sites were not analyzed in our study, we cannot exclude that these core OTUs may be present in the area independently of the contamination. However, in partial support of our hypothesis, we observed that several core OTUs displayed lower importance in those stations displaying the lower contamination, which suggests that contamination may have promoted the growth and selection of these OTUs. This appears true, in the Po River Prodelta, for several OTUs (among which Thermoanaerobacter, several Deltaproteobacteria and Methanomicrobia) in the station presenting the lowest level of contamination (PO10). The same was observed in the Mar Piccolo of Taranto, where more than half of the core OTUs were less abundant (up to 28 times) in the less contaminated station (TA1) than in the other ones. Future studies will have to provide support to the existence of a core microbiome in contaminated areas, and to elucidate the role that this core set of microbes plays in the ecosystem functioning and biogeochemical cycles in coastal areas under anthropogenic stress.

### Gradients and Type of Contaminants Shape Bacterial Diversity and Community Composition

The distLM analyses showed that the environmental and chemical variables, with the only exclusion of some metals (Al, Cd, and Fe), significantly shaped community composition at the two sites, by explaining a large fraction of observed variance (73.3 and 72.1% at phylum and OTU level, respectively). Our results corroborates previous findings reporting significant effects of contaminants in driving sediment bacterial community composition (Sun et al., 2012, 2013; Jonhston and Leff, 2015). Sun et al. (2012) reported a significant role of metals and PAHs in driving changes in bacterial communities. Paissé et al. (2008) showed that 32% of variation in bacterial communities along a hydrocarbon contamination gradient was significant explained by PAHs. The fraction of explained variance in our study appears relatively high compared with previous studies, suggesting that contaminants and environmental variables are playing a significant role in structuring communities in the investigated sites. However, the remaining fraction of unexplained variance is likely to depend upon other factors which were not investigated in this study, such as biotic factors (predation, viral lysis and completion), deserving future investigations.

The further dbRDA plot showed that the environmental and chemical parameters divided the stations into two separate clusters corresponding to the study sites, suggesting the existence of important within-site differences. These differences are likely dependent upon a different role played by the geographical and environmental characteristics on communities, and upon the presence of spatial gradients in the concentration and type of pollutants. In the Po River Prodelta, the environmental and chemical parameters structured the 19 stations into several clusters, which corresponded to groups of stations located along gradients of pollution, and characterized by different contamination level. Stations PO1 and PO2 were located northern of the delta main distributary mouth (Po della Pila), in an area under minor influence of river discharge, as river waters discharged to the region are typically flowing south (Boldrin et al., 2005). This finding is confirmed by the average concentration of certain contaminants (such as Ni, Mn, Al, and PCBs) which appears lower in these two stations than in remaining ones located southward. Accordingly, the second cluster of stations (from PO3 to PO7), which was located in front or south of the main distributary mouth, were the most contaminated in terms of more than half of the measured contaminants, and particularly by PCBs (on average 25 ng g<sup>−</sup>1). Thus, communities in this cluster of station are likely to have been shaped together by the higher contaminant pressure, resulting also from the recent flood event which preceded our sampling. The remaining clusters of stations in this site appeared to be structured differently according to contaminant levels, type of anthropogenic influence and distance from land and from the other tributary mouths.

In the Mar Piccolo site, the environmental and chemical parameters structured the 11 stations into three major clusters and some outlier stations. One outlier was TA1, which is located right in front of the "Navigabile" channel connecting the Mar Piccolo with the adjacent open sea. This station was characterized by different environmental characteristics (higher % of sand and lower organic matter content), lower contaminant level (evident especially in terms of metals and PCBs) and a more marked influence of the marine waters entering into the inlet. This influence is likely the reason for the differences in community composition between this station and the other ones, evident for instance in terms of lower importance of Firmicutes and higher contribution of other taxa, such as Vibrionales (which were up to seven times more abundant at this station). The other major clusters of stations were also related with different characteristics. One cluster comprised stations TA4 and TA5, which were located in the nearby of the arsenal of the Italian Navy, and displayed the highest concentration of Hg, As, and PCBs. Station TA13 clustered apart, being characterized by the highest concentrations of PAHs and of total P of the entire study area. Another cluster included TA7, TA8, and TA14, which were located in the middle of the First Inlet, and likely received a similar combined effect of multiple contaminants (including PCBs) which may have shaped similar communities. A remaining cluster grouped together all those stations located in the Second Inlet. This Inlet is less polluted than the First in terms of most of the analyzed contaminants. In this cluster, the station TA12 showed an increased relevance of Clostridia (phylum Firmicutes), accounting for more than 20% of the sequences. Within this class, two of the five dominant OTUs observed in the entire study site (Thermoanaerobacter and Tissierella/Soehngenia) reached percentages up to 3 and 12 times (respectively) higher than the average values in the other stations. The high relevance of Clostridia in this station may be related to the presence of dense mussel culture facilities, which supports previous results in the same study area stressing the role of these bacteria as bioindicator of aquaculture impact (Zaccone et al., 2005).

The two sites here studied also differed significantly in terms of PCBs concentration (ANOVA, *p <* 0.01), with the Mar Piccolo site showing the average concentration of 466.72 ng g−<sup>1</sup> (standard error 89.29 ng g−1), which was up to 25 times higher than in the Po River Prodelta (average 18.71 ng g<sup>−</sup>1, standard error 1.19 ng g−1). This striking difference was reflected in the microbial community composition, and in the increased relevance of certain taxa which are well known as PCB-degraders. The largest difference was observed in terms of Chloroflexi, which were on average twofold more abundant in Mar Piccolo than the Po River Prodelta. PCBs are known to be degraded by several prokaryotes under both aerobic and anaerobic types of metabolism (Furukawa and Fujihara, 2008; Pieper and Seeger, 2008). PCB-degrading bacteria span a variety of taxonomic groups, among which the gram negative genera *Pseudomonas*, *Alcaligenes*, *Achromobacter*, *Burkholderia*, *Comamonas*, *Sphingomonas*, *Ralstonia*, and *Acinetobacter*, and the Gram-positive genera *Rhodococcus*, *Corynebacterium*, and *Bacillus* (Field and Sierra-Alvarez, 2008). Enrichment cultures experiments indicate that several *Dehalococcoides* sp. and other microorganisms within the Chloroflexi phylum can grow by linking the oxidation of H2 to the reductive dechlorination of PCBs (Fagervold et al., 2007). The percentage of Chloroflexi in our samples was significantly and positively related with the concentration of PCBs (*r* = 0.42, *p <* 0.05) suggesting the potential selection of PCB-degrading communities in the more polluted stations. This hypothesis was further corroborated by the presence of significant correlations between Chloroflexi and certain PCBs congeners, such as 99, 110, and 126 (*r* = 0.44, *r* = 0.43, and *r* = 0.61, *p <* 0.05, respectively). It is clear that future studies should use more accurate quantification techniques, such as quantitative PCR targeting specific genes (the 16S rRNA gene of *Dehalococcoides*), to quantify the abundance of degrading populations, and confirm the significance of these relationships. At the same time, future researches are needed to prove the functional capacity of these populations and to identify their active fraction, basing on the analysis of the 16S rRNA rather than 16S rDNA or other approaches to quantify functions (Blazewicz et al., 2013). Chlorinated biphenyls are potentially fully biodegradable, a process which is made possible in a sequence of anaerobic reductive dechlorination reactions, followed by aerobic mineralization of the lower chlorinated products (Field and Sierra-Alvarez, 2008). Our results suggest the possible role of certain autochthonous populations as key agents in sediment PCBs degradation processes, and as potential targets for future remediation actions to improve the environmental quality of the investigated sites, and to design and test on-site bioremediation actions by the sediment naturally dechlorinating microorganisms, as recently demonstrated in laboratory experiments in the Mar Piccolo site (Matturro et al., 2015).

## Conclusion

We conclude that sediment contamination by multiple chemicals do significantly shape benthic prokaryotic diversity in coastal areas, and suggest considering the potential contribution of resident microbes to actions of contaminant remediation. While our results add new knowledge about the anthropogenic influence on coastal habitats and the response of benthic microbes, future studies will need to elucidate how and whether multiple contaminants influence the functioning of the sediment microbiome, to identify microbial indicator taxa for human alteration, and to evaluate the ecological and biogeochemical consequences of chemical pollution on coastal ecosystems.

### Acknowledgments

This work was possible thanks to funds granted to GML by the Italian National Flag Programme RITMARE (SP3-WP2- AZ2 "Strumenti innovativi per la valutazione degli effetti di contaminanti emergenti sulle comunità biologiche") and to

### References


funds granted to DC by the same project (SP3-WP2-AZ3 "Caratterizzazione ambientale integrata e bonifica dei siti contaminati costieri"). We sincerely thank our colleagues Dr. Tiziano Minuzzo (for precious advices and crucial technical assistance with the bioinformatic analyses) and Dr. Francesca Garaventa (ISMAR-CNR, Venice) for the precious help during sampling activities in both the study sites. Dr. Nicola Cardellicchio, Dr. Rosa Cavallo, Dr. Loredana Stabili, and Dr. Marcella Narracci (IAMC-CNR, Taranto) are acknowledged for providing laboratory facilities and spaces during the sampling campaign in the Mar Piccolo of Taranto, and similarly Dr. Maria G. Marin and Dr. Carlotta Mazzoldi (University of Padova) for providing laboratory spaces at the Marine Station in Chioggia during the Po River Prodelta sampling campaign.

### Supplementary Material

The Supplementary Material for this article can be found online at: http://journal*.*frontiersin*.*org/article/10*.*3389/fmicb*.* 2015*.*01053


Elliott, J. E., and Elliott, K. H. (2013). Tracking marine pollution. *Science* 340, 556–558. doi: 10.1126/science.1235197


in vitro following exposure to the Mississippi Canyon Block 252 oil (MC252). *3 Biotech.* 5, 1–14.


**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 Quero, Cassin, Botter, Perini and Luna. 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 role of biofilms as environmental reservoirs of antibiotic resistance

### *José L. Balcázar1\*, Jéssica Subirats1 and Carles M. Borrego1,2*

*<sup>1</sup> Catalan Institute for Water Research, Girona, Spain, <sup>2</sup> Group of Molecular Microbial Ecology, Institute of Aquatic Ecology, University of Girona, Girona, Spain*

Antibiotic resistance has become a significant and growing threat to public and environmental health. To face this problem both at local and global scales, a better understanding of the sources and mechanisms that contribute to the emergence and spread of antibiotic resistance is required. Recent studies demonstrate that aquatic ecosystems are reservoirs of resistant bacteria and antibiotic resistance genes as well as potential conduits for their transmission to human pathogens. Despite the wealth of information about antibiotic pollution and its effect on the aquatic microbial resistome, the contribution of environmental biofilms to the acquisition and spread of antibiotic resistance has not been fully explored in aquatic systems. Biofilms are structured multicellular communities embedded in a self-produced extracellular matrix that acts as a barrier to antibiotic diffusion. High population densities and proximity of cells in biofilms also increases the chances for genetic exchange among bacterial species converting biofilms in hot spots of antibiotic resistance. This review focuses on the potential effect of antibiotic pollution on biofilm microbial communities, with special emphasis on ecological and evolutionary processes underlying acquired resistance to these compounds.

Keywords: aquatic ecosystems, biofilms, mobile genetic elements, antibiotic resistance genes, aquatic resistome

# ENVIRONMENTAL BIOFILMS

Nature is often unpleasant. It is then better to face environmental uncertainties under the principle of "strength through unity". In many habitats, either natural or artificial, microorganisms attach themselves to surfaces, either abiotic or biotic, forming a complex matrix of biopolymers known as biofilm that protect them from environmental hazards (Costerton et al., 1978). Biofilms may be composed of a single bacterial species (e.g., *Vibrio cholerae*, Teschler et al., 2015) but more frequently they are formed by a complex and diverse community of microorganisms (bacteria, algae, fungi and protozoa) embedded in an extracellular matrix of polysaccharides, exudates, and detritus (Costerton et al., 1978; Wimpenny et al., 2000). Many microbial species are able to change their lifestyle (free-living vs. attached) depending on their physiological status and the physicochemical conditions in their surroundings, taking advantage of the greater availability of organic matter in suspended particles and surfaces (Simon et al., 2002; Grossart et al., 2004; Grossart, 2010; Teschler et al., 2015). In aquatic habitats, biofilms develop not only in benthic substrata, such as streambed cobbles and sand (epilithic and epipsammic biofilms, respectively), but also on floating macro– and microaggregates (Simon et al., 2002). From an ecological perspective, microorganisms in environmental biofilms actively participate in organic matter decomposition, nutrient dynamics and biogeochemical cycling, being a key component of

### *Edited by:*

*Maurizio Labbate, University of Technology Sydney, Australia*

### *Reviewed by:*

*Michael R. Twiss, Clarkson University, USA Hans-Peter Grossart, IGB-Leibniz-Institute of Freshwater Ecology and Inland Fisheries, Germany*

> *\*Correspondence: José L. Balcázar jlbalcazar@icra.cat*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 09 June 2015 Accepted: 19 October 2015 Published: 31 October 2015*

### *Citation:*

*Balcázar JL, Subirats J and Borrego CM (2015) The role of biofilms as environmental reservoirs of antibiotic resistance. Front. Microbiol. 6:1216. doi: 10.3389/fmicb.2015.01216* ecosystem functioning (Sabater and Romaní, 1996; Sabater et al., 2002; Simon et al., 2002; Battin et al., 2007; Romaní, 2010). Moreover, streambed biofilms are considered as good indicators of the overall water quality and the ecological status of the system (i.e., ecosystem health) (Burns and Ryder, 2001; Sabater et al., 2007). It is then of special interest to assess how biofilm communities respond to anthropogenic pollution of aquatic environments (e.g., rivers, lakes, and reservoirs) considering the increasing amount of chemical compounds (metals, personal care products and drugs used in veterinary and human medicine) released into these waterbodies mainly through wastewater treatment plant (WWTP) effluents and agricultural run-off (Pruden et al., 2006; Sarmah et al., 2006; Baquero et al., 2008). This review focuses on the role of streambed biofilms as reservoirs of antibiotic resistant bacteria and resistance genes, providing a general overview of the causes and consequences of a chronic exposure of biofilm communities to sub-inhibitory concentrations of antibiotics and their role in the spread and persistence of antibiotic resistance.

# BIOFILMS AND ANTIBIOTICS

Biofilms show an increased survival and resistance to environmental and chemical stressors (e.g., antibiotics) mainly, but not only, by the protection conferred by the extracellular polysaccharide matrix (Mah and O'Toole, 2001; Stewart and Costerton, 2001; Donlan, 2002; Donlan and Costerton, 2002; Stewart, 2002; Hall-Stoodley et al., 2004; Høiby et al., 2010). In biofilms, bacterial cells exhibit 10 to 1,000 times less susceptibility to specific antimicrobial agents compared with their planktonic counterparts (Gilbert et al., 2002). This reduced susceptibility is caused by a combination of different factors, namely: (i) a poor antibiotic penetration into the polysaccharide matrix; (ii) the arbitrary presence of cells showing a resistant phenotype (known as "persisters"); and (iii) the presence of either non-growing cells or cells that triggered stress responses under unfavorable chemical conditions within the biofilm matrix (Stewart and Costerton, 2001; Stewart, 2002). These protective mechanisms act synergistically to those responsible for conventional resistance linked to the presence of antibiotic resistance genes (ARGs) in bacterial genomes or extrachromosomal elements, yielding an overall increased resistance of biofilms to antimicrobial compounds. For instance, β-lactamase producing bacteria offered increased protection in biofilms because the β-lactam antibiotic, such as ampicillin, was inactivated by those β-lactamases (Anderl et al., 2000). Moreover, the *ampC* gene of *Pseudomonas aeruginosa* biofilms was strongly induced by exposure to antibiotics, such as imipenem (Bagge et al., 2004). Additionally, biofilm formation may result as a defensive reaction to the presence of antibiotics. Hoffman et al. (2005) found that sub-inhibitory concentrations of aminoglycosides induce biofilm formation as part of a defense response in *Escherichia coli* and *P. aeruginosa*. Similar results were described by Salcedo et al. (2014), who observed that sub-inhibitory concentrations of tetracycline and cephradine induce biofilm formation and enhance the transfer rate of the pB10 plasmid among the biofilm biomass (*E. coli* and *P. aeruginosa*) at rates 2–5 times faster than without antibiotic treatment. Since biofilm formation is also common for most bacterial pathogens, the enhanced resistance of biofilms to antibiotics is a serious concern for human health as many chronic infections are linked to biofilm growth on either natural surfaces (e.g., teeth, lungs) or foreign-body devices (e.g., pacemakers, catheters, prosthetic heart valves). The characteristics, composition, growth dynamics, and resistance mechanisms of clinically relevant biofilms have been reviewed in detail by several authors (Donlan and Costerton, 2002; Parsek and Singh, 2003; Hall-Stoodley et al., 2004; Høiby et al., 2010), and are out of the scope of this review. In clear contrast, lesser is known about the role of environmental biofilms as natural reservoirs of ARGs, their contribution to ARGs spreading among biofilm inhabitants and their transfer to free-living bacteria, increasing the risk for their transmission to aquatic microorganisms and potential human pathogens (Vaz-Moreira et al., 2014 and references therein).

# ENVIRONMENTAL BIOFILMS UNDER CHEMICAL STRESS

Many aquatic systems (rivers, lakes, reservoirs) are affected by human activities such as continuous discharges from WWTP effluents. Under such conditions, macro- and microorganisms inhabiting these waterbodies are exposed to a low but constant concentration of a wide range of chemical pollutants (antibiotics but also analgesics, anti-inflammatory, and psychiatric drugs, β-blockers, pesticides, etc.) that alter their behavior at different levels, with consequences that we are only beginning to grasp (Bernier and Surette, 2013; Boxall, 2014). Several studies have demonstrated the effects of the so-called emerging pollutants on the composition, activity, and resilience of streambed biofilms (Bonnineau et al., 2010; Ricart et al., 2010; Proia et al., 2011, 2013a,b; Osorio et al., 2014), although the ecological implications of such background pollution are difficult to envisage. A serious drawback arises when comparing the environmental concentrations of antibiotics measured in polluted aquatic habitats (from ng/L to μg/L) to those used to treat bacterial infections (i.e., therapeutic concentrations, which are usually ≥1 mg/L). Since environmental concentrations of antimicrobial compounds are several orders of magnitude below the minimum inhibitory concentration (MIC) of most bacterial pathogens, their antibiotic effect is doubtful, if any (Waksman, 1961; Davies, 2006; Davies et al., 2006; Davies and Davies, 2010). Current data strongly suggest that antibiotics, at these sub-MIC concentrations, act as signaling molecules mediating a wide variety of cell processes (gene transcription and expression, quorum sensing, inter- or intra-species communication, biofilm formation, among others; Davies, 2006; Romero et al., 2011; Sengupta et al., 2013; Andersson and Hughes, 2014), instead of causing growth arrest or cell death. Moreover, low concentration of antibiotics may also trigger different stress responses that might accelerate horizontal gene transfer (HGT) and the spread of ARGs in a broad range of bacterial species (Beaber et al., 2004; Miller et al., 2004; Maiques et al., 2006). Under this perspective, the chronic exposure to subinhibitory antibiotic concentrations that occurs in most aquatic ecosystems offers new avenues for research that deserve exploration. For instance, is the effect of this chronic exposure strong enough to shape the composition of microbial communities? Or is it buffered by the many other physico-chemical constraints that microbes face in their habitat? Is the antibiotic pollution adding a background noise that interferes with normal communication among bacterial cells in their habitats (e.g., biofilms)? If so, how can this noise effect be measured? And what about activity? Does antibiotic pollution have measurable effects on biogeochemical cycles at both local and global scales? In this regard, Roose-Amsaleg and Laverman (2015) have recently reviewed 31 articles dealing with the effects of antibiotics on microorganisms involved in biogeochemical cycles to ascertain if environmental concentrations of these compounds have side-effects on such cycles, with special focus on N cycling (anammox, denitrification, and nitrification). Despite the few studies available and the variability in terms of antibiotic types and conditions tested, conclusions of their work point to a clear alteration of microbial activity in key biogeochemical cycles, thus affecting ecosystem functioning at different levels.

Despite these considerations, it is now clear that chronic exposure to antibiotics, even at very low concentrations, promotes and maintains a pool of resistance genes in natural microbial communities (Séveno et al., 2002; Allen et al., 2010; Sengupta et al., 2013; Andersson and Hughes, 2014). It should be mentioned, however, that most of these genes, although conferring a resistant phenotype when expressed, are probably not "true" resistance genes (Martinez et al., 2015) thus having a function distantly related to that under therapeutic conditions (Allen et al., 2010; Martinez et al., 2015). Notwithstanding this, current data indicate that the extensive use of antibiotics over the last century has generated a selective pressure that has accelerated the acquisition and spread of ARGs among environmental bacteria posing a risk for human health assuming the striking capacity of microbes to share genes.

# ACQUISITION AND SPREAD OF ARGs IN BIOFILMS

Susceptible bacteria may become resistant to antibiotics through chromosomal mutations or by HGT, being the latter the major contributor to the spread of antibiotic resistance determinants. The significance of HGT to microbial adaptation was initially recognized when antibiotic-resistant pathogens were identified (Sobecky and Hazen, 2009). HGT is mediated by mobile genetic elements (MGEs), which play an important role in the evolution and adaptation of bacterial species to new and/or changing environmental conditions (Frost et al., 2005). MGEs are segments of DNA encoding a variety of enzymes and proteins that mediate their movement within the host genome (intracellular mobility) or between bacterial cells (intercellular mobility). Interchange of DNA fragments between a cell donor and a receptor takes place through conjugation, transformation, or transduction, whereas intracellular movement

is facilitated by integrons and transposons (Modi et al., 2014).

Together with phage transduction and natural transformation, the exchange of genetic material through conjugation is one of the most efficient pathways to disseminate antibiotic resistance among bacterial cells, where donor and recipient cells are in close contact. Conjugation is mainly mediated by the so-called "conjugative plasmids", although "conjugative transposons" are also capable of triggering the process. One of the most important aspects of conjugative plasmids is that they can be exchanged among both related and phylogenetically distant bacteria (Dionisio et al., 2002). The high cell density and close contact among cells within the biofilm matrix together with increased genetic competence and accumulation of MGEs in these habitats convert them into an optimal scenario for the acquisition and spread of ARGs (Fux et al., 2005). Several studies have shown increased conjugation efficiencies in biofilms when compared to free-living bacterial cells. In fact, conjugation of the broad-host-range plasmid RP4 between two species of *Pseudomonas* occurred in a biofilm reactor at high frequencies (Ehlers and Bouwer, 1999). *In situ* assessment of gene transfer rates in biofilms using automated confocal laser scanning microscopy revealed conjugation rates 1,000-fold higher than those determined by classical plating techniques (Hausner and Wuertz, 1999). Molin and Tolker-Nielsen (2003) also showed that the efficiency of gene transfer seems to be correlated with the biofilm surface, suggesting that a high surface/volume ratios favor transfer within or between biofilm populations.

The diversity and abundance of ARGs in environmental biofilms have been investigated by several authors to unveil differences in the concentration of target genes between planktonic and benthic compartments. Less information is available, however, on the contribution of MGE to the acquisition and spread of ARGs among biofilm inhabitants and between them and free-living bacteria. **Table 1** summarizes some relevant studies dealing with the presence, diversity and abundance of ARGs in biofilms from different environmental settings such as rivers exposed to WWTP effluent discharges, WWTP and drinking water network pipelines, experimental mesocosm, and sand filters. Although not exhaustive, **Table 1** provides a general overview of results obtained by different research groups studying the role of environmental biofilms as hot spots for the accumulation and transfer of ARGs. Schwartz et al. (2003) demonstrated that the *vanA* gene, which confers a highlevel resistance to vancomycin, was detected in drinking water biofilms in the absence of any vancomycin-resistant enterococci, suggesting a potential gene transfer from them to autochthonous bacteria in drinking water systems. Gillings et al. (2008) investigated the presence of a MGE, the class 1 integrase (*intI1*) gene, in bacterial isolates collected from diverse environmental samples near Sydney. Authors found that 1 to 3% of bacterial isolates from lake sediments were *intI1* positive, while in biofilms from a groundwater treatment plant, the number of *intI1*-positive isolates reached 30% despite no antibiotics were used as selective agents for culturing. Moreover, Engemann et al. (2008) found that the abundance of six genes conferring resistance to tetracycline


 biofilms.

TABLE

*polymorphism.*

was reduced at different rates in the water column, and some genes, particularly *tetW*, readily migrated into biofilms. Transfer to biofilms did not, however, completely explain disappearance of *tet* genes from the planktonic compartment and other factors such as sunlight and potential microbial degradation would probably contributed (Engemann et al., 2006, 2008). In a similar experimental approach but using periodical piglet waste loadings, Zhang et al. (2009) observed that *tet* genes migrate rapidly to biofilms, where they persist longer than in adjacent waters. Recently, Farkas et al. (2013) also observed that 9.4% of isolates from drinking water biofilms harbored class 1 integrons, which were mainly detected in bacteria (e.g., *Enterobacteriaceae*) that may be associated with microbiological contamination.

Because biofilms play an important role as reservoirs for ARGs, they could be considered as biological indicators of antibiotic resistance pollution in the same way as river ecologists use streambed biofilms as indicators of the overall "ecological status" of the river ecosystem (Sabater et al., 2007). The chronic exposure to sub-MIC concentration of antibiotics exerts a selective pressure on biofilm bacterial communities that may stimulate the emergence and spread of antibiotic resistance (Allen et al., 2010; Andersson and Hughes, 2014; Marti et al., 2014a; Chow et al., 2015). The presence of other pollutants, such as heavy metals from feed additives, organic, and inorganic fertilizers, pesticides and anti-fouling products, also contributes in the coselection of antibiotic resistance because the close location of genes encoding for these resistance phenotypes in the same MGE (Seiler and Berendonk, 2012). Such exposures may eventually have consequences on the selection and abundance of MGEs, thereby facilitating the spread of ARGs among different species; different biofilm compartments (e.g., epilithic, epipsammic, and hyporheic streambed); or even between different prokaryotic communities as recently assessed by plasmid metagenomics (Sentchilo et al., 2013). Besides, several studies provided evidence that ARGs tend to accumulate in biofilms rather than in the planktonic compartment. In this regard, Börjesson et al. (2009) found a high proportion of genes encoding resistance to aminoglycosides and tetracyclines in biofilm samples collected at a WWTP. Winkworth (2013) demonstrated that, while the levels of ARGs in biofilm samples collected along the Taieri River were low, sites subjected to combined influences of greater human activity and intensive dairy farming showed an increased level of ARGs. Likewise, a study carried out by our research group clearly showed the effect of WWTP effluents on the prevalence of several ARGs in the Ter River, accompanied by a significant increase in their relative abundance in biofilm samples collected downstream the WWTP discharge point (Marti et al., 2013). Moreover, we have investigated the prevalence of plasmid-mediated quinolone resistance (PMQR) determinants in ciprofloxacin-resistant strains isolated in biofilm and sediments from a WWTP discharge point and its receiving river (upstream and downstream sites). We observed that, while the number of strains harboring PMQR determinants was higher in sediments, PMQR-positive strains were also detected in biofilm samples, especially in those from the WWTP discharge point and downstream sites (Marti et al., 2014b). In a study carried out in a horizontal subsurface flow constructed wetland, Nõlvak et al. (2013) found that copy numbers of *tetA* and *sul1* genes in the wetland biofilms were one order of magnitude higher than in the effluent water, despite the fact that this facility had a similar efficiency to conventional WWTP in removing ARGs from wastewater. Altogether, these studies undoubtedly demonstrate the contribution of biofilms in the acquisition and spread of ARGs.

# ANTIBIOTIC RESISTANCE IN BIOFILMS ASSESSED BY METAGENOMICS

Until the last decade our knowledge of antibiotic resistance has largely depended on data provided by traditional culture-based methods (Cockerill, 1999). Although useful, these data are limited and biased towards cultivable members of the community. Recent advances in genomics and metagenomics are now providing new avenues for understanding evolutionary processes controlling antibiotic resistance mechanisms and their spreading among microbial populations.

To date, several thousand metagenomes have already been sequenced from a large variety of environments, and this number is set to grow rapidly in the forthcoming years. Most of these metagenomes are publically available through various databases and annotation platforms, such as MG-RAST (Meyer et al., 2008), CAMERA (Sun et al., 2011), and IMG/M (Markowitz et al., 2012), which provide additional insight in the function of complex microbial communities through comparative analyses. Moreover, the availability of specialized databases such as the ARG Database (ARDB; Liu and Pop, 2009), the Comprehensive Antibiotic Resistance Database (CARD; McArthur et al., 2013), the Integron Database (INTEGRALL; Moura et al., 2009), the Bush, Palzkill, and Jacoby's collection of curated β-lactamase proteins (http:// www.lahey.org/Studies/), and the implementation of highthroughput sequence analysis tools such as BLAT (Kent, 2002), USEARCH (Edgar, 2010), and DIAMOND (Buchfink et al., 2015), provide a comprehensive molecular toolbox that allow a better understanding of the evolution, ecology, and spread of antibiotic resistance in different organisms and ecosystems.

We have conducted a comparative analysis of selected metagenomes corresponding to several projects and environments publically available in the MG-RAST database (http://metagenomics.anl.gov/) to provide an overall insight on the prevalence of MGEs and ARGs in environmental biofilms. This analysis showed that MGEs-related sequences, such those from phages and plasmids, were found in a lower proportion in metagenomes from river biofilms than those from WWTPs and river water environments. Remarkably, transposons were detected in a higher proportion in WWTPs and river biofilms than those from river water environments (**Figure 1**). Similarly, sequences related to genes conferring resistance to β-lactam antibiotics were also detected more frequently among microbial communities from WWTPs and streambed river biofilms than those from river water environments. Sequences related to genes conferring resistance to tetracyclines were also abundant in WWTPs and river biofilms, but to a lesser extent than β-lactams. Finally, no differences in the proportion of genes conferring resistance to sulfonamides were observed among the examined environments.

Interestingly, the analysis of the selected metagenomes also showed that two acid mine drainage biofilm samples from the Richmond Mine (4441138.3 and 4441137.3) yielded a high proportion of sequences related to genes conferring resistance to β-lactam antibiotics (5.7 to 7.2%). These relatively high values of β-lactamases might be related to the higher proportion of transposons in these acidophilic biofilms (0.5 to 1.6%) than those detected in environments close to neutral pH such as riverbed biofilms, WWTPs and freshwater systems (**Figure 1**).

A recent study revealed a remarkable abundance and diversity of genes encoding transposases in the metagenome of a hydrothermal chimney biofilm (Brazelton and Baross, 2009). The comparative analysis between this metagenome (4461585.3) and the metagenomes mentioned above confirmed these observations (8.1% of transposase sequences), but similar proportions were observed for β-lactamases between the hydrothermal vent biofilms and those from river water environments. The high relative proportion of transposases may favor an enhanced gene transfer between bacterial genomes that confer new and useful accessory functions, including resistance to heavy metals or antimicrobial compounds. The presence of genes conferring resistance to β-lactams in environments not subjected to antibiotic pollution such as deep sea vents or pristine systems raises interesting questions not only about the origin and ecological function of these genes in nature but also the criteria that researchers adopt when defining a resistance gene (Martinez et al., 2015).

4511199.3; and river biofilms: 4528142.3, 4528143.3, 4528144.3, 4528145.3, 4528146.3, 4528147.3, 4589537.3, 4589538.3, 4589539.3, 4589540.3,

4589541.3, and 4589542.3.

## FINAL REMARKS AND FUTURE PROSPECTS

Biofilms occur in almost any submerged surface in both natural and man-made systems providing a suitable and optimal environment for the growth, activity, and interaction of different bacterial species. Biofilms also provide a shelter where to cope with transient or permanent stress conditions, also favoring metabolic interactions and genetic interchange between different bacterial species struggling for survival in a changing environment. Punctual or continuous discharges of pharmaceutical compounds into aquatic systems might constitute not only a selective pressure on aquatic bacterial communities that stimulate the transmission and spread of ARG, but also a chronic source of background biochemical noise that may potentially interfere the communication networks that microbes finely tuned during evolution. Although little information is available on the actual capacity of aquatic bacteria to transfer antibiotic-resistance determinants to potential human pathogens, current data corroborate that environmental biofilms are true reservoirs of ARGs. Further research is needed; however, to elucidate to which extent such hot spots of antibiotic resistance may constitute a serious concern for human health, how the diversity and abundance of ARG

### REFERENCES


change between different biofilm compartments, how this resistance genetic pool moves among communities and how this gene transfer varies in response to the amount of chemical pollution (antibiotics but also other stressors such as heavy metals and xenobiotic compounds) in the receiving waters. The continuous refinement of sequencing technologies (e.g., metagenomics, metatranscriptomics) and bioinformatic tools and the availability of specialized and properly curated databases may help to reach these goals and hit new research targets. Answering these (and other) questions will provide a better knowledge of the transfer dynamics of resistance genes at ecosystem level (between species, communities, and/or habitats), yielding clues to fight against antibiotic resistance and the threat that it poses to the environment and to the human health.

### ACKNOWLEDGMENTS

This work has been supported by the European Communities seventh Framework Programme Funding under Grant agreement no. 603629-ENV-2013-6.2.1-GLOBAQUA. JB acknowledges the Ramon y Cajal research fellowship (RYC-2011-08154) from the Spanish Ministry of Economy and Competitiveness.


**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 Balcázar, Subirats and Borrego. 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.*

# Potential impacts of aquatic pollutants: sub-clinical antibiotic concentrations induce genome changes and promote antibiotic resistance

Louise Chow, Liette Waldron and Michael R. Gillings \*

Emma Veritas Laboratory, Department of Biological Sciences, Macquarie University, Sydney, NSW, Australia

### Edited by:

Maurizio Labbate, University of Technology, Sydney, Australia

### Reviewed by:

Julian Davies, University of British Columbia, Canada Jose L. Martinez, Centro Nacional de Biotecnología, Spain

### \*Correspondence:

Michael R. Gillings, Department of Biological Sciences, Genes to Geoscience Research Centre, Macquarie University, Sydney, NSW 2109, Australia michael.gillings@mq.edu.au

### Specialty section:

This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology

Received: 27 May 2015 Accepted: 22 July 2015 Published: 05 August 2015

### Citation:

Chow L, Waldron L and Gillings MR (2015) Potential impacts of aquatic pollutants: sub-clinical antibiotic concentrations induce genome changes and promote antibiotic resistance. Front. Microbiol. 6:803. doi: 10.3389/fmicb.2015.00803 Antibiotics are disseminated into aquatic environments via human waste streams and agricultural run-off. Here they can persist at low, but biologically relevant, concentrations. Antibiotic pollution establishes a selection gradient for resistance and may also raise the frequency of events that generate resistance: point mutations; recombination; and lateral gene transfer. This study examined the response of bacteria to sub-inhibitory levels of antibiotics. Pseudomonas aeruginosa and Pseudomonas protegens were exposed kanamycin, tetracycline or ciprofloxacin at 1/10 the minimal inhibitory concentration (MIC) in a serial streaking experiment over 40 passages. Significant changes in rep-PCR fingerprints were noted in both species when exposed to sub-inhibitory antibiotic concentrations. These changes were observed in as few as five passages, despite the fact that the protocols used sample less than 0.3% of the genome, in turn suggesting much more widespread alterations to sequence and genome architecture. Experimental lines also displayed variant colony morphologies. The final MICs were significantly higher in some experimental lineages of P. protegens, suggesting that 1/10 the MIC induces de-novo mutation events that generate resistance phenotypes. The implications of these results are clear: exposure of the environmental microbiome to antibiotic pollution will induce similar changes, including generating newly resistant species that may be of significant concern for human health.

### Keywords: antibiotic resistance, microbiome, antibiotic pollution, SOS response, evolution

# Introduction

Antibiotic resistance has been identified as one of the greatest threats to human health for the twenty-first century by the World Health Organisation (WHO, 2014). Overuse and misuse of antibiotics in the medical and agricultural sectors have contributed to the problem, and it is estimated that 70% of pathogens now exhibit resistance to at least one or more antibiotics (Berdy, 2012). In most cases the risk of death is doubled if the individual is infected with a resistant strain of bacteria. In the United States in 2013, there were 23,000 confirmed deaths due to Antibiotic resistance (US CDC) and Europe reports 25,000 deaths per year (2007, ECDC).

The primary human use of antibiotics is medicinal, where they are used to treat a range of bacterial infections. However, misuse and overuse of antibiotics are contributing to the development of antibacterial resistance. Incorrect prescription of antibiotics, unnecessarily high dosages and over-use all promote resistance (Campoccia et al., 2010; Andersson and Hughes, 2012; Hvistendahl, 2012; Witte, 2013). Antibiotics are also extensively used in agriculture and aquaculture to prevent disease and as a growth promoter (Hilbert and Smulders, 2004; Bednorz et al., 2013). It has been estimated that 50–70% of antibiotics produced in the United States of America are used in agriculture (Lipsitch et al., 2002; Berge et al., 2005).

A relatively small amount of the antibiotics consumed by humans and animals are actually absorbed, with some 30–90% of antibiotics excreted unchanged and released into waste treatment facilities or directly into the environment (Sarmah et al., 2006). Antibiotics, along with heavy metals, disinfectants and genes conferring resistance are disseminated into the environment via human waste streams, agricultural run-off (Su et al., 2014) and effluent from antibiotic production factories (Li et al., 2009, 2010). Current waste treatment methods are often unable to remove these substances, and the water is either reclaimed (Wang et al., 2014) or released into the environment via rivers (Pruden et al., 2006; Storteboom et al., 2010), estuaries or the ocean (Lapara et al., 2011; Wang et al., 2014). The release of these substances into the environment should be thought of as a significant component of soil and water pollution.

Waste water treatment facilities and aquatic environments can become hotspots for the generation and acquisition of resistance. The presence of selective agents such as antibiotics, heavy metals and disinfectants, combined with genes conferring resistance, mobile elements such as transposons, plasmids, and integrons, and diverse microorganisms creates an ideal environment to generate resistance through mutation or lateral gene transfer.

Many studies have investigated the effect of clinical, or inhibitory levels, of antibiotics on the generation of antibiotic resistance. However, there is increasing evidence that subinhibitory levels of antibiotics may have significant effects on bacterial populations. A gradient of antibiotic concentration forms around human activities. Within the human microbiome there may be a gradient along the digestive tract, while dissemination of antibiotics via waste water will generate a gradient of antibiotic concentration spreading outwards from human population centers.

Sub-inhibitory levels of antibiotics are known to trigger the SOS response, a broad response to DNA damage that has been documented in many bacterial species. It may play a significant role in the generation of antibiotic resistance, as it can increase the rates of mutation and lateral gene transfer (Baharoglu and Mazel, 2014). It is triggered by the occurrence of single stranded DNA resulting from DNA damage, or inhibition of the processes involved in DNA replication. The SOS response is mediated by the LexA repressor. Under normal conditions, LexA prevents SOS genes from being expressed. Under stressful conditions, the protein RecA is recruited onto single stranded DNA where it stimulates cleavage of the LexA repressor, inactivating it and therefore allowing the expression of approximately 40 SOS genes. SOS genes are often involved in DNA repair (Laureti et al., 2013; Baharoglu and Mazel, 2014).

It is well documented that lethal concentrations of antibiotics can induce the SOS response in bacteria (Miller et al., 2004; Michel, 2005). It has also been suggested that sub-inhibitory levels of antibiotics, as those discussed above, may be more relevant to the problem of antibiotic resistance than lethal concentrations of antibiotics (Andersson and Hughes, 2012; Hughes and Andersson, 2012; Laureti et al., 2013). Lethal concentrations exert a strong selective pressure on bacteria, whereby they either die or they acquire mutations allowing them to survive. When exposed to sub-inhibitory levels of antibiotics, most bacteria survive with little effect on growth, and the SOS response is initiated. This, in turn, increases general rates of mutation and lateral gene transfer amongst all bacteria in a population, adding to any extant diversity upon which natural selection can operate. It is also thought that humans may be inadvertently selecting for lineages of bacteria with a greater ability to evolve through increased basal rates of mutation and lateral gene transfer (Gillings and Stokes, 2012).

Sub-inhibitory concentrations of antibiotics polluting areas surrounding human activity may be affecting: (i) the rates at which bacteria can generate variation; and (ii) the rates at which advantageous mutations fix in natural environments. However, there has been little or no empirical testing of these ideas.

In this study, two species of Pseudomonas were passaged as single colony transfers on media containing 1/10 their respective minimum inhibitory concentrations for three different classes of antibiotics. This experiment was designed to test the genotypic and phenotypic effects of realistic levels of antibiotic pollution.

# Materials and Methods

### Bacterial Isolates

Isolates of two species were selected for this study: Pseudomonas aeruginosa strain PA14; and P. protegens strain PF-5. These species were chosen as they encompass both clinical and environmental representatives of the genus. Both strains have been genome sequenced (GenBank: AY273869.1 GenBank: CP000076.1, He et al., 2004; Paulsen et al., 2005). P. aeruginosa PA14 is an opportunistic bacterium that causes infections in hospitals and cystic fibrosis patients. P. protegens PF-5 (formerly Pseudomonas fluorescens PF-5) is a common soil bacterium studied for its potential biocontrol properties (Loper et al., 2012).

P. aeruginosa PA14 was obtained from Professor Joyce Loper, Oregon State University and P. protegens PF-5 was obtained from Professor Ian Paulsen, Macquarie University. Bacteria were maintained on LB Agar plates (0.01% tryptone, 0.005% yeast extract, 0.005% sodium chloride, 0.015% Agar) at 25◦C. A second isolate of P. protegens PF-5 was obtained that had been routinely maintained of 100µg/ml ampicillin, which is a common laboratory practice. This isolate was examined to determine whether maintenance on ampicillin affects the resistance of P. protegens PF-5, and will be referred to as P. protegens PF-5A. Single colonies were re-suspended in equal parts 30% glycerol and M9 salts and held at -80◦C for long term storage.

### Antibiotic Treatments

Three antibiotics were selected for this study, each with different modes of action: kanamycin; tetracycline; and ciprofloxacin. Kanamycin is an aminoglycoside antibiotic which binds to the 30S ribosomal subunit and inhibits prevents protein synthesis (Misumi and Tanaka, 1980). Tetracycline is a polyketide antibiotic that is similar to kanamycin in that it binds to the 30S ribosomal subunit, however it prevents aminoacyl-tRNAs attaching to the ribosome, which in turn prevents addition of amino acids to growing polypeptide chains (Chopra and Roberts, 2001). Ciprofloxacin is a second generation fluoroquinolone used to treat a broad spectrum of infections. It inhibits DNA gyrase, which in turn prevents DNA replication (Lebel, 1988).

### Determination of Minimum Inhibitory Concentration

The minimum inhibitory concentration (MIC) was determined for each isolate against the three antibiotics following established methodology (Wiegand et al., 2008). MICs were determined in microtitre trays containing a serial dilution of the relevant antibiotic in Luria-Bertani medium (0.01% tryptone, 0.005% yeast extract, 0.005% sodium chloride). Wells were inoculated with bacteria prepared from an overnight culture and diluted to an optical density of 0.01. The concentration of antibiotic in test wells ranged from 32 to 0.0156 mg/L for ciprofloxacin and 512– 0.0156 mg/L for tetracycline and kanamycin. A growth control containing only the suspension of bacteria and a sterility control containing only medium were included on each plate. Plates were incubated at 25◦C for 24 h and then the optical density was read on a Pherastar FS spectrometer at 540 nm. Relative optical density was plotted against antibiotic concentration to determine the MICs, which were defined as no visible growth in the wells.

To determine statistical significance of differences in MIC, a One-Way analysis of variance (ANOVA) was performed. Growth data were expressed as the ratio of growth in the presence of antibiotics against growth in the control. This standardized the data prior to the ANOVA.

### DNA Extraction

DNA was extracted from bacterial cultures using a bead-beating method (Yeates and Gillings, 1998; Gillings, 2014). Briefly, a single, well isolated colony from an overnight culture was resuspended in a lysing matrix tube with sodium phosphate buffer and MT buffer (MP Biomedicals) or with CLS-TC buffer (MP Biomedicals). Preliminary testing indicated no significant difference between sodium phosphate/MT buffer and CLS-TC buffer, therefore CLS-TC buffer was used for the remainder of the study as it was more economical. Cells were physically lysed by treatment in a FastPrep FP120 (BIO 101 Savant) machine for 30 s at 5.5 m/s before being centrifuged in an Eppendorf 5417C, for 5 min at 14,000 g. Protein precipitation, binding, washing and subsequent elution of DNA in TE buffer were as previously described (Yeates and Gillings, 1998; Gillings, 2014). Purified DNA was stored at −20◦C.

### Repetitive Element PCR

DNA fingerprints were generated using ERIC-PCR, REP-PCR or BOX-PCR (Versalovic et al., 1991; Martin et al., 1992) with the modifications previously outlined (Gillings and Holley, 1997). One <sup>µ</sup>L of DNA was mixed with 9µL of Genereleaser™ (Bioventures Inc.) in a 0.5 mL PCR strip tube, and heated on high for 7 min in a 650 W microwave oven with a microwave sink. Tubes were then held at 80◦C for 5 min in an Eppendorf Master Cycle Epigradient S PCR machine, before 40µL of PCR master mix was mixed into each tube. The PCR master mix per reaction was as follows: 11µL PCR water, 25µL GoTaq <sup>R</sup> white (Promega), 2.5µL 25 mM MgCl2, 0.5µL 1 mg/ml RNAse, 1µL 50µM of the relevant rep-PCR primer. Negative controls containing Genereleaser™ only and water only were included in each PCR. The appropriate PCR cycle was then performed (**Table 1**). BOX, ERIC, and REP primers were synthesized by Sigma-Aldrich Inc.

### Agarose Electrophoresis

PCR products were separated on 2% agarose gels poured in Tris-Borate-EDTA (TBE) buffer (Russell and Sambrook, 2001). DNA samples were loaded with one quarter volume of bromophenol blue loading dye (0.45 M Tris-borate, 0.01 EDTA, 40% sucrose, 0.25% bromophenol blue). A 100 base pair ladder (Crown Scientific) was included on each gel. Gels were run in TBE at 110 v for 50–80 min. Gels were stained with GelRed™ (Biotium) and DNA visualized under UV light. Gel images were captured using a Gel logic 2200 PRO camera and Carestream MI computer software.

### Serial Plating Experiments

A single, well isolated colony of each species was chosen to (as far as possible) eliminate any extant variation amongst cells. This single colony was then used to inoculate the control LB agar plates; LB plates containing 1/10 the MIC for kanamycin; LB plates containing 1/10 the MIC for tetracycline; and LB plates containing 1/10 the MIC for ciprofloxacin, each in triplicate

TABLE 1 | Thermal cycling programs and primers used to generate DNA fingerprints using Rep-PCR.




(**Table 2**). Plates were incubated at 25◦C for 48 h, referred to here, for convenience, as one generation.

After incubation for 48 h, a single well-separated colony from each plate was used to continue the serial plating. After five generations, three single colonies were randomly selected from each plate for DNA extraction and PCR analysis, the first of these three would also be used to continue the serial plating. Repetitive Element PCRs were carried out to monitor changes in DNA patterns and to monitor for possible contamination of the cultures.

### DNA Banding Analysis

Images captured of the gels were analyzed to identify changes in the banding patterns, indicative of changes in the genome of the sample. Changes were scored against a control profile to calculate the similarity coefficient (F) using the formula devised by Nei and Li (1979):

$$\mathbf{F} = 2\mathbf{N\_{xy}}/(\mathbf{N\_x} + \mathbf{N\_Y})$$

Where Nx and Ny are the number of bands in lane x and lane y respectively and Nxy is the number of bands that lane x and lane y share. Samples with an F-value of 1 are identical while a value of 0 indicates no similarity. Scoring of the bands was carried out blind by an individual not involved in the Rep-PCR process, to remove the possibility of bias. The F-values for the various antibiotic treatments were plotted as a scatter graph to illustrate the spectrum of variation.

### Changes in Colony Morphology

To examine colony morphology at the end of the experiment, colonies of all lines from generation 40 were streaked onto LB agar plates and incubated for 48 h at 25◦C. Images of single colonies were captured using a Motic BA300 compound microscope with a 4x lens, mounted with a Moticam 2 2.0MP camera and were analyzed using DigiLabII-C and Motic Images Plus 2.0 computer programs.

### Results

### Colony Morphology Changes

Images captured of colonies at generation 40 show significant morphological changes between treatment groups. The three control lines of P. aeruginosa PA14 displayed no significant changes, kanamycin line 2, tetracycline lines 2 and 3, and ciprofloxacin line 3 exhibited significant changes to their colony morphology (**Figure 1**). The three control lines of P. protegens PF-5 displayed no significant changes, kanamycin line 3 and tetracycline line 3 exhibited significant morphological differences. The three ciprofloxacin lines were relatively unchanged (Figure S1). The three control lines and three tetracycline lines of P. protegens PF-5A had similar colonies. All three kanamycin lines had significantly changed colonies, as had lines 2 and 3 of the ciprofloxacin treatment (Figure S2).

### Detectable Genome Changes

BOX, ERIC, and REP-PCRs were carried out to detect genome changes. The basis of these PCRs is explained in Gillings and Holley (1997), but, in brief, relies on amplification of regions between two random, but reproducible priming sites. Consequently, amplicons are sensitive to both mutations in the priming sites and indels across the amplified regions. After testing both species with ERIC, REP and BOX primers, BOX-PCR was determined as the best method to examine changes. BOX-PCRs were conducted on triplicates of all lines every five generations. Experimental lines often exhibited changes in banding patterns, while the control lines remained the same, indicating that the changes were due to exposure to 1/10 MIC antibiotics (**Figure 2**). Changes were apparent after as few as five passages (evidence not presented), and increased in frequency as the experiment progressed, until they were present in the majority of experimental lines after 40 passages (**Figure 2**, Figures S3–S6).

Two features were notable in the lines exposed to subinhibitory antibiotic concentrations. In general, polymorphisms were commonly exhibited in experimental lineages, and often, replicates from single lineages exhibited diversity, demonstrating an ongoing instability within each generation. Further, similar changes to banding patterns were often observed in independent lineages, suggesting that similar events (such as transpositions or prophage activation) were being promoted within independent lines by the antibiotic treatment (**Figure 2**).

To determine the degree of polymorphism amongst the individual experiments, F statistics were calculated. A scatter plot of the F-statistics shows that control lines maintained a uniform BOX-PCR pattern across all three bacterial isolates (PA14, PF-5, and PF-5A) for the 40 generations of the experiment (**Figure 3**). Amongst the lineages treated with sub-inhibitory antibiotic concentrations, only the kanamycin treatment of PA14 maintained a stable BOX-PCR pattern. All other treatments generated polymorphic banding patterns in at least some of the replicates. The approximate degree to which polymorphisms were generated was in the order of Kan < Tet < Cipro (**Figure 3**).

### Changes in the MIC

The MIC of each line was determined at passage 40 in order to detect any significant differences in MICs from the control line and from the starting MIC. There were no significant differences in the MIC of P. aeruginosa PA14 for any of the treatment lines. In contrast, there were some significant differences in the MIC of P. protegens PF-5 and P. protegens PF-5A. A representative sample of MIC graphs are displayed in **Figure 4**. **Figure 4A** shows

the MIC for ciprofloxacin for all control and experimental lines of P. protegens PF5. One line of P. protegens PF5 that had been exposed to 1/10 the MIC of ciprofloxacin over the serial plating experiment exhibited a 10-fold increase in MIC for ciprofloxacin (DF = 11, F = 11.94, P < 0.0001). A similar phenomenon was seen in P. protegens PF5 (**Figure 4B**) and P. protegens PF5A (**Figure 4C**) when tested on kanamycin. All six lines that had been exposed to kanamycin over the serial plating experiment had four to eight fold increases in their MIC for kanamycin (DF = 11, F = 1.96, P > 0.05 and DF = 11, F = 46.04, P < 0.0001 respectively). Similar tests conducted on a subset of the kanamycin treated lines at passage 20 did not detect any elevation in MIC.

### Discussion and Conclusions

The role of antibiotics as environmental pollutants is attracting more attention, as more concern is being raised about their effects at sub-inhibitory concentrations (Gillings and Stokes, 2012; Andersson and Hughes, 2014). Specific issues include their potential effects on environmental microorganisms, and their potential for triggering complex interactions with the environmental resistome, thereby generating new opportunistic pathogens of relevance to human health (Gillings, 2013). Here we set out to test whether sub-inhibitory concentrations of antibiotics affect the genotype and phenotype of representative clinical and environmental pseudomonads.

BOX-PCR was performed on generation 40 Pseudomonas protegens. Lanes are labeled as follows: m = 100 bp ladder. Antibiotic treatments are noted as independent lines within each treatment (1, 2, or 3). Three colonies were tested from each line. For further examples see Supplementary Material (Figures S3–S6).

P. aeruginosa and P. protegens were serially plated on agar containing 1/10 the experimentally determined MIC for representatives of three antibiotic classes. This antibiotic concentration was chosen because it induces maximum transcriptional activity (Davies et al., 2006). Exposure to 1/10 the MIC for the panel of antibiotics tested had significant genotypic and phenotypic effects.

Effects on the genomes were immediate and readily detectable. Changes to rep-PCR DNA fingerprints could be detected after as few as five serial transfers on sub-inhibitory antibiotic

concentrations. This result is even more remarkable, since BOX-PCR is a fairly insensitive measure of genomic variation, although it generates highly reproducible DNA fingerprints. In this series of experiments, the BOX assay sampled between 15 and 20 kb of DNA, amounting to less than 0.3% of the ∼7 Mb pseudomonad genome. If the genome changes are similar in the un-sampled portion of genome, sub-inhibitory antibiotic concentrations are having a widespread and significant effect on DNA sequence, genome architecture, or both.

Sub-inhibitory antibiotic concentrations also induced phenotypic changes. After 40 generations of serial transfer, many of the experimental lines exhibited changes in colony morphology. Perhaps of most significance, all six lines of P. protegens maintained on 1/10 the MIC for kanamycin showed up to eight-fold elevation in their MICs for kanamycin by 40 generations. Similarly, one line held on 1/10 MIC for ciprofloxacin also showed an elevated ciprofloxacin MIC by 10-fold. Sub-lethal ciprofloxacin exposure has previously been shown to induce resistance in hypermutable strains of P. aeruginosa (Jørgensen et al., 2013). Whether the resistance observed in our experiment is also mediated by mutations in gyrA or gyrB will have to await sequence analysis.

The changes in MIC we observed are not likely to be the result of selection on pre-existing mutations in the single colony we used to initiate each experiment. A suspension of a single, well isolated colony was used as inoculum for both control and experimental lines of the three pseudomonads tested (PA14, PF-5, and PF-5A). All six kanamycin treated lines (three each for the two independent strains of P. protegens) showed elevated MICs for kanamycin by the 40th passage. One line from the ciprofloxacin treatments also showed a ten-fold increase in MIC. For these outcomes to have arisen from pre-existing mutants in the generation zero colony, each of the three kanamycin lines for both strains of P. protegens must have been the recipient of an appropriate mutant cell, as must have been the ciprofloxacin lineage. All of these putative mutations must have arisen during the growth of the time zero colonies from a single cell, which seems unlikely. Further, testing of a subset of the lines at passage 20 did not detect any increase in MIC in the kanamycin treated, or any other lines. By passage 20, any pre-existent mutant should have gone to fixation. The most parsimonious explanation for our results is that the changes in MIC were due to de-novo mutation.

If our findings are generally applicable, it suggests that similar phenotypic and genotypic changes will occur in all environments where antibiotics reach concentrations of 1/10 the MIC, and that these effects will potentially apply to all members of the environmental microbiota.

The effects of sub-inhibitory antibiotic concentrations observed in our experiments might be driven by the bacterial SOS response, which is known to induce processes that increase mutation, transposition and recombination rates (Gillings, 2013; Andersson and Hughes, 2014). Certainly, ciprofloxacin is a potent inducer of the SOS response, and generated the most extreme changes in BOX fingerprints observed in our experiments. Aminoglycosides such as kanamycin also induce the SOS response, but here tetracycline had an even greater effect on genomic architecture as assessed by BOX-PCR. However, there is no evidence that the diversity we observed is entirely due to the SOS response. The advantage of the approach we have taken here is that all mechanisms that generate variation, including the SOS response, and other potentially novel mechanisms, can be captured.

The concentrations of antibiotics used here may represent typical of levels of antibiotic pollution. There is limited knowledge about the concentrations of antibiotics found in the environment, however it is now known that antibiotics can persist in the environment longer than previously thought. The time that an antibiotic can persist in the environment differs depending on the class of antibiotic and the environmental conditions. Closed bottle tests provide a simple way to measure the biodegradability of antibiotics and indicate whether or not the antibiotic will readily degrade. Classes of antibiotics such as the β-lactams, tetracyclines, macrolides, lincosamides, penicillin, aminoglycosides, carbapenems, nitroimidazoles, polyeneantimycotics, quinolones, sulphonamides, and glycopeptides have all been found to persist over a 28 day testing period (Al-Ahmad et al., 1999; Alexy et al., 2004). High temperatures and exposure to UV light can cause degradation of some antibiotics. Fluoroquinolone antibiotics can degrade in sunlight, however they are readily absorbed onto sediments, where they have been documented persisting up to 80 days with less than 1% of degradation (Marengo et al., 1997). It would be convenient if resistant organisms destroyed or inactivated antibiotics, however the mechanisms that usually allow resistance involve mutation of binding sites or efflux pumps, meaning that antibiotics are not physically altered and may persist in the environment (Levy, 2002). Following one application of manure, antibiotics and antibiotic resistance genes can persist in the soil for approximately 6 months, depending on environmental conditions, during which time it could be dangerous to consume products that have had direct contact with the soil (Marti et al., 2014). Given the significant time frame in which antibiotics can persist in the environment it is highly likely that they will exist at concentrations near 1/10 the MIC.

The concentration at which antibiotics may occur in the environment is affected by several factors: substrate, proximity to source of antibiotics, environmental conditions and the antibiotics themselves. Testing of rivers and oceans have detected the presence of antibiotics, most notably sulphonamides and quinolones which were found at high concentrations in a number of environments. Sulphonamides were detected in water (0.86– 1563µg/L) (Hirsch et al., 1999; Luo et al., 2011; Li et al., 2012; Zhang et al., 2013) and quinolones were detected in sediments and plants (65.5–1166 and 8.37–6532 µg/kg, respectively) (Li et al., 2012). The antibiotic concentration of 1/10 the MIC easily falls into the ranges of antibiotic pollution detected in these waterways, indicating that the results of this study are likely to represent the rates of mutation and recombination taking place in environments polluted by antibiotics.

Very small concentrations of common antibiotics can induce significant genotypic and phenotypic changes in bacterial species. Given the huge quantities of antibiotics that are entering the environment, it is likely that this antibiotic pollution is generating antibiotic resistant organisms that may be a source of newly emerging opportunistic pathogens. These may then pose significant threats to human and animal life. Changes need to be made at every level of antibiotic use, by the individual, in medical practice, in pharmaceutical production, government monitoring and waste treatment, otherwise modern medicine is at a risk of facing a post antibiotic era where infections are harder, and in some cases, impossible to treat.

### Acknowledgments

We would like to gratefully acknowledge Marita Holley and Hilal Varinli for their technical support and assistance throughout the course of this study.

### Supplementary Material

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmicb. 2015.00803

# References


of Pseudomonas aeruginosa. Antimicrob. Agents Chemother. 57, 4215–4221. doi: 10.1128/AAC.00493-13


of antimicrobial substances. Nat. Protoc. 3, 163–175. doi: 10.1038/nprot. 2007.521


**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 Chow, Waldron and Gillings. 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.

# **Multidrug resistance found in extended-spectrum beta-lactamase-producing** *Enterobacteriaceae* **from rural water reservoirs in Guantao, China**

*Hongna Zhang <sup>1</sup> \*, Yufa Zhou 2,3, Shuyuan Guo <sup>1</sup> and Weishan Chang <sup>1</sup> \**

*<sup>1</sup> College of Animal Science and Technology, Shandong Agricultural University, Taian, China, <sup>2</sup> College of Animal Science and Technology, Shanxi Agricultural University, Taigu, China, <sup>3</sup> Animal Husbandry Bureau of Daiyue District, Taian, China*

### *Edited by:*

*Justin R. Seymour, University of Technology, Sydney, Australia*

### *Reviewed by:*

*Lisa Moore, University of Southern Maine, USA Michael Gillings, Macquarie University, Australia*

### *\*Correspondence:*

*Hongna Zhang and Weishan Chang, College of Animal Science and Technology, Shandong Agricultural University, Daizong Street 61, Taian 271000, China zhang186982@126.com; weishanchang@126.com*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

> *Received: 31 January 2015 Accepted: 18 March 2015 Published: 31 March 2015*

### *Citation:*

*Zhang H, Zhou Y, Guo S and Chang W (2015) Multidrug resistance found in extended-spectrum beta-lactamase-producing Enterobacteriaceae from rural water reservoirs in Guantao, China. Front. Microbiol. 6:267. doi: 10.3389/fmicb.2015.00267* Extended-spectrum beta-lactamase (ESBL)-producing *Enterobacteriaceae* have been isolated from humans and animals across the world. However, data on prevalence of ESBL-producing *Enterobacteriaceae* from rural water reservoirs is limited. This study aimed to isolate and characterize ESBL-producing *Enterobacteriaceae* in rural water reservoirs in Guantao, China. ESBL-producing *Enterobacteriaceae* were found in 5 (16.7%) of 30 sampled rural water reservoirs. Sixty-six individual isolates expressing an ESBL phenotype were obtained in the present study. Species identification showed that 42 representatives of *Escherichia coli*, 17 *Klebsiella pneumoniae*, 4 *Raoultella planticola*, and 3 *Enterobacter cloacae*. Twenty isolates contained a single *bla* gene, including CTX-M (17 strains), TEM (2 strains), and SHV (1 strain). Forty-six isolates contained more than one type of beta-lactamase genes. ESBL-producing *Enterobacteriaceae* isolated in this study were all multidrug resistant. These findings indicated that the serious contamination of ESBL-producing *Enterobacteriaceae* in rural water reservoirs existed in Guantao, China.

### **Keywords: ESBL,** *Enterobacteriaceae***, rural water reservoirs, multidrug resistance,** *bla* **genes**

# **Introduction**

The rational use of antibiotics helps control infectious diseases of humans and animals. Abuse and overuse of antibiotics in clinical practice has selected drug resistant bacteria and "superbugs" (Thompson et al., 2007; Yong et al., 2009; Pruden et al., 2012). Extended-spectrum beta-lactamases (ESBLs), resulting from amino acid substitutions in TEM-1, TEM-2, and SHV-1 enzymes were described in the 1980s and 1990s (Bush and Jacoby, 2010). ESBLs can hydrolyze penicillins, oxyimino-cephalosporins (e.g., cefotaxime, ceftazidime, ceftriaxone, cefuroxime, cefepime) and aztreonam but not cephamycins (e.g., cefoxitin, cefotetan) or carbapenems (Bush and Jacoby, 2010; El Salabi et al., 2013). ESBLs are predominantly found among *Enterobacteriaceae*, which are inhabitants of intestinal flora and important pathogens in nosocomial and community settings (Laurent et al., 2008; Azap et al., 2010; Song et al., 2011; Kang et al., 2013).

Extended-spectrum beta-lactamase-producing *Enterobacteriaceae* can spread between humans via contaminated food or water (Oteo et al., 2010; Piednoir et al., 2011) and acquire resistance to antibiotics by plasmids, transposons or other mobile vectors that carry resistance elements (Oteo et al., 2010; Peirano et al., 2012). Water environments are considered as important reservoirs for

resistance genes (Gao et al., 2009), and maybe play an important role in transfer of drug-resistant genes between bacteria (Malakoff, 2002; Kummerer, 2004). More importantly, once ESBL-producing *Enterobacteriaceae* enter the intestine of humans and animals via drinking water, these bacteria could lead to the spread of resistance genes and to serious infections.

To date, numerous studies on ESBL-producing *Enterobacteriaceae* isolated from water environments have focused on wastewaters of hospitals and animal farms, and waters from rivers and lakes (Cabello et al., 2013; Varela and Manaia, 2013; Yang et al., 2013; Zurfluh et al., 2013; Haque et al., 2014). However, data on ESBL-producing *Enterobacteriaceae* isolated from drinking water in rural areas is very limited. In China, the main drinking sources for rural residents in many villages are water reservoirs. Therefore, the present study was conducted to describe the isolation and characterization of ESBL-producing *Enterobacteriaceae* in rural water reservoirs in Guantao, China.

# **Materials and Methods**

### **Sampling Sites and Water Sample Collection**

Between July and September of 2013, water sampling was conducted in Guantao, China (**Figures 1 and 2**). Five samples each were collected at six locations for a total of 30 samples. The water samples were collected from 50 cm below the water surface using sterile bottles (100 ml/bottle, one bottle/each reservoir). The collected water samples were stored on ice and immediately transported to our lab for further analyses within 3 h.

### **Microbiological Analysis**

Hundred milliliters of water was filtrated through a sterile 0.45µm membrane (Millipore, Billerica, MA, USA), and then the filters were incubated in 20 ml of enterobacteria enrichment (EE) Broth (Becton Dickinson, Heidelberg, Germany) at 37°C for 24 h. One loopful of enrichment cultures was spread onto chromogenic Brilliance ESBL agar (Oxoid, Hampshire, UK) and incubated at 37°C for 24 h. The colonies with different color and morphology

**FIGURE 2 | Sampling sites.** Blue triangles **(A–C,F)** represent the communities where no ESBL-producing *Enterobacteriaceae* were detected in water samples. Red triangles **(D, E)** represent the communities where ESBL-producing *Enterobacteriaceae* were found in water samples. Yellow circle represents urban areas of Guantao County.

were picked and sub-cultured on sheep blood agar for 24 h at 37°C (Huang et al., 2010). Conventional biochemical methods and API ID 32 E (bioMérieux, Marcy l'Etoile, France) were used to identify the isolates. If isolates showed doubtful results, they were subjected to genetic identification based on sequencing of *rpoB* gene fragments (Mollet et al., 1997).

### **Antimicrobial Susceptibility Testing and ESBL Confirmation**

According to the protocols of the Clinical and Laboratory Standards Institute (CLSI, 2011), the disk diffusion method was used to test susceptibility of the isolates against 17 antimicrobial agents. The tested antibiotics were: ampicillin (AMP), cefaclor (CEC), cefazolin (CFZ), cefepime (FEP), cefotaxime (CTX), ceftazidime (CAZ), ceftriaxone (CRO), cefuroxime (CXM), aztreonam (AZT), ciprofloxacin (CIP), gentamicin (GEN), imipenem (IPM), ofloxacin (OFX), piperacillin (PIP), amikacin (AMK), chloramphenicol (CHL) and tetracycline (TET). According to the manufacturer's protocols, Etest-ESBL strips (bioMérieux, Marcy l'Etoile, France) were used to confirm ESBL production. Isolates showing resistance to three or more antibiotic classes were defined as multidrug resistant (MDR). *E. coli* ATCC 25922 and *K. pneumoniae* ATCC 700603 were used as quality control strains.

### **Polymerase Chain Reaction (PCR) to Detect** *bla* **Genes**

The DNA of the isolates confirmed for producing ESBLs was extracted separately using a DNA extraction kit (Biospin plasmid

**TABLE 1 | Species composition of ESBL-producing** *Enterobacteriaceae* **from water samples.**


extraction, Bioflux, Japan). According to previously published work, PCR was used to detect *bla*TEM, *bla*CTX-M, and *bla*SHV genes using specific primers (Chen et al., 2010).

# **Results**

## **Detection of ESBL-producing** *Enterobacteriaceae*

Extended-spectrum beta-lactamase-producing *Enterobacteriaceae* were detected in five water reservoirs of two rural communities (D: 2, E: 3). The five water reservoirs were all located close to chicken farms (approximately 12–15 m). No ESBL-producing *Enterobacteriaceae* were found in the other reservoirs, which were far away from rural villages and animal farms (approximately 1.0–1.5 km).

Sixty-six different isolates exhibiting an ESBL phenotype were obtained (D: 28, E: 38). The results of species identification showed that 42 *E. coli*, 17 *K. pneumoniae*, 4 *Raoultella planticola*, and 3 *Enterobacter cloacae* (**Table 1**).

### **Antibiotic Susceptibility of ESBL-producing** *Enterobacteriaceae*

Extended-spectrum beta-lactamase-producing *Enterobacteriaceae* isolates displayed similar drug-resistant trends. Nearly all ESBL-producing *Enterobacteriaceae* were resistant to the first- and second-generation cephalosporins (cefazolin, cefaclor, cefuroxime). These isolates were also resistant to the thirdgeneration cephalosporins: cefotaxime (91.5%), ceftriaxone (67.9%), and ceftazidime (31.1%). Moreover, 48.1% of the isolates were resistant to cefepime (the fourth-generation cephalosporin), 51.9% to aztreonam (a monocyclic β-lactam antibiotic), and 89.6% to ampicillin.

Extended-spectrum beta-lactamase-producing *Enterobacteriaceae* were resistant to non-β-lactam antibiotics: resistant to ciprofloxacin (78.3%), gentamicin (60.4%), ofloxacin (76.4%), piperacillin (68.9%), chloramphenicol (55.7%), and tetracycline (65.1%). But the majority of ESBL-producing *Enterobacteriaceae* isolates were susceptible to amikacin (95.3%) and imipenem (97.2%) (**Figure 3**).

### **Characterization of** *bla* **Genes in ESBL-producing** *Enterobacteriaceae*

All 66 ESBL-producing *Enterobacteriaceae* carried *bla* genes. Among 66 ESBL-carriers, 20 strains carried only one *bla* gene (20/66, 30.3%), including 17 carrying *bla*CTX-<sup>M</sup> (17/66, 25.8%), 2 isolates carrying *bla*TEM (2/66, 3.0%), and 1 carrying *bla*SHV (1/66, 1.5%). The other 46 isolates carried at least two *bla* genes (46/66,

**FIGURE 3 | Antibiotic resistance rates of ESBL-producing**

*Enterobacteriaceae.* R: Resistance; I: intermediate; S: susceptible; AMP, ampicillin; CEC, cefaclor; CFZ, cefazolin; FEP, cefepime; CTX, cefotaxime; CAZ, ceftazidime; CRO, ceftriaxone; CXM, cefuroxime; AZT, aztreonam; CIP, ciprofloxacin; GEN, gentamicin; IPM, imipenem; OFX, ofloxacin; PIP, piperacillin; AMK, amikacin; CHL, chloramphenicol; TET, tetracycline.

### **TABLE 2 |** *bla* **gene types of ESBL-producing** *Enterobacteriaceae* **from water samples.**


69.7%), including 38 carrying *bla*TEM+CTX-<sup>M</sup> (38/66, 57.6%), 6 carrying *bla*SHV+CTX-<sup>M</sup> (6/66, 9.1%), and the other 2 carrying three *bla* genes (2/66, 3.0%) (**Table 2**).

# **Discussion**

In this study, ESBL-producing *Enterobacteriaceae* were detected in five out of 30 rural reservoirs (5/30, 16.7%), but not found in the other water reservoirs far away from villages and animal farms. This suggests that ESBL-producing *Enterobacteriaceae* are being shed into reservoirs from animal farms and anthropogenic activities.

Extended-spectrum beta-lactamase-producing *Enterobacteriaceae* isolates in this study were MDR. These isolates showed high resistance against the third-generation cephalosporins: cefotaxime (91.5%), and ceftriaxone (67.9%). Importantly, 48.1% of these isolates were resistant to cefepime, the fourth-generation cephalosporin. In addition, antibiotics resistance rates of these isolates to non-β-lactam antibiotics were also worrisome. But 95.3% and 97.2% of these isolates were respectively susceptible to amikacin and imipenem, which may be related with relatively low use of these medicines in this region.

In mainland China, previous investigations about ESBLproducing *Enterobacteriaceae* in water bodies and food-producing animals showed that *bla*CTX-<sup>M</sup> gene was the dominant ESBL producer (Jin and Ling, 2006; Chen et al., 2010; Rao et al., 2014). Our data also identified *bla*CTX-<sup>M</sup> gene as an important ESBL producer. Additionally, 46 out of 66 ESBL-producing *Enterobacteriaceae* isolates carried at least two *bla* genes and *bla*CTX-<sup>M</sup>+TEM

has become the dominant phenotype of ESBL, which was different from the previous result in clinical isolates (Zhang et al., 2014).

There were some limitations in this study: water sampling was carried out only in 30 rural water reservoirs, so the results may not be representative of the whole area; sequence analyses of *bla* genes encoding TEM, CTX-M, and SHV were not further conducted

# **References**


to identify variants of these types of enzymes; the risk factors associated with rural reservoir water carriage of ESBL-producing *Enterobacteriaceae* were not further analyzed.

In summary, these findings indicated that the contamination of ESBL-producing *Enterobacteriaceae* in rural water environments existed in Guantao, China, and the pollution may be closely related to local animal farms and anthropogenic activities.


**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 Zhang, Zhou, Guo and Chang. 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.*

# Vertical Segregation and Phylogenetic Characterization of Ammonia-Oxidizing Bacteria and Archaea in the Sediment of a Freshwater Aquaculture Pond

Shimin Lu1, 2, Xingguo Liu<sup>1</sup> \*, Zhuojun Ma<sup>3</sup> \*, Qigen Liu<sup>2</sup> , Zongfan Wu<sup>4</sup> , Xianlei Zeng1, 2 , Xu Shi <sup>1</sup> and Zhaojun Gu<sup>1</sup>

### Edited by:

*Mark Vincent Brown, University of New South Wales, Australia*

### Reviewed by:

*Lucas Stal, Netherlands Institute of Sea Research, Netherlands Wei Xie, Tongji University, China*

### \*Correspondence:

*Xingguo Liu liuxingguo@fmiri.ac.cn; Zhuojun Ma mazj@cafs.ac.cn*

### Specialty section:

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

Received: *23 May 2015* Accepted: *21 December 2015* Published: *20 January 2016*

### Citation:

*Lu S, Liu X, Ma Z, Liu Q, Wu Z, Zeng X, Shi X and Gu Z (2016) Vertical Segregation and Phylogenetic Characterization of Ammonia-Oxidizing Bacteria and Archaea in the Sediment of a Freshwater Aquaculture Pond. Front. Microbiol. 6:1539. doi: 10.3389/fmicb.2015.01539*

*<sup>1</sup> Fishery Machinery and Instrument Research Institute, Chinese Academy of Fishery Sciences, Shanghai, China, <sup>2</sup> College of Fisheries and Life, Shanghai Ocean University, Shanghai, China, <sup>3</sup> Chinese Academy of Fishery Sciences, Beijing, China, <sup>4</sup> Tongren Municipal Agricultural Commission (Government, Public), Tongren, China*

Pond aquaculture is the major freshwater aquaculture method in China. Ammoniaoxidizing communities inhabiting pond sediments play an important role in controlling culture water quality. However, the distribution and activities of ammonia-oxidizing microbial communities along sediment profiles are poorly understood in this specific environment. Vertical variations in the abundance, transcription, potential ammonia oxidizing rate, and community composition of ammonia-oxidizing bacteria (AOB) and ammonia-oxidizing archaea (AOA) in sediment samples (0–50 cm depth) collected from a freshwater aquaculture pond were investigated. The concentrations of the AOA *amoA* gene were higher than those of the AOB by an order of magnitude, which suggested that AOA, as opposed to AOB, were the numerically predominant ammonia-oxidizing organisms in the surface sediment. This could be attributed to the fact that AOA are more resistant to low levels of dissolved oxygen. However, the concentrations of the AOB *amoA* mRNA were higher than those of the AOA by 2.5- to 39.9-fold in surface sediments (0–10 cm depth), which suggests that the oxidation of ammonia was mainly performed by AOB in the surface sediments, and by AOA in the deeper sediments, where only AOA could be detected. Clone libraries of AOA and AOB *amoA* sequences indicated that the diversity of AOA and AOB decreased with increasing depth. The AOB community consisted of two groups: the *Nitrosospira* and *Nitrosomonas* clusters, and *Nitrosomonas* were predominant in the freshwater pond sediment. All AOA *amoA* gene sequences in the 0–2 cm deep sediment were grouped into the *Nitrososphaera* cluster, while other AOA sequences in deeper sediments (10–15 and 20–25 cm depths) were grouped into the *Nitrosopumilus* cluster.

Keywords: freshwater aquaculture pond, ammonia-oxidizing archaea, ammonia-oxidizing bacteria, sediment, depth distribution

# INTRODUCTION

China is the world's largest producer, consumer, processor, and exporter of fish. China alone accounts for >60% of the global aquaculture volume and roughly half of the global aquaculture value (Cao et al., 2015). Currently, there are 2,623,180 ha of freshwater aquaculture ponds, and freshwater pond culturing is the major culture method in China (National Bureau of Statistics of China 2014, http://www.stats.gov.cn/english/statisticaldata/ Quarterlydata/). To obtain more benefits from aquaculture, higher stocking densities are becoming prevalent. At the same time, large residual feed and feces are deposited into aquaculture sediments (Cao et al., 2015). A large amount of ammonia will be produced and released into the aquaculture water during the mineralization of organic matter. Ammonia not only significantly contributes to the eutrophication of aquaculture pond ecosystems, but is also one of the most toxic substances in intensive fish farming (Ackefors and Enell, 1994; Randall and Tsui, 2002). The high concentration of ammonia in aquaculture water has become a limitation for pond culturing in China.

Nitrification, the biological conversion of ammonia (NH3) to nitrate via nitrite (NO<sup>−</sup> 2 ), is a key process in nitrogen cycling in aquatic ecosystems (Merbt et al., 2012). Currently, the oxidation of NH<sup>3</sup> to NO<sup>−</sup> <sup>2</sup> —the first and rate-limiting step of nitrification is considered to be conducted by ammonia-oxidizing bacteria (AOB) and ammonia-oxidizing archaea (AOA; Koops and Pommerening-Röser, 2001). AOB fall into two phylogenetic lineages within the β- and γ-Proteobacteria (Kowalchuk and Stephen, 2001) and mainly belong to the genera Nitrosomonas, Nitrosospira, and Nitrosococcus (Kowalchuk and Stephen, 2001; Wang et al., 2014a). Based on genomic level comparisons, AOA were classified into a newly proposed branching phylum of the Archaea, named the Thaumarchaeota (Pester et al., 2011), and a recent study showed that AOA could be grouped into five major clusters: the Nitrosopumilus cluster (also called group I.1a AOA), the Nitrosotalea cluster (also called group I.1a-associated AOA), the Nitrososphaera cluster (also called group I.1b AOA), the Nitrososphaera sister cluster, and the Nitrosocaldus cluster [also called thermophilic AOA (ThAOA); Pester et al., 2012]. AOB and AOA both contain a homologous ammonia monooxygenase (AMO) that is responsible for catalyzing the first step in ammonia oxidation. The amoA gene, encoding the alpha subunit of AMO, has been used widely as a functional gene marker for tracking ammonia oxidizers in environmental samples (Rotthauwe et al., 1997; Francis et al., 2005). Thus far, many studies of AOA and AOB have been conducted; however, most of them have focused on large ecosystems, such as soils (Leininger et al., 2006; Tourna et al., 2011), oceans (Wuchter et al., 2006; Horak et al., 2013), lakes (Ye et al., 2009; Auguet et al., 2011, 2012; Zhao et al., 2013), and rivers (Jin et al., 2011; Sun et al., 2013; Wang et al., 2014b). There is a lack of studies of AOA and AOB in aquaculture ponds. Pond environments are smaller in area and shallower in depth, have limited water circulation, and are subject to large depositions of feeding debris. Moreover, the hypolimnion dissolved oxygen concentration was very low, although the supersaturation of oxygen usually occurs in the top layer during the daylight period (Chang and Ouyang, 1988). The trophic status and sediment properties make freshwater aquaculture ponds a good model system for studying the vertical distribution of AOA and AOB.

The oxidation of ammonium mainly occurs in the pond sediments, probably because of photoinhibition, and we previously found a low abundance of ammonia-oxidizing microorganisms in freshwater aquaculture water throughout the year (Lu et al., 2015). Hence, the aims of the present study were to investigate the activity and biodiversity in different sediment layers of a selected freshwater aquaculture pond in East China, and to quantitatively assess its AOA and AOB. In order to better understand the vertical distribution of AOA and AOB, we also partially characterized the physical and chemical factors [pH, dissolved oxygen (DO), total organic carbon (TOC), ammonium (NH<sup>+</sup> 4 ) and NO<sup>−</sup> 2 ] of the sediment.

# MATERIALS AND METHODS

## Sediment Samples and Background

Samples were collected from a freshwater aquaculture pond located at the Research Center for Pond Ecosystem Engineering, Chinese Academy of Fishery Sciences [30◦ 56′ N, 121◦ 09′ E], Shanghai, China. The sampling pond had a surface area of ∼5000 m<sup>2</sup> and an average depth of about 1.6 m. Wuchang bream (Megalobrama amblycephala), grass carp (Ctenopharyngodon idella), silver carp (Hypophthalmichthys molitrix), and bighead carp (Hypophthalmichthys nobilis) were raised in the pond for commercial use from 2008 to 2011 and 2014 to 2015, and the production of fish was about 1200 kg km−<sup>2</sup> per year. From 2012 to 2013, the submerged plant Chara fragilis Desv. was widely cultivated, and crab, whose production was about 150 kg km−<sup>2</sup> per year, were raised in the sampled pond for commercial use. The sampled pond was dry during winter.

Three sediment cores (5 cm diameter and 50 cm depths) were collected from the aquaculture pond in October 2014 using a polyvinylchloride pipe. Then, the sediment cores were placed in sterile plastic bags, sealed, and transported to the laboratory on ice. Later, they were sectioned to 2 cm from 0 to 10 cm depths, and to 5 cm at 10–50 cm depths, and then we mixed the different cores from each sample for each depth. One portion was incubated to determine the ammonia oxidation activities immediately after arrival, another portion was used for an analysis of chemical components, and subsamples were stored at −80◦C for subsequent DNA and RNA extractions and molecular analysis.

# Chemical Analytical Procedures of Sediments

Ammonium (NH<sup>+</sup> <sup>4</sup> <sup>−</sup>N) and nitrite (NO<sup>−</sup> <sup>2</sup> −N) were extracted from the sediments with 2 M KCl and measured photometrically using Nessler's reagent, and spectrophotometrically using N-(1- Naphthyl)-ethylenediamine dihydrochloride, respectively (Hou et al., 2003; Lu et al., 2015). The pH of sediment was determined after mixing it with water at a ratio (sediment/water) of 1:2.5, and sediment organic matter was determined using a total carbon analyzer (Vario TOC, Elementar, Germany; Zhu et al., 2011). In July 2015, sediment samples were collected in the same pond as previously described, and the DO concentration in fresh sediments was measured immediately on a fishing boat using an OXY Meter S/N 5015 with an microelectrode sensor (OX-50µm, Unisense, Aarhus, Denmark), as described by Gundersen et al. (1998).

# Measuring the Potential Ammonia Oxidation Rate

Potential ammonia oxidation rates were measured using the chlorate inhibition method (Kurola et al., 2005). Briefly, 5.0 g of fresh sediment was added to 50 ml centrifuge tubes containing 20 ml of phosphate buffer solution (NaCl, 8.0; KCl, 0.2; Na2HPO4, 0.2; NaH2PO4, 0.2 g L−<sup>1</sup> ) containing 1 mM (NH4)2SO4. Potassium chlorate was added to the tubes to a final concentration of 10 mM to inhibit nitrite oxidation. The suspension was incubated with shaking (300 rpm) for 0.5 h at 25◦C in the dark; then, the suspension was incubated without shaking for 24 h at 25◦C in the dark; afterwards, nitrite was extracted with 5 ml of 2 M KCl and determined spectrophotometrically at 540 nm using N- (1-Naphthyl) ethylenediamine dihydrochloride. The potential ammonia oxidation rates were calculated based on the change in the nitrite concentrations.

# Nucleic Acid Extraction, Quantitative Polymerase Chain Reaction (qPCR), and Reverse Transcription

Extraction of DNA from the sediment samples was conducted, and two controls were performed to estimate the possible inhibition of qPCR performance by the co-extracted polyphenolic compounds or humic acids in the sediment, as described by Lu et al. (2015). Total RNA was extracted from the sediment samples using the E.Z.N.A. <sup>R</sup> Soil RNA Mini Kit (Omega Bio-Tek, Norcross, GA, USA) according to the manufacturer's instructions. RNA was reverse transcribed into cDNA using the PrimeScript RT Master Mix (Perfect Real Time; TaKaRa Biotechnology Dalian Co., Ltd., Dalian, China). Absence of contamination from DNA and chemical reagents was verified by performing the same reactions without reverse transcriptase or template, respectively. The obtained cDNAs were stored at −80◦C for further analysis. qPCR was used to estimate the abundance of ammonia-oxidizing microorganisms' amoA mRNA and DNA, as well as total bacterial and Crenarchaeota 16S rRNA genes. qPCR was performed using a SLAN real-time PCR detection system (Hongshi Medical Technology Co. Ltd., Shanghai, China). The primers and reaction conditions for qPCR are listed in **Table 1**. qPCRs were conducted in a total volume of 20µL containing 10µL of SYBR Premix Ex Taq II for the AOB amoA gene or SYBR Premix Ex Taq for the AOA amoA gene and total bacterial and Crenarchaeota 16S rRNA genes (Takara), 1µL of DNA template, and 0.2 mg mL−<sup>1</sup> bovine serum albumin (BSA). A negative control without DNA template was subjected to the same procedures to exclude or detect any possible contamination. After qPCR, the specificity of the amplification was verified by a melting curve analysis and agarose gel electrophoresis. All measurements were performed in triplicate.

Standard curves for qPCR were developed as described previously (Lu et al., 2015). External standard curves ranging from 10<sup>1</sup> to 10<sup>5</sup> copies per microliter of the archaeal and bacterial amoA genes or 10<sup>4</sup> to 10<sup>8</sup> copies of the bacterial and Crenarchaeota 16S rRNA genes were generated during the process of qPCR. Standard curve coefficients of variation and efficiencies were as follows: AOA (R <sup>2</sup> = 0.999, efficiency = 94.4%), AOB (R <sup>2</sup> = 0.999, efficiency = 90.9%), bacterial 16S rRNA (R <sup>2</sup> = 0.998, efficiency = 90.0%) and crenarchaeota 16S rRNA gene (R <sup>2</sup> = 0.999, efficiency = 82.7%). The results of the real-time PCR were expressed as the number of amoA or 16S rRNA gene copies g−<sup>1</sup> of sediment (dry weight).

# Cloning, Sequencing, and Phylogenetic Analysis

The purified PCR products were ligated and cloned using the pMD ™18-T Vector (Takara). In total, 105 and 76 clones of the AOA and AOB amoA gene PCR products, respectively, were successfully picked and sequenced. Operational taxonomic units (OTUs) were defined as sequence groups in which sequences differed by ≤2% for AOA and ≤3% for AOB. Neighbor-joining phylogenetic trees were constructed using MEGA 5.05 (Kumar et al., 2008).

# Statistical Analysis

The estimated coverage of the constructed amoA gene libraries was calculated as C = [1 − (n/N)] × 100%, where n is the number of unique OTUs and N is the total number of all clones in a library. Indices of the amoA genotype diversity (Shannon–Wiener, H), richness estimations (Chao index), and rarefaction analysis were calculated using DOTUR (Schloss and Handelsman, 2005). Correlations between AOA abundance and environmental factors and One-way analysis of variance (ANOVA) were analyzed using SPSS 16.0 software.

# Nucleotide Sequence Accession Numbers

The nucleotide sequences obtained in this study were deposited in the GenBank database under accession nos. KR081161– KR081236 for AOB and KR081056–KR081160 for AOA.

# RESULTS

# Abundances, Ammonia Oxidation Rates, and Expression of AOA and AOB

The vertical distribution profiles of TOC, NH<sup>+</sup> <sup>4</sup> <sup>−</sup>N, NO<sup>−</sup> <sup>2</sup> −N, and pH in every sediment core are shown in **Figure 1**. Briefly, the TOC, NH<sup>+</sup> <sup>4</sup> <sup>−</sup>N, and NO<sup>−</sup> <sup>2</sup> −N concentrations were high in 0–6 cm sediments, and decreased rapidly from 6 to 10 cm.

The concentrations of TOC, NH<sup>+</sup> <sup>4</sup> <sup>−</sup>N, and NO<sup>−</sup> <sup>2</sup> −N ranged from 1.81 ± 0.04 to 17.60 ± 0.85 g kg−<sup>1</sup> , 3.88 ± 0.45 to 64.09 ± 11.01 mg kg−<sup>1</sup> , and 0.05 ± 0.01 to 0.47 ± 0.16 mg kg−<sup>1</sup> , respectively. The pH ranged from 7.42 to 8.33, and the lowest and the highest value occurred at the 0–2 and 45–50 cm, respectively.


TABLE 1 | Primers used for PCR amplification for library construction and real-time PCR quantification.

The depth of the DO detection limit was 500µm, and the DO concentrations ranged from 0 to 48.01µmol L−<sup>1</sup> (**Figure 2**).

To detect the presence of AOA, AOB, as well as the Crenarchaeota and total bacterial, the amoA and 16S rRNA genes from sediment core samples were amplified. The results showed that the depth limits for detecting the AOA and AOB amoA genes were 25 and 6 cm, respectively, and that the concentrations of the AOA amoA gene (ranging from 6.82 ± 2.28 × 10<sup>4</sup> to 7.79 ± 3.88 × 10<sup>5</sup> ) were higher than those of the AOB (1.88 ± 0.39 × 10<sup>3</sup> to 3.60 ± 0.91 × 10<sup>4</sup> ) by an order of magnitude, which suggested that the AOA, as opposed to the AOB, were the numerically predominant ammonia-oxidizing organisms in the surface sediment (**Figure 3**). Additionally, a positive PCR product was obtained for the Crenarchaeota and total bacteria in every sample, and the 16S rRNA concentrations ranged from 4.94 ± 2.12 × 10<sup>6</sup> to 6.18 ± 1.33 × 10<sup>8</sup> , and 6.10 ± 0.36 × 10<sup>7</sup> to 1.62 ± 0.04 × 10<sup>11</sup> copies g−<sup>1</sup> , respectively (**Figure 3**). Linear relationships between different environmental factors and the amoA gene abundance of the AOA and AOB were characterized using Pearson's correlation coefficient. It was found that the abundance of AOA and AOB positively correlated with the TOC in the sediments (R <sup>2</sup> = 0.838, P < 0.01; R <sup>2</sup> = 0.852, P < 0.01), while the AOA and AOB abundances were negatively correlated with pH (R <sup>2</sup> = −0.755, P < 0.01; R <sup>2</sup> = −0.787, P < 0.05, respectively). No significant correlations were detected between the AOA and AOB abundances and the concentrations of NH<sup>+</sup> <sup>4</sup> <sup>−</sup>N and NO<sup>−</sup> <sup>2</sup> −N in sediments.

To obtain more detailed information about the AOA and AOB in the freshwater aquaculture pond sediments, the potential ammonia oxidation rate was obtained from every sample, and it ranged from 0.0014 ± 0.0001 to 0.0386 ± 0.0028 mg kg−<sup>1</sup> h −1 . The potential ammonia oxidation rates in 0–6 cm deep sediments were significantly higher than those in other sediment layers (P < 0.01; One-way ANOVA; **Figure 3A**). The expression of the AOA and AOB amoA genes was calculated using the abundance of the PCR products that were amplified from cDNAs. Despite the fact that the depth of the AOB detection limit was 6 cm, AOB amoA gene expression could be detected at 8–10 cm in sediment cores, and it was higher than that of AOA in 0–10 cm depths by 2.5–39.9-fold (**Figure 4**).

# Diversity of AOA and AOB

To investigate the diversity and community composition of ammonia-oxidizing populations, sediment layers with depths of

0–2 and 4–6 cm, as well as 0–2, 10–15 cm, and 20–25 cm, were selected for the construction of clone libraries of bacterial and archaeal amoA genes, respectively. Five clone libraries of the amoA gene were constructed to explore the diversity of AOB and AOA. The estimated coverage (C) of the five clone libraries ranged from 91 to 100%, which, together with the rarefaction analysis (**Figure 5**), indicated that the bacterial and archaeal amoA genotypes in the sediments could be well-represented by these clone libraries. As shown in **Figure 5A**, the OTU numbers, Chao estimate, and Shannon index of AOB in the 4– 6 cm deep sediment were all less than those of AOB at a 0– 2 cm depth, which indicated that the diversity of AOB decreased with increasing sediment depth. A phylogenetic analysis of bacterial amoA sequences suggested that the AOB community in aquaculture pond sediments consisted of two groups: the Nitrosospira and Nitrosomonas clusters (**Figure 6A**). The sequences related to Nitrosomonas spp. were predominant over those of Nitrosospira in AOB communities in the freshwater pond sediment.

An obvious variation in the AOA community and structure with sediment depth was also observed. As shown in **Figure 5B**, the diversity of AOA decreased with increasing sediment depth. All the AOA amoA gene sequences in 0–2 cm deep sediments were grouped into the Nitrososphaera cluster, while all the AOA sequences in 10–15 cm deep sediments, as well as a portion of the AOA sequences in the 20–25 deep sediments were grouped into a branch that belongs to the Nitrosopumilus cluster; the other AOA sequences in the 20–25 cm deep sediments were grouped into another branch of the Nitrosopumilus cluster. No sequences belonging to the ThAOA and Nitrosotalea clusters were detected (**Figure 6B**).

### DISCUSSION

## Abundances, Ammonia Oxidation Rates, and Expression of AOA and AOB

A significant positive correlation between the abundance of AOA and TOC was observed. This may indicate that AOA are able to assimilate organic substrates and thereby be able to grow mixotrophically or even heterotrophically. This view is supported by studies of archaeal isolates from soil and marine sediments (Tourna et al., 2011; Qin et al., 2014), although our results are quite different from those that showed a negative correlation between AOA abundance and TOC concentrations in the sediments of a eutrophic lake and river (Wu et al., 2010; Wang et al., 2014b).

Because of large depositions of feeding debris and feces in the aquaculture pond, the surface sediments were rich in organic substances and exhibited a high NH<sup>+</sup> <sup>4</sup> −N concentration (**Figure 1**). The abundance of the AOA amoA gene was one order of magnitude higher than that of the AOB in the surface sediment (0–6 cm depth), which suggested that AOA, as opposed to AOB, were the numerically predominant ammoniaoxidizing organisms in the surface sediment of the freshwater aquaculture pond. We observed the same phenomenon in 10 other Chinese freshwater pond sediments (Lu et al., 2015). Our result contradicts the concept that AOA prefer lower NH<sup>+</sup> 4 concentration environments because of their higher specific affinity for NH<sup>+</sup> 4 , whereas AOB prefer higher NH<sup>+</sup> 4 concentrations (Martens-Habbena et al., 2009; Habteselassie et al., 2013).

AOA, rather than AOB, were the numerically predominant ammonia-oxidizing organisms in the surface sediment. This could be attributed to the fact that AOA are more resistant to low levels of DO (Coolen et al., 2007; Molina et al., 2010; Bouskill

et al., 2012). The oxygen dynamics in aquaculture ponds differ from those of other aquatic systems, as pond environments are smaller, have limited water circulation, and are subjected to large deposits of feeding debris. The hypolimnion DO concentration was rarely >62.5µ mol L−<sup>1</sup> in an aquaculture pond, although the super-saturation of oxygen usually occurs during the daylight period (Chang and Ouyang, 1988). Moreover, for example, the DO concentration at the water-sediment interface was only 48.1µ mol L−<sup>1</sup> during another season (July 2015), and it reached zero when at depths >500µm.

Apart from the NH<sup>+</sup> <sup>4</sup> −N and TOC concentrations, pH has been suggested to be an important environmental factor that influences the distribution of AOB and AOA (He et al., 2007; Yao et al., 2011; Hu et al., 2014; Jiang et al., 2015). In this study, a significant negative correlation was found between pH and the abundance of the AOA amoA gene, indicating that the number of AOA decreased with decreasing pH-values. This finding is consistent with the physiological features of isolated AOA strains (Jong-Geol et al., 2012; Qin et al., 2014) and previous studies conducted in soil (Nicol et al., 2008; Hu et al., 2014). This effect

*(Continued)*

### FIGURE 6 | Continued

(*n* = 1000 replicates); Scale bar represents 0.1 (for AOB) or 0.05 (for AOA) nucleic acid substitutions per nucleotide position; Clones obtained from this experiment are highlighted in bold and are designated by sample names; the numbers in brackets represent the number of clones.

might be associated with the reported requirement for the use of NH3, not NH<sup>+</sup> 4 (Martens-Habbena et al., 2009). The sediment pH was significantly negatively correlated with the abundance of the AOB amoA gene, indicating that pH was an important factor that controlled the AOB abundance in aquaculture pond sediment. This finding is consistent with the result obtained from an alkaline sandy loam (Shen et al., 2008). A study of an AOB isolate from freshwater showed that it could grow in a wide pH range, although the highest growth rate occurred at pH 7–7.5 (Elizabeth et al., 2012). In this study, the pH in the surface (0- 6 cm depth) sediment ranged from 7.4 to 7.7. Although the pH only increased by 0.3 units with depth, the average abundance of AOB decreased 19-fold. The increase in pH may have decreased AOB growth and abundance.

To better understand the activity of the ammonia-oxidizing community in different sediment layers, potential ammonia oxidation rates were measured in the laboratory. Variations in the potential ammonia oxidation rates were not explained by the concentrations of amoA genes or mRNA in different sediment layers, as the rates did not exhibit any positive correlation with the concentrations of amoA or mRNA. Perhaps, the potential ammonia oxidation rates should be determined not only by the abundance and expression of AOB and AOA, but also by the phylotypes of AOB and AOA, as shown in **Figures 6A,B**, both of which consisted of different phylotype clusters, which may have different growth and nitrification rates (Bollmann et al., 2002, 2005; Tourna et al., 2011; Jong-Geol et al., 2012).

The results for the expression of amoA mRNA showed that the concentrations of AOB amoA mRNA was higher than that of AOA by 2.5- to 39.9-fold in the surface sediments (0–10 cm depth; **Figure 4**), although the copy numbers of the AOA amoA gene were higher than those of AOB by an order of magnitude in the surface sediments (0–6 cm depth; **Figures 3B,C**). The results indicated that ammonia oxidation was mainly carried out by AOB in surface sediments (0–10 cm depth), and that AOA might be the dominant ammonia-oxidizing microorganisms in deeper sediments (>10 cm depth), where only the AOA amoA gene was detected. A similar phenomenon was found in an agricultural soil, where AOB, rather than AOA, mainly conducted the ammonia oxidation, despite the fact that AOA amoA genes were more numerous than AOB amoA genes, and which was demonstrated by DNA-stable isotope probing (Jia and Ralf, 2009). In addition, in a temperate forest soil, it was also suggested that AOB are more involved than AOA in net nitrification in the top 5 cm of soil in July, and that AOA amoA genes are more numerous than AOB amoA genes in the topsoil (Onodera et al., 2010).

### Diversity of AOA and AOB

The diversity of AOB has been studied in various ecosystems with molecular tools, and it has been shown that AOB exhibit apparently high biodiversity in many aquatic ecosystems (Nicol et al., 2008; Wu et al., 2010; Jin et al., 2011; Sun et al., 2013). In this study, there were two AOB clusters: the Nitrosospira and Nitrosomonas clusters were found in sediments, and the latter was predominant in both sediment layers (**Figure 6A**). These results are consistent with previous studies of rhizoplanes of floating aquatic macrophytes, as well rice soils (Nicolaisen et al., 2004; Wang et al., 2009; Wei et al., 2011). Nitrosomonas were often detected in high-nitrogen environments, such as wastewater treatment plants (Geets et al., 2006; Stephanie et al., 2015), and some other studies have suggested that high concentrations of NH<sup>+</sup> <sup>4</sup> −N could enhance the development of Nitrosomonas spp. relative to Nitrosospira spp. (Bollmann et al., 2002, 2005).

Like AOB, the archaeal amoA gene was detected in different sediment layers, and all AOA fell within the Nitrososphaera and Nitrosopumilus clusters of the Thaumarchaeota phylum, with the latter being the dominant type (**Figure 6B**). A similar observation was found in the hyporheic zone of a eutrophic river (Wang et al., 2014b), where two distinct monophyletic clusters were also found, and the diversity of AOA decreased slightly with increasing sediment depth, but the Nitrososphaera cluster was the dominant cluster of archaeal ammonia oxidizers. In addition, the Nitrososphaera cluster represented the majority of AOA in many wastewater treatment plants, where Nitrosopumilus and Nitrososphaera clusters were jointly found (Limpiyakorn et al., 2013). The archaeal sequences were assigned into two branches with a clear difference between the surface and deeper samples, which may be attributed to the higher concentrations of TOC and NH<sup>+</sup> <sup>4</sup> −N in surface sediment. There was evidence that the Nitrososphaera cluster could bear higher amounts of TOC (Chen et al., 2008; Tourna et al., 2011; Liu et al., 2013) and NH<sup>+</sup> <sup>4</sup> −N (Tourna et al., 2011) than the Nitrosopumilus cluster. A similar phenomenon was also found in the sediments of the Dongjiang and Qiantang rivers (Liu et al., 2013; Sun et al., 2013). The AOA in deeper pond sediment were all grouped into the Nitrosopumilus cluster, which could be inhibited by organic carbon and prefer relatively lower carbon contents (Könneke et al., 2005).

In summary, our results showed that diversity of AOA and AOB decreased with increasing sediment depth and different dominant species were found at the different depths sampled. AOA were less active than AOB in surface sediments (0–10 cm depth) of the freshwater aquaculture pond, however, where AOA, as opposed to AOB, were the most abundant ammonia-oxidizing organisms. AOA might be the dominant ammonia-oxidizing microorganisms in deeper sediments, where only the AOA amoA gene was detected. This could be attributed to the fact that AOA are more resistant to low levels of DO. These results provide some useful information toward our understanding of freshwater pond sediment and their management, especially for the process of ammonia oxidation in fish pond sediment.

# ACKNOWLEDGMENTS

This study was financially supported by the National Natural Science Foundation (grant no. 31372570), a project in the National Science & Technology Pillar Program during the Twelfth Five-year Plan Period (no. 2012BAD25B01), and an

### REFERENCES


Open Fund of the Key Laboratory of Fishery Equipment and Engineering, Ministry of Agriculture (grant no. 2014006) of China. We also thank Dr. Liao Ming-jun (College of Resource and Environmental Engineering, Hubei University of Technology, Wuhan 430068, China) for reading the manuscript.


microorganisms in the Qiantang River, China. Appl. Environ. Microbiol. 13, 4065–4071. doi: 10.1128/AEM.00543-13


Dongjiang river, a drinking water supply for Hong Kong. Microbes Environ. 28, 457–465. doi: 10.1264/jsme2.ME13066


**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 Lu, Liu, Ma, Liu, Wu, Zeng, Shi and Gu. 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.

# Turbulence-driven shifts in holobionts and planktonic microbial assemblages in St. Peter and St. Paul Archipelago, Mid-Atlantic Ridge, Brazil

### *Edited by:*

Maurizio Labbate, University of Technology Sydney, Australia

### *Reviewed by:*

Matthew Schrenk, Michigan State University, USA Zhenfeng Liu, University of Southern California, USA Katherine Dafforn, University of New South Wales, Australia

### *\*Correspondence:*

Fabiano L. Thompson, CCS, BIOMAR, Laboratório de Microbiologia, Institute of Biology, Cidade Universitária, Av. Carlos Chagas F◦ . S/N, Bloco A (Anexo) A3, sl 102, Rio de Janeiro, RJ, CEP 21941-599, Brazil fabianothompson1@gmail.com

### *Specialty section:*

This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology

*Received:* 01 June 2015 *Accepted:* 11 September 2015 *Published:* 02 October 2015

### *Citation:*

Moreira APB, Meirelles PM, Santos EO, Amado-Filho GM, Francini-Filho RB, Thompson CC and Thompson FL (2015) Turbulence-driven shifts in holobionts and planktonic microbial assemblages in St. Peter and St. Paul Archipelago, Mid-Atlantic Ridge, Brazil. Front. Microbiol. 6:1038. doi: 10.3389/fmicb.2015.01038 Ana Paula B. Moreira<sup>1</sup> , Pedro M. Meirelles <sup>1</sup> , Eidy de O. Santos <sup>2</sup> , Gilberto M. Amado-Filho<sup>3</sup> , Ronaldo B. Francini-Filho<sup>4</sup> , Cristiane C. Thompson<sup>1</sup> and Fabiano L. Thompson<sup>1</sup> \*

<sup>1</sup> Laboratory of Microbiology, Institute of Biology, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil, <sup>2</sup> Fundação Centro Universitário Estadual da Zona Oeste (Uezo), Rio de Janeiro, Brazil, <sup>3</sup> Diretoria de Pesquisa Científica, Instituto de Pesquisas Jardim Botânico do Rio de Janeiro, Rio de Janeiro, Brazil, <sup>4</sup> Department of Environment and Engineering, Federal University of Paraíba, Rio Tinto, Brazil

The aim of this study was to investigate the planktonic and the holobiont Madracis decactis (Scleractinia) microbial diversity along a turbulence-driven upwelling event, in the world's most isolated tropical island, St Peter and St Paul Archipelago (SPSPA, Brazil). Twenty one metagenomes were obtained for seawater (N = 12), healthy and bleached holobionts (N = 9) before, during and after the episode of high seawater turbulence and upwelling. Microbial assemblages differed between low turbulence-low nutrient (LLR) and high-turbulence-high nutrient (HHR) regimes in seawater. During LLR there was a balance between autotrophy and heterotrophy in the bacterioplankton and the ratio cyanobacteria:heterotrophs ∼1 (C:H). Prochlorales, unclassified Alphaproteobacteria and Euryarchaeota were the dominant bacteria and archaea, respectively. Basic metabolisms and cyanobacterial phages characterized the LLR. During HHR C:H << 0.05 and Gammaproteobacteria approximated 50% of the most abundant organisms in seawater. Alteromonadales, Oceanospirillales, and Thaumarchaeota were the dominant bacteria and archaea. Prevailing metabolisms were related to membrane transport, virulence, disease, and defense. Phages targeting heterotrophs and virulence factor genes characterized HHR. Shifts were also observed in coral microbiomes, according to both annotation–indepent and -dependent methods. HHR bleached corals metagenomes were the most dissimilar and could be distinguished by their di- and tetranucleotides frequencies, Iron Acquision metabolism and virulence genes, such as V. cholerae-related virulence factors. The healthy coral holobiont was shown to be less sensitive to transient seawater-related perturbations than the diseased animals. A conceptual model for the turbulence-induced shifts is put forward.

Keywords: metagenomics, Atlantic Ocean, oceanic islands, Scleractinia, *Madracis decactis*

# Introduction

Marine microbial communities are recognized as engines of globally important processes,such as the marine carbon, nitrogen and sulfur cycles (Falkowski et al., 2008; Fuhrman, 2009). Only recently with the introduction of molecular techniques have satisfactory descriptions of natural microbial assemblages been generated (Fierer and Jackson, 2006; Rusch et al., 2007; Costello et al., 2009). Nevertheless, most marine ecosystems are understudied. As a result, there is limited information on the diversity of microbial assemblages in changing environments and on the environmental drivers of microbial diversity shifts (Karl, 2002).

Nutrient dynamics in the sea is inextricably linked to variations in physical processes. Either enhanced nutrient delivery from turbulent mixing or upwelling, or enhanced stratification can lead to shifts in microbial assemblages, with significant consequences for nutrient cycling (Cullen et al., 2002). Episodic mixing events must occur in order to balance supply and demand (Hayward, 1987, 1991). Because open-ocean microbial assemblages are metabolically active with a potential for relatively high specific growth, they are poised to respond quickly and effectively to environmental perturbations (Karl, 2002). A model of turbulence-nutrients regimes decoupled characteristics and adaptations of phytoplankton assemblages and how they relate to food web structure: (i) high turbulence—low nutrients: low biomass, slow turnover, adaptations for efficient use of light and nutrients (e.g., iron-limited, high latitude waters); (ii) LLR: smaller cells, high turnover, competition for nutrients, retention by recycling (microbial loop); (iii) low turbulence high nutrients: larger cells, higher biomass, slower turnover, selective pressure to sequester nutrients and minimize losses (e.g., noxious toxic blooms); and (iv) HHR: larger cells, higher biomass, transient, and self-limiting selection for rapid growth (e.g., diatoms). According to this model, the microbial loop is present in all regimes but it dominates the biomass in the low turbulence—high nutrients regime (Cullen et al., 2002).

Some physical mechanisms that vertically supply nutrients from below to the oligotrophic oceanic surface layers are: (i) internal waves and tides, (ii) cyclonic mesoscale eddies, (iii) winddriven Ekman pumping, and (iv) atmospheric storms (Karl, 1999). Internal waves are ubiquitous in deep-ocean environments (Garrett and Munk, 1972), and it appears that the high vertical shear of low-frequency internal waves contributes to occasional pulses of vertical mixing (Gregg et al., 1986; Sherman and Pinkel, 1991). Stochastic events, that may be short-lived, are of major ecological significance. They are undoubtedly undersampled by ship-based observation programs (Platt et al., 1989). Even a monthly sampling schedule, as adopted in the Hawaii Ocean Time-series (HOT) program, is considered too infrequent to register important but intermittent nutrient injections (Karl, 1999). Furthermore, most studies are restricted to the Pacific, Caribbean, and North Atlantic.

St. Peter and St. Paul Archipelago (SPSPA) are the smallest and most isolated tropical islands in the world. It comprises the most seaward Brazilian oceanic islands (∼1000 km from mainland) and the unique on the northern hemisphere, lying 100 km off the equator (00◦ 55′N; 29◦ 22′W). Data from the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) evinces the chlorophyll pattern characteristic of mesotrophic waters (0.1–0.3 mg.m−<sup>3</sup> Chla) (Supplementary Figures 1A,B), according to the classification of Shushkina et al. (1997). SPSPA is located in the biogeochemical Western Tropical Atlantic Province (WTRA) (Longhurst et al., 1995), under the influence of the Intertropical Convergence Zone. Its formed by minute summits of the Mid-Atlantic Ridge (MAR) within the St Paul Fault Zone (FZ), between the South American and the African Plates (Supplementary Figures 2A,B), where seismicity is frequent (Campos et al., 2010). The islets are devoid of shore and consist entirely of steep scarpments extending to 60–100 m depth, with the most limited area of shallow habitat among oceanic islands (∼200 m) (Robertson, 2001). Within 2 km diameter bathyal depths are reached and within 5 km depths fall within the abyssal range (−3.600 to 5.000; Supplementary Figure 2B). The South Equatorial Current (SEC) flows E-W superficially and the Atlantic Equatorial Undercurrent (EUC) flows W-E at 60–100 m depth (Edwards and Lubbock, 1983). The latter is one of the fastest, varying and least predictable among the Atlantic currents, which reaches 120 cm.seg−<sup>1</sup> above the thermocline (Philander, 1986). A permanent thermocline may prevent deep water masses to emerge (Macedo et al., 2009) nevertheless the eventual presence of deep water Pleurommama spp. and Heterorhabdus spp at the shallow layer during the diurnal period is contradictory to the former assumption (Macedo-Soares et al., 2009). There are no reports on local hydrodynamics but, seemingly, intermittent vertical flow of deep water masses should result from the violent interaction with the geomorphology—MAR—a perpendicular barrier to the currents. In principle, the friction between the SEC and EUC masses, which flow in opposing directions, should promote and intensify episodic extensions of the thermohaline to the photic zone.

Moreira et al. (2014)reported a significant enrichment process along an 8 days period during which an ever-growing turbulence with surge pulses was observed in SPSPA. The process occurred along the lunar phase from crescent to full moon. The work performed the first (and unique to date) characterization of the culturable heterotrophic bacterial community of SPSPA. Bacterial counts (colony forming units, CFU) correlated positively with nutrients in seawater, which in turn correlated positively with turbulence—energy and frequency of the surges. In the present work we analyzed the metagenomic composition and diversity of both the planktonic microbial assemblages and in the scleractinian coral M. decactis along the same period in the same locale. The aim was to characterize the microbial diversity during an upwelling-driven nutrient enrichment. We

**Abbreviations:** EUC, Atlantic Equatorial Undercurrent; BLAST, Basic Local Alignment Search Tool; chla, chlorophyll a; CFU, colony forming unit; DMSP, dimethylsulfoniopropionate; HOT, Hawaii Ocean Time-series; HHR, high turbulence-high nutrients regime; LLR, low turbulence-low nutrients regime; MAR, Mid-Atlantic Ridge; NCBI, National Center for Biotechnology Information; NADR, North Atlantic Drift Province; PHAST, PHAge Search Tool; SeaWiFS, Sea-viewing Wide Field-of-view Sensor; SATL, South Atlantic Gyral; SAO, South Atlantic Ocean; SPSPA, St Peter and St Paul Archipelago; FZ, St Paul Fault Zone; SEC, South Equatorial Current; SCM, sub-superficial chlorophyll maximum; VFs, virulence factors; VFDB, Virulence Factor Data Base; WTRA, Western Tropical Atlantic Province; WORMS, World Register of marine Species.

did not expect to find bleached corals in SPSPA. The coral holobionts were targeted in this survey to investigate whether there was a correlation between the metagenomic features and seawater parameters (vibrio counts, nutrients, bacterioplankton composition). Sampling was performed before, during and after a turbulence surge. It was a short-lived event, which is locally recurrent. We analyzed 21 metagenomic samples of seawater (n = 12), healthy and bleached corals (n = 9). The findings are summarized in a model of the physico-chemical-biological dynamics in SPSPA, where a cyclic recurrent pattern with extreme regimes of low turbulence-low nutrients (LLR) and high turbulence-high nutrients (HHR) is hypothesized to contribute to structure the marine ecosystem in that barren archipelago.

### Materials and Methods

### Field Sampling (Performed by Moreira et al., 2014)

In brief, sampling was performed by SCUBA diving at the Sub-caulerpa zone (mesophotic), according to the zonation of (Edwards and Lubbock, 1983; Moreira et al., 2014). The survey took place along the NW side of the archipelago from the inlet (ca 4500 m<sup>2</sup> ) to Belmonte islet's contiguous vertical wall. The satellite view and topograghy of the inlet are shown in Supplementary Figures 1C, 2B, respectively. The turbulence surge occurred along the lunar phase from crescent (14/Sep/2010) to full moon (22/Sep/2010). The peak of the surge overlapped with cloudiness, wind and rain. Rain and swash flushed guano, a possible additional source of nutrients (Gagnon et al., 2013), from the cays into the bay. Samples were obtained from the onset (LLR, t<sup>1</sup> = 14/Sep/2010), while enhancing (t<sup>2</sup> = 15; t<sup>3</sup> = 18/Sep/2010, HHR) and almost to recovery of LLR condition, or recovery for short (t<sup>4</sup> = 22/set/2010). During climax (20–21/Sep/2010) the strong vortex precluded diving (Supplementary Video 1). Henceforth samples will be referred as 14, 15, 18, and 22. In total, 12 colony fragments (10 × 10 cm) of M. decactis (healthy: n = 8; bleached: n = 4) were collected with hammer and chisel. On days 14 and 15 bleached corals were not found. Coral samples processed for metagenomics were: (i) healthy corals (Mad): 14 (n = 1), 15 (n = 2), 18 (n = 1), and 22 (n = 1); and (ii) bleached corals (MadBle): 18 (n = 2) and 22 (n = 2). Seawater was sampled from the water column immediately above the corals (<1 m) (4 samples: 14, 15, 18, and 22; 20 L/sample; 3 sterivex/sample). All samples were taken immediately to the Scientific Staion (SS) Laboratory, 20 m from the pier (Supplementary Figure 1D; view of the pier and the SS from the water at LLR). Seawater was filtered. Filters and coral samples were preserved in liquid nitrogen until DNA extraction (no longer than 3 months after).

### *Madracis Decactis* (Lyman, 1859) (Scleractinia: Pocilloporideae)

Is a colonial zooxanthellate scleractinian coral. It has a variable bathymetric distribution, from 3 to 30 and up to 100 m (Neves and Johnsson, 2009). It is widespread in Brazil (N to SE), in Caribe, Gulf of Mexico and locally found in the Southeastern Atlantic (West Africa) (World Register of marine Species— WORMS: www.marinespecies.org). Free-living colonies of M. decactis display an unique formation off southern Brazil, at Galé Island. Spheroid shape, a.k.a. circumrotatory colonies, form the first corallith site discovered in the subtropical South Atlantic Ocean (SAO), at 6–15 m depth over 3400 m<sup>2</sup> (coralreeefs-2012). In São Paulo (SE Brazil) it's a major contributor of reef structures, where bleaching has been seriously affecting its populations (Migotto, 1997), adding interest to the study of this coral species. In SPSPA, its one of the two scleractinian species locally found, mostly at the mesophotic zone. Healthy, bleached and with scars left by fish predation (Stegastes sanctipauli, Pomacentridae; Halichoeres radiatus, Labridae) in SPSPA are shown in Supplementary Figure 3.

### Seawater: Temperature, Nutrients, and Microbial Abundance

Environmental parameters were analyzed by standard oceanographic methods with at least three replicates for each parameter and determined by Moreira et al. (2014). Temperature was recorded in situ with a HOBO UA-002-64/date Logger and UEMIS dive computer from 5, 15, 33, 45, and 65 m depth (n = 5 for each depth), during September/2010 and June/2011 (published in Crespo et al., 2014). Environmental data from (Moreira et al., 2014) and temperature from (Crespo et al., 2014) are summarized in Supplementary Figure 4, for aid in data interpretation.

### Metagenomic DNA Extraction

Corals' DNA extraction was performed as in Trindade-Silva et al. (2012). Seawater was sequentially pre-filtered (100 and 20µm) by gravity and then filtered on the Sterivex (0.22 µm) by positive pressure using Niskin system (2 L/Sterivex). The microbes collected at Sterivex filters were preserved with SET buffer (20% sucrose, 50 mM EDTA and 0.5 mM Tris–HCl). Metagenomic DNA extraction was performed using lysozyme (1 mg/mL) for 1 h at 37◦C. Then, proteinase K (0.2 mg/mL) and sodium dodecyl sulfate (SDS) (1% v/v) were added and incubated (55◦C; 60 min) under agitation. The lysate was rinsed into SET buffer (1 mL). Organic extraction was performed with one volume of phenol:chloroform:isoamyl alcohol (25:24:1). DNA precipitation was performed with ethanol (2,5 volumes) and amnonium acetate (0.7 M) at −20◦C overnight. After centrifugation the pellet was washed twice with ethanol (70%) and air-dried. Elution was done in TE buffer (1X). Three libraries were prepared for each Sterivex and coral sample and pyrosequenced subsequently.

### Metagenomic Library Construction

Metagenomes were obtained by pyrosequencing technology using a 454 GS Junior instrument (Roche) (Margulies et al., 2005). Shotgun libraries were generated with 500 ng of whole metagenome samples, sheared into fragments by nebulization. End-repair and adaptor ligation were performed using GS FLX Titanium kit (Roche). Quality control and quantification were performed with Agilent 2100 Bioanalyzer (Agilent Technologies) and TBS 380 Fluorometer (Turner Biosystems), respectively. After the libraries construction, approximately 10<sup>6</sup> molecules/metagenome were denatured and amplified by emulsion PCR.

### Metagenomic Data Analysis

Raw sequences were submitted to quality control using PRINSEQ Standalone Lite (version 0.20.4; available at http://sourceforge. net/projects/prinseq/files/). We analyzed 21 metagenomic samples of seawater (n = 12) and corals (n = 9) from 4 days (t<sup>1</sup> = 14, t<sup>2</sup> = 15, t<sup>3</sup> = 18, t<sup>4</sup> = 22, of set/2010) along the enrichment process (t<sup>4</sup> − t<sup>1</sup> = 8 days). Annotation was performed by Meta-Genome Rapid Annotation using Subsystems Technology (MG-RAST) server (Meyer et al., 2008) version 3.0, using (SEED) Subsystems Technology and the GenBank database for functional and organismal classifications, respectively. For this purpose, all BLAST queries were conducted with a maximum cutoff E-value 0.00001, a minimum identity of 60%, and a minimum alignment length of 20 measured in aa for protein and bp for RNA databases.

### Metagenomes Comparison Trough Annotation-independent Analysis

Dinucleotide odds ratio and Karlin distances (δ) between the metagenomes, based on the dinucleotide relative abundances differences (according to Karlin et al., 1997), were calculated using Perl scripts as in Willner et al. (2009). The values in the Karlin matrix were multiplied by a 1000 for easier comparison. Tetranucleotide frequencies were calculated using a Perl script as in Albertsen et al. (2013). The divergence between the observed and expected tetranucleotide frequencies was transferred into z-scores and pairwise comparison of the metagenomic sequences was performed by computing the Pearson's correlation coefficients of the z-scores, both through Python scripts, according to (Teeling et al., 2004). Seawater metagenomes (Sw-14, -15, -18, and -22) were compared among each other, as well as coral metagenomes (Mad14, Mad15, Mad18, MadBle18, Mad22, MadBle22).

### Phage Detection

The 21 metagenomic libraries were searched for phages using the PHAge Search Tool (PHAST), available at http://phast. wishartlab.com (Zhou et al., 2011). Briefly, pyfasta 0.5.2 (available at http://pypi.python.org/pypi/pyfasta/) was used to split metagenomes into smaller subsets without splitting individual fasta entries, after which PHAST was used to phage search. The tool provides an ensemble of ORF prediction and translation (via GLIMMER; Salzberg et al., 1998), protein identification (via BLASTP; Altschul et al., 1997), phage sequence identification (via BLAST matching to a specific database), tRNA identification (using tRNAscan-SE; Lowe and Eddy, 1997), attachment site recognition (with ARAGORN; Laslett and Canback, 2004) and gene clustering density measurements using density-based spatial clustering of applications with noise (DBSCAN; Ester et al., 1996), and sequence annotation text mining. PHAST's database encloses protein sequences from two sources: the NCBI phage database and the prophage database (Srividhya et al., 2006). Specific keywords (e.g., "protease," "integrase," and "tail fiber") are used for screening. Matched phage or phage-like sequences with E < 0.0001 are saved as hits and their positions tracked for subsequent evaluation for local phage density by DBSCAN. Phage schemes shown in Supplementary Figures 11–18 were generated with PHAST.

### Sequence Comparison to the Virulence Factor Database

The 21 metagenomic libraries were compared to the virulence factor database (VFDB) (Chen et al., 2005) (http://www.mgc.ac. cn/VFs/) using BLASTX (E < 0.0001).

### Statistics

Statistical analysis was conducted using R Version 3.1.3 (Team, 2012) with a suite of packages. The comparison of the correlation coefficients of the z-scores obtained for the tetranucleotide frequencies was visualized through heatmaps using rpy2 and gplots (Gautier, 2008). An exploratory analysis aiming to correlate samples with nutrients' concentrations and metabolisms, according to the level 1 SEED classification, was performed by means of a principal component analysis (PCA) using the rda function of Vegan package (Oksanen et al., 2013). Abundance plots were drawn using the ggplot2 and reshape packages (Wickham, 2007; Wickham and Chang, 2009). The cluster analysis was performed with the APE package (Paradis et al., 2004) using Pearson correlation and ward distance.

### Sequence Data

The metagenomic data that we generated are available in the MG-RAST v3 server (http://metagenomics.anl.gov/ metagenomics.cgi) under the unique identifiers: 4461594.3, 4463932.3, 4463930.3, 4463939.3, 4463927.3, 4461593.3, 4463933.3, 4463931.3, 4463928.3, 4463926.3, 4463925.3, 4468639.3, 4486661.3, 4486665.3, 4486669.3, 4486662.3, 4486667.3, 4486668.3, 4486664.3, 4486663.3, and 4486666.3; and in the Brazilian Marine Biodiversity database (BaMBa) (pmeirelles.18.1).

### Sampling Permit

Ministério do Meio Ambiente (MMA), Instituto Chico Mendes de Conservação da Biodiversidade (ICMBio) Number 10112-2.

### Results

In this study 21 metagenomic libraries were obtained from seawater (n = 12) and corals (n = 9) with a total of 446,129 sequences (2.13 × 10<sup>8</sup> bp) (**Table 1**). Through annotation-independent analysis, samples Sw14, -15, and -22 were distinguished from -18 (**Table 1**). Average Karlin distances for pairs of seawater samples Sw14, -15, -18 and -22 defined three categories based on δ ranges. (i) δ < 11 grouped pairs 14- 14 and 22-22, indicating high degree of genetic similarity; (ii) 11 < δ < 30 grouped pairs 15–15, 18–18, 14–15, 14–22, and 15–22, indicating intermediate genetic similarity, and (iii) δ > 30 grouped pairs 14–18, 15–18, and 18–22, indicating low genetic similarity. The most dissimilar samples (δ = 50.6) were Sw14 (LLR) and Sw18 (HHR) (Supplementary Figure 5). Similarly,


### TABLE 1 | General features of the metagenomes.

\*Day of set/2010; Sw, seawater; Mad M. decactis; MadBle bleached M. decactis, Avg, Average; SE, standard error.

average Karlin distances for coral samples Mad14, -15, -18, and -22 defined three categories. (i) δ < 11 grouped pairs 14–22, 14–15, 15–22, and 22–22; (ii) 11 < δ < 20 grouped pairs 15–15; and (iii) δ > 20 grouped pairs 14–18, 15–18, 18–22, and 18–18. Samples Mad18 were the most dissimilar in comparison with -14, -15, -22 and also among each other (Supplementary Figure 6). The same pattern was revealed by the analysis that resulted from the tetranucleotide frequencies estimated for seawater and coral metgenomes. Samples Sw18 (HHR) were the most dissimilar amongst seawater metagenomes (Supplementary Figure 7A) and metagenomes from HHR bleached corals (MadBle18) were the most dissimilar amongst the holobionts' metagenomes (Supplementary Figure 7B).

To analyze the overall relationships among the most abundant taxa we performed a clustering analysis (**Figure 1**). The hierarchical clustering of the 21 metagenomes corroborated the binning based in sequence composition (di- and tetranucleotides frequencies). Two major branches split seawater and coral metagenomes. Two seawater branches were defined, with samples 14 (and Sw15-1) representing the LLR. Samples 15 (Sw15-2 and 4), 18 and 22 reflected the turbulence surge (**Figure 1**). Coral metagenomes were split into healthy and bleached (with only two exceptions: MadBle18-1 grouped into the healthy corals branch and Mad22 into the bleached corals branch). Healthy corals also showed a trend to group according to the enrichment gradient. The healthy corals branch split Mad14 and Mad15-1 from the remainders Mad15-2, Mad18, and Mad22.

### Taxonomic Assignment of Seawater Metagenomes

Bacteria was the most abundant domain in all samples with overall average relative abundance of 97.0 ± 1.8% (Mean ± SD), followed by 1.11 ± 0.69% Eukarya (Supplementary Figure 8). Archaeal sequences accounted for ∼0.4% of the overall sequences. The main groups were Euryarchaota (∼58%) and Thaumarchaeota (21%). Viruses represented ∼0.4% of sequences. The most abundant order was Caudovirales (∼52%). The overall relative abundance of the most prevalent bacterial phyla were ∼61% Proteobacteria, 32% Cyanobacteria followed by 2.4% Bacteroidetes (100% Flavobacteriales), 1.6% Firmicutes, and 1.3% Actinobacteria (**Figure 1**).

### Shifts in Planktonic Assemblages

The taxonomic classification of seawater samples revealed a clear difference between LLR and the remainder groups (Sw15, -18, -22). In all samples the dominant groups were Cyanobacteria (C) and Proteobacteria (P). At LLR (Sw14), these groups were equally abundant in seawater (C:P ∼1). Along the turbulence gradient, Proteobacteria members increased and Cyanobacteria

abundant orders is presented at right.

decreased to a vanishingly small proportion (Sw18, HHR; C:P << 0.05), with a subsequent recovery (Sw22; C:P > 0.5). The highest proportion of Proteobacteria occurred at HHR (92%) (**Figure 2A**). At LLR Alphaproteobacteria presented the highest relative abundance (Alpha 28.3% and Gamma 25.1%; ANOVA, P < 0.05), whereas Gammaproteobacteria was dominant in all the following samples (Sw15, -18, and -22: Gamma 29.4% and Alpha 24.8%; P < 0.05) (Supplementary Figure 9). Proteobacterial groups whose relative abundances enhanced during enrichment were Alteromonas, Vibrio and Pseudomonas (Gamma); Ruegeria, Roseobacter, and Candidatus Pelagibacter (Alpha) (**Figure 2B**). Cyanobacteria and unclassified Alphaproteobacteria (mostly SAR11) decreased correspondingly (**Figure 1**). Oceanospirillales was absent in the top ten rank at LLR, but appeared in Sw15-4 (2.5%), in HHR (5.6–7.3%), and Sw22 (2.5–2.7%). Archaeal groups also shifted dominance. The most abundant phyla in Sw14-15 was Euryarchaeeota, whereas in Sw18-22 it was Thaumarchaeota (**Figure 2C**).

### Functional Assignment of Seawater Metagenomes

The overall most abundant subsystems were Protein Metabolism (9.8 ± 1.0%), Amino Acids and Derivatives (9.4 ± 0.7%), Carbohydrates (9.4 ± 0.7%), Cofactors, Vitamins, Prosthetic

Groups, Pigments (7.2 ± 0.7%), and RNA Metabolism (5.2 ± 0.7%) (Supplementary Figure 10). The most abundant gene per metagenome was the TonB-dependent receptor, of the subsystem Iron Acquisition and Metabolism, in samples Sw18 (HHR) (Supplementary Table 1). A phage protein encoding gene was the most abundant gene in five samples: Sw22-1,2 (recovery) and Sw14-1,2,3 (LLR). Phage detection through specific databases (using PHAST) revealed that the most probable hosts were cyanobacteria for Sw14 and -22 (LLR and recovery, respectively), and Proteobacteria for the remainders (Sw15 and HHR), with only two exceptions (**Table 2**; Supplementary Figures 11–18). The range of unknown proteins was 17.9–42.4% (Sw18-2 and Sw14-1, respectively).

### Shifts in Metabolisms in Seawater Metagenomes

To investigate the metabolic profile of planktonic dwelling microbes and to correlate differences with the enrichment gradient, we performed a PCA analysis (**Figures 3A,B**). Clustering of samples Sw14 (LLR) was not explained by nutrients concentrations, contrarily to the remainder samples (**Figure 3A**). Basic cell functions such as Carbohydrates and Respiration explained the clustering of samples Sw14 (LLR), whether enrichment samples Sw15 and Sw18 (HHR) clustered in response to Virulence, Disease, and Defence; Membrane Transport and Nitrogen Metabolism subsystems (**Figure 3B**). Samples Sw15 were grouped by the concentration of the nutrients organic phosphorous, ammonia, orthophosphate, and nitrite. The former three showed the highest concentration at Sw15 (**Figure 3A**, Supplementary Figure 4). HHR samples (Sw18) were coupled by the heterotrophic-characteristic metabolisms, Membrane Transport and Virulence, Disease and Defense. Accordingly, their most abundant gene was the TonB-dependent receptor (Virulence, Disease and Defense). Recovery samples (Sw22) were coupled by the concentration of nitrate (the highest, **Figure 3A**, Supplementary Figure 4) and Nitrogen Metabolism (Sw22-1; **Figure 3B**), correspondingly.



Sw, seawater; Mad M. decactis; MadBle bleached M. Decactis; #, number.

FIGURE 3 | Principal component analysis (PCA) diagrams. (A) Diagram generated for the bacterioplankton samples and nutrients in seawater. (B) Diagram generated for the bacterioplankton samples and metabolisms in seawater.

### Taxonomic Assignment in *Madracis Decactis* Metagenomes

Bacteria was the most abundant identified Domain in all samples with overall relative abundance of 57.15%, followed by Eukarya (39.63%). Archaeal sequences represented 1.92% and viruses 0.1%. Within the domain Eukarya, 33.59% of the sequences were Cnidaria (Supplementary Figure 19A). Main overall archaeal phyla were Euryarchaeota (45.61%), Thaumarchaeota (42.11%), and Crenarchaeota (10.96%). (**Figure 2C**). There were marked differences in comparison with seawater metagenomes (**Figure 1**). The most abundant bacterial phyla were Proteobacteria (∼48% of the counts), Firmicutes (17%), Actinobacteria (10%), Bacteroidetes (6%), and Cyanobacteria (4%). The most abundant proteobacterial orders were Rhizobiales (16.29%), Rhodobacterales (12.44%); Burkholderiales (10.77%), Pseudomonadales (4.85%), Pasteurellales (4.82%), Alteromonadales (4.68%), Desulfuromonadales (4.38%), Enterobacteriales (4.12%), Myxococcales (4.01%), and Chromatiales (3.46%). Cyanobacteria was mostly represented by Chroococcales (50.23%), Oscillatoriales (24.35%), Nostocales (20.03%), and Prochlorales (5.24%). Interestingly, Rhodobacterales, Rhizobiales, Actinomycetales, Burkholderiales, and Clostridiales were abundant in healthy and diseased M. decactis.

### Shifts in Microbial Assemblages in *Madracis Decactis* Metagenomes

At HHR bleached corals (MadBle18) showed more sequences affiliated to Proteobacteria than the healthy corals (Mad18) (Supplementary Figure 19B). The presence of Pasteurellales among the ten most abundant orders in five out of nine metagenomes (in Mad15, -18, and, -22 with 11–14% of counts) appeared to be a diagnostic feature (**Figure 1**). Pasteurellales was found in low relative abundance (<1%) in the remainder corals and in seawater metagenomes. Thaumarchaeota affiliates in all M. decatis samples (Mad14, -15, -18, and -22) approximated Sw18-22 abundance levels (**Figure 2C**). Cnidaria and Nematoda metagenomic sequences were more abundant in healthy and bleached corals, respectively (Supplementary Figure 8).

### Functional Assignment in *Madracis Decactis* Metagenomes

The overall most abundant subsystems were Carbohydrates (16.6 ± 2.8%), Amino Acids and Derivatives (14.3 ± 1.2%), Protein Metabolism (9.66 ± 1.6%), Cofactors, Vitamins, Prosthetic groups, Pigments (8.9 ± 1.6%), and DNA Metabolism (6.6 ± 2.4%) (Supplementary Figure 20).

### Shifts in Metabolisms is *Madracis Decactis* Metagenomes

To investigate major differences in metabolisms the corals' metagenomes were pooled (according to **Figure 1** and Supplementary Figure 7B) in samples (i) healthy 14–15 (Mad14 and Mad15-1,2), (ii) healthy 18–22 (Mad18 and Mad22), and (iii) diseased 18–22 (MadBle18-1,2 and MadBle22-1,2). Seawater samples were included as reference (Sw14–15, Sw18– 22). Prevalence of motility and chemotaxis in seawater compared to corals was the main difference observed between type samples. DNA metabolism prevailed in corals Mad18–22 (healthy and bleached) relative to the remainder samples (Supplementary Figure 20). DNA metabolism was the fourth most represented in corals. The overall most abundant gene (EC 2.1.1.72) in corals was affiliated to this subsystem, which was the most abundant gene in three metagenomes of the Mad18–22 group, and in none of the Mad14–15 group (Supplementary Table 1). The Iron Acquisition subsystem was investigated further by pooling coral samples according to health status in (i) healthy and (ii) diseased. The relative abundance range was lower in healthy corals (5.56–9.09%) than in diseased corals (5.15–16.36%) (P < 0.05). Iron Acquisition in Vibrio was the most abundant function in overall samples, with a relative abundance range of 18.2–83.3% (MadBle22-1 and Mad22, respectively). Healthy corals presented higher relative abundances (60–83%) than diseased corals (18.2–34.4%) for this function. Six functions were represented only in diseased corals: Campylobacter Iron Metabolism (6.3–40.0%); Heme, Hemin Uptake and Utilization Systems in GramPositives (0–20.0%); Siderophore Pyoverdine (0–18.2%), Transport of Iron (0–14.9%); Iron Acquisition in Streptococcus (0–10.3%) and Iron(III) Dicitrate Transport System (0–3.1%). Heme, Hemin Uptake and Utilization Systems in Gram-Negatives was overrepresented in diseased (16.2%) compared to healthy corals (7.3%) (P < 0.05).

### Shifts in the Profiles of Seawater- and *Madracis Decactis*- Dwelling Communities

The shift from autotroph:heterotroph-balance to offset was further investigated using virulence factors (VFs) as indicators of heterotrophy and risk or threat for corals. In total there were 21,230 significant similarities against the VFDB (**Table 3**). When normalized to library size, virulence genes were overrepresented in samples Sw 15, -18 (16.0–24.3) and, to a lesser extent in Sw22 (16.6–16.8) (recovery), compared to LLR (Sw14; 12.7–14.2); and in bleached (MadBle; 0.8–8.5) when compared to healthy corals (Mad; 0.6–3.1). We further investigated the iron uptake system, which is a nonspecific virulence system related to competition skills, and thus suitable to reflect the overall heterotrophic community. Iron related virulence genes comprised ∼9% (n = 1817) of the total hits to the VFDB. Sw18 samples (HHR) presented the lower percentage of iron related genes relative to the total virulence hits per metagenome (P < 0.05), suggesting that other than iron uptake genes were most representative of the surplus heterotrophs. The VFDB lacks genes related to iron from vibrios (which was overly represented in healthy corals), but encompasses genes related to iron from Haemophilus (Pasteurellales). Among all virulence genes related to iron, ∼33% (n = 606) fell into this category. V. cholerae related virulence genes, which confer infective skills, comprised ∼6% (n = 1298) of the total virulence hits. The percentage range of these genes relative to total VFs per metagenome in seawater samples Sw14, -15, and -22 was 4.4–6.5% and in Sw18 (HHR) was 6.6–7.7%. Similarly, in healthy corals this range was 0.1–6.9%, and higher in diseased corals: 4.7–8.1% (**Table 3**). Heterotrophic populations that overgrew in response to turbulence-nutrient pulses were better represented by pathogenic (e.g., V. cholerae-related) than by non-specific (e.g., iron acquisition) VFs.

# Discussion

Microbial assemblages during LLR (Sw14) were comparable to those previously described for the surface layers in the western SAO (South Atlantic Gyral - SATL) (Alves Junior et al., 2015) and within the WTRA (Heywood et al., 2006; Schattenhofer et al., 2009), where the dominant groups (= 50%) detected were also Prochlorales and unclassified Alphaproteobacteria or SAR11 and related. Following LLR the microbial assemblages observed increasingly differed from previous studies focusing the surface layers of those most neighboring locations (Heywood et al., 2006; Schattenhofer et al., 2009; Swan et al., 2011). Alteromonadales appeared as the second most abundant group after Prochlorales, prevailing over unclassified Alphaproteobacteria, and Vibrionales emerged as a new group with >5% relative abundance (Sw15). Comparable Alteromonadales relative abundances, combined with lower abundances of unclassified Alphaproteobacteria (SAR11) were previously reported for the sub-superficial chlorophyll maximum (SCM) layer at higher depths (48–82 m) in the SAO, where, instead of Vibrionales, Pseudomonadales, and Mamiellales emerged as differing groups compared to the surface layers (Alves Junior et al., 2015). At HHR (Sw18), other Gammaproteobacteria appeared with > 5% relative abundance, i.e., Pseudomonadales and Oceanospirillales, whereas Thaumarchaota reached Euryarchaeota relative abundance levels. Pseudomonadales, Oceanospirillales, and Thaumarchaeota characterized deep waters (236–1200 m) in the SAO, and in which water masses Prochlorales was not amongst the 10 most abundant orders (AlvesJr-14). (Schattenhofer et al., 2009) reported a Gammaproteobacteria bloom in the North Atlantic Drift Province (NADR), with a maximum relative abundance of >50% of all picoplankton in surface waters, compared to the average values of 2–5% for all the other Atlantic provinces. Only a minor fraction was identified (Alteromonas/Colwellia and Pseudoalteromonas: 2–5%; Vibrio: 1%, and Oceanospirillum: 4%). The Gammaproteobacteria bloom was attributed to the end of the spring phytoplankton bloom, indicated by declining chlorophyll values. Massive growth of Bacteroidetes was concomitant and deep water Archaea presence at surface was not observed. Gammaproteobacteria have the potential to


### TABLE 3 | Abundances of virulence factor genes from microbes in the bacterioplankton and *M. decactis* in SPSPA.

Sw, seawater; Mad M. decactis; MadBle bleached M. Decactis; #, number.

respond to sudden nutrient pulses released from phytoplankton (Cottrell and Kirchman, 2000). Members of Alteromonas, Pseudoalteromonas, and Vibrio are well known to rapidly respond to excess nutrient supply (Bano and Hollibaugh, 2002; Beardsley et al., 2003; Allers et al., 2007, 2008). Thaumarchaeota are typically more abundant at depths of ≥100 m, as oposed to Euryarchaeota, which is known for decreasing abundance below 100 m (Delong, 1992; Zhang et al., 2009; Santoro et al., 2010; Tseng et al., 2015). In SPSPA, concomitant with Thaumarchaeota increase, ammonia levels decreased, possibly due to its ammonia-oxidizing ability (Francis et al., 2005); and Flavobacteriales (Bacteroidetes) relative abundance decreased. Next (Sw22), the five most abundant groups at LLR recovered relative abundances almost to LLR (Sw14) levels (Prochlorococcus, unclassified Alphaproteobacteria— SAR11, Rhodobacterales, Rhizobiales, and Chroococcales), but Alteromonadales, Pseudomonadales, Vibrionales, and Oceanospirillales remained amongst the 10 most abundant groups, as well as Thaumarchaeota remained abundant, which is a distinctive assemblage for those geographical coordinates (Schattenhofer et al., 2009; Swan et al., 2011; Alves Junior et al., 2015). Shifts in planktonic assemblages at the mesophotic zone in SPSPA were possibly driven by the turbulence surge, meaning that microbes from progressively deeper layers could hitchhike with the vertical flux along the surge. The upwelling is also supported by the enrichment and by the wide variation of water temperatures registered for the sampling depth (Supplementary Figure 4).

### A Model of Physical-Chemical-Biological Dynamics in SPSPA

Although sequence similarities to genes do not represent levels of gene expression, metagenomes have been shown to be strong predictors of the biogeochemical conditions driving the microbial community (Dinsdale et al., 2008). According to the lines of evidence garnered the microbiome of the mesophotic waters in SPSPA undergoes cyclic transient shifts in relation to turbulence-nutrients regimes. A microbial succession resulting from the interplay between physical and chemical factors is a plausible scenario. Two extreme turbulencenutrient regimes can be clearly distinguished and alternate with intermediate conditions determining microbial assemblages: (i) When turbulence is low (LLR) at least 50% of the microbiome is composed of Prochlorococcus, followed by unclassified Alphaproteobacteria (SAR11 and related), which are small sized cells, highly adapted to oligotrophic conditions and starvation. Rhodobacterlaes, Rhizobiales, and Chroococcales are typical. In this environment phage genes are the most abundant in seawater, mostly from Prochlorococcus and Synechococcus, following the hosts' abundances. The viral shunt is probably less active toward relatively scarce cells. The proportion of unknown genes is the highest; (ii) Episodic surges promote vertical mixing from the immediate lower water mass to the mixed layer. Waves also wash guano from the cays flushing phosphates and ammonium into the inlet. Heterotrophs (Alteromonadales, Vibrionales) respond quickly and surpass autotrophs, motility, and chemotaxis related genes stand out; (iii) Ongoing eddies and intensified high-energy waves promote entrainment of deep water organisms such as Thaumarchaeota and nutrients (nitrite, nitrate) (HHR). Eventual cloudiness, winds and rain cope with turbulence, irradiance is intermittent and turbidity is enhanced. Heterotrophy predominates with dominance of Alteromonadales, Pseudomonadales, and Oceanospirillales. Gammaproteobacterial groups approximate 50% of the microbial assemblage, resembling the end of the spring phytoplankton bloom in higher latitudes (e.g., NADR). The gene pool in surface waters reflects the shift with membrane transport and virulencerelated genes (e.g., TonB-dependent receptor, V. cholerae virulence genes) surpassing cyanobacterial phages and basic metabolisms genes. Phages targeting heterotrophs are active. The proportion of unknown genes is the lowest; (iv) Turbulence alleviates (e.g., after moon changes toward new). Larger cell sized heterotrophs begin to decline as viral lysis and predation by grazers overrides growth, which is constrained by the paucity of limiting nutrients (e.g., phosphorus). The microbial loop is most prominent at this stage. Autotrophs respond to irradiance and retake growth (if rain, wind, and cloudiness mitigate this response is accelerated). Nitrogen metabolism is intensive. A reversal to autotrophy:heterotrophy equilibrium is triggered (**Figure 4**). The short-lived but recurrent turbulence-nutrient pulses might be responsible for structuring the marine ecosystem in a bottom-up manner in SPSPA. These pulses might be indispensable to warrant the energy and carbon flow to the higher trophic levels concurring to the observed pelagic fishes biomass around the barren islets (Luiz and Edwards, 2011). On a stable LLR the growth of phytoplankton is largely supported by regenerated nutrients, so only a small proportion of primary production is available to higher trophic levels or for export to the deep sea (Cullen et al., 2002; Karl, 2014). Turbulence is physically forcing the co-ocurrence of nutrients and light in SPSPA, on the other hand, the fact that nutrient resupply is shortlived might concur to retain the local mesotrophic condition. Bacterioplankton shifts were shown to be transient, following the cyclic nutrient-turbulence pulses and other physical parameters (rain, cloudiness, winds, turbidity).

### The Holobiont *Madracis Decactis*

Some bacterial taxa prevalent in bleached corals (Rhodobacterales, Rhizobiales, and Clostridiales) have been previously associated with opportunistic diseases (Frias-Lopez et al., 2002; Rosenberg et al., 2007; Sekar et al., 2008; Sunagawa

contribute to enrichment and the shift to heterotrophy. Proteobacterial phages (red poligons) reflect the abundance of the hosts as well as Iron acquisition (red circle) and virulence factor (red crosses) genes. Vibrios abundance (blue elipses) reflect seawater parameters. Episodic surges frequently correlate with full moon. When turbulence pulses mitigate and weather assuages a recovery takes place in seawater, both in terms of nutrients concentrations and microbial assemblages (e.g., after moon changes toward new).

et al., 2009; Mouchka et al., 2010). On the other hand, a study of the corals microbiome, aiming at distinguishing the core, the symbiotic and the whole community microbiome, suggested that Rhodobacterales pertains to the latter. Conversely, Rhizobiales members were suggested to belong to the symbiotic coral microbiome. Actinomycetales and Burkholderiales (both also prevalent in all coral samples) were characterized as part of the coral core microbiome (D Ainsworth et al., 2015). Pasteurellales was one of the most abundant bacteria in the coral metagenomes, contrasting to its dwindling relative abundance in seawater. Pasteurellales members can cause disease in a wide range of domestic and wild animals (Wilson and Ho, 2013). They are commonly found in fish tissues (Birkbeck et al., 2002). Reef fishes (Chaetodontidae) have been characterized as major vectors of coral diseases (Raymundo et al., 2009) and damselfish (Stegastes spp.) was shown to increase the prevalence of the coral Black Band Disease (BBD) (Casey et al., 2014). It is plausible that Pasteurellales are frequently transmitted to corals through fishes, possibly by fish bites, since M. decactis is frequently predated by S. sanctipauli, H. radiatus and other fishes in SPSPA (Supplementary Figure 3). This hypothesis explains the uneven distribution of Pasteurellales between healthy and diseased corals, as well as the disconnection to seawater parameters. Bleached corals were distinguished by enhanced Iron Acquisition metabolism. Six functions within this subsystem were represented only in bleached corals. HHR bleached corals samples (MadBle18) were the most dissimilar in terms of sequence composition (di- and tetranucletides frequencies), higher counts of Proteobacteria (including Vibrionales), and higher relative abundance of hits to the VFDB, including V. cholerae-related VFs. This dissimilarity, including the HHR healthy coral (Mad18), indicates that the healthy coral holobiont might be less sensitive to transient seawater-related perturbations than the diseased animals. The distinguishing characteristics of HHR bleached corals agree with the bacterioplankton and seawater features during HHR, reported both in the present and former study (Supplementary Figure 4; Moreira et al., 2014): sequence composition, higher relative abundance of motility and chemotaxis, and of membrane transport and virulence genes (e.g., Ton-B dependent receptor of the Iron Acquisition metabolism, V. cholerae-related VFs), higher vibrio counts and nutrients in seawater. Taken together, the datasets suggest coupling between the benthic and pelagic compartments, as previously reported (Chimetto Tonon et al., 2015).

### Caveats

Owing to the remote nature of this site, we do not have complete data sets. Resampling will be needed to strengthen the

# References


link between turbulence-upwelling and the shifts in microbial assemblages.

# Conclusions

This work analyzed shifts in microbial composition related to physical forcings (turbulence-upwelling and storms) in SPSPA. LLR is characterized by the equilibrium between autotrophyheterotrophy and microbial assemblages that resemble those of surface tropical waters previously characterized in the SAO. At HHR microbial communities shift to heterotrophic and deep-sea characteristic organisms (Thaumarchaota). HHR diseased corals are distinguished by sequence composition and enhanced VFs hits, suggesting some level of coupling between planktonic and coral microbial communities.

### Acknowledgments

This work was supported by CNPq, CAPES and FAPERJ. Comissão Interministerial para os Recursos do Mar partially supported the expedition. The authors are grateful to the Brazilian Navy for maintenance of the Scientific Station and training; to Ericka Coni for joining the expedition and participating in sampling; to the Transmar III crew, Douglas Abrantes and Camilo Ferreira for general logistical support, and to Prof. Rob Edwards (San Diego State University) for the valuable ideas and helpful discussion.

# Supplementary Material

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmicb. 2015.01038


habitats across space and time. Science 326, 1694–1697. doi: 10.1126/science. 1177486


**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 Moreira, Meirelles, Santos, Amado-Filho, Francini-Filho, Thompson and Thompson. 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.

# Rhodopsin gene expression regulated by the light dark cycle, light spectrum and light intensity in the dinoflagellate *Prorocentrum*

*Xinguo Shi1,2, Ling Li1, Chentao Guo1, Xin Lin1, Meizhen Li1 and Senjie Lin1,3\**

*<sup>1</sup> State Key Laboratory of Marine Environmental Science, College of Ocean and Earth Sciences, Xiamen University, Xiamen, China, <sup>2</sup> College of the Environment and Ecology, Xiamen University, Xiamen, China, <sup>3</sup> Department of Marine Sciences, University of Connecticut, Groton, CT, USA*

### *Edited by:*

*Mark Vincent Brown, University of New South Wales, Australia*

### *Reviewed by:*

*Jose M. Gonzalez, University of La Laguna, Spain Gurjeet Singh Kohli, University of Technology, Sydney, Australia*

### *\*Correspondence:*

*Senjie Lin, Department of Marine Sciences, University of Connecticut, Groton, CT 06340, USA senjie.lin@uconn.edu*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 03 March 2015 Accepted: 20 May 2015 Published: 02 June 2015*

### *Citation:*

*Shi X, Li L, Guo C, Lin X, Li M and Lin S (2015) Rhodopsin gene expression regulated by the light dark cycle, light spectrum and light intensity in the dinoflagellate Prorocentrum. Front. Microbiol. 6:555. doi: 10.3389/fmicb.2015.00555* The proton pump rhodopsin is widely found in marine bacteria and archaea, where it functions to capture light energy and convert it to ATP. While found in several lineages of dinoflagellates, this gene has not been studied in Prorocentrales species and whether it functionally tunes to light spectra and intensities as in bacteria remains unclear. Here we identified and characterized this gene in the bloom-forming *Prorocentrum donghaiense*. It is a 7-helix transmembrane polypeptide containing conserved domains and critical amino acid residues of PPR. This gene is phylogenetically affiliated to the xanthorhodopsin clade, but seems to have a distinct evolutionary origin. Quantitative reverse transcription PCR showed that in regular cultures, the transcript abundance of the gene exhibited a clear diel pattern, high abundance in the light period and low in the dark. The same diel pattern was observed for protein abundance with a Western blot using specific antiserum. The rhythm was dampened when the cultures were shifted to continuous dark or light condition, suggesting that this gene is not under circadian clock control. Rhodopsin transcript and protein abundances varied with light intensity, both being highest at a moderate illumination level. Furthermore, the expression of this gene responded to different light spectra, with slightly higher transcript abundance under green than blue light, and lowest abundance under red light. Transformed *Escherichia coli* over-expressing this rhodopsin gene also exhibited an absorption maximum in the blue–green region with slightly higher absorption in the green. These rhodopsinpromoting light conditions are similar to the relatively turbid marine habitat where the species forms blooms, suggesting that this gene may function to compensate for the light-limited photosynthesis in the dim environment.

Keywords: rhodopsin, gene expression, light dark cycle, light intensity, light spectrum, *Prorocentrum*

# Introduction

Microbial rhodopsin is a type of photoreceptor widespread in the marine ecosystem, widely reported in marine bacteria and archaea (de la Torre et al., 2003; Man et al., 2003; Venter et al., 2004; Giovannoni et al., 2005). It also occurs in some eukaryotic algae (Gualtieri et al., 1992; Nagel et al., 2002; Sineshchekov et al., 2005; Frassanito et al., 2010; Lin et al., 2010; Slamovits et al., 2011; Guo et al., 2014). Based on functional variations, rhodopsin can be classified as light-driven proton pumps, chloride pumps, Na+ pumps and signal transducers (Ernst et al., 2014). Among these functional types, the proton pump rhodopsin (PPR), in association with all-trans retinal, absorbs light and drives light-activated proton across cell membranes to generate an outward proton gradient. This results in proton outflux and production of ATP (Martinez et al., 2007). Thus PPR enables cells to acquire energy from light independently of plastid photosystems (Sharma et al., 2006; Walter et al., 2007; Gomez-Consarnau et al., 2010). Compared with the photochemical reaction in photosynthetic organisms, the rhodopsin-based phototrophic mechanism is more efficient due to the simplicity of the molecular machinery needed.

Proton pump rhodopsin was first discovered in archaea *Halobacterium salinarum* in early 1970s (Oesterhelt and Stoeckenius, 1971), and is now known to exist in marine γ-Proteobacteria (Beja et al., 2000) as well as a wide range of other bacteria (de la Torre et al., 2003; Venter et al., 2004). In recent years, PPR-like coding genes were found in some cultured and uncultured dinoflagellates, haptophytes, and diatoms (Lin et al., 2010; Marchetti et al., 2012). The detection of PPR genes in phylogenetically diverse taxa of dinoflagellates suggests that PPR occurs widely in this phylum (Lin et al., 2010). Yet, this gene has not been reported in the order of Prorocentrales, which are common in the world's oceans.

Factors that regulate rhodopsin gene expression in eukaryotes remain poorly studied. Researchers have mainly examined the effect of nutrient and light conditions on PPR in bacteria. For example, it has been reported that in *Vibrio* strain AND4, rhodopsin gene expression was affected by nutrient limitation; in nutrient limited media, rhodopsin gene expression in this strain was strongly up-regulated leading to increased survival of the strain (Akram et al., 2013). Similar studies have been conducted for other bacteria to detect physiological functions of rhodopsin and their effects on population growth (Lami et al., 2009; Gomez-Consarnau et al., 2010; Steindler et al., 2011). These studies indicate that at least in some bacterial strains, rhodopsin enhances survival of the host species of bacteria under nutrient deficiency. Meanwhile, light has also been shown to influence rhodopsin gene expression in bacteria (Gomez-Consarnau et al., 2010). In *Dokdonia* sp. strain MED134, rhodopsin gene expression increased in light-cultivated cultures compared to dark-grown cultures (Gomez-Consarnau et al., 2010). In freshwater microbial communities, a metatranscriptome study showed that rhodopsin was expressed at a higher level in the light than in the dark (Vila-Costa et al., 2013). Furthermore, the rhodopsin gene also has been reported to respond to different spectra at different water depths. Shallow seawater tends to favor green-light-absorbing rhodopsin, while deeper waters favor bluelight-absorbing rhodopsin (Man et al., 2003; Fuhrman et al., 2008). The two types of rhodopsins in SAR86, with a single amino acid residue substitution at position 105, display different maximal absorbance spectra potentially enabling adaptation to their respective environments (Beja et al., 2001; Man et al., 2003).

In this study, we identified a rhodopsin gene from the Prorocentrales dinoflagellate *P. donghaiense*, which is a dominant harmful algal bloom (HAB) species in the East China Sea, where it forms HABs almost every year (Lu et al., 2005). As HABs formed by this species have been linked to relatively weak light field in the subsurface layer of a turbid water column (Sun et al., 2008), it is of interest to examine how the expression of this rhodopsin gene responds to changes in light conditions. We investigated the expression dynamics of *P. donghaiense* rhodopsin under different light/dark regimes, light intensities, and light spectra to deduce the potential contribution of this gene to enhanced fitness in *P. donghaiense*. To more closely link the gene transcriptional pattern to function, we also developed an antiserum and used it to determine *P. donghaiense* rhodopsin protein abundance in the cultures grown under different light intensities.

# Materials and Methods

### Algal Culture and Sample Collection for Rhodopsin Gene Identification

*Prorocentrum donghaiens* was originally isolated from a HAB event in East China Sea in 2009, and was provided by the Center for Collections of Marine Algae in Xiamen University (source culture number:CCMAXU-364). In this study, the culture was first grown in L1 medium with an antibiotic cocktail (200 mg/L ampicillin, 100 mg/L kanamycin and 100 mg/L streptomycin). The culture was verified to be bacteria free microscopically by DAPI staining (Life Technologies, Grand Island, NY, USA) of filtered samples and molecularly by 16S rDNA PCR of extracted DNA. To perform our experiments, the axenic culture of *P. donghaiense* was then transferred into fresh autoclaved L1 seawater medium (without silicate) at 20 ± 1◦C under a 14:10 h light:dark cycle with a photon flux of 100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1. The experiments were carried out in Yiheng incubators (Yiheng Technical Co., Ltd, China), with illumination provided by fluorescent light bulbs (Foshan Illumination Company, China). Cell counts were taken daily using a Sedgwick-Rafter counting chamber under the microscope, and growth curves were plotted to indicate growth stages. When the culture entered the mid exponential growth phase, cells (∼107 cells per sample) were harvested by centrifugation at 3000 × *g* at 20◦C for 10 min. The cell pellets were resuspended in 1 ml TRIzol Reagent (Invitrogen, Carlsbad, CA, USA) and stored at −80◦C for subsequent RNA extraction.

### Light Manipulation to Study Rhodopsin Expression Pattern in *P. donghaiense*

The first experiment was carried out under a 14:10 h light dark regime. Culture was first synchronized as previously reported (Shi et al., 2013), and the synchronized culture was then transferred into 7.5-L L1 medium in triplicate. Three days later, when the cultures were in the exponential phase, 400 ml samples were collected as described above every 2 h over a 24-h light/dark cycle. Cell pellets were thoroughly resuspended in 1 ml TRIzol Reagent by vortex and stored at −80◦C until RNA extractions.

In the second experiment, in order to measure rhodopsin expression levels under continuous light and continuous darkness, the synchronized culture grown as described above was split into two groups, with three replicates of each sample. One group was transferred to continuous illumination and the other group to continuous darkness. Twenty-four hours later, a sample was taken every 2 h for a 24 h period. This set of samples has been used previously on Rubisco gene expression (Shi et al., 2013).

In the third experiment, cultures were grown under different spectra to determine the response of *P. donghaiense* rhodopsin expression to chromatic variations. Triplicated cultures were grown under red (T8 30W/R, wavelength 622–700 nm, peak at 660 nm), green (T8 30W/G, wavelength 492–577 nm, peak at 560 nm) and blue lights (T8 30W/B, wavelength 455– 492 nm, peak at 470 nm) provided by fluorescent lamps (Foshan Illumination Company, China) at equal intensities (100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1) measured using digital luxmeter (TES1332A, Taiwan). Cell concentrations were determined daily as described earlier. From day 3 to day 7, samples were collected daily from each culture at the same time of the day using centrifugation as described above.

The fourth experiment was carried out to study *P. donghaiense* rhodopsin expression under different light intensities. A synchronized culture was split into three groups, with three replicates of each. The cultures were grown under 14:10 light dark cycle, at 100 (for convenience named normal light here), 20 (low light), and 200 μE·m−2·s−<sup>1</sup> (high light). Daily sampling and cell concentration determination were performed as described above.

To further measure *P. donghaiense* rhodopsin protein abundance under different light intensities and light-dark cycle, the synchronized culture was split into groups (three replicates each) that were grown under four light intensities (25, 50, 100, and 200 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1). Cells were collected 4 h after the onset of the light period and 2 h after the onset of the dark period. Samples were harvested as described above. The cell pellets were resuspended in 0.5 ml PBS (phosphate-buffered saline) for subsequent protein extraction.

### RNA Extraction and cDNA Synthesis

Total RNA was extracted using TRI-Reagent (Molecular Research Center, Inc., Cincinnati, OH, USA) coupled with Qiagen RNeasy Mini kit (Qiagen) following previously reported protocol (Lin et al., 2010). The potential DNA contaminant was eliminated using RQ1 DNase (Promega) according to the manufacturer's protocol and the resultant RNA was purified using Qiagen RNeasy Mini kit. RNA concentrations were measured using NanoDrop ND-2000 Spectrophotometer (Thermo Scientific), and the qualities were assessed using the absorbance ratios of 260/280 nm and 260/230 nm.

For each sample, 300 ng total RNA was used in cDNA synthesis. GeneRacer oligo-dT (Invitrogen, Carlsbad, CA, USA) was used as the primer in the case where the resultant cDNA was for rhodopsin gene amplification. For cDNA to be used in RT-qPCR, oligo-(dT)16 primer was used.

### Identification of *P. donghaiense* Rhodopsin cDNAs and Sequence Analysis

A rhodopsin gene fragment was obtained from a transcriptomic dataset generated using GS-FLX+ Titanium (454 Life Sciences, Roche, Branford, CT, USA) sequencing (unpublished data). A gene specific forward primer was designed based on the partial gene sequence (*PCDH-rhod-F*, **Table 1**) and was used with cDNA 3-end adaptor primer to amplify the 3 end of the cDNA. PCR was run under the program consisting of initial denaturation at 95◦C for 3 min, followed by 35 cycles 95◦C 15 s, 56◦C 30 s, 72◦C 45 s, and a final step of 72◦C for 5 min. The cDNA amplicon was purified, cloned, and sequenced. A gene specific reverse primer (*Pdrhod*-qR, **Table 1**) was paired with DinoSL as the forward primer (Lin et al., 2010) to amplify the 5 end of the cDNA.

Transmembrane helical structure was analyzed using ProteinPredict (Yachdav et al., 20141 ). Protein secondary structure was displayed by web software TOPO22 . Multialignment was conducted to identify conserved amino acid residues that are presumed to form retinal pocket and those that are known to be functional residues in other species.

### Phylogenetic Analysis

To determine the affinity of the *P. donghaiense* rhodopsin, phylogenetic trees were constructed using the amino acid sequences of this gene. Reference protein sequences identified from BLAST results were retrieved from NCBI to combine with the sequences generated from this study. Alignment of these sequences was carried out using ClustalX (Larkin et al., 2007). Phylogenetic trees were inferred using neighbor-joining (NJ; Saitou and Nei, 1987) and Bayesian (BE; Huelsenbeck and Ronquist, 2001) methods. BE analysis was run for 100,000 generations with trees sampled every 100 cycle and the first 25,000 were discarded as burn-in. Tree topology was shown by the result of NJ analysis (with JTT amino acid substitution method) and support of the nodes was obtained from both BE and NJ analyses.

### Gene Expression Analysis Using Reverse Transcription Quantitative PCR (RT-qPCR)

Reverse transcription quantitative PCR was performed with cDNA templates prepared from samples collected from the various experiments described above, using iQTM SYBR<sup>R</sup> Green Supermix in 96-well plates on a CFX96 Real-time PCR System (BioRad, USA). Each reaction was carried out in a total volume of 12 μl containing 250 nM of each primer, 5 μl cDNA or DNA, and 6 μl 2×SYBR<sup>R</sup> Green Super mix. To prepare a standard curve, PCR amplicon of *P. donghaiense* rhodopsin was obtained from a plasmid containing the whole coding region. To achieve accurate standards, amplicon or restriction digested plasmid, instead of whole plasmid were used (Hou et al., 2010). The amplicon was purified and quantified using NanoDrop, and then serially diluted by 10-fold to obtain a gradient of 102–107 gene copies per 5 μl. The standard series and the experimental cDNA samples were run on the same PCR plates using the thermo cycle program as reported previously (Zhang and Lin, 2003). All reactions were carried out in three technical replicates.

<sup>1</sup>www*.*predictprotein*.*org

<sup>2</sup>http://www*.*sacs*.*ucsf*.*edu/cgi-bin/open-topo2*.*py


### TABLE 1 | Primers used in this study.

Data were analyzed using CFX software (Bio-Rad, Hercules, CA, USA). In order to normalize rhodopsin gene expression across different samples, several reference genes, including calmodulin (*calm*), glyceraldehyde 3-phosphate dehydrogenase (*gapdh*), a-tubulin and mitochondrial cytochrome b (*cob*), were selected to compare their expression stability (Supplementary Figure S3). *calm* showed the greatest stability and was selected to normalize expression levels of rhodopsin. The expression level of *calm* was also determined on the same qPCR plates as rhodopsin, with its standard curve (Supplementary Figure S4) prepared as previously reported (Shi et al., 2013).

### Rhodopsin Antibody Preparation and Western Blot Analysis

A synthetic peptide with the sequence CVTYAKSNKDGALLA, identical to the C terminus of the rhodopsin, was produced and used (peptide–KLH conjugate) to immunize two rabbits at Genscript Corporation (Piscataway, NJ, USA). The resultant polyclonal antibodies (PdRHODab1 and PdRHODab2) were affinity-purified and tested for titer using enzymelinked immunosorbent assay. PdRHODab1, with a high titer (1:512,000), was chosen for use in this study.

Samples collected from the light dark regime and light intensity experiments were homogenized in PBS buffer as for RNA extraction. The homogenate was then centrifuged at 3000 × *g* at 4◦C for 15 min, and the supernatant equivalent to <sup>5</sup> <sup>×</sup> <sup>10</sup><sup>5</sup> cells from each sample was mixed with Laemmli buffer, and incubated at 95◦C for 5 min. The samples were then loaded in 10% SDS-PAGE gels (Bio-Rad) and electrophoresed at 100 V for 1 h. The resolved proteins were transferred to a polyvinylidene difluoride (PVDF) membrane (Millipore, Bedford, MA, USA) using Trans-Blot SD Semi-Dry Transfer Cell (Bio-Rad, USA) at 25 v for 30 min. Membranes were blocked with 5% non-fat milk for 2 h and then incubated with the rhodopsin antiserum with 5000-fold dilution in PBS for 2 h at room temperature. Following three time washes with PBS containing 0.05% Tween 20 (PBST), the membrane was incubated with a biotinylated goat anti-rabbit IgG (TransGen Biotech, Beijing, China) in 10,000 fold dilution for 1 h at room temperature and then washed seven times in PBST. Finally, the membrane was washed and incubated with a horseradish peroxidase-labeled streptavidin solution (Beyotime Institute of Biotechnology, Shanghai, China). The immunoreactive bands were detected using the enhanced chemiluminescent (ECL) Substrate (Invitrogen, Carlsbad, CA, USA). The immunodetection procedure was essentially the same as we previously reported (Lin et al., 1994). The protein band image was captured using Bio-red Gel Doc XR. Following the same procedure, glutaraldehyde phosphate dehydrogenase (GAPDH; Ku et al., 2013) was detected on a protein blot prepared in parallel to the one used for rhodopsin protein detection. The primary antibody against GAPDH provided by Sangon (Cat #: AB90090, Shanghai, China) was used at 1:1000 dilution. Band intensity was measured using Bio-red Gel Doc XR equipped with Quantity One software (Bio-Rad Laboratories, ShangHai, China).

### Spectroscopy to Determine *P. donghaiense* Rhodopsin Absorption Optima

*Prorocentrum donghaiense* rhodopsin encoding sequence was amplified from the full-length rhodopsin cDNA using primer Pdrhod-cF and Pdrhod-cR (**Table 1**). The product was cloned into pEASY-E1 expression vector (TransGen Biotech) and transformed into *Escherichia coli* strain BL21. The *E. coli* culture was grown with all-*trans* retinol (final concentration 0.01 mM) and IPTG (final concentration 1 mM) to induce *P. donghaiense* rhodopsin expression for 3 h at 30◦C with shaking at 200 rpm. Three miniliters of the culture were used to measure absorption spectrum in standard 1-cm cuvettes on Cary 100 spectrophotometer (Varian Instruments). Absorption was scanned from 450 nm to 650 nm at 0.1 nm intervals. Rhodopsin absorption spectrum was obtained by subtracting the spectrum of a negative control from the spectrum of the experimental culture; the negative control was *E. coli* strain BL21 transformed with the cloning vector without a target gene insert.

### Results

### *P. donghaiense* Rhodopsin Identification, Function Prediction and Phylogenetic Inference

A full-length cDNA (1013 bp, with spliced leader in the 5 -UTR and polyA in the 3 -UTR region was obtained (GenBank accession number, KM282617), and BLAST result showed that it was a rhodopsin. The cDNA encoded a protein of 258 amino acid residues with predicted molecular mass of 28.8 kDa. Transmembrane domain analysis predicted that *P. donghaiense* rhodopsin has seven transmembrane domains (Supplementary Figure S1A), the conserved feature of rhodopsin.

This gene also contains the same conserved functional residues as *Oxyrrhis marina* rhodopsin of the proton pump type (Slamovits et al., 2011). As shown in Supplementary Figure S1B, position 96 is an Asp (101 in *O. marina*), which is predicted to be a proton acceptor; position 107 (112 in *O. marina*) is Glu, a proton donor; and position 235 (237 in *O. marina*) is Lys, which is predicted to form the retinal pocket to harbor the retinal. There is a Leu residue at position 104, equivalent to position 105 in eBAC31A08 that has been shown to be a green-light-absorption-tuning switch residue (Man et al., 2003). Further, there is a Trp in position 155 (156 in *Salinibacter*), a hallmark of Xanthorhodopsin subgroup II, in contrast to Gly at this position in subgroup I (Vollmers et al., 2013). Substitution of Gly in this position by the bulky Trp abolishes binding of keto-carotenoids (Imasheva et al., 2009; Slamovits et al., 2011).

Phylogenetic analysis of amino acid sequences showed that *P. donghaiense* rhodopsin clustered with most of the dinoflagellate rhodopsins in a clade that otherwise consisted exclusively of rhodopsins from proteobacteria. This dinoflagellate rhodopsin clade belongs to xanthorhodopsin subgroup II (**Figure 1**). The only obvious exception was *Karlodnium veneficum* (formerly *K. micrum*), which was affiliated in a separate clade mainly composed of proteorhodopsins from bacteroidetes *Winogradskyella* and proteobacteria *Pelagibacter*. Additionally, we also found a very strong bootstrap support (100% from NJ and 0.99 from Mr. Bayes) for the affinity between the xanthorhodopsin clade and the proteorhodopsin clade. The apparent monophyletic grouping of major dinoflagellate rhodopsins was disrupted by rhodopsin recently reported from the haptophyte *Phaeocystis globosa* that branched with *P. donghaiense* rhodopsin into a distinct subclade, separated from the major dinoflagellate subclade. However, bootstrap support of the separation was not significant, probably due to too few taxa in the *Prorocentrum*/*Phaeocystis* cluter.

FIGURE 1 | Phylogenetic relationship of dinoflagellate rhodopsin with other typical rhodopsins based on amino acid sequences. Tree topology shown is from neighbor-joining (NJ) analysis, which is similar to that produced by Bayesian (BE) analysis. Values shown at nodes are bootstrap support of NJ/posterior probability of BE analyses (only values

*>*50%/0.50 are shown). General grouping of microbial rhodopsin is shown in separate light gray boxes with names placed on the lower right of the box. Dark gray box highlights *P. donghaiense* rhodopsin sequences obtained in this study. Bracket and arrow depict dinoflagellate proton pump rhodopsin (PPR) groups.

### *P. donghaiense* Rhodopsin Expression Profile Under LD, DD, and LL Light Regimes

Rhodopsin transcript abundance relative to reference gene *calm*, exhibited a clear diel rhythm when the culture was cultured under the LD cycle with a photon flux of 100 μE·m−2·s−<sup>1</sup> (**Figure 2A**). At the beginning of the dark period (h0), the transcripts abundance was in a lower level. It increased slowly from the middle (h4) to the late part of the dark period (h8). After the light period began (h10), *P. donghaiense* rhodopsin transcript abundance increased rapidly to reach a maximum in the middle of the light period (h12). Thereafter, the transcript level declined until the dark period. Throughout the LD cycle, the amplitude of the rhodopsin transcript dynamics was 4.8 fold. The same expression dynamic pattern was observed when rhodopsin transcript abundance was normalized to total RNA

(Supplementary Figure S2A), except that the peak appeared 2 h later, and the amplitude of the dynamics was 6.5-fold.

In the DD and LL cultures, the expression level of *P. donghaiense* rhodopsin did not exhibit the same rhythm as that under LD (**Figures 2B,C**). The transcript abundance fluctuated only slightly, with a fold change of 1.7 and 1.6 in the DD cultures whereas 1.9 and 2.4 in the LL cultures when normalized to the reference genes *calm* and total RNA, respectively.

### *P. donghaiense* Rhodopsin Expression Profile Under Different Light Spectra and Intensities

Rhodopsin transcript abundance normalized to *calm* was significantly higher when the cultures were exposed to white, blue, and green light spectra than when exposed to red light (onetailed *t-*test, *<sup>p</sup> <sup>&</sup>lt;* 0.01, *<sup>n</sup>* <sup>=</sup> 6; **Figure 3A**, Supplementary Table S1). The average expression level under green and blue light was about 1.55-fold and 1.23-fold higher respectively than culture

exposed to red light. Cultures exposed to green light showed a slightly higher expression level than exposed to blue light, but not significantly (one-tailed *t*-test, *n* = 6).

Under the "normal" photon flux (100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1), rhodopsin transcript abundance maintained a relatively stable level (**Figure 3B**, Supplementary Table S2). In contrast, when the culture was shifted to a lower (20 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1) or a higher light intensity (200 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1), gene expression decreased markedly (one-tailed *t*-test, *p <* 0.01, *n* = 6), both to a relatively stable and similar level, more so when normalized to *calm* (**Figure 3B**).

### *P. donghaiense* Rhodopsin Protein Abundance Under Different Light Intensities and Light-Dark Cycle

*Prorocentrum donghaiense* rhodopsin was significantly more abundant in the light period than in the dark period, except when the culture was in 200 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup><sup>1</sup> (**Figure 4**). Under <sup>25</sup> <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1, 50 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1, and 100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1, the protein levels in the light period were 1.07–2.42 folds higher than in the dark period when normalized to the amount of total proteins and the fold change increased to 1.75–3.32 when normalized to the reference protein GAPDH. In the <sup>200</sup> <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup><sup>1</sup> treated cultures, *P. donghaiense* rhodopsin was 1.63-folds higher in the dark than in the light period when normalized to the amount of total proteins, and the fold change was 1.44 when the expression level was normalized to GAPDH.

Meanwhile*, P. donghaiense* rhodopsin abundance in the light period showed a parabolic profile with light intensity. It was in a low level under 25 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1. The expression increased steadily with light intensity increase, reaching the highest level at 100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1. At this light intensity level, the protein abundance was about 2.8-folds higher than that under 25 μE·m−2·s−1. Yet at the light intensity of 200 μE·m−2·s−<sup>1</sup> the expression level decreased (**Figure 4**). When the protein level was normalized to GAPDH, a similar expression profile was detected, except for the abundance at 50 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup><sup>1</sup> being slightly higher than in 100 μE·m−2·s−1. The effect of light intensity seemed to extend to the dark period. Compared to the expression level in 25 μE·m−2·s−1, 2.3, 1.77 and 3.89-folds up regulation ware detected in 50 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1, 100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1, and 200 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup><sup>1</sup> respectively (**Figure 4A**). The trend remained the same when the expression levels were normalized to GAPDH (**Figure 4C**).

### Blue-Green Absorption Spectrum of Over-Expressed *P. donghaiense* Rhodopsin

Spectroscopic analysis showed that the bacteria over-expressing *P. donghaiense* rhodopsin had a broad absorption peak in the blue-green spectrum, with green absorption slightly higher than blue absorption (**Figure 5**).

### Discussion

Photochemical reaction centers and retinal-activated proton pumps (PPR) are two different mechanisms used by organisms to harness solar energy. The former usually involves at least 30 plastid enzymes to form a complex system to harvest solar energy and fix carbon dioxide to provide energy for cell growth. The latter, in contrast, employs a simple mechanism to form a proton gradient to activate ATPase using a single membrane protein rhodopsin (Beja et al., 2001; Finkel et al., 2013), which is presumably more efficient. Therefore, it would seem favorable for an organism to harbor this light energy harvesting mechanism

biological triplicates. *p* values of pairwise comparison (*t*-test) are shown on

solid lines.

the Western blot shown in (A). The rhodopsin protein expression level is relative to that under light intensity of 25 <sup>μ</sup>E·m−2·s−<sup>1</sup> during the light period

to supplement photosynthetic apparatus. That may explain why PPR is so widespread in the aquatic ecosystem (Venter et al., 2004). As an ecologically successful group of aquatic eukaryotic microbes, it is not surprising that PPR exists widely in dinoflagellates. The presence of the conserved critical residues (those making retinal pocket, electron donor and acceptor) and light-responding features observed in this and previous studies suggests that dinoflagellate rhodopsins of this kind (aside from the sensory type) likely have a similar function to bacterial PPR (Lin et al., 2010; Slamovits et al., 2011). As dinoflagellate PPRs are believed to have been acquired through horizontal gene transfer (HGT) from bacteria (see next section), functional conservation is expected. Both groups of organisms use PPR for the same reason. However, the functional extrapolation of bacterial PPR to dinoflagellate rhodopsin should be taken with caution due to the significant difference in cellular and molecular machinery between bacteria and eukaryotes. Direct experimental evidence, e.g., measured proton pump activity, is still required to verify the physiological function of rhodopsin in this species.

### The Affiliation of *P. donghaiense* Rhodopsin with Xanthorhodopsin Subgroup II and Evolution of Dinoflagellate Rhodopsins by Horizontal Gene Transfer

The presence of DinoSL at the 5 end of *P. donghaiense* rhodopsin cDNA indicates that the sequence was indeed from a dinoflagellate rather than from bacteria. All dinoflagellate rhodopsins except some in *O. marina* (which possesses both sensory and proton pump types of rhodopsin) belong to PPR type. Within this type, there is a xanthorhodpsin group, which is further divided into subgroups I and II (Vollmers et al., 2013). Our phylogenetic inference clearly placed *P. donghaiense* rhodopsin, along with all other dinoflagellate PPRs (except those in *K. veneficum*), in xanthorhodopsin subgroup II. This affiliation has strong statistical support (96% NJ bootstrap/1.00 BE posterior probability). This suggests that dinoflagellate rhodopsins do not bind to the 4-keto-carotenoid antenna pigments (Vollmers et al., 2013).

The separation of *K. veneficum* rhodopsin from typical dinoflagellate rhodopsins in our phylogenetic tree agrees with previous findings and lending further support to the proposition that dinoflagellate rhodopsins have arisen at least twice independently through HGT (Lin et al., 2010; Slamovits et al., 2011). Furthermore, our observation that *P. donghaiense* rhodopsin branched with a haptophyte rhodopsin in a distinct subclade suggests that *P. donghaiense* rhodopsin might have been acquired in yet another HGT event, and shares with the haptophyte *Phaeocystis* a common bacterial rhodopsin progenitor. This requires more rigorous phylogenetic analysis in the future with broader taxon sampling to achieve significant bootstrap support. Whereas *K. veneficum* plastid is haptophyte originated (Tengs et al., 2000), it is curious that *K. veneficum* rhodopsin is so distantly separated from haptophyte rhodopsin. However, this is not entirely surprising given that rhodopsin and chloroplast have independent evolutionary histories.

### The Expression of Rhodopsin is Light Dark Cycle-Dependent in *P. donghaiense*

Understanding how the expression of rhodopsin responds to illumination variability sheds light on the protein's function in the organism. Both rhodopsin transcript and protein in *P. donghaiense* showed higher abundances in the light period under normal light intensity (100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1) than in the dark. Our qPCR result showed a remarkable diel oscillation in rhodopsin gene transcription: low transcript abundances in the dark with a minimum in the mid-dark period, and high transcript abundances in the light period with a peak in the mid-light period. The same diel rhodopsin expression pattern also has been detected in a meta-transcriptomics study on phosphorus limited lake microbial community, where the dominant type of rhodopsin was bacterial PPR (Vila-Costa et al., 2013). Similarly, rhodopsin protein abundance in the light period was significantly higher than in the dark period when the cells were cultured under normal light intensity (100 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1), as shown in our Western blot result. All the data suggest that rhodopsin expression may be light-dark cycle related.

The observed diel rhythm may be attributed to a circadian clock or simply to a light/dark-triggered oscillation. Circadian clock regulated rhythm would persist at least 2–3 days after the shifting from light dark (LD) cycle to continuous light (LL) or continuous darkness (DD) (Hwang et al., 1996). Therefore, we transferred the LD-grown culture to LL and DD, respectively, and analyzed rhodopsin transcript abundance over a 24 h period. The transcript abundance under LL and DD did not display the diel rhythm that was observed under LD. This result suggests that rhodopsin expression is directly influenced by light dark conditions, and is not under circadian clock control. This illumination-responsive feature would enable the organism to promptly tune to varying light conditions in its habitat.

It is interesting to note that the transcript abundance of *P. donghaiense* rhodopsin in LL and DD were basically the same in our study. This is different from rhodopsin gene expression profile in bacteria such as Flavobacteria and SAR11 (Gomez-Consarnau et al., 2007; Lami et al., 2009), in which the abundance of rhodopsin transcripts was dramatically higher under LL or LD than under DD (Lami et al., 2009). While PPR depression in DD seen in bacteria is as what would be expected for a light absorbing protein, the lack of difference between DD and LL in *P. donghaiense* cannot be explained. There is a possibility that the expression of this gene is controlled by light dark transition cues as previously proposed for phytoplankton (Fuhrman et al., 2008). This requires further investigation.

### Effects of Spectrum and Light Intensity: Potentially Adaptive to Natural Light Field in *P. donghaiense* Habitat

Cruise surveys and field experiments of *P. donghaiense* in East China Sea have revealed that the highest cell density occurred at middle water depths (deep chlorophyll maximum layer, usually located at 10–50 m in *P. donghaiense* bloom area) prior to a bloom outbreak (Sun et al., 2008; Chen et al., 2011; Wen et al., 2012). At this depth, long wavelength spectra such as red and yellow would have largely disappeared due to absorption by particles and water molecules, leaving green and blue light as the major available spectra (Kirk, 1994). Our qPCR results of samples collected from cultures grown under different spectra suggest that the rhodopsin transcript level under green light was somewhat higher than it was under blue light, but both were significantly higher (by 1.67 ± 0.41 folds when data from all sampling days were considered and 1.83 ± 0.30 folds if the last day data were excluded) than under red light (*p <* 0.05). In accordance, *P. donghaiense* rhodopsin is presumably a greenlight-absorbing type rhodopsin because of the Leu residue at the position 104 (equivalent to position 105 in eBAC31A08). PPR with Leu at this position has been shown to have an absorption maximum in the green light spectrum in bacteria such as SAR86 and *Dokdonia sp.* strain MED134 (Man et al., 2003; Gomez-Consarnau et al., 2007), which is believed to promote bacterial growth under green light condition. Furthermore, the absorption spectrum from *P. donghaiense* rhodopsin cloned into and overexpressed in *E. coli* also showed the maximum absorption in the green spectrum although it has a broad absorption peak from blue to green. As chloroplast mainly absorbs blue light, green light dominates coastal waters, making the green shift of absorption spectrum highly adaptive in the coastal marine environment (Beja et al., 2000).

Previous reports have suggested that light intensity is one of the most important factors influencing the bloom dynamics of *P. donghaiense* in East China Sea (Chen et al., 2006). This species normally blooms at relatively muddy sea areas where light penetration is relatively low (∼<sup>175</sup> <sup>±</sup> 17.4 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup>1; Sun et al., 2008). As PPR can putatively function as a source of energy subsidy, it is of interest to examine how the expression of this protein/gene responds to different light intensities. When the cultures were treated with different light intensities, the transcript abundance of *P. donghaiense* rhodopsin showed the highest level in cells grown under a moderate light intensity (100 μE·m−2·s−1) and decreased considerably when the cultures were transferred to low (20 μE·m−2·s−1) or high (200 μE·m−2·s−1) light conditions. The same expression pattern was detected in two consecutive days. Our Western blot results also showed that the encoded protein too was most abundant at moderate to lower light intensities (100 μE·m−2·s−<sup>1</sup> when it was normalized to total protein, 50 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup><sup>1</sup> when normalized to reference GAPDH). Thus, both the qPCR and the Western blot data in concert suggest an adaptation of the PPR system in *P. donghaiense* to this organism's commonly occurring turbid habitat. The decrease in both transcript and protein abundances of *P. donghaiense* rhodopsin at 200 <sup>μ</sup>E·m−2·<sup>s</sup> <sup>−</sup><sup>1</sup> perhaps should not be a surprise, in light of a previous study on bacterium *Psychoflexus torquis* showing higher PPR expression levels under dim light than under high light or darkness (Feng et al., 2013). The authors in that study suggested that *P. torquis* PPR expression under high light was influenced by photooxidative stress, because the bacteria cell abundance and growth rate was lower under this illumination condition. However, it seems unlikely in our case, because both cell abundance and growth rate were not decreased under 200 μE·m−2·s−1, at which photooxidative stress was suggested not to be so likely to take place in *P. donghaiense* (Xu et al., 2010). Based on current data, it is not clear why *P. donghaiense* rhodopsin transcript and protein abundance decreased under high light.

In summary, our qRT-PCR and Western blot results showed that *P. donghaiense* rhodopsin expression profile (high expression in the medium light intensity, during light periods, and under green/blue light wavelength) implies that this chloroplastindependent light energy harvesting system will enhance fitness of this organism under bloom conditions. These are consistent with what would be expected of a functional PPR. The light intensity as well as chromatic optima of rhodopsin *in P. donghaiense* expression is likely a consequence of evolutionary adaptation to the organism's living environment, including subsurface layer of a turbid water column. Therefore, this protein may provide *P. donghaiense* a fitness advantage, allowing it to outgrow other phytoplankton, which rely on chloroplast light harvesting system, and form intense blooms in turbid subsurface seawater. Even in calm clear water column, PPR with the light responding features of this protein will allow a dinoflagellate (able to migrate with the aid of its flagella) to photosynthetically utilize the more abundant nutrients at the nutricline depth even though light is dim there. However, further experiments, such as measure proton pump activity of *P. donghaiense* rhodopsin using techniques such as laser flash photolysis and gene knockout, are needed to prove this hypothesis.

### Acknowledgments

The work was supported by a 1000Plan Award of the Chinese government, a State Key Laboratory of Marine Environmental Science grant for exploratory research (MELRI1203), the Chinese National Science Foundation grant 41330959 and 41176091, and US National Science Foundation Early Concept Grant for Exploratory Research (EAGER) grant #OCE-1212392 (to SL), and the China Scholarship Council (to XS). We thank Dr. Steven Kelley from Scripps Institution of Oceanography for his help with English.

### References


# Supplementary Material

The Supplementary Material for this article can be found online at: http://journal*.*frontiersin*.*org/article/10*.*3389/fmicb*.* 2015*.*00555/abstract


**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 Shi, Li, Guo, Lin, Li and Lin. 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.*

# Differences in Intertidal Microbial Assemblages on Urban Structures and Natural Rocky Reef

*Elisa L.-Y. Tan1,2\*, Mariana Mayer-Pinto1,2, Emma L. Johnston1,2 and Katherine A. Dafforn1,2*

*<sup>1</sup> Evolution and Ecology Research Centre, School of Biological, Earth and Environmental Sciences, University of New South Wales, Sydney, NSW, Australia, <sup>2</sup> Sydney Institute of Marine Science, Mosman, NSW, Australia*

Global seascapes are increasingly modified to support high levels of human activity in the coastal zone. Modifications include the addition of defense structures and boating infrastructure, such as seawalls and marinas that replace natural habitats. Artificial structures support different macrofaunal communities to those found on natural rocky shores; however, little is known about differences in microbial community structure or function in urban seascapes. Understanding how artificial constructions in marine environments influence microbial communities is important as these assemblages contribute to many basic ecological processes. In this study, the bacterial communities of intertidal biofilms were compared between artificial structures (seawalls) and natural habitats (rocky shores) within Sydney Harbour. Plots were cleared on each type of habitat at eight locations. After 3 weeks the newly formed biofilm was sampled and the 16S rRNA gene sequenced using the Illumina Miseq platform. To account for differences in orientation and substrate material between seawalls and rocky shores that might have influenced our survey, we also deployed recruitment blocks next to the habitats at all locations for 3 weeks and then sampled and sequenced their microbial communities. Intertidal bacterial community structure sampled from plots differed between seawalls and rocky shores, but when substrate material, age and orientation were kept constant (with recruitment blocks) then bacterial communities were similar in composition and structure among habitats. This suggests that changes in bacterial communities on seawalls are not related to environmental differences between locations, but may be related to other intrinsic factors that differ between the habitats such as orientation, complexity, or predation. This is one of the first comparisons of intertidal microbial communities on natural and artificial surfaces and illustrates substantial ecological differences with potential consequences for biofilm function and the recruitment of macrofauna.

Keywords: biofilm, artificial structures, rocky shores, seawalls, 16S rRNA sequencing, Sydney Harbour

# INTRODUCTION

Coastal zones have great socioeconomic value supporting industry, trade, and growing urban populations. These systems also support great biological diversity and provide important ecosystem services such as nutrient cycling and food (Costanza et al., 1997; Ortega-Morales et al., 2010). Increasingly, urbanization and industrialization are transforming coastal zones around the

### *Edited by:*

*Justin Robert Seymour, University of Technology Sydney, Australia*

### *Reviewed by:*

*Steven Smriga, Massachusetts Institute of Technology, USA Bonnie Laverock, University of Technology Sydney, Australia*

> *\*Correspondence: Elisa L.-Y. Tan elisa.tly@hotmail.com*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 20 June 2015 Accepted: 31 October 2015 Published: 20 November 2015*

### *Citation:*

*Tan EL-Y, Mayer-Pinto M, Johnston EL and Dafforn KA (2015) Differences in Intertidal Microbial Assemblages on Urban Structures and Natural Rocky Reef. Front. Microbiol. 6:1276. doi: 10.3389/fmicb.2015.01276*

globe, and a substantial literature now documents the ecological changes to plant and animal communities associated with this physical transformation (Blockley, 2007; Glasby et al., 2007; Bulleri and Chapman, 2010). However, advances in molecular biology have only recently enabled us to study the impact that intense coastal development is having on the largely hidden, yet ecologically important, microbial communities.

More than 50% of the coastline has been modified in some regions of Japan (Koike, 1996), Australia (Chapman and Bulleri, 2003), USA (Davis et al., 2002), and Europe (Airoldi et al., 2005; Airoldi and Beck, 2007). Modifications include the addition of infrastructure to support activities, such as commercial and recreational boating, fishing, tourism, and waterside living (Airoldi and Bulleri, 2011; e.g., jetties, pilings, and pontoons) and coastal defense structures (e.g., seawalls and breakwaters; Vaselli et al., 2008; Browne and Chapman, 2011). This extensive coastal armouring has significant local and regional effects on natural systems, from loss of natural habitats, decreases in diversity, and increase in non-indigenous species to homogenization of systems (e.g., Dafforn et al., 2013).

Artificial structures usually differ from natural habitats in their substrate material, substrate age, orientation, shading, and habitat complexity (Glasby and Connell, 2001; Bulleri and Chapman, 2004; Bellou et al., 2012). Seawalls, for instance, generally have homogenous surfaces compared to natural rocky shores as they lack microhabitats such as rock pools, crevices, and overhangs (Chapman, 2006). Furthermore, as opposed to the relatively horizontal natural rocky shores, seawalls are often built vertically and have a steep slope (Chapman, 2006). Consequently, although they can support relative diverse assemblages of organisms, they do not mimic natural habitats and there is substantial evidence that they do not function as surrogates for rocky shores (Glasby and Connell, 1999; Bulleri and Chapman, 2010). While we have achieved substantial advances in understanding how and why macrofaunal communities differ between natural habitats and artificial structures (Glasby and Connell, 2001; Blockley, 2007; Glasby et al., 2007), the consequences of this habitat modification for microbial communities remain largely unknown.

The earliest microbial colonizers of any surface form biofilms (Dang and Lovell, 2000; Davey and O'Toole, 2000; Wahl et al., 2012) and their formation can be divided into several stages. Physico-chemical interactions allow for initial cell attachment on surfaces (e.g., plant surfaces, plastic, and sediment particles) and a cell monolayer is formed (Decho, 2000). Cells in the monolayer undergo proliferation, attracting other microbes for attachment, forming an active biofilm of microcolonies. Development of a mature biofilm through secretion of a matrix of mucilaginous extracellular polymers allows for cells to become motile and undergo chemotaxis. As a result, spreading of biomass and horizontal gene transfer occurs (Henschel and Cook, 1990; Singh et al., 2006). These complex aggregates of microbes consisting of mucus, microalgae, and bacteria (Wahl, 1989; Decho, 2000; Ortega-Morales et al., 2010) form the basis of many ecological processes that maintain the biosphere such as biogeochemical cycling of carbon and nutrients (Azam et al., 1993; Narváez-Zapata et al., 2005), primary production in intertidal systems (Moriarty, 1997; Decho, 2000), organic matter degradation (Davey and O'Toole, 2000), contaminant remediation (Singh et al., 2006) and trophic linkage (Ortega-Morales et al., 2010).

In marine systems, biofilms have at least three clearly defined roles – (1) as settlement, attachment and metamorphosis cues for a variety of sessile marine invertebrate larval (Holmström et al., 1992; Wieczorek and Todd, 1997; Thompson et al., 2005), such as mussels (Yang et al., 2014) and bryozoans (Dahms et al., 2004); (2) as primary attachment sites for plant and animal propagules (Wahl, 1989); and (3) the basis of intertidal food webs, as it is the major source of primary production and the largest biomass consumed *in situ*, in particular by grazers (Thompson et al., 2000, 2005). Due to their short generation time, biofilm bacteria are often at the forefront in responding to and recovering from environmental stressors (Lowe and Pan, 1996), altering their community composition and relative abundance in response to changes (Nocker et al., 2004). For example, biofilms are extremely sensitive to environmental changes – such as contamination variations in pH, nutrient and oxygen availability (Singh et al., 2006; Lear and Lewis, 2009; Mayer-Pinto et al., 2011).

Artificial structures may affect microbial communities directly through abiotic factors such as hydrodynamics, changing contaminant concentrations, material, shading, and habitat complexity (Geesey et al., 1992; Stoodley et al., 1998; Davey and O'Toole, 2000). Past studies have shown that biofilms that were exposed to laminar flows developed a different morphology to those exposed to turbulent flows. An increase in biofilm biomass has also been linked to increase in nutrient (carbon and nitrogen) and metal (zinc) concentrations (Stoodley et al., 1998; Mayer-Pinto et al., 2011). Artificial structures may also affect microbial communities indirectly via the assemblages that such structures support, e.g., grazers (Davey and O'Toole, 2000; Skov et al., 2010). Although there is direct removal of biofilm by grazing, there is also the potential for grazing to boost the photoautotrophic biomass of biofilms by removing the biofilm canopy and allowing greater light and nutrient penetration (Skov et al., 2010).

Studies of microbial community response to anthropogenic environmental change have given particular focus to artificial substrates in freshwater systems (Tien et al., 2009) and estuaries (Jones et al., 2007); while studies of artificial structures have identified changes to intertidal macrofauna compared with natural habitats (Bulleri and Chapman, 2004; Chapman, 2006; Bilkovic and Mitchell, 2013). Microbial biofilm formation and factors that regulate their assemblages have also been conducted on rocky shores (Thompson et al., 2004, 2005; Narváez-Zapata et al., 2005). However, no previous study has used amplicon sequencing to compare bacterial communities between artificial structures and natural shores and investigate the effects of increasing marine urbanization. Amplicon sequencing is a powerful tool for assessing structural changes in the biofilm communities colonizing artificial structures and the potential functional consequences (DeLong et al., 2006). We investigated whether potential structural changes to biofilm communities are related to the habitats or the local environmental conditions associated with the habitats. Microbial communities were identified using amplicon sequencing of the 16S rRNA gene to target bacteria, which form the primary component of biofilms (Davey and O'Toole, 2000).

# MATERIALS AND METHODS

## Experimental Design

The study was conducted in Sydney Harbour, New South Wales, Australia in the austral winter of 2014. Sydney Harbour is a highly urbanized area, with approximately half of the harbour's foreshore replaced by seawalls (Blockley, 2007). Sampling was done in the intertidal zone from two types of habitat; natural rocky shores and artificial seawalls. Four locations of each type of habitat were selected for this study based on their availability within the harbour and substantial effort was made to ensure that seawalls and rocky shores were interspersed. Natural rocky shores were studied at Taylors Bay, Chowder Bay, Farm Cove, and Shark Bay while studies on artificial seawalls were located at Bradleys Head, Kirribilli, Kurraba, and Watsons Bay (**Figure 1**). Seawalls were all vertical while rocky shores were mostly horizontal with a gentle slope (*<*20◦) because this is the most common type of rocky shore in Sydney Harbour.

This study investigated the variation in bacterial communities on the surface of natural rocky shores and artificial seawalls using recently cleared plots of existing surfaces and deployed surfaces (recruitment blocks). Six 10 cm × 10 cm plots were randomly chosen and cleared at each type of habitat, in each location, close to the high water mark, 0.8–1.1 m from the low tide line. Plots were marked using 6.5 mm raw plugs drilled at two opposing corners of each plot. All organisms found within the plot were removed using a hammer and a chisel. Plots were cleared thoroughly by scraping and scrubbing with an iron brush to ensure removal of all organisms and the existing biofilm on the substrata. This was done to ensure consistency of biofilm age across all samples. Newly formed biofilms were sampled 3 weeks after clearing. Mature biofilms are formed within days (Singh et al., 2006) and pilot setup of the experiment was tested to determine a suitable time frame for this study. At the same time that plots were cleared, four "recruitment blocks" were deployed for 3 weeks at each location to standardize the time frame available for biofilm colonization on all surfaces. Each "recruitment block" consisted of two 390 long × 140 deep × 190 mm high concrete bricks that were fastened together using cable ties to increase weighting and reduce the chance that replicates would be lost in high wave action. Blocks were placed adjacent to artificial or natural habitats and within 5 m of the nearest scraped plot. Recruitment blocks were deployed to account for possible differences in communities associated with habitat orientation, substrate material and age, as

Seawalls were studied at Bradleys Head, Kirribilli, Kurraba, and Watsons Bay.

well as the local history of the natural substrata found at each location.

# Environmental Parameters

Environmental parameters including temperature, salinity, dissolved oxygen (DO), and pH were measured adjacent to seawalls and rocky shores during the high tide immediately after biofilm samples were collected from the plots to investigate abiotic differences that might influence bacterial communities. Three replicate measurements were collected *in situ* at all locations using a water quality probe calibrated prior to data collection (YSI-Sonde 6600-v2, Yellow Springs, OH, USA).

# Bacterial Biofilm Community

Three weeks after set-up and deployment, a randomly selected 3 cm × 3 cm area (Thompson et al., 2004, 2005) of biofilm was swabbed for 20 s from the upper horizontal surface of each recruitment block and cleared plot at each location using sterile cotton tips. All biofilm samples were collected within 1–2 h of low tide to ensure evenness of residual seawater across all samples. Swabs from the cleared plots were randomly pooled into three replicates (each replicate consists of swabs from two separate plots). This was done because single replicate plots yielded insufficient DNA for individual sequencing. Swabs from recruitment blocks yielded sufficient DNA material for sequencing and were therefore not pooled. This provided four replicates from each location. Swabs were immediately placed in separate cryogenic vials and stored in liquid nitrogen in the field then frozen at −80◦C until DNA extraction, which was done within 2 weeks of sampling. Genomic DNA was extracted with the PowerBiofilm<sup>R</sup> DNA Isolation Kit (Mo Bio Laboratories Inc., Carlsbad, CA, USA), according to the manufacturer's instructions.

Amplicon sequencing of the 16S rRNA gene was done at the Molecular Research DNA Lab (MR DNA; Shallowater, TX, USA). Bacterial 16S primers 104F (Bertilsson et al., 2002) and 530R (Lane, 1991) were used to generate amplicon libraries for paired-end sequencing on the Illumina Miseq platform. Analysis of paired-end sequence data was processed using a proprietary analysis pipeline (MR DNA, Shallowater, TX, USA). Briefly, barcodes attached to sequences were removed. Sequence data underwent denoising and sequences shorter than 150 base pairs in length, and those with ambiguous bases and homopolymers exceeding 8 were removed to generate operational taxonomic units (OTUs). Next, chimeric sequences were identified and removed. OTUs were clustered at 3% divergence (97% similarity) to remove potential errors in sequence data. Taxonomy for the remaining OTUs was classified using the sequence alignment tool BLASTn against a curated GreenGenes database (DeSantis et al., 2006). Chloroplasts and OTUs with ≤5 occurrences were removed. This provided a total of 6497 OTUs sampled from plots and 8937 OTUs sampled from recruitment blocks.

# Statistical Analyses

The factors considered in the analyses were Habitat (fixed, two levels – natural rocky shores or seawalls) and Locations (random, four levels and nested within habitat). Due to differences in sampling methodology, bacterial community data collected were analyzed separately for plots and recruitment blocks and used to construct resemblance matrices using the Bray– Curtis dissimilarity index for a relative abundance-weighted measure of how similar the bacterial communities are in terms of their community structure (i.e., relative abundance and composition of species). Resemblance matrices were also constructed using Jaccard similarity index, for a comparison between communities based solely on the presence and absence of bacterial OTUs, for a measure of bacterial community composition. Principal co-ordinates analysis (PCO) was done to visualize the multivariate patterns in biofilm bacterial community structure and composition based on the data generated from each biofilm sample. Vector plots were overlaid to illustrate the relationship (*R >* 0.5) of bacterial classes to differences. Differences in bacterial community structure and composition were investigated with permutational multivariate analysis of variance (perMANOVA). Where significant differences were observed among habitats, the contribution of each bacterial class to the similarity/dissimilarity within/between habitats was further investigated with similarity percentage analysis (SIMPER; Clarke and Warwick, 2001; Clarke and Gorley, 2006).

Data were also rarefied to a common number of OTUs to investigate differences in diversity among habitats. Briefly, the lowest number of OTUs found in any one sample was assessed (770 OTUs) and the datasets were randomly subsampled to a common number of OTUs for a comparison of both alpha (OTU (species) richness, Shannon's diversity and Pielou's evenness) and beta (dispersion) diversity among habitats. Univariate diversity data, environmental parameters, were analyzed with permutational analysis of variance (perANOVA), using Euclidean distance. Beta diversity was calculated as the mean distance of individual observations for habitat to the group centroid for both community structure (Bray–Curtis) and community composition (Jaccard) using permutational analysis of multivariate dispersions (PERMDISP; Anderson, 2006).

All perMANOVAs were performed using Type III sum of squares to account for unbalanced data and 9999 permutations of residuals under a reduced model for raw data. Heterogeneity of dispersions was tested using PERMDisp. All data were analyzed using PRIMER-6 software (Clarke and Gorley, 2006) and its PERMANOVA+ add-on (Anderson et al., 2008).

# RESULTS

# Environmental Parameters

Temperature (◦C; Supplementary Figure S1A), salinity (psu; Supplementary Figure S1B), DO (mg/L; Supplementary Figure S1C) and pH (Supplementary Figure S1D) measurements did not range widely and did not differ significantly between seawalls and rocky shores, but were variable among locations (**Table 1**). Temperature ranged from 17.6 to 18.8◦C, while salinity ranged from 35.0 to 36.0 psu. DO varied from 12.2 to 15.1 mg/L and pH values varied from 8.3 to 9.2.

TABLE 1 | Permutational multivariate analysis of variance (perMANOVA) comparing environmental parameters (A) Temperature, (B) Salinity, (C) Dissolved Oxygen, (D) pH sampled at high tide from natural rocky shores and artificial seawalls.


*Factors include Habitat (Ha: fixed, 2 levels) and Locations (Lo: random, 4 levels, nested in habitat). Ns* = *not significant at p < 0.05;* ∗∗*p < 0.01.*

### Bacterial Biofilm Community

The structure of microbial communities (i.e., relative abundance and composition of OTUs) on seawalls was different (SIMPER: 84.89%) to those sampled on rocky shores (Ha: *Pseudo-F* = 1.35, *<sup>P</sup> <sup>&</sup>lt;* 0.05 **Table 2**, **Figure 2A**) and varied significantly among locations within habitat types [Lo (Ha): *Pseudo-F* = 3.28, *<sup>P</sup> <sup>&</sup>lt;* 0.01; **Table 2**, **Figure 2A**]. Communities within rocky shore plots were less similar (SIMPER: 19.63%) than communities within seawalls (24.25%). Differences in structure among habitats were separated along the PCO2 axis with seawalls positively correlated to PCO2 and rocky shores negatively correlated to PCO2 (**Figure 2A**). When the vector plot was overlaid the bacterial classes contributing most to structural differences separated out along the PCO1 axis. This suggests that the bacterial classes identified by the vector analysis primarily explained variation among locations that also separated out along PCO1, with the bacterial class *Alphaproteobacteria* most abundant in locations including Bradleys Head, Shark Bay, and Farm Cove (**Figure 2A**). The bacterial classes *Nitriliruptoria, Synechococcophycideae, Rubrobacteria, GN02, Subsection II, Deinococci,* and *Sphingobacteria* were negatively correlated with PCO1 axis indicating an increased abundance in Kurraba and Watsons Bay (**Figure 2A**).

SIMPER analysis indicated that 10 bacterial classes contributed to 92.52% of the dissimilarity among habitats sampled from plots (**Table 3**). Most of the dissimilarity was explained by the *Alphaproteobacteria* (28.31%)*, Subsection II* (20.13%), and *Bacilli* (10.02%) classes (**Table 3**). *Alphaproteobacteria, Bacilli, Gammaproteobacteria, Oscillatoriophycideae, Subsection IV, Flavobacteria,* and *Deltaproteobacteria* were most abundant on rocky shores, while *Subsection II, Synechococcophycideae,* and *Sphingobacteria* were most abundant on seawalls (**Table 3**). These taxa contributed to 19.43% of the dissimilarity among habitats. Bacterial community composition (based on presence/absence data only) differed only among locations and was not significantly different between plots sampled from seawalls and from rocky shores [Ha: *Pseudo-<sup>F</sup>* <sup>=</sup> 1.25, *<sup>P</sup> <sup>&</sup>gt;* 0.05; Lo (Ha): *Pseudo-F* <sup>=</sup> 1.89, *<sup>P</sup> <sup>&</sup>lt;* 0.01; **Table 2**, **Figure 2B**].

TABLE 2 | perMANOVA comparing bacterial community structure (Bray–Curtis) and composition (Jaccard) sampled directly from natural rocky shores and artificial seawalls ((A) Plots) and from substrate experimentally deployed in the habitat ((B) Recruitment blocks).


*Factors include Habitat (Ha: fixed, 2 levels) and Locations (Lo: random, 4 levels, nested in habitat).*

*Ns* = *not significant at p < 0.05;* <sup>∗</sup>*p < 0.05;* ∗∗*p < 0.01.*

Both the structure and composition of bacterial communities recruiting to deployed blocks did not differ significantly between seawalls and rocky shores (*<sup>P</sup> <sup>&</sup>gt;* 0.05, **Table 2**, **Figures 3A,B**), only among locations (*<sup>P</sup> <sup>&</sup>lt;* 0.01, **Table 2**).

Overall, biofilm communities sampled from both plots and blocks were dominated by *Alphaproteobacteria* (33.3 and 33.1% respectively, **Figures 4A,B**). *Gammaproteobacteria* (11.5% of plots, 19.8% of blocks) and *Cyanobacteria (Subsection II;* 22.4% of plots, 12.9% of blocks) were also major components of the biofilm community sampled from seawalls and rocky shores (**Figures 4A,B**).

Alpha diversity (species richness, Shannon's diversity, and Pielou's evenness) did not differ among habitats for both plot and recruitment blocks (*<sup>P</sup> <sup>&</sup>gt;* 0.05, **Table 4**, **Figure 5**). Beta diversity for both community structure (PERMDisp *P* = 0.2292 *>* 0.05) and community composition (PERMDisp *P* = 0.3494 *>* 0.05) on plots also did not differ between seawalls and rocky shores. Similarly, beta diversity on recruitment blocks for both community structure (PERMDisp *P* = 0.5706 *>* 0.05) and composition (PERMDisp *P* = 0.2482 *>* 0.05) did not differ between habitats.

Seven bacterial phyla recruited to the deployed blocks that were not present in plots sampled directly from seawalls and rocky shores (Supplementary Table S1). This included the phyla *Aquificae, Chlorobi, KSB1, NC10, NKB10, SBR1093,* and *WS3* (Supplementary Table S1).

### DISCUSSION

Our results showed that natural rocky shores and seawalls were colonized by similar bacterial taxa, but in different abundances. Much of the structural difference could be explained by greater relative abundances of *Proteobacteria* on rocky shores than seawalls. We also found that bacterial community composition and structure did not differ between artificial and natural habitats when differences in substrate orientation and surface were removed through the deployment of standard recruitment

species) and composition ((B) Jaccard – presence absence data only) sampled directly from natural rocky shores and artificial seawalls (Plots). Vector plot of bacterial taxa (at level of class) most strongly related (*R >* 0.5) to differences in community structures are also presented. Lengths of vectors indicate the strength and direction of relationships to measured variables.

### TABLE 3 | Results of SIMPER analysis giving dissimilarities among habitats sampled from plots.


*Listed are the 10 bacterial classes contributing most to the dissimilarity with respect to average abundance (Av. Abund), average dissimilarity (Av. Diss), quotient of dissimilarity and standard deviation (Diss/SD), % contribution to differences (Contrib%) and cumulative % contribution to differences (Cum.%). Abundance values highlighted in bold indicate the highest abundance of that class in a particular habitat.*

blocks. In addition, temperature, salinity, DO, and pH did not differ between types of habitats. Differences found in the intertidal bacterial communities are, therefore, likely due to intrinsic differences between artificial and natural habitats, e.g., surface complexity [hardness and texture (Coombes et al., 2011)] or biotic factors, rather than differences in the local environment, such as wave energy, water quality or other physico-chemical variables. Indeed, temperature, salinity, DO, and pH did not differ between types of habitats, so differences cannot be attributed to these environmental factors. The current study, however, did not quantify hydrodynamics and nutrient concentrations; thus there is a possibility that local flow regimes and nutrient concentrations were contributing drivers of the differences found in community structure.

# Bacterial Community Differences between Natural and Artificial Habitats

Microbial biofilms play an important role in biogeochemical processes. Differential abundances of bacterial taxa on natural rocky shores and artificial seawalls can have potential changes to function and influence on overall biological processes. Heterotrophic bacteria such as members of the *Proteobacteria* (e.g., *Alpha-, Gamma-,* and *Delta-proteobacteria*) are metabolically extremely diverse (Tujula et al., 2010). They are crucial to nutrient cycling and perform transformations and remineralization of material such as organic carbon and nitrogen (Azam et al., 1993; Azam, 1998). In the current study, all of these classes of *Proteobacteria* were relatively more abundant on natural rocky shores than seawalls. *Alphaproteobacteria* dominated the biofilm communities in this study and are often highlighted as the primary colonizing group of hard substrates, but are relatively less abundant in soft sediments. Given that a major impact of increasing marine urbanization is the replacement of soft sediments with artificial hard structures (Airoldi et al., 2009; Dafforn et al., 2013), this has the potential to alter the overall abundance of *Alphaproteobacteria* in the ecosystem. However, as highlighted in this study, seawalls may not provide a surrogate for rocky shores with respect to the bacterial taxa they support. While *Alphaproteobacteria* dominated the seawall biofilms, they were still relatively

less abundant than on rocky shores. This may represent reduced biogeochemical processing of carbon and nitrogen although further investigation, e.g., using metagenomics tools (Dinsdale et al., 2008) would be required to determine this.

Primary productivity is a principal process occurring in the oceans and represents the first stage in the flow of energy and matter through ocean systems (Cotner and Biddanda, 2002). In most aquatic habitats, the cyanobacteria are responsible for the majority of bacterial primary production (Capone et al., 1997; Liu et al., 1997). In the current study, several classes of *Cyanobacteria* (*Subsection II, Synechococcophycideae, Oscillatoriophycideae, Subsection IV*) were important in differentiating between rocky shore and seawall communities.


TABLE 4 | perMANOVA comparing alpha diversity (species richness, Shannon's diversity, Pielou's evenness) sampled directly from natural rocky shores and artificial seawalls ((A) Plots) and from substrate experimentally deployed in the habitat ((B) Recruitment blocks).

*Factors include Habitat (Ha: fixed, 2 levels) and Locations (Lo: random, 4 levels, nested in habitat). Ns* = *not significant at p < 0.05;* <sup>∗</sup>*p < 0.05;* ∗∗*p < 0.01.*

*Subsection II* (*Pleurocapsales*) cyanobacteria are known to survive extreme desiccation and UV radiation (Billi et al., 2000) and many have evolved salt tolerance mechanisms or actually require salt for growth (Cumbers and Rothschild, 2014). Together with *Synechococcophycideae*, these autotrophic cyanobacteria were more abundant on seawalls. In contrast, *Subsection IV* (*Nostocales*) and *Oscillatoriophycideae* were relatively more abundant on natural rocky shores than artificial seawalls. Members of the *Nostocales* group of cyanobacteria are filamentous and vegetative cells may differentiate into heterocysts that are then important in nitrogen fixation under aerobic conditions (Tomitani et al., 2006). The lower relative abundance of *Subsection IV* on seawalls compared to rocky shores may have implications for nutrient cycling if this translates to reduced potential for nitrogen fixation on artificial structures. However, additional manipulations and measurements would be required to quantify any functional differences. In addition, it has been shown that vegetative cells of the *Subsection IV* cyanobacteria may differentiate into akinetes (resting cells resistant to environmental stress) in response to local conditions (Tomitani et al., 2006). Since intertidal systems represent extreme environments where high temperature, desiccation, high levels of UV-radiation and increased wave action place significant stress on local communities, it remains uncertain which cell type dominates in the *Nostocales* from these assemblages, and how this might translate into potential functions. Having taxa capable of coping with these conditions is important for community resilience. This resilience might be decreased on seawalls, where this taxa was found in lower abundances.

# Ecological Effects of Seawalls

Artificial structures result in the introduction of substrates that are often alien to natural conditions or that differ, for example, with respect to composition, age, orientation, and material (Glasby, 1999c, 2000, 2001). The uniformity of construction has been shown to result in homogeneity across terrestrial landscapes (McKinney, 2006), although less is known in marine systems. We found little evidence that seawalls supported more homogeneous communities than natural rocky shores. Measures of alpha and beta diversity were similar among habitats, although there was significant variation in species richness, diversity, and evenness among locations. This possibility of biotic homogenisation in the marine environment should be further investigated at higher trophic levels by examining patterns in the macroinvertebrates recruiting to seawalls and other urban structures. The evaluation of macroinvertebrate homogenization is particularly relevant to biofilm homogeneity due to invertebrate grazers that regulates biofilm structure and function as would be discussed below.

Due to the size of many artificial structures, their construction design and orientation, they often have different illumination levels from shading (Glasby, 1999b) and UV exposure. Artificial structures typically have vertical surfaces or horizontal surfaces facing downward and few have a surface that is analogous to horizontal rocky reef. Differences in light between rocky shores and seawalls have been observed in previous studies (Blockley, 2007) with shading by vertical seawalls found to reduce light levels at certain times of the day. Given the abundance of autotrophic bacteria identified in this study, shading by seawalls and the associated reduction in light availability is more likely to have affected biofilms than physico-chemical measures such as temperature and salinity that did not differ among habitats.

The occurrence of several groups of extremophiles in the current study may be linked to the stressful environments represented by intertidal hard substrates (Thompson et al., 2002). The bacterial group *Bacilli* has several representatives that require high temperatures for growth or can survive a range of high and low temperatures (Rothschild and Mancinelli, 2001). In the current study we found *Bacilli* to be relatively more abundant on rocky shores than seawalls and this may reflect increased tolerance to high temperatures and UV radiation on horizontal shores (Thompson et al., 2002). However, the slope of substrates may mitigate these stressors and also affect the relative abundance of bacteria comprising biofilms. Tidal submersion is longer on gently sloping rocky shores than vertical seawalls and may help reduce exposure to stressful environmental conditions (Bulleri and Chapman, 2010) that lead to desiccation. Differences between patterns observed in the microbial community structure may have been stronger had the full range of photosynthetic organisms including diatoms and microalgae been investigated with the bacteria (Consalvey et al., 2005).

Artificial structures such as seawalls constructed in marine environments create islands of hard-substrate (Airoldi and Beck, 2007) and these structures can change local hydrodynamic conditions which may alter the rate at which nutrients and organic material are delivered or entrained (Glasby, 1999a). While these environmental factors were not measured in the current study, the dominance of the *Proteobacteria* groups in all assemblages suggests that nutrient availability could be an important factor influencing the differences in relative abundances of biofilm taxa. Thus, the presence of *Proteobacteria* groups in lower abundances on seawalls suggests that nutrient cycling may be altered. Increasingly, studies seek to measure rates of biogeochemical processes in relation to artificial structures. Indirect effects on nitrogen gas production derived from the facilitation of invasive macroalgae by coastal defense structures have been found (Geraldi et al., 2014). Future studies might consider quantifying the potential for urban structures to support biofilm communities that perform important ecosystem functions.

Artificial structures may affect microbial communities indirectly via the benthic assemblages that such structures support, composed of algae, invertebrates, and fish. Past research has also shown that several species of mobile grazers commonly found on rocky shores are rare or absent from seawalls (Chapman, 2006). Differences in grazer assemblages could have an effect on the relative abundance of bacterial taxa on natural and artificial structures observed in the current study (Skov et al., 2010). Grazing allows direct removal of biofilm as well as the potential bloom in photo-autotrophic biomass when greater light and nutrient penetration are enabled through the removal of biofilm canopy (Skov et al., 2010). Consequently, grazing is an important driver in the structure and function of biofilms (Burns and Ryder, 2001; Thompson et al., 2004) and *Cyanobacteria* have previously been found to be exploited by populations of macrograzers such as echinoderms, polyplacophorans and gastropods (Hutchinson et al., 2006; Liess and Kahlert, 2009). Therefore the differences in relative abundances of *Cyanobacteria* may be related to differential grazing pressure between seawalls and rocky shores. Grazers were not quantified in the current study, but would provide useful information about changes in trophic dynamics related to marine urban seascapes.

# CONCLUSION

This study found that while the identity of communities in rocky shore and seawall habitats are similar, relative abundances of bacterial taxa differed. These differences were not due to slope, ecological history or material of the structure under study, but are probably a consequence of surface complexity, benthic assemblages and/or environmental variables. As the requirement for coastal defense structures is expected to increase in the coming years with increasing urbanization of the coastal zone, this study forms a crucial baseline of the ecological consequences of urban seascapes at the microbial scale. Thus, investigation of microbial communities on other widespread artificial structures such as pilings and pontoons should also be conducted. Future research of microbial communities in urban seascapes might consider an approach that includes targeted sequencing to investigate the eukaryotic component of the biofilm. Metatranscriptomics or measurements of processes such as productivity, respiration, and nitrogen cycling could also be used to assess functional changes in bacterial communities in response to marine urbanization. Additionally, it would be worthwhile to study the differences in microbial communities between natural and artificial substrates alongside factors that regulate microbial community assemblages such as grazing, flow movements, and light.

# REFERENCES


# ACKNOWLEDGMENTS

This research was funded by an ARC Linkage Grant awarded to KD and EJ. Fieldwork was conducted under NSW Department of Primary Industries permit number P13/0007-1.0. The authors would like to thank Melanie Sun and Simone Birrer for their lab support and to Michael Sutherland, Jamie-Louise Morrison, volunteers and members of the Applied Marine and Estuarine Ecology laboratory for their field assistance.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal*.*frontiersin*.*org/article/10*.*3389/fmicb*.* 2015*.*01276


community associated with a green marine Ulvacean alga. *ISME J.* 4, 301–311. doi: 10.1038/ismej.2009.107


**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 Tan, Mayer-Pinto, Johnston and Dafforn. 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.*

# **Innovative biological approaches for monitoring and improving water quality**

*Sanja Aracic <sup>1</sup> , Sam Manna <sup>1</sup> , Steve Petrovski <sup>1</sup> , Jennifer L. Wiltshire <sup>1</sup> , Gülay Mann <sup>2</sup> and Ashley E. Franks <sup>1</sup> \**

*<sup>1</sup> Applied and Environmental Microbiology Laboratory, Department of Physiology, Anatomy and Microbiology, La Trobe University, Melbourne, VIC, Australia, <sup>2</sup> Land Division, Defence Science and Technology Organisation, Melbourne, VIC, Australia*

Water quality is largely influenced by the abundance and diversity of indigenous

### *Edited by:*

*Maurizio Labbate, University of Technology, Sydney, Australia*

### *Reviewed by:*

*Steven Ripp, University of Tennessee, USA George S. Bullerjahn, Bowling Green State University, USA*

### *\*Correspondence:*

*Ashley E. Franks, Applied and Environmental Microbiology Laboratory, Department of Physiology, Anatomy and Microbiology, La Trobe University, Plenty Road, Melbourne, VIC 3086, Australia a.franks@latrobe.edu.au*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 25 May 2015 Accepted: 27 July 2015 Published: 12 August 2015*

### *Citation:*

*Aracic S, Manna S, Petrovski S, Wiltshire JL, Mann G and Franks AE (2015) Innovative biological approaches for monitoring and improving water quality. Front. Microbiol. 6:826. doi: 10.3389/fmicb.2015.00826* microbes present within an aquatic environment. Physical, chemical and biological contaminants from anthropogenic activities can accumulate in aquatic systems causing detrimental ecological consequences. Approaches exploiting microbial processes are now being utilized for the detection, and removal or reduction of contaminants. Contaminants can be identified and quantified *in situ* using microbial whole-cell biosensors, negating the need for water samples to be tested off-site. Similarly, the innate biodegradative processes can be enhanced through manipulation of the composition and/or function of the indigenous microbial communities present within the contaminated environments. Biological contaminants, such as detrimental/pathogenic bacteria, can be specifically targeted and reduced in number using bacteriophages. This mini-review discusses the potential application of whole-cell microbial biosensors for the detection of contaminants, the exploitation of microbial biodegradative processes for environmental restoration and the manipulation of microbial communities using phages.

**Keywords: synthetic biology, bioremediation, electromicrobiology, phage-therapy, water quality**

# **Introduction**

Anthropogenic activities, such as manufacturing, mining and farming, have led to an increase of a wide range of chemical (organic and inorganic) and biological (e.g., bacteria, yeast, fungi, and algae) contaminants in aquatic environments (Carpenter et al., 2011). Contamination of lakes, rivers, oceans, reservoirs, and groundwater affects not only the organisms living within these bodies of water but can also impact the entire biosphere. A major threat to aquatic ecosystems worldwide is eutrophication, the over-enrichment of water with nutrients or organic matter (Woodward et al., 2012). High concentrations of nitrogenous compounds and phosphates result in the formation of algal blooms which negatively impact water quality and ecosystems (Grizzetti et al., 2012). Inorganic contaminants, in particular heavy metals, are also a prominent environmental concern because they are not biodegradable and can accumulate in living organisms (Fu and Wang, 2011). The toxicity of heavy metals depends not only on the particular element but also its chemical speciation and oxidation state. The threat posed to higher organisms as a consequence of contaminated water can be reduced by microbial degradation of organics and re-speciation of heavy metals into less toxic forms (Kot and Namiesńik, 2000).

The current approaches for removal of contaminants include sedimentation, membrane filtration, coagulation-flocculation, adsorption, chemical precipitation and ion-exchange (Otte and Jacob, 2006; Fu and Wang, 2011). These processes depend on existing infrastructure and can be impractical to implement in developing countries and remote areas. As a result, there is an increasing need to detect and treat contaminated water *in situ* using sustainable approaches that are inexpensive and environmentally-friendly. Researchers are currently investigating the use of native biota for the detection and degradation or reduction of contaminants as a cheaper and sustainable alternative. This mini-review addresses the potential application of biological approaches, and their limitations, as complementary methods to chemical techniques for the detection of contaminants and treatment of contaminated water.

# **Monitoring of Water Quality**

### **Detection of Contaminants Using Naturally-existing Whole-cell Microbial Biosensors**

Environmental and microbiological research has driven the growing interest in real-time monitoring of water quality using whole-cell microbial biosensors. Yeast, algae and bacterial wholecell biosensors have been applied to domestic wastewater and natural waters to detect phenols, non-ionic surfactants, pesticides, heavy metals and effluents from the chemical industry (Girotti et al., 2008). Microbial whole-cell biosensors produce a measurable signal enabling detection and quantification of contaminants (Lagarde and Jaffrezic-Renault, 2011). Growth characteristics, enzymatic activity or other measureable outputs can be monitored in response to the presence of specific contaminants. Given their ubiquity in aquatic systems algae have been utilized as bioreporters that are capable of detecting contaminants and nutrient fluxes in water. The abundance of specific benthic algae (16 of 21 species tested) directly correlated with the total phosphorus present, providing information on *in situ* levels (Rott and Schneider, 2014). The morphological responses of cyanobacteria to specific nutrients also provides information on nutrient levels, in the absence of nitrogen these organisms form an abundance of nitrogen-fixing heterocysts, whereas in the absence of phosphorus they produce elongated filaments (Whitton and Potts, 2007).

Some microorganisms possess innate characteristics, such as luminescence or the ability to generate electrical current, which can be utilized to measure metabolic response to environmental contaminants (**Figure 1A**; Daunert et al., 2000). Luminescence produced by the marine bacterium *Vibrio fischeri* has been exploited for the detection of phenols in water (Stolper et al., 2008). The presence of phenols in water results in a quantifiable reduction of luminescence of the microorganism (a 90% reduction of luminescence was observed in the presence of 100 mg L *−*1 3,4-dichlorophenol) (Stolper et al., 2008). While *Vibrio* uses luminescence as a phenotypic indicator, there are other naturally existing biosensors that use non-luminescent based strategies to report the presence or absence of contaminants in water.

Current production in microbial fuel cell systems, a measure of electron flow from central metabolism, is a direct measure of metabolic activity and can be used to monitor changes in metabolic activity over time (Aracic et al., 2014). This approach has been utilized for *in situ* monitoring of the metabolic activity of complex microbial communities in a variety of subsurface anoxic environments (Williams et al., 2009). The indigenous microbial population may utilize many contaminants as electron donors and cause an increase in microbial metabolism. Wastewater contains a large volume of organic compounds that can stimulate microbial growth. Metabolic activity, as measured as current production in microbial fuel cells, has been shown to be capable of providing real-time monitoring of organic effluents in relation to their chemical oxygen demand (Di Lorenzo et al., 2009). Likewise, a simple anode-resistor-cathode device can monitor *in situ* rates of anoxic subsurface microbial activity providing continuous metabolic rates in response to the presence of organic contaminants (Wardman et al., 2014). The integration of nanomaterials (gold particles, magnetic beads and carbon nanotubes) as well as electron mediators in electrochemical biosensors has resulted in improved limits of detection of numerous water pollutants (Lagarde and Jaffrezic-Renault, 2011). Immobilization of electroactive *Pseudomonas putida* cells on electrodes using carbon nanotubes resulted in an 80-fold increase in sensitivity and 2.8-fold increase in response time to trichloroethylene (Hnaien et al., 2011). In more recent years biosensor research has moved from naturally existing wholecell biosensors to synthetically-derived microbial biosensors to optimize the detection of contaminants. Advances in synthetic biology have allowed the stability, specificity and sensitivity of whole-cell biosensors to be improved.

### **Detection of Contaminants Using Synthetically-derived Microbial Biosensors**

Synthetic biology is now allowing the systematic design of whole-cell biosensors. Typically, a reporter gene is placed under the control of a promoter that is transcriptionally active in the presence of a specific contaminant (**Figure 1B**). Numerous regulatory elements (promoters and their cognate transcriptional regulators) have been identified which respond to specific organic contaminants and heavy metals found in contaminated water such as arsenic, cadmium and mercury (Bereza-Malcolm et al., 2014). The regulatory elements control the transcription of reporter genes whose expression produces a detectable and quantifiable fluorescent, luminescent or electrochemical signal. Syntheticallyderived microbial biosensors are often created using common laboratory strains of *Escherichia coli*. While these systems are functional in *E. coli*, a significant challenge is encountered in the real-world application of these biosensors for the detection of contaminants in aquatic settings. This is because *E. coli* lacks many of the physiological characteristics that are required for its survival and proliferation in these niche environments. As a consequence, biosensors are being developed using microbes that inhabit the aquatic environment of interest. Cyanobacteria, which inhabit marine and freshwater environments, have been engineered to detect and provide a measurable signal in response to a range of contaminants. Genetically-modified cyanobacteria

have been used as*in situ* bioindicators for excess nutrients, organic contaminants and heavy metals in water (Gillor et al., 2010; Mateo et al., 2015). An extensive range of strains have been engineered as general ecotoxicity, nitrogen, ammonium, nitrate, nitrite, phosphorus and heavy metals reporters using luciferase in a light-on or lights off response (see Mateo et al., 2015, for a recent extensive review).

The use of electroactive microorganisms as whole-cell biosensors allows the integration of microbial outputs into electrochemical signals. The ability of these organisms to transfer electrons to and from electrodes has been extensively studied for numerous biotechnological processes including their potential use in autonomous sensing devices (Rosenbaum and Franks, 2014). A *Shewanella oneidensis* biosensor has been constructed using outer membrane cytochrome complexes capable of electron transfer under the control of a promoter responsive to arabinose. The concentration-dependent detection of arabinose was linked to the expression of cytochromes involved in direct extracellular electron transfer (Golitsch et al., 2013). Proteins involved in electron transfer (flavins, shuttles, and cytochromes) produce specific electrochemical signals detectable through electrochemical techniques (Golitsch et al., 2013). The required instrumentation for electrochemical detection can be miniaturized, providing portable analytical devices for simple *in situ* measurements (Badihi-Mossberg et al., 2007).

Whole-cell biosensors can detect contaminants with sensitivities comparable to those of conventional chemical methods and/or optical bioassays (Lagarde and Jaffrezic-Renault, 2011). Therefore they could serve as complementary techniques to conventional chemical methods currently used to monitor levels of contaminants by the Environmental Protection Agency and World Health Organization. Furthermore, microbial biosensors measure the bioavailability of the contaminants without the need for further *in vitro* bioassays. Nevertheless, several limitations still remain in the application of biosensors including the variability and complexity of flowing water, the wide range of concentrations of contaminants, as well as the sensitivity, specificity and robustness of the biosensors.

## **Improving Water Quality**

### **Enhanced Removal of Organic Contaminants and Heavy Metals Via Exploitation of Macrophyte-associated Microbial Communities**

Excessive nitrogen and phosphorus in water causes proliferation of algae resulting in reduced oxygen availability, food resources and habitats that fish and other aquatic life need to survive. Algal blooms are also harmful because they produce elevated toxins and can result in increased bacterial growth. The reversal of eutrophication (oligotrophication) and whether reduction of nitrogen, phosphorus or both is the most cost-effective approach to improve water quality has been extensively reviewed (Grizzetti et al., 2012; Schindler, 2012; Woodward et al., 2012). Nevertheless, specific processes are known to be crucial for the removal of these elements. Nitrification, performed mainly by bacteria from the *Nitrobacteraceae* family, is a process which reduces nitrogenous compounds from the environment (Hagopian and Riley, 1998). Under anoxic conditions the efficiency of nitrification is restricted (Yoo et al., 1999). This rate-limiting step can be enhanced fourfold via the use of vascular plants that live in or near water (macrophytes) which create oxygenated microenvironments in the rhizosphere and promote nitrification (Soana and Bartoli, 2014). The presence of contaminant-tolerant oxygen-releasing plants can stimulate nitrification by increasing microbial activity

in the rhizosphere of otherwise anoxic sediments (**Figure 2A**). Artificial floating islands (i.e., soil-less structures constructed with floating macrophytes) have been applied in the treatment of stormwater, household effluent and industrial waste (Yeh et al., 2015). Numerous in-field studies have demonstrated that macrophyte-microbe interactions are responsible for the removal of organic matter in a range of polluted water environments (Yeh et al., 2015). For example, a floating bed of perennial grasses removed 60% of the total nitrogen and 56% of the total phosphorus, of which only 2.8 and 4.5%, respectively, were due to plant uptake (Li et al., 2012). The majority of the removal was due to the plant–microbe interactions, which were enhanced by the growing plants. The introduction of plants also resulted in 50% reduction of total petroleum hydrocarbons (from 1700 to 871 mg L *−*1 ). This approach is an environmentally-friendly strategy for removing excess nutrients and organic pollutants.

In addition to degrading excess nutrients and organic pollutants, macrophyte-associated microorganisms also facilitate heavy metal uptake, a crucial step in the phytoremediation process (Jing et al., 2007). This bioremediation strategy for treating contaminated water has an added benefit because harvested macrophytes can be processed into biomaterials such as biogas (Ahalya et al., 2003) and animal feed (Li et al., 2012). The protein and fiber content of macrophytes meet the national feed thresholds and the content of toxic heavy metal ions is below the critical levels for animal feeds (Li et al., 2012). The removal of macrophytes impedes nitrification and other microbial processes requiring oxygenated environments, however, if not removed, the decaying macrophytes can also serve as a source of pollution (Xie et al., 2013). An equilibrium between removing heavy metals while promoting the growth of a diverse range of microbes that degrade complex organic compounds is required for the process to be successful on a large scale. Further elucidation of the interplay between the chemical and biological processes that occur between macrophytes and microbes will lead to improved water bioremediation processes. The influence of environmental changes on these interactions as well as biotic and abiotic processes that compete for oxygen and nutrients, limiting their availability to the important microorganisms needs to be further explored.

### **Enhanced Removal of Organic Contaminants and Reduction of Heavy Metals Via Exploitation of Microbial Communities: Biodegradative and Electroactive**

Bioremediation of contaminated water can occur through innate processes performed by microbes, which involve the utilization of contaminants as a nutrient or energy source (Singh et al., 2014). Biodegradative microbes may be indigenous to the site or exogenous species used for bioaugmentation (Dixit et al., 2015). The growth and biodegradative abilities of these microbes can be enhanced through the addition of certain nutrients which promote proliferation of beneficial bacterial species or addition of terminal electron acceptors/donors. The stimulated microbial communities are able to convert organic and inorganic contaminants into non-hazardous or less hazardous forms through oxidation or reduction (Dixit et al., 2015). Often a consortium of microorganisms, including bacteria, yeast and fungi, performs these processes sequentially. The genetic architecture of these microbial communities allows biodegradation, biotransformation, biosorption and bioaccumulation of contaminants to occur in unison. These innate abilities can be further enhanced through genetic modification of regulatory and/or metabolic genes (Dixit et al., 2015). Heavy-metal-tolerant bacteria containing the arsenic(III) *S*-adenosylmethionine methyltransferase (ArsM) are able to methylate toxic inorganic arsenic(III) to the less toxic arsenic(V) (Chen et al., 2014). Arsenic volatilization can be improved ninefold by genetically engineering strains to overexpress ArsM. Although this approach has not yet been applied to treat arsenic contaminated water, the laboratory-scale study conducted by Chen et al. (2014) provides promising data to tackle arsenic contaminated environments.

Metagenomic studies are providing information about the composition and the genetic potential of microbial communities involved in the removal of excess elements and toxic contaminants (Bai et al., 2014). The presence of genes involved in xenobiotic degradation would indicate the presence of these contaminants in the aquatic environment. However, metatranscriptomics or gene expression studies are necessary to confirm the link between the abundance of degradation genes and biodegradation rates. RT-qPCR studies revealed that expression of hydrocarbon degrading genes of *Pseudomonas* and *Rhodococcus* species was 1000-fold higher in contaminated soil environments and were low or undetected in uncontaminated samples (Yergeau et al., 2012). Similarly, Yu and Zhang (2012) analyzed the presence and expression of genes involved in nitrification, denitrification, ammonification, and nitrogen fixation processes using metagenomic and metatranscriptomic analyses of microbial communities in wastewater (Yu and Zhang, 2012). This study revealed that although denitrification genes were most abundant (78.6%) of the four processes and nitrification genes were the least abundant (0.9%), the ratio of cDNA to DNA was highest for the nitrification process (0.18 and only 0.03 for denitrification). Therefore gene expression studies are essential in determining the potential for bioremediation of contaminated aquatic environments.

Electroactive microorganisms can use heavy metals and trace elements as terminal electron acceptors or reduce them through detoxification mechanisms. Dissimilatory metal reducing bacteria, such as species of *Geobacter*, utilize insoluble iron(III) or manganese(IV) as final electron acceptors (Lovley and Phillips, 1988). *Geobacter* spp. can also reduce soluble uranium(VI) to insoluble uranium(IV), thus precipitating the ions out of the water table (Lovley and Phillips, 1992; Anderson et al., 2003; Gregory and Lovley, 2005). Electron donors and acceptors can become limiting factors during degradation in anoxic environments. The innate capacity of these microorganisms to reduce/oxidize heavy metal ions and other contaminants can be improved for optimal removal of chemical contaminants using solid surfaces such as anodes as electron acceptors and cathodes as electron donors (**Figure 2B**; Du et al., 2007; Zhang et al., 2010). The transfer of electrons from microorganisms to the solid surface of an electrode enables the catabolism of chemical contaminants to take place at a significantly greater efficiency. For example the presence of an electrode as an electron acceptor instead of iron(III) oxide increased the speed of toluene oxidation 100-fold in petroleum-contaminated sediments (Zhang et al., 2010). While the catabolism of a wide range of organic substrates found in wastewater has been demonstrated (Daunert et al., 2000; Aulenta et al., 2009), significant challenges remain in the implementation of this system on a large scale. A better understanding of the mechanisms involved in electron transfer between mixed communities and electrode surfaces are needed to optimize the electron transfer *in situ* which would consequently increase the degradation efficiency (Aracic et al., 2014).

### **Controlling Abundance of Biological Contaminants Using Phages**

Phage-therapy is a process that allows the removal of a select subset of microbial populations within a water-associated community (**Figure 2C**). Excessive proliferation of filamentous bacteria in activated sludge systems is a major concern, affecting the crucial separation of the solid and liquid phases in settling tanks (De Los Reyes, 2010; Wanner et al., 2010). These processes are essential for successful wastewater treatment and are referred to as sludge bulking and foaming. Sludge bulking occurs when filamentous bacteria form aggregates (flocs) which can extend and interact with adjacent flocs thus disrupting the settleability of the solid particles. The proliferation of hydrophobic bacteria in aeration tanks consisting of surfactants results in the formation of stable foams (De Los Reyes, 2010; Petrovski et al., 2011a). In the absence of surfactants, hydrophobic cells form a stable scum layer on the top of the aeration tank (Petrovski et al., 2011a). The organisms responsible for stabilizing the foams are mainly the unbranched Gram positive filamentous bacterium "*Candidatus Microthrix parvicella*" and short Gram positive branching filamentous bacteria (the mycolata; Kragelund et al., 2007; Seviour et al., 2008; De Los Reyes, 2010). The latter includes members from the genera *Gordonia*, *Skermania*, *Rhodococcus*, *Nocardia*, *Tsukamurella*, and *Mycobacterium* (Seviour et al., 2008). Although improved culturing practices and molecular methods (such as 16S/23S rRNA targeted probes) have allowed the isolation and identification of numerous filamentous bacteria which are responsible for the formation of the stable foams (Speirs et al., 2009, 2011; Seviour and Nielsen, 2010), there are currently no reliable strategies to control sludge bulking during water treatment.

One attractive approach to preventing or eliminating sludge bulking and foaming caused by biological contaminants in treatment plants is to use phage-therapy (Thomas et al., 2002). The use of lytic phages that target the causative bacteria is specific, relatively inexpensive and environmentally-friendly. Most mycolata-specific phages that have been isolated can prevent foaming in pure cultures under laboratory conditions (Petrovski et al., 2011b,c,d, 2012). However, the isolation of phages for *Gordonia amarae* and *Gordonia defluvii*, two of the major foaming bacteria, has been unsuccessful. It is possible that these organisms have evolved sophisticated mechanisms to prevent successful infection, such as clustered regularly interspaced short palindromic repeats systems, however, this is yet to be elucidated. The phages isolated and described so far possess long noncontractile tails, characteristic of the *Siphoviridae* family, and have a broad host range for several mycolata genera responsible for foaming (Thomas et al., 2002).

Some wastewater treatment methods rely on the use of antimicrobial agents to control proliferation of bacteria however, this is not a reliable strategy due to the elimination of bacteria beneficial to wastewater treatment processes. Phage-therapy can therefore be used as an efficient complementary technique

# **References**


to control the proliferation of recalcitrant bacteria to prevent sludge bulking and foaming (Thomas et al., 2002; Withey et al., 2005). Although phage-therapy is effective in controlling the proliferation of wastewater bacteria in the laboratory, full-scale studies are required prior to the implementation of this approach in an industrial setting. Activated sludge systems in treatment plants are complex engineered ecosystems that support the growth of a wide diversity of microbes, which are poorly understood (Withey et al., 2005). Even less information is available on phage diversity and ecology in wastewater treatment systems. Phage survival rate in activated sludge has not been studied, and the ability of phages to infect their hosts may be inactivated or blocked by chemicals in wastewater. Furthermore, the host may evolve to develop resistance to the phage and a combination of phages will be required to tackle a large proportion of bacteria.

# **Concluding Remarks**

Our understanding of the structure and function of microbial communities in contaminated aquatic environments is allowing the manipulation of their innate biosensing and biodegradative abilities. Currently used physico-chemical methods to remove heavy metals are not effective at low concentrations but could be overcome with the use of biological approaches. The biological approaches described in this mini-review could serve as complementary techniques to chemical treatments and allow real-time measurements for prompt detection of contaminants (**Figure 1**) and treatment of contaminated water (**Figure 2**). The innovative approaches described can be used in conjunction with existing technologies to improve biosensing and biodegradation processes. In addition, the implementation of these approaches as an autonomous platform could allow for continuous *in situ* monitoring of contaminants at remote locations and could overcome some of the limitations of the current processes. Phagetherapy could be used to target pathogenic/detrimental bacterial species, thus serving as a cost-effective and environmentallyfriendly approach to treatment of biologically-contaminated water.

# **Acknowledgments**

The Applied and Environmental Microbiology Laboratory receives support from the Defence Science Institute, Defence Science and Technology Organisation, Office of Naval Research Global (Award no. N626909-13-1-N259) and the Australian Research Council (Award no. LP140100459).


denitrification (SND) via nitrite in an intermittently-aerated reactor. *Water Res.* 33, 145–154. doi: 10.1016/S0043-1354(98)00159-6


**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 © 2015 Aracic, Manna, Petrovski, Wiltshire, Mann and Franks. 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.*

# Long-term impacts of disturbance on nitrogen-cycling bacteria in a New England salt marsh

### *Anne E. Bernhard\*, Courtney Dwyer, Adrian Idrizi, Geoffrey Bender and Rachel Zwick*

Biology Department, Connecticut College, New London, CT, USA

### *Edited by:*

Maurizio Labbate, University of Technology Sydney, Australia

### *Reviewed by:*

Nick Bouskill, Lawrence Berkeley National Laboratory, USA Lauren Frances Messer, University of Technology Sydney, Australia

### *\*Correspondence:*

Anne E. Bernhard, Biology Department, Connecticut College, 270 Mohegan Avenue, Box 5327, New London, CT 06320, USA e-mail: aeber@conncoll.edu

Recent studies on the impacts of disturbance on microbial communities indicate communities show differential responses to disturbance, yet our understanding of how different microbial communities may respond to and recover from disturbance is still rudimentary.We investigated impacts of tidal restriction followed by tidal restoration on abundance and diversity of denitrifying bacteria, ammonia-oxidizing bacteria (AOB), and ammoniaoxidizing archaea (AOA) in New England salt marshes by analyzing nirS and bacterial and archaeal amoA genes, respectively. TRFLP analysis of nirS and betaproteobacterial amoA genes revealed significant differences between restored and undisturbed marshes, with the greatest differences detected in deeper sediments. Additionally, community patterns indicated a potential recovery trajectory for denitrifiers. Analysis of archaeal amoA genes, however, revealed no differences in community composition between restored and undisturbed marshes, but we detected significantly higher gene abundance in deeper sediment at restored sites. Abundances of nirS and betaproteobacterial amoA genes were also significantly greater in deeper sediments at restored sites. Porewater ammonium was significantly higher at depth in restored sediments compared to undisturbed sediments, suggesting a possible mechanism driving some of the community differences. Our results suggest that impacts of disturbance on denitrifying and ammonia-oxidizing communities remain nearly 30 years after restoration, potentially impacting nitrogen-cycling processes in the marsh. We also present data suggesting that sampling deeper in sediments may be critical for detecting disturbance effects in coastal sediments.

**Keywords:** *amo***A, disturbance,** *nir***S, restoration, salt marsh**

### **INTRODUCTION**

Nitrification, the oxidation of ammonia to nitrate, and denitrification, the reduction of oxidized nitrogen to dinitrogen gas, play critical roles in determining the availability of nitrogen in salt marshes. Nitrogen is arguably one of the most important nutrients regulating the high productivity reported in many salt marshes (Vitousek et al., 2002), and its availability is controlled primarily by the activity of microorganisms. Additionally, nitrification and denitrification are often coupled (Risgaard-Petersen, 2003), and limited evidence from studies of nitrifiers and denitrifiers suggests that process rates are linked to diversity and community composition of the microbes responsible (Cavigelli and Robertson, 2000; Horz et al., 2004; Webster et al., 2005; Bernhard et al., 2007). Despite their importance to ecosystem productivity, our understanding of what regulates nitrifying and denitrifying communities, how they interact with each other, and how they may respond to ecosystem perturbations remains relatively enigmatic.

New England salt marshes have undergone extensive disturbances in the last century, including nutrient loading, ditching to control mosquitoes, and other modifications to normal tidal flow for recreational and commercial purposes (Gedan et al., 2009). In the last few decades, as we have realized the important roles that salt marshes play in maintaining healthy coastlines, extensive efforts have been underway to restore these vital habitats to their original state. Much of the research on the recovery of salt marshes has focused on changes in macrobiota and biogeochemical processes (see reviews by Warren et al., 2002; Callaway, 2005), with potential impacts on the microbiota only recently coming under scrutiny.

Restricting access of tidal waters to marshes, commonly by diking or impounding, obviously reduces salinity, but also leads to other changes in hydrology and biogeochemical processes important to salt marsh ecology. As the marsh freshens, the water table is lowered, resulting in higher oxygen levels and increased aerobic decomposition, leading to sediment subsidence (Portnoy, 1999). Reduction in salinity also means a reduction in the supply of sulfate, leading to a switch from sulfate reduction to methanogenesis as the primary pathway of anaerobic decomposition (van Proosdij et al., 2010). Because methanogenesis is less energetically favorable, decomposition decreases, as does mineralization of N and P, leaving sediments with higher organic matter. Consequently, sediment chemistry has been shown to differ significantly in impounded marshes compared to unrestricted ones (Portnoy, 1999). When saltwater is restored to the marsh, sulfides increase and oxygen decreases. Depending on the hydrology of the marsh, sediments may also experience changes in pH and availability of inorganic nutrients (Portnoy and Giblin, 1997). Changes in grain size and porosity of sediment due to altered sedimentation patterns may also impact retention of N (Zedler and Kercher, 2005). Changes in sediment chemistry are expected to have significant impacts on the sediment microbial communities that drive much of the biogeochemical processes.

To date, a handful of studies have investigated the impacts of disturbance on salt marsh microbes and, surprisingly, have yielded few significant impacts. In response to nutrient manipulations, minimal impacts on bacterial communities have been reported (Lovell et al., 2001; Bowen et al., 2009). In another study, Bowen et al. (2011) showed no significant differences in the total bacterial communities or the denitrifying communities in two marshes after acute and chronic fertilization. In a study of nitrifying microbes in one of the same marshes sampled by Bowen et al. (2011), significant impacts of fertilization were detected on ammonia oxidizing bacteria, but not on archaea (Peng et al., 2013). And, Bernhard et al. (2012)reported no differences in microbial community composition, but greater variability, in impounded and subsequently restored marshes compared to undisturbed marshes. The minimal impacts of disturbance on salt marsh microbial communities are somewhat surprising given that others have reported significant impacts in other habitats (Downing and Leibold, 2010; Shade et al., 2011; Berga et al., 2012), and that in some cases, significant differences were reported for biogeochemical processes in the marsh (e.g., Hamersley and Howes, 2005). The limited data from salt marsh studies suggest that microbial communities may exhibit different levels of resilience or resistance (as defined in Allison and Martiny, 2008) to perturbations or that we have not sampled adequately, and that our understanding of how disturbance theory applies to microbial communities is still rudimentary (see Shade et al., 2012).

In this study, we investigated the long-term impacts of impoundment (tidal restriction) and subsequent tidal flow restoration (considered press, or chronic, disturbances) on ammonia-oxidizing bacteria (AOB), ammonia-oxidizing archaea (AOA), and denitrifying bacteria in Connecticut salt marshes. We chose nitrifiers and denitrifiers as the targets partly because of differential responses to fertilization these groups have shown in previous studies (Bowen et al., 2011; Peng et al., 2013). Furthermore, a recent study in the same Connecticut marshes (Bernhard et al., 2012) showed greater spatial and temporal variability of bacterial communities in restored marshes, but whether such variability might impact ecosystem services, such as nutrient cycling, in the marshes remains uncertain. Our goals in this study were to determine if the disturbances to the marshes (impoundment and subsequent tidal flow restoration) have left lasting effects on microbial communities involved in two critical nitrogen-cycling processes, and if the response patterns of the functional groups are similar, suggesting a broader-scale impact to nitrogen-cycling in the marsh.

### **MATERIALS AND METHODS**

### **SITE DESCRIPTION**

Samples were collected from the Wequetequock-Pawcatuck (known as Barn Island) and Cottrell salt marshes in southeastern Connecticut. Complete site descriptions and the full management history of the sites have been previously described (Warren and Neiring, 1993; Bernhard et al., 2012). Briefly, four marshes within

the Barn Island system were impounded (hereafter referred to as Impoundments 1–4) in the late 1940s. Starting in 1978, tidal flow was restored to the four marshes over a period of 13 years, with Impoundments 1 and 2 restored first, followed by Impoundment 4 in 1987, and finally, Impoundment 3 in 1991. Two undisturbed marshes in Barn Island (Wequetequock Cove and Headquarters) and two additional sites in the nearby Cottrell marsh were selected as reference marshes for comparison. Dominant vegetation at all sites was *Spartina patens*.

### **SAMPLE COLLECTION**

Triplicate sediment cores (6.5 cm diameter) were collected from each of the eight sites in July 2006 and sectioned into 0–2 cm and 6–8 cm horizons. Each horizon was homogenized and aliquoted for DNA extraction, porewater analyses (salinity, pH, % water, and NH4 +), and dry wt determination. Methods for DNA extraction, porewater analyses, and dry wt determination have been previously published (Nelson et al., 2009; Bernhard et al., 2012).

### **QUANTITATIVE PCR**

*nir*S genes were quantified by real-time PCR using the primers *nir*S-1F and *nir*S-3R (Braker et al., 1998). All 20 μl reactions were run in an iCycler (BioRad) with ca. 5–10 ng DNA, SYBR Green I super mix (BioRad), 0.5 μM of each primer, 0.008% bovine serum albumin using the following conditions for 40 cycles: 95◦C for 15 s, 54◦C for 20 s, 72◦C for 30 s. To monitor product specificity, we conducted melt curve analysis (95◦C for 1 min, 54◦C for 1 min, and then 0.5◦C increase every 10 s, with fluorescence read continuously) after each run. Gene abundances were estimated by comparison to known concentrations of a plasmid containing a cloned *nir*S gene. Concentrations of the plasmid ranged from 1 pg to 1 fg. Bacterial 16S rRNA genes and *amo*A genes were quantified as previously described (Moin et al., 2009; Bernhard et al., 2012; Peng et al., 2013). PCR efficiencies were 94.3 ± 0.14% (*nir*S), 93.4 ± 4.7% (AOB), 99.6 ± 10.0% (AOA), and 89.2% (bacterial 16S rRNA).

### **COMMUNITY FINGERPRINTS**

The *nir*S gene was amplified from all samples in triplicate using the primers *nir*S-1F and *nir*S-6R (Braker et al., 1998). The *nir*S-1F primer was labeled at the 5 -end with 6-FAM. PCRs were run with the following cycle conditions: 30 cycles of 95◦C for 15 s, 54◦C for 30 s, and 72◦C for 90 s, followed by a final elongation of 5 min at 72◦C. PCR products were confirmed by comparison to a DNA molecular weight ladder by electrophoresis analysis in a 1% agarose gel. Positive PCR products were then digested with 10 units of *Hha*I overnight at 37◦C followed by by ethanol precipitation. The restriction endonuclease *Hha*I was found to provide the greatest discrimination based on *in silico* analysis of *nir*S sequences (data not shown). An additional advantage of using *Hha*I, is that it produces a 3 overhang, and therefore only yields the original terminal restriction fragment (TRF) without producing artifacts due to residual polymerase activity (Hartman et al., 2007).

Betaproteobacterial and archaeal *amo*A genes were processed for Terminal restriction fragment length polymorphism (TRFLP) analysis as previously described (Bernhard et al., 2005; Peng et al., 2013). Digests of *amo*A and *nir*S genes were resuspended in 5 μl of deionized H2O, 0.2 μl of the internal size standard, GS500-ROX (Applied Biosystems Inc., Fremont, CA, USA), and 10 μl of Hi-Di Formamide (ABI) and sent to the Biotechnology Resource Center at Cornell University (http://cores.lifesciences.cornell.edu/brcinfo/) for analysis on an Applied BioSystems 3730xl DNA Analyzer.

Terminal restriction fragment sizes and relative abundances for both *amo*A and *nir*S genes were estimated using GeneMarker software, version 1.4 (SoftGenetics, State College, PA, USA). Since betaprotebacterial and archaeal *amo*A diversity in New England salt marshes has been relatively well-characterized (Bernhard et al., 2005; Moin et al., 2009; Peng et al., 2013), we included only TRFs previously identified from published *amo*A sequences in our analysis to minimize the impact of TRFLP artifacts. However, the *nir*S genes from New England salt marshes have not been as well characterized as the *amo*A genes, so we analyzed two different data sets for *nir*S. The first data set included all detectable TRFs in the analysis, but likely included artifacts due to chimera or heteroduplex formation. The second data set included only the TRFs that were confirmed by sequence analysis from our samples or from publically available sequences. Because the results were similar between the two data sets based on multivariate analyses (data not shown), we present the results from the second data set that included only TRFs that were represented by a *nir*S sequence. We acknowledge that this data set may not include all *nir*S TRFs since the *nir*S sequence database for salt marshes is likely incomplete, but we chose to focus on the more conservative approach to avoid including potential artifacts.

### **STATISTICAL ANALYSES**

Terminal restriction fragment length polymorphism profiles were compared using PC-Ord version 6 (McCune and Mefford, 1999). The relative abundance data were transformed by an arcsine square root function to reduce skew. Non-metric multidimensional scaling (NMS; Kruskal, 1964) was used to ordinate samples in gene fragment space, using the SØrenson's distance measure. The autopilot option was set to the slow and thorough level for all ordinations. Monte Carlo tests were run to confirm that results obtained were significantly better than would be obtained from randomized data. Additionally, the proportion of variance explained by each axis and the cumulative variance explained was determined by calculating the coefficient of determination between distances in ordination space and distances in the original *p*-dimensional space. Correlation coefficients in the ordination

space were determined for environmental variables and TRFs by rotating the ordination to maximize the coefficient on one axis (Varimax rotation) in order to facilitate detecting clusters of samples (McCune and Grace, 2002).

Multi-response permutation procedure (MRPP), a nonparametric test, was used to test for differences between restored and undisturbed sites and among different sites. MRPP is a variant of analysis of similarity and provides a measure of the effect and *p*-value when testing for differences between two or more groups defined by the user (McCune and Grace, 2002).

Differences in relative abundance of TRFs and abundance of genes between restored and undisturbed marshes were detected by Student's *t*-tests on arcsin square root transformed data using InStat 3.0b (GraphPad Software, Inc.). In some cases, data transformation was not sufficient to meet the assumptions of the *t*-test, so we applied the nonparametric Mann–Whitney test. Statistical significance for all analyses was set at α = 0.05.

### **SEQUENCE ANALYSIS OF** *nir***S GENES**

Since betaproteobacterial and archaeal *amo*A genes have been previously characterized in the marshes studied here (Moin et al., 2009), we focused our sequencing efforts only on the *nir*S genes. Our intention was not to fully describe the phylogeny of *nir*S genes, but rather to identify TRFs of the most frequently detected *nir*S populations to include in community analysis.

The *nir*S gene was amplified from DNA from 0 to 2 cm and 6 to 8 cm horizons from cores collected from Wequetequock Cove and Impoundments 1 and 4 using the primers *nir*S1F and *nir*S6R (Braker et al., 1998). PCR products were cloned into the pSC vector using the StrataClone PCR Cloning kit (Stratagene, Agilient Technologies, Santa Clara, CA, USA) following the manufacturer's recommendations. Thirty-three clones from each of the six libraries were screened with the vector-specific primers M13F and M13R. Clones containing the correct size insert were sequenced by High Throughput Sequencing Solutions (University of Washington, Department of Genome Sciences, Seattle, WA, USA) using the vector-specific primers T3 and T7. Nucleotide sequences for *nir*S have been deposited in Genbank under the accession numbers KF895915-KF896071.

### **RESULTS**

### **POREWATER ANALYSIS**

Analysis of porewater chemistry revealed some significant differences in conditions between restored and undisturbed sites and between surface and deep sediments (**Table 1**). Both pH and

**Table 1 | Mean (±SE) salinity, pH, ammonium, and % water in porewater of sediments from 0 to 2 cm and 6 to 8 cm from samples collected in restored and undisturbed marshes.**


Different letters after the values indicate significantly different values (P < 0.05). No pH data were collected for sediments from 6 to 8 cm.

ammonium were significantly higher in restored sites compared to undisturbed sites, but only in the deeper sediment for ammonium and surface sediments for pH (pH data for deep sediments was not available). We also detected significant differences in salinity between surface and deeper sediment, but only at the restored sites. We did not measure nitrate concentrations from the samples in this study, but previous measurements from other sampling dates at the same sites do not indicate significant differences in nitrate concentrations between restored and undisturbed sites (Bernhard, unpublished).

### **GENE ABUNDANCE PATTERNS**

There was a consistent abundance pattern in relation to sediment depth and restoration status for all three genes with lowest abundances found in undisturbed deep sediment (**Figure 1**). For all three genes, abundance was significantly higher in restored sites compared to undisturbed sites in deep sediment only, and similar patterns were found when the data were normalized to bacterial 16S rRNA gene abundance (**Figure 1**).

Abundance of *nir*S genes ranged from 9.8 <sup>×</sup> <sup>10</sup><sup>6</sup> to 2.1 <sup>×</sup> 109 copies/gdw at the surface and 6.2 <sup>×</sup> <sup>10</sup><sup>7</sup> to 4.6 <sup>×</sup> <sup>10</sup><sup>9</sup> at depth, and represented as much as 9% of the total bacterial community (based on 16S rRNA gene abundance) at the surface and 36% at depth. Archaeal *amo*A genes were about one order of magnitude lower in abundance compared to *nir*S genes, and ranged from a low of 8.9 <sup>×</sup> 105 in deeper sediment to 1.2 <sup>×</sup> 109 at the surface. Betaproteobacterial *amo*A genes showed similar patterns, but were, on average, an order of magnitude lower than archaeal *amo*A genes, ranging from a low of 6.8 <sup>×</sup> <sup>10</sup><sup>4</sup> (6−8 cm) to 1.2 <sup>×</sup> <sup>10</sup><sup>8</sup> (0−2 cm). AOB and AOA comprised up to 1.6 and 4.8%, respectively, of the total bacterial community at the surface and as much as 0.6 and 1% at depth.

Abundance of *nir*S genes was positively correlated with porewater pH (Pearson's correlation coefficient, *r* = 0.44, *P* = 0.03) and water content (*r* = 0.54, *P* = 0.007) at the restored sites, but not at the undisturbed sites. We did not detect any significant correlations between betaproteobacterial *amo*A abundance and porewater salinity, pH, ammonium, or water content.

Abundance of archaeal *amo*A genes was significantly negatively correlated with pH (*r* = −0.53, *P* = 0.0083) at the surface when restored and undisturbed sites were combined.

### **COMMUNITY COMPOSITION**

Ordination analysis of TRFLP profiles for denitrifiers, AOA, and AOB from marshes in southeastern Connecticut indicated different responses to disturbance among the functional groups. In all cases, over 80% of the variability was explained by the first two axes and the final stress of the ordinations suggest a low risk of drawing false inferences (McCune and Grace, 2002). Initially, we analyzed all samples combined to identify significant patterns of community composition for each gene. Using depth as the grouping variable, MRPP analysis showed significantly different communities in surface sediments compared to deeper sediments for all three genes (**Table 2**). However, we identified significant differences among communities only for *nir*S and AOB when restoration status was used as the grouping variable (both depths combined).

Further analysis of *nir*S TRFLP profiles from restored and undisturbed marshes at each depth revealed restoration effects in both sediment depths, but the effects were more striking in the deeper sediment (**Figure 2**). MRPP analysis confirmed that the denitrifier communities in restored sites were significantly different from those in undisturbed sites (**Table 2**). In surface sediments, community patterns of *nir*S genes were not distinguishable among the four impoundments, but we detected a significant site

restored and undisturbed sites are indicated by different letters.


**Table 2 | Results from multiresponse permutation procedure (MRPP) based on TRFLP fingerprints of** *nir***S, betaproteobacterial** *amo***A, and archaeal** *amo***A genes.**

Effects of restoration and sediment depth were tested with both depths combined. Effects of restoration only were then tested for each depth separately. Significant

effects (<sup>α</sup> <sup>≤</sup> 0.05) are indicated by P-values in bold. <sup>a</sup>A is the intragroup average distance; when all items are identical within groups, A <sup>=</sup> 1. <sup>b</sup><sup>T</sup> <sup>=</sup> ( <sup>δ</sup>−m)/s <sup>=</sup> (observed–expected)/s. dev. of expected, where m and s are the mean and SD of <sup>δ</sup> under the null hypothesis. <sup>c</sup>P is the probability of a smaller or equal <sup>δ</sup>.

(*P* = 0.005) and marsh (*P* = 0.005) effect among the undisturbed sites from Barn Island and Cottrell marshes.

In the deeper sediments, we observed community patterns of denitrifiers indicating a chronological shift among the restored marshes from Impoundments 3 and 4 (restored in 1991 and 1987, respectively) to Impoundments 1 and 2 (restored in 1978). Impoundments 1 and 2 were more similar to each other and to undisturbed marshes, while Impoundments 3 and 4 were distinct from each other and from the other marshes (**Figure 2B**). Similar to surface sediments, undisturbed sites in Barn Island were significantly different from Cottrell marsh sites (*P* = 0.002), but undisturbed sites within each marsh were not different. When undisturbed sites from each marsh were removed from the analysis, restored sites were still significantly different from the undisturbed Barn Island marsh (*P* < 0.0001) and the Cottrell marsh (*P* < 0.0001) sites.

Analysis of archaeal *amo*A genes by TRFLP revealed few community differences between restored and undisturbed sites, between surface and depth, or between undisturbed Barn Island sites and Cottrell sites (**Figure 3**; **Table 2**). AOB communities, however, at restored and undisturbed marshes in surface sediments were not different, but a significant disturbance effect was detected in deeper sediment (**Figure 4**; **Table 2**). The patterns of AOB communities in deeper sediments was somewhat similar to the patterns observed for *nir*S communities, with communities at Impoundments 1 and 2 grouping together and Impoundments 3 and 4 grouping together (**Figure 4B**), but no significant differences were detected among the undisturbed sites.

Diversity of denitrifier communities based on *nir*S TRFLP profiles revealed significantly higher evenness, but not richness, in surface sediments in restored sites compared to undisturbed sites (**Table 3**). Similar patterns were also observed in the deeper sediment, but the differences were not quite significant. We detected no differences in diversity, however, of AOB or AOA communities between restored and undisturbed sites.

### **PATTERNS OF TRF ABUNDANCE**

Thirteen TRFs representing *nir*S genes were identified from analysis of over 200 *nir*S sequences from clone libraries created from both depths at one undisturbed site (Wequetequock

### **Table 3 | Mean (SE) diversity indices calculated from TRF relative abundance data.**


Numbers in bold indicate significantly different values between restored and undisturbed marshes (P < 0.05).

Cove) and two restored (Impoundments 1 and 4) sites as well as from analysis of closely related published sequences (Figure S1, Table S1). Relative abundance of only one *nir*S TRF in the surface sediments was significantly different between restored and undisturbed marshes, while abundance of nine TRFs in the deeper sediments showed significant patterns related to disturbance (**Figure 5A**). Of the TRFs that showed significant differences between restored and undisturbed marshes, three of them (70, 142, and 277) showed patterns in the deeper sedimentsfrom Impoundments 1–4 that correspond to the chronology of restoration (data not shown). For example, in the deeper sediment, abundance of TRF 70 was significantly greater in undisturbed sites compared to restored sites, and was also significantly greater (*P* = 0.004) in Impoundments 1 and 2 (restored in

1978) compared to Impoundments 3 and 4 (restored 10–12 years later) when abundance from each impoundment was analyzed separately.

Archaeal *amo*A TRFs 170 and 296 were the dominant fragments detected at all sites, each comprising 40–50% of the community (**Figure 5B**). These TRFs correspond to sequences affiliated with *Nitrosopumilus* group I (TRF170) and *Nitrosopumilus* group 2 (TRF296), as reported in Peng et al. (2013). There were no significant differences in relative abundance of individual TRFs between restored and undisturbed sites. However, TRF 119 was significantly higher (*P* = 0.016) in deeper sediments compared to surface sediments at restored sites, but no differences were detected at undisturbed sites.

Betaproteobacterial *amo*A TRFs 127 and 130 were the dominant TRFs at both restored and undisturbed sites (**Figure 5C**) and have been shown in previous studies to correspond primarily to *Nitrosospira*-like *amo*A sequences (Bernhard et al., 2005; Peng et al., 2013). TRF 127 also represents a small number of sequences closely related to *Nitrosomonas* sp. NM143 (Peng et al., 2013). TRF 127 was significantly greater in relative abundance at restored sites at depth compared to undisturbed sites (*P* = 0.004). Conversely, TRF 98 was significantly greater at undisturbed sites (*P* < 0.0001), comprising nearly 50% of the community, while TRF 98 made up less than 6% of the community at other sites. TRF 98 represents sequences affiliated with the *Nitrosospira*-like cluster (Peng et al., 2013). TRFs 343 and 462 were significantly greater in surface sediments compared to deeper sediments (*P* = 0.001 and *P* = 0.003, respectively), but no differences with restoration status were detected for these two TRFs.

### **DISCUSSION**

undisturbed (U) marshes.

In this study, we report significant impacts of chronic disturbance on coastal sediment microbial communities involved in nitrogen cycling. Our results suggest that disturbances may differentially impact different groups of microbes, and that sediment depth may be an important factor in characterizing impacts and subsequent recovery of sediment microbial communities. In a previous study of bacterial 16S rRNA genes from surface sediments from the same marshes surveyed in this study, communities were more variable at restored sites compared to undisturbed sites, but the overall community composition was similar and there were no differences in bacterial abundance (Bernhard et al., 2012). Our focus on specific functional groups, however, suggests that after nearly 30 years of restored tidal flow, there are detectable differences in abundance and community composition between restored and undisturbed marshes.

Abundance of *nir*S genes was similar to abundances reported in other salt marshes and estuaries, although in some cases our values were slightly higher than previously reported. Bowen et al. (2011) reported *nir*S abundances ranging from 10<sup>4</sup> to 10<sup>5</sup> copies per ng of DNA (our data converted to these units ranged from <sup>4</sup> <sup>×</sup> 104 to 5 <sup>×</sup> <sup>10</sup><sup>6</sup> gene copies/ng DNA) in two New England salt marshes, and ratios of *nir*S to bacterial 16S rRNA genes of 1–4%. Similarly, our *nir*S abundances were similar to those reported in San Francisco Bay estuarine sediments (Mosier and Francis, 2010), when converted to copies of *nir*S per gram wet weight. Abundances of *nir*S genes in Elkhorn Slough sediments (Smith et al., 2014), however, are at the low range of our data, with many of our samples being 1–2 orders of magnitude higher. Elkhorn Slough is an agriculturally impacted estuary, and may have different nitrogen-cycling dynamics compared to the salt marshes in Long Island Sound. The studies in San Francisco Bay and Elkhorn Slough are also from unvegetated sediments, which would also likely have different nitrogen dynamics compared to vegetated sediments in salt marshes.

Archaeal and betaproteobacterial *amo*A gene abundances at the surface are within the ranges reported in other estuaries and salt marshes (see review by Bernhard and Bollmann, 2010). Ratios of AOA and AOB to bacterial 16S rRNA genes are much lower than *nir*S ratios, which is expected for obligate chemoautotrophs. Numbers for AOA and AOB at depth have not been previously reported in these ecosystems to our knowledge. Because ammonia oxidation is an aerobic process, most studies have focused on surface sediments. However, extensive root systems in salt marshes that may provide oxygen (Mendelssohn et al., 1981; Howes et al., 1986), as well as bioturbation by invertebrates (Dollhopf et al., 2005), so it is likely that there may be aerobic micropockets that can support smaller populations of these aerobic microorganisms. The lack of a significant difference of relative AOB to bacterial 16S rRNA between restored and undisturbed marshes in deeper sediments suggests that the AOB may be less impacted compared to AOA and *nir*S, and may reflect a more general effect on the total microbial abundances rather than a specific effect on AOB.

Increased abundance of nitrifiers and denitrifiers in restored marshes compared to undisturbed marshes in deeper sediment suggests a more general impact of disturbance on nitrogen-cycling in the marsh. Increases in abundance could indicate higher rates of nitrification and denitrification at these sites. Some studies have shown strong correlations between *amo*A gene abundance and nitrification rates in estuaries for both AOA and AOB (recently reviewed in Bernhard and Bollmann, 2010). We should also note that differences in gene abundance could reflect differences in gene copy numbers per cell, rather than an increase in the population size. Some denitrifiers and AOB are known to have multiple copies of *nir*S (Etchebehere and Tiedje, 2005; Jones et al., 2008) or *amo*A (Norton et al., 1996), respectively. Multiple copies of *amo*A in AOA, however, have not been reported (Zhalnina et al., 2014).

Finding significantly greater effects of disturbance on community composition and abundance in deeper sediments relative to surface sediments suggests that the impacts of disturbance may be greater or have much longer lasting effects in deeper sediment. Impacts of disturbance may be more pronounced deeper in the sediments, since surface sediments are likely to be resuspended with each tidal cycle and redistributed across the marsh landscape, thus obscuring evidence of disturbance. Significantly different salinity between surface and deep sediments at the restored marshes, but not in the undisturbed marshes, suggests there may still be significant differences in the hydrology of the marsh. Salinity has previously been reported to be an important factor in driving nitrifier (see Bernhard and Bollmann, 2010 and references cited within) and denitrifier (Yoshie et al., 2004; Santoro et al., 2006; Bulow et al., 2008) communities in estuaries and salt marshes. Differences in hydrology would also be expected to impact oxygen levels. When marshes are tidally restricted, the water table drops, allowing increased oxygen penetration, and the sediment subsides (Portnoy, 1999). Once seawater is restored, there may be significant differences in porewater chemistry due to the lower elevation.

Additionally, Swamy et al. (2002) reported significantly different densities of some benthic invertebrates in restored marshes in Barn Island compared to reference marshes, which may significantly impact sediment turnover and C and N distributions (Wang et al., 2010), as well as oxygenation of deeper sediments. Furthermore, since root and rhizome densities may vary with sediment depth (e.g., Bertness, 1985), there may also be depthspecific differences in the relative importance of plant activity on how microbes grow and recover after disturbance. Previous studies in soils and sediments have suggested a significant effect of plant root exudates on denitrifying activity (Kaplan et al., 1979; Christensen and Sørensen, 1986; Risgaard-Petersen and Jensen, 1997; Henry et al., 2008). Root exudates are also known to significantly impact oxygen in sediments (Mendelssohn et al., 1981; Howes et al., 1986).

Significantly different NH4 + concentrations between restored and undisturbed marshes in deeper sediments suggests differences in resource availability for nitrifiers. If nitrification and denitrification are coupled (Risgaard-Petersen, 2003), nitrifier resource availability would be expected to impact denitrifiers as well. Higher ammonium in restored sites compared to undisturbed sites may reflect differences in mineralization of N due to increased decomposition. Others have reported increased decomposition via sulfate reduction in impounded and restored marshes compared to reference marshes (van Proosdij et al., 2010). Portnoy and Giblin (1997) also reported increased N and P mineralization and increased sulfides, suggesting accelerated rates of sulfate reduction in restored marshes.

The recovery trajectory implied by the community patterns for *nir*S and betaproteobacterial *amo*A genes support a repeatable pattern for recovery of these genes in deep sediment. The lack of a similar pattern in surface sediments may indicate that only certain communities show reproducible recovery patterns or that certain edaphic conditions are more likely to lead to these patterns. Bach et al. (2010) reported a directional shift in microbial communities in silty clay soils, but not in sandy loam soils. Similarly, Levine et al. (2011) reported a recovery trajectory for methanotrophs, but not for heterotrophic bacteria, while Banning et al. (2011) reported broad support for successional changes at the phylum level in soil bacterial communities. These seemingly contrary results suggest that either sampling or methodology was inadequate to detect the patterns, or that recovery or succession of some communities is not as predictable or reproducible as others. Recently, Pagaling et al. (2014) demonstrated that historical contingencies also play a role in determining predictable patterns of succession.

The directional shift in *nir*S communities may provide some insight into how long it takes for communities to recover, or at least become indistinguishable from communities in undisturbed marshes. Tidal inundation was restored to Impoundments 1 and 2 in 1978, and the denitrifying communities at these sites are more similar to those in undisturbed marshes compared to the denitrifying communities in Impoundments 3 and 4, which did not have tidal flow restored until 10–12 years later. Others have reported similarly long-lasting effects of disturbance on soil microbial communities (Bach et al., 2010; Levine et al., 2011) and have found strong correlations of microbial community recovery with soil texture (Bach et al., 2010), edaphic conditions such as pH, C, N, and P availability (Banning et al., 2011), and microbial immigration rates (Lambert et al., 2012).

We also identified specific denitrifier populations (i.e., specific TRFs) that were significantly different between restored and undisturbed sediment. Unfortunately, in almost all cases, TRFs represent polyphyletic groups, so we cannot relate specific phylogenetic clusters with recovery. However, several TRFs show patterns that are consistent with a recovery trajectory. TRFs 71, 142, and 277 are significantly more abundant in the undisturbed marshes, and also show a statistically significant increase from their abundance in Impoundments 3 and 4 to Impoundments 1 and 2, as you would predict if these denitrifiers are representative of undisturbed or recovered marshes. Following changes in these TRFs may provide a monitoring mechanism for the recovery of denitrifying communities in salt marshes. Future sampling in the Barn Island marshes will determine whether the recovery trajectory of denitrifiers is on track to return to resemble those of undisturbed marshes.

Significant site effects among the undisturbed sites for *nir*S genes in the surface sediments suggest that restoration to predisturbance conditions may be impossible to fully assess given the variation among undisturbed sites. However, since there was no site effect among the restored sites and all of the undisturbed sites were significantly different from the restored sites, the data still support a significant disturbance effect in the marsh.

Although the patterns for AOB do not indicate an obvious recovery trajectory among the four impoundments as for denitrifiers, the relative abundance patterns for TRFs 98 and 127 in deeper sediment suggest differential responses to disturbance. Additional experiments, however, are necessary to explore this hypothesis.

Disturbance effects were also detected in diversity patterns for *nir*S genes. However, we interpret the data with caution since *nir*S TRFs show little correlation to phylogeny and each TRF represents multiple sequence types and sequence types are represented by multiple TRFs. However, Shannon indices were just slightly lower than those previously reported for estuarine sediments based on *nir*S sequences rather than TRF patterns (Santoro et al., 2006). Other studies have shown significant effects of disturbance on microbial diversity (Wittebolle et al., 2009; Levine et al., 2011). More sequences are necessary to fully describe *nir*S diversity.

Recent studies of disturbance on salt marsh microbial communities have revealed surprisingly few effects (Bowen et al., 2009, 2011; Bernhard et al., 2012; Peng et al., 2013), even though the studies were conducted in different marshes with different disturbance regimes, used different methods, and focused on different microbial groups. It is possible that because the studies were done many years after the initial disturbances, any alterations in the microbial communities may have been missed. However, in some cases, macrobiota and biogeochemical processes were significantly different following disturbance and have remained different many years post-disturbance, yet analysis of the microbial communities from the same areas suggest a surprising degree of resilience or resistance to perturbations. It is also possible that the microbes present in these marshes show a surprising degree of phenotypic plasticity as the conditions change, thus leading to changes in process rates, but no detectable change in the community composition.

An alternative hypothesis, however, is that changes may vary with sediment depth. One commonality of the studies is that they focused only on surface (the top 2 cm or less) sediments, yet we found greater impacts of disturbance in deeper (6–8 cm) sediments. In a previous study of the same sites as presented here, Bernhard et al. (2012) reported no differences in abundance of bacterial 16S rRNA genes in surface sediments. Based on the depth-specific patterns we observed for functional genes in the current work, we decided to measure bacterial 16S rRNA gene abundance from the same deep sediments. In support of our hypothesis, we found significantly higher abundance of bacterial 16S rRNA genes in the 6–8 cm sediments from restored sites (Bernhard, unpublished), thus echoing the patterns of abundance for *nir*S in restored and undisturbed sites. Additionally, in a study of recovery in a freshwater wetland (Meyer et al., 2008), most of the changes in N and C pools occurred within the surface across a 10-year chronosequence.

It will be interesting to conduct additional community composition studies based on bacterial 16S rRNA to determine if the bacterial communities may also differ significantly at depth. Although the data are limited, we believe further investigation is warranted and may change the conclusions from previous studies of disturbance on salt marsh microbial communities if more "indepth" analyses are conducted in the future. We must also report that recently, Bowen et al. (2013) was able to detect small differences in the gene sequences of denitrifying communities in surface sediments (0–1 cm) in fertilized plots compared to unfertilized

plots after analyzing more than 60,000 *nir*S sequences, suggesting that we may just need to look harder to identify differences in surface sediments.

In conclusion, our study suggests that salt marsh nitrogencycling microbial communities may be impacted by disturbance, but it is uncertain whether the altered communities are functionally similar to those in undisturbed marshes. If there is a link between community composition, gene abundance, and process rates, as some studies have documented for nitrifiers and denitrifiers (e.g., Cavigelli and Robertson, 2000; Webster et al., 2005; Bernhard et al., 2007; Philippot et al., 2013), then our data point strongly toward differences in nitrogen-cycling in disturbed areas of the marsh compared to undisturbed marshes. However, in a study of disturbance on soil denitrifiers, Wertz et al. (2007) found that a community may recover functionally even though the diversity remains altered, suggesting some level of functional redundancy. Similarly, Cao et al. (2008) reported that denitrifier community composition and function were uncoupled in a California salt marsh. Whether the communities of nitrifiers and denitrifiers function similarly in disturbed and undisturbed marshes or not, it is important to understand how microbial communities respond to disturbance and to identify underlying principles that govern the changes.

### **AUTHOR CONTRIBUTIONS**

Anne E. Bernhard designed the project, provided QPCR data, and wrote the manuscript; Courtney Dwyer and Adrian Idrizi provided community fingerprints, analyses, and interpretation; Rachel Zwick and Geoffrey Bender provided sequence data, phylogenetic analyses, and interpretations, and Geoffrey Bender helped with sample collection and processing. All authors contributed to the intellectual content and editing of drafts.

### **ACKNOWLEDGMENTS**

This work was supported in part by the National Science Foundation award DEB-0814586 (Anne E. Bernhard ). We also thank Roberta Sheffer for her technical assistance on all laboratory analyses. Additional support was provided by the George and Carol Milne Endowment and R. F. Johnson funds at Connecticut College.

### **SUPPLEMENTARY MATERIAL**

The Supplementary Material for this article can be found online at:http://www.frontiersin.org/journal/10.3389/fmicb.2015.00046/ abstract

### **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.

*Received: 22 October 2014; accepted: 13 January 2015; published online: 04 February 2015.*

*Citation: Bernhard AE, Dwyer C, Idrizi A, Bender G and Zwick R (2015) Long-term impacts of disturbance on nitrogen-cycling bacteria in a New England salt marsh. Front. Microbiol. 6:46. doi: 10.3389/fmicb.2015.00046*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology.*

*Copyright © 2015 Bernhard, Dwyer, Idrizi, Bender and Zwick. 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.*

# **A network-based approach to disturbance transmission through microbial interactions**

*Dana E. Hunt <sup>1</sup> \* and Christopher S. Ward 1,2*

*<sup>1</sup> Marine Laboratory, Duke University, Beaufort, NC, USA, <sup>2</sup> Integrated Toxicology and Environmental Health Program, Duke University, Durham, NC, USA*

Microbes numerically dominate aquatic ecosystems and play key roles in the biogeochemistry and the health of these environments. Due to their short generations times and high diversity, microbial communities are among the first responders to environmental changes, including natural and anthropogenic disturbances such as storms, pollutant releases, and upwelling. These disturbances affect members of the microbial communities both directly and indirectly through interactions with impacted community members. Thus, interactions can influence disturbance propagation through the microbial community by either expanding the range of organisms affected or buffering the influence of disturbance. For example, interactions may expand the number of disturbance-affected taxa by favoring a competitor or buffer the impacts of disturbance when a potentially disturbance-responsive clade's growth is limited by an essential microbial partner. Here, we discuss the potential to use inferred ecological association networks to examine how disturbances propagate through microbial communities focusing on a case study of a coastal community's response to a storm. This approach will offer greater insight into how disturbances can produce community-wide impacts on aquatic environments following transient changes in environmental parameters.

### *Edited by:*

*Mark V. Brown, University of New South Wales, Australia*

### *Reviewed by:*

*Eric Fouilland, Centre National de la Recherche Scientifique, France Alexander Eiler, Uppsala University, Sweden*

> *\*Correspondence: Dana E. Hunt dana.hunt@duke.edu*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 27 May 2015 Accepted: 12 October 2015 Published: 27 October 2015*

### *Citation:*

*Hunt DE and Ward CS (2015) A network-based approach to disturbance transmission through microbial interactions. Front. Microbiol. 6:1182. doi: 10.3389/fmicb.2015.01182* **Keywords: interaction networks, disturbance, phytoplankton, anthropogenic, storms**

# **MICROBES AS IMPORTANT RESPONDERS TO ECOSYSTEM CHANGES**

Most people and development reside near water bodies, so human activities profoundly affect both freshwater and marine ecosystems (Vitousek et al., 1997). In these aquatic environments, microbes are the numerically- and often the biomass-dominant organisms, thus how they respond to anthropogenic impacts determines both ecosystem health and biogeochemical rates. Although a large body of research explores microbial responses to long-term human alteration of the environment (e.g., climate change, ocean acidification), here we focus on pulse disturbance events that disrupt "ecosystem, community, or population structure and [change] resources, substrate availability or the physical environment" (White and Pickett, 1985). High levels of diversity and short generation times make aquatic microbes a sensitive model system to explore disturbance, but also complicate tracking the impacts and progression of disturbance. The wide range of pulse disturbances affecting aquatic environments including storms, snowmelt, mixing/upwelling, and chemical or sewage spills allows microbial ecologists to probe community responses to environmental changes.

In general, microbial communities are not resistant, which is defined by Allison and Martiny (2008) as the degree to which microbial composition remains unchanged in the face of disturbance. This low resistance is likely due to the wide range of genetic and physiological targets present in diverse microbial communities as well as microbes' short generation times, which allow observation of both positive (increased growth) and negative (death, impaired growth) responses. Following a disturbance, community resistance and resilience are generally determined by comparing community composition at specific time points (Shade et al., 2012). A metric of community recovery, microbial resilience is generally defined as a return to the initial community composition (Allison and Martiny, 2008; Shade et al., 2012). However, aquatic microbial communities are highly dynamic and continually change in response to seasonal environmental variables (e.g., light and temperature) or subsequent disturbances (Chow et al., 2013; Needham et al., 2013; Yung et al., 2015). Thus, we define the resilience of an aquatic microbial community as the rate at which the community composition returns to a *non-disturbed* state following a disturbance. This definition of resilience requires understanding the disturbance-independent temporal dynamics of microbial communities. Although marine microbial communities exhibit regular seasonal patterns at monthly timescales (Fuhrman et al., 2006; Gilbert et al., 2012; Giovannoni and Vergin, 2012), high resolution and repeated annual sampling reveals shorter-term and inter-annual variability at the days to weeks time scale of disturbance responses (El-Swais et al., 2015), complicating differentiation of disturbance responses from annual patterns and stochasticity. However, even if the seasonal community trajectory is known, challenges to measuring microbial responses to disturbance include confounding factors such as unrelated changes in environmental variables, stochasticity in response and recovery, dispersal limitation, genomic evolution to become resistant to disturbances, and microbial interactions with other organisms (Shade et al., 2012; Nemergut et al., 2013). We propose to begin addressing the importance of microbial interactions to gain new insights into the mechanisms underlying the resistance and resilience of microbial communities.

# **MICROBIAL INTERACTIONS IN COMMUNITY ASSEMBLY**

As identifying the drivers of microbial community composition is complex, most investigators first consider environmental selection, and generally secondarily address other aspects of community assembly: dispersal, drift (stochasticity), and diversification (mutation; Vellend, 2010; Hanson et al., 2012; Nemergut et al., 2013). However, dispersal may limit the viable population present even when conditions favor growth (Caporaso et al., 2011; Hanson et al., 2012) or alternately, environmental changes may not persist long enough for viable cells to respond (Hutchinson, 1961). An emphasis on deterministic processes also ignores the role of stochasticity in community assembly and the potential for communities with different compositions to carry out the same processes at the same rates (e.g., functional redundancy; Werner et al., 2011; Bissett et al., 2013; Hellweger et al., 2014; Zhou et al., 2014). Further, microbial genomes evolve in response to disturbance; they can develop resistance to disturbances such as antibiotics or heavy metals, alter metabolic capabilities, and change physiological niche width (Riehle et al., 2003; Davies and Davies, 2010). Although microbial communities are shaped by a combination of selection, drift, dispersal and evolution, there is value in addressing subsets of these factors, here we focus on selection via biological interactions following disturbance.

Microbial ecology research currently emphasizes the role of interactions in the community response to environmental changes and disturbances (Faust et al., 2012; Bissett et al., 2013; Fuhrman et al., 2015). Although some examples of relationships between specific taxa and environmental variables exist (Field et al., 1997; Johnson et al., 2006; Yung et al., 2015), interactions between aquatic microbes have not been well explored. Even for predation by viruses and grazers, one of the best studied microbial interactions, much still remains to be discovered about the interaction specificity (Sullivan et al., 2003; Apple et al., 2011). Furthermore, the nature of biological interactions may be dictated by characteristics of dominant aquatic bacteria; the most abundant marine populations (e.g., *Pelagibacter*, *Prochlorococcus*) are known for their streamlined genomes, small cell sizes, and efficient use of resources (Giovannoni et al., 2014). Some of the evolutionary success of these organisms may be due to their conservation of limited resources by shedding genes encoding critical functions and outsourcing these functions to other members of the community (Black Queen Hypothesis; Morris et al., 2012). For example, both *Pelagibacter* and *Prochlorococcus* have lost the gene for catalase which protects cells from hydrogen peroxide; as hydrogen peroxide diffuses through cell membranes, other members of the microbial community can protect catalase non-producers (Morris et al., 2011, 2012). Yet aquatic organisms with complex genomes have also evolved required interactions with other organisms; many eukaryotic algae have a B12-dependent methionine synthase rather than the B12-independent version, despite the fact that B<sup>12</sup> is only synthesized by prokaryotes. This suggests that interactions with other organisms evolve due to net ecological advantage rather than solely genome streamlining.

Although outsourcing key requirements may be ecologically advantageous, long distances between cells, on average *∼*100 µm (Hunt et al., 2010), may exclude specific types of biological interactions for planktonic organisms such as syntrophy where physical coupling allows efficient transfer between cells (Boetius et al., 2000; Malfatti and Azam, 2009). For truly free-living organisms, interactions likely involve diffusible compounds, suggesting that interaction partners may not be highly specific or involve complex regulation. Experimental evidence supports complementation of lost capabilities by non-specific interaction partners: a range of reduced sulfur sources can be used by SAR11 (Tripp et al., 2008) and many bacteria can provide B<sup>12</sup> for auxotrophs (Croft et al., 2005). Additionally, some obligate relationships, at least in artificial laboratory conditions, do not involve regulation or signaling (Durham et al., 2015), while others are regulated (Kazamia et al., 2012), suggesting a number of potential strategies for interactions. Although outsourcing key functions is thought to be evolutionarily adaptive, interactions also incur costs: B<sup>12</sup> additions have been shown to stimulate phytoplankton, implying that an interaction limits algal growth (Sañudo-Wilhelmy et al., 2006; Bertrand et al., 2007). While experimentally-verified interactions between microbes remain rare, the success of aquatic organisms may stem at least partially from outsourcing key functions. Thus, increasingly, microbial ecologists are incorporating interactions into our understanding of microbial communities, including interactionmediated transmission of disturbance, resistance, and resilience.

# **USING ASSOCIATION NETWORKS TO EXPLORE DISTURBANCE**

In general, microbial interactions cannot be directly observed, thus ecological relationships are instead inferred based on environmental observations of co-occurrence patterns and synchronous population dynamics (Ruan et al., 2006; Steele et al., 2011; Faust et al., 2012). Patterns of microbial relative abundance obtained from communities sampled over spatial or temporal gradients are used to generate correlation-based association networks of potential interactions between operational taxonomic units (OTUs) and between OTUs and environmental variables (Barberan et al., 2012; Faust et al., 2012; Fuhrman et al., 2015). These correlations are interpreted to capture biological mutualisms such as cross-feeding and exchange of metabolites (Kazamia et al., 2012; Morris et al., 2012), functional redundancy (Eiler et al., 2012; Needham et al., 2013), or antagonism through competition or predation (Pernthaler and Amann, 2005). In addition to the well-known biases of DNA extraction, PCR amplification and in inferring patterns of organismal abundance from library relative abundance data (Polz and Cavanaugh, 1998; Acinas et al., 2004; Friedman and Alm, 2012), association networks also suffer from a number of networkspecific limitations. First, association networks assume that 16S rRNA-based OTUs are ecologically coherent in spite of known microdiversity (Hunt et al., 2008) and physiologically identical under all environmental conditions, e.g., does not account for phenotypic plasticity based on environmental conditions (Nemergut et al., 2013; Worden et al., 2015). Second, associations may serve as proxies for specific environmental conditions or niches rather than indicating true interactions (Fuhrman et al., 2015). Finally, metrics of association strength are not standard and depend on the metric chosen, number of samples, taxa relative abundance, beta diversity, and data normalization (Ruan et al., 2006; Faust et al., 2012; Friedman and Alm, 2012; Berry and Widder, 2014). Currently, this field also lacks methods to add additional support for interactions such as observed physical associations to networks (Malfatti and Azam, 2009; de Vargas et al., 2015). While acknowledging the limitations of correlationbased association networks, we believe this technique has the potential to inform our understanding of aquatic microbial community dynamics.

Recently, association networks were employed to predict the bacterial response to disturbance (Bissett et al., 2013); expanding on this work, we propose to use network approaches to quantitatively examine the importance of interactions in altering the taxa affected by disturbance. Of particular promise are techniques developed in information technology and social learning, where interactions transmit signals between nodes, much in the same way that initial disturbance-induced changes in an OTU's abundance may in turn affect the abundance of its interaction partners at later time points. One technique to look at disturbance transmission, information flow analysis can model the transmission of disturbance through the interaction network using the interaction strength and considering all possible paths in a network (Missiuro et al., 2009). Information flow analysis accounts for the strength of inferred interactions, enabling prediction of how changes in the relative abundance of a specific organism or value of an environmental variable will affect the microbial community, and thus provides a metric of predicted community resistance. Additionally, network-based diffusion analysis could be used to determine quantitatively whether association networks help to explain the propagation of disturbance through the community (Franz and Nunn, 2009). Operationally, association networks would be used to predict the temporal dynamics of microbial community composition following disturbance. The effects of disturbance on the rest of the community (changes in OTU relative abundances) can be predicted using information flow analysis. This predicted community composition would be compared to the actual community composition following a disturbance and community changes predicted from a randomized network generated by preserving the association network topology but repeatedly, randomly assigning OTUs to network nodes. Thus if the association network's inferred interactions are truly important in the community's disturbance response, the true association network should more closely match the observed community responses compared to a set of randomized networks. These methods will quantify the importance of interactions and predict community responses to specific environmental conditions, enhancing our understanding of the role interactions play in disturbance.

Although these techniques are potentially powerful methods to track community responses to disturbance, there are a number of logistical considerations in using association networks to follow the propagation of disturbance through microbial communities. First, network-based analyses require large datasets both pre- and post-disturbance synoptic with community changes to develop an association network and track the disturbance response, respectively. As disturbance-responsive taxa are often rare, they may not be well-represented in association networks which generally require taxa to be present in most samples (Shade et al., 2014). Moreover, taxa which can respond quickly to environmental changes may exhibit fewer, or different types of biological interactions than the streamlined genome oligotrophs which dominate many aquatic environments (Polz et al., 2006). Additionally, microbial community composition, generally measured using small subunit ribosomal RNA genes, may not be sufficiently sensitive to detect a disturbance response due to the time for cells to reproduce or predation of responsive taxa, necessitating the use of alternative metrics such as activity measurements (Berga et al., 2012; Hunt et al., 2013). Finally, dispersal may limit the response of taxa even under conditions which favor growth. With the relatively short time scales of pulse disturbances, it may be necessary to include prior relative abundance in predicting an OTU's potential responsiveness to disturbance. With all of these caveats in place—we suggest first studying time periods when disturbances are predicted to produce large changes in the microbial community.

Theoretically, anthropogenic disturbances should have the greatest impact when highly-connected taxa change their abundance or activity. Research on networks has shown that disturbances that target central "keystone" nodes dramatically alter the rest of the network (Albert et al., 2000; Montoya and Solé, 2002). Ecology posits the existence of keystone taxa—which may impact multiple members of the community through either positive interactions (production of substrates or co-factors utilized by other microbes) or competitive exclusion, predation, disease, or habitat modification (Power et al., 1996). Keystone organisms are often defined as those with disproportionate ecological roles given their relative abundance (Power et al., 1996); however as microbial ecology lacks techniques to remove specific OTUs and quantify the ecosystem effect, here we operationally define keystones as taxa located at the hubs of association networks with an increased number of network connections relative to abundance (high mean degree); however, other metrics take into account the betweenness and closeness centralities of the node as well as strength of interactions (Missiuro et al., 2009; Bissett et al., 2013; Berry and Widder, 2014; Peura et al., 2015). Yet many network hubs may be artifacts of network construction rather than true keystone taxa (Berry and Widder, 2014). Although the concept of keystone taxa has not been thoroughly explored in microbial ecology, previous studies have suggested that microbial community activity and succession is driven by interactions with phytoplankton (Azam et al., 1983; Kent et al., 2007). The factors that promote phytoplankton growth are generally well known: light, inorganic nutrients, specific temperature ranges; and phytoplankton are the dominant primary producers in most aquatic systems. These photosynthetic organisms shape the microbial community through primary production, but at the same time outsource the production of essential functions (e.g., hydrogen peroxide detoxification) to the broader community (Cole, 1982; Kazamia et al., 2012; Morris et al., 2012). Other taxa interact with phytoplankton through photosynthate consumption, degradation of detrital material, symbiosis, and predation (Croft et al., 2005; Stocker et al., 2008; Morris et al., 2011; Teeling et al., 2012; Durham et al., 2015). Finally, phytoplankton serve as hubs in association networks (Steele et al., 2011) and could function as keystone organisms in aquatic ecosystems. While the ecological role of some potential keystone taxa has been identified, e.g., nitrogenfixing bacteria (Tyson et al., 2005), for most network hubs there is no known keystone function (Steele et al., 2011; Bissett et al., 2013). Thus the phytoplankton, where growth-promoting factors and relationships with other microbes are relatively well-characterized, represent an ideal model system in which to explore the biological interactions that underlie association networks during pulse disturbances.

# **USING STORMS TO EXPLORE DISTURBANCE PROPAGATION**

Storms represent complex, pulse disturbances that integrate both natural and human impacts. Storm-driven rain and wind events increase turbidity and introduce nutrients, organic material, and microbes from both the benthos and land into aquatic systems; while anthropogenic activity increases nutrient fluxes, impacts the timing of freshwater inputs, and contributes other chemical pollutants. Thus storms are multi-faceted disturbances; yet, unlike some discrete disturbances (e.g., Deepwater Horizon oil spill), they occur frequently enough to allow comparison across different storms, environments, and microbial communities (Berga et al., 2012; Yeo et al., 2013). Here we use storms as a model disturbance to explore using association networks to track the propagation of disturbance through the microbial community.

To investigate this concept further, we will follow the progression of storm-mediated impacts on a simplified microbial community association network where an alga serves as a keystone microbe and a network hub. In our model system (**Figure 1**), the major storm impact is an increase in nutrients (Iluz et al., 2009; Johnson et al., 2013); and the first microbial community responder is the keystone algal OTU, which is positively correlated with nutrient levels. Using association networks prepared from non-disturbance data (**Figures 1A,B**), we can infer which other OTUs are likely to respond to a change in algal abundance. With high resolution post-storm sampling, we can observe changes in OTUs correlated with the early responders, as shown by lines (edges) connecting these taxa to the alga, which should exhibit changes in activity or relative abundance at intermediate time points if that OTU is dependent on the alga, e.g., through metabolism of photosynthate (**Figures 1C,D**: yellow circles). At still later time points, the disturbance may propagate to taxa which interact with the yellow OTUs (**Figure 1D**: green circles). Alternately, at this same time point, OTUs with inferred relationships with the alga, but utilizing detritus associated with bloom termination rather than photosynthate from active algal cells may exhibit increases in relative abundance (**Figures 1C,D**: green circle; Teeling et al., 2012). Thus network-based approaches can offer biological insights into phytoplankton-bacterial interactions, the propagation and persistence of disturbance (**Figure 1**), and community stability (Carpenter et al., 2011; Veraart et al., 2012). Even anecdotal observations of how OTUs respond to disturbance can generate hypotheses that can be verified using more controlled laboratory or manipulation experiments.

Here, we have presented a cartoon storm as a pulse of nutrients, in reality storms and other ecological disturbances are complex. In addition to nutrients, storms introduce human pollutants into aquatic ecosystems, including pesticides, oil, untreated human waste, etc., that will have direct and interactionmediated effects on the microbial community. Unlike our simple example in **Figure 1**, there may be multiple, competing impacts on our keystone algal OTU. For example, chemical herbicides such as atrazine impact phytoplankton due to the conservation of photosystem II between cyanobacteria, algae, and plants (Huber, 1993). While, the specific impacts of most chemicals

direction of inferred disturbance propagation through the network based on the timing of observed changes in OTU relative abundance.

are correlated with concentration; another herbicide class of synthetic auxins (e.g., 2,4-dichlorophenoxyacetic acid) is toxic to cyanobacteria at high concentrations but stimulates growth at lower levels (Mishra and Pandey, 1989), a subtlety which is not readily incorporated into association networks. Among other anthropogenic pollutants, fungicides are generally less specific than herbicides, targeting highly-conserved cellular processes such as respiration and thus directly affect a range of microbes (Casida, 2009; Yang et al., 2011). Thus, along with nutrients, storms introduce a cocktail of chemicals to aquatic environments, complicating evaluation of direct and indirect community effects on the microbial community.

# **CONCLUSIONS**

Here, we discuss the potential for association networks to track the propagation and persistence of disturbance in a microbial community. We have identified two major opportunities afforded by this approach: (1) to quantify the importance of interactions in a microbial community's response to disturbance and (2) to generate biological hypotheses about the network's inferred interactions. However, a major challenge of this approach is that to characterize a microbial community's resistance and resilience we first need to understand disturbance-independent microbial community dynamics (Shade et al., 2012), suggesting the need for long-term monitoring of key study sites. Although the vast amounts of data required can appear daunting, specific taxa have been shown to repeatedly respond to storms (Jones et al., 2008) and the field is beginning to identify general characteristics of disturbance-responsive organisms (Shade et al., 2014), suggesting that there are conserved rules that govern microbial communities' disturbance responses. However, to tease apart the effects of factors that tend to co-vary in the environment, for example, separating the stimulatory effects of increasing nitrogen versus organic carbon, there is an additional role for controlled, replicated manipulations of natural aquatic communities. Beyond community changes, these experiments will also provide predictions about the alteration and restoration of ecosystem function following a disturbance, either by linking

### **REFERENCES**


specific taxa to functions or by identifying the types of disturbance which may be most likely to disrupt specific processes (Amend et al., 2015). An association network-based approach to analyzing microbial community disturbances and experimental manipulations will provide a basis to mechanistically predict community response to both pulse and press environmental changes.

### **ACKNOWLEDGMENTS**

This work was supported grants from the NSF (OCE 1322950; OCE 1416665) and the Gordon and Betty Moore Foundation (GBMF3768) to DH and by a NSF Graduate Research Fellowship to CW.

communities over a decade in the Southern California Bight. *ISME J.* 7, 2259–2273. doi: 10.1038/ismej.2013.122


factoring in the multifarious lifestyles of microbes. *Science* 347:1257594. doi: 10.1126/science.1257594


fluidic ecosystem. *Proc. Nat. Acad. Sci.* 111, E836–E845. doi: 10.1073/pnas. 1324044111

**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 Hunt and Ward. 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.*

# Next-generation sequencing (NGS) for assessment of microbial water quality: current progress, challenges, and future opportunities

*BoonFei Tan1, Charmaine Ng2, Jean Pierre Nshimyimana1,3,4, Lay Leng Loh1,2, Karina Y.-H. Gin2 and Janelle R. Thompson1,5\**

### *Edited by:*

*Mark Vincent Brown, University of New South Wales, Australia*

### *Reviewed by:*

*Lisa Moore, University of Southern Maine, USA Rebecca Gast, Woods Hole Oceanographic Institution, USA*

### *\*Correspondence:*

*Janelle R. Thompson, Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Building 48, Cambridge, MA 02139, USA jthompson@mit.edu*

### *Specialty section:*

*This article was submitted to Aquatic Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 13 July 2015 Accepted: 10 September 2015 Published: 25 September 2015*

### *Citation:*

*Tan B, Ng C, Nshimyimana JP, Loh LL, Gin KY-H and Thompson JR (2015) Next-generation sequencing (NGS) for assessment of microbial water quality: current progress, challenges, and future opportunities. Front. Microbiol. 6:1027. doi: 10.3389/fmicb.2015.01027* *<sup>1</sup> Center for Environmental Sensing and Modelling, Singapore-MIT Alliance for Research and Technology Centre, Singapore, Singapore, <sup>2</sup> Department of Civil and Environmental Engineering, National University of Singapore, Singapore, Singapore, <sup>3</sup> Singapore Centre on Environmental Life Sciences Engineering, Nanyang Technological University, Singapore, Singapore, <sup>4</sup> School of Civil and Environmental Engineering, Nanyang Technological University, Singapore, Singapore, <sup>5</sup> Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, Cambridge, MA, USA*

Water quality is an emergent property of a complex system comprised of interacting microbial populations and introduced microbial and chemical contaminants. Studies leveraging next-generation sequencing (NGS) technologies are providing new insights into the ecology of microbially mediated processes that influence fresh water quality such as algal blooms, contaminant biodegradation, and pathogen dissemination. In addition, sequencing methods targeting small subunit (SSU) rRNA hypervariable regions have allowed identification of signature microbial species that serve as bioindicators for sewage contamination in these environments. Beyond amplicon sequencing, metagenomic and metatranscriptomic analyses of microbial communities in fresh water environments reveal the genetic capabilities and interplay of waterborne microorganisms, shedding light on the mechanisms for production and biodegradation of toxins and other contaminants. This review discusses the challenges and benefits of applying NGS-based methods to water quality research and assessment. We will consider the suitability and biases inherent in the application of NGS as a screening tool for assessment of biological risks and discuss the potential and limitations for direct quantitative interpretation of NGS data. Secondly, we will examine case studies from recent literature where NGS based methods have been applied to topics in water quality assessment, including development of bioindicators for sewage pollution and microbial source tracking, characterizing the distribution of toxin and antibiotic resistance genes in water samples, and investigating mechanisms of biodegradation of harmful pollutants that threaten water quality. Finally, we provide a short review of emerging NGS platforms and their potential applications to the next generation of water quality assessment tools.

Keywords: next-generation sequencing, water quality, fecal indicator, antibiotic resistance, harmful algal bloom, sewage, biodegradation

### The Role and Precedent of "Proxies" in Water Quality Analysis

Surface freshwaters including lakes, rivers and streams, are important aquatic ecosystems and are a source of drinking water in many countries. In recent decades, increase in human population size and urbanization have exerted immense pressure on the use of these water resources for human recreation and consumption. Contamination by sewage, toxic chemicals, nutrients, and resultant harmful algal blooms can render water unfit for human consumption or recreational activities. Bio-monitoring using sentinel or indicator species in the form of microorganisms and aquatic macroinvertebrates are frequently used by water management authorities to infer water quality, ecosystem health status, and to protect public health from waterborne risks (Carew et al., 2013). The occurrence and abundance of indicator organisms serve as proxies, i.e., easily measured quantities that are correlated to often unknown agents that directly mediate waterborne risk such as pathogens, biotoxins and chemicals. Advances in molecular methods and next-generation sequencing (NGS) have ushered in new opportunities for water quality assessment through analysis of waterborne microbial communities for the development of indicators and sentinels, new markers for microbial source tracking, and observation of microbially mediated processes. However, as these new tools are developed, biases and uncertainties associated with nucleic-acid based methods and NGS must be considered.

For decades, water quality for drinking and recreational purposes has been largely assessed based on culture-based enumeration and detection of fecal indicator bacteria (FIB), e.g., total coliforms, *Escherichia coli,* or *Enterococci*: a practice that has been regarded as the "gold standard" in the assessment of microbial safety of water (Figueras and Borrego, 2010). FIB are at high concentration in human feces, thus their presence in freshwater serves as a proxy for associated pathogens, where risks of exposure and has been shown to correlate to incidence of disease in exposed populations (e.g., Haile et al., 1999; Zmirou et al., 2003; Colford et al., 2007; Harwood et al., 2014). However, these proxies are imperfect as they may be derived from non-human sources and/or may be subjected to ecological or environmental interactions that compromise their predictive power as proxies for pathogens. In addition to predicting contamination from sewage, microbial proxies are used to predict the origin or source of sewage since fecal material of human-origin poses the most significant risk to water quality due to its potential to transmit human pathogens. Research into "source tracking" has established at least two species within the genus *Bacteroides* as indicators of humanorigin, which have been shown to associate with and perhaps be symbionts of the human gut (Bernhard and Field, 2000; Shanks et al., 2007, 2009, 2010; Yampara-Iquise et al., 2008). Finally, direct detection and enumeration of specific pathogens such as waterborne *Leptospira, Campylobacter*, *Legionella*, viruses (e.g., Norovirus) and parasites (e.g., *Cryptosporidium*) enables quantitative microbial risk assessment (QMRA; Hass et al., 2014), where environmental concentrations of pathogens are compared to models of infectivity to characterize the risk to exposed human populations.

Development and validation of methods for enumeration of indicators and/or pathogens presents major challenges. Initial methods for identification of waterborne indicators and pathogens relied on selective culturing and enumeration of presumptive isolates, followed by species confirmation using biochemical, serological or molecular genetic methods. Such efforts lead to the standardization of simple and routine procedures for quantification of FIB, which are relatively easy to culture (Dufour, 1984; Simpson et al., 2002). However, cultivation methods for detection of specific pathogens are often time-consuming, and may fail to detect some organisms due to fastidious growth or requirement for a host (in the case of intracellular pathogens; Evangelista and Coburn, 2010). In addition, some pathogens are known to exist in a dormant, but still infective state that cannot be quantified by standard cultivation-based methods, i.e., the viable but not-culturable (VBNC) state (Ramamurthy et al., 2014).

To circumvent problems associated with quantification of VBNC and otherwise difficult to culture bacteria, cultivationindependent molecular methods were developed for detection and quantification of specific bacteria (recently reviewed by Ramamurthy et al., 2014). Such methods target DNA extracted from environmental and clinical samples which is subsequently subjected to analysis for the presence or abundance of genes from indicator species or pathogens of interest. The small subunit ribosomal RNA gene (SSU rRNA), consisting of the 16S rRNA gene for bacteria and the 18S rRNA gene for eukarya, has emerged as one of the most frequently used target genes for molecular analysis. This is due to its ubiquity in all organisms and sequence structure which includes both highly conserved and variable/hypervariable regions to promote alignment of the DNA sequence across diverse organisms, and allow for more finescale taxonomic identification (e.g., Guo et al., 2013). Use of the polymerase chain reaction (PCR) and quantitative PCR (qPCR) targeting regions of the SSU rRNA and functional genes for microbial source tracking and direct detection and quantification of target indicator strains has been recently reviewed by Harwood et al. (2014).

Cultivation-independent analyses based on direct detection of nucleic acids or amplification of targeted genes using PCR circumvent some problems associated with culture-based methods, while raising additional challenges. First, detection and quantification of environmental DNA from individual species is often assumed to be derived from living organisms, however, naked or free DNA may also be detected in such methods, making the correlation between DNA copies and cell abundance imperfect. Second, DNA-based methods used for quantification of microbial risk agents may be confounded by the highly dynamic and diversified genomes of pathogens and prevalence of strain-specific virulence factors. Thus, quantification of pathogenic taxa based on occurrence of biomarker DNA such as the SSU rRNA may not correlate to public health risk if the strain detected lacks virulence genes. Nevertheless, PCR-based detection and quantification of organisms in environmental DNA has proven useful for enumeration of sewage indicators

and one DNA-based method is approved for quantification of the fecal indicator *Enterococcus* by the US Environmental Protection Agency (USEPA, 2009, 2013) while additional DNAbased methods are currently under evaluation (Green et al., 2014). A third challenge is raised by the ability of nucleic-acid based analyses to detect as little as one DNA target molecule; thus, it is possible to detect trace levels of nucleic acids from microbial risk agents (e.g., pathogens or virulence factors). Risk assessment frameworks to define thresholds of "acceptable risk" will be necessary before incorporating quantification of microbial risk agents by DNA-based approaches into decision-making for water quality management.

### Brief Review of NGS Technologies and Analysis Methods

Advances in NGS enabling massively parallel analysis of DNA sequence information from PCR amplicons, or environmental nucleic acids, ushers in a new era of proxy development for water quality assessment. In clinical research, massively parallel sequencing (MPS) has been demonstrated as a screening tool used to complement or circumvent conventional diagnostic methods (e.g., culturing, microscopy and Gram-staining) for the detection and identification of etiological agents in disease (Kumar et al., 2013; Salipante et al., 2013). While quantitative tests for water quality assessment (e.g., qPCR, culture-based FIB quantification kits) are appropriate for estimating exposure to biological risk agents or sewage contamination, application of NGS surveys can be a first step to focus on more specific exposure assessment of appropriate targets.

Surveys of waterborne microbial communities using NGS thus far, have relied upon targeted sequencing of the hypervariable regions of SSU rRNA gene (e.g., V1, V3, V4, V6 regions) and Large Subunit (LSU) rRNA gene (e.g., Guo et al., 2013). Apart from SSU and LSU rRNA genes, other genes with taxonomic signals such as *nirS* (denitrification) and *nifH* (nitrogen fixation) indicative of biochemical cycles (e.g., Farnelid et al., 2011; Bowen et al., 2013), as well as plastid SSU rRNA (Steven et al., 2012) have also been adopted for NGS-based microbial profiling.

In early studies of the relationship between waterborne microbial communities and water quality, 454 pyrosequencing (Roche) emerged as the preferred platform of choice (e.g., McLellan et al., 2010; Vandewalle et al., 2012) due to relatively long sequence read lengths (i.e., initial read lengths of 110 bp, now currently ∼1000 bp van Dijk et al., 2014a) and generally better optimized sequencing conditions and bioinformatic workflows (e.g., Sergeant et al., 2012). The Illumina (Solexa) platform was introduced to the market with read lengths of 35 bp with a focus on genome sequencing (van Dijk et al., 2014a). However, as the Illumina technology improved and read lengths achieved by merging paired-end reads began to rival pyrosequencing, it has gained in stature for use as a platform for analysis of environmental samples. Other NGS platforms including Ion Torrent and single molecule real-time sequencing (SMRT) such as the Pacific Biosciences have also been used in amplicon sequencing for microbial community profiling (e.g., Marshall et al., 2012; Yergeau et al., 2012), although these technologies have not been

widely adopted. An emerging number of studies in recent years have used the Illumina MiSeq platform in shotgun amplicon sequencing of SSU rRNA (e.g., Newton et al., 2015), with several studies demonstrating improved performances (e.g., sequencing depth, coverage, detection sensitivity, false positive detection) compared to 454 pyrosequencing and Ion Torrent (Loman et al., 2012; Li et al., 2014; Sinclair et al., 2015).

Several open-source bioinformatic platforms including MOTHUR (Kozich et al., 2013) and QIIME (Caporaso et al., 2010), which are among some of the most commonly used software packages in the analyses of amplicon sequences, have been updated to improve on sequence analyses using paired-end sequence data generated from the Illumina platform. As Illumina MiSeq technology has become one of the most widely used sequencing platforms worldwide, and with the announcement by Roche to withdraw the GS FLX 454 pyrosequencing platform, several studies have embarked on further refining data analyses for Illumina platforms by improving on methods in library preparation (Kozich et al., 2013; Esling et al., 2015; Shishkin et al., 2015) and quality control of sequence reads (Kozich et al., 2013; Nelson et al., 2014; Schirmer et al., 2015). The success of using MPS of the SSU rRNA gene for routine monitoring and general sample comparative purposes, however, will hinge on efforts to streamline processes in samples processing, which at the current stage, can vary from one laboratory to another.

Small subunit rRNA gene amplicon sequences generated through MPS are generally clustered into operational taxonomic units (i.e., OTUs) based on nucleotide identity thresholds (e.g., 95–99%). In some cases, however, OTU clustering may fail to segregate highly identical sequence variants into ecologically- or environmentally- relevant groups. A method termed "oligotyping" has been developed to classify and group sequences based on sequence minimum entropy decomposition (MED), at the resolution of single nucleotide polymorphisms (SNPs), and has been shown to be particularly useful in tracing different microbial populations in sewage treatment facilities to their environmental origins (Eren et al., 2013, 2014). Community composition cataloged using NGS can serve as a baseline or reference for monitoring environmental perturbations or biodegradation. Furthermore, NGS-enabled biostatiscal analyses (e.g., principle coordinate analyses, nonmetric multidimensional scaling or correspondence analysis) have been adopted to correlate the occurrence and distribution of microbial taxa, OTUs or oligotypes to environmental metadata, thereby allowing identification of oligotypes or OTUs that can serve as bioindicators for environmental quality (e.g.,Yergeau et al., 2012).

In order for NGS to become a useful tool for water quality monitoring purposes, long term sequence data collection and management will be crucial in establishing databases that are important for inter-laboratory data comparison and comparative metagenomics studies. Using a combination of NGS approaches (e.g., metagenomics, metatranscriptomics, single-cell genomics, and comparative genomics) in systematic studies of freshwater microbiomes can be expected to yield a wealth of information crucial in water quality assessment and management.

### Can NGS be Quantitative?

### Massively Parallel Sequencing of SSU rRNA Gene Amplicons

Similar to proxies based on cultivation or cultivationindependent enumeration of indicators and pathogens, MPS of targeted genes such as the SSU rRNA has challenges which will prevent proxies developed with this technology from being 100% accurate. Moreover, the role of NGS in water quality assessment is in its infancy and has not yet been integrated into an epidemiological framework to link trends observed with NGS studies to adverse effects in human populations. One requirement of microbial risk assessment is the ability to accurately quantify biological risks/agents within the framework of QMRA (Hass et al., 2014). While a number of molecular tests including qPCR and digital PCR are known to be highly quantitative, providing results in the form of absolute gene copy number, profiles of species composition generated from amplicon sequencing are generally regarded as being qualitative. Biases in PCR amplification due to secondary structure or GC content of the resulting amplicons, generation of false diversity from sequencing error or chimera formation, choice of primers targeting different SSU rRNA hypervariable regions (Kozarewa et al., 2009; Quail et al., 2012; Kozich et al., 2013; Nelson et al., 2014; van Dijk et al., 2014b; Schirmer et al., 2015), as well as the presence of multiple copies of the SSU rRNA gene in some bacterial species (Angly et al., 2014) all influence the relative abundance of taxa observed by PCR-based methods with more pronounced biases associated with increased PCR cycle numbers (Murray et al., 2015; Schirmer et al., 2015; Sinclair et al., 2015). Sequencing errors associated with reads generated from the Illumina MiSeq platform may also result from differences in library preparation methods (Schirmer et al., 2015). Downstream bioinformatic data processing including methods in chimera removal (Kennedy et al., 2014) and OTU clustering (e.g., UCLAST, CD-HIT, UPARSE, Nelson et al., 2014; Schmidt et al., 2014; Sinclair et al., 2015) can similarly contribute to biases in determining the relative abundance and diversity of microbial taxa. Studies of mock bacterial communities suggest that these variations can sometimes result in inflated OTU numbers and therefore skew estimates of species richness and evenness (Kennedy et al., 2014; Nelson et al., 2014).

Despite the concerns discussed above, several studies have demonstrated that MPS of the SSU rRNA gene does offer a good approximation of the microbial species composition and relative abundance in samples (Ong et al., 2013; Wang et al., 2013), though the accuracy is likely sample-dependent and may be influenced by factors such as methods in library preparation and primer choice, as discussed above. Due to inherent differences in sample types, and associated microbiota, it is likely that no single universal approach can be best applied to all samples types to achieve quantitative measurement (Schmidt et al., 2014; Murray et al., 2015).

### Quantification of Gene Copy Number in Multi-Omics Datasets

Due to concern with biases in PCR amplification, several studies have carried out amplification-independent analysis of microbial communities using SSU rRNA genes extracted from metagenomic datasets. For example, Logares et al. (2014) recently assessed the diversity of marine plankton communities using three different NGS platforms (i.e., metagenomic shotgun sequencing by Illumina HiSeq and Roche 454, and amplicon sequencing of SSU rRNA gene by Roche 454). The diversity and composition of SSU rRNA genes observed in metagenomes prepared by the Illumina HiSeq platform provided higher and more even estimates of community diversity relative to the metagenomes or amplicons sequenced using the Roche 454 platform. In the same study, the relative abundance of microbial taxa observed by Illumina HiSeq was comparable to relative abundances observed by catalyzed reporter deposition fluorescence *in* situ hybridization (CARD-FISH) and flow cytometry (i.e., positive Pearson correlation and *p <* 0.01).

The choice of NGS platform may yield results with varying resolution and quantitative power due to differential sequence throughput and coverage. Frey et al. (2014) recently compared three sequencing platforms (i.e., Roche 454, Illumina MiSeq, and Ion Torrent) to characterize the relative abundance of viral and bacterial pathogens spiked into blood at known, serially diluted, concentrations (i.e., Dengue virus Types 1 and 2; Swine Influenza A, and *Bacillus anthracis*). The blood samples were lysed using commercial kits, and one portion of DNA sample was sequenced using the three independent NGS platforms, while the second portion was subjected to qPCR for the quantification of specific virus and bacterial targets. Sequence reads from NGS were mapped to the reference genomes of the tested organisms, and all three NGS platforms were found to produce results comparable to the qPCR assay for relative quantification. Illumina and Ion Torrent provided the highest sensitivity to detect sequences at the lowest dilutions due to their higher throughput than Roche 454 (Frey et al., 2014).

Several protocols have been developed to quantify absolute gene copy numbers or transcripts within a metagenome or metatranscriptome, respectively (Gifford et al., 2011; Satinsky et al., 2014a,b) by benchmarking to standards that were added to sample prior to total RNA/DNA extraction, library construction, and high throughput sequencing (Gifford et al., 2011). In several metatranscriptomics studies (Gifford et al., 2011; Satinsky et al., 2014a), *in vitro* RNA internal standards were prepared using commercially acquired plasmid DNA, linearized and restrictionenzyme digested, after which plasmid fragments were transcribed *in vitro*. Following this, the absolute copy number of the internal standard was quantified by spectrophotometry before addition into a sample prior to cell lysis and RNA extraction. Benchmarking of sequenced transcripts to the copy number of internal standard was used to estimate sequencing depth and absolute quantity of differential expressed gene or transcripttype. Similarly, DNA standards in known concentration (Satinsky et al., 2014b) or a biological agent (e.g., virus or bacterial species) in known quantity (e.g., titer or cell count) could be added into a sample prior to sample processing (i.e., cell lysis) and shotgun sequencing, so that the absolute quantity of metagenomics reads recovered from a sample can be quantified by benchmarking to the spiked-in internal control. These methods have been used to quantify gene and transcript abundance in microbial communities in the Amazon river plume (Satinsky et al., 2014a,b), bathypelagic marine bacterioplankton communities after Deep Horizon Oil Spill (Rivers et al., 2013) and costal water in the USA (Gifford et al., 2011).

The studies discussed above demonstrate the possibility of using NGS platforms for both quantification of the relative abundance of microbial genes or taxonomic groups in a sample or absolute quantification through the use of internal standards. Nevertheless, potential biases and uncertainties introduced by use of nucleic-acid based methods and NGS (especially shotgun amplicon sequencing) require in-depth consideration during data analyses and interpretation.

# Application of NGS to Analysis of Water Quality

To date, integrated multi-omics approaches encompassing genomics, metagenomics, metatranscriptomics and MPS of targeted genes (e.g., SSU rRNA genes) have been used to unravel the functions of microbial communities in a variety of freshwater (reviewed by Fondi and Liò, 2015; Franzosa et al., 2015) and marine environments (reviewed by Bourlat et al., 2013). For microbial water quality assessment, these multi-omics analyses have been used in combinations to investigate the microbial compositions and their ecological functions (e.g., dissemination of antibiotic genes and pollutant degradation) in the urban water cycle, such as lakes and rivers, engineered waste tailings ponds, sewage, water distribution systems and finished waters, amongst others (**Table 1**). These studies yield insights into the activities mediated by microbial communities in aquatic systems and potential biological risk factors in the form of specific microbial populations and their genes for virulence, toxin biosynthesis, or antibiotic resistance (AR), which can be important in microbial water quality management.

In this section we review recent studies that have leveraged NGS to improve upon the current understanding of factors mediating water quality in mainly freshwater environments. In particular, we will discuss ongoing work to (i) identify indicators of human sewage and human fecal contamination for water quality assessment (see Discovery of New Indicators for Human Sewage Contamination), (ii) examine the fate and transport of human pathogens in water and wastewater systems (see Relationship between Fecal Indicator Bacteria and Pathogen-Like Sequences and Microbial Safety of Drinking Water), (iii) understand the ecological drivers of harmful algal bloom persistence and toxin production/degradation (see Toxin Production and Degradation in Cyanobacterial Blooms), (iv) observe the spread and distribution of AR in freshwater environments (see Tracking Antibiotic Resistance through Metagenomics), and (v) investigate mechanisms of natural and stimulated biodegradation of harmful pollutants that threaten water quality (see Understanding Biodegradation of Pollutants that Threaten Water Quality). While NGS-based studies are providing unprecedented insights into the structure and function of waterborne microbial communities, an open challenge remains

to convert trends observed in these studies into actionable data for water quality managers.

### Discovery of New Indicators for Human Sewage Contamination

Contamination of freshwater bodies intended for human consumption or recreational use by fecal materials continues to be a major public health concern in many countries (Figueras and Borrego, 2010). New approaches are being developed to establish evidence of fecal contamination in surface freshwaters through alternative DNA-based indicators. MPS of the SSU rRNA gene has been used to characterize the microbial composition in raw sewage entering wastewater treatment plants in several different countries. Bacterial taxa associated with sewage infrastructure could be differentiated from those present in human fecal materials and other environmental sources (McLellan et al., 2010; Unno et al., 2010; Shanks et al., 2013; Cai et al., 2014). Comparison to a database of microbial sequences from feces in the Human Microbiome Project1 suggested that only ∼10–15% of sewage microbiomes in these surveyed sites were of human-fecal origin (McLellan et al., 2010, 2013; Shanks et al., 2013; Newton et al., 2015). Microbiomes of human-fecal origin detected in all sewage treatment plants appeared to be largely congruent and were represented by Firmicutes (e.g., *Lachnospiracea*, *Ruminococcaceae*; McLellan et al., 2013; Shanks et al., 2013), Bacteroidetes (e.g., *Bacteroidaceae, Porphyromonadaceae, Prevotellaceae*; Newton et al., 2015) and to a lesser extent by other mostly anaerobic microbes, e.g., *Bifidobacteriaceae*, *Coriobacteriaceae* (McLellan et al., 2010; Shanks et al., 2013; Newton et al., 2015). Amplicon sequences (V4–V5) obtained from municipal sewage communities and human stool samples identified 27 human fecal oligotypes (predominantly *Bacteroidaceae*, *Prevotellaceae*, or *Lachnospiraceae*/*Ruminococcaceae*) that are commonly and abundantly present in most surveyed sewage treatment facilities across the USA (Newton et al., 2015). The relative abundance of sequences within the dominant families of *Bacteroidaceae*, *Prevotellaceae*, or *Lachnospiraceae*/*Ruminococcaceae* could be statistically correlated to the obesity rate of the surveyed cities; with higher obesity rates corresponding to higher representation of *Bacteroidaceae* in the sewage microbiomes. This finding is particularly interesting as high representation of *Bacteroidaceae* in the human gut microbiome has been previously linked to consumption of a diet high in fat (David et al., 2014).

In contrast, a major proportion of the microbiome associated with sewage appeared to be adapted to the sewage infrastructure (∼80%) and varied in diversity and abundance according to geographical location (i.e., latitude) and air temperature (Vandewalle et al., 2012; Shanks et al., 2013; Newton et al., 2015). The microbiomes associated with sewers were also predominantly unique in taxonomy compared to those associated with animal hosts, surface freshwaters and other environmental sources (McLellan et al., 2010, 2013; Vandewalle et al., 2012; Shanks et al., 2013), and had in general higher diversity and

<sup>1</sup>http://hmpdacc*.*org/




 *waste water treatment plant; ARG, antibiotic resistance genes.*

**236**

species richness compared to stool samples (Newton et al., 2015). Such microbial species found in either human feces or sewage, but either absent, or present in very low abundance, in surface freshwaters, are potentially good indicators for sewage contamination (Koskey et al., 2014; McLellan and Eren, 2014). For example, development of qPCR probes based on *Lachnospiraceae* sequences enriched in human sewage were used to evaluate sewage contamination in an urban freshwater harbor where occurrence of this new indicator group was highly correlated with a more established marker of human fecal contamination (i.e., HF183; Newton et al., 2011). Analysis of profiles of microbial SSU rRNA genes (V4/V6 regions) with a source estimation program employing Bayesian statistics (SourceTracker) allowed identification of human fecal and sewage signatures which correlated to the distribution of humansource markers Lachno2 and HF183 along the coastline of Lake Michigan (Newton et al., 2013). Use of bacterial taxonomic groups identified through NGS-based surveys as alternative indicators may be site-specific as several studies have shown that the microbial composition in different sewer systems can differ according to several instances: origin of the waste materials (Ye and Zhang, 2013), climatic variations due to latitudinal locations of treatment facilities (Shanks et al., 2013), or infiltration of rainwater and storm water inputs (McLellan et al., 2010).

Collectively, studies using MPS of the SSU rRNA gene have shown that wastewater treatment is, in general, capable of removing most microbial populations associated with humanfeces (Ye and Zhang, 2011; Cai et al., 2014). Even so, a small proportion of potential pathogens may still be present in sewer effluents (Ye and Zhang, 2011; Cai et al., 2014). In areas without widespread sewage treatment, direct contamination of waters with human waste is still a problem. The cited studies have shown that assessment of waterbodies using NGS can reveal impacts from sewage and fecal contamination. In addition, with advancement in high throughput sequencing technology and bioinformatics, surveys of the sewage microbiome may 1 day be used to assess and monitor the overall health status of human populations.

# Relationship between Fecal Indicator Bacteria and Pathogen-Like Sequences

While NGS has shown potential for tracking water impairment through discovery and detection of indicators of human fecal or sewage pollution, a more direct approach to evaluate waterborne biological risks is through evaluation of pathogen diversity and abundance. However, tracking individual waterborne pathogens is costly, methodologically challenging, and requires knowledge of which pathogens to target (Varela and Manaia, 2013). The use of NGS to screen for sequences with high identity to waterborne pathogens allows identification of potential risk agents and, depending on biases in sample preparation, may also provide semi-quantitative inference of their relative abundance in wastewater and other environments (McLellan et al., 2010; Cai and Zhang, 2013; Ibekwe et al., 2013; Cai et al., 2014; Lu et al., 2015). Identification of pathogen species was not possible in early studies that were based on analysis of the V6 region of the SSU rRNA gene due to short sequence length (∼60 bp) which failed to

provide taxonomic resolution beyond the family or genus level. However, as sequence read lengths through NGS continues to increase, confident classification of NGS-generated sequences to the level of bacterial species has become attainable.

A recent study of bacterial pathogens in a wastewater treatment plant applied a combination of NGS and qPCR of genetic markers to track the occurrence of bacterial pathogens through stages of wastewater treatment process (Lu et al., 2015). This analysis revealed that raw sewage was enriched in potential pathogens most closely related to *Arcobacter butzleri*, *Aeromonas hydrophila,* and *Klebsiella pneumonia.* The *Arcobacter* genus represented over 43.5– 97.37% of all pathogen-like sequences in the treatment plant (sewage influent, primary effluent, activated sludge, secondary effluent and final sand filter effluent) while the effluent contained only *Arcobacter butzleri* with other non-detected species presumed to have been removed mainly by biological processes during treatment. Quantification of genetic markers with qPCR confirmed the trends in the distribution of pathogenic species identified by sequencing in both raw and treated sewage (Lu et al., 2015). In a previous study of sewage treatment plants across nations (Canada, USA, China, and Singapore), 454 pyrosequencing revealed that the most abundant sequences related to pathogenic bacteria in raw sewage corresponded to the genera *Aeromonas* and *Clostridium* with species *Aeromonas veronii, Aeromonas hydrophila, Clostridium perfringens,* and *Corynebacterium diphtheria* (Ye and Zhang, 2011).

Several studies have employed NGS to characterize the distribution of potential pathogens in natural and humanimpacted areas. Nine sites sampled in urban and agricultural watersheds along the Santa Ana River, CA, USA were screened for potential bacterial pathogens by pyrosequencing of the V1 and V2 hypervariable regions of the SSU rRNA gene (Ibekwe et al., 2013). Sequences related to human pathogens comprised a greater percent of the sampled microbial community in urban runoff and agricultural waters than in waters collected from sites with little to no human activity. Similarly, a study of an urban tropical area in Singapore used Illumina MiSeq (V3 and V4 hypervariable regions of the SSU rRNA gene) to evaluate the variation of bacterial communities as a function of land use and water quality (Nshimyimana et al., 2015). The relative abundance of pathogen-like sequences identified to genera and species varied with land use and with the abundance of FIB, where moderate positive correlation to measured levels of *E. coli* suggested higher incidence of potential pathogens in sites with higher measured levels of FIB.

Despite the capability for simultaneous detection of multiple human bacterial pathogens in environmental waters, NGS has not yet been integrated into a framework of QMRA, in part because, as discussed in Section 1, (1) amplificationbased biodiversity screens are qualitative, although studies of mock communities indicate that these may provide tentative indications of relative abundance, (2) identification of pathogens to species-level has only recently been possible with longer reads afforded by NGS, and (3) species-level identification may not be sufficient to assess risk from pathogens with strain-specific virulence.

Recent studies have attempted to monitor human sewage to track outbreaks of disease-causing viruses. While published studies have not yet integrated NGS, such methods could complement conventional approaches such as PCR, qPCR, and Sanger sequencing for the identification and quantification of pathogens. This method combination was recently employed for analysis of eight viral strains in sewage in Gothenburg, Sweden and led to an early warning of a potential outbreak of Hepatitis A and norovirus (Hellmér et al., 2014) which were also detected in an ongoing outbreak in Scandinavia (Hellmér et al., 2014). Clinical research and diagnostic studies in Africa also combined three methods (Sanger sequencing, PCR, and qPCR) to understand the spread and distribution of the polio virus and to discern wild-type from vaccine-derived strains (Gumede et al., 2013). Future studies have the potential to apply NGS to screen for and identify pathogens that could subsequently be monitored by targeted diagnostic methods such as PCR and qPCR.

Reviewed studies demonstrate that NGS and related sequencebased methods are being used to track human bacterial pathogens and/or waterborne viral strains. Monitoring the distribution of viral strains in the environment shows promise to improve vaccination campaigns and provide early warnings of disease outbreaks in a given community. NGS profiling of bacterial pathogens in the natural and built environment provides general insights into how the distribution and relative representation of potential human pathogens varies with environmental conditions, anthropogenic impacts, and the implementation of water treatment technologies. Such information may be leveraged to guide engineers and watershed managers in selection of best practices to reduce human exposure to potential pathogens.

### Microbial Safety of Drinking Water

Next-generation sequencing has been used to survey the microbial composition in drinking water distribution systems including source waters (Chao et al., 2013), end-point taps (Chao et al., 2013; Shi et al., 2013; Huang et al., 2014), and various stages of the drinking water treatment and distribution process (Gomez-Alvarez et al., 2012; Chao et al., 2013; Shi et al., 2013; Huang et al., 2014), as well as biofilm matrices associated with distribution pipelines (**Figure 1A**; Hong et al., 2010). While most microorganisms in treated drinking water are harmless, outbreaks of diseases linked to pathogens in drinking water may be due to compromised water treatment and contaminated source waters (MacKenzie et al., 1994; Howe et al., 2002). Pathogens that have evaded treatment and disinfection processes may persist in water distribution pipelines as biofilms, leading to dissemination to end users through the process of sloughing (Figueras and Borrego, 2010). A recent study noted an increase in microbial diversity in end-point drinking water relative to source water which was linked to dispersal of biofilm-associated microbes during passage through the water distribution pipeline (Huang et al., 2014). Consumption of untreated well-water poses an

FIGURE 1 | Environments where studies of water quality have been advanced by NGS. (A) Fluorescence microscopy showing biofilm grown on a stainless steel surface in a laboratory potable water biofilm reactor for 14 days (image from Donlan, 2002). Drinking water is thought to contain a mixture of both planktonic and dissociated biofilm bacteria. Using NGS methods, a repertoire of microbial species including potential pathogens have been detected in water distribution systems, raising the concern of the effectiveness of water treatment on microbial water safety. (B) A water reservoir in Singapore impacted by a cyanobacterial bloom characterized by Penn et al. (2014). With global warming, cHAB is predicted to occur more regularly with likely severe consequences (e.g., toxin loading) that can diminish water quality. (C) An oil sands tailings pond in Alberta, Canada characterized by Tan et al. (2015). Inset (D) shows microcosms established to study degradation of unrecovered hydrocarbons deposited in oil sands tailings pond. NGS has been used to characterize the microbial communities in oil sands tailings ponds (http://www*.*hydrocarbonmetagenomics*.*com/) in order to better understand mechanisms of pollutant degradation.

additional risk, where waterborne pathogens such as *Legionella* and *Campylobacter* are among some of the main well-waterborne disease agents in the USA (CDC, 2013). Other causative agents of waterborne diseases in both developed and developing countries include parasites such as *Cryptosporidium parvum*, *Toxoplasma gondii*, *Cyclospora cayetanesis*, *Giardia lamblia* and viruses such as norovirus (Ashbolt, 2015). Metagenomic surveys of water within distribution systems and end-point drinking waters have detected DNA indicative of several of these potential opportunistic pathogens (e.g., *Legionella*, *Mycobacterium*, *Pseudomonas,* and *Leptospira*) and virulence factors (e.g., AR and pathogenicity islands, Gomez-Alvarez et al., 2012; Shi et al., 2013; Huang et al., 2014). Several recent studies have used NGS to shed light on the fate of microbial populations, including pathogens, during various stages of the water treatment process.

Microbial communities in drinking water systems tend to be present at low abundance (e.g., *<*10<sup>5</sup> cells per ml, personal observation) hence studies of microbial communities in drinking water require collection of microbial biomass from large amounts of water, e.g., 100–2000 L (Gomez-Alvarez et al., 2012; Chao et al., 2013; Shi et al., 2013; Huang et al., 2014) where DNA extraction yields may still be too low for direct sequencing, requiring amplification (of genomes or amplicons) prior to sequencing (Gomez-Alvarez et al., 2012). Studies of water distribution systems reveal microbial communities dominated by the phyla Proteobacteria (i.e., Alpha-, Beta-, and Gammaproteobacteria), Firmicutes, Nitrospirae, and Actinobacteria (Gomez-Alvarez et al., 2012; Chao et al., 2013; Huang et al., 2014). Application of disinfectants (e.g., free chlorine or monochloramine) appear to have selective effects on the microbial assemblages in water distribution pipelines and end-point drinking water (Gomez-Alvarez et al., 2012; Chao et al., 2013; Huang et al., 2014). In water treated with monochloramine collected from a water distribution simulator in the USA, the microbial composition based on taxonomic assignment of 454 reads, was dominated by Actinobacteria (28%; e.g., *Mycobacterium*), Betaproteobacteria (25%; e.g., *Acidovorax*), and Alphaproteobacteria (23%), whereas freechlorine treated water had a lower proportion of Actinobacteria (6%), and was dominated largely by Alphaproteobacteria (35%; e.g., *Caulobacter*, *Rhodopseudomonas, Bradyrhizobium*) and Cyanobacteria (*Synechococcu*s; Gomez-Alvarez et al., 2012). Chlorination appeared to selectively deplete Gamma- and Betaproteobacteria, resulting in higher relative abundance of Alphaproteobacteria in drinking water (Chao et al., 2013; Huang et al., 2014). Relative to untreated water, treatment with chlorine was found to reduce waterborne microbial diversity (Huang et al., 2014). Importantly, chlorination was associated with the removal of most potential pathogens from the detected diversity (Huang et al., 2014) although genomic signatures of some potential pathogens (e.g., *Pseudomonas aeruginosa* and *Leptospira interrogans*) persisted in the tap water (Huang et al., 2014). Protective functions for detoxification and modulation of oxidative stress (Chao et al., 2013), as well as mobile genetic elements (MGEs; Shi et al., 2013) were also found to be more highly represented in metagenomes obtained from

water following disinfection (Shi et al., 2013; Huang et al., 2014).

### Toxin Production and Degradation in Cyanobacterial Blooms

Cyanobacterial harmful algal blooms (cHABs) are destructive to aquatic ecosystems, deteriorating surface water quality and rendering it unsafe for use by humans and livestock. cHABs are associated with overgrowth of cyanobacterial species many of which are capable of producing cyanotoxins, e.g., microcystin, cylindrospermopsin, anatoxin (Ferrão-Filho and Kozlowsky-Suzuki, 2011) and off-flavor/odorous compounds, e.g., geosmin or 2-methylisoborneol (Li et al., 2012). Rising global temperature as a result of climate change is expected to promote the frequency and severity of cHAB events in the future (Paerl and Huisman, 2009). NGS technologies are being applied to the systematic understanding of the ecology and control of cHAB using methods in metagenomics, comparative genomics, and metatranscriptomics (Li et al., 2011; Steffen et al., 2012, 2015; Penn et al., 2014).

A cyanobacterial bloom consists of both primary producers and heterotrophs including consumers and grazers, resulting in nutrient cycling and recycling. It is likely that different populations of microbial heterotrophs may influence the dynamics of a cHAB and may benefit from nutrients and metabolites released during a cHAB event (Edwards and Lawton, 2009; Mou et al., 2013), or consumption of dead cyanobacterial biomass after a bloom die-off (Xing et al., 2011). The interplay of this complex ecosystem shapes the bloom community structure and may control the adverse effects of the bloom (e.g., bloom persistence and the balance between production and biodegradation of toxins and offflavor/odor compounds). Though this interplay has long been hypothesized (Christoffersen et al., 1990), detailed insights from community gene expression can link specific populations, or groups of populations, to specific activities via metagenomic and metatranscriptomic approaches. In contrast to oligotrophic lakes where phototrophic Proteobacteria and Actinobacteria appear to be the most active light harvesters (Yurkov and Beatty, 1998; Sharma et al., 2008), in a cHAB, which generally manifests during eutrophic conditions, solar energy is captured primarily by algae and cyanobacteria. Metagenomic studies conducted during bloom events in Lake Taihu (China), Lake Erie (N. America), and Grand Lake St. Marys (GLSM; OH, USA) revealed a high proportion of cyanobacterial sequences (25– 88%) including common bloom members *Chroococcales* (e.g.*, Cyanothece*, *Synechococcus*, *Crocosphera;* ca. 40–45%), *Nostocales* (ca. 10%), and *Oscillatoriales* (e.g., *Lyngbya*, *Trichodesmium*like*, Arthrospira*, etc; ca. 17–38%; Steffen et al., 2012). Metatranscriptomics studies of other freshwater systems suggest that the most active cyanobacterial taxa were also from the same cyanobacterial orders, dominated by *Microcystis* in an eutrophic Singaporean reservoir (Penn et al., 2014; **Figure 1B**) and in Lake Erie with co-dominant taxa including *Synechococcales* and *Gloeobacterales* (Steffen et al., 2015). Several recent omics studies showed that the non-cyanobacterial members of cHAB communities were dominated by Proteobacteria (*>*90% of non-cyanobacterial sequences) with a smaller portion of Bacteroidetes (Li et al., 2011; Steffen et al., 2012; Penn et al., 2014). Eukaryotic algal species (e.g., *Streptophyta*, *Euglenids*, *Chlorophyta*, *Bacillariophyta*) were also observed within blooms, and may co-dominate, as revealed using MPS of the plastid 23S rRNA gene, specific for cyanobacterial and eukaryotic algal species (Steven et al., 2012). In addition, viruses/phage, protozoa, and fungi are present and are increasingly recognized as key players in bloom persistence and decay (Gerphagnon et al., 2013; Xia et al., 2013; Ger et al., 2014).

Studies based on NGS have provided unprecedented insights into the evolution of cyanobacterial genomes, pointing to the role of nutrient-enrichment as a factor that can accelerate genome evolutionary rates and potential niche expansion of cyanobacterial species. Comparative genomic analyses reveal signatures of genome rearrangement through both homologous and non-homologous recombination (Frangeul et al., 2008; Humbert et al., 2013; Yang et al., 2015) with such variation reflected in natural bloom populations. Comparison of the genome of *Microcystis aeruginosa* strain NIES 843 to metagenomes from Lake Erie, Lake Taihu, and GLSM identified metagenomic islands (Mi's) within the NIES 843 genome (defined as regions of ≥10 kb with low coverage in the metagenomes) that were not observed in these three sites and which contained transposase and MGEs (Steffen et al., 2012). Variation in toxin biosynthesis gene content among closely related cyanobacterial strains is well-documented (Mikalsen et al., 2003; Tanabe et al., 2004; Christiansen et al., 2008) and is linked to genomic dynamism mediated by homologous and non-homologous recombination. Transposase expression in strain NIES 843 has been shown to be upregulated by nutrient-enrichment, particularly in the presence of organic nitrogen (urea; Steffen et al., 2014). Thus, conditions of eutrophication may promote genome rearrangement within this group, contributing to the mosaic patterns in gene order, MGE content and the distribution of toxin genes observed in these strains. The dynamic genomes of cHAB strains may influence their fitness, promoting local adaptation and contributing to the widespread distribution of closely related strains (Mikalsen et al., 2003; Yoshida et al., 2008; Acinas et al., 2009).

It is of interest to water quality managers to know which waterborne microbial populations are responsible for the production and biodegradation of toxins and off-flavor/odorous compounds. Transcripts for genes in the biosynthesis pathways for multiple cyanotoxins (i.e., microcystin, aeruginosin, and cyanopeptolin), and structurally related compounds including a newly identified secondary metabolite gene cluster were detected throughout a 24 h sampling campaign in Singapore, suggesting that toxin gene expression was continuous in this system, despite dissolved toxin levels measured near the limit of detection (Penn et al., 2014). Biosynthesis and release of toxins may be balanced by their biodegradation. Biodegradation of microcystin by natural microbial communities from Lake Erie was studied through metagenomic analysis of mesocosms (Mou et al., 2013). The authors found *Methylophilales* and *Burkholderiales* were significantly enriched in microcystin-amended microcosm, while the gene encoding for the only known mechanism of microcystin biodegradation (mlr) was not, prompting the authors to speculate that new, undiscovered pathways for microcystin biodegradation were being utilized. As additional omics-enabled studies with a focus on water quality emerge, it is likely that new pathways for the production and biodegradation of such compounds will be discovered, paving the way for better strategies to control their expression and improve source water quality.

### Tracking Antibiotic Resistance through Metagenomics

Antibiotic resistance in pathogenic bacteria is a growing public health threat (Berendonk et al., 2015). Antibiotic resistance determinants (ARDs) including antibiotic resistance genes (ARGs), and MGEs that catalyze the transfer of such genes, occur widely in environmental bacteria. ARD appear to be enriched in microbial communities including sediment microbiota near wastewater treatment plant effluent outfalls (Czekalski et al., 2014; Port et al., 2014), sediments of humanimpacted estuaries (Chen et al., 2013) and in a variety of sample types (e.g., soil, water, sediments, human and animal fecal samples, sludge, wastewater) derived from heavily anthropogenic impacted environments (Li et al., 2015). Although environmental surveillance of AR is not currently part of water quality monitoring frameworks, there is an urgent need to better control the spread and evolution of AR and a better understanding of the occurrence and characteristics of environmental reservoirs of ARD will advance that goal (Bush et al., 2011).

To gain a better understanding of the types of ARGs in environmental bacteria and their co-localization with MGEs (e.g., transposons, plasmids, and integrons) complementary approaches of sequence-based metagenomics and functional metagenomics has overcome the impediments of earlier culturing and molecular techniques. With sequence-based metagenomics, total DNA from an environment is directly extracted and randomly sequenced. The sequenced reads are then interrogated against a reference database containing known ARG sequences to predict the resistance potential originating from a metagenome. Functional metagenomics involves cloning randomly sheared DNA fragments into an expression vector and transforming them into a host (e.g., *E. coli*) and selecting transformants which exhibit resistance to the selected antibiotic (Allen et al., 2009; Sommer et al., 2009). The advantages of this dual approach include (i) the ability to identify highly divergent genes from known ARGs; (ii) direct evidence of resistance phenotypes associated with expressed genes and (iii) no reference gene sequences required for gene identification (Pehrsson et al., 2013). Combining sequence based metagenomics and functional metagenomics has led to the discovery of novel ARGs in various microbial communities including soil (Riesenfeld et al., 2004; Allen et al., 2009; Torres-Cortés et al., 2011), freshwater lakes (Bengtsson-Palme et al., 2014), human gut microbiomes (Sommer et al., 2009; Hu et al., 2013), oral microbiome (Diaz-Torres et al., 2006), animal gut microbiomes (Kazimierczak et al., 2009) and activated sludge (Mori et al., 2008; Parsley et al., 2010).

The growing sophistication, accuracy and speed in downstream processing pipelines of NGS datasets provides a significant step forward in facilitating large-scale environmental studies to assess the emerging threat imposed by AR. With the inundation of metagenomic sequence data over the past decade from variety of environmental microbiomes (e.g., IMG, MG-RAST, Sequence Read Archive), metagenomic analysis tools have been developed to mine larger and more expansive reference sequence databases for ARG signatures. These tools include the Comprehensive Antibiotic Resistance Database (CARD2 , McArthur et al., 2013), the Antibiotic Resistance Database (Liu and Pop, 2009), the beta-lactamase database (BLAD, Danishuddin et al., 2013), and ResFinder3 . To determine the relationship between environmental and human-associated resistomes (i.e., antibiotic resistance genes and bacteria), Gibson et al. (2015) developed Resfams, a curated protein family database and associated profile hidden Markov models (HMMs), organized by ontology specifically applied to AR functions, with a subset of these AR proteins functionally verified using protein assays. This method circumvents the common approach of assigning AR functions using pairwise sequence alignment to AR databases, instead, HMM and consensus models were used for AR functional assignment, which significantly increases prediction sensitivity and specificity.

Several recent studies have employed NGS and bioinformatic tools, such as those described above, to characterize the AR profiles of microbial communities in aquatic environments. A network analysis was conducted to investigate the broad-spectrum profiles of ARGs and their co-occurrence patterns in 50 samples spanning water, soil, sediments, wastewater, sludge, and human and fecal samples. This study concluded that the abundant ARGs were associated with antibiotics commonly administered in human or veterinary medicine (i.e., aminoglycoside, bacitracin, betalactam, chloramphenicol, macrolide-lincosamide-streptogramin, quinolone, sulphonamide, and tetracycline) and abundances of these ARGs were up to three magnitudes higher in the most heavily anthropogenic-impacted environments (Li et al., 2015). Similarly, metagenomic profiles of ARGs in sediments of a heavily human impacted estuary [Pearl River Estuary (PRE) in China] revealed a higher diversity of both genotypes and resistance genes for sulphonamides, fluoroquinolones, and aminoglycosides relative to a pristine deep ocean bed in the South China Sea (SCS; Chen et al., 2013). In addition, this study showed parallel trends between the distribution of ARGs and MGEs, where MGE's may function as vectors for dissemination of ARGs in the aquatic environment.

Antibiotic resistance is a public health threat of heightened concern and the central theme of current global monitoring efforts is limited to tracking antibiotic consumption and antibiotic resistant bacteria (ARBs) isolated from clinical and public health laboratories (Grundmann et al., 2011). While clinical settings may be the source of highest antibiotic and ARB loads, the natural environment has recently drawn attention as a reservoir of transferable ARGs potentially implicating environments highly contaminated with ARGs as a human health risk (Ashbolt et al., 2013). ARGs in particular are

2http://arpcard*.*mcmaster*.*ca/

3https://cge*.*cbs*.*dtu*.*dk//services/ResFinder/

increasingly viewed as emerging pollutants and emphasis of their occurrence and distribution in natural environments are being considered in development of antibiotic surveillance frameworks (Codex Alimentarius Commission, 2011; Berendonk et al., 2015). Incorporating metagenomics into frameworks for monitoring the threat of AR in aquatic environments provides a novel approach for environmental health monitoring and pushes boundaries on improving current risk assessment models. Recently, a metagenomic-based approach was used to develop a multivariate index to quantify the AR potential in published metagenomes (Port et al., 2014). This index was based on the abundance of ARGs, MGEs, pathogenic potential and metal resistance genes (implicated in co-selection of ARGs). Congruent with studies examining the distribution of ARG, the multivariate index was shown to differentiate aquatic environments based on humanimpact. This study exemplifies the utility of NGS to advance characterization of water quality within the context of AR, which will ultimately assist in risk evaluation and mitigation by public health authorities.

### Understanding Biodegradation of Pollutants that Threaten Water Quality

Chemical pollution is one of the major causes of diminished water quality (Schwarzenbach et al., 2010). Strategies for pollutant removal through stimulation of indigenous microbial community for bioremediation can be effective to remove, or immobilize, toxic compounds (Major et al., 2002; Gieg et al., 2014; Koenig et al., 2014). The presence of pollutants can exert selective pressures that enrich for microbial populations that are capable of coping with associated stresses and/or are able to utilize chemical contaminants as a source for carbon, nutrients, or as an electron acceptor for respiration (Hemme et al., 2010; Smith et al., 2012; An et al., 2013a,b; Rivers et al., 2013; Tan et al., 2015). Recent work in the area of marine metagenomics has demonstrated that the enrichment of microbial genes in distinct environments reflects the key biogeochemical processes in these systems (Coleman and Chisholm, 2010; Ulloa et al., 2012; Kelly et al., 2013). Similarly, microbial communities and gene expression profiles in polluted waters reveal pathways for transformation of contaminants, thus serving as an indicator of *in situ* biodegradation processes and highlighting metabolic pathways to target and possibly stimulate for accelerated pollutant clean up (Kimes et al., 2013; Engel and Gupta, 2014; Lamendella et al., 2014; Mason et al., 2014).

Analysis of microbial populations and genes enriched in polluted sites using "-omics" enabled approaches has revealed key microbial processes involved in contaminant tolerance and transformation. For example, in an underground water system heavily polluted by heavy metals, nitric acid and organic solvents, the microbial community surveyed through metagenomics showed limited diversity consisting mainly of Beta- and Gammaproteobacteria, with streamlined metabolic capabilities including the capacity to utilize heavy metal ions for lithotrophy (Hemme et al., 2010). A similar link between microbial gene content and contaminant transformation is observed in systems contaminated with hydrocarbons. In the highly anaerobic environments of oil sands tailings ponds constructed to store waste tailings from bitumen extraction, metagenomic data identified a repertoire of facultative and strict anaerobes that were capable of utilizing the pollutant hydrocarbons as a carbon source (**Figure 1C**; An et al., 2013a; Tan et al., 2015). Similarly, NGS surveys of open oceans and coastal shorelines in Gulf of Mexico after Deep Horizon Oil spill in 2010 showed that within the water column, a microbial community consisting of mainly marine Gammaproteobacteria (e.g., *Colwelliaceae*, *Oceanospirillaceae*, *Piscirickettsiaceae*, *Methylococcaceae*) dominated the oil plumes, and were the key players in the aerobic degradation of oil and gas (Mason et al., 2012; Rivers et al., 2013). MPS of SSU rRNA genes from microbial communities originating from beach areas before and post-oil impact indicated that the community shifted from a baseline of predominantly enteric-type consortia associated with human-impact to marine-associated taxa (e.g.*, Oceanospirillales*, *Rhodospirillales*, and *Rhodobacterales*) linked to remediation efforts (e.g., sand washing), including populations with potential oil degradation capability (Engel and Gupta, 2014). Oil plumes have been shown to have an impact on microbial communities in marine sediments and other beach areas, with NGS detecting genes and transcripts indicative of pollutant (monoaromatics and alkanes) degradation in marine sediments (Kimes et al., 2013; Mason et al., 2014) and areas along the shorelines (Lamendella et al., 2014).

Several studies have used microcosms and enrichment cultures inoculated from contaminated sites (i.e., polluted groundwater or tailings ponds containing toxic waste) to establish evidence of pollutant degradability, followed by NGS studies incorporating gene expression (i.e., metatranscriptomics) or comparative metagenomics approaches (Hug et al., 2012; Waller et al., 2012) in order to identify (novel) key processes relevant to pollutant degradation (e.g., Abu Laban et al., 2010; Hug et al., 2012; Luo et al., 2014; **Figure 1D**). NGS in tandem with other functional analyses (e.g., proteomics and transcriptomics) have been employed to successfully identify novel genes and microbes involved in anaerobic activation and degradation of recalcitrant compounds, e.g., carboxylation of benzene (Abu Laban et al., 2010; Luo et al., 2014), dechlorination of chlorinated ethene (Hug et al., 2012; Waller et al., 2012) and fumarate addition activation of hydrocarbons (Tan et al., 2015). Comparative metagenomics and genomecentric metagenomics have been used to tease out the functional roles of different members in complex community in bioreactors (Albertsen et al., 2013; Haroon et al., 2013). Microbially mediated processes, such as dechlorination of chlorinated compounds often requires the presence of a mixed community to provide supporting roles such as production of essential nutrients, e.g., H2, corrinoid, and methionine that are needed by primary degraders (e.g., Dehalococcoidetes) in substrate mineralization. Hug et al. (2012) compared the metagenomes of three enrichment cultures involved in the dechlorination of chlorinated ethene, and identified genes required for the production of nutrients essential for Dehalococcoidetes in dechlorination. Assigning these gene to taxa using a homology-approach; revealed a mutualistic relationship between a repertoire of microorganisms (e.g., Firmicutes, methanogens, and Deltaproteobacteria), which is not easily achieved using conventional microbiology methods involving isolation. The functions of key members involved in the degradation of pollutants could also be inferred through targeted genome enrichment using strategies such as stable-isotope probing (SIP), in which labeled recalcitrant compounds can be used to enrich for key degraders, followed by (meta)genome sequencing and analysis allowing investigation of the genetic capabilities or metabolic modeling of uncultivated pollutant degraders (e.g., Saidi-Mehrabad et al., 2013).

# Future Technologies on the Horizon and Impact for Water Quality Assessment

Next-generation sequencing platforms for single molecule realtime sequencing technology (SMRT; e.g., Pacific Biosciences and Oxford Nanopore Technologies) are capable of producing sequences with read lengths exceeding 1000 bp, surpassing those generated using Illumina and Roche 454. In addition, real-time data acquisition make these technologies attractive targets for integration into sensors for monitoring environmental sequence data. To date, these technologies have not been widely adopted for surveys of highly diverse bacterial communities due to high rates of randomly distributed sequencing errors, which would lead to artificially inflated community diversity (Schloss et al., 2015). However, recent studies employing SMRT sequencing with improved sequencer hardware and chemistry have yielded improved accuracy and read lengths (i.e., *>*1400 bp) making community surveys with amplicon sequencing more feasible (Marshall et al., 2012; Mosher et al., 2014). To date, Pacific Bioscience (PacBio) sequencing has been used extensively for genome sequencing, where the high coverage and long sequence read length allows for generation of highly contiguous consensus sequences with low error rates, where microbial genomes can often be closed within a single run (Koren et al., 2013). PacBio sequencing is often used in parallel with other NGS platforms (e.g., Illumina, Roche 454, and SoLiD) to allow for scaffolding and phishing to produce finished/close genomes with high sequence quality (Koren et al., 2013). MinION, a newer SMRT based on Nanopore technology (Venkatesan and Bashir, 2011), has garnered a lot of interest due to the cost-effectiveness and pocket-size mobility. Hundreds of units were shipped to members of the MinION Access Program to test the device in a wide range of sequencing applications. Sequencing results produced by some of the laboratories showed that MinION could generate single read lengths of up to 5500 bp in a single run, though riddled with a high sequence error rate (∼30%; Ashton et al., 2014; Mikheyev and Tin, 2014; Madoui et al., 2015). Nanopore sequencing holds particular promise for online detection of waterborne nucleic acids due to its high potential portability and real-time data output, allowing future work to develop real-time sensing platforms for water quality monitoring. Indeed, as different NGS technologies advance toward enhanced sequence chemistry for improved sequence read length with higher throughput and reduced error rate, online real-time detection using NGS will breach the gaps for its use in real-time monitoring of genetic parameters for water quality.

The synergy between multiple NGS platforms and the increased accuracy and data processing speed of downstream bioinformatics, biostatistical and machine-learning pipelines continue to push the boundaries of genomic and metagenomic data analyses. Advancements in the field of multi-omics have enabled optimization of bioinformatics workflows (Kozich et al., 2013; Nelson et al., 2014); allowing acquisition of uncultivated microbial genomes from a complex metagenome (Huson et al., 2007; Dick et al., 2009; Hess et al., 2011; Albertsen et al., 2013; Haroon et al., 2013); deduction of microbial community composition based on single copy genes (Sunagawa et al., 2013); inference of metabolic interactions among microbial community members (Hanson et al., 2014); identification of phylogenetic markers and genes encoding important processes (Stark et al., 2010; Darling et al., 2014), and from a broader public health aspect, have augmented our knowledge of the population genomics and dissemination of virulent strains of waterborne pathogens such as *Vibrio cholerae* (Dutilh et al., 2014).

The influx of newly sequenced microbiomes associated with various natural and engineered water environments progressively expand sequence information available in public databases such as MG-RAST (Meyer et al., 2008) and IMG (Markowitz et al., 2014), providing a bigger data pool for mining and comparative metagenomic analyses. This is evident in ecological studies of freshwater microbiomes where taxonomic composition and functions of waterborne microbial communities and pathogens are dependent upon *a priori* databases. In addition, integrating physiological methods through use of microcosms, enrichment cultures and single cell genomics, coupled with NGS has allowed the discovery of new enzymes and genes in complex microbial communities that are important in water quality preservation

### References


(e.g., toxin degradation, nutrient cycling), which otherwise would be overlooked since many of these microbes have remained uncultivated.

Microbial water quality assessment through NGS-based molecular detection of FIB, pathogens, or virulence factors, as well as genes encoding biogeochemical processes such as pollutant biodegradation, have the potential to be translated into actionable data for water quality managers. However, crucial relationships between the occurrence, detection and quantification of nucleic acids in the environment through NGS-based approaches and potential impacts to human or environmental health must be established before profiles based on the distribution(s) of microbial genes are relied upon to replace currently used bioindicators. Current frameworks for water quality assessment heavily rely upon proxy measurements (e.g., quantification of culturable FIB) with a deep literature substantiating such approaches with epidemiological data. As the field of NGS-based water quality assessment matures, such further studies will be needed to establish whether proposed NGS-based indicators for pollution can improve upon the existing state of the art in water quality assessment.

### Acknowledgments

This research was funded by the National Research Foundation Singapore through the Singapore MIT Alliance for Research and Technology's (SMART) Center for Environmental Sensing and Modeling (CENSAM) research program. The authors would like to thank Professor Emerita Julia Foght (University of Alberta) for providing **Figures 1C,D** in the main text.


health status. *Mar. Pollut. Bull.* 74, 19–31. doi: 10.1016/j.marpolbul.2013. 05.042


cultures: the role of the non-dechlorinating community. *BMC Genomics* 13:327. doi: 10.1186/1471-2164-13-327


cultures. *Water Res.* 46, 5165–5173. doi: 10.1016/j.watres.2012.0 6.008


and impact on microcystin production in *Microcystis* strains. *J. Bacteriol.* 185, 2774–2785. doi: 10.1128/JB.185.9.2774-2785.2003


**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 Tan, Ng, Nshimyimana, Loh, Gin and Thompson. 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.*