# DENDRITIC CELL AND MACROPHAGE NOMENCLATURE AND CLASSIFICATION

EDITED BY: Florent Ginhoux, Martin Guilliams and Shalin Naik PUBLISHED IN: Frontiers in Immunology

#### *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 978-2-88919-918-1 DOI 10.3389/978-2-88919-918-1

## 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

# **DENDRITIC CELL AND MACROPHAGE NOMENCLATURE AND CLASSIFICATION**

#### Topic Editors:

**Florent Ginhoux,** Singapore Immunology Network (SIgN), A\*STAR, Singapore **Martin Guilliams,** Ghent University, Belgium **Shalin Naik,** The University of Melbourne, Australia

The blue background represents a cluster of human plasmacytoid dendritic cells, coloured scanning electron micrograph (SEM), and the purple foreground represents a myeloid CD1c dendritic cell.

Image by Benoit Malleret.

The mononuclear phagocyte system (MPS) comprises dendritic cells (DCs), monocytes and macrophages (MØs) that together play crucial roles in tissue immunity and homeostasis, but also contribute to a broad spectrum of pathologies. They are thus attractive therapeutic targets for immune therapy. However, the distinction between DCs, monocytes and MØ subpopulations has been a matter of controversy and the current nomenclature has been a confounding factor.

DCs are remarkably heterogeneous and consist of multiple subsets traditionally defined by their expression of various surface markers. While markers are important to define various populations of the MPS, they do not specifically define the intrinsic nature of a cell population and do not always segregate a bona fide cell type of relative homogeneity. Markers are redundant, or simply define distinct activation states within one subset rather than independent subpopulations. One example are the steady-state CD11b+ DCs which are often not distinguished from monocytes, monocyte-derived cells, and macrophages due to their overlapping phenotype. Lastly, monocyte fate during inflammation results in cells bearing the phenotypic and functional features of both DCs and MØs significantly adding to the confusion. In fact, depending on the context of the study and the focus of the laboratory, a monocyte-derived cell will be either be called "monocyte-derived DCs" or "macrophages". Because the names we give to cells are often associated with a functional connotation, this is much more than simple semantics. The "name" we give to a population fundamentally changes the perception of its biology and can impact on research design and interpretation.

Recent evidence in the ontogeny and transcriptional regulation of DCs and MØs, combined with the identification of DC- and MØ-specific markers has dramatically changed our understanding of their interrelationship in the steady state and inflammation. In steady state, DCs are constantly replaced by circulating blood precursors that arise from committed progenitors in the bone marrow. Similarly, some MØ populations are also constantly replaced by circulating blood monocytes. However, others tissue MØs are derived from embryonic precursors, are seeded before birth and maintain themselves in adults by self-renewal. In inflammation, such differentiation pathways are fundamentally changed and unique monocyte-derived inflammatory cells are generated.

Current DC, monocyte and MØ nomenclature does not take into account these new developments and as a consequence is quite confusing. We believe that the field is in need of a fresh view on this topic as well as an upfront debate on DC and MØ nomenclature. Our aim is to bring expert junior and senior scientists to revisit this topic in light of these recent developments. This Research Topic will cover all aspects of DC, monocyte and MØ biology including development, transcriptional regulation, functional specializations, in lymphoid and non-lymphoid tissues, and in both human and mouse models. Given the central position of DCs, monocytes and MØs in tissue homeostasis, immunity and disease, this topic should be of interest to a large spectrum of the biomedical community.

**Citation:** Ginhoux, F., Guilliams, M., Naik, S., eds. (2016). Dendritic Cell and Macrophage Nomenclature and Classification. Lausanne: Frontiers Media. doi: 10.3389/978-2-88919-918-1

# Table of Contents


*104 The complex myeloid network of the liver with diverse functional capacity at steady state and in inflammation*

Christoph Eckert, Niklas Klein, Miroslaw Kornek and Veronika Lukacs-Kornek

**Dendritic cells and macrophages across tissues**


Thien-Phong Vu Manh, Jamila Elhmouzi-Younes, Céline Urien, Suzana Ruscanu, Luc Jouneau, Mickaël Bourge, Marco Moroldo, Gilles Foucras, Henri Salmon, Hélène Marty, Pascale Quéré, Nicolas Bertho, Pierre Boudinot, Marc Dalod and Isabelle Schwartz-Cornil

*156 Investigating evolutionary conservation of dendritic cell subset identity and functions*

Thien-Phong Vu Manh, Nicolas Bertho, Anne Hosmalin, Isabelle Schwartz-Cornil and Marc Dalod

**Human dendritic cells and macrophages**


Bayan Sudan, Mark A. Wacker, Mary E. Wilson and Joel W. Graff

# Editorial: Dendritic Cell and Macrophage Nomenclature and Classification

#### *Florent Ginhoux1 \*† , Martin Guilliams2,3 \*† and Shalin H. Naik4,5 \*†*

*1 Singapore Immunology Network (SIgN), A\*STAR, Singapore, 2 Laboratory of Immunoregulation, VIB Inflammation Research Center, Ghent University, Ghent, Belgium, 3Department of Respiratory Medicine, Ghent University, Ghent, Belgium, 4Molecular Medicine Division, The Walter* + *Eliza Hall Institute of Medical Research, The University of Melbourne, Melbourne, VIC, Australia, 5Department of Medical Biology, The University of Melbourne, Melbourne, VIC, Australia*

Keywords: dendritic cells, macrophages, monocytes, nomenclature

#### **The Editorial on the Research Topic**

#### **Dendritic Cell and Macrophage nomenclature and classification**

#### *Edited and Reviewed by:*

*Peter M. Van Endert, Université Paris Descartes, France*

#### *\*Correspondence:*

*Florent Ginhoux florent\_ginhoux@immunol. a-star.edu.sg; Martin Guilliams martin.guilliams@irc.ugent.be; Shalin H. Naik naik.s@wehi.edu.au*

*† Florent Ginhoux, Martin Guilliams, and Shalin H. Naik contributed equally to the manuscript.*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 24 March 2016 Accepted: 18 April 2016 Published: 02 May 2016*

#### *Citation:*

*Ginhoux F, Guilliams M and Naik SH (2016) Editorial: Dendritic Cell and Macrophage Nomenclature and Classification. Front. Immunol. 7:168. doi: 10.3389/fimmu.2016.00168*

Mononuclear phagocytes that include dendritic cells (DCs), monocytes, and macrophages constitute a group of cell types crucial for the control of pathogens and induction of immune responses as well as for the support of tissue functions. These properties make them highly relevant targets for immune therapy, vaccination, and treatment of autoimmune and inflammatory diseases (1, 2). However, exactly how many cell types exist in the mononuclear phagocyte system (MPS), or whether they even combine to constitute a family, has been a matter of contention for many years. Historically, cells of the MPS have, at one time or another, been referred to as erythrophagocytes, pyrrhol cells, adventitia cells, rhagiocrine cells, polyblasts, clasmatocytes, and histiocytes (Yona and Gordon) prior to their current terminology established in 1972 (3). The seminal discovery of a new cell type termed DCs in the 1970s by the late Ralph Steinman that were distinct from macrophages added even more complexity in the MPS classification (4). However, some time passed before DCs were fully accepted as true member of the MPS. Over time, there was appreciation that there were not just one but multiple DC subtypes, each with a specialized role (5). So, while a "*dendritic-shaped cell that can process and present antigen to activate naive T cells*" was a good initial working definition (6), it did not take into account the inconsistent observations that other cells can be dendritic in appearance or activate naive T cells, and that not all "DCs" are immunostimulatory nor dendritic (7). As a result, many different cell types have been given a DC moniker over the years, such as monocyte-derived DCs, conventional DCs (cDCs), and plasmacytoid DCs (8). This appreciation of multiple subtypes has both clarified and confused the field. Importantly, we do not consider the classification and nomenclature issues as trivial semantics. Indeed, classification is of very practical importance in allowing researchers to work to a common framework as highlighted by Norma Lang "If we cannot name it, we cannot control it, finance it, teach it, research it, or put into public policy (page 109)" (9).

The idea behind this Frontiers Research Topic on "Dendritic Cell and Macrophage Nomenclature and Classification" was to have an open debate on the advantages and disadvantages of different classification systems of cells within the MPS. In this Research Topic, 17 contributions from international experts cover the complexity of the MPS, from its ontogeny and transcriptional regulation, its classification in different tissues and different species.

First, in a historical perspective, Yona and Gordon examine the early origins and development of macrophage research from Ilya Metchnikoff 's discovery to the establishment of the MPS nomenclature half a century ago.

In an opinion article, Vremec and Shortman discuss issues of DC subset definition encountered in their past work.

In a hypothesis and theory article, Guilliams and van de Laar discuss the practical application of our recently proposed classification system based on ontogeny (8).

Hoeffel and Ginhoux cover the ontogeny of tissue-resident macrophages and discuss evidence suggesting that hematopoietic stem cell-independent embryonic precursors transiently present in the yolk sac and the fetal liver give rise to long-lasting selfrenewing macrophage populations.

Tussiwand and Gautier discuss the developmental pathways of murine MPS cells, with a particular emphasis on the transcriptional factors that regulate their development and function.

Poltorak and Schraml review experimental approaches taken to fate map DCs and discuss how these have shaped our understanding of DC ontogeny and lineage affiliation.

In a perspective article, Gottschalk and Kurts review the complexity of the renal MPS, and how to distinguish DCs and macrophages in the kidney from the nephrologist's point of view.

Gross et al. discuss origins and functions of intestinal DCs and macrophages and their respective subsets, focusing largely on the mouse and cells residing in the lamina propria.

Greter et al. discuss myeloid cells in the brain and the difficulties to delineate resident microglia from infiltrating myeloid cells using currently known markers and the recent advances that have helped to make clear definitions between phenotypically similar, yet functionally diverse myeloid cell types of the brain.

Cassado et al. review the heterogeneity of peritoneal macrophages, which exhibit distinct phenotypes, functions, and origins.

Eckert et al. summarize the multiple roles of macrophages and DCs in chronic liver diseases and outline the currently known marker combinations for the identification of these cell populations for the study of their role in liver immunology.

### REFERENCES


Moving to human cells, Reynolds and Haniffa review the parallel organization of human and mouse mononuclear phagocyte networks. They also discuss the strategies, power, and utility of comparative biology approaches to integrate recent advances in human and mouse mononuclear phagocyte biology, and its potential to drive forward clinical translation of this knowledge.

In a research article, Vu-Manh et al. extend our knowledge of the homology of the MPS across species through comparative transcriptomics. They present an approach combining refined cell sorting and integrated comparative transcriptomics analyses, which revealed conservation of the mononuclear phagocyte organization across human, mouse, sheep, pigs, and chicken.

In a complementary review, Vu-Manh et al. discuss the highly significant conservation during evolution of DC subsets cell surface phenotyping, expression analysis of hallmark genes, and functions.

Ziegler-Heitbrock reviews human blood monocyte heterogeneity and their subdivision into classical, intermediate, and non-classical monocytes, and how these proportions change during inflammation and discuss its relevance to management of disease.

Durand and Segura review recent advances in our understanding of the human DC network and discuss some remaining gaps and future challenges of the human DC field.

Finally, in an original research article, Sudan et al. identify novel markers of activated human macrophages through the analysis of gene-expression profiles for human macrophages of a single donor subjected to 33 distinct activating conditions.

Altogether, the many contributions to this Frontiers Research Topic not only underline the complexity of the MPS system but also highlight the similarities between MPS cells of different tissues. Moreover, classifying MPS cells based on their geneexpression profiles results in a classification system that is close to the classification of cells based on their cellular origin and development. Although a final basis for MPS classification has not been defined yet, we hope this Frontiers Research Topic will pave the way toward a wider consensus within the field.


**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 Ginhoux, Guilliams and Naik. 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.*

# **From the reticuloendothelial to mononuclear phagocyte system – the unaccounted years**

*Simon Yona<sup>1</sup> \* and Siamon Gordon<sup>2</sup>*

*<sup>1</sup> University College London, London, UK, <sup>2</sup> Sir William Dunn School of Pathology, The University of Oxford, Oxford, UK*

It is over 125 years since Ilya Metchnikoff described the significance of phagocytosis. In this review, we examine the early origins and development of macrophage research continuing after his death in 1916, through the period of the reticuloendothelial system. Studies on these cells resulted in a substantial literature spanning immunology, hematology, biochemistry, and pathology. Early histological studies on morphology and *in situ* labeling laid the foundations to appreciate the diversity and functional capacity of these cells in the steady state and during pathology. We complete this phagocyte retrospective with the establishment of the mononuclear phagocyte system nomenclature half a century ago.

*Edited by:*

*Peter M. Van Endert, Université Paris Descartes, France*

#### *Reviewed by:*

*Edda Fiebiger, Harvard Medical School, USA Richard A. Kroczek, Robert Koch-Institute, Germany Jean M. Davoust, Institut National de la Santé et la Recherche Médicale, France*

#### *\*Correspondence:*

*Simon Yona, University College London, 5 University Street, London WC1E 6JF, UK s.yona@ucl.ac.uk*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 20 April 2015 Accepted: 11 June 2015 Published: 01 July 2015*

#### *Citation:*

*Yona S and Gordon S (2015) From the reticuloendothelial to mononuclear phagocyte system – the unaccounted years. Front. Immunol. 6:328. doi: 10.3389/fimmu.2015.00328* **Keywords: macrophage, monocyte, Metchnikoff, phagocytosis, dendritic cells, inflammation**

# **Introduction**

The earliest accounts of macrophage research are closely linked with the widespread introduction of the microscope in the mid-nineteenth century, 300 years following the seminal microscopic observations of Antony van Leeuwenhoek (1700) (1). In the histological accounts, von Kölliker (1847) detected cells in the spleen containing particles; later Preyer (1867) observed the internalization of erythrocytes by splenic cells and proposed that this occurred by an active process (2, 3). However, investigators at the time did not associate such observations with a defense mechanism. In fact, Klebs (1872) believed just the opposite, proposing that these cells assist the transport of bacteria to lymphatic tissue (4). Koch (1878) also concluded that these cells provide a suitable microenvironment for bacilli to multiply and disseminate to other tissues, – the so-called Trojan horse theory – after observing numerous bacilli within leukocytes, while studying frogs treated with anthrax (5). Therefore, although cytological observations of the mid-nineteenth century recognized the ability of leukocytes to devour (*fressen*) erythrocytes and microorganisms, opinion at the time did not associate this event with host defense, nor was there a consensus that the process was active or merely the penetration of foreign material into cells to aid infection.

By the late nineteenth century, Metchnikoff (1892), the Russian zoologist and forefather of cellular immunity, established the idea of the phagocyte (6–8). Metchnikoff was the first to fully appreciate the capabilities and purpose of phagocytosis, by performing a series of classical studies spanning from simple unicellular organisms to complex vertebrates. The description of Metchnikoff 's discovery of phagocytosis documented by his wife Olga, now rests in the pantheon of immunology legend.

*. . .* One day when the family had gone to see some performing apes at the circus, Metchnikoff with his microscope introduced a rose thorn into the transparent body of a starfish larva, Metchnikoff observed the accumulation of phagocytes surrounding the foreign material and attempting to devour the splinter*. . .*(9).

It is important to remember that Metchnikoff started his career as an evolutionary developmental embryologist, influenced by Darwin's recent publication On the Origin of Species in 1859. As an embryologist, Metchnikoff modeled the early formation of the embryo in primitive organisms, such as sponges, and proposed that an inner "*parenchymella*" contained wandering cells of mesodermal origin capable of taking up particulate matter during embryogenesis. These studies may have been the foundation for Metchnikoff 's phagocytosis theory. Later, Metchnikoff recognized the multiple tasks performed by phagocytosis; as an embryologist, the reabsorption of tissue during embryogenesis, as a zoologist, a common feeding mechanism of unicellular organisms and as a pathologist, its role in host defense. Therefore, when Metchnikoff performed his most notable study, the rose thorn experiment at Messina culminated in the phagocytic process we know today. Metchnikoff was one of, if not the, earliest to demonstrate the evolutionary functional adaptation of a particular biological process, in this case phagocytosis, from a simple feeding mechanism for unicellular organisms, to a developmental requirement during embryogenesis and finally as a necessity for host defense (3, 10).

Metchnikoff 's phagocytes comprised two populations he termed macrophages (large eaters) and microphages (small eaters, later known as polymorphonuclear leukocytes). Contrary to Rudolf Virchow's impression that inflammation is a continuous life threatening menace, Metchnikoff regarded it as a healing or salutary reaction as postulated 100 years earlier by the Scottish surgeon and collector John Hunter (1794) (11). Therefore, Metchnikoff concluded that the ability of cells to engulf foreign microorganisms acts as an active defense mechanism, giving rise to the concept of cellular innate immunity. At the time, this triggered extensive debate between humoral and cellular schools of thought. Two major events at the turn of the twentieth century helped to reconcile this dispute. First, in 1908, the Nobel Prize in Physiology or Medicine was awarded jointly to Metchnikoff, advocate of the cellularists and to Ehrlich, the champion of humoralist dogmas "*in recognition of their work in immunity*". Second, in 1903, Wright and Douglas proposed the concept of "opsonization" as a humoral mechanism to increase the susceptibility of bacteria to phagocytosis. These investigators claimed that humoral and cellular functions were not mutually exclusive, rather interdependent (12). This theory was spoofed by George Bernard Shaw, in the introduction to his play "The Doctor's Dilemma" in 1906.

*. . .* Sir Almroth Wright, following up one of Metchnikoff 's most suggestive biological romances, discovered that the white corpuscles or phagocytes, which attack and devour disease germs for us, do their work only when we butter the disease germs appetizingly for them with natural sauce which Sir Almroth Wright named opsonin*. . .*(13).

# **The Reticuloendothelial System**

By the early decades of the twentieth century descriptions of the phagocyte system had become chaotic, not least since the term macrophage had become synonymous with adventitia cell, anode cell, clasmatocyte, dictocyte, erythrophagocyte, histiocyte, polyblast, pyrrhol cell, and rhagiocrine cell; the many terms bestowed on these cells (*>*30 different names) (14) revealed that the divergence of opinion at the time as to the relationship of these cells to one another and from tissue-to-tissue. Not only were tissue phagocytes given a variety of bewildering names but also their origin remained unknown. From time-to-time, historic discoveries are lost in the ether of *a priori* thought; this is certainly true for histological techniques that assisted in the classification of blood cytology. Until Ehrlich's early effort to develop leukocyte cytological staining, scholars of blood operated solely on fresh samples. Ehrlich's aim was to take advantage of the known chemical structures of dyes and their interaction with cellular bodies to map and characterize the anatomy of blood cells. By using aniline dyes in combination with neutral dyes and the morphology of the nucleus, he was able to divide cells of the blood into mononucleated lymphocytes, some of which were large, large mononuclears with indented nuclei (now known as monocytes) and polymorphonuclear cells that were neutrophilic (neutrophils), acidophilic (eosinophils), or basophilic granules (basophils/mast cells) (15). By the early twentieth century, Ribbert (1904) had performed studies with lithium carmine solution injected into the peripheral circulation and observed the specific uptake and storage by a group of cells, which became *vitally stained* (16). These were subsequently demonstrated to be mononuclear cells phagocytosing particulate matter. Clark and Clark (17) described these large mononuclear cells in tissues to be the same as "clasmatocytes," described by Louis-Antoine Ranvier, the "Polyblasts" of Alexander Maximow and the "Histiocytes" defined by Kenji Kiyono. Following these early observations, it became apparent that a large number of histological dyes including trypan blue, neutral red, isamine blue, and other colloids discriminated phagocytes from fibroblasts. The systematic analysis of tissues and dyes led Karl Albert Ludwig Aschoff (1924) to coin the term "reticuloendothelial system" (RES) to describe this group of cells, with their ability to incorporate vital dyes from the circulation (18). Reticulo refers to the propensity of these large phagocytic cells in various organs to form a network or a *reticulum* by cytoplasmic extensions; *endothelial* refers to their proximity to the vascular endothelium (19), from which they were sometimes believed to arise, these cells formed Aschoff 's unified system throughout the organism. The capture and clearance of unwanted particulate material from blood and lymph were considered to be the major function of the RES. Although opinions about the origin of cells of the RES will be discussed later in this series; at the time, Aschoff considered that the cells of the RES were derived locally and that both histiocytes and reticulum cells shared a common origin.

# **Cells of the Reticuloendothelial System**

Metchnikoff had previously described the dissemination of macrophages throughout the organism and Aschoff 's system implied a common function of the cells of the RES even in the absence of inflammation. In the next section, we highlight some of the tissue locations and possible functions assigned at the time.

# **Kupffer Cells**

The macrophages of the liver are located within the sinusoids, which is composed of four cell types, each with its own morphology and function. Karl Wilhelm von Kupffer (1876) observed "*Sternzellen*" (star cells) in the liver and believed them to be an integral part of the hepatic endothelium. Later, Tadeusz Browicz (1899) identified Kupffer's cells as the phagocytes of the liver (20) (sometimes known as Browicz–Kupffer cells) and observed that they could take-up a large percentage of vital stain. In the early 1930s, Peyton Rous developed an ingenious method to isolate Kupffer cells of the liver. Rous and Beard injected a suspension of gamma ferrous oxide i.v., light in weight but highly magnetic particles, Kupffer cells efficiently phagocytosed these particles. They then perfused and processed the liver and the phagocytes were then selected by magnetic force, to the best of our knowledge the first description of magnetic cell sorting (21, 22), enabling the extraction of macrophages from a solid tissue for examination *in vitro*. The origin of Kupffer cells like all cells of the RES at the time remained a source of continued confusion and debate. At the American Association of Anatomists conference in 1925, M. R. Lewis presented a paper comparing Kupffer cells isolated from frogs, thought to be derived from endothelial cells, side-by-side with an examination of clasmatocytes and concluded these cells were identical in morphology and function (23).

In 1950s, Baruch Benacerraf, Nobel laureate in 1980 for his work on MHC with George Snell and Jean Dausset (24), teamed up with Guido Biozzi, a young Italian in the Halpern laboratory in Paris in a productive collaboration. They developed techniques to study clearance of particulates from blood and formulated equations that govern this in mammals. In subsequent work, Biozzi bred strains of mice differing in the quantitative antibody response to various antigens. Biozzi mice are still in use to study autoimmune inflammatory neurological disease (25). These studies in mice and guinea pigs helped to introduce genetic approaches to the role of macrophages in innate and adaptive immunity.

# **Microglia and the Origin of the RES**

Virchow (1858) acknowledged that the central nervous system (CNS) was composed of both neurones and interstitial cells, which he termed neuroglia (26, 27). By the end of the nineteenth century, the Scottish pathologist William Ford Robertson confirmed that neuroglia were indeed composed of multiple cell types (28). Robertson continued to investigate this heterogeneous population of cells; with the aid of platinum staining techniques he was able to distinguish a novel cell type in the brain he termed mesoglia (as he believed that they were mesodermal in origin). Finally, Robertson deduced that mesoglia possessed phagocytic properties (29). In fact, Robertson had identified oligodendroglia and mesodermal derived cells, under the term mesoglia. In 1913, Santiago Ramon y Cajal described a group of cells derived from the mesoderm as the "*third element*" of the CNS, the first element being neurones and the second element the astrocytes, derived from ectoderm. It was the Spanish pathologist Pio Del Rio-Hortega who revolutionized our understanding of neuroglia from a series of detailed studies using silver carbonate impregnation staining. He uncovered a homogeneous group of cells within the CNS with tree like projections and predicted that they possess phagocytic functions within the CNS; he termed these cells as microglia (26, 27). Hortega laid the groundwork for microglia research in a lecture given at University of Oxford, microglia enter the CNS during development from mesodermal origin where they disseminate throughout the CNS and take-up a branched ramified cytological appearance. He went on to explain that they remain evenly spaced in the steady state, while pathological insults cause microglia to take on an amoeboid morphology, express the ability to phagocytose and to migrate (30). These studies confirmed that the microglia of the CNS belonged to the RES. The account he gave in Oxford remains accurate to this day. Interestingly, although microglia were unable to take-up vital dyes because of the blood brain barrier, they were known to be highly phagocytic, reticuloendothelial cells readily stained by silver carbonate. Although Hortega described microglia to be derived from cells of the mesoderm during embryogenesis, this still remained a matter of great debate, until recently. Early observations in the late nineteenth and several studies in the early twentieth described microglia during neurodegenerative diseases without a clear understanding of their origin.

# **Osteoclasts**

The histological identification of a cell that resorbs bone can be traced back to the early 1850s. Tomes and De Morgan (31) described within a section of diseased femur, cavities that were associated with an increase in nucleated cells (31). By 1873, Kölliker described multinucleated giant cells involved in bone absorption that he termed *Ostoklast* and anticipated that these cells are involved in homeostatic and pathological bone degradation (32, 33). The notion of a bone-resorbing cell became widely accepted (34). Over the next 50 years, the morphology of the osteoclast was refined and interestingly these large multinucleated cells showed variation in size and nuclear content; in pigs, they contained as many as 100 nuclei (35) while human osteoclasts could contain up to 50 nuclei (32). John Loutit an Australian born pioneer in radiation biology from the late 1940s studied not only osteoclast origin from blood precursors but also the biology of bone marrow transplantation after irradiation (36) in a long and productive career at Harwell MRC laboratories.

# **Alveolar Macrophages and Phagocytosis**

The lung also contains many mononuclear phagocytes, which are associated with the alveoli and the alveolar space (37). The macrophages within the alveolar space were initially known as "dust cells" because of their content of intracellular carbon particles. There is a constant requirement to keep the alveolar space free of particles and pathogens allowing for optimal oxygen transfer, the major role of these cells. As the lung occupies a unique accessible position among internal organs, it is constantly in contact with the external world. In the late 1950s, Karrer observed the efficient phagocytosis of India ink exclusively by free alveolar macrophages, similar to the previous observations of increased carbon particles in these cells of city dwellers (38, 39). The question of the origin of the macrophage troubled cytologists and immunologists for most of the twentieth century; this was no different in the lung.

One of the best-studied pathologies in the first half of the twentieth century in relation to macrophages was pulmonary tuberculosis. The initial stage of tuberculosis displays the transient influx of neutrophils described by Maximow in 1925 (40). However, these cells are unable to destroy the bacilli and monocyte/macrophages remain the most prominent infected host cells. From 1920s, Sabin, the first full female Member of the Rockefeller Institute and first female elected to the National Academy of Sciences, considered the monocyte response to tuberculosis the most significant process, "*cellular and immunological reactions in tuberculosis center around the monocyte*," when she first proposed this she was mocked by her peers. Sabin observed monocytes to become epithelioid cells that develop into macrophage giant cells (41), previously described by the German pathologist Theodor Langhans as a hallmark of tuberculous granulomata already in the nineteenth century. In 1930s, Max Lurie, an advocate of the monocyte theory, used inbred rabbits to study their susceptibility to bovine tuberculosis. Lurie observed resistant inbred rabbits went on to develop cavitary tuberculosis while susceptible families went on to develop disseminated tuberculosis (42). The Australian immunologist, George Mackaness studied the role of anti-TB drugs on infected macrophages when a student at the Sir William Dunn School of Pathology, University of Oxford with Howard Florey in the early 1950s. His subsequent studies in the sixties at the Trudeau Institute in Saranac Lake defined T lymphocyte-dependent activation of macrophages by BCG and *Listeria* infection (43, 44), Mackaness coined the term macrophage activation, the so-called "angry" macrophages (45). Dannenberg has been another pioneer of macrophage research in experimental and clinical tuberculosis, especially in the characterization of the granuloma (46).

Other resident phagocytic populations were described in many tissues during this period, for example, in the skin (Langerhans cells), gut, lympho-hemopoietic tissues, reproductive and endocrine organs, and placenta (Hoffbauer cells). We draw attention to specialized macrophages in bone marrow stroma, where they appear at the center of hematopoietic islands, first described in detail by Marcel Bessis and his collaborators (47). John Humphrey drew attention to the marginal metallophilic macrophages located in a zone between the red and white pulp of spleen, especially in rodents (48, 49). They line a sinusoidal space where they sample circulating blood for viruses, for example, and play an important role in clearance of T cell-independent immunogenic polysaccharides. Tingible body macrophages (TBM) were identified by Walther Flemming in 1885; located in germinal centers. TBM contain phagocytosed apoptotic cell debris (tingible bodies) and are involved in the clearance of apoptotic lymphocytes (50), these observations were confirmed by electron microscopy in the early 1960s (51). Finally, the peritoneal macrophages of the mouse, responsible for much of our knowledge of macrophage immunobiology, were first described as a tractable cell population by Cohn only in 1962 (52).

# **The Origin of Macrophages**

As Aschoff was formulating the requirements of the RES, a number of research groups were searching for the origin of macrophages. In 1914, Awrorow and Timofejewskij concluded from the outgrowth of leukocytes from leukemic blood that the lymphocyte is the progenitor from which macrophages arise (53). A few years later, several *in vitro* studies described the differentiation of blood monocytes into macrophages (53–55); Carrel and Ebeling (55) and Lewis and Lewis (23) observed that blood cultures over time developed into macrophages that had phagocytosed the debris of other blood cells, concluding that these monocyte-derived cells became actively phagocytic and were indistinguishable from macrophages in staining with neutral red (54, 55). At the same time, in 1925, Sabin took a cytological approach using neutral red staining to examine resident macrophage populations in connective tissue (clasmatocytes), concluding that a proportion of macrophages were derived from bone marrow-derived monocytes (56). However, the first *in vivo* study to examine how mammalian blood monocytes behave during an acute pathological insult was performed by Ebert and Florey (57) at the University of Oxford, using the rabbit ear chamber, observing diapedesis of blood monocytes toward the site of tissue injury. These monocytes transformed into macrophages during the inflammatory process; they concluded "*The cells originating from monocytes eventually became cells which we are calling histiocytes, which are indistinguishable from the socalled resting wandering tissue-cell of Maximow*" (57). Twenty-five years later, Volkman and Gowans (58) confirmed these findings using thymidine autoradiography and parabiosis inferring that macrophage precursors are rapidly dividing cells derived from a remote site during inflammation (58). Takahashi mentions in his comprehensive review on macrophage ontology how the Japanese pathologist Amano with the aid of supravital staining at the inflammatory foci observed blood monocytes to be precursors of the macrophage (59). Finally, Marchesi and Florey employing electron microscopy were able to distinguish the earliest phase of monocyte migration during mild inflammation, which occurred during the maximal efflux of neutrophils (14, 60). The conclusion from these studies suggests that macrophages derived from circulating monocytes were based on inflammatory models. Therefore, a more accurate conclusion would be that during inflammation monocytes become effector cells by concentrating at the site of injury with the ability to produce large quantities of inflammatory mediators (61).

Another important and well-studied population of recruited monocyte-derived cells are the foam cells, a hallmark of atheromatous pathology. A pupil of Maximow in St. Petersburg and later the student of Aschoff in Freiburg (62), Anitschkow (63, 64) showed that simply by feeding rabbits purified cholesterol caused vascular changes leading to the formation of lesions similar to those seen during atherosclerosis in humans (63, 64). Anitschkow decided that these cholesterol-laden cells were of leukocyte origin (65). Anitschkow's work on lipid storage was compared to the work by Robert Koch on the tubercle bacillus (66). It was mainly as a result of the work by Russell Ross in 1970s that these monocyte-derived cells have been categorized as part of a fat modified inflammatory process (67). Recruited monocytes can also give rise to multinucleated giant cells, not only a feature of tuberculosis. They are found in many granulomatous inflammatory diseases, including viral and parasitic infections, and in responses to foreign bodies and fat necrosis (Touton cells), named after the German dermatologist Karl Touton (1885) (68).

The accumulation of data and the introduction of new techniques highlighted that the cells of the RES differ in morphology, function, and origin (14). In addition, the underlying processes involved in these functional and morphologic alterations remained unknown. Is there a proliferating mononuclear phagocyte population within the RES, constantly differentiating in the steady state? These questions continued to puzzle scientists throughout the twentieth century.

# **The Mononuclear Phagocyte System**

*. . .* The most immature cell of the mononuclear phagocyte system *. . .* is the promonocyte *. . .* that by dividing gives rise to monocytes*. . .* Monocytes in the circulation constitute a mobile pool of relatively immature cells on their way from the place of origin to the tissues. At sites where conditions are favourable for phagocytosis, these cells become macrophages*. . .*(69).

As knowledge accumulated, the term RES was regarded as insufficient to describe resident phagocytes and their antecedents. At a scientific meeting in Leiden in 1969, a group of prominent pathologists/immunologists proposed the term "mononuclear phagocyte system" (MPS) as a more accurate term (**Figure 1**) (69). The MPS at the time comprised monocytes and macrophages derived from the bone marrow derived monocytes. Nevertheless, little evidence existed to suggest that monocytes differentiate into resident macrophages under steady state conditions. Interestingly, Maximow proposed on the basis of embryonic studies on amphibians and mammals that macrophages and leukocytes may arise from distinct lineages (70).

While the MPS was being formulated in 1960s, immunologists were in pursuit of the "*third cell*" (71) a requirement for adaptiveimmune responses. Steinman and Cohn in their landmark study (1973) identified and characterized the dendritic cell (DC) as distinct from macrophages (72, 73); over time, the DC became accepted as the third arm in the trinity we know today as the

MPS (74). Monocytes, macrophages, and DCs are distinguished on the basis of morphology, function, and origin, yet collectively constitute the MPS.

As more data accumulated in the early decades of the twentyfirst century, it emerged that most tissue macrophage populations in adults in the steady state are maintained independent of the bone marrow and rely almost exclusively on self-renewal (75, 59, 76–86). These data facilitated the reexamination of the concept of the MPS (87).

# **Conclusion**

We have highlighted only a few of the many milestones of macrophage biology from its early origins to the establishment of the MPS nomenclature in 1968. Studies during this period resulted in a substantial literature spanning immunology, hematology, and pathology. A number of important issues emerge from a retrospective analysis of the literature. First, we learn that rarely in science do revolutions occur from a single Eureka moment rather years of observation culminate in new findings. While Metchnikoff 's phagocytosis theory seems to have emerged from his experiments on starfish larvae, he had previously observed cells capable of taking up particulate matter during embryogenesis (10). Second, the first half of the twentieth century, blighted by two World Wars, had profound impacts on science, resulting in a geographical and common language shift of scientific research. Third, the techniques used routinely in the laboratory shifted from the pathologists' tool box of the microscope and microbiology to the immunologists' introduction of cell transfer, thymidine autoradiography, immunohistochemistry, parabiosis, electron microscopy, later flow cytometry, cell, and molecular biology. However, if one was able to transport Metchnikoff, Aschoff, or Cohn to a conference in 2015 on mononuclear phagocytes they would perhaps not appreciate the cytokines, chemokines, blots, and plots; however, the fundamental questions and discussions remain familiar; what is the origin of these cells? How do they phagocytose? Do macrophages proliferate *in situ*? How is particulate material recognized and cleared? This is why it is important to examine the history of our field since our research questions today are more closely linked to the past than we may appreciate.

The macrophage story is not over. In recent months, further strides have been made in examining the molecular signatures, characterizing the MPS in the steady state and upon enhanced recruitment of monocytes during inflammation (88–94). These studies highlight collective attributes of the macrophages; however, they also show significant local adaptations associated with particular functions within a specific organ. The next stage on this journey will include recreating *in vitro* the phenotypes of these specific populations using induced pluripotent stem cells, and gaining a greater insight into how these cells behave under steady state conditions *in situ*, as well as during and after the inflammatory response. Finally, the role of the circulating monocyte is also undergoing a re-evaluation; previously, monocytes were viewed as the bridge from bone marrow progenitors to fully differentiated tissue macrophages not only after inflammation, injury, and infection but also for resident populations in the absence of inflammation, as stated in Van Furth's description of the MPS "*Monocytes in the circulation constitute a mobile pool of relatively immature cells on their way from the place of origin to the tissues*" (69). Moreover, monocytes should now be further investigated as distinct precursors of only newly recruited monocyte-derived cells and as effector cells in their own right.

# **References**

*A number of the citations in this review are historical and may be unavailable to readers. If you are interested in obtaining a copy of a particular publication please contact the corresponding author. In addition, please be aware the authors have cited each paper as it appeared at the time of publication, many journal names have changed overtime papers are cited as they appeared at the time of print.*


# **Acknowledgments**

We have cited only a fraction of investigations during this period. We dedicate this review to all the scientists who have contributed to the field especially E. Metchnikoff and Z. A. Cohn, whose contributions continue to impact research today.


**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 Yona and Gordon. 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.*

# **What's in a name? Some early and current issues in dendritic cell nomenclature**

*David Vremec<sup>1</sup> and Ken Shortman1,2,3 \**

*1 The Walter and Eliza Hall Institute, Melbourne, VIC, Australia, <sup>2</sup> Department of Medical Biology, The University of Melbourne, Melbourne, VIC, Australia, <sup>3</sup> Burnet Institute, Melbourne, VIC, Australia*

**Keywords: dendritic cells, DC subsets, monocytes, macrophages, nomenclature**

The name dendritic cell (DC) was given by Steinman to describe the unusual cell type he saw in spleen cell suspensions. This morphological description is not sufficient to specify the cell of so much interest to immunologists; many cells can adopt a similar form. A useful functional definition evolved as Steinman and colleagues explored the immunological properties of this novel cell type (1). DCs were considered as antigen collecting and processing cells able to present antigen on MHC molecules and efficiently activate even primary T-cells. Nowadays, immunologists would likely add to this definition, a capacity to sense the context in which the antigen was collected, via receptors for pathogen or damaged cell-derived material. Why might we need to go beyond the name "dendritic cell" for cells with these well-understood functions? Some limitations of this single name arose early in DC research. This article surveys some problems of definition encountered in past work from our own laboratory. The problems we encountered arose from two sources, the first the discovery of different DC subsets and the need to determine whether these represented different maturation states or separate sub-lineages. The second was the difficulty in distinguishing these DC subsets from macrophages.

Our first hint that there could be distinct types of DCs came from our studies with Wu and Ardavin on thymic T and DC development (2). We were surprised to find that a high proportion of mouse thymic DCs stained with antibodies against characteristic T-cell markers, such as CD8α; it was a relief to find they did not stain with antibodies against CD3 or the T-cell receptor! Pickup of material from thymocytes was eliminated as an explanation. We then found a similar but less frequent DC subset staining for surface CD8α among the DCs in mouse spleen and these DCs were shown to express mRNA for CD8α (3). Others had already reported some staining of DCs with anti-CD8; our work emphasized that these CD8<sup>+</sup> DCs were a distinct population, CD8α expression being positively correlated with expression of DEC205 but inversely correlated with expression of other markers such as CD4, CD11b, and, as illustrated in **Figure 1**, SIRPα (4, 5).

Immunological interest in the CD8<sup>+</sup> and CD8*<sup>−</sup>* DC subsets increased when it became apparent from the work of many laboratories that these DCs differed in immunological functions. Differences were apparent in the expression of toll-like and other microbial pattern recognition receptors, in the cytokines produced on activation, in the fate of the T-cells they stimulated, in their capacity to phagocytize dead cells, and in the processing of antigens for MHC class I versus MHC class II presentation [reviewed in Ref. (7)]. The key findings from our laboratory were that the CD8<sup>+</sup> DCs, when appropriately stimulated, were the most potent producers of IL12p70 (8), and that the CD8<sup>+</sup> DCs have a strong bias to cross-presenting exogenous antigens, both soluble and particulate, for MHC class I presentation (9, 10).

An important issue became whether these functionally distinct DC types represented different lineages, or were simply different maturation states within one very plastic lineage. There was direct evidence, confirmed by us, that some CD8*<sup>−</sup>* DCs could on adoptive transfer, produce CD8<sup>+</sup> DCs. However, these CD8<sup>+</sup> DCs proved to be generated from a small number of early members of the

#### *Edited by:*

*Martin Guilliams, Ghent University – VIB, Belgium*

#### *Reviewed by:*

*Elodie Segura, Institut Curie, France Steffen Jung, Weizmann Institute of Science, Israel*

> *\*Correspondence: Ken Shortman shortman@wehi.edu.au*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 10 April 2015 Accepted: 14 May 2015 Published: 29 May 2015*

#### *Citation:*

*Vremec D and Shortman K (2015) What's in a name? Some early and current issues in dendritic cell nomenclature. Front. Immunol. 6:267. doi: 10.3389/fimmu.2015.00267*

CD8<sup>+</sup> DC lineage that had not yet acquired CD8α expression; the bulk of CD8*<sup>−</sup>* DCs did not give rise to CD8<sup>+</sup> DCs (11). Although sequential maturation states were found within the CD8<sup>+</sup> DC lineage, with early forms lacking CD8α expression (12) and the earliest stages lacking capacity for antigen cross-presentation (13), there was a clear developmental separation from the bulk of CD8*−* DCs. A further distinct DC type in mouse spleen became evident when the mouse equivalent of the human type 1 interferonproducing plasmacytoid dendritic cell (pDC) was identified (14). Although some pDCs expressed CD8α (15), they were clearly a separate lineage from the CD8<sup>+</sup> conventional DCs (cDCs). Our subsequent work with Naik showed that spleen CD8<sup>+</sup> DCs and CD8*−* DCs represented separate cDC sub-lineages derived via pre-DC populations from a bone marrow pro-DC or common dendritic cell precursor (CDP) (16–18). Thus, because of differences in surface phenotype, immunological functions, and developmental pathways, these two spleen cDC populations had to be distinguished, and the terms CD8<sup>+</sup> cDC and CD8*<sup>−</sup>* cDC became current.

In line with the pioneering work of Salomon et al. (19) and Anjuere et al. (20), we extended our analysis of DC subtypes from mouse spleen to mouse lymph nodes (LNs) (4, 21). Spleen should contain only what we termed the "lymphoid tissue resident" cDCs whereas LNs should contain both these and the "migratory" cDC type arriving via the lymph from other tissues. It was immediately apparent that the level of DC heterogeneity was greater than in spleen. One source of heterogeneity was the existence of different activation states within even one DC lineage. In particular, the DC that had migrated even in steady state from peripheral tissues such as skin into LNs were more activated than those remaining in skin, and more activated than their lymphoid tissue resident counterparts; the DCs that had migrated expressed higher surface levels of MHC class II and of co-stimulator molecules such as CD86. Although they were first called "mature" DCs they proved to be not fully activated but "semi-mature"; they were not producing cytokines and were likely tolerogenic (22). A similar transformation termed "spontaneous activation" occurred when spleen cDCs were isolated and placed in culture (23). In both cases, further signals, such as given by microbial products interacting with TLR ligands, were required to produce a fully active, cytokine secreting immunogenic DC. However, even when these different activation states were considered, further cDC subsets not found in spleen were apparent, such as the epidermal Langerhans cell-derived LN DCs. The full lineage complexity of LN has now been well delineated by other laboratories, a notable finding being the existence of a migratory form of the CD8<sup>+</sup> DC lineage but lacking CD8α expression, commonly termed as the CD103<sup>+</sup> cDCs (24–26).

Our second problem with DC nomenclature arose as we attempted to distinguish DCs from macrophages, a particularly difficult exercise in inflamed tissues. It was also difficult to relate the DC populations we isolated from steady state mouse spleen with the DCs produced by culture of monocytes with granulocyte-macrophage colony stimulating factor (GM-CSF), a well-established model of DC generation (27). At that time, it was generally assumed that all DCs and macrophages would be bone marrow derived cells, with monocytes as the common late precursor. Some questioned whether it was valid even to consider DCs as a separate entity rather than as a macrophage variant (28). We had some sympathy with this view, since in experiments with Metcalf we had difficulty in segregating DCs from macrophages in the peritoneal fluid from mice expressing high levels of GM-CSF (29). Although cells with DC function and surface phenotype could be segregated from macrophages at the extremes of the distribution, there appeared to be a continuum of phenotypes rather than two discrete populations. For us the clarification came when, with Naik, the immediate precursor of the spleen cDC was isolated and shown to be distinct from monocytes and unable to produce macrophages (16). We termed these pre-DCs. This led to the view that there were two different routes to cells with DC antigen presenting function, one via monocytes and more often found under conditions of inflammation, the other derived from CDP/pro-DC precursors in bone marrow then via pre-DC to the types of DC found in steady state lymphoid tissue (17). The culture model finally developed for generation of the type of DCs found in steady state became bone marrow stimulated with Flt3 ligand, rather that with GM-CSF (30, 31). Thus, the developmental pathway leading to DC functions became a major criterion for segregating and naming DC subtypes.

It was then possible to segregate DCs derived from monocytes from the cDCs found in steady state spleen. However, it is evident from the account above that our previous nomenclature of the subsets of spleen cDCs based on CD8α expression was inadequate. Certain pDC subsets also expressed CD8α. Early DCs of "CD8<sup>+</sup> cDC" lineage in spleen did not express CD8α. The migratory version of the same lineage, the CD103<sup>+</sup> DCs, did not express it. And finally, CD8 was not expressed by human DCs. A major advance was the demonstration in several laboratories of an equivalent of the mouse "CD8<sup>+</sup> cDC" lineage within human DCs, and the finding that the chemokine receptor XCR1 and the C-type lectin-like molecule Clec9A, rather than CD8, served as common DC surface markers crossing this species barrier [commentary in Ref. (32)]. The proposed designation of this DC subtype as DC1 overcomes the previous nomenclature problems (33).

In contrast to these advances in understanding the DC1 subset, the CD8*<sup>−</sup>* CD11bhigh SIRPα high cDCs (designated as DC2) have been less studied and still present nomenclature issues. We had already separated spleen CD8*−* DCs into two subsets based on CD4 expression (5), but the significance of this remains obscure. A more meaningful separation can now be made based on surface expression of Clec12A (DCAL2, MICL) versus DCIR2 or ESAM (6, 34, 35). An example of such segregation is shown in **Figure 1**. Importantly, these DC subsets differ in both developmental requirements and immunological characteristics; formation of DCIR2<sup>+</sup> ESAMhigh Clec12A*<sup>−</sup>* DCs requires Notch2 signaling and this subset selectively responds to flagellin and induces Th2 responses. Will these differences demand a further division into DC2 and DC3 subtypes? Or will one of these, particularly the

# **References**


Clec12A<sup>+</sup> subset, prove to be part of the monocyte-derived group? These questions require further work.

It is notable that ontogeny has led to a better understanding and provided one logical basis for DC classification (33). Will ontogeny be the best guide for DC nomenclature in future? We can foresee one area where it may cause confusion. A proportion of mouse pDCs and the CD8α-expressing subset of cDCs in the mouse thymus have a potential route of development from lymphoid rather than myeloid precursors (36, 37). These DC types have D–J rearrangements in their Ig heavy chain genes, a characteristic of lymphoid-origin cells (38). The extent to which a lymphoid route contributes to their development in steady state is still unclear, but the potential is there. Yet, the thymic CD8<sup>+</sup> DCs are similar to the splenic CD8<sup>+</sup> DCs of myeloid origin, and pDCs developing from myeloid or lymphoid precursors have similar surface phenotype and immunological functions. Should they have separate names according to their developmental origin, or should this "convergent" development lead to cells with the same name? There may yet be fine differences in function that eventually will be important to specify, but at present they are called by the same name. One resolution of this paradox comes from the likelihood that, despite the differences in bone marrow precursor surface markers, a common molecular program for pDC or for CD8<sup>+</sup> cDC formation has been initiated, with transcription factors that override any previous precursor orientation. Considering ontological origin in terms of these final molecular programs, rather than by the surface markers on the precursor cells, should overcome the paradox resulting from apparent convergent differentiation.

# **Acknowledgments**

Our research was supported by the National Health and Medical Research Council, Australia and was made possible through the Victorian State Government Operational Infrastructure Support and Australian Government NHMRC IRIISS.


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

# A hitchhiker's guide to myeloid cell subsets: practical implementation of a novel mononuclear phagocyte classification system

*Martin Guilliams 1,2 \* and Lianne van de Laar 1,2 \**

*<sup>1</sup> Laboratory of Immunoregulation, VIB Inflammation Research Center, Ghent University, Ghent, Belgium, 2Department of Respiratory Medicine, University Hospital Ghent, Ghent, Belgium*

The classification of mononuclear phagocytes as either dendritic cells or macrophages has been mainly based on morphology, the expression of surface markers, and assumed functional specialization. We have recently proposed a novel classification system of mononuclear phagocytes based on their ontogeny. Here, we discuss the practical application of such a classification system through a number of prototypical examples we have encountered while hitchhiking from one subset to another, across species and between steady-state and inflammatory settings. Finally, we discuss the advantages and drawbacks of such a classification system and propose a number of improvements to move from theoretical concepts to concrete guidelines.

Keywords: nomenclature, dendritic cells, macrophages, monocytes, classification

# Introduction

In the science fiction series created by Douglas Adams (1), the Hitchhiker's Guide to the Galaxy starts as follows: "*Space is big. Really big. You just won't believe how vastly hugely mind-boggling big it is. You may think it's a long way down the road to the chemist's, but that's just peanuts to space*." Given the complexity of the mononuclear phagocyte (Star)system (MPS), one could easily give a similar warning to readers who are trying to make some sense of the huge number of hypothetically distinct dendritic cell (DC) and macrophage (MΦ) subsets. At the last International DC Symposium (DC2014, Tours – France), we counted at least 28 different DC subsets that were described using various surface markers and nomenclature systems in distinct species. If one would add the different MΦ subsets and the Cytof technology allowing to measure the expression of more than 30 different surface markers per cell, one could with a bit of luck end up with "42" as answer to the ultimate myeloid question of how many mononuclear phagocyte subsets exist in life, the universe, and everything. Although this would be great for fans of Douglas Adams, without Babel Fish to help us make some sense of so many different subsets, this evolution will not be beneficial for communication among myeloid cell experts, let alone for the communication toward pharmaceutical companies, scientific editors, medical doctors, or graduate students. We will here try to simplify this apparent complexity through a number of practical examples and theoretical concepts. Having hitchhiked from MΦ to DC labs studying myeloid cells in various tissues and in distinct inflammatory conditions, we would, in accordance with the Hitchhiker's Guide to the Galaxy, advise the following: do not panic and bring your towel along.

#### *Edited by:*

*Ken J. Ishii, National Institute of Biomedical Innovation, Japan*

#### *Reviewed by:*

*Richard A. Kroczek, Robert Koch-Institute, Germany Kazuhiro Suzuki, Osaka University, Japan*

#### *\*Correspondence:*

*Martin Guilliams and Lianne van de Laar, Laboratory of Immunoregulation, VIB Inflammation Research Center, Ghent University, Technologiepark 927, Zwijnaarde, 9052 Ghent, Belgium martin.guilliams@irc.vib-ugent.be; lianne.vandelaar@irc.vib-ugent.be*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 13 June 2015 Accepted: 24 July 2015 Published: 11 August 2015*

#### *Citation:*

*Guilliams M and van de Laar L (2015) A hitchhiker's guide to myeloid cell subsets: practical implementation of a novel mononuclear phagocyte classification system. Front. Immunol. 6:406. doi: 10.3389/fimmu.2015.00406*

# Members of the Mononuclear Phagocyte System

In the original MPS model proposed by Ralph van Furth, James Hirsch, and Zanvil Cohn, MΦs were proposed to derive from circulating monocytes (2). A couple of years later, Ralph Steinman and Zanvil Cohn identified DCs, which were also included in the MPS (3). The fact that DCs could be derived from human and mouse monocytes in GM-CSF-driven *in vitro* cultures (4–8) and *in vivo* upon inflammation or in barrier tissues (9–15) supported this concept. For a historical overview of the MPS field, we redirect the readers to the review of Simon Yona and Siamon Gordon in this issue (16). The identification of mouse hematopoietic precursors committed to the DC lineage called the common DC progenitors (CDPs – giving rise to pDCs and cDCs) and precDCs (giving rise to cDCs) that are distinct from monocytes and can give rise to the so-called conventional DCs (cDCs) induced a first conceptual revolution in the field (12, 17–20). Moreover, Flt3-L, and not GM-CSF, was shown to be critically involved in the development of cDCs *in vitro* (8, 21–23) and *in vivo* (24–28). Recently, two additional committed precursors were identified in mice: the pre-pDC precursor that preferentially differentiates into pDCs (29), and the monocyte-committed common monocyte progenitor (cMop) (30). Importantly, the human equivalent of the pre-cDC, CDP, and cMop was recently identified (31, 32). A second conceptual revolution in the field was driven by the finding that most tissue-resident MΦs do not derive from circulating HSC-derived monocytes but develop from embryonic precursors, i.e., the yolk-sac MΦs (YS MΦs) or fetal liver (FL) monocytes (33–39). The relative contribution of YS MΦ-derived and FL monocyte-derived MΦs seems to vary from one organ to another (40–42). It was recently demonstrated that almost all MΦs have a YS origin [either directly from YS MΦs or through YS-derived EMPs (39)]. This may seem in contradiction with the proposed partial origin from FL monocytes (35, 43). However, it is now clear that YS-derived EMPs seed the FL and go through a FL monocyte intermediate before differentiating into most tissueresident MΦs (44), reconciling most of the apparent discrepancies in the field. Together, these findings have challenged the MPS dogma and revealed that most DCs and MΦs derive from distinct committed precursors rather than from circulating HSC-derived monocytes (**Figure 1**).

# Revisiting the Classification of Mononuclear Phagocytes

Historically, mononuclear phagocytes were classified as DCs or MΦs based on a restricted set of surface markers (CD11c and MHCII for DCs versus F4/80 for MΦs), proposed functional specialization (antigen-presentation and migration to lymph nodes for DCs versus phagocytosis for MΦs) and/or morphological features (dendritic-shaped cells for DCs versus large vacuolar cells for MΦs). However, these features are often not mutually exclusive. For example, although CD11c and MHCII are typically associated with DCs, alveolar MΦ are CD11chi and MHCII is expressed by intestinal MΦs (35, 45). Ideal surface markers allowing identification of the distinct myeloid cell subsets across tissues and species are still incomplete. Markers typically associated with some myeloid cell subsets can be lost or acquired by other subsets. The monocyte-associated

marker Ly-6C is rapidly down-regulated on many monocytederived cells (MCs) upon entrance in the tissues (45–48) and is expressed on pDCs (and lowly expressed on some cDCs). The pDC-associated marker mPDCA1 (stained with 120G8) can be acquired by MCs during inflammation (49). Alveolar MΦs (50) and Kupffer Cells (unpublished data) can upregulate CD11b during inflammation. Finally, BDCA3 is expressed on both human cDC1s and MCs (51). Thus, the inability to consistently identify myeloid cell subsets irrespective of tissue, species, or inflammatory state makes surface markers unattractive as basis for classification.

We would also propose to avoid a classification based primarily on functional specialization. First, each myeloid cell subset can perform more than one prototypical function. MΦs are often linked to phagocytosis of dead cells and pathogens but also have important immunomodulatory and metabolic functions. Second, subsets can acquire or lose functional capacities during inflammation as recently demonstrated for cDC2s that acquire crosspresentation capacities upon TLR stimulation (52). Therefore, we propose to disregard function as a basis for classifying cells.

Instead of surface markers, functional specialization, or morphology, we have recently suggested to classify cells based on their cellular origin, which could allow a more robust classification system (53). This would yield three big groups of cells (**Figure 1**): (i) embryonic progenitor-derived MΦs, (ii) CDP-derived DCs (that would be subdivided into cDC1s, cDC2s, and pDCs), and (iii) MCs. As these precursors have now been identified in both the mouse and the human, this allows one classification system across tissues and species.

Although precursor-based classification would provide a robust and species-conserved system, at the end of the day the function of the cells is what really matters for converting our knowledge into therapeutic advances for patients. Regrouping all the DCs into three big subsets of cDC1s, cDC2s, and pDCs will thus have the disadvantage of lumping together cells that may be in very different functional activation states. Similarly, MCs have been shown to be particularly plastic cells (54). Therefore, we propose to add a second classification level to the fixed ontogeny-based Level1 (**Figure 2**). Addition of a Level2 allows specification of the cellular activation state, the micro-anatomical localization or simply the surface markers utilized to identify the cells in a particular study. Of note, when defining the Level2 it will be important to avoid generalizations as a given function is often performed by only a fraction of the cells studied. We would thus propose to restrict the Level2 to objective criteria that can be measured at the single-cell level.

# Practical Implementation for DCs

Historically, DCs were divided into subsets based on surface markers that differed between tissues and species, such as CD207 (Langerin) in the skin, CD103 (IntegrinαE) in the intestine, CD11b (IntegrinαM) in the lungs, CD4/CD8α in the spleen, and CD24/CD172α for *in vitro* differentiated DCs (**Figure 3**). Human DCs, on the other hand, have been divided into CD141<sup>+</sup> (BDCA3) and CD1c+ (BDCA1) DCs. pDCs are identified by the expression of BDCA4 and BDCA2 in human, but by B220, mPDCA1 (BST2, recognized by 120G8), or Siglec-H in mice (53). Technical advances in multi-color flow cytometry have made matters worse with evermore "novel DC subsets" based on the expression of additional surface markers. By comparing the gene-expression profile of DCs isolated from various tissues and species, one can appreciate three big clusters of DCs (55–61). The pDC cluster includes mouse PDCA1<sup>+</sup> pDCs and human BDCA2<sup>+</sup>BDCA4<sup>+</sup> pDCs. The cDC1 cluster comprises dermal CD207<sup>+</sup>CD103<sup>+</sup> cDC1s, lung CD103<sup>+</sup>CD11b**<sup>−</sup>**cDC1s, splenic CD8a<sup>+</sup>CD4**<sup>−</sup>** cDC1s, intestinal CD103<sup>+</sup>CD11b**<sup>−</sup>** cDC1s and human blood BDCA3<sup>+</sup> cDC1s. Dermal CD207**<sup>−</sup>**CD11b<sup>+</sup> cDC2s, lung CD103**<sup>−</sup>**CD11b<sup>+</sup> cDC2s, splenic CD8a**<sup>−</sup>**CD4<sup>+</sup> cDC2s, intestinal CD103<sup>+</sup>CD11b<sup>+</sup> cDC2s and human blood BDCA1<sup>+</sup> cDC2s form the cDC2 cluster (62, 63). This bio-IT-driven analysis also revealed that within the cDC population, XCR1 and Sirpα are, respectively, expressed by all cDC1s and cDC2s across tissues, allowing an improved identification of these cells (51, 64–70). Note, however, that Sirpα is also expressed by other myeloid cells than cDC2s, showing the need for correct cDC identification prior to using this marker to distinguish cDC2s from cDC1s. Strikingly, this gene-expression-based division is supported by the existence of distinct pre-committed precursors (29, 71, 72) and by differential developmental transcription factor requirement of cDC1s,

cDC2s, and pDCs in the mouse. cDC1s, but not cDC2s, require BATF3 (71, 73, 74), ID2 (28, 75, 76), NFIL3 (77), and IRF8 (28, 71, 78–80) for their development, while cDC2s, but not cDC1s, are dependent on RELB (81), RBPJ (82), and IRF4 (79, 83–85). pDC development has been shown to be driven by E2-2 (86, 87).

The subdivision of DCs in three distinct Level1 groups is thus supported by their gene-expression profiles, cellular origin, and transcription factor requirement. However, these cells can acquire a distinct functional activation state from one tissue to another and in distinct inflammatory settings, underlining the need for a Level2 system. This can be illustrated by the capacity of intestinal cDCs to produce retinoic acid and promote the generation of induced regulatory T cells (iTREGs) (88–91). Identification of DCs with superior iTREG inducing ability is clinically relevant as the prevalence of food allergies, celiac disease and inflammatory bowel diseases is currently rising throughout the western world. Originally, it was described that CD103+ but not CD103**−** intestinal DCs excel in iTREG generation in a retinoic acid-dependent manner (89, 90). It is now clear that CD103+ intestinal DCs comprise two ontogenically distinct subsets, CD103<sup>+</sup>CD11b**<sup>−</sup>** cDC1s and CD103<sup>+</sup>CD11b<sup>+</sup> cDC2s (74). Interestingly, rather than being associated with either of the two subsets, about half of the intestinal CD103<sup>+</sup>CD11b**<sup>−</sup>** cDC1s were shown to possess the capacity to produce retinoic acid, while only one-third of the cDC2s do (91). Moreover, on a per cell basis retinoic acid producing CD103<sup>+</sup> cDC1s were the best at inducing iTREGs (**Figure 4**). These data reveal that CD103<sup>+</sup> cDC1s, although broadly considered as a homogeneous subset, consist of 50% cells that are very efficient at inducing iTREGs and 50% cells that are not. Interestingly, dermal cDC2s have higher retinoic acid-dependent iTREG induction activity than dermal cDC1s (91). We hypothesize that this functional heterogeneity may be explained by the existence of distinct micro-environments within organs, inducing diverse functional modules on DCs. The finding that important functional modules can be acquired by only a fraction of cDC1s and/or cDC2s, which can moreover differ from one organ to another, illustrates the need for a Level2 nomenclature for DCs.

Another example of functional heterogeneity within DCs concerns the cDC2s. Splenic cDC2s contain a subpopulation that expresses CD4 and is specifically localized in the bridging channels (92). This localization has been shown to be EBI2-driven and essential to drive antibody production by B cells. The development of these CD4<sup>+</sup> cDC2s is Notch2 dependent. Note also that Notch2 deficiency is associated with defects in TH17 induction (93, 94). In addition, it was found that KLF4 controls the development of a subpopulation of CD24loCD11bloSirpαhi cDC2s in the dermis (95). Importantly, loss of KLF4 was associated with loss of TH2 induction. Thus, although cDC2s have been proposed to excel at both the induction of TH2 (47, 96) and TH17 responses (84, 85, 93), it may well be that these functional modules are in fact expressed by subpopulations of cDC2s (controlled by KLF4 and Notch2, respectively). In conclusion, although the current knowledge of early DC development in the bone-marrow seems to support only three big groups of DCs (cDC1s, cDC2s, and pDCs), it appears that a second layer of tissue-specific signals imprint operative gene modules on a fraction of DCs. Depending on their micro-localization, subpopulations of cDC1s, cDC2s, or pDCs will acquire distinct functional properties, requiring a flexible Level2 to classify and describe functionally distinct subpopulations.

A final example of a need for a Level2 classification involves inflammation-induced changes of surface marker expression. When mice are infected with the influenza virus, there is a transient change in the CD103 versus CD11b expression profile of lung cDCs, yielding four instead of two lung DC subsets (**Figure 5**). If one considers these as four distinct DC subsets, one could conclude that influenza infection disrupts hematopoiesis in the bone-marrow, as has been shown for Toxoplasma infection (97). Alternatively, these novel CD103/CD11b expression patterns may represent distinct local activation states of cDC1s or cDC2s. We have studied the cellular origin of the "novel" DC subsets arising during influenza infection (Neyt et al., manuscript in preparation). Our preliminary data suggest that CD103<sup>+</sup>CD11b<sup>+</sup> cells are cDC2s that acquire CD103 expression during inflammation rather than a completely new subset. In conclusion, although we cannot rule out the existence of additional DC subsets that specifically develop during inflammation, when in doubt we propose to first evaluate whether cells with a novel surface receptor expression profile represent a different activation state of cDC1s or cDC2s before assuming the existence of a novel cDC3.

# Practical Implementation for Embryonic Macrophages

In our classification system based on ontogeny all mononuclear phagocytes of embryonic origin are grouped together under a single Level1 as "macrophages" (**Figure 1**). This includes liver-resident Kupffer cells, brain-resident microglial cells, lungresident alveolar MΦs but also epidermis-resident Langerhans cells. In effect, this would thus classify Langerhans cells as MΦs and not as DCs, based on the fact that these cells derive from embryonic precursors that seed the epidermis around birth and then self-maintain throughout life (43, 98). We propose to keep the historical names for MΦs with undisputed identities. Mouse Kupffer cells, for example, do not require a different nomenclature since these cells have a well-defined cellular origin [embryonic (34, 38, 39)], localization (i.e., the liver sinusoids), and gene-expression profile (99, 100). However, we would like to emphasize that not any F4/80<sup>+</sup> cell in the liver should be categorized as Kupffer cell. MCs infiltrating

the liver during acetaminophen-induced injury also express F4/80 but are short-lived and acquire a gene-expression profile that is strikingly different from resident Kupffer cells (100). Similarly, MCs infiltrating the central nervous system during inflammation are short-lived and do not acquire the specific gene-expression profile of embryonic microglia (101–103). As such, any MΦ-like cell in the liver or the brain should not be classified as Kupffer cell or microglia, respectively, as is often the case. Unfortunately, tools to correctly distinguish MCs from resident MΦs have long been lacking. In a way, this is surprising given the huge difference in gene-expression profile between resident embryonic MΦs and recruited MCs in these disease models. We have now identified several surface markers that are expressed by Kupffer cells but not MCs recruited during liver injury (Scott et al. manuscript in preparation) and we expect that given the striking heterogeneity of tissue-resident MΦs (104, 105) many of these MΦ-specific markers will be found. This will facilitate the correct classification of these cells and pave the way toward unraveling the functional differences between recruited MCs and tissue-resident embryonic MΦs during inflammation.

# Practical Implementation for Monocyte-Derived Cells

Monocytes are particularly plastic cells. This can be appreciated using *in vitro* culture systems. Monocytes cultured with GM-CSF express some DC-like characteristics and have therefore long been referred to as moDCs. By contrast, culturing monocytes with M-CSF induces their differentiation into MΦ-like cells (moMΦs). Adding IL-4 or IFN-γ to M-CSF cultures further polarizes MCs into the so-called "classically activated MΦs/M1s" or "alternatively activated MΦs/M2s" (106), with strikingly different gene-expression profiles and metabolic modules (107). In a nomenclature system based on ontogeny, moDCs, M1s, or M2s are however first classified as MCs (Level1). In theory, this does not prevent further Level2 classification as "dendritic MCs," "classically activated MCs," or "alternatively activated MCs." However, we feel this polarized classification implies functional characteristics that are often not assessed experimentally. For example, MCs classified as M1s are typically associated with pathogen killing, M2s with wound healing, and moDCs with antigen-presentation (**Figure 6**). However, the identification of "dendritic MCs/moDCs," "classically activated MCs/M1s," or "alternatively activated MCs/M2s" *in vivo* turned out to be very challenging. In fact, profiling of MCs isolated from various inflamed tissues or *in vitro* culture systems reveals that monocytes can acquire a much broader transcriptional repertoire than suggested by the three-way M1/ M2/moDC model. In recent efforts to further characterize the heterogeneity of MC activation states, Schultze and colleagues compared the gene-expression profile of MCs stimulated with a vast array of cytokines and TLR ligands. Instead of yielding a polarized model, the unbiased bio-informatics-driven clustering approach revealed a spectrum model (54). In our view, this spectrum model can be taken one step further and include the "dendritic MCs/moDCs" derived from GM-CSF-induced bone-marrow cultures as yet another extreme of the continuum of cellular faiths that can be acquired by monocytes. Rather

lungs are shown. The appearance of CD103+CD11b+ DCs and CD103−CD11b− DCs is transient as schematically represented.

than unique end points, bacterial killing, wound-healing, and antigen-presentation represent three of many functional modules that can be acquired by MCs in a spectrum model that can be graphically represented by a continuous circle (**Figure 6**).

One important consequence of the herein-described classification would be the regrouping of moDCs and moMΦs under a single MC Level1. We feel this will represent an improvement for the field due to the lack of clear, mutually exclusive features

that can be used to objectively separate moDCs from moMΦs (53). This problem can be illustrated by the MCs present within inflamed mesenteric lymph nodes during an experimental model for colitis (45). In this study, we called these cells moMΦs because they were CD64hiF4/80hi and excelled at phagocytosis (**Figure 7**). However, we could perfectly, like the Powrie group (108), have classified these cells as moDCs based on their CD11chiMHCIIhi profile, their localization within the T cell zone, their antigenpresentation capacity, or their dendritic morphology (**Figure 7**). By classifying these cells as CD64hiF4/80hi and/or CD11chiMHCIIhi MCs, they are recognized as one lineage, which will promote understanding and simplify communication between different research groups without preventing the study of DC-like or MΦlike properties of specific MCs.

Finally, we do not expect MCs to be homogeneous in inflamed tissues. We and others found iNOS, the enzyme that is used by MΦs to produce NO and that is classically associated with an M1 activation state, to be typically expressed by only 10% of MCs (11, 45, 109). Since NO is bactericidal, suppressive for T cells, and can induce serious tissue damage, it makes perfect sense to study the factors that induce the expression of iNOS on a fraction of MCs. But by classifying these cells as "iNOS<sup>+</sup> MCs" instead of "M1 macrophages" or "TIP-DCs" one avoids to associate functions that have not been proven for these cells such as the antigenpresentation activity typically expected from DCs [in fact mice lacking monocytes showed identical T cell priming suggesting that TIP-DCs are not essential for this function (11)].

# The Tough Cases Part I: moDCs as Fourth DC Subset?

MCs fitting the complete list of characteristics attributed to moDC, including migratory and antigen-presentation capacities comparable to cDCs, are not easily identified *in vivo*. We have described migratory MCs upon house-dust mite (HDM) exposure in the lungs (47), but their migration is much less efficient as compared to cDCs and required very high (and non-physiological) doses of HDM. In fact, we found that the majority of HDM-induced pulmonary MCs are not migratory cells but instead play an important role in the secretion of inflammatory chemokines that orchestrate the local immune responses. Similarly, a low-grade migration of CCR2<sup>+</sup>CD64int MCs was described upon DSS inflammation in the skin but this was minor as compared to cDC migration (48). Moreover, compared to cDCs these dermal CCR2<sup>+</sup>CD64int MCs displayed a rather modest antigen–antigen presentation capacity.

The most convincing pieces of evidence for cDC-like features of MCs come from *in vitro* culture systems. Bone-marrow cells cultured with GM-CSF yield cells with excellent antigenpresentation capacity that acquire CCR7, the chemokine receptor controlling migration of cDCs from tissues to lymph nodes, upon TLR stimulation and can migrate to the lymph nodes upon *in vivo* transfer (110, 111). This culture system has been used in many labs and is globally accepted to yield a homogeneous population of moDCs. This concept was first challenged by a study using single-cell transcriptomics (112). Among LPS-stimulated GM-CSF-induced bone-marrow-derived moDCs, the majority of cells were found to show high expression of inflammatory genes such as TNF, IL1, and CXCL10, while a smaller subset had much lower expression of these genes but displayed a signature reminiscent of "mature DCs," including high expression of CCR7 (113, 114). This was originally interpreted as functional heterogeneity among moDCs. However, in what we consider a landmark paper, Reis e Sousa and colleagues now demonstrate that this minor "mature" population in fact represents cDC2s that contaminate these cultures. These cDC2s displayed lower production of inflammatory cytokines but much better CCR7-ligand-induced migration and antigen-presentation as compared to GM-CSF-induced MCs (115, 116). This implies that many of the DC-like features of GM-CSF-induced moDCs should in fact not be attributed to MCs, but to a minor contaminating cDC2 population. All in all, both *in vitro* and *in vivo* data thus point toward a lower migration and antigen-presentation capacity of MCs as compared to cDCs, but conversely a higher production of inflammatory cytokines and chemokines. We therefore propose that in an inflamed organ the core business of cDCs will be migration to the draining lymph nodes and activation of naïve T cells, whereas MCs will primarily orchestrate local inflammatory responses. Note that this has important consequences for DC-based vaccination strategies as this may explain why MC-based vaccines have only yielded modest clinical responses (117). The recent advances in cDC culture systems and the proper identification of committed circulating DC-precursors (31, 32, 118) may therefore pave the way toward more efficient cDC-based vaccination strategies.

# The Tough Cases Part II: Steady-State MCs Versus Embryonic Macrophages

Most MΦ-like cells present in steady-state tissues are of embryonic origin (33–36, 38, 39, 44). However, puzzling exceptions have been reported. Although embryonic MΦs colonize the intestine and the heart before birth, these cells are thereafter progressively replaced by MCs. Importantly, these cells are relatively short-lived and continuous monocyte-recruitment is required to maintain the MC pool in these tissues (45, 46, 119–121). Similarly, monocytes are continuously recruited to the steady-state dermis (48). Therefore, while in some steady-state tissues, including the lung and the spleen, monocytes have been proposed to remain in an undifferentiated state (37, 122); in others, including the intestine, the skin, and the heart, they acquire a MΦ-like phenotype. The classification of MCs that differentiate in these steady-state organs and that replace the embryonic MΦs is difficult. They do not fit the profile of the MCs that are recruited to inflamed tissues, including pulmonary infection (36), auto-immune brain inflammation (101, 102), and acute liver injury (100), since in these inflammatory settings MCs do not replace the embryonic MΦs and display a very different gene-expression profile. Future research will be required to compare the functional properties and gene-expression profile of the embryonic MΦs present in the intestine, the skin, and the heart to the ones from the MCs that replace them with time. It will be interesting to compare the influence of tissue-imprinting to the intrinsic differences associated with their distinct cellular origin. Embryonic MΦs were recently compared to their bone-marrow-derived counterparts that replace them after irradiation-induced depletion. It was found that both cells share between 50 and 90% of the tissue-specific epigenetic landscape (105). This emphasizes the importance of tissue-imprinting, but at the same time implies that between 10 and 50% of the epigenetic landscape could be governed by the cellular origin of the cells. Future research will be required to assess the functional relevance of these findings. In the meantime, the classification of MΦ-like MCs in steady-state tissues remains difficult.

# The Way Forward

The Level1 that forms the scaffold of the herein proposed classification system is in part based on elegant murine fate-mapping systems developed to study the cellular origin of MΦs (33, 34, 38, 39, 43, 44) and DCs [(123) and (124) in this issue]. Although the recent identification of committed DC-precursors distinct from monocytes in humans suggests that many of the principles identified in mice apply to the human immune system, this remains to be formally proven. Moreover, many of these murine fate-mapping systems label only a small fraction of the cells per population, rendering functional studies difficult. In cases where classification as cDC1, cDC2, pDC, MC, or MΦ is not obvious, a core set of signature genes that are specific for each cell type could facilitate correct Level1 identification. However, such signature genes are not easily identified. In addition, identification based on surface receptors would be most practical since it would allow the sorting of living cells through flow cytometry for functional assays. Ideally, such markers would be conserved across species. We are currently data-mining the gene-expression profiles of cells from various tissues and species to try to identify such markers. This can however represent a catch22. To find markers specifically expressed by the different populations, one requires pure geneexpression profiles, but correct sorting of the cells without contamination by other populations for RNA profiling requires the very markers we are looking for. Recent technological advances in single-cell RNA sequencing will allow to profile the gene expression of mixed populations. This may at last disentangle mixed myeloid populations and will hopefully provide the field with new markers that can then be validated with the current fate-mapping systems.

Although the current classification systems should thus be seen as work in progress, we are confident that in the near future better markers will be found which faithfully translate the cellular origin of cells and will form a practical base for the Level1 classification of myeloid cells. The Level2 classification should in our view be kept as flexible as possible to allow researchers to focus on one particular functional attribute of their cells of interest without implying too many additional features that have not been studied. Finally, it is noteworthy that in parallel to our proposition a nomenclature system for MΦs was proposed (106). In this proposition, terms implying functional specialization such as "classically activated macrophages" (pro-inflammatory) or "alternatively activated macrophages" (anti-inflammatory) were replaced by an objective description of how a MΦ is cultured *in vitro* [e.g., MΦ (IL-10)] or identified *in vivo* (e.g., "Relmahi MΦ"). Thus, the common core Level1 would be MΦ and the added description provides the Level2. This and our classification system thus share three important principles: (i) elimination of terms that imply functional specialization as much as possible, (ii) introduction of a fixed Level1 system across species and tissues, and (iii) permitting flexibility through a Level2 system. Irrespective of which system is used to define the Level1 [ontogeny as we propose, or gene-expression profile as proposed by the Dalod group in this issue (60, 61)], we feel these three principles should be maintained for a future and hopefully definitive classification system.

# Acknowledgments

We thank Bernard Malissen, Hugues Lelouard, Bart Lambrecht, and Katrijn Neyt for the use of unpublished data. MG is supported by a Marie Curie Reintegration grant, an Odysseus grant, and several FWO grants of the Flemish Government. LL is supported by an EMBO long-term fellowship and a Marie Curie intra-European fellowship.

# References


and the inflamed setting. *Annu Rev Immunol* (2013) **31**:563–604. doi:10.1146/ annurev-immunol-020711-074950


mouse dendritic cell subsets. *Eur J Immunol* (2010) **40**:2089–94. doi:10.1002/ eji.201040498


cells in the spleen and intestine. *Immunity* (2011) **35**:780–91. doi:10.1016/j. immuni.2011.08.013


through distinct mechanisms at steady state and during inflammation. *Immunity* (2014) **40**:91–104. doi:10.1016/j.immuni.2013.11.019


**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 Guilliams and van de Laar. 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.*

# Ontogeny of tissue-resident macrophages

#### *Guillaume Hoeffel\* and Florent Ginhoux\**

*Singapore Immunology Network (SIgN), Agency for Science, Technology and Research (A\*STAR), Singapore, Singapore*

The origin of tissue-resident macrophages, crucial for homeostasis and immunity, has remained controversial until recently. Originally described as part of the mononuclear phagocyte system, macrophages were long thought to derive solely from adult blood circulating monocytes. However, accumulating evidence now shows that certain macrophage populations are in fact independent from monocyte and even from adult bone marrow hematopoiesis. These tissue-resident macrophages derive from sequential seeding of tissues by two precursors during embryonic development. Primitive macrophages generated in the yolk sac (YS) from early erythro-myeloid progenitors (EMPs), independently of the transcription factor c-Myb and bypassing monocytic intermediates, first give rise to microglia. Later, fetal monocytes, generated from c-Myb+ EMPs that initially seed the fetal liver (FL), then give rise to the majority of other adult macrophages. Thus, hematopoietic stem cell-independent embryonic precursors transiently present in the YS and the FL give rise to long-lasting self-renewing macrophage populations.

Keywords: macrophages, monocytes, fetal liver, yolk sac, C-Myb, erythro-myeloid progenitors, hematopoiesis, hematopoietic stem cells

# Introduction

Ilya (Elie) Metchnikoff first described the mechanism of phagocytosis and the cells responsible for this process over a century ago. These professional phagocytic cells were named "macrophages" (from the Greek derivation macro = large and phage = devouring, "large devouring cells"). These were separate from "microphages," which included polymorphonuclear phagocytes (1). Determining the role of macrophages in pathogenic infections was one of the fundamental observations leading to the concept of cellular immunity (2). Through this seminal work, Metchnikoff anticipated the central role of macrophages in tissue inflammation and homeostasis. We recommend an elegant historical review for more details about Metchnikoff 's work by Yona and Gordon in this issue (3).

Since then, the definition of the phagocyte system has been continuously refined, and our understanding of the wide-ranging functions of macrophages has been substantially expanded. It is now clear that, in addition to their classical function in the activation and resolution of tissue inflammation, macrophages also play roles in tissue-specific functions, tissue remodeling during angiogenesis and organogenesis, and wound healing, to name a few (4). Macrophages are exquisitely adapted to their local environment, acquiring organ-specific functionalities during developmental stages and the steady state (4). Macrophages are able to support multiple tissue functions, integrating cues from both the outside environment and their microenvironment to act as rheostatic cells of tissue function. Thus, tissue-resident macrophages represent an attractive target for modern medicine to treat a wide spectrum of diseases in which they have been implicated, including atherosclerosis, autoimmune diseases, neurodegenerative and metabolic disorders, and tumor growth (5–8). Understanding the

#### *Edited by:*

*Peter M. Van Endert, Université Paris Descartes, France*

#### *Reviewed by:*

*Meredith O'Keeffe, Burnet Institute for Medical Research, Australia Jean M. Davoust, Institut National de la Santé et la Recherche Médicale, France*

#### *\*Correspondence:*

 *Guillaume Hoeffel and Florent Ginhoux, Singapore Immunology Network (SIgN), Agency for Science, Technology and Research (A\*STAR), 8A Biomedical Grove, IMMUNOS Building #3-4, Biopolis, 138648 Singapore guillaumehoeffel1@gmail.com; florent\_ginhoux@immunol. a-star.edu.sg*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 22 July 2015 Accepted: 07 September 2015 Published: 22 September 2015*

#### *Citation:*

*Hoeffel G and Ginhoux F (2015) Ontogeny of tissue-resident macrophages. Front. Immunol. 6:486. doi: 10.3389/fimmu.2015.00486*

origin and developmental pathways of macrophages will help to design novel intervention strategies targeting these cells in tissuespecific sites.

A number of observations now indicate that certain macrophage populations derive from embryonic precursors sequentially seeding tissues during development (9–13). Two macrophage progenitors, yolk sac (YS) macrophages and fetal monocytes, have been described in the embryo, but their exact nature and origin were not fully understood until recently (14, 15). Here, we discuss recent developments in our understanding of the origin of adult tissue-resident macrophages, exploring the sequence of progenitors generated during embryonic and adult hematopoiesis. We focus on the relative contributions of YS macrophages and fetal or adult monocytes, including a discussion of our own recent data exploring the heterogeneity of fetal monocyte developmental pathways.

# Early Concepts

Macrophages form part of the mononuclear phagocyte system (MPS), which also includes circulating monocytes and dendritic cells (16). Until recently, our vision of macrophage origin and homeostasis was largely based on seminal studies that used *in vivo* radioisotope labeling and radiation chimera experiments. These studies led to the early dogma that resident macrophages were constantly replenished from circulating bone marrow (BM) derived monocytes as a continuum of differentiation (17–19).

In agreement with that concept, studying the ontogeny of the MPS revealed that monocytes and macrophages derived from macrophage and dendritic cell progenitors (MDPs) present in the BM, which are phenotypically defined as lineage-c-kit<sup>+</sup>CX 3CR1<sup>+</sup>Flt3<sup>+</sup>CD115<sup>+</sup> (20). MDPs further differentiate through a newly described common monocyte precursor (cMoP), phenotypically defined as lineage<sup>−</sup>c-kit<sup>+</sup>CX3CR1<sup>+</sup>Flt3<sup>−</sup>CD115<sup>+</sup> (21), that gives rise to the two main subsets of circulating monocytes distinguished by the expression of Ly6C (22).

Specific tissue macrophages, such as dermal, gut, and heart macrophages, seemed to follow the model of Van Furth, that macrophages are derived from monocytes (23–25). However, this model did not fit in all cases and evidence also emerged indicating that macrophages were long-lived cells, able to self-renew locally. Hashimoto was the first to speculate that Langerhans cells (LCs) represented a self-perpetuating "intraepithelial phagocytic system" (26). Performing a human skin transplantation assay onto nude mice, Krueger et al. described the remarkable longevity of LCs, which were able to persist in the grafts for more than 2 months (27). Their ability to self-renew through proliferation was later described using DNA densitometry (28). Similar conclusions were drawn soon after regarding alveolar macrophages (29). The dominant concept of "the monocytic origin" of tissue macrophages was also challenged through experiments in animals with prolonged monocytopenia following strontium-89 monocyte depletion, in which liver Kupffer cells were shown to maintain cell numbers by increasing local proliferation (30, 31).

More recently, the use of long-term parabiotic mice and subsequent fate-mapping models have challenged the MPS paradigm and revealed that, unlike all other hematopoietic cells, which rely on hematopoietic stem cell (HSC)-derived BM progenitors, certain macrophage populations possess the unique ability to self-renew locally independently of circulating precursors (32–36). Initial studies describing the presence of macrophages in embryonic tissues suggested that tissue macrophages derived from embryonic progenitors. In rodents, macrophage-like cells first described in the brain rudiment and in the developing skin (37, 38) were named "fetal macrophages" and found to exhibit a high capacity for proliferation (39). These observations suggested that adult macrophages derive from fetal macrophages established during early development. However, whether these fetal macrophages were maintained until adulthood or were replaced postnatally was not addressed until recently. In addition, the exact nature and the origin of fetal macrophage progenitors remained unclear.

# Embryonic Hematopoiesis

Mammalian embryos produce several transient waves of hematopoietic cells before the establishment of HSCs in the BM during late gestation (40, 41). The multiple embryonic waves are differentially regulated in time and space and exhibit distinct lineage potentials. Importantly, they contribute to hematopoietic populations that persist until adulthood. These waves include primitive hematopoiesis in the YS, and definitive hematopoiesis, which comprises a transient definitive stage, generating multi-lineage erythro-myeloid progenitors (EMPs) and lympho-myeloid progenitors (LMPs), and a definitive stage characterized by the production of HSCs in the aorta-gonad-mesonephros (AGM). These transient progenitors establish themselves transiently in the fetal liver (FL) during the mid to late stages of hematopoiesis. The sequential waves of hematopoiesis can overlap in time and space (**Figure 1**) and remain difficult to separate clearly, even with the most recent fate-mapping tools available.

#### Primitive Hematopoiesis

In mice, the first hematopoietic progenitors appear in the extra-embryonic YS blood islands at around embryonic age 7.25 (E7.25), where primitive hematopoiesis is initiated, producing mainly nucleated erythrocytes. This observation linked the myeloid progenitors observed in the YS at E7 with the emergence of YS macrophages after E9.0 [**Figure 2**; Ref. (42–45)]. Primitive hematopoiesis was also shown to produce megakaryocyte progenitors (46). The denomination "primitive" was given to reflect the production of embryonic erythroblasts, like those observed in lower species such as fish, amphibians, and birds, and remaining nucleated throughout their life span (47–49). This denomination was extended to macrophages in the YS due to their concomitant development prior to FL hematopoiesis. Interestingly, no clear evidence of monocytic intermediates was reported at this stage, although the seminal study of Cline and Moore did mention the existence of local intermediate cells between progenitors and functional macrophages (43). Studies by Naito and Takahashi et al. clarified the emergence of primitive macrophages in the YS blood islands in the mouse and rat, observing an absence of endogenous peroxidase activity as a surrogate marker for an absence of monocytic intermediates, such as those found in the

BM (50–52), suggesting a unique developmental pathway for YS macrophages (53, 54).

#### Transient Definitive Hematopoiesis

the placenta.

The quest to elucidate the origins of embryonic HSCs led to the discovery of earlier lineage-restricted HSC-independent progenitors seeding the FL at E10.5. These progenitors arise concurrently with the transition of primitive to definitive erythropoiesis and were thus considered to form a transient stage of definitive hematopoiesis (45, 47, 55). Transient definitive hematopoiesis consists of progenitors sequentially acquiring myeloid, then lymphoid potential, without exhibiting the long-term reconstitution potential of HSCs. Seminal work from Palis and colleagues on embryonic erythropoiesis in the YS described the parallel emergence of multiple myeloid lineage potential progenitors from E8.25 in the YS (45). Palis et al. first observed the emergence of definitive progenitors for mast cells and a bipotential granulocyte/macrophage progenitor. These progenitors then migrated to the FL through the bloodstream after E8.5, once circulation was established (44, 56). From this pattern of development, the authors concluded that definitive hematopoietic progenitors arise in the YS, migrate through the bloodstream, and seed the FL to rapidly initiate the first phase of intra-embryonic hematopoiesis. Similarly, primitive and definitive erythropoiesis, associated with myelopoiesis, was also shown to emerge prior to HSC in the zebrafish embryo (57). Bertrand et al. showed that definitive hematopoiesis initiates in the posterior blood island with only transient proliferative potential. Because these HSC-independent definitive progenitors were observed to produce definitive erythroid and myeloid

local macrophage population, likely important for sustaining enucleation of primitive erythrocytes passing through the sinusoid prior to the establishment of definitive

cell types, but not to colonize the zebrafish thymus (implying that they are devoid of lymphoid potential), this population was termed EMPs (57). Interestingly, EMPs can also emerge from the hemogenic endothelium (HE) located in the placenta and umbilical cord (58) and colonize the FL from E9.5 (55) to participate in definitive hematopoiesis. Further studies advanced the field significantly by identifying CD41 as an early marker (pre-CD45) for defining hematopoietic progenitors, including EMPs, emerging from the YS (59, 60). Altogether, these important studies provided phenotypic and functional analyses of the first hematopoietic progenitors and demonstrated that definitive hematopoiesis proceeds through two distinct waves during embryonic development (**Figure 3**).

hematopoiesis and the generation of fetal monocyte-derived macrophages in the fetal liver.

In parallel, several groups have also identified other multipotential progenitors with lymphoid- or myelo-lymphoid-restricted potential in the YS and the developing para-aortic splanchnopleura (P-Sp) prior to HSCs (61, 62). Lacaud et al. also described AA4.1 (CD93)<sup>+</sup> multipotential progenitors present in the E14.5 FL with T cell, B cell, and macrophage potential (63), although the precise origin of these progenitors was not addressed. At the same time, the team of Jacobssen identified Flt3<sup>+</sup> lympho-myeloid progenitors (LMPs) devoid of erythrocyte and megakaryocyte capacity (64). Later, cells with myelo-erythroid and lymphoid lineage potential, such as B-1 cells present in the adult spleen, were associated with E9.5 YS progenitors expressing AA4.1 and CD19 (65). A year later, the same team also identified T cell potential within the E9.5 YS progenitors (66). Using a Rag-1-Cre fate-mapping model, Boiers et al. confirmed that these LMPs emerged at approximately E9.5 in the YS, seeding the FL by E11.5 to give rise to T and B cells, as well as granulocytes and monocytes in the E14.5 FL, prior to HSCs (67). Finally, lympho-myeloid progenitors isolated from the dorsal aorta at E9.0 were shown to acquire long-term reconstitution capacity after a few days of *in vitro* culture with stromal cells, and were called immature HSCs (68). Without preculture, these multipotential progenitors can only engraft natural killer (NK)-deficient Rag2γc−/− mice.

monocytes.

maintain the capacity to self-renew throughout adulthood in certain tissues, such as the liver or the lung, where they will not be replaced by adult BM-derived

Whether these immature HSCs arise from LMPs or represent a distinct wave of progenitors remains to be clarified. However, these seminal studies provided strong evidence that lymphoid potential can emerge from the YS, prior to HSC-budding from

the AGM (69). Because the emergence of EMPs and LMPs overlaps in time and space, they could not be distinguished clearly until recently. Previous reports had suggested that lymphoid potential was restricted to the CD41-negative cell fraction (59, 65). However, CD41 is also expressed in a sub-fraction of FL HSCs, and so this phenotypical distinction spread some confusion (47). Finally, a recent report from the group of Palis clarified this point by showing that co-expression of c-kit, CD41, and CD16/32 defines EMPs and allows their separation from other progenitors with lymphoid potential, such as those giving rise to the B-1 cell (70). McGrath et al. extended the notion of EMPs by showing their potential to generate neutrophils, megakaryocytes, macrophages, and erythrocytes. Finally, transplantation of EMPs in immunecompromised adult mice can also provide transient adult red blood cell reconstitution (70).

To conclude, commitment to hematopoietic fates begins during gastrulation in the YS, which represents the only site of primitive erythropoiesis and also serves as the first source of transient definitive hematopoietic progenitors. HE develops from the YS to various intra-embryonic sites, and acquires myeloid and then lymphoid lineage potentials in overlapping waves, highlighting the complexity of the hematopoietic output. Whether some of these progenitors arise from independent sources or represent different maturation stages of a shared hematopoietic wave, culminating with the generation of HSCs, needs to be further clarified. However, it is tempting to speculate that the clear contrasts in differentiation/lineage potential do not reside in their intrinsic potential, but rather in the extrinsic signals provided by the local environment.

#### Definitive Hematopoiesis

The complex hierarchy of stem and progenitor cells in the BM is first established during embryonic development starting with the emergence of small numbers of HSCs from the AGM at E10.5 in murine embryos or at 5 weeks in human embryos (71, 72). After E9.5 in the mouse, with the determination of the intra-embryonic mesoderm toward a hematopoietic lineage, new waves of hematopoietic progenitors emerge within the HE of the embryo proper (**Figure 4**), first in the P-Sp region and the umbilical and vitelline arterial regions of the embryo, then in the AGM region and the placenta (55, 73, 74). The hematopoietic activities of the P-Sp and AGM first generate immature HSCs and then mature HSCs, which are defined by their capacity to reconstitute adult conventional mice (long-term reconstitution; LTR). Both immature and mature HSCs seed the FL at approximately E10.5 (68, 71, 75, 76) to establish definitive hematopoiesis (40, 77, 78). A maturation step seems necessary for immature HSCs to express their LTR activity in full, which is then maintained until adulthood (68)**.** However, further investigations using a fate-mapping system would be necessary to confirm this model.

The FL becomes the major hematopoietic organ after E11.5, generating all hematopoietic lineages. Importantly, the FL itself does not produce progenitors *de novo*, but rather recruits progenitors derived from the YS and other hemogenic sites, to initiate definitive hematopoiesis (79) in parallel with the expansion of the definitive HSC population before their migration to the spleen and BM (80).

The contribution of HSCs to FL hematopoiesis is complex to evaluate, partly because of the lack of specific fate-mapping models, and also the relatively limited knowledge regarding embryonic HSC maintenance and homeostasis in this environment. The capacity for long-term reconstitution, which defines functional HSCs, is present in the AGM by E10.5 (76). However, lineage-specific commitment may not occur *in vivo* immediately after reaching the FL environment. A number of other progenitors generated during transient definitive hematopoiesis, as discussed above, are already present and able to give rise to almost all cell lineages, which could prevent HSC consumption and differentiation (**Figure 5**). Evaluation of HSC contribution has long been based on the assumption that all hematopoietic cells in the FL were derived from HSCs as is the case in the BM (81). Many multipotential progenitors share the same phenotype with pre-HSC and HSCs, such as the expression of CD41 and AA4.1 (60), adding to this confusion. The combination of the marker Sca-1 and new markers such as those from the SLAM family (82) have greatly helped to clarify the characterization of HSCs, defined now as Lin<sup>−</sup>ckit<sup>+</sup>Sca-1<sup>+</sup>CD150<sup>+</sup>CD48<sup>−</sup>CD244<sup>−</sup>. However, no specific fate-mapping model exists to characterize embryonic HSC progeny with the exception of the Flt3-Cre

FIGURE 4 | Transition between fetal and adult hematopoiesis. Hemogenic endothelial cells from extra and intra-embryonic hematopoietic tissues generate C-Myb-dependent multipotential progenitors, such as LMPs and pre-HSCs, between E9.0 and E10.5, culminating with the emergence of mature HSCs with long-term reconstitution-bearing potential. CD93 (AA4.1) expression is associated with the emergence of lymphoid potential, whereas Sca-1 is the hallmark of HSCs. These progenitors seed the fetal liver around E10/E11, expanding and giving rise to the various lineages of the hematopoietic system, including fetal monocytes. These late fetal monocytes continue to participate in the tissue-resident macrophage network until hematopoiesis switches completely from the fetal liver to the bone marrow. Once adult hematopoiesis begins to take place in the bone marrow generating monocytes, certain tissues, such as the dermis, heart peritoneum, and the gut, continue to recruit adult monocytes to generate resident macrophages and replace with time the embryonic-derived macrophages.

model (83), which was used until now with the assumption that embryonic and adult HSCs follow similar differentiation pathways. Our recent report suggests that the Flt3-Cre model can also be used to follow the progeny of LMPs (15). Furthermore, in the nascent BM, the long-term repopulation (LTR) capacity that characterizes functional HSCs is only observed at around E17.5 (84). Considering the time required to initiate full HSC differentiation, these data suggest that proper adult HSC-derived hematopoiesis does not take place in the BM until a few days after birth. Characterization of the functional specificities and regulatory pathways of HE that give rise to HSCs versus those that generate EMPs and other multipotential progenitors could aid the development of new fate-mapping models and improve our understanding of this process (85). Use of other fate-mapping models such as the Runx1-Mer-Cre-Mer (Runx1-iCre) (86), Tie2-Mer-Cre-Mer mice (14), and the c-kit-Mer-Cre-Mer mice (87) provided complementary results, although a careful analysis of the targeted cells in time and space is not yet fully available for the last two models. We present here our best interpretation of the data provided in these two recent studies that have used these models in light of the literature and our own results and experience using the Runx1-iCre model (**Figure 6**).

# Embryonic and Adult Precursors of Adult Tissue-Resident Macrophages

# Yolk Sac Macrophages

Yolk sac macrophages first appear in the YS blood islands at E9 (albeit in small numbers) with a unique pattern of differentiation that bypasses the monocytic intermediate stage seen in adult macrophages (50, 52). YS-derived primitive macrophages spread into the embryo proper through the blood as soon as the circulatory system is fully established (from E8.5 to E10) (56), and migrate to various tissues, including the brain. Importantly, this occurs before the onset of fetal monocyte production by the FL, which starts around E11.5/E12.5 (92). These primitive macrophages retain the high proliferative potential observed in the YS as they colonize various tissues (52, 93–95). Primitive macrophages may contribute to many fundamental processes during mid and late embryogenesis, such as clearance of dead cells or tissue maturation. In this regard, the developmental process of interdigital cell death removal during the mouse footplate remodeling that occurs between E12.5 and E14.5 is of interest as the interdigit regions become heavily populated by macrophages and most of the dead cells were shown to be rapidly engulfed by macrophages (96). However, mouse models devoid of primitive macrophages such as the colony-stimulating factor 1 receptor (CSF-1R) KO (Florent Ginhoux, unpublished data,) and PU.1 KO (97) appears to exhibit a normal interdigit web tissue. Wood et al. observed that interdigit web tissue in PU.1 KO was only slightly retarded, suggesting that other cell type such as neighboring mesenchymal cells were compensating (97). In addition, we recently showed that depletion of primitive macrophages and hence of embryonic microglia, affected the progression of dopaminergic axons in the forebrain and the laminar positioning of subsets of neocortical interneurons, likely through phagocytic mechanisms (98).

Schulz et al. highlighted further differences between primitive and definitive hematopoiesis, showing that the latter relies on the transcription factor Myb, while YS-derived macrophages are Myb-independent, and are instead dependent on PU.1 (12). This again reinforces the view that YS-derived macrophages constitute an independent lineage, distinct from the progeny of definitive HSCs. Schulz et al. exploited the differential dependence of primitive versus definitive hematopoiesis on the transcription factor c-Myb and reported that E16.5 tissue macrophage populations were not affected by the loss of c-Myb. Using a CSF-1R-iCre fate-mapping model of YS macrophages, they also reported the persistence of YS macrophages progeny in adult tissue-resident macrophage populations (lung, liver, and pancreas, as well as in the brain and skin), although the level of labeling was minimal (below 3–5%) and decreased with time. The authors concluded that tissue-resident macrophages were therefore derived from a c-Myb-independent lineage via YS macrophages (12), data supporting the initial report showing that microglia arise from

FIGURE 6 | Fate-mapping systems. (A) The Runx-iCre fate-mapping model (86) used in our study targets the hemogenic transition (88), hence, labeling specifically progenitors in the process of budding out from the hemogenic endothelium. The Runx1 expression decreases in progenitors once they start to express Vav thus reducing the chances of tagging released progenitors from precedent waves (88). As a consequence, Runx-iCre tagging is restricted to a short time window in the lifespan of a given progenitor and allows a sharp definition of each hematopoietic wave. However, this model also restricts the tagging to only a small fraction of the targeted progenitor wave. (B) Tie2 is expressed in all endothelial cells that constitute the hemogenic endothelia even before the hemogenic transition (89). Thus, all endothelial cells and their progeny (non-hematopoietic and hematopoietic cells) will be labeled after tamoxifen injection using the Tie2-iCre model. As a consequence, an early tamoxifen injection (such as at E7.5) will result in the tagging of all hematopoietic cells emerging before the time of analysis. This will include progenitors from the primitive, the transient definitive, and the definitive waves if, for example, the analysis is done at E11.5. A late injection (such as at E10.5) will restrict the tagging to only the latest hematopoietic stem cells wave as they are just budding from HEs (90). Thus, this model might not be suitable to clearly separate the primitive from the transient definitive waves of hematopoiesis. However, this model could be important to study late HSC progeny as no other progenitors than HSCs emerge from HE after E10.5 (91). (C) C-kit is expressed by all hematopoietic progenitors and does not label endothelial cells that constitute the HEs (89). An early tamoxifen injection (such as at E7.5) will restrict the labeling to early progenitors making suitable the c-kit-iCre model to study the primitive hematopoiesis. However, the FL recruits progenitors of each hematopoietic wave from E8.5 until E11 (79). These progenitors still express c-kit and coexist after seeding the FL during the time necessary for their differentiation (47, 55). A later tamoxifen injection (such as at E9.5) might thus result in the cumulative labeling of undifferentiated primitive and definitive progenitors, including the transient wave of EMPs and LMPs. Thus, such model may not be suitable to resolve the complexity of the different embryonic hematopoietic waves characterized by short time windows of emergence and strong overlapping tendencies. Primitive hematopoietic progenitors are rapidly consumed and the engagement of EMPs and LMPs in FL hematopoiesis reduces the expression of c-kit on their surface. Thus, later tamoxifen injection (such as at E11.5) could restrict the labeling to newly derived HSCs expressing high level of c-kit without labeling precedent progenitor waves (87). Such model might be interesting to study the progeny of late HSCs although the risk of tagging the progeny of EMPs and LMPs or later committed progenitors derived from HSCs remains high and difficult to exclude. Further analysis would be necessary to clarify the potential of such model.

YS macrophages (9). Embryonic origin of macrophages was further supported by the work of Yona et al. and Hashimoto et al. showing that adult monocytes do not substantially contribute to tissue macrophages under steady-state conditions (13, 33). Furthermore, Yona et al. suggested the existence of a CX3CR1<sup>+</sup> precursor for some of the monocyte-independent macrophages, although the exact nature of this precursor was not elucidated. In fact, both YS macrophages and fetal monocytes express CX3CR1 (9, 11, 15) and could therefore correspond to the unidentified precursors suggested by Yona et al. However, using a CSF-1RiCre fate-mapping model, also used by Schulz et al. (12), another study noted that the YS macrophage contribution in the brain, the adult liver, and the heart was maintained although at a minimal level that decreased with time (99). Interestingly, the level of labeling was always higher in microglia than that in the liver or the heart, suggesting that the level of YS macrophage contribution may differ between tissues and that YS macrophages may be differently replaced over time by later waves of progenitors, which follow tissue-distinct kinetics (discussed below). Our own report using the Runx1-iCre fate-mapping model (86) indicated that only microglia, specialized macrophages of the central nervous system, were derived solely from primitive macrophages while all other tissue macrophages derived from definitive hematopoiesis (9).

To understand whether YS macrophages might be the sole progenitors of every other adult macrophages, we asked what impact their *in utero* depletion would have on the subsequent generation of fetal tissue macrophages. CSF-1R is expressed on YS macrophages and fetal monocytes, but only the development of the former is actively dependent on CSF-1R (9, 11). Thus, we attempted to deplete YS macrophages by transiently inhibiting the CSF-1R signaling pathway using a blocking anti-CSF-1R antibody, as recently described (98). Importantly, after complete depletion of primitive YS macrophages in E10.5 embryos and thus of most macrophages in treated embryos at E14.5, tissue macrophages (including microglia) were able to repopulate to normal levels before birth. These data suggest that YS macrophages are dispensable for the generation of tissue-resident macrophages in the embryo, and that another CSF-1R-independent embryonic precursor can functionally replace YS macrophages during development (15, 98). Using a combination of both the CSF-1RiCre and the Runx1-iCre fate-mapping models, we noted that although YS macrophages infiltrate all tissues (including lung, liver, kidney, skin, gut, heart, pancreas, and stomach) until E13.5, a second wave of precursors, with a monocytic morphology and phenotype, supersedes them after E14.5 with the exception of the brain where YS macrophages are maintained until adulthood (15). A fuller understanding of this process may help to resolve some of the earlier discrepancies regarding the contribution of YS macrophages.

### Fetal Monocytes

Fetal monocytes were described by Naito et al. (92). Focusing their study on liver Kupffer cells (the resident macrophages of the liver) during embryonic development, they exploited the endogenous peroxidase activity of monocytes and pro-monocytes granules described earlier by van Furth et al. (18, 92). Naito et al. observed the transient appearance of peroxidase activity, a signature for monocyte and pro-monocyte granule activity, during the *in vitro* generation of macrophages from a preparation of FL-dissociated cells (92). In the YS and at early stages of FL development, no peroxidase activity was observed, suggesting that primitive macrophages first seed the FL. At a later stage, the peroxidase activity increased, suggesting the presence of monocytic intermediates. *In vitro* clonal expansion assays confirmed the existence of two types of colonies, those containing fetal monocytes and those devoid of them. This provided early evidence for the existence of two distinct developmental pathways leading to the generation of Kupffer cells, although at this stage, direct differentiation of fetal monocytes into macrophages *in vivo* had not been demonstrated (100).

To investigate the developmental event leading to the emergence of tissue-specific macrophages, we initially focused on the LC, the specialized myeloid population of the epidermis. While YS macrophages seed the embryonic skin before E13.5, we discovered that the major fraction of adult LCs is in fact derived from fetal monocytes that are generated in the FL from E12.5 and are then recruited into fetal skin at E14.5 (11). These cells share a similar phenotype to their adult counterparts; however, they are generated independently of CSF-1R expression (9, 11). They possess high proliferative potential, and, in contrast to their adult counterparts, express few genes related to pathogen recognition and immune activation (15). Further studies should clarify whether such differences reflect monocyte immaturity imposed by a sterile fetal environment, or rather dedicated functional specializations that have yet to be unraveled. *In utero* adoptive transfers combined with fate-mapping studies unequivocally confirmed *in situ* differentiation of fetal monocytes into adult LCs (11). Fetal monocytes were then demonstrated to be the precursor of adult macrophages in lung alveoli by intranasal injection (10, 101). Fetal monocytes were also shown to be involved in the generation of adult macrophages of the heart (99). In fact, fetal monocytes become the major leukocyte within the blood circulation after E13.5, spreading to all tissues. This occurred independently of the CCL2/CCR2 axis (15), suggesting an alternative mechanism of exit from the FL and/or recruitment by fetal tissues. Moreover, we were able to fate-map, from before birth to adulthood, the local differentiation of fetal monocytes into resident macrophages, by taking advantage of the specific expression of S100a4 in fetal monocytes compared to YS macrophages (15). Only the brain remained free from fetal monocyte infiltration, possibly resulting from the isolation of the brain by the nascent blood–brain barrier as early as E13.5 (15, 102). Thus, these data now reveal that fetal monocytes are the major circulating embryonic precursor for all macrophages, with the exception of the brain. The absence of monocyte precursor contribution to the microglial pool could result from a lack of intrinsic potential or a lack of access to the developing brain due to the nascent blood–brain barrier. Interestingly, we observed a major influx of monocytes in the brain at E14.5 in our YS macrophage depletion model, and preliminary data using our fetal monocyte S100a4- Cre/WT fate-mapping model combined with *in utero* depletion of YS macrophages suggest that fetal monocytes are capable of giving rise to microglia under certain conditions (Hoeffel & Ginhoux, personal communication). Whether this atypical fetal monocyte infiltration reflects a compensatory mechanism to fulfill an empty niche in the brain or results from a disruption of the blood–brain barrier remains to be investigated.

#### Adult Monocytes

BM-derived circulating monocytes were considered the only precursors for all tissue-resident macrophages since the seminal work of van Furth et al. (17–18). Although this dogma was entirely revisited recently with the emergence of sophisticated fate-mapping tools as well as parabiotic models, the physiological contribution of circulating adult monocytes to the adult macrophage network remains valid at least in certain tissues. The continuous recruitment of circulating monocytes to the dermis has been shown to shape the adult dermal macrophage network (25). Although this study did not employ fate-mapping techniques, Tamoutounour's data suggest the existence in the dermis of both a prenatal pool of macrophages and a second pool derived from adult blood monocytes. The authors argue that the dermis, in contrast to the epidermis, continues to recruit circulating monocytes in adulthood, most likely facilitated by its high level of vascularization. The macrophage network in the intestine follows a similar model. Data from Bain et al. suggest that embryonic macrophages do not persist in adulthood in the gut, and are replaced constantly by circulating adult monocytes (23), convincingly showing that adult monocytes are the source of intestine-resident macrophages. The role of commensal microbiota in this process is supported by the observation that the use of germ-free animals or treatment with broad-spectrum antibiotics results in a significant reduction in the recruitment of Ly6C<sup>+</sup> monocytes to the colon (23). The macrophage network of the heart has also been shown to contain a component of YS macrophages and fetal monocyte-derived macrophages, both of which are maintained in adulthood (99). However, similar to the dermis and the gut, adult monocytes seem to replace embryonic macrophages progressively over time (24). The decreasing capacity for self-renewal of embryonic macrophages with age observed by Molawi et al. may explain the requirement for continuous recruitment of monocyte-derived macrophages to the heart in the absence of inflammation. It remains to be clarified whether this phenomenon occurs in other tissues as a result of aging. In agreement, proliferation of YS macrophages and fetal monocytes is very high during development (20–40% before E14.5) but decreases progressively to 10% few days after birth in most tissues and decreases to almost undetectable levels in adults (15). Interestingly, macrophage turnover seems different from one tissue to another. Following BrdU incorporation at steady state, almost no proliferation was observed in adult gut macrophages (23), while 2–5% was measured in adult heart macrophages (24). Macrophage proliferation activity can also be mobilized upon inflammation. For example, peritoneal macrophages can increase their proliferation rate from 1 to 9% in response to parasite infection or in response to IL-4 stimulation (35), while enhanced local proliferation of macrophages in atherosclerotic lesions sustain disease progression (103). The characterization of local signals regulating macrophage proliferation as well as the presence of specialized tissue niches that sustain macrophage survival, proliferation, or even "stemness" will be fundamental to better understand their tissue homeostasis.

The macrophage network of the lymphoid system seems to follow a similar pattern than in the gut and dermis. Although the lymph nodes (LN) start to develop very early in the embryo (104), they become functionally active only within the first week after birth recruiting and organizing B and T cell areas when follicles start to shape with connections to afferent lymphatics via the subcapsular sinus (105). Although macrophages are known to participate in lymphangiogenesis during development, notably by the production of VEGF (106, 107), the precise origin of the different LN macrophage populations remain poorly understood (108). The high level of foreign antigens passing through the LN during the lifespan, support the model of a constant replenishment of the local macrophage pool by circulating adult monocytes. However, the work of Jakubzick et al. suggests otherwise as tissue-patrolling monocytes at steady state seem to enter the LN without any sign of local differentiation to macrophages or dendritic cells (34). Further studies using fate-mapping systems should be addressed to clarify this point. Spleen macrophages are generated prenatally (13, 33). However, red pulp macrophages and marginal zone macrophages seem highly dependent, respectively, on the transcription factor SPI-C (109) and on the nuclear receptor LXR (110), also expressed by circulating monocytes and suggest again that embryonic-derived macrophages are replaced over time by adult monocytes-derived macrophages. The use of the S100a4-Cre fate-mapping model in our hands supports these observations and similar conclusions were obtained for BM and peritoneal macrophages (15). Although tissue microenvironment shapes certain macrophage functional specificities (111), through an ontogenic point of view, the composition of each tissue-resident macrophage pool evolves throughout life and the respective origins of each macrophage population may account for some of their key functions and cellular behaviors in a given tissue. Hence, a new challenge is to understand if an embryonic or adult origin matters for the function and the activation states of tissue-resident macrophages.

# Origin and Development of YS Macrophages and Fetal Monocytes

#### Origin of YS Macrophages

Bertrand et al., in line with the seminal work of Palis (45), described two sequential myeloid waves within the early YS (42). Using an *in vitro* culture reporter system, Bertrand et al. observed a first wave of monopotent progenitors that gave rise only to macrophages, followed by a second wave that gave rise to a mix of granulocytes, monocytes, and macrophages. More recently, Kierdorf et al. revisited the work of Bertrand et al. exploiting organotypic embryonic brain slices to demonstrate that microglial cells derived from YS EMPs (112). Kierdorf et al. also showed that these EMPs did not express the transcription factor c-Myb, associating them with the progenitors reported by Schulz et al. (12), although a direct link with the generation of microglia *in vivo* in adulthood was not conclusively demonstrated. More recently, Perdiguero et al. used the CSF-1R-iCre fate-mapping model to show that YS macrophages are derived from CSF-1R<sup>+</sup> EMPs (14). Hence, these two studies suggest that YS macrophages, and thus microglia, would originate from c-Myb-independent CSF-1R<sup>+</sup> EMPs. Furthermore, Perdiguero et al. demonstrated that CSF-1R<sup>+</sup> EMPs were able to seed the FL by E10.5, suggesting that these progenitors could later populate other tissue niches and produce YS-like macrophage later during development in others tissues. Nevertheless, these data do not explain the low percentage of labeled adult macrophages observed by Schulz et al. using the CSF-1R-iCre fate-mapping model (12). Later observations by Epelman et al. (99), and more recently by our group using the same fate-mapping model (15), indicated that the ability of CSF-1R+ EMP to reach the FL could explain the surprising maintenance of primitive macrophages until E16.5 in c-Myb null embryos, where primitive macrophages generated in the YS as well as in the FL would be able to fulfill the empty niche left by the absence of c-Myb-dependent myeloid cells, that include fetal monocytes. However, this may not reflect the physiological situation and may instead result from a compensatory mechanism to ensure the presence of macrophages in all tissues in the absence of c-Myb activity and fetal monocytes. Using the same CSF-1R-iCre fate-mapping model (15), we were able to follow the maintenance of microglia in the brain by self-renewal from E10.5 until adulthood, linking them with CSF-1R<sup>+</sup> EMPs and confirming the previous observations of Perdiguero et al. (14). However, for all other macrophage populations, the reduction of fate-mapping reporter labeling after E13.5 confirmed the progressive replacement of YS macrophages by another unlabeled precursor arising from a different hematopoietic wave.

We previously showed that Runx1<sup>+</sup> YS progenitors that emerged at E7.5 give rise to YS macrophages and microglia (9, 11). Using both the Runx1-iCre and the CSF-1R-iCre fate-mapping models, we showed that these E7.5 Runx1<sup>+</sup> YS progenitors were in fact the same CSF-1R<sup>+</sup> EMPs described by Perdiguero et al. and Kierdorf et al., which contributed to the generation of YS macrophages and, to a lesser extent, those seeding the FL (14, 15, 112). However, we also observed their disappearance from the FL after E11.5 indicative of a rapid local consumption/differentiation rather than long-term maintenance. Our results also suggest that these early CSF-1R<sup>+</sup> EMPs are able to contribute to a short-term maintenance of macrophages in the FL (**Figure 2**), but do not contribute to other tissue macrophages as evidenced by their rapid disappearance from the blood circulation after E14.5 (15). This transient population in the FL may be due to a local immediate requirement for macrophages, at least during the onset of FL hematopoiesis, to perform efficient enucleation of primitive erythrocytes passing through the FL sinusoids (100, 113). Combining historical evidences showing their direct lineage connection with the emergence of YS macrophages and recent findings showing their independence with c-Myb activity, we propose that CSF-1R<sup>+</sup> EMPs should be designated as primitive EMPs.

#### Origin of Fetal Monocytes

Because adult monocytes are derived from HSCs in the BM, it would be reasonable to assume that embryonic HSCs might also give rise to fetal monocytes in the developing liver. In agreement with this hypothesis, we have identified a population in the FL similar to adult MDPs that have the potential to generate fetal cMoPs and monocytes following *in vitro* culture (15). Exploiting the Flt3-Cre tomato fate-mapping model (83), we then followed the progeny of embryonic HSCs. However, the poor labeling observed between E14.5 and E17.5 in FL monocytes and macrophages contrasted with the strong labeling of FL MDPs, suggesting that HSCs had limited involvement in the generation of fetal monocytes (15). Nonetheless, the limited but significant labeling in fetal monocytes and macrophages at birth suggested an increasing derivation from fetal HSCs, assuming that fetal HSCs follow a similar Flt3-dependent differentiation pathway as adult HSCs. In parallel, gene array analysis highlighted a strong lymphoid signature within fetal MDPs (15), indicative of their derivation from the recently described YS-derived LMPs (67). Thus, LMPs may be important for the generation of a small but significant proportion of fetal monocytes prior to the expansion of mature HSCs (**Figure 4**). Further investigations using more specific fate-mapping models will be necessary to elucidate the exact contribution of LMPs as well as the hematopoietic transition between the FL and the BM.

Importantly, we observed that fetal monocytes were not tagged with the CSF-1R-Cre model that label early CSF-1R<sup>+</sup> EMPs, suggesting that fetal monopoiesis is not dependent on CSF-1R<sup>+</sup> EMPs, consistent with our previous data (9, 11) and with our YS macrophage depletion results (15, 98). Furthermore, the Runx1-iCre fate-mapping model allowed us to identify two waves of EMPs that arise sequentially before LMPs in the YS. These included an early wave, arising at E7.5 that differentiates locally into YS macrophages; and a later wave tagged at E8.5, that migrates and seeds the FL following the establishment of the blood circulation before E9.0. Early EMPs tagged at E7.5 were therefore related to those described previously by Kierdorf and Perdiguero (14, 112). The late EMPs tagged at E8.5, however, expressed c-Myb, expanded more efficiently in the FL, and differentiated *in vivo* into fetal cMoPs, constituting the major component of the fetal monocyte population as well as the fetal monocyte-derived macrophage population (**Figure 3**), which was able to maintain itself in all tissues tested (15).

The existence of two distinct EMP waves is in agreement with Bertrand et al. who reported an early wave of macrophage progenitors restricted to the YS, and a second wave that was able to reach the FL to participate in definitive hematopoiesis (42). The differential expression of c-Myb between early and late EMPs is in agreement with previous reports indicating that primitive hematopoiesis can occur in the absence of c-Myb, especially for the generation of monopotent macrophage progenitors (114), whereas EMPs from definitive hematopoiesis express and are dependent on c-Myb activity (45, 62, 115).

Notably, a previous study showed that c-Myb ablation strongly compromises definitive hematopoiesis (116). Palis et al. observed that c-Myb is expressed prior to and during the early development of definitive erythrocyte progenitors (45). Thus, late EMPs and LMPs, as well as HSCs, express c-Myb (15, 45, 61, 62), suggesting that the entire fetal monopoiesis machinery is reliant on this transcription factor. In agreement, the CD11bhiF480lo population, which in our hands contains fetal monocytes, was completely absent in the c-Myb-deficient embryo (12, 116). As a consequence, the contribution of c-Myb-dependent progenitors to tissue-resident macrophage populations could not be evaluated in c-Myb-deficient embryos, where c-Myb-independent YS macrophages maintain themselves as a compensatory mechanism due to the absence of c-Myb-dependent fetal monocytes that normally outcompete them. Because c-Myb expression is upregulated during the successive steps of fetal monopoiesis (15), the switch in EMP localization between the YS and the FL may indeed be orchestrated by c-Myb. As a consequence, most tissue-resident macrophages derived from fetal monocytes would therefore rely on c-Myb activity. Altogether we propose that c-Myb<sup>+</sup> EMPs giving rise to the first circulating monocytes should be designated as definitive EMPs.

# Conclusion

Recent reports have drastically changed the view of the development of the MPS and shed light on the multiple layers that define fetal hematopoiesis. It is now evident that fetal monocytes form the major precursors of most adult tissue-resident macrophages, and further investigations are now necessary to clarify how they shape macrophage heterogeneity. Examining how tissues imprint specific fates in these circulating precursors will aid our

# References


understanding of the mechanisms that control the tissue-specific functions of macrophages in the steady state, and thus may uncover new therapeutic opportunities in diverse pathological settings such as metabolic diseases, fibrosis, and carcinogenesis.

# Funding

This work was supported by the Singapore Immunology Network (SIgN) core grant.


blood, is a signature of TH2 inflammation. *Science* (2011) **332**:1284–8. doi:10.1126/science.1204351


in the zebrafish embryo. *Development* (2007) **134**:4147–56. doi:10.1242/ dev.012385


**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 Hoeffel and Ginhoux. 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.*

# Transcriptional regulation of mononuclear phagocyte development

#### *Roxane Tussiwand1 \* and Emmanuel L. Gautier2*

*1Department of Biomedicine, University of Basel, Basel, Switzerland, 2 INSERM UMR\_S 1166, Sorbonne Universités, UPMC Univ Paris 06, Pitié-Salpêtrière Hospital, Paris, France*

Mononuclear phagocytes (MP) are a quite unique subset of hematopoietic cells, which comprise dendritic cells (DC), monocytes as well as monocyte-derived and tissue-resident macrophages. These cells are extremely diverse with regard to their origin, their phenotype as well as their function. Developmentally, DC and monocytes are constantly replenished from a bone marrow hematopoietic progenitor. The ontogeny of macrophages is more complex and is temporally linked and specified by the organ where they reside, occurring early during embryonic or perinatal life. The functional heterogeneity of MPs is certainly a consequence of the tissue of residence and also reflects the diverse ontogeny of the subsets. In this review, we will highlight the developmental pathways of murine MP, with a particular emphasis on the transcriptional factors that regulate their development and function. Finally, we will discuss and point out open questions in the field.

#### *Edited by:*

*Martin Guilliams, Ghent University, Belgium*

#### *Reviewed by:*

*Elodie Segura, Institut Curie, France Vuk Cerovic, RWTH Aachen, Germany*

*\*Correspondence: Roxane Tussiwand r.tussiwand@unibas.ch*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 15 April 2015 Accepted: 02 October 2015 Published: 19 October 2015*

#### *Citation:*

*Tussiwand R and Gautier EL (2015) Transcriptional regulation of mononuclear phagocyte development. Front. Immunol. 6:533. doi: 10.3389/fimmu.2015.00533*

Keywords: transcription factors, development, dendritic cells, macrophages, immunity

# INTRODUCTION

The mononuclear-phagocyte system (MPS), which comprises dendritic cells (DCs), macrophages, and monocytes, is a heterogeneous group of myeloid cells. The complexity of the MPS is equally reflected by the plasticity in function and phenotype that characterizes each subset depending on their location and activation state. Specialized subsets of mononuclear phagocytes (MP) reside in defined anatomical locations, are critical for the homeostatic maintenance of tissues, and provide the link between innate and adaptive immune responses during infections. The ability of MP to maintain or to induce the correct tolerogenic or inflammatory milieu also resides in their complex subset specialization. Such subset heterogeneity is obtained through lineage diversification and specification, which is controlled by defined transcriptional networks and programs. Understanding the MP biology means to define their transcriptional signature, which is required during lineage commitment, and which characterizes each subset's features. This review will focus on the transcriptional regulation of the MPS; in particular, what determines lineage commitment and functional identity; we will emphasize recent advances in the field of single-cell analysis and highlight unresolved questions in the field.

# THE MONONUCLEAR-PHAGOCYTE SYSTEM NETWORK

As summarized in **Table 1**, the MPS is a rather heterogeneous group of myeloid cells, which includes DC, monocytes, and macrophages (1). DCs are mostly short lived and characterized by a half-life that varies between few days up to few weeks (2). This subset of MPs is equipped with pattern recognition receptors (PRR) and is specialized in antigen capture and presentation to T cells (3). At least three different DC subsets have been identified: plasmacytoid DCs (pDCs), and two common or conventional DC (cDC) subsets; cDC1, which express CD24, and CD8α in lymphoid tissues, or CD103 in peripheral organs; and cDC2, which express CD4, CD11b, and CD172 (1). This latter subset of cDCs is heterogeneous and seems to comprise also monocyte-derived DCs and activated macrophages, which have acquired a DC phenotype and most likely function (4).

Steady-state monocytes are short-lived MPs. They are subdivided into two major subsets: patrolling and inflammatory monocytes, which are characterized by low and high expression of Ly-6C, respectively (5). Inflammatory monocytes are recruited and extravasate into infected tissues. They play a role in maintaining the correct inflammatory milieu, are important in the resolution of inflammation and in certain tissues monocytes will replenish the pool of resident macrophages (5–7). The role of patrolling monocytes is less clear but they are certainly involved in the homeostasis of the endothelium (8, 9).

The last subset of MP comprises the mostly long-lived tissueresident macrophages (10). This subset is present in every developing as well as mature tissue, which is highly heterogeneous in terms of phenotype and function, reflecting the physiological needs of the organ of origin (11). Macrophages are thought to be required for the correct development and maintenance of tissues. This topological-related feature is possibly the reason for their extreme heterogeneity and their tissue specialization (12).

Collectively, MPs are highly plastic myeloid cells, which can perform very diverse functions. **Table 1** summarizes the mostly used surface markers in mice and the function attributed to the different MP subsets.

# TRANSCRIPTIONAL REGULATION OF DENDRITIC CELLS DEVELOPMENT

As shown in **Figure 1**, lineage development of hematopoietic progenitor cells along DC lineage occurs through an orchestrated expression pattern of transcription factors (TF), yet the precise molecular mechanisms of lineage restriction and determination remains largely unexplained (2, 13–17). The analysis of genetargeted mice has revealed the functional importance of a few

TABLE 1 | Summarized are the three major murine MPs: dendritic cells, monocytes, and macrophages.


*Each MP subset can be further subdivided into different subsets based on surface marker expression and function and as indicated.*

critical TFs in DC development, with some of them affecting all DCs and some affecting specific subsets (18). DC progenitors are present within the fms-related tyrosine kinase 3 (Flt3)-expressing bone marrow fraction and sustained Flt3 signaling can be considered as instructive for DC development (19–22). Consistently, Flt3-ligand (Flt3L) supports the *in vitro* differentiation of progenitor cells into both pDCs and cDCs (23, 24). Genetic deletion of Flt3L, its receptor, or treatment of mice with Flt3 inhibitors leads to a 10-fold reduction of lymphoid-organ pDCs and cDCs (25, 26). Moreover, Flt3L injection or overexpression of Flt3L results in the expansion of both pDCs and cDCs in all lymphoid and non-lymphoid organs (27, 28). Engagement of Flt3 by Flt3L induces Stat3 phosphorylation and activation, identifying Stat3 as the critical checkpoint of Flt3-induced DC development and proliferation (29, 30). Mirroring Flt3 deficiency, Stat3-deficient mice have severely reduced DC progenitors and mature cells (29). Similarly, deletion of the transcriptional repressor growth factor independent 1 (Gfi1) results in impaired DC development (31). Gfi1-deficient mice show reduced Stat3 phosphorylation and nuclear translocation, with increased expression levels of the Stat3 negative regulators SOCS3 and PIAS3 suggesting that Gfi1 is downstream of Stat3 signaling in the Flt3-Flt3L-induced DC developmental pathway (31). However, the role of Gfi1 is more complex since mice deficient for this repressor show multiple hematopoietic impairments (32, 33). The defects related to Gfi1 deficiency can partially be related to dysregulation of Id2 expression (34–36). However, further studies using subset-specific deletion models will be instrumental to precisely dissect specific transcriptional requirements within the MP lineage. Similarly, despite the experimental evidence of DC expansion following sustained Flt3 signaling, the instructive mechanism promoting DC development is still unclear, given the broad expression of Flt3 on all short-term uncommitted hematopoietic progenitors (ST-HSC) (37, 38). A long non-coding RNA (lncRNA), named lnc-DC, was recently suggested to be the missing key element regulating Stat3 activity exclusively in DCs (39). lnc-DC RNA is expressed by mature DCs and by monocyte-derived DCs and seems to directly interact with Stat3 preventing its de-phosphorylation by SHP1. Furthermore, knockdown experiments of lnc-DC *in vitro* showed impaired DC development from mouse BM progenitors. The conservation of this lnc-DC in terms of function and of its consensus elements at the promoter region across species supports the hypothesis of a new level of regulation present in DC development. However, in mice the transcript seems translated into a highly expressed protein in adipose tissue (40). Further studies are therefore needed to understand potential species-specificities as well as its requirement *in vivo* under steady-state conditions.

Proceeding along the DC developmental pathway, three major branches of mature DCs are identified: pDCs, CD24<sup>+</sup> cDC1, and CD11b<sup>+</sup> cDC2 (3, 16). pDCs and cDC1 both express and depend on the transcription factor interferon regulatory factor 8 (Irf8), while cDC2 express and are partially dependent on Irf4 (1, 18, 41–44). Despite major advances in our understanding of the transcriptional requirement during DC development, we are still unable to draw a clear developmental map (**Figure 2**) (13, 18). This may reflect subset heterogeneity as well as the plasticity, which characterizes DCs. Also, the expression of the different TFs

is not unique and can change during differentiation and activation further complicating the picture.

During early stages of DC development, a progenitor that expresses high levels of Irf8 and shows developmental potential toward all DCs can be identified (42). It is likely that the first branching choice will determine whether pDCs or cDCs commitment occurs. The balance of the E-protein transcription factor 4 (Tcf4), also known as E2-2, and the E-protein inhibitor of DNA binding 2 (Id2) seems to determine lineage development toward pDCs or cDCs, respectively (45–50). Constitutive or inducible deletion of E2-2 in CD11c-expressing cells blocks the development of pDCs but not cDCs, while overexpression of Id2 inhibits pDC development (47, 48). E2-2 is required not only during development but also for lineage maintenance of pDCs (47, 51). Several targets of E2-2 have been identified such as SpiB, Irf8, and Irf7, and all contribute to pDCs lineage specification (47, 51). Despite the requirement for E2-2 during pDCs commitment, how Id2 and E2-2 are conversely induced and regulated is still an open question. Recently, the eight-twenty-one (ETO) protein 2 or Mtg16 (also referred as core-binding factor, runt domain, alpha subunit 2, translocated to 3 Cbfa2t3) was suggested to target and repress Id2 together with E2-2 and inhibit Irf8-expressing cDC1 development, while favoring pDC commitment (52). Consistently, Id2 and Mtg16 double-deficient mice show restored pDC potential (52). However, Mtg16 seems to act together with E2-2 leaving the question on how lineage determination toward E2-2- or Id2-expressing progenitors occurs, still open. On the other side, one other candidate, which could be involved in reinforcing lineage fate toward Irf8-expressing cDC1 at the expenses of pDCs could be the leucine zipper transcription factor E4BP4, also referred as Nfil3 (53). Mice deficient for this TF show increased pDC and reduced cDC1 development (53). The mechanism of action remains to be elucidated since Id2 expression does not appear to be perturbed and only the basic leucine zipper transcription factor ATF-like 3 (Batf3) expression was shown to be reduced (53). Phenotypically, a bias toward pDC development has been observed within the macrophage-colony-stimulating factor receptor (M-CSFR) negative progenitors, whereas cDCs precursors are enriched within the M-CSFR expressing BM fraction (54, 55). These results may suggest that under sustained M-CSF stimulation uncommitted progenitors may lose the potential toward pDCs. Alternatively, as recently suggested, the absence of GM-CSF signaling, which induces STAT5 phosphorylation, could be the permissive condition to promote pDC development (56). Accumulation and/or withdrawal of specific cytokines during proliferation and differentiation as well as regulation of TF levels through division of progenitor cells could partially explain how BM niches influence development and lineage commitment (57, 58).

Proceeding along DC development, a common cDC progenitor able to differentiate *in vivo* into both CD24<sup>+</sup> cDC1 and CD11b+ cDC2 was identified (55, 59–61). And recently, lineagetracing studies allowed further dissection of cDC commitment and resulted fundamental to establish the transcriptional

requirements during development of clonogenic cDC progenitors (62, 63). The expression pattern of the zinc finger and BTB domain containing 46 transcription factor Zbtb46 (also called Btbd4) can be considered as cDC-lineage specific within hematopoietic cells (64–66). This TF is not present on pDCs and is induced on monocyte-derived DCs, supporting on the one hand early divergence of pDCs during DC commitment, and on the other hand suggesting a developmental convergence between cDCs and monocyte-derived DCs (4, 66). Similarly, lineagetracing experiments were performed using mice expressing Cre recombinase under the control of Clec9a also referred as Dngr-1 (67). Although labeling is not absolute on all cDCs subsets, it seems to be restricted to pre-cDC progeny, without marking inflammatory-derived DCs (67). The use of these reporter mouse models will help us better characterize the ontogeny of specific cDCs subsets also depending on the tissue of origin and whether under steady-state or inflammatory conditions.

The CD24+ cDC1 branch of cDCs depends on the transcription factors Irf8, Id2, Nfil3, and Batf3 (68). The generation of mice deficient for Batf3 has revealed the common origin and the lineage identity of Irf8-expressing cDC1 cells, also referred as CD8a<sup>+</sup> or CD103<sup>+</sup> across all lymphoid and peripheral organs (69, 70). However, while only Irf8 was shown to be necessary for commitment, Id2, Nfil3, and Batf3 are dispensable under certain conditions (71, 72). Lineage choice seems influenced by high and sustained levels of Irf8 during cDC1 commitment. Binding of Batf3 and Irf8 to an AP1-IRF composite element (AICE) within the Irf8 super-enhancer in CD24- or Zbtb46-gfp-expressing immediate progenitors leads to sustained Irf8 expression and cDC1 development (62). In the absence of Batf3, reduced Irf8 levels, redirect commitment of a CD24-expressing cDC1 progenitor toward the Irf4-expressing cDC2 lineage (62).

Despite the recent advances, how the branching of cDC1 and cDC2 occurs is still an open question. The recent identification of a committed cDC2 progenitor might help to identify the key factors involved in this process: we still need to understand how expression of Irf4 progressively replaces Irf8, and how those two TF determine the identity of these subsets. Furthermore, the cDC2 lineage, as already mentioned, is highly heterogeneous and possibly contains multiple subsets (1, 2, 11, 16, 18). Mature CD11b-expressing cDC2 express high levels of Irf4, suggesting an important role for this TF within this lineage. And indeed, absence of Irf4 impairs the development as well as the function of cDC2 (42, 44, 73–77). In mice lacking IRF4 in CD11c-expressing cells, cDC2 numbers are reduced in lung and small intestinal DCs, while no difference is reported for skin (44, 74). However, reduction in lung and lamina propria cDCs is only observed upon deletion of Irf4 in early progenitors (44; 74). Despite, normal numbers of skin DCs in Irf4 deficient animals, migration to draining lymph nodes is impaired as a consequence of defective induction of CCR7 (78). Furthermore, reduced up-regulation of MHC-II and co-stimulatory molecules is also associated with Irf4 deficiency (75, 77, 78). Collectively, Irf4 shows a broad action across different tissues and potentially subsets, and further studies are required to be able to understand the specific requirement of this TF during development.

Other TFs reported to display a reduction of cDC2 are RelB, Notch2, RbpJ, and the Kruppel-like Factor 4 (Klf4) (79–85). Notch2 is required for terminal differentiation of endothelial cell-selective adhesion molecule (ESAM)-expressing splenic cDC (81, 83). Similar to Notch2 deficiency, mice compromised in Runx3 (86, 87) and in the alternative NF-kB pathway show a reduction in the development of ESAM<sup>+</sup> cDCs (80, 88). However, a survival disadvantage in competitive settings appears to be present in mice with compromised NF-kB signaling, suggesting caution in proposing the requirement for NF-kB during DCs development (80, 88). Klf4 deficiency results in impaired development of the so-called "double negative" DCs in skin draining LN and a partial reduction of Sirp-α but not splenic ESAM-expressing cDC2 across all the organs (84). In these mice, cDC progenitors are impaired in their ability to down-regulate Irf8 and up-regulate Irf4. However, the *in vitro* differentiation potential of Irf4-expressing cDCs as well as expression of Irf4 on peripheral cDCs is not compromised. This can be explained by the existence of at least two cDC2 subsets, where only the Klf4/Irf4-dependent one is developmentally impaired. Alternatively, a different maturation/activation state, which requires Klf4, may exist within the Irf4-expressing cDC2 subset.

Collectively, a partial reduction associated with the lack of one or the other TF confirms the developmental, and supports the subset-specific heterogeneity observed in single-cell sequencing experiments for the Irf4 and CD11b-expressing cDC2 cells (89– 92). The transcriptional diversity, which characterizes these cDC2 cells, results and reflects a functional heterogeneity (**Figure 2**). Notch2 cDC2 are required for anti bacterial Th17/IL-22 immunity, while Klf4 deficiency results in impaired Th2 immunity (83, 84, 93). Expression of Irf4 in cDCs is necessary for both Th17 and Th2 responses further highlighting the complexity of this TF in DC biology (44, 73, 74, 77). Understanding whether the absence of a subset or a functional defect caused by a transcriptional deficiency on the remaining subset could account for the observed phenotypes will require subset-specific deletion. Furthermore, we also need to explore more in detail the influence of tissues on the different subsets. Are tissue-specific cues driving the expression of a transcriptional signature in a similar way as recently revealed for macrophages? (12) Are the differences reflecting a developmental or a functional heterogeneity? Is a developmental convergence between cDCs and monocyte-derived DCs creating the confusion within this branch of cDCs. We need a better characterization of the different subsets, which fall under the broad umbrella of CD11b or Irf4-expressing cDC2 and some progress has certainly been made with the introduction of new reporter mice as previously discussed as well as the recently identified committed progenitor. Teasing this heterogeneous pool of Irf4-expressing cDC2 apart is currently an active field of investigation (90, 91, 94). And new technologies will be instrumental to improve our comprehension of the molecular clues, which regulate lineage commitment. A recent report analyzed stage and subset-specific expression of mi-RNAs during DC development and miR-142 was identified as a key regulator of cDC2 differentiation, further adding additional complexity to our current understanding of DC development (91).

Better genetic models are needed and will possibly be soon developed as a result of the recently published single-cell analysis (16, 89). Identifying TFs or surface markers, which would compromise or trace the development of one lineage independently of the anatomical localization, as previously done in Batf3<sup>−</sup>/<sup>−</sup> mice for Irf8-dependent cDC1 would be of great advantage (69, 70).

# TRANSCRIPTIONAL REGULATION OF MONOCYTE DEVELOPMENT

The molecular regulation, which defines monocyte differentiation and lineage commitment, is poorly understood (95). Most of the identified TFs, that result in impaired monocyte development, also show an effect on other hematopoietic lineages. The transcription factors Irf8, Sfpi1 (PU.1), Egr-1, Stat3, Gfi1, Gata2, Gbx2, Nur77, retinoic acid receptors, C/EBPα and C/EBPβ, Klf4, and c-Maf as well as members of the NF-κB family members are all involved in monocyte differentiation, however their function is often redundant, certainly not limited to monocytes and in some cases mediating proliferative and/or survival rather than instructive cues (96). Most of the TFs involved in monocyte differentiation are shared within the myelo-monocytic branch. Some of them were already mentioned as important during DC development; others are involved in macrophage and or granulocyte commitment; we are therefore aware that we can only provide here a simplified transcriptional path, which leads to monocyte development and that more efforts are required to better understand.

Expression of the ETS family transcription factor Sfpi1 or PU.1 at early stages is suggested to antagonize on the one hand key regulators of other developmental pathways, such as GATA-1 for erythroid lineage, and on the other hand activate myeloid-specific factors such as Irf8, Klf4, and Erg1 (95). A critical step in monocyte differentiation is the induction of Csf1R expression at the cell surface. This seems to be regulated by Klf4 and Irf8, however both factors are also involved in cDC development, as previously discussed, (85). Furthermore, Csf1R is also needed for macrophage development.

The identification of a committed progenitor with monocyterestricted potential called cMoP confirmed high expression levels of the above-mentioned TF (97). However, none of those is unique to monocyte differentiation and potentially complex genetic models will be required to unravel the transcriptional map required for monocyte lineage specification.

# ORIGIN OF TISSUE-RESIDENT MACROPHAGES

As discussed above for DCs, similar questions arise considering tissue-resident macrophage origin and development. Lineagetracing studies recently revisited their origin and revealed how their maintenance in adult tissues is mostly independent from monocytes and adult definitive hematopoiesis (10). Indeed, tissue-resident macrophages were proposed to develop from a Myb-independent but Sfpi1 (PU.1)-dependent fetal progenitor present in the yolk sac (YS) (5, 6, 15, 98–100) and capable of seeding the developing embryo and self-renewing during adulthood. This developmental path was first described for microglia, the brain-resident macrophages (98, 99), but still remained elusive for a number of other macrophage populations. Using similar tools, the contribution of YS progenitors to a number of adult tissue-resident macrophage populations was next assessed and only very limited input was found in most tissues tested (101). In parallel, other studies conducted in the lung and skin found that resident alveolar macrophages and Langerhans cells originated from fetal monocytes (99, 102). A recently described hypothesis is now trying to bridge these findings by proposing the existence of erythro-myeloid progenitors (EMP) distinct from hematopoietic stem cells (HSCs), which develop in YS (E8.5) and colonize the fetal liver at E16.5 giving rise to fetal erythrocytes, macrophages, granulocytes, and monocyte (103, 104). Such progenitors would generate microglia early during fetal development and participate to Kupffer cells and Langerhans cells development, but its definitive participation to the generation of other tissue-resident populations, as well as its long-term persistence, still remains to be firmly established. Indeed, a very recent study is now arguing that fetal HSCs, and not YS progenitors or EMPs, give rise to most tissue macrophage populations, except microglia known to originate from YS progenitors other than HSCs (105). This study also highlighted that while most tissue macrophages subsets maintain by self-renewal in the adult, peritoneal, dermal, and colonic residents macrophages needed continuous HSCs input to be maintained during lifetime. Accordingly, gut macrophages, most likely a specific population of macrophage residing in the serosa (106), were shown to derive from HSC-derived circulating monocytes (107). Moreover, blood monocytes can participate to the maintenance of heart macrophages in the adult (108). During inflammation, in addition to tissue-resident macrophages, some macrophages found in tissue differentiate from locally recruited Ly-6Chi monocyte. Such monocyte-derived macrophages reside only for a short period of time in the tissue until inflammation resolves, and are cleared through local cell death (109). Overall, these studies suggest that there is probably more than a single developmental pathway to generate tissue macrophages and to support their self-renewal potential and unique long-term maintenance ability (**Figure 3**).

# PATHWAYS ALLOWING FOR TISSUE-RESIDENT MACROPHAGE DEVELOPMENT AND MAINTENANCE

Proceeding along development, Runt-related transcription factor 1 (RUNX1) is required at early stages of myeloid lineage specification and regulates the expression of Sfpi1 (PU.1) which has to be expressed at high levels to allow for development and maintenance of macrophage differentiation (57). One of the most crucial target genes of PU.1 during macrophage development is *Csf1r*, which encodes the receptor for M-CSF and IL-34 (110). Signaling of M-CSFR through either M-CSF or IL-34 allows for the maintenance of tissue-resident macrophages (111, 112). Other TFs required for macrophage development, which co-operate with PU.1 in lineage determination, are AML1 and CCAAT enhancer-binding proteins (C/EBP) (113, 114). Overall, our understanding of the molecular pathways controlling tissue macrophage development in general, as well as their maintenance, remains poorly defined and further studies are need to better characterize how their development is regulated. While for DCs, deletion of a subset might result in minor consequences, macrophages are thought to be critical for the organogenesis and organ homeostasis, therefore deletion of a subset could be deleterious for the life of the individual or only compatible with compensation through alternative subsets or pathways.

# TISSUE-RESIDENT MACROPHAGES DIVERSITY AND TISSUE-SPECIFIC TRANSCRIPTION FACTORS CONTROLLING RESIDENT MACROPHAGE DEVELOPMENT AND MAINTENANCE

Our analysis of the transcriptional landscape of tissue-resident macrophages revealed wide heterogeneity across tissues, leading to the definition of population-specific signatures. These specific signatures were recently shown to rely on distinct enhancer landscapes shaped by the tissue microenvironment (115, 116). Using the Immunological Genome database, the reconstruction of lineage-specific regulation from gene-expression profiles across lineages (117) revealed gene modules selectively associated with a single tissue macrophage population (12). Additionally, TFs were predicted to regulate these modules, and thus could potentially influence the development of resident macrophages in a tissue-specific manner (12). Among others, predicted regulators included Spi-C for red pulp macrophages, which confirmed precedent findings (118) and thus validated the predictive power of the algorithm. Indeed, Spi-C is a TF closely related to Sfpi1 and highly expressed in spleen red pulp macrophages compared to other phagocytes (12, 118). Mice deficient for Spi-C lack splenic red pulp macrophages (118), leading to defective red blood cells recycling and iron accumulation in the spleen. At which levels Spi-C acts to control the differentiation and/or survival of red pulp macrophage remains to be determined. LXRα is another TF needed for splenic red pulp macrophage development (119), and whether Spi-C and LXRα interact together in this process is not known. Interestingly, intracellular heme accumulation following erythrocytes uptake induced Spi-C expression by stimulating the degradation of its transcriptional inhibitor Bach1 (118). Thus, heme-induced Spi-C controls the functionality of splenic red pulp macrophages, but also their maintenance albeit by an undetermined mechanism.

Similarly, PPARγ was identified as a regulator for lung macrophages (120). It is a ligand-controlled TF of the nuclear receptor family known for its role in lipid metabolism (121). Previous work has shown that PPARγ expression is important to maintain lung macrophages functionality and surfactant catabolism (122). We reported that conditional deletion of PPARγ in lung macrophages strikingly altered their transcriptome (120). Dysregulated expression of a number of genes involved in lipid metabolism was observed (120), and many of these genes were known targets of the sterol-responsive transcription factor LXR. Accordingly, increased sterol accumulation was observed in lung macrophages lacking PPARγ, as well as decreased expression of genes involved in inflammation and immunity (120). Using a different gene deletion

approach, it was recently shown that PPARγ could also be key in controlling the development of this subset. Such discrepancy between models might relate to the different temporal induction of the cre expressing strains used in these two studies (123).

Finally, GATA6 was identified as a specific peritoneal macrophage regulator and we observed that its expression was selectively found in F4/80+ peritoneal macrophages across many lineages tested (12), suggesting that it may represent the master regulator of tissue-resident peritoneal macrophages. Interestingly, GATA6 expression by resident peritoneal macrophages was dependent on retinoic acid signaling *in vivo* (12) and mice lacking GATA6 in macrophages, generated by crossing Gata6fl/fl mice with Lyz2-cre, showed a strong reduction in F4/80+ peritoneal macrophages (115–117). Additionally, Th2 inflammation following parasitic infection failed to increase peritoneal macrophage numbers in Lyz2-cre × Gata6fl/fl mice (124), as described for wild-type mice (125). Impaired steady-state numbers of peritoneal macrophages in the absence of GATA6 was accompanied by impaired self-renewal, marked increased in S/G2-M cell cycle phases and accumulation of multinucleated macrophages due to impaired cytokinesis (126). While reduced survival of peritoneal macrophage already explains the strong contraction in their number, impaired cytokinesis will likely further exacerbate the phenotype. GATA6 deficiency in F4/80+ peritoneal macrophages led to the down-regulation of Aspa mRNA, which encodes an aspartoacylase generating acetyl-CoA, a central cellular metabolite, from *N*-acetylaspartate (124). Interestingly, mice lacking Aspa showed reduced F4/80+ peritoneal macrophages. Overall, a tissue-specific transcriptional network driven by GATA6 controls multiple pathways all required for the maintenance of F4/80+ peritoneal macrophages.

# CONCLUDING REMARKS

In the past few years, major advances have been made in our understanding how the development of myeloid cells occurs. DNA, RNA protein sequencing and characterization on entire tissues and populations is now a more accessible technology. This combined with improved multicolor flow cytometry and CyTOF technology has allowed us to better understand which TFs identify specific subsets and developmental stages during hematopoietic development (89, 90, 124, 127). However, as it is often the case, the better our analysis tools become, the more complex the picture appears. And despite these advances, we are now starting to perceive how many more gaps need to be filled in order to be able to draw a definitive road map for every MP subset. Significant progress has been made in defining, which TF are needed during DC and macrophage development in specific tissues. Monocyte development, however, is still elusive and most of the factors identified rather compromise their survival, making hard to discriminate between developmental and survival defects. Moreover, in the past few years, it has become obvious how tissues are able to influence not only the phenotype but also the function of the different subsets. This observation translates in changes in the transcriptional signature, which identifies each subset in a given tissue. Tissue-associated hallmarks have been mostly studied in macrophages, however profound consequences appear to matter also within DC subsets. It will therefore be important to discriminate between tissue- versus subset-specific transcriptional identity, to define intrinsic properties, and functional potential for every subset across and within the different tissues. On the one hand, it is attractive to think that subset specialization similar as for T and innate lymphoid cells is also present within the myeloid compartment. On the other hand, we are aware that myeloid cells are characterized by an elevated intrinsic functional plasticity. Anatomical compartments, pathogen and antigen dose as well as small micro-environmental cues might drastically influence the phenotype, the transcriptional landscape as well as the function of the different subsets during immune responses and we are just starting to explore in depth the complexity of the different subset in response to an immunological insult (90, 127). For DCs, a model has been recently suggested which takes into account the development of a subset with its immunological function. As shown in **Figure 2**, expression of Irf8 and Batf3 is needed in response to pathogens or immunological conditions where IFN-γ is required. On the other side, Irf4 is essential to stimulate Th17, Th2, and IL-22 responses. Within the Irf4 response, Notch2 and Klf4 are specifically required for Th17/IL-22 or Th2 immunity, respectively. The scenario, which appears, is consistent with functional modules of transcription across different cell types, i.e., Klf4 is also necessary for goblet cell development and polarization of M2 macrophages, whereas Notch is required for ILC3 development. Similarly, Nfil3 is important not only for cDC1 but also for NK and ILC1 cells development.

Several TFs that are required for MP development have been characterized; we can draw a map for their temporal requirement along development but for most of them the precise mechanism of

## REFERENCES


action and their targets still need to be identified. Furthermore, since the developmental as well as the functional requirements for a transcriptional pathway are most often shared, caution is necessary to ascribe a specific role to a TF. Recently, a Waddington landscape was suggested to explain the plasticity in DC development (92). A similar concept may reflect and be applied to the entire MP system, where lineage commitment, specific functions, as well as subset identity could depend on the achievement of a threshold of a pool of TFs, rather than a unique master regulator. This concept would explain the so-called "graded-commitment" obtained from barcoding individual progenitors and performing lineage-tracing experiments (128). A second level of complexity is characterized by the fact that multiple subsets share the same TF, though the functional requirements are different. For example, Irf4 seems to regulate migration in skin DCs but not in other peripheral tissues, such as lungs. The functional outcome might be shared; such as in both cases antigen presentation is impaired, however it is important to understand the different requirements depending on the tissue of origin.

The study of MP is characterized by blurry phenotypic boundaries, which do not allow for unequivocal identification of the different subsets. The absence of specific markers leads to the absence of specific genetic tools and sometimes conflicting or unclear results are present in the literature. For lineage-specific deletion within cDC, we still relay on CD11c–cre mice, despite the evident limitations of this model. For monocytes as well as macrophages we lack genetic models, which would allow for selective and specific depletion of inflammatory or patrolling monocytes as well as tissue macrophages. Efforts to generate better lineagedeleter mouse models are therefore required and should be a priority in the next future to better understand the development as well as the contribution of MPs during an immune response.

# ACKNOWLEDGMENTS

RT is supported by the Swiss National Fund with an SNF Professorship (NBM 1570). EG is funded by grants (LIPOCAMD and MACLEAR) from the National Agency of Research (ANR) and by a grant from the Fondation de France (00056835).

effector cells and migratory antigen-presenting cells. *Immunity* (2012) **37**:1076–90. doi:10.1016/j.immuni.2012.08.026


and CD4/CD8 lineage decision in the thymus. *J Exp Med* (2003) **197**:831–44. doi:10.1084/jem.20021417


dendritic cells drive mucosal T helper 17 cell differentiation. *Immunity* (2013) **38**:958–69. doi:10.1016/j.immuni.2013.03.009


cells in the spleen and intestine. *Immunity* (2011) **35**:780–91. doi:10.1016/j. immuni.2011.08.013


self-renewal in the absence of the transcription factor C/EBP alpha. *Immunity* (2004) **21**:853–63. doi:10.1016/j.immuni.2004.11.006


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The Guest Associate Editor Martin Guilliams declares that, despite having published one manuscript together with Roxane Tussiwand, the review process was handled objectively.

*Copyright © 2015 Tussiwand and Gautier. 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.*

# **Fate mapping of dendritic cells**

#### *Mateusz Pawel Poltorak and Barbara Ursula Schraml\**

*Institute for Medical Microbiology, Immunology and Hygiene, Technische Universität München, Munich, Germany*

Dendritic cells (DCs) are a heterogeneous group of mononuclear phagocytes with versatile roles in immunity. They are classified predominantly based on phenotypic and functional properties, namely their stellate morphology, expression of the integrin CD11c, and major histocompatibility class II molecules, as well as their superior capacity to migrate to secondary lymphoid organs and stimulate naïve T cells. However, these attributes are not exclusive to DCs and often change within inflammatory or infectious environments. This led to debates over cell identification and questioned even the mere existence of DCs as distinct leukocyte lineage. Here, we review experimental approaches taken to fate map DCs and discuss how these have shaped our understanding of DC ontogeny and lineage affiliation. Considering the ontogenetic properties of DCs will help to overcome the inherent shortcomings of purely phenotypic- and function-based approaches to cell definition and will yield a more robust way of DC classification.

#### *Edited by:*

*Shalin Naik, Walter & Eliza Hall Institute, Australia*

#### *Reviewed by:*

*Sam Basta, Queen's University, Canada Theresa T. Lu, Weill Cornell Medical Center, USA*

#### *\*Correspondence:*

*Barbara Ursula Schraml, Institute for Medical Microbiology, Immunology and Hygiene, Technische Universität München, Trogerstraße 30, Munich 81675, Germany barbara.schraml@tum.de*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 31 January 2015 Accepted: 13 April 2015 Published: 04 May 2015*

#### *Citation:*

*Poltorak MP and Schraml BU (2015) Fate mapping of dendritic cells. Front. Immunol. 6:199. doi: 10.3389/fimmu.2015.00199* **Keywords: dendritic cell, ontogeny, fate mapping, lineage tracing, mononuclear phagocyte**

# **Introduction**

Dendritic cells (DCs) were originally identified in mouse spleen for their unique stellate morphology, their ability to adhere to certain glass surfaces and their superior capacity to activate naïve T lymphocytes that distinguished them from macrophages (MØs) (1–3). Mostly for historical reasons, DCs are considered part of the mononuclear phagocyte (MP) system, which groups all highly phagocytic cells derived from monocytes or their precursors based on the premise that tissue MØs arise from monocytes (4–9). This presumed relatedness of DCs, monocytes, and MØs coupled to the lack of reliable ways to distinguish MP subtypes has caused continuous debates over accurate cell-type identification and has led some to question whether DCs in fact constitute an independent cell lineage (6, 7, 10–14). However, today we have conclusive evidence demonstrating that DCs, monocytes, and MØs have distinct cellular origin and we further distinguish plasmacytoid DCs (pDCs) from two subsets of so-called conventional or classical DCs (cDCs) based on unique developmental requirements (7, 15–19). Nonetheless, DCs remain defined based on phenotypic and functional properties that often overlap with those of monocytes or MØs (19), although some have suggested a shift in paradigm toward a nomenclature that takes cell ontogeny into account (6, 7, 10).

Dendritic cells are generally identified by their high expression of major histocompatibility complex class II molecules (MHCII) and of the integrin CD11c, as well as their superior capacity to migrate from non-lymphoid to lymphoid organs and stimulate naïve T cells (3, 20–22). However, these characteristics are not absolute and can change in situations of inflammation or infection, thus complicating cell identification (6, 7, 23, 24). For instance, CD11c, considered the hallmark surface marker of DCs, is also found on B, T, and NK cells as well as some monocytes, MØs, and eosinophils (25–32). Dendritic protrusions have also been observed in some MØs and T cells (33–35). Further, surface markers, such as F4/80, CD14, or CD64 (Fc-gamma receptor 1), generally associated with monocytes or MØs can be found on DCs (36–38). One might argue that the most defining feature of DCs is their ability to activate T cells, however such definition discounts the fact that DCs potently regulate innate immune responses independent of their ability to migrate to lymphoid organs or stimulate T cells (39–44). Conversely, non-DCs can carry antigen to lymph nodes and activate naïve T cells in some instances (45–47).

Therefore, morphological and functional properties, as well as the expression of surface markers are insufficient to clearly distinguish DCs from monocytes and MØs, raising the necessity to find a more robust way of cell identification. Recent studies in mouse and human indicate that DCs, MØs, and monocytes have unique ontogenetic properties and thus can be considered distinct cell lineages (36, 48–54). Here, we review approaches that have been employed to track and define the progeny of DC precursors *in vivo* and discuss how such "fate mapping" approaches have improved our understanding of DC heterogeneity and ontogeny. These studies lay the foundation for moving toward cell ontogeny as a major lineage-determining criterion, which will allow for a more reliable and precise classification of DCs and DC subsets.

# **DC Development**

Dendritic cells are short-lived and their maintenance relies on constant replenishment from bone marrow progenitors that originate from hematopoietic stem cells (HSCs) (19, 55). In the classic model of DC development monocytes and DCs arise from bipotent progenitors, so-called MØ and DC progenitors (MDPs) (**Figure 1**) (56). MDPs further give rise to common DC progenitors (CDPs) restricted to the generation of pDCs and cDCs (**Figure 1**) (57, 58). pDCs terminally differentiate in the bone

**FIGURE 1 | Classic model of DC development**. DCs and monocytes are ancestrally related and arise from bi-potential MDPs residing in the bone marrow. MDPs further differentiate into monocytes and CDPs, which are restricted to the generation of various types of DCs. CDPs give rise to pDCs, which fully develop in the bone marrow, and pre-DCs, which migrate through the blood to tissues, where they fully differentiate into CD11b*<sup>−</sup>* (including CD8α <sup>+</sup> cDCs in lymphoid tissue and migratory CD103<sup>+</sup> cDCs in non-lymphoid tissue) and CD11b<sup>+</sup> cDCs. Monocytes complete their development in the bone marrow and reach peripheral tissues via the bloodstream. There they further differentiate into monocyte-derived DCs (mo-DCs) or MØs (mo-MØs) in response to environmental cues.

marrow, thus exit the bone marrow as fully developed cells and reach peripheral organs via the blood stream (**Figure 1**) (15, 59). In contrast, cDCs arise from another developmental intermediate termed pre-DC, which exits the bone marrow and migrates through the blood to seed lymphoid and nonlymphoid tissues (60, 61). There, pre-DCs terminally differentiate into cDCs, including the main CD11b*<sup>−</sup>* and CD11b<sup>+</sup> subtypes (**Figure 1**) (60–63). In lymphoid tissues these are CD8α <sup>+</sup>CD11b*<sup>−</sup>* and CD11b<sup>+</sup> resident cDCs, whereas in non-lymphoid tissues they comprise CD103+CD11b*<sup>−</sup>* and CD11b<sup>+</sup> migratory cDCs (3, 60–63). Like pDCs, monocytes complete their development in the bone marrow but in tissues they differentiate into cells with DC- or MØ-like features (**Figure 1**) (23, 24, 64, 65). This plasticity is remarkably prominent in inflammatory or infectious environments, when monocyte-derived cells with qualities of DCs have been referred to as TNF-α/iNOS-producing DCs (Tip-DCs), monocyte-derived DCs (mo-DCs), and/or inflammatory DCs (23, 24, 64, 65).

Although most of our knowledge concerning DC development is derived from mouse studies, developmental parallels have been observed in other species (66–73). Especially the identification of putative equivalent DC progenitor populations in human holds promise for future research (72, 73). Yet, some uncertainties remain. Common lymphoid progenitors (CLPs) can give rise to DC descendants upon adoptive transfer (74), although it is now thought that DCs originate predominantly from myeloid progenitors (75, 76). Nonetheless, some pDCs, but not cDCs, show evidence of VDJ gene rearrangements, potentially indicating lymphoid lineage heritage (15, 59, 77). However, it remains unclear whether evidence of *Rag* gene expression history necessarily means that pDCs have dual lymphoid and myeloid origin. Contrary to the dogma that monocytes and DCs share a common immediate ancestor, recent data suggest that lineage divergence of HSC-derived myeloid cells occurs much earlier than previously predicted and that monocytes and DCs might arise independent of a bi-potential developmental intermediate (49, 78, 79). Elucidating such unresolved aspects pertaining to DC ontogeny may solve uncertainties in determining lineage affiliation, which, in turn, will aid to further decipher the unique functions of DCs in immunity.

# **Fate Mapping**

Understanding cell development requires models with which the relationship of a precursor cell and its progeny can be defined *in vivo*. Such "fate mapping" can be achieved in various ways and relies on the selective labeling of the cell(s) of interest so that consequently the development of the marked cell can be followed in its natural environment (80). Tracing progenitors *in vivo* also offers the possibility to determine the fate of populations when lineage affiliation is most heavily debated, namely following experimental manipulation to generate conditions of inflammation or infection. While most fate mapping strategies follow the progeny of bulk cell populations, recently developed techniques have enabled the tracing of single cells, thus providing valuable information regarding their developmental potential at the clonal level (80, 81). In all fate mapping experiments, it is important to consider that their interpretation is dependent on the use of select, faithful and stable markers (82).

## **Precursor Transfers**

The transfer of purified and pre-marked precursor cells into congenic recipients is the most accessible form of fate mapping as a variety of labeling options can be used to distinguish between donor and host cells (**Figure 2A**) (80). As a result, precursor transfers are commonly used to study cell development and lineage relationships and remain a standard protocol for defining the stemness of progenitor cells (80). Such experiments rely on the ability to purify sufficient precursors that, after cell isolation, retain the capacity to home to the appropriate anatomical niche and expand sufficiently into detectable progeny. To circumvent such limitations transfer studies are often combined with protocols to induce leukopenia, such as irradiation, in order to increase the niche available for cell engraftment (**Figure 2A**) (80). However, these manipulations can alter developmental signals, which, in turn, might impact on the interpretation of results (18, 54, 83). To best mimic the endogenous cellular environment, progenitors have been returned directly to their organs of origin, for instance by intra-bone injection (84).

The DC progenitors MDP, CDP, and pre-DC were in part defined by assessing their developmental potential after adoptive transfer into mice (56–58, 60, 61, 84–86). In such experiments, MDPs give rise to DCs and monocytes, whereas CDPs and pre-DCs are restricted to the generation of DCs but do not generate monocytes or other leukocyte lineages (56–58, 61, 84–86). In combination with experiments assessing the differentiation potential of single progenitors *in vitro* (56–58), these studies have significantly shaped our view of DC development (**Figure 1**). Surprisingly, the existence of MDP as a bi-potential intermediate for DCs and monocytes has recently been questioned when single CX3CR1<sup>+</sup> MDPs were unable to generate both DCs and monocytes upon differentiation *in vitro* (78). The authors further found that adoptively transferred CX3CR1<sup>+</sup> MDPs, not only gave rise to DCs and monocytes but also neutrophils (78). However, such multi-potency of MDPs was not observed in earlier studies (52, 56, 61, 85, 86) and is not evident in genetic CX3CR1 fate mapping experiments (50). It is possible that these discrepancies may be explained by experimental variation such as differences in cell isolation, the timing of analysis or variances in the niche available for cell engraftment following irradiation (18, 54, 83). In light of these results it is noteworthy, however, that upon adoptive transfer MDPs exhibit pDC potential only in some studies (52, 86) but not others (56, 85), whereas the presumed downstream CDPs produce both pDCs and cDCs (57, 58). Taken together these experiments raise some doubt about the existence of a MDP as a key developmental intermediate for monocytes, cDCs, and pDCs. However, resolving this matter will require the use of better models to trace single cells *in vivo* as experiments relying on the isolation and analysis of bulk progenitor populations are inherently prone to disparities in gating strategy or cell purity.

In DC ontogeny, these issues are augmented because MDP and CDP exhibit substantial phenotypic overlap: both lack lineagedefining markers, are characterized by expression of CX3CR1, CD115 (M-CSFR, Csf1R) as well as CD135 (FMS-like tyrosine kinase 3, FLT3) and, until recently, CDP could only be distinguished from MDP by lower expression of the receptor tyrosine kinase CD117 (c-kit) (56–58, 61, 86). We have recently found that the C-type lectin receptor DNGR-1 (Clec9a) marks cells resembling CDPs (36). Surprisingly, upon adoptive transfer, DNGR-1+CD115<sup>+</sup> progenitors exhibit cDC-restricted differentiation potential and do not generate pDCs (36), suggesting that DNGR-1 marks cDC-restricted progenitors. These data are in line with a recent study demonstrating a strong bias for CD115<sup>+</sup> CDPs to generate cDCs, whereas pDCs arise predominantly from CD115 negative cells (79). Therefore, cDCs and pDCs appear to have distinct developmental intermediates that can be distinguished by expression of CD115 (79) and DNGR-1 (36). Since CD115<sup>+</sup> CDPs presumably express DNGR-1 (36), it is unclear why some CD115<sup>+</sup> CDPs show combined cDC and pDC potential in clonal assays (57, 58, 79). It is possible that antibody-mediated triggering of DNGR-1 or growth factor receptors, such as CD115, during cell isolation skews DC differentiation toward a particular DC sub-lineage in an unforeseeable manner. The developmental potential of progenitors may also be influenced by the specific culture conditions used (78) or DCs could exhibit a degree of developmental plasticity (87). Nonetheless, the existence of a putative intermediate monocyte-restricted progenitor downstream of MDP (common monocyte progenitor, cMoP) (52) alongside the aforementioned pDC- and cDC-restricted progenitors supports a model in which monocytes, cDCs, and pDCs develop independently. The genuine point of lineage divergence, however, remains to be determined.

Questions regarding the lineage affiliation of DCs have been muddled significantly by the developmental plasticity of monocytes (6, 24). The phenotypic transformation of monocytes into DC-like cells is most prominent in inflamed environments (8, 19, 23, 24). It can also be mimicked *in vitro* by culturing monocytes in the presence of GM-CSF (granulocyte-macrophage colonystimulating factor) *±* IL-4 (Interleukin-4) (88, 89). However, *in vivo* the inflammation-induced differentiation of monocytes into cells with attributes of DCs appears GM-CSF-independent (90), highlighting that the developmental requirements underlying this phenotypic conversion *in vitro* might differ from those involved *in vivo*. In the absence of experimentally induced infection or inflammation, adoptively transferred monocytes readily acquire CD11c and MHCII expression as well as functional features of DCs in non-lymphoid tissues (91–95). This phenotypic conversion is also observed after adoptive transfer into unirradiated hosts, which most closely mimics steady-state conditions (63). In contrast, transferred monocytes do not generate DCs in lymphoid organs, even if the niche for engraftment is opened by depletion of CD11c<sup>+</sup> cells (84). Importantly, in nonlymphoid tissues monocytes exclusively generate CD11b+, but not CD103+CD11b*<sup>−</sup>* cells, which is in contrast to CDPs and pre-DCs that generate CD11b<sup>+</sup> as well as CD103+CD11b*<sup>−</sup>* cDCs (63, 91–95). Therefore, CD11c+MHCII+CD11b<sup>+</sup> cells in nonlymphoid tissues appear to constitute a population of mixed cellular origin that can arise from monocytic progenitors as well as pre-DCs. Adoptive transfer experiments do not allow to determine the relative contribution of each progenitor to this population,

#### **FIGURE 2 | Continued**

**(A)** Progenitors are adoptively transferred to assess their differentiation in the physiological context. Graft-derived cells are distinguished from host cells based on pre-defined labels, for instance congenic markers. This method is often combined with strategies to increase the niche available for cell engraftment, such as irradiation. **(B)** In transgenic approaches, lineage-restricted promoters can be used to drive a reporter gene. Target cell populations can be visualized by the expression of fluorescent proteins or can be depleted. In the latter case, cell-restricted expression of DTR allows for conditional cell ablation following DT injection. **(C)** Progenitors are transduced *in vitro* with semi-random DNA sequences (barcodes) by retro- or lentiviral vectors and subsequently

although surrogate markers such as CD64 or Mar-1 can serve to distinguish monocyte-derived cells from bona fide pre-DCderived cDCs (46, 93, 94).

Notably, in irradiated hosts transferred monocytes can also generate CD11c+MHCII<sup>+</sup> cells of the epidermis, which resemble Langerhans cells (LCs) (96–99). LCs exhibit many phenotypic and functional features of DCs, such as the capacity to migrate to lymphoid organs and stimulate naïve T cells, and have long been considered a prototypical DC population (96–99). However, we now realize that the majority of LCs is established before birth and maintained under steady-state conditions by self-renewal from local progenitors (96, 97, 99–102). These properties thus ontogenetically separate LCs from bone marrow-derived DCs or monocyte-derived cells. Moreover, monocytes may not necessarily adopt features of DCs or MØs upon entry into tissues, as a recent study indicates that monocyte can also exist in tissues without further differentiation (45). When considering this immense plasticity it will be crucial to elucidate the environmental cues that shape the diverse fates of monocytes to further dissect the full functional spectrum of monocytes and monocyte-derived cells.

# **Lineage Restricted Reporters**

When the availability of isolatable progenitor cells is limiting and when populations are ontogenetically heterogeneous or might be influenced by alterations in their surroundings, determining lineage affiliation requires models to trace cells directly in their natural environment. One way to achieve this is by engineering models in which lineage-restricted promoters or genetic elements drive the expression of reporter genes (**Figure 2B**) (80, 82, 103). It is important to bear in mind that such experiments assume that the expression of the selected marker is restricted to the cell lineage in question and therefore, the choice of stable and specific markers is essential (80, 82, 103). Additionally, the genetic elements used to drive expression of the reporter must faithfully mimic endogenous gene expression (80, 82, 103).

Genetic elements of the *Itgax* gene, which encodes CD11c, have extensively been used to generate reporters to study DCs (82, 103). As such, transgenic mice in which the CD11c promoter drives the expression of fluorescent proteins (**Figure 2B**) have been key to visualizing the distribution and cellular interactions of DCs in a variety of tissues, including lymphoid organs, heart, lung, and skin (103–107). But fate mapping can also be achieved by cell deletion. Transgenic expression of primate diphtheria toxin receptor (DTR) renders murine cells susceptible to diphtheria toxin (DT)-induced cell death and, thus, enables inducible target transferred into irradiated congenic recipients. After differentiation, cell progeny are analyzed for their barcode repertoire using deep sequencing or microarray. The representation of a given barcode in multiple cell populations indicates multi-potency of the transferred cell. **(D)** Expression of Cre recombinase is driven by a lineage or cell-specific promoter. Additionally, a reporter gene, usually a fluorescent protein, is placed under control of a constitutive promoter. Expression of the reporter is blocked by inserting a *loxP*-flanked *STOP* cassette (1). Cre expression leads to excision of the *STOP* cassette resulting in expression of the reporter gene (2). Since the promoter-driving reporter gene expression is constitutively active, the target cell is irreversibly labeled irrespective of continuous Cre expression (3).

cell depletion (**Figure 2B**) (82, 108). In this sense mice in which DTR expression is controlled by the elements of the CD11c promoter have been widely used to characterize the *in vivo* functions of DCs (28, 109–112). In part through analyzing such reporter mice, however, it has become evident that CD11c expression is not entirely restricted to DCs. It is also expressed on alveolar MØs, Ly6Clow as well as activated monocytes, plasmablasts, NK cells, and some T cells (25–29, 113). In addition, CD11c-driven fate reporter expression varies depending on the specific promoter elements used for transgenesis. CD11c.DTR mice, which were generated by conventional transgenesis using a 5.5-kb promoter element of the *Itgax* gene (109, 114), efficiently deplete most CD11c-expressing cDCs, LCs, alveolar, splenic marginal zone, and metallophilic MØs, as well as plasmablasts and T cells (27, 109, 115). However, DT-induced cell depletion in these mice is incomplete and spares certain cell types that transcribe their endogenous *Itgax* allele, including pDCs and NK cells (82, 115). Additionally, prolonged cell depletion using CD11c.DTR mice requires the use of bone marrow chimeras, possibly because of aberrant DTR expression on non-immune cells (82, 108, 112). Notably, this is not the case in CD11c.DOG and CD11c.LuciDTR mice, which were generated using bacterial artificial chromosome (BAC) transgenesis to place DTR under control of the extended regulatory region of the *Itgax* gene and in which DTR expression seems to more faithfully represent endogenous CD11c expression (28, 112, 115, 116). In all models, the occurrence of systemic neutrophilia and monocytosis following CD11c<sup>+</sup> cell depletion (28, 115, 117) adds another layer of complexity to deciphering the cellular function and lineage affiliation of DCs.

The realization that CD11c is not restricted to DCs in all instances nurtured the search for more specific lineagedefining markers. Two groups simultaneously identified the transcription factor Zbtb46 (zDC, Btbd4) as ideal candidate to distinguish cDCs, as it is expressed in pre-DCs and cDCs but not in pDCs or their precursors (37, 38). Consistently, CD8α <sup>+</sup> and CD11b<sup>+</sup> cDCs in lymphoid organs as well as CD103<sup>+</sup> cDCs in non-lymphoid organs uniformly express Zbtb46 as assessed in Zbtb46-GFP (37) and Zbtb46-DTR (38) reporter mice generated by site-directed mutagenesis. In contrast, CD11c+MHCII+CD11b<sup>+</sup> cells in non-lymphoid organs, including lung, small intestine, and kidney, exhibit partial Zbtb46 expression (37, 38) indicating that they represent a heterogeneous population. This is consistent with reports demonstrating that these cells are of mixed monocyte and pre-DC origin (63, 91, 92, 95). Subsequently, Zbtb46 reporter mice have been used to help establish lineage relationships in a variety of tissues including heart, pancreas, tumors, and thymus (118–121). The fact that Zbtb46 expression is also found in human DCs suggests that it may also help to identify DCs across species (48, 122).

Nevertheless, the use of Zbtb46 as lineage-defining marker requires a note of caution. Zbtb46 expression is downregulated after DC stimulation and it is found in some non-immune cells (37, 123). Despite its prominent expression in the cDC lineage, Zbtb46 appears largely dispensable for cDC development (37, 123). Instead, it may reinforce DC-specific transcriptional programs (37) and/or suppress DC activation (123). Interestingly, monocytes activated in the presence of GM-CSF *±* IL-4 uniformly induce Zbtb46 expression, whereas monocyte-derived Tip-DCs that are generated following infection with *Listeria monocytogenes* do not (37). This raises the possibility that Zbtb46 may control DC-like features of monocyte-derived cells in some inflammatory situations and it will be interesting to determine if Zbtb46 controls transcriptional programs in monocytes. These data also highlight that despite its selective expression on cDC progenitors and their descendants, Zbtb46 is not necessarily an indicator of cell ontogeny.

## **Identifying Common Developmental Requirements**

Establishing that the development and/or delineation of a cell type depends on a certain transcription or growth factor constitutes a powerful way of fate mapping that has extensively been applied to MPs (42, 51, 63, 124–141). We can now clearly delineate DCs into distinct subpopulations based on the transcriptional programs that govern their development. pDCs are distinguished from two subsets of cDCs by their dependence on E2-2 (67, 142). The differentiation of pre-DCs into CD8α <sup>+</sup> cDCs in lymphoid organs and CD103+CD11b*<sup>−</sup>* cDCs in non-lymphoid tissues is controlled by a set of transcription factors, including Irf8, Nfil-3, Id2, and Batf3 (124–128). Therefore, CD8α <sup>+</sup> cDCs and CD103<sup>+</sup> cDCs represent a developmentally related lineage of cDCs (6, 7). Notably, these cells also exhibit a degree of functional relatedness that is, for instance, exemplified by their superior capacity to activate CD8<sup>+</sup> T cells (124, 143–145). In contrast, the development of CD11b<sup>+</sup> cDCs from pre-DCs is controlled by distinct transcription factors, including RelB, RbpJ, PU.1, and Irf4 (42, 129–136). Notably, expression of CD24 separates pre-DCs into cells that preferentially generate either CD8α <sup>+</sup> or CD11b<sup>+</sup> cDCs in spleen (60) suggesting a stepwise differentiation of pre-DCs into cDCs. It will be interesting to determine whether such heterogeneity of pre-DCs also exists in the bone marrow. Notably, the extent of transcription factor dependence is linked to the genetic background of the particular mouse strain analyzed (146–148), indicating that transcriptional requirements are not always absolute or redundant factors exist (148). Consistently, CD8α <sup>+</sup> DCs can develop in the absence of Batf3, Id2, and Nfil-3 (149). The local microenvironment may also contribute to shaping the diversity of the DC compartment, as in some tissues, such as the spleen and intestinal system, CD11b<sup>+</sup> cDCs can be divided into ontogenetically and functionally distinct subpopulations (36, 42, 91, 95, 131). Importantly, some of the transcription factors controlling DC differentiation in mice have also been implicated in the development of human DCs (67, 69, 71) and putative equivalent DC subpopulations exist in rat, chicken, sheep, and pig (150–153), highlighting that DC populations are conserved across species.

While several growth factors have been linked to DC differentiation, the development of all DC subsets is strongly dependent on FLT3 ligand (FLT3L) and downstream signaling events (7, 18, 154). FLT3L administration potently expands pDCs and cDCs in mice and humans (72, 73, 85, 155–157). *In vitro,* FLT3L promotes the differentiation of bone marrow progenitors from mice, humans, and pigs into functional subsets of DCs (66, 158, 159). Mice lacking FLT3L display a severe deficiency in DCs, which is also apparent, although to a lesser extent, in mice lacking its receptor CD135 or mice treated with CD135 inhibitors (63, 137, 160, 161). In contrast, FLT3L appears largely dispensable for monocyte and MØ development (137) and, therefore, FLT3L dependency is often used delineate DCs *in vivo* (18, 65, 162). The interpretation of fate mapping using mice deficient in CD135 or its ligand is however complicated by the fact that these animals also exhibit abnormalities in other hematopoietic lineages, including B, T, and NK cells (137, 163) and show evidence of systemic neutrophilia and monocytosis, as has been reported in other DC-deficient models (112, 117).

Despite the prominent expression of CD135 on DC progenitors it remains to be clarified exactly at what stage of cellular differentiation FLT3L impacts on DC development. Consistent with a role for FLT3L early in development, a reduction of bone marrow CDPs in FLT3L deficient animals has been reported but ranges from a mere twofold decrease (164) to near complete absence (78). In contrast, the numbers of MDPs and splenic pre-DCs appear largely unaffected by CD135 deficiency (85). The observation that pre-DC frequencies in non-lymphoid organs of FLT3L-deficient mice are reduced (63) and that transfer of DCs into a FLT3Ldeficient environment decreases their homeostatic proliferation (85) indicates a role for FLT3L in the peripheral expansion of DCs rather than their differentiation. This interpretation would equally be consistent with the observation that DCs that develop in the absence of FLT3L are functional (137). In light of this finding it will be interesting to determine, to what extent FLT3L impacts on the development and functional regulation of other MPs. Addition of FLT3L to purified human monocytes cultured with GM-CSF *±* IL-4 increases their T cell stimulatory capacity (165), although it is not clear whether this is also the case for murine monocytes. Culture of murine bone marrow with GM-CSF and IL-4 presumably mimics monocyte differentiation under the same conditions (166). When FLT3 signaling is inhibited in such bulk cultures the T cell stimulatory capacity of the output cells is reduced (161). Therefore, these data raise the possibility that FLT3L might influence monocyte differentiation into cells with functional properties of DCs also in the murine system, although a direct causality remains to be demonstrated. Further, comparative gene expression profiling revealed that upon migration to lymph nodes LCs induce CD135 expression (167), indicating that they might be capable of responding to FLT3L. Therefore, it is conceivable that FLT3L may control certain functional aspects generally associated with DCs, such as antigen presentation, in ontogenetically distinct MP subtypes, which will be interesting to formally address in the context of FLT3L or CD135 deficiency.

Dendritic cell progenitors also express CD115, the receptor for MØ colony-stimulating factor (M-CSF) (56–58, 61, 86). However, compared to the dominant role of FLT3L in DC differentiation, M-CSF-deficiency only mildly impacts on DC development (168). M-CSF deficient osteopetrotic (op/op) mice exhibit a two- to threefold reduction in splenic cDCs and pDCs, respectively, but the remaining DCs are capable of stimulating a mixed lymphocyte reaction and induce costimulatory molecules upon activation, thus appear functional (168). In contrast, M-CSF is strongly required for monocyte and MØ development (141, 169). Therefore, the observation that mice lacking CD115 exhibit reduced frequencies of CD11c+MHCII+CD11b<sup>+</sup> cells in non-lymphoid organs (63, 91) likely reflects the ontogenetic heterogeneity of this population (63, 91–95). Consistently, M-CSF is also required for the generation of monocyte-derived cells with features of DCs during inflammation (90). Nonetheless, M-CSF may play a role in DC development. It can promote DC differentiation *in vitro* and *in vivo* even in the absence of FLT3L, although DCs generated by M-CSF alone phenotypically and functionally differ from those induced by FLT3L (170). M-CSF-induced DC poeisis is also more efficient in FLT3L-sufficient conditions (170). *In vivo*, antibody-mediated blockade of M-CSF in pregnant mice reduces pre-DC extravasation, translating into a reduction of CD11b<sup>+</sup> DCs in the pregnant uterus (171). Whether M-CSF affects pre-DC migration also in other tissues and whether it acts in a cell intrinsic manner or by promoting the production of chemotactic factors by other cells remains to be determined (171).

In purified monocytes, GM-CSF induces phenotypic and functional attributes of DCs (88, 89, 172). Similarly, purified CD115<sup>+</sup> MDPs respond to GM-CSF by differentiating into CD11c+MHCII<sup>+</sup> DCs (85) and GM-CSF deficiency leads to a slight reduction of bone marrow MDPs and CDPs (164). However, GM-CSF is dispensable for the differentiation of lymphoid tissue DCs (85, 173) and, therefore, it seemed likely that GM-CSF would selectively regulate the differentiation of monocytes into cells resembling DCs (23). This speculation also lead to the hypothesis that monocytes cultured in the presence of GM-CSF represent the counterpart of mo-DCs generated under conditions of inflammation/infection *in vivo* (23). Surprisingly, GM-CSF does not appear to control monocyte differentiation *in vivo* (90) and thus, GM-CSF elicited monocyte-derived cells are unlikely to be fully equivalent to inflammatory monocyte-derived cells. Rather, GM-CSF influences the homeostasis of cDCs in a variety, but not all, non-lymphoid tissues, most likely by promoting cell survival (90). Importantly, GM-CSF deficiency leads to a greater reduction of CD103<sup>+</sup> cDCs than of CD11b<sup>+</sup> cDCs (90). However, the extent of cDC reduction in the absence of GM-CSF apparently relates to the markers used for cell identification (90, 147, 164). This is most likely because GM-CSF regulates certain phenotypic as well as functional features of DCs, such as CD103 expression (174) or their ability to cross-present antigen (90, 174, 175). Therefore, the above-mentioned growth factors not only influence lineage decisions but also impact on the functional regulation of DCs, monocytes, and MØs. Elucidating the exact roles of FLT3L, GM-CSF, and M-CSF in each cell type will help to decipher the functional heterogeneity of MPs.

# **Cellular Barcoding**

The biggest challenge for fate mapping is to trace the developmental plasticity of individual cells. This can now be achieved using "cellular barcoding," in which progenitors are tagged *in vitro* with semi-random, non-coding DNA sequences by transduction using retro- or lentiviral vectors (**Figure 2C**) (81). Therefore, the barcodes are heritable and by choosing conditions of low transduction efficiency one can ensure that each cell receives only a single barcode. Subsequently, barcode-labeled progenitors are adoptively transferred in numbers low enough to minimize the chance that two identically barcoded cells are transferred into the same recipient (**Figure 2C**). After differentiation *in vivo*, cell progeny are analyzed for their barcode repertoire using deep sequencing or custom microarray. Since each barcode represents an individual progenitor, the presence of the same barcode in more than one cell type indicates that they were generated from a single precursor (multi-potent or bi-potent, **Figure 2C**). On the other hand, if a barcode is only found in one cell type, the progenitor generated only a single cell lineage (mono-potent, **Figure 2C**) (81).

During maturation, HSCs are thought to progressively lose their self-renewal ability and become increasingly limited in their differentiation potential, ultimately giving rise to lineagerestricted progenitors (55, 176). Lymphoid primed multi-potent progenitors (LMPPs) are developmental intermediates downstream of HSCs that can give rise to various, but not all, cell lineages and are thus considered multi-potent (55, 176). Surprisingly, in barcoding experiments only a minority (3%) of single LMPPs exhibits true multi-potency, defined as the ability to generate all of the following cell lineages: B cells, DCs, and myeloid cells (monocytes and neutrophils) (49). Rather, single LMPPs differ drastically in terms of their cellular output: 10% of the progenitors contribute primarily to B cells, 10% primarily to myeloid cells but about 50% of transferred LMPPs produce predominantly DCs (49). The remaining fraction of progenitors exhibits bipotentiality to generate combinations of the examined cell lineages (49). Therefore, LMPPs are multi-potent when analyzed as a population, however single cells exhibit unexpected lineage bias that is imprinted early in development. Why the majority of LMPPs is DC-committed (49), even though DCs constitute a minority lineage compared to B cells, remains to be clarified, although it is possible that some progenitors proliferate better than others or have certain competitive advantages. A major lineage divergence toward DCs seems to occur before or at the LMPP stage, as most HSCs analyzed by the same method are multi-potent, although even HSCs exhibit a degree of lineage bias (49, 177). Since CDPs might arise directly from LMPPs without additional developmental intermediates (79), these data infer that DCs diverge as a developmental lineage distinct from other myeloid cells early on (49).

This, again, questions the existence of a bi-potential MDP as central intermediate in the development of DCs and monocytes. Yet, it is noteworthy that even though DC-biased LMPPs are fivefold more frequent than bi-potent myeloid/DC LMPPs, monopotent and bi-potent progenitors contribute equally to the final DC pool (49). Therefore, bi-potent progenitors seem to play a significant part in generating DCs, potentially because they have a proliferative advantage. Resolving these issues will require further refinement of the technique at hand. The differentiation potential of progenitors may be influenced by cell isolation, processing or *in vitro* manipulation (80) and virus-mediated transformation might skew cell fate in an unforeseeable manner, as evidenced by the fact that barcoded LMPPs cannot generate T cells (49, 81). This also means that barcoding does not yet uncover the full potential of single progenitors. The early lineage bias of HSCs and LMPPs suggests that cell development may follow a model of graded commitment rather than proceeding in a truly stepwise manner (178). It will be interesting to determine, to what extent this process is regulated by epigenetic modification and how inflammatory processes might impact on lineage divergence. Future studies will benefit from the development of models allowing for *in vivo* barcoding of single cells but the labor-intensive quantification and analysis of barcoding experiments makes it difficult to follow populations in real time.

### **Genetic Lineage Tracing**

Dynamic mapping of populations of distinct origin *in vivo* can be achieved using genetic lineage tracing based on *Cre-loxP* technology (**Figure 2D**) (80, 179). It relies on inducible reporter genes that are placed under the control of constitutively active promoters, such as the *Rosa26* locus. The reporter is most commonly a fluorescent protein that is preceded by a *loxP-*flanked *STOP* cassette and, therefore, its expression is induced only after Cre recombinase (Cre) mediated excision of the stop codon (**Figure 2D**). Since this form of labeling is genetic it is also heritable, meaning that any cell expressing Cre will pass on the label to all progeny, irrespective of continuous recombinase expression (**Figure 2D**). Since the promoter driving the reporter gene is constitutively active, labeling is irreversible and not affected by fluctuations in gene expression (**Figure 2D**) (80).

By crossing mice expressing Cre under the control of the *Clec9a* locus to *Rosa26*-*STOP-flox*-enhanced-yellow fluorescent protein (YFP) reporter mice (180), we have recently generated the first genetic model to trace the progeny of DNGR-1<sup>+</sup> CDPs and pre-DCs (36). In these mice, YFP expression is restricted to DCs but is not found in monocytes or MØs even in inflammatory conditions, as tested after intestinal inflammation or infection with *L. monocytogenes* (36). Nonetheless, certain limitations need to be taken into account. DNGR-1 is also expressed on CD8α <sup>+</sup>/CD103<sup>+</sup> cDCs and to a lower extent on pDCs (36, 71, 181, 182) and, therefore, in these populations labeling is not a strict indicator of cell ontogeny. Further, labeling of CDP and pre-DC progeny in mice heterozygous for Cre is incomplete, possibly due to a delay in Cre protein synthesis and DNA recombination in rapidly cycling progenitors (36). Consistently, penetrance of the YFP label is increased in mice homozygous for Cre (36). The efficiency of lineage tracing experiments in such cases or when Cre expression is low may be improved by using alternate reporter constructs in which the *loxP* sites are positioned closer together, thus facilitating recombination (183).

Genetic lineage tracing does not require prior knowledge of which markers are expressed by the output cells and, thus, enables unbiased monitoring of cell ontogeny. Therefore, we were able to identify CDP-derived cells in cell populations previously thought to constitute monocytes/MØs based on the expression of surface markers, such as CD64 (36). CD64<sup>+</sup> CDP-derived cells do not express *Clec9a* message and are especially frequent in kidneys, although the presence of few YFP<sup>+</sup> cells in the CD64<sup>+</sup> component of lung and small intestine indicates that atypical CDP-derived cells also exist in other tissues (36). CD64<sup>+</sup> kidney DCs resemble yolk sac-derived F4/80hi tissue-resident MØs, appear to lack Zbtb46 expression (37) and their affiliation as DCs or MØs has been debated (184). We, therefore, used adoptive transfer as additional method to confirm cell ontogeny. Surprisingly, neither purified DNGR-1<sup>+</sup> CDPs nor total bone marrow generated F4/80hiCD64<sup>+</sup> CDP progeny in kidneys 1 week after adoptive transfer into irradiated recipients (36). Since kidney DCs reportedly have a slow turnover (185), it is possible that CDPs had insufficient opportunity to reach their renal niche and expand during short-term transfer experiments. Consistent with this notion, F4/80hiCD64<sup>+</sup> kidney leukocytes were efficiently generated from bone marrow progenitors in long-term reconstitution experiments (36). Therefore, our data strongly support a CDP origin of CD64<sup>+</sup> kidney leukocytes, despite their phenotypic resemblance to monocytes or MØs (36). These data exemplify the power of lineage tracing in following cell ontogeny in an unbiased way, although it is possible that DNGR-1 is expressed on yet unidentified developmental intermediates.

Addressing this possibility might require tamoxifen-inducible Cre constructs that can be used to pulse label progenitor populations (80). In the future, combinatorial approaches, such as "split-Cre" fragments controlled by two different promoters (186) or an intersection where Cre and the inducible reporter are driven by two cell-specific promoters (187, 188) may be of benefit to generate improved models to lineage trace DCs. The identification of CDP-derived cells with attributes of monocytes/MØs exemplifies the insufficiency of phenotypic properties, such as surface markers, as means of accurate cell identification of MPs. It also raises the question why cells of distinct ontogeny but overlapping phenotype exist in the same tissue. Further elucidation of the specific functions of MPs in immunity will benefit from lineage tracing approaches that result in target cell deletion through the use of inducible DTR or DT subunit modules (82, 112, 189, 190).

# **Conclusion**

The studies discussed above have significantly advanced our understanding of DC ontogeny but have also uncovered some uncertainties (**Figure 3**). While the bone marrow origin of DCs and monocytes is undisputed, the exact developmental intermediates and branching points between HSCs and DC progenitors remain to be clarified. Current data indicate that lineage imprinting toward DCs and monocytes may occur as early as LMPPs, potentially through epigenetic modification (**Figure 3**). This realization constitutes a major conceptual shift as it puts in question the existence of a bi-potential MDP and the resulting relatedness of DCs and monocytes. A definitive resolution of this question requires increasingly refined methods to genetically trace single progenitors or select DC and monocyte lineages. Nonetheless, it is clear that cDCs, pDCs, and monocytes can be separated based on their descendance from committed

including CD135, DNGR-1, Ly6C, and CD115 (as indicated in black). While pDCs fully develop in the bone marrow, cDCs arise via pre-DCs, which migrate through the blood stream to lymphoid and non-lymphoid tissues,

be delineated from cMoPs based on the expression of surface markers,

developmental intermediates (**Figure 3**). Their differentiation is further driven by unique factors indicating that their developmental paths are distinct (**Figure 3**). In stark contrast to pDCs, cDCs, and monocytes, most tissue MØs arise from embryonic progenitors and are predominantly maintained by self-renewal into adulthood (**Figure 3**).

Taken together, these data unequivocally establish that DCs, monocytes, and MØs develop as unique cellular entities and although one could argue that most of this knowledge is derived from mouse studies, developmental parallels have been observed in other species (66–73). Despite these advances, we are at a loss for a universal definition of DCs that is readily accessible to

monocyte-derived cells, and MØs in mice are shown. (\*) indicates that the specified marker can be expressed on ontogenetically distinct MP subtypes in some instances. (+/*−*) indicates markers with heterogeneous or tissue-dependent expression on the specified MP sublineage.

Poltorak and Schraml Fate mapping of dendritic cells

experts within and outside the field of MP biology. In light of this recognition, it has been suggested to revise the current nomenclature of MPs into a system that takes cell ontogeny into account when defining subpopulations (6). Such system would greatly aid our understanding of phagocyte biology as it remains uncertain to what extent the cellular origin of DCs, monocytes, and MØs determines the unique functionality of these cells in immunity and/or tissue homeostasis. While global profiling has revealed a role for the local tissue microenvironment in shaping the transcriptional landscape of DCs, monocytes, and MØs from different organs, certain gene signatures and transcriptional features are set by ontogeny (167, 191–193). Therefore, the full functional diversity of DCs, monocytes, and MØs is likely shaped by both nature (ontogeny) and nurture

# **References**


(the environment). Since ontogeny is immutable it provides a more robust common denominator for cell definition that enables deciphering cellular functions without assuming preconceived functional or phenotypic relationships. DC classification based on cell ancestry is a work in progress but its implementation will ultimately yield a more robust and transparent way of cell definition.

# **Acknowledgments**

BS and MP are funded by the German Research Foundation (Emmy Noether Grant: Schr 1444/1-1). We thank members of the Schraml lab and Clarissa Prazeres da Costa for critical reading of the manuscript.


with lymphocyte hyperactivation. *Eur J Immunol* (2011) **41**(2):291–8. doi:10. 1002/eji.201041169


cells during differentiation and regulates their expansion. *J Immunol* (2005) **175**(3):1399–405. doi:10.4049/jimmunol.175.3.1399


local microenvironment. *Cell* (2014) **159**(6):1312–26. doi:10.1016/j.cell.2014. 11.018

193. Gosselin D, Link VM, Romanoski CE, Fonseca GJ, Eichenfield DZ, Spann NJ, et al. Environment drives selection and function of enhancers controlling tissue-specific macrophage identities. *Cell* (2014) **159**(6):1327–40. doi:10. 1016/j.cell.2014.11.023

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

*Copyright © 2015 Poltorak and Schraml. 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 debate about dendritic cells and macrophages in the kidney

#### *Catherine Gottschalk and Christian Kurts\**

*Institute of Experimental Immunology, Rheinische Friedrich-Wilhelms-Universität Bonn, Bonn, Germany*

The mononuclear phagocyte system includes macrophages and dendritic cells (DCs), which are usually classified by morphology, phenotypical characteristics, and function. In the last decades, large research communities have gathered substantial knowledge on the roles of these cells in immune homeostasis and anti-infectious defense. However, these communities developed to a degree independent from each other, so that the nomenclature and functions of the numerous DC and macrophage subsets overlap, resulting in the present intense debate about the correct nomenclature. This controversy has also reached the field of experimental nephrology. At present, no mutually accepted way to distinguish renal DC and macrophages is available, so that many important roles in acute and chronic kidney disease have been ascribed to both DCs and macrophages. In this perspective article, we discuss the causes and consequences of the overlapping DC–macrophage classification systems, functional roles of DCs and macrophages, and the transferability of recent findings from other disciplines to the renal mononuclear phagocyte system from the nephrologist's point of view.

#### *Edited by:*

*Martin Guilliams, Ghent University – VIB, Belgium*

#### *Reviewed by:*

*Barbara Ursula Schraml, Technische Universität München, Germany Molly Ingersoll, Institut Pasteur, France*

#### *\*Correspondence:*

 *Christian Kurts, Institute of Experimental Immunology, Rheinische Friedrich-Wilhelms-Universität Bonn, Sigmund-Freud-Str. 25, Bonn D-53127, Germany ckurts@uni-bonn.de*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 15 May 2015 Accepted: 11 August 2015 Published: 01 September 2015*

#### *Citation:*

*Gottschalk C and Kurts C (2015) The debate about dendritic cells and macrophages in the kidney. Front. Immunol. 6:435. doi: 10.3389/fimmu.2015.00435*

Keywords: dendritic cells, macrophages, kidney, flow-cytometry, glomerulonephritis

# Introduction

The current intense debate regarding the classification and nomenclature of dendritic cells (DCs) and macrophages has reached also the field of experimental nephrology. Numerous kidney diseases are immune mediated, such as the different forms of glomerulonephritis, and research over the last years has described important, yet overlapping roles of both cells types.

Macrophages and DCs are often considered distinct cell types based on their morphology and function. Macrophages were defined as large vacuolar cells that are highly phagocytic and modulate immune responses by production of immune mediators (1, 2), whereas DCs were characterized as stellate migratory cells that act as sentinels in non-lymphoid tissues and enter lymphoid tissues upon antigen encounter, present antigen and subsequently activate naïve T lymphocytes (3–5). Following these original descriptions, two research areas developed that more or less independently studied macrophages and DCs. This artificial separation has contributed to the emergence of different names for similar or the same cell types, thereby adding to the current confusion about their identity and function. In particular, advances in multi-color flow cytometry and gene-analyses enabled researchers to define many DC and macrophage subsets by the expression of a variety of surface molecules (6). As cell surface markers are easy to determine, they are widely used to classify mononuclear phagocytes, although they are rather unspecific and their expression patterns in the murine and human systems differ substantially.

Also in the kidney, surface markers and functional parameters have been used to propose several classification systems of mononuclear phagocytes. However, these systems overlap, comparable to the situation in other non-lymphoid organs, resulting in great uncertainty among experimental nephrologists regarding the correct terminology. Here, we discuss the present state of our knowledge on renal mononuclear phagocytes in health and disease and problems resulting from the current nomenclature debate from a nephrologist point of view.

# The Network of Renal DCs and Macrophages

The kidney parenchyma consists of the outer renal cortex and the inner renal medulla. Numerous individual functional units, the nephrons, span both compartments. The cortex contains glomeruli and proximal tubuli of the nephrons, which generate the primary urine. The medulla contains the loop of Henle, which generates a high osmolarity that is required for water reabsorption from the primary urine. The distal tubules end in collecting ducts through which the concentrated urine is transported into the renal pelvis and on through the ureters into the bladder. The space between the tubules is known as tubulointerstitium and contains blood vessels, fibroblasts, and numerous cells of the hematopoietic system that had been classified by pathologists as constituents of the reticuloendothelial system [reviewed in Ref. (7)].

Early immunological studies had classified the tubulointerstitial mononuclear cells as macrophages due to their F4/80 expression (8). During the early 1990s, several groups reported that these tubulointerstitial cells morphologically resembled DCs in humans and rodents (9–12), whereas cells with the typical morphology of macrophages were described to reside mainly in the kidney capsule, the intravasal lumina, and the pelvic wall of healthy kidneys (13). The use of CX3CR1-reporter mice and live cell imaging illustrated the intricate tubulointerstitial network of dendritiform processes that these cells use to constantly probe the environment, suggestive of DCs in action (14–16). The nomenclature debate intensified when it became clear that the vast majority of renal mononuclear phagocytes possess the phenotype CD11c<sup>+</sup> CD11b<sup>+</sup> F4/80<sup>+</sup> CX3CR1<sup>+</sup> (17), which allows classification of both macrophages and DCs.

Notably, CX3CR1 exhibits relative organ specificity for renal mononuclear phagocytes: these cells were >50% reduced in the kidneys, but not in other organs (except the intestine) of CX3CR1 deficient mice. This may be explained by the comparatively high renal expression of its ligand CX3CL1 (18). Notably, those CX3CR1+ phagocytes that co-express CD11c and exert DC functionality were reduced even by more than 75% (18–20). This may result from an effect of CX3CR1 on CD11c expression, but this has yet to be shown. Interestingly, CX3CR1 regulated the numbers of the CD11c<sup>+</sup> and the CD11c<sup>−</sup> renal mononuclear phagocytes by different mechanisms: it promoted homeostatic and inflammatory recruitment of the former, whereas it prevented *in situ* proliferation of the latter under inflammatory conditions (20). Assuming that CD11c distinguishes renal DCs and macrophages, this difference would be consistent with recent reports that the number of tissue macrophages is regulated by local proliferation (21), whereas DC numbers are usually thought to be regulated by immigration and emigration (22).

The kidney also contains a minor subset of CD103<sup>+</sup> DCs, which constitute <5% of all renal CD11c<sup>+</sup> phagocytes and lack expression of CX3CR1, CD11b, and F4/80 (23), whose function currently is unclear. There are neither CD11b<sup>+</sup> CD103<sup>+</sup> DCs nor plasmacytoid DCs in the healthy kidney (24).

# Functionality and Phenotype of Renal Mononuclear Phagocytes

Researchers from both, the DC and the macrophage fields, have investigated kidney mononuclear phagocytes defined by cell surface markers in homeostasis and models of renal disease. Many important roles were shown in models of acute renal injury and in chronic immune-mediated kidney disease (**Table 1**), such as cytokine production or T cell-crosstalk in response to tissue injury or infection (17, 25–33). However, none of these functions is generally accepted to be exclusive for DCs or macrophages. Moreover, many nephrologists trained by the DC and macrophage communities still use CD11c and F4/80 to identify DCs and macrophages, respectively (see **Table 1**), even though 70–90% of renal mononuclear phagocytes co-express these two markers (17), implying that they studied cellular subsets that largely overlap. Also, the tools used for loss-of-function studies cannot clearly discriminate between DCs and macrophages: CD11c–DTR mice are used to deplete kidney DCs, CD11b–DTR mice for depleting kidney macrophages but the expression of CD11c and CD11b on kidney mononuclear phagocytes is too heterogeneous for this black-and-white thinking (34). Clodronate liposomes are used for both purposes (35–38). All kidney mononuclear phagocytes are phagocytic (34) which might render them sensitive to clodronate liposomes.

The consequence of this overlap is well illustrated by two recent studies examining how CX3CR1 affects renal disease: both studies agreed that mononuclear phagocytes are substantially reduced in the kidneys of CX3CR1-deficient mice. However, one of them noted a higher susceptibility to renal candidiasis and attributed this to the loss of renal macrophages (19), while the other documented protection against glomerulonephritis and assigned this to the loss of renal DCs (18). A possible explanation for this different classifications is that glomerulonephritis is driven mostly by phagocytes in the kidney cortex, in which glomeruli are located, whereas anti-infectious activity seem to be primarily due to phagocytes in the medulla, through which pathogens enter the kidney (18). Medullary phagocytes express significantly less CD11c than those in the cortex, which may bias their classification as DCs. The causes for these phenotypical and functional differences between medullary and cortical mononuclear phagocytes are unknown, but may result from differences in osmolarity, pH, and oxygen tension between these compartments, to which the mononuclear phagocytes may adapt. This would be in line with the current view that the tissue microenvironment dictates the organ-specific plasticity of macrophages (39, 40), and thus, perhaps also of renal mononuclear phagocytes.

# Re-Defining Kidney Mononuclear Phagocyte Nomenclature

The current definitions of renal DCs and macrophages are not mutually exclusive, so that renal mononuclear phagocytes may fulfill the definitions of both cell types simultaneously. This creates


TABLE 1 | Summary of the functions of mononuclear phagocyte subsets in renal diseases, which have been attributed to either renal DC or macrophages, based on marker expression and/or disease attenuation or aggravation after cell depletion.

#### Pro-inflammatory Pro-inflammatory

I. Crescent formation *DC Macrophages* (71) MHCII+, CD11c+, F4/80− (72)

III. Chemokine expression *Macrophages*

II. T cell infiltration and activation MHCII+, CD11c+ CD11b+, CD8−, B220−; depletion in CD11c–DTR mice; antigen presentation and T cell activation function (32) *DC* (32, 72) Chemokine expression by CD11b+ CD11c+ DC was analyzed in lymphoid organs (73) *DC* (73) Sensitive to diphtheria toxin in CD11b-DTR mice, CD68+ (71)

*(Continued)*

#### Frontiers in Immunology | www.frontiersin.org


confusion, especially among those nephrologists that are more interested in disease relevance than in semantics. A recent proposal for a unified nomenclature has been based on cellular ontogeny: it proposes an initial division of mononuclear phagocytes into macrophages, monocytes and monocyte-derived cells and DCs (so-called "level 1 nomenclature") (41). This classification was based on the following facts: (1) most adult macrophages in tissues are successors of an embryonic precursor and maintained through self-renewal (42–46), (2) a common monocyte progenitor (cMoP) exists, which gives rise to monocytes (47), and (3) conventional DC (cDC) and plasmacytoid DC but not monocytes or macrophages arise from a common DC precursor (CDP) (48, 49). Thus, tissue-resident macrophages were classified by their origin from embryonic (yolk sac and fetal monocytes) derived erythro-myeloid progenitors (46, 50) and DC were classified as cells arising from hematopoietic stem cell-derived precursors, identified by genetic tracing via DNGR1 (CLEC9A) (51), which are distinct from monocyte/macrophage precursors. Finally, monocyte-derived cells differentiate from cMoP that can exert macrophage- or DC-like functions and express markers associated with either (41). This classification does not resolve the question whether monocyte-derived macrophages and monocyte-derived DCs are ontogenically distinct or whether one cell type displays high plasticity in different microenvironments. To include cell function, location, and morphology, the authors suggested to add a "level 2" nomenclature to the level 1 classification (41).

While this nomenclature proposal might bring order into the ever increasing numbers DC and macrophage subsets, one major concern remains: without fate mapping tools, the origin of a phagocyte in a given tissue is usually not apparent, so that surrogate markers need to be used. Several markers for distinguishing phagocytes derived from different precursors are currently being discussed, but as we shall see below, they fail to discriminate renal DCs and macrophages.

One of these markers, CD64, alone or in combination with CCR2 or MerTK, has been reported to identify monocyte-derived macrophages and to be able to discriminate DC from non-DC in the intestine, the muscle and spleen (52–55), and the skin (56). DNGR1, when combined with genetic fate mapping technology, was shown to mark CDP and pre-DC (51), whereas Csf1r can be used for fate mapping of yolk sac derived (myb independent) tissue macrophages (46). In the kidney, most mononuclear phagocytes express CD64, low levels of CD11b and high levels of F4/80, which is not the case in other organs. However, 30% of CD64<sup>+</sup> cells co-expressed the DNGR1-fate mapper, indicating that CD64 expression, despite the evidence for specificity in other organs, does not differentiate CDP-derived from monocytederived cells in the kidney (51). Similarly, another fate mapping study that used Myb and PU.1 dependency for defining CD11bhi monocytes or macrophages and F4/80bright tissue macrophages derived by adult or embryonic hematopoiesis, respectively, found a dual origin in kidney macrophages as well (45). These findings highlight the difficulties when basing cellular classification solely on ontogeny when ontogeny is based on surrogate markers. Furthermore, transferring ontogeny-based nomenclature to human mononuclear phagocytes in tissue might prove impracticable.

A classification approach based on transcriptome analysis reported that CD11c<sup>+</sup> MHC II<sup>+</sup> cells in the kidney expressed a set of core DC markers characteristic of DCs in non-lymphoid tissues, that is absent from macrophages, including Zbtb64, Flt3, and CCR7 (57). These "core DC markers" had been defined by analyzing cDCs except the CD11b<sup>+</sup> non-lymphoid tissue–DC, because of the great heterogeneity of CD11b<sup>+</sup> cells. However, these constitute the vast majority of kidney mononuclear phagocytes.

Another classification approach is based on mononuclear phagocyte functionality. However, observed functions generally represent a snapshot of a cell within a specific context and time frame. Demonstrating that a phagocyte performs a given function under certain conditions at a certain time-point does not imply that this is a general feature of this cell. Furthermore, there is no clear demarcation between exclusive DC and macrophage functions. For example, macrophages phagocytose and degrade material. However, under certain conditions DCs do that too, albeit less efficiently [reviewed in Ref. (58)]. On the other hand, DCs classically activate naïve T cells, but macrophages can do that too, albeit less efficiently (59, 60). Furthermore, the ability to stimulate T cells is difficult to determine on a single cell basis. A recent study differentiated renal mononuclear phagocytes into five phenotypically and functionally distinct populations (34). In that study, mononuclear phagocyte populations were differentiated by CD11c, CD11b, F4/80, CD103, CD14, CD16, and CD64 expression in juvenile and adult mice of different strains. Functional analyses and fate mapping studies were used for further characterization. In line with the complexity of kidney mononuclear phagocyte subsets observed by others and us (17, 45, 51), the study revealed that all subsets expressed CD68 that is usually used to identify macrophages and that all subsets were phagocytic but showed differences in their antigen presentation capacity. Fate mapping experiments identified one population with a dual origin, two populations that were closely related to monocytes, whereas the remaining two were not. Notably, the largest population not only showed the phenotypical and functional characteristics of reparative macrophages (M2) but also had significant antigen presentation function and most likely emigrated from the kidney under inflammatory conditions. Additionally, this population differed significantly between mouse strains, which might explain immunological differences between those strains. The authors concluded that functions are more related to context than separate lineage and suggested their marker combination as an unbiased approach to identify kidney mononuclear phagocyte populations (34). These findings are consistent with recent concepts that macrophage fine differentiation is shaped by the tissue microenvironment (39, 40).

# Concluding Remarks

As a consequence of the separate development of the DC and macrophage research communities, the functional and phenotypic definitions of these cell types overlap substantially. Thus, scientists from both communities often study the same cells, perhaps unaware of, or ignoring progress and concepts in the other field. The false assumption that classifying a mononuclear phagocyte as a macrophage implies that it is not a DC, and vice versa, hampers communication between researchers from both fields. Some studies have focused on arguing about subsets and semantics (61), perhaps hoping to "claim territory" for their own communities. This may result in highly citable or controversial publications, but it does not advance our understanding of mononuclear phagocytes, neither in the kidney nor elsewhere.

An overlapping classification system, such as the existing one, is certainly not desirable. An improvement is needed. It is unrealistic to assume that either the DC or the macrophage community

# References


will accept the nomenclature of the other field. Drawing a line that segregates mononuclear phagocytes into DCs or macrophages will unlikely be acceptable to both fields. Furthermore, there are currently no unambiguous discriminatory parameters; for any new parameters introduced, exceptions are reported quickly, such as for CD64 and DNGR1-fate tracking in the kidney. Still, an improved classification system is needed. How can we reach a consensus?

First, the purpose of the revised classification system needs to be defined. Clinicians are interested in cellular entities that are useful for diagnostic or therapeutic purposes and translational immunologists often study the functions of cellular subsets. Basic immunologists may favor ontogeny, which is biologically the cleanest and most logical approach. However, mononuclear phagocytes adapt their gene enhancer landscape according to the tissue of residence independently of the precursor they originated from (39), an ontogeny-based nomenclature may lead to different cell types with similar functionality, or to cells of the same name with different functionality depending on the organ they reside in. Moreover, the origin of a mononuclear phagocyte in a given tissue is not obviously apparent, because unique discriminatory parameters are missing. Thus, ontogeny, although theoretically logical, will be difficult to use for routine research. At the end of the day, a classification system needs to be convenient and feasible, or it will not be used.

The late Ralph Steinman remarked "The DC is a functional state" (personal communication). Indeed, at the age of single cell transcriptomics, it becomes clear that several transcriptional programs may run simultaneously in individual mononuclear phagocytes, and confer a spectrum of functionalities that are more or less consistent with the current concepts of a DC, of a macrophage, or both. Current technical advances will undoubtedly allow distinguishing far more functional states of mononuclear phagocytes. In the field of renal immunology, experts coming from the DC and macrophage communities have jointly suggested avoiding the DC–macrophage controversy altogether by referring to mononuclear phagocytes (preferentially using a "catchier" name for these cells), with different degrees of DC- or macrophage-, or other functionalities (62). It remains to be seen whether basic immunologists and scientists studying mononuclear phagocytes in other organs feel that this is useful or not.

# Acknowledgments

We thank Natalio Garbi and Wolfgang Kastenmüller for helpful discussions. This work was supported by the Deutsche Forschungsgemeinschaft (grants KFO228 and SFBTR57). CK is a member of the Excellence-Cluster ImmunoSensation in Bonn.


<sup>3.</sup> Steinman RM, Cohn ZA. Identification of a novel cell type in peripheral lymphoid organs of mice. I. Morphology, quantitation, tissue distribution. *J Exp Med* (1973) **137**(5):1142–62. doi:10.1084/jem.137.5.1142


**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 Gottschalk and Kurts. 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.*

# **Guardians of the gut – murine intestinal macrophages and dendritic cells**

*Mor Gross 1,2, Tomer-Meir Salame1,2 and Steffen Jung1,2 \**

*<sup>1</sup> Department of Immunology, Weizmann Institute of Science, Rehovot, Israel, <sup>2</sup> Biological Services, Weizmann Institute of Science, Rehovot, Israel*

Intestinal mononuclear phagocytes find themselves in a unique environment, most prominently characterized by its constant exposure to commensal microbiota and food antigens. This anatomic setting has resulted in a number of specializations of the intestinal mononuclear phagocyte compartment that collectively contribute the unique steady state immune landscape of the healthy gut, including homeostatic innate lymphoid cells, B, and T cell compartments. As in other organs, macrophages and dendritic cells (DCs) orchestrate in addition the immune defense against pathogens, both in lymph nodes and mucosa-associated lymphoid tissue. Here, we will discuss origins and functions of intestinal DCs and macrophages and their respective subsets, focusing largely on the mouse and cells residing in the lamina propria.

#### *Edited by:*

*Martin Guilliams, Ghent University – VIB, Belgium*

#### *Reviewed by:*

*Masaaki Murakami, Hokkaido University, Japan Amir Ghaemmaghami, The University of Nottingham, UK*

#### *\*Correspondence:*

*Steffen Jung, Weizmann Institute of Science, Department of Immunology, Rehovot 76100, Israel s.jung@weizmann.ac.il*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 23 March 2015 Accepted: 07 May 2015 Published: 02 June 2015*

#### *Citation:*

*Gross M, Salame T-M and Jung S (2015) Guardians of the gut – murine intestinal macrophages and dendritic cells. Front. Immunol. 6:254. doi: 10.3389/fimmu.2015.00254* **Keywords: gut, dendritic cells, macrophages, homeostasis, inflammation, IBD**

# **The Unique Characteristics of the Gut Landscape**

Intestinal mononuclear phagocytes are located in a unique anatomic environment that necessitated the evolution of special functional adaptations of these cells. Exposure to commensal bacteria and harmful pathogens, as well as nutrients and food antigens, in the intestinal lumen force the immune system to continuously weigh tolerogenic and protective immune response. Disruption of this critical and delicate balance can result in devastating inflammatory reactions, e.g., hyper-reactivity to food components (1) or inflammatory bowel diseases (IBD), such as Crohn's disease or ulcerative colitis (2).

Both dendritic cells (DC) and macrophages are found spread throughout the connective tissue that underlies the epithelial layer of the gut, the lamina propria. Moreover, representatives of the two main mononuclear phagocyte families are also located in mucosa-associated lymphoid tissue (MALT), including Peyers' Patches and isolated lymphoid follicles (ILFs) (3). DC and macrophages have distinct, yet complementary roles in maintaining gut homeostasis and immune defense. In keeping with their migratory capacity, DC translocate from the lamina propria via the lymphatics to the gut-draining mesenteric lymph nodes (MsnLNs), where they present antigens to naïve T cells, polarize them toward effector fates, and thus establish the adaptive branch of the immune system (4).

Macrophages, on the other hand, are believed to contribute to the local clearance of bacteria from the tissue, translate alert signals to other immune cells, secrete cytokines to establish the local homeostatic immune cell network, and participate in T cell re-stimulation and maintenance within the lamina propria (5).

DC and macrophages can, as discussed in detail below, be divided into several subpopulations with defined origins, overlapping and distinct surface marker profiles, functions and


#### **TABLE 1 | Mononuclear phagocytes and their respective subsets in the lamina propria of the mouse intestine**.

locations. The best characterized DC and macrophage subsets and their key features are summarized in **Table 1**.

With this review, we provide an overview on the characteristics and function of intestinal macrophages and DC in the mouse, including specific roles of their subpopulations. We will discuss distinct origins, roles in maintaining gut homeostasis, and the interactions between these cells and other immune cells. Finally, we will review their communication with their nonimmune microenvironment and elaborate on emerging roles of macrophages and DC in inflammation.

# **Intestinal Macrophages**

Macrophages are the most abundant mononuclear phagocytes in the steady-state gut lamina propria (3, 24). Intestinal macrophages are currently best characterized by their expression of CD64, the Fcγ receptor 1 (FcγRI) (25), and the chemokine receptor CX3CR1 (18), as well as the F4/80 antigen (EGF-like module containing mucin-like hormone receptor-like 1-EMR1) and the integrins CD11b and CD11c (26). Due to the high surface expression levels of the chemokine receptor CX3CR1 by gut macrophages, these cells can also be readily detected, isolated, and studied *in situ* using intra-vital microscopy on mice harboring a GFP reporter gene inserted into the CX3CR1 locus (27).

# **Ontogeny**

Like other tissue macrophages (28), also intestinal macrophages are first established before birth from precursors originating in the yolk sac or fetal liver (29). However, unlike macrophages in most other tissues, these embryo-derived cells are replaced in the gut shortly after birth by cells that derive from Ly6C<sup>+</sup> blood monocytes (29). The adult monocyte-derived cells display a uniquely short half-life for macrophages (30) indicating their continuous renewal. The monocytic origin of intestinal macrophages was first established in adoptive transfer experiments, involving the transfer of CX3CR1gfp monocyte-precursors and monocytes into CD11c-DTR transgenic mice, whose CD11cexpressing cells, including intestinal macrophages, were depleted by a diphtheria toxin challenge (11, 19, 31). During their differentiation into gut macrophages, monocytes lose Ly6C expression, while other surface markers, such as MHCII, F4/80, CD64, CD11c, and CX3CR1 are up-regulated (25, 32, 33). Moreover, the cells acquire a characteristic anti-inflammatory gene expression profile (32, 34), whose timely establishment and maintenance are critical for gut homeostasis (35). This includes the expression of IL-10, TREM-2, IRAK-M, and tumor necrosis factor (TNF)AIP3 genes, but also of TNFα, which has both proand -anti-inflammatory activity (32). Of note, this expression profile is robust, as it seems to withstand acute challenges, such as the ones associated with oral dextran sulfate sodium (DSS) exposure (32). The molecular cues that drive the "education" of the macrophages in various regions of the gut remain to be defined, but the epithelium is likely to play a role in this process. Epithelial cells could control macrophage differentiation by secretion of immune-regulatory factors, such as thymic stromal lymphopoietin (TSLP), transforming growth factor-β (TGF-β), and prostaglandin E-2 (PGE-2) (36). In addition, recent findings suggested that semaphorin 7A, which is secreted by epithelial cells, contributes to the induction of IL-10 expression by CX3CR1<sup>+</sup> intestinal macrophages (37). Also, colony-stimulating factor 2 (Csf-1; previously named macrophage colony-stimulating factor, M-CSF) and colony-stimulating factor 2 (Csf-2; previously named granulocyte-macrophage colony-stimulating factor, GM-CSF) play a role in the development of macrophages. Csf-1 is a crucial factor for monocyte development, as Csf-1-deficient osteopetrotic (op/op) mice display reduced levels of F4/80<sup>+</sup> cells in the small and large intestine after the first few days of life (28, 38, 39). Csf-2-depleted mice were shown have reduced numbers of CD11c<sup>+</sup> colonic macrophages (20).

Of note, Ly6C<sup>+</sup> monocytes fail to acquire the characteristic macrophage quiescence during intestinal inflammation, but under this condition respond to local factors that trigger pattern recognition receptors, such as TLRs and NLRs, giving rise to proinflammatory macrophages (32). These pro-inflammatory cells, which in acute inflammation outnumber the resident macrophage population, secrete IL-12, IL-23, TNF-α, and inducible nitric oxide synthase (iNOS) (32).

A key suppressor of macrophage-associated inflammation is the IL-10/IL-10 receptor (IL-10R) axis, as mice bearing mutations in IL10-Ra in intestinal CX3CR1<sup>+</sup> macrophages developed severe colitis (35) comparable to the pathology reported for IL-10-deficient animals (40). This central critical role of IL-10 in maintaining the non-inflammatory state of macrophages, and thereby, gut homeostasis is also supported by research conducted on samples from humans with loss of function mutations in IL-10R (41). The latter provides an explanation for the severe early onset of colitis observed in pediatric patients harboring nonsense and missense mutations in IL-10R, which reduce IL-10R expression and hamper its signaling cascades (42). Interestingly though, IL-10 production by intestinal macrophages, although also prominent, seems to be redundant for the maintenance of gut homeostasis (35); rather the system seems to rely on alternative IL-10 sources, such as Treg cells (43).

Homeostatic monocyte recruitment to the gut is thought to depend on the chemokine receptor CCR2, as CCR2-deficient mice display less intestinal macrophages and CCR2-deficient intestinal macrophages are underrepresented in mixed bone marrow chimeras (24, 25). The exact factors and mechanisms that ensure homeostatic Ly6C<sup>+</sup> monocyte recruitment to the steady state gut are, however, still unknown. While they are likely related to the microbiota exposure of the tissue, analysis of germ-free animals has yielded conflicting results (29, 34, 44, 45). The latter could be due to intestinal embryo-derived macrophages that might persist in the absence of arising competition by an adult monocyte influx.

# **Macrophage Heterogeneity**

Interestingly, emerging evidence suggests that intestinal macrophages are more heterogeneous than previously thought. Monocyte-derived CD11b<sup>+</sup> CX3CR1<sup>+</sup> cells in the gut comprise both CD11c<sup>+</sup> and CD11c*<sup>−</sup>* cells. While differential functions of these cells remain to be established, studies into this matter might profit from the recent finding that generation of CD11c+, but not CD11c*<sup>−</sup>* CX3CR1<sup>+</sup> intestinal macrophages requires Notch signaling (21). A subpopulation of CD169-expressing CX3CR1<sup>+</sup> macrophages has been reported to be associated with the intestinal crypts (22), although these cells will require further functional characterization. Bogunovic and colleagues recently reported an intriguing CX3CR1<sup>+</sup> macrophage subpopulation that resides in the muscularis layer and communicates with enteric neurons to regulate gastrointestinal motility (23). Importantly, we and others have recently shown that macrophages isolated from distinct tissues, such as the liver, lung, brain, and peritoneum, differ considerably with respect to their gene expression profile (46, 47). As expected, this diversity is also prominently reflected in the differential enhancer usage of these cells, as inferred from highly divergent histone modifications (47). Moreover, given that the number of regulatory elements by far exceeds the number of genes (48, 49), this heterogeneity is even more pronounced, including both active and poised enhancer states (47). This applies, albeit to a lesser extent, also to macrophages located in proximal and distal segments of the gut (47). Epigenetic heterogeneity of intestinal macrophages likely reflects monocyte exposure to distinct environmental cues in ileum and colon during their local differentiation (32, 47). In-depth understanding of how these macrophage identities are established, including the hierarchy of induced transcription factors, could yield valuable insights into monocyte differentiation that might be applicable to other tissues and inflammatory settings. PU.1 is a pioneering factor, which induces c-fms transcription and is hence required for macrophage differentiation (50). Intestinal macrophages are furthermore characterized by prominent expression of the Runt-related transcription factor 3 (Runx-3) (47). Interestingly, mice that harbor Runx3 deficiency develop spontaneous colitis (51). Other candidates that might be involved in the establishment of the intestinal macrophage signature are the interferon regulatory factors 4 and 5 (Irf-4, Irf-5), shown to be associated with classical and alternative macrophage activation, respectively (52–54).

# **Macrophage Interactions with Their Environment**

# **Macrophage Communication with the Epithelial Cell Layer**

Pioneering studies by Rescignio and colleagues revealed that certain intestinal mononuclear phagocytes can penetrate the intestinal epithelium by virtue of expression of tight junction proteins and formations of dendritic projections (55). These structures, later termed trans-epithelial dendrites (TEDs) (56), were subsequently ascribed to macrophages expressing CX3CR1 (18) and allegedly allow these non-migratory cells to sense, and potentially sample, the luminal content (18, 56). TED formation by macrophages in the terminal region of the ileum was found to be dependent on expression of both CX3CR1 macrophages and its membrane-tethered ligand CX3CL1/Fractalkine by selected epithelial cells (57). CX3CR1-deficient and CX3CL1-deficient mice were reported to be relatively protected from acute, DSSinduced colitis (58) – a phenotype that might be related to TED formation (57). Likewise, CX3CR1-deficient mice were shown to display impaired oral tolerance, which was related to impaired IL-10 production by intestinal macrophages, though not their TED formation (59). Finally, there is evidence for a potential role of CX3CR1<sup>+</sup> macrophages in the capture of luminal bacteria (60) and even the transport of the latter to lymph nodes, at least under conditions of dysbiosis (61). However, the exact definition of macrophage contributions in their native tissue context remains challenging, because it requires their accurate discrimination from closely related and phenotypically similar monocyte-derived DC.

Apart from their role in maintaining intestinal immune homeostasis, gut macrophages also contribute critically to epithelial wound healing. Macrophages associated with the crypts of Lieberkuehn in the colon were reported to assist, following tissue damage, the proliferation and survival of epithelial progenitor cells in a Myd88-dependent manner (62–64). Moreover, in a murine model of acute epithelial regeneration in the colon, activated macrophages supported tissue repair by up-regulating expression of IL-3 and IL-4, while inhibiting secretion of TNF and interferon-γ (IFN-γ) in the lamina propria (3, 65). Macrophages also appear to be able to influence the permeability of the epithelium barrier via the secretion of IL-6 and NO, thereby potentially increasing the invasion of pathogens (66).

# **Communication with Immune Cells**

Macrophages are inferior to DC in their ability to prime naïve T cells (67). This might be due to their rapid degradation of ingested proteins, which impairs their ability to retain antigens for presentation (68). Moreover, at least in steady state, intestinal CX3CR1<sup>+</sup> macrophages lack expression of CCR7, i.e., the chemokine receptor required for migration to the MsnLNs (25, 69). Rather, the cells that reside in the lamina propria have been proposed to maintain the functionality of FoxP-3<sup>+</sup> T regulatory cells that migrated back from the MsnLNs into the tissue (59). Thus, while Treg cell generation of CX3CR1-deficient mice is unimpaired, these animals harbor reduced Treg cell numbers in the lamina propria, a phenotype that is associated with impaired oral tolerance (59). In light of other data (70), the authors of this study linked the reduced FoxP-3<sup>+</sup> Treg cell numbers to impaired production of IL-10 by CX3CR1<sup>+</sup> macrophages (59). However, the latter might have to be revised, since CX3CR1Cre:IL10fl/fl mice were shown to harbor unimpaired FoxP-3<sup>+</sup> Treg cell numbers (35). Also, interactions between CX3CR1<sup>+</sup> macrophages and Th17 cells, which are rarely found in intestinal lymphoid tissues and, though primed in the MsnLN, might terminally differentiate in the lamina propria, remain incompletely defined. On one hand, it was shown that intestinal CD70hi CX3CR1<sup>+</sup> macrophages are activated by commensal-derived ATP and drive the *in vitro* differentiation of Th17 cells (71, 72). On the other hand, intestinal macrophages were reported to counteract Th17 generation that is promoted by CD103+CD11b<sup>+</sup> DC (73, 74). Of note, CD103+CD11b<sup>+</sup> DCs and Th17 cells co-localize in the intestinal tract, as the number of both cells drop from the duodenum to the ileum, and they are scarce in the colon. By contrast, CX3CR1<sup>+</sup> macrophages and FoxP3<sup>+</sup> Treg cells are most abundant in the colon (74).

Recent findings revealed an intriguing cross-talk between intestinal macrophages and innate lymphoid cells (ILC). Thus, in response to luminal stimuli and using a signaling pathway involving the TLR adaptor Myd88, macrophages were shown to secrete IL-1β and in turn induce production of csf-2 by RORγt + type 3 ILC (20). Mice lacking Csf-2 display reduced numbers of colonic macrophages and DC, associated with a hampered Treg cell compartment (20). Moreover, in a *Citrobacter* infection model CX3CR1<sup>+</sup> macrophages were shown to promote ILC production of IL-22 via secretion of IL-23 (75), in line with another report (76). Interestingly, CX3CR1<sup>+</sup> macrophage-derived IL-23 not only induces IL-22 but also seems to concomitantly suppress IL-12 production by CD103<sup>+</sup> CD11b*<sup>−</sup>* DC and thereby prevents otherwise detrimental immunopathology (77). Notably, the latter finding provides first evidence for the existence of a direct crosstalk among intestinal mononuclear phagocytes in tissue context, a topic that clearly deserves further study.

# **Intestinal Dendritic Cells**

Dendritic cells are specialized in communicating with T cells, curbing autoreactivity and activating T cell immunity in response to threats. Specifically, DC provide T cells with antigenic peptides that are presented in MHC context, co-stimulation and instructing cytokines that govern T cell polarization into effector cells (67). In order to maintain homeostasis and avoid inflammatory responses toward innocuous antigens, gut DC employ tolerogenic mechanisms that allow them to dampen adaptive immunity. MsnLNand lamina propria-resident CD103<sup>+</sup> DC secrete, for example, retinoic acid (RA) and transforming growth factor-β (TGF-β), which promote the generation of Foxp3<sup>+</sup> Treg cells and contribute to the differentiation of plasma cells, which secrete IgA (78, 79).

# **Classification and Ontogeny**

Intestinal DC in mice are characterized by the surface expression of the integrins CD11c (αX) and CD103 (αEβ7) (11, 19, 69). More recently, CD24 and Sirpα have been introduced for the better discrimination of DC from macrophages (8, 10). CD103<sup>+</sup> DC in the gut arise from dedicated DC precursors, or preDC, and accordingly, mice deficient for fms-related tyrosine kinase-3 receptor (Flt-3) or its ligand Flt-3L have significantly decreased levels of intestinal DC (7, 19). Other, currently though less wellcharacterized DC progenitors are α4β7 <sup>+</sup> so-called "pre-μDC," which are generated in the bone marrow and were shown to give rise to classical CD103<sup>+</sup> DC and CCR9<sup>+</sup> plasmacytoid DC (80).

Classical CD103<sup>+</sup> DC are divided into two major subpopulations according to their expression of CD11b (αM) (81). CD103<sup>+</sup> CD11b<sup>+</sup> DC and CD103<sup>+</sup> CD11b*<sup>−</sup>* DC display distinct abundance in small and large intestine, present different additional surface markers, and require different growth factors for their development (82, 83).

CD103<sup>+</sup> CD11b<sup>+</sup> DC are developmentally related to CD11b<sup>+</sup> CD8α *−* splenic DCs (15) and found in the lamina propria of the small and large intestine. They can migrate in CCR7-dependent manner (84) to the MsnLNs, where they present luminal antigens to T cells. CD103<sup>+</sup> CD11b<sup>+</sup> DCs likely represent a heterogeneous population, as a fraction of them is Csf-2-dependent (3). Development of CD103<sup>+</sup> CD11b<sup>+</sup> DC, but not of CD103<sup>+</sup> CD11b*<sup>−</sup>* DC, is hampered in Csf-2R-deficient mice (85) and when expression of Notch-2 (12, 76) or IRF-4 (14) is impaired. Moreover, CD103<sup>+</sup> CD11b<sup>+</sup> DC numbers are also reduced in absence of RA and under conditions of vitamin A deprivation (15).

CD103<sup>+</sup> CD11b*<sup>−</sup>* DC are more prevalent in lymphoid organs – the Peyer's Patches, MsnLNs, and ILFs (7, 69). However, they can be found also in animals lacking these structures, and are hence not limited to lymphoid tissues (3). Similar to classical CD8α + DC in the spleen, CD103<sup>+</sup> CD11b*<sup>−</sup>* DC depend on the expression of the transcription factors BatF-3 and Irf-8 (6, 15). Like the former, they also express the chemokine receptor XCR1 that has emerged as a universal marker for this DC subset in mouse and human (8, 9). The connection between CD103<sup>+</sup> CD11b*<sup>−</sup>* DC and CD8α <sup>+</sup> DC is also supported by the fact that the number of CD103<sup>+</sup> CD11b*<sup>−</sup>* DC was shown to increase, alongside with splenic CD8α <sup>+</sup> DC, in mice that display constitutive β-catenin activation (86). Moreover, like splenic CD8α <sup>+</sup> DC (87), also CD103<sup>+</sup> CD11b*<sup>−</sup>* DC are specialized in cross-presentation (88).

The exact definition of intestinal DC is complicated, since monocyte-derived cells can acquire phenotypic and functional DC hallmarks. Studies have described a population of CD103*−*CX3CR1+CD11b<sup>+</sup> DC, which resides in the lamina propria (11, 16). These cells are CSFR-1 dependent and appear to be derived from Ly6Chigh monocytes (11). Recent studies also reported that under inflammatory conditions, these CD103*−*CX3CR1+CD11b<sup>+</sup> DC expressed CCR7 and migrated in the intestinal lymph, similar to classical intestinal DC, and induced the differentiation of IL-17 and IFN-γ producing T cells (16, 17).

# **Antigen Sensing and Uptake**

CD103<sup>+</sup> DC, present in the lamina propria and associated with the intestinal epithelium lining the villi, provide surveillance of the luminal environment (30). They detect foreign and inflammatory signals, acquire and present antigens and interact with T cells by migrating to secondary lymphoid organs (3). Located deep in the core of the villous lamina propria, CD103+CD11b<sup>+</sup> DC would seemingly have limited access to luminal signals, unless antigens or bacteria cross the epithelium or are imported into the lamina propria by other cells, e.g., macrophages, epithelial M cells, or small intestine goblet cells (36, 89, 90). However, lamina propria-resident CD103<sup>+</sup> DC were shown to migrate into the epithelial cell layer and capture bacterial antigens (90).

# **DC Migration**

Mucosal T cell priming, arguably one of the primary roles of gut DC, is believed to be restricted to lymphoid tissues (3). Intestinal DC are hence bound to migrate from the lamina propria to the MsnLNs, or within Peyer's Patches into T cell zones. Indeed, CD103<sup>+</sup> DC were detected in the intestinal lymph under homeostatic conditions (69, 84). In addition, after systemic BrdU administration, labeled CD103<sup>+</sup> DC were found in the lamina propria before they could be discerned in the MsnLNs (30). LNresident CD103<sup>+</sup> DC are thus derived from the tissue and constantly immigrate (30, 91). Interestingly, steady state migration of intestinal CD103<sup>+</sup> DC does not appear to be induced by the microbiota or by TLR signaling (92), but may rather depend on a low, tonic release of inflammatory cytokines, or result from spontaneous DC maturation. Nevertheless, entry of CD103<sup>+</sup> DC into the MsnLNs is of course considerably enhanced by proinflammatory cytokines or TLR ligands (93, 94). Migration of intestinal DC depends on CCR7, both in steady state and under inflammatory conditions. Accordingly, CCR7 expression is upregulated in DC before their migration from the tissue into the MsnLN (84) and CCR7 deficient DC fail to migrate (69, 84, 95). Moreover, it was recently shown that DC can also migrate from the lamina propria into the epithelial layer (90) and can thus gain direct access to antigen and luminal bacteria. Hence, following challenge with *Salmonella*, accumulation of the bacteria was first observed in DC of the epithelial fraction and only subsequently in DC in the lamina propria (90).

# **DC and the Epithelium**

DC intimately interact with the epithelial layer of the intestine by a variety of mechanisms. Small intestinal goblet cells were shown to transfer small soluble antigens from the intestinal lumen to CD103<sup>+</sup> DC (89). Chemokines secreted by enterocytes in response to TLR ligand exposure can induce the above-mentioned relocation of lamina propria DC to the epithelium (90). In addition, it is becoming more and more evident that epithelial cells play a critical role in maintaining DC in a tolerogenic state, compatible with gut homeostasis. Epithelial and stromal cells secrete factors, which are thought to induce DC tolerance, such as RA, TGF-β, PGE-2, and TSLP (3, 82, 96–99). In parallel to ILC (20), intestinal epithelial cells regulate retinal dehydrogenase (RALDH) expression by CD103<sup>+</sup> DC that the cells need to metabolize retinoids. Specifically, epithelial cells express a critical cytosolic retinoid chaperone, the cellular retinol binding protein II, which is required for *in vivo* imprinting of gut DC by lumenal retinoids (99, 100). Supporting this notion, the *in vitro* coculture of bone marrow- or spleen-derived DC with epithelial cells results in the up-regulation of CD103 and RALDH, together with TGF-β imprinted homing potential on T cells (101–103). These data establish the potential of intestinal epithelial cells to educate intestinal DC, although further *in vivo* studies and higher resolution, with respect to cell subsets, are required to better elucidate the underlying mechanisms.

# **DC Communication with Intestinal T Cells**

Intestinal CD103<sup>+</sup> DC, found in lamina propria, Peyer's Patches, and the MsnLNs program T cells to express the gut-homing factors CCR9 and α4β7 integrin (101, 104, 105). Concomitantly, DC can also induce the development of FoxP-3<sup>+</sup> and IL-10 producing Treg cells (106) and prime Th17 cells (17, 107, 108). The majority of these DC-governed priming events require TGF-β signaling and RA, which are generated in the DC by enzymatic conversion of all-trans-retinal, a derivative of vitamin A, using RALDH2 (101, 109, 110). Indeed, RA has emerged as the critical conditioning factor for intestinal DC, as vitamin A is crucial for the activity of the enzyme RALDH in DC. Without RALDH, the ability of DC to imprint T cells is hampered, and restored only after vitamin A administration (111). The balance between RA and TGF-β levels seems to determine the fate of Treg cells primed by DC, as presence of both RA and TGF-β favor the development of FoxP-3<sup>+</sup> cells, while RA induces the generation of IL-10 producing T cells (106).

Other enzymes that influence the outcome of T cell priming are indoleamine 2,3 dioxygenase (IDO) and TSLP. IDO is expressed also by DC in other tissues and was shown to inhibit the development of effector T cells and promote Treg cell generation (112, 113). TSLP is, as mentioned above, secreted by epithelial cells, but also by the intestinal DC, themselves. In the presence

**FIGURE 1 | CyTOF analysis of CD45<sup>+</sup> cells from murine colon**. Cells were isolated from colon of 6–9 weeks old WT female C57Bl/6 mice and stained with a panel of 26 cell surface markers. The results were gated for live, single, CD45<sup>+</sup> cells. Bh-SNE analysis and clustering were performed by Accense (http://www.cellaccense.com/) and the results were processed by GIMP. Colors indicate high levels of the following markers:

green – TCRβ, CD3e (T cells), Orange – B220 (B cells), light blue – Ly6G (granulocytes), pink –Ly6C (monocytes), purple – CD64, F4/80 (macrophages), blue – clustered by Accence, different DC populations, gray – non-identified or non-specific cells. Red populations in zoom-in black squares indicate high levels of the marker written. Representative of at least four separate, independent experiments.

of TSLP, Th17 responses are restricted due to a reduced ability to produce IL-17, and Treg cell differentiation is up-regulated (107). The ability of the intestinal DC compartment to generate Th17 cells seems to be associated with CD103<sup>+</sup> CD11b<sup>+</sup> DC, as the frequency of Th17 cells is reduced in mice lacking these DC due to either IRF-4 or Notch-deficiency (10, 12, 14), or as a result of conditional ablation of this DC subset (13). Interestingly though, a recent study showed that also another subpopulation of DC, i.e. CCR2<sup>+</sup> CD103*<sup>−</sup>* CD11b<sup>+</sup> DC can induce IL-17a production in CD4<sup>+</sup> T cells and effectively prime Th17 cells, probably via IL-12/IL-23p40 secretion (17).

# **Intestinal DC, Inflammation, and Immune Response**

In steady state, intestinal DC are probably mainly tolerogenic. Under inflammatory conditions, however, they can become highly effective T cell activators (114). Induction of experimental colitis results in the accumulation of CD103<sup>+</sup> DC with an inflammatory profile in the MsnLNs (114). These DC express less RALDH and TGF-β and instead of promoting Treg cell formation, now induce Th1 inflammatory responses (114). While Th17 polarization might be carried out by CD103<sup>+</sup> CD11b<sup>+</sup> DC (12), differentiation of CD8<sup>+</sup> effector T cells under inflammatory conditions seems to be dependent on CD103<sup>+</sup> CD11b*<sup>−</sup>* CD8α + DC that migrated into the lymph (88).

Flagellin stimulation causes TLR-5<sup>+</sup> CD103<sup>+</sup> DC in the small intestine to promote differentiation of Th17 cells and secrete IL-23, which in turn induces IL-22 production by ILC3 and subsequent epithelial up-regulation of antibacterial peptides (115).

In summary, DC are major players in maintaining homeostasis in the intestine. While tolerogenic at steady state, under inflammatory conditions they tip the scales and activate the immune system. They can migrate between different compartments of the intestine – from the lamina propria to the epithelium and into the MsnLNs – and execute different immune responses in each tissue. Further research regarding the location of DC, their functions and characteristics should shed new light on the role of these cells in the intestine.

# **Concluding Remarks and a Glimpse to the Future**

In summary, macrophages and DC critically contribute to intestinal homeostasis and immune defense. Both cellular

# **References**


compartments have been subdivided into discrete subpopulations, which though currently mainly phenotypically defined, in some cases have been assigned distinct activities. The challenge ahead is to better define precise roles of these subsets both in health and under inflammatory conditions, first in the mouse but then also in the human. This task is complicated by the fact that many of the used markers used to distinguish between subpopulations of DC and macrophages are shared by the two types of mononuclear phagocytes. Moreover, under inflammatory conditions monocyte-derived cells further blur the picture. Collectively, this highlights the need to define cells by multiple parameters, including both surface and intracellular markers. Single cell transcriptome analysis is likely to help with this task (116, 117). However, classic flow cytometry analysis using fluorescent dye-coupled antibodies allows only a very limited simultaneous panel of markers due to the few dyes available and the spectral overlap of their emission. This problem might, in the near future, be solved by spectral cytometry systems that use ultrafast optical spectroscopy combined with flow cytometry to differentiate between the emission curves of different fluorophores, thus enabling the use of dozens of antibodies in one sample (118). Moreover, a new cell analyzer has been introduced, which uses mass cytometry instead of flow cytometry and is termed cytometry by Time-Of-Flight, or CyTOF (119). Instead of conjugations to fluorophores, this machine uses conjugations to heavy metal isotopes. Such metals do not exist naturally in the cells, so background is insignificant. The stained cells are injected into the CyTOF and are evaporated in a plasma chamber. The metals are ionized, hit the TOF detector, and their mass is measured, allowing the machine to determine the expression levels of the markers on each cell. This multiple-parameter approach enables to explore entire immune cell populations and subpopulations from the same tissue. As exemplified in **Figure 1**, such global analysis methods might well hold the key for the better definition and understanding of the cellular make-up of the intestine. No doubt, that with the recent development in the fields of cell cytometry and RNA sequencing, more pieces of this complex puzzle of the characteristics and roles of mononuclear phagocytes in the gut will be detected and put in place.

# **Acknowledgments**

This work was supported by the European Research Council (340345).


XCR1 and SIRPalpha. *Front Immunol* (2014) **5**:326. doi:10.3389/fimmu.2014. 00326


and green fluorescent protein reporter gene insertion. *Mol Cell Biol* (2000) **20**:4106–14. doi:10.1128/MCB.20.11.4106-4114.2000


identity and diversity of mouse tissue macrophages. *Nat Immunol* (2012) **13**:1118–28. doi:10.1038/ni.2419


dendritic cell conditioning. *Mucosal Immunol* (2009) **2**:340–50. doi:10.1038/ mi.2009.13


**Conflict of Interest Statement:** The Guest Associate Editor Martin Guilliams declares that, despite having collaborated on a paper with Steffen Jung in May 2013, the review process was handled objectively. 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 Gross, Salame and Jung. 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.*

# **Microglia versus myeloid cell nomenclature during brain inflammation**

*Melanie Greter\*, Iva Lelios and Andrew Lewis Croxford*

*Institute of Experimental Immunology, University of Zurich, Zurich, Switzerland*

As immune sentinels of the central nervous system (CNS), microglia not only respond rapidly to pathological conditions but also contribute to homeostasis in the healthy brain. In contrast to other populations of the myeloid lineage, adult microglia derive from primitive myeloid precursors that arise in the yolk sac early during embryonic development, after which they self-maintain locally and independently of blood-borne myeloid precursors. Under neuro-inflammatory conditions such as experimental autoimmune encephalomyelitis, circulating monocytes invade the CNS parenchyma where they further differentiate into macrophages or inflammatory dendritic cells. Often it is difficult to delineate resident microglia from infiltrating myeloid cells using currently known markers. Here, we will discuss the current means to reliably distinguish between these populations, and which recent advances have helped to make clear definitions between phenotypically similar, yet functionally diverse myeloid cell types.

#### *Edited by:*

*Martin Guilliams, Ghent University – VIB, Belgium*

#### *Reviewed by:*

*Steffen Jung, Weizmann Institute of Science, Israel Marco Prinz, University of Freiburg, Germany*

#### *\*Correspondence:*

*Melanie Greter, Institute of Experimental Immunology, University of Zurich, Winterthurerstrasse 190, Zurich 8057, Switzerland greter@immunology.uzh.ch*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 26 February 2015 Accepted: 07 May 2015 Published: 26 May 2015*

#### *Citation:*

*Greter M, Lelios I and Croxford AL (2015) Microglia versus myeloid cell nomenclature during brain inflammation. Front. Immunol. 6:249. doi: 10.3389/fimmu.2015.00249* **Keywords: microglia, macrophage, monocyte, dendritic cell, CNS inflammation**

# **Introduction**

Most tissues are populated by incredibly diverse and abundant myeloid cells. By contrast, the central nervous system (CNS) harbors comparatively few myeloid cell subsets. This is likely due to the immune privilege and relative isolation enjoyed by the CNS compared to other non-lymphoid tissues such as the gut or the lung, which are continually confronted with foreign entities. In the steady state, the CNS houses several populations of myeloid cells with distinct localizations including perivascular, choroid plexus, and meningeal macrophages/dendritic cells (DCs) and microglia, which are the most abundant (1). Microglia are considered the resident macrophages of the brain given that they are the only myeloid cells present in the CNS parenchyma. Microglia perform both homeostatic and immune-related functions and constitute about 5–20% of all cells in the CNS (2). They use their "ramified" morphology to act as immune sentinels, extending specialized processes, and sampling the local environment for foreign bodies (3, 4). Numerous recent reports have unmasked additional functions for microglia other than being simply the brain's intrinsic immune system. For example, microglia are also critical for neuronal development, adult neurogenesis, learning-dependent synapse formation, and brain homeostasis (5–7). Microglia are classified as tissue resident macrophages but are clearly ontogenically distinct from other members of the mononuclear phagocyte system (MPS), which includes DCs, monocytes, and macrophages. Microglia originate from primitive macrophages that derive from erythro-myeloid precursors in the yolk sac (8–10). These primitive yolk sac macrophages colonize the developing brain in mice as early as embryonic day 9.5 (8). Throughout adult life microglia remain of embryonic origin in the healthy CNS and maintain themselves locally without any detectable contribution from circulating myeloid progenitors including monocytes. Yolk sac macrophages and microglia precursors in the developing brain express high levels of the fractalkine receptor (CX3CR1) and are positive for the integrin alpha M (Itgam, also know as CD11b; macrophage-1 antigen, Mac-1), F4/80, and the macrophage-colony stimulating factor receptor 1 (Csf-1R, CD115) similar to adult microglia as described below (8). Compared to adult microglia, however, microglia precursors are CD45hi. The development of microglia is dependent on Csf-1R (CD115), the transcription factors PU.1 and Irf8 but is independent of Myb, which is crucial for the development of hematopoietic stem cells (HSCs) (8, 9, 11, 12). In contrast to microglia, recent adoptive transfer and fate-mapping studies revealed that other macrophage populations are either embryonically derived from definitive hematopoiesis (e.g., alveolar or heart macrophages) or are constantly replaced by circulating monocytes (e.g., dermal or gut macrophages) (10, 13–18). Aside from the unique ontogeny of microglia within the MPS, a clear classification of microglia compared to other tissue macrophages in terms of phenotype and function has been difficult. Only recently, transcriptome and epigenetic analysis identified genes uniquely expressed and regulated by microglia but not by other macrophage populations (19–24). These studies might be useful to classify and distinguish microglia from other myeloid cells.

# **Microglia Markers in Steady State**

In steady state conditions, microglia express surface markers typically present on many other tissue macrophages and/or monocytes such as CD11b, F4/80, Fc-gamma receptor 1 (CD64), and CD115 (Csf-1R), ionized calcium-binding adapter molecule 1 (Iba-1) and proto-oncogene tyrosine-protein kinase MER (MerTK) (**Figure 1**) (19). In contrast to microglia, which are γ-irradiation resistant, perivascular myeloid cells are replaced by bone marrow (BM) derived precursors after total-body irradiation and BM transplantation (25–28). However, the exact ontogeny of (non-microglia) myeloid cells associated with the CNS and whether they are also able to maintain themselves locally is, to date, not known (29). These perivascular cells are equipped to present antigen (varying levels of MHCII and CD11c). Whether they represent a homogeneous distinct population or a heterogeneous population of macrophages and/or DCs is not entirely resolved. In the past, a cell expressing F4/80 was deemed to be a macrophage, whereas a cell expressing CD11c was considered a DC. It is clear now that subsets of DCs can also express F4/80 and certain macrophage populations express CD11c. Upon Flt3L treatment, a CD11c+MHCII<sup>+</sup> population in the meninges and choroid plexus expanded, which is indicative of the DC lineage whose development is dependent on Flt3L signaling (27, 28). In addition, a limited number of CD11c<sup>+</sup> myeloid cells were also described to be in a juxtavascular location in the CNS parenchyma (30). These cells might, however, represent *bona fide* microglia expressing CD11c in certain regions of the brain. Further studies are required to dissect the ontogeny and characterize these elusive myeloid cells associated with the CNS in the steady state.

As common to many other macrophage populations including microglia, most of the CNS-associated myeloid cells also express CD11b, CD115, Iba-1, and F4/80 (31). Therefore, apart from their location, the only available means to unequivocally distinguish microglia from other CNS-resident myeloid populations (CNS-associated macrophages/DCs) and circulating monocytes

is by the reduced expression of the common leukocyte antigen CD45, which is readily detectable by flow cytometry. Adult microglia, unlike most other tissue macrophages, constitutively express high levels of the fractalkine receptor CX3CR1 (32). A major advance in the microglia field has been the generation of *Cx3Cr1creER* mice (32–34). This tamoxifen-inducible Crerecombinase under the CX3CR1 promoter allows for microgliaspecific gene targeting. Despite the fractalkine receptor being expressed by monocytes and myeloid precursors in the BM, microglia remain a self-contained population in the CNS and therefore remain targeted long after ceasing of tamoxifen administration, returning the short-lived, circulating myeloid cells to their wild-type origin.

Only recently, gene expression studies have identified surface markers and transcription factors specifically expressed by steady state microglia but not by other macrophage populations or monocytes. These include, for example, sialic acidbinding immunoglobulin-type lectin H (Siglec-H), Fc receptorlike S (Fcrls), and purinergic receptor P2Y G-protein coupled 12 (P2ry12) (20, 21). Furthermore, microglia seem to be the only hematopoietic cell population that specifically expresses Sal-like 1 (Sall1), a transcription factor that plays a crucial role in kidney development (35). While previous studies have reported expression of Sall1 only by stromal cells, in the adult CNS this factor is expressed exclusively by microglia. These aforementioned gene expression studies have compared the transcriptome of microglia to either macrophages derived from the spleen, the lung, the peritoneum, or to monocytes. Whether these microglia core signature markers are also expressed by CNS-associated macrophages/DCs remains to be shown.

# **Microglia Markers in Inflammation**

In contrast to the healthy brain, during neuro-inflammation the picture becomes far more complicated. A hallmark of microglia is their rapid activation after a CNS insult, resulting in their migration toward injury, proliferation, and their change in morphology. They take on a more "amoeboid" shape with shorter and thicker processes, display increased immunoreactivity for Iba-1 and upregulate CD45. Experimental autoimmune encephalomyelitis (EAE), which is a mouse model for multiple sclerosis (MS), is characterized by infiltration of T cells, monocytes, and neutrophils. Monocytes and their progeny [macrophages/monocytederived DCs (moDCs)] are undoubtedly the prevailing cell type in the lesions (see below). However, activated microglia are also clearly detected in the vicinity of the inflammatory lesions. The downregulation of Ly6C by monocytes upon their differentiation adds complexity to the separation of these two distinct cell types based on the commonly used cell surface markers. Additionally, molecules involved in antigen presentation and T cell stimulation, which are barely detectable in steady state microglia, are expressed to some level by microglia already at disease onset and retain expression throughout disease progression. These markers include major histocompatibility complex class II (MHCII), CD11c (also known as integrin alpha X, Itgax), CD80 (B7- 1), CD86 (B7-2), and CD40 (36–38). Under these conditions, it is considerably more difficult to distinguish these activated microglia from inflammatory monocyte-derived cells. Similar changes in microglia surface markers have been observed in mouse models of neurodegenerative disease such as Alzheimer's disease (AD), Parkinson's disease, and amyotrophic lateral sclerosis (ALS) (20, 31, 39).

A recent study, however, has used differentially expressed chemokine receptors on the surface of microglia and monocytes to distinguish those two myeloid populations and study their function during neuro-inflammation. Microglia were identified by their high expression of CX3CR1 whereas infiltrating monocytes, which subsequently differentiate into macrophages/DCs, were defined by their high expression of C–C chemokine receptor type 2 (CCR2), a receptor mediating monocyte recruitment to sites of inflammation. Gene expression profiles from macrophages versus "embryonically derived" microglia at different stages of EAE show that despite some similarities between these inflammatory cell types, microglia exhibit a distinct molecular signature (40). This genetic distinction reflects a different function of resident microglia and infiltrating monocytes under pathological circumstances. While monocyte-derived macrophages seemed to be the effector cell type causing CNS damage, microglia might have a regulatory function and could play a role in tissue repair and homeostasis (40). Another report also showed that monocytes recruited to the CNS in EAE do not acquire microglia-signature genes (21). These studies will unquestionably help attributing unique functions to microglia and CNS-invading myeloid cells in different pathological conditions in the brain.

Whether microglia-specific surface markers and transcription factors alter their expression between steady state and inflammation remains unclear. The microglia-specific ATP receptor P2ry12 was downregulated under inflammatory conditions such as lipopolysaccharide (LPS) systemic injection or in SOD1 mousemodel of ALS (41, 42). On the other hand, P2ry12 and Fcrls continue to be expressed by microglia in EAE but are not expressed by infiltrating monocytes (21). Therefore, further studies need to be undertaken in order to better characterize these "new" phenotypic microglia markers under pathological conditions.

# **Monocyte-Derived Microglia**

Even though microglia homeostasis is maintained through local self-renewal, under certain conditions circulating precursors can give rise to microglia-like cells. For example, early studies using BM-chimeras showed that up to 10–20% of microglia were reconstituted by donor-derived cells 6-12 months after total-body irradiation and BM transplantation (43). However, this engraftment of BM-derived microglia can only be seen upon blood–brainbarrier (BBB) disruption (e.g. irradiation) and becomes minimal in models where the BBB is unperturbed (e.g. protection of the head during irradiation and parabiosis) (44, 45). This clearly indicates that under steady state conditions, monocytes or BM-derived myeloid progenitors do not infiltrate the CNS parenchyma and thus do not give rise to adult microglia. Similarly, as described above in experimental models of neuroinflammation, monocytes infiltrate the CNS and differentiate into effector cells resembling phenotypically activated microglia. Despite these similarities, monocyte-derived cells do not persist in the CNS after inflammation has been resolved and thus do not contribute to long-lived microglia (46).

Finally, local administration of ganciclovir to transgenic mice expressing the thymidine kinase of herpes simplex virus under the CD11b promoter (CD11b-HSVTK) leads to a rapid depletion of microglia (47, 48). Subsequently, BM-derived cells enter the CNS and differentiate into long-lived microglia-like cells. Notably, while these cells form a network filling the niche for embryonically derived microglia, they do not obtain a complete microglia phenotype. Monocyte-derived "microglia" in this model show a less-ramified morphology and a higher expression of CD45 compared to yolk sac-derived microglia resembling more activated microglia (48). It has not yet been investigated whether these monocyte-derived "microglia" functionally resemble embryonically derived microglia or whether they acquire microglia-signature genes (Siglec-H, Fcrls, P2ry12) as described above.

Recent studies used a Csf-1R inhibitor to deplete microglia. Upon treatment stop, microglia were repopulated within 1 week (49, 50). These studies showed that new "microglia" were derived from CNS-resident nestin-positive precursors and resembled embryonically derived microglia in response to an inflammatory stimulus. Animals with newly repopulated "microglia" did not display any impairment in behavior, cognition, or motor function compared to control animals (50).

# **Monocytes and Monocyte-Derived Cells in CNS Inflammation**

Brain inflammation or "encephalitis" invariably results in a reshaping of the myeloid cell populations inhabiting the CNS. An inflammatory response brought on by either infection or autoimmune manifestations results in a rapid increase in bloodderived cellularity to this otherwise dormant site. Despite EAE being fully dependent on T helper cells (51), the vast majority of the inflammatory infiltrate seen in EAE is of myeloid derivation. Two types of monocytes exist including the classical monocytes (Ly6ChiCCR2+CX3CR1lo) and the non-classical monocytes (Ly6CloCCR2*−*CX3CR1hi). Here, we will only discuss Ly6Chi monocytes given that during neuro-inflammation, this is the subset recruited to the brain. Engraftment of phagocytes derived from circulating CCR2<sup>+</sup> monocytes has also been shown in an AD mouse model (39). Ly6Chi monocytes egress from the BM and cross the BBB in a CCR2-dependent manner (52, 53) followed by their differentiation into macrophages/moDCs and upregulation of a set of cell surface markers (e.g., MHCII, CD11c) expressed on a wide variety of MPS members. Likewise, microglia progressively alter their phenotype to resemble more classically activated macrophages during CNS inflammation, infection, and neuronal or myelin damage (54).

Ly6C<sup>+</sup> monocytes were shown to migrate into the CNS prior to disease onset and precede the development of paralysis and subsequent clinical manifestations of EAE, when "DC-like" cells are found in abundance in the inflamed tissue (55, 56). This corresponds well with a previous report showing CD205<sup>+</sup> myeloid cells accumulating in the meninges, choroid plexus, and subpial space of the spinal cord and in perivascular cuffs in demyelinating lesions during acute disease (57). CD11b<sup>+</sup> DCs within the inflamed CNS were demonstrated to be critical for the propagation of EAE (27, 58, 59). Further phenotypical characterization would be required to demarcate their lineage whether they resemble moDCs or are more similar to classical DCs. Indeed, monocyte-derived antigen presenting cells (APCs) have been shown to be required for optimal priming of T cells in models of infection (60). Current evidence suggests that phenotypically similar macrophages in the CNS can not only contribute to the generation of inflammatory lesions and perform a pathogenic role in the demyelination process but also contribute to regenerative repair mechanisms to resolve inflammation (61, 62). These studies emphasize that distinct functions are attributed to the different subsets of myeloid cells in the course of a CNS inflammation. As such, a complete understanding of cell types based on surface phenotype alone would be of great benefit both in preclinical models of CNS inflammation and also in human patients.

Even with the knowledge we now possess on myeloid cell diversity, it is still commonplace in the literature using animal models of CNS inflammation to use a simplistic CD45hiCD11bhi gating strategy to separate CNS infiltrating, blood-derived myeloid cells from CNS-resident, embryonically derived microglia (CD45low) (63). Efforts to sort cells using a broad CD45hiCD11bhi surface phenotype from within the inflamed CNS will inevitably result in analysis of multiple cell types, lacking any of the desired specificity. Indeed, without the removal of Ly6Ghi cells during sorting, a mixed population is inevitable and expression profiles subsequently attributed to moDCs are either confused with, or heavily influenced by, an abundant neutrophil contamination. Even if the effort is taken to remove neutrophils, moDCs at various stages of development will be incorporated. This distinction is increasingly important given that both neutrophils and moDCs have been shown to mediate BBB permeability and demyelination, and that different pathogenic mechanisms are likely active in the two populations during the same inflammation (40, 64).

We know that at least four clearly distinct cell types share this rather non-specific CD45hiCD11bhi surface phenotype in an inflamed CNS, namely neutrophils (CD11b+Ly6G+), monocytes (CD11b+Ly6ChiCX3CR1low), and their progeny such as moDCs and/or activated macrophages (**Figure 1**). The latter two cell types represent most likely the same population with just different names assigned by different studies. Ly6Chi monocytes that have migrated into the CNS can further be subdivided into numerous differentiation stages characterized by the upregulation of CD11c and MHCII, with the concomitant downregulation of Ly6C and CCR2. Upon differentiation and upregulation of MHCII, monocytes are then called moDCs/activated macrophages. Thus, moDCs in the CNS are characterized by the expression of CD11b+F4/80+MHCII+CD11cintLy6C+/*−*. These moDCs/activated macrophages also express CD64 and likely also MerTK, which both are universally expressed by tissue macrophages including microglia (19). Interestingly, it has been shown that monocytes recruited to the CNS during EAE do not express the newly identified microglia markers Fcrls and P2ry12 highlighting again the diverse ontogeny of these cell types and suggesting that the microglia-signature genes are indeed specific to microglia rather than location (CNS) specific (21). The Fc*ε*RIα (MAR-1) has been suggested to represent a moDC marker. Whether moDCs in the inflamed CNS express MAR-1 has so far not been analyzed (65). Perhaps a more functional distinction should be drawn on the level of relevance for the inflammatory process to persist. CNS-infiltrating myeloid cells with DC-like morphology express MHCII, CD40, and CD86, all of which have critical roles in multiple inflammatory models (66). The CD86/CD28 interaction between T cells and APCs is of critical importance for T cell activation. Furthermore, the CD40/CD40L interaction induces a maturation pathway within the inflamed CNS, resulting in further costimulatory capabilities and proinflammatory cytokine expression (67, 68). The levels of CD40 on monocyte-derived cells in the inflamed CNS are variable but generally not as high as on classical DCs.

After activation, inflammatory macrophages can not only express a wide range of inflammatory cytokines but also oxygenbased chemically reactive molecules involved in host defense. The route an activated macrophage takes depends largely on the T cell and/or NK cell-derived cytokines present during their activation. For example, activation in the presence of LPS and IFN-γ leads to a "classical" activation (often called "M1"), resulting in secretion of high levels of TNF-α, iNOS, IL-1, IL-6, and IL-12. Conversely, activation of the same cells in the presence of IL-4 and IL-10 will result in rapid upregulation of IL-10, production of Arginase-1, and upregulation of the mannose receptor CD206, generating a macrophage capable of suppressing T cell activity (called "M2") (69). This intracellular divergence in phenotype illustrates that an apparently similar cell expressing F4/80, CD64, CD11b, MHCII, and CD11c on its surface may, in fact, differ greatly in its function. Indeed, markers identifying both M1 and

## **References**


M2 macrophage populations have been shown synergistically in CNS biopsies obtained from MS patients. CD40, CD64, CD86, and CD32, mannose receptor and CD163 were co-expressed in the large majority of foamy macrophages found in lesional CNS (70). Therefore, surface characterization of inflammatory macrophages would appear insufficient and may mask different macrophage populations in direct opposition to each other, depending on the type of inflammation taking place. Generally in the steady state, tissue macrophages display an "M2-like" phenotype and are critical for tissue homeostasis. Interestingly, in a model of spinal cord injury, it was shown that M2 macrophages (CD11b+F4/80+CX3CR1hiLy6Clo) are beneficial and promote recovery (62).

# **Conclusion**

Under steady state conditions, site specific and phenotypic characteristics exist to distinguish between microglia and other CNSassociated macrophages. As with almost all innate and adaptive immune cell types, consensus with respect to nomenclature in CNS-resident versus CNS-infiltrating myeloid cells has not been effectively reached under inflammatory conditions. The advent of microarray technology and next generation sequencing will serve to provide more useful ways to distinguish between these two apparently similar, yet ever more functionally diverse cell types. An ever-increasing variety of previously unappreciated, and non-immune homeostatic functions performed by macrophages are now beginning to emerge, making a more detailed separation of these cell types highly desirable (71). Ultimately, better characterization and dissection of the various myeloid cells in an inflamed brain will help deciphering the specialized functions of the different members of the MPS in pathological conditions.


through distinct mechanisms at steady state and during inflammation. *Immunity* (2014) **40**:91–104. doi:10.1016/j.immuni.2013.11.019


**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 Greter, Lelios and Croxford. 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.*

# **Revisiting mouse peritoneal macrophages: heterogeneity, development, and function**

*Alexandra dos Anjos Cassado<sup>1</sup> \*, Maria Regina D'Império Lima<sup>1</sup> and Karina Ramalho Bortoluci 2,3*

*<sup>1</sup> Departamento de Imunologia, Instituto de Ciências Biomédicas, Universidade de São Paulo, São Paulo, Brazil, <sup>2</sup> Centro de Terapia Celular e Molecular (CTC-Mol), Universidade Federal de São Paulo, São Paulo, Brazil, <sup>3</sup> Departamento de Ciências Biológicas, Campus Diadema, Universidade Federal de São Paulo, São Paulo, Brazil*

#### *Edited by:*

*Florent Ginhoux, Singapore Immunology Network, Singapore*

#### *Reviewed by:*

*Xinjian Chen, University of Utah, USA Thomas Marichal, University of Liège, Belgium*

#### *\*Correspondence:*

*Alexandra dos Anjos Cassado, Departamento de Imunologia, Instituto de Ciências Biomédicas, Universidade de São Paulo, Av. Professor Lineu Prestes, 1730, Cidade Universitária, São Paulo, SP 05508-900, Brazil alecassado@hotmail.com*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 27 February 2015 Accepted: 26 April 2015 Published: 19 May 2015*

#### *Citation:*

*Cassado AA, D'Império Lima MR and Bortoluci KR (2015) Revisiting mouse peritoneal macrophages: heterogeneity, development, and function. Front. Immunol. 6:225. doi: 10.3389/fimmu.2015.00225* Tissue macrophages play a crucial role in the maintenance of tissue homeostasis and also contribute to inflammatory and reparatory responses during pathogenic infection and tissue injury. The high heterogeneity of these macrophages is consistent with their adaptation to distinct tissue environments and specialization to develop niche-specific functions. Although peritoneal macrophages are one of the best-studied macrophage populations, recently it was demonstrated the co-existence of two subsets in mouse peritoneal cavity (PerC), which exhibit distinct phenotypes, functions, and origins. These macrophage subsets have been classified, according to their morphology, as large peritoneal macrophages (LPMs) and small peritoneal macrophages (SPMs). LPMs, the most abundant subset under steady state conditions, express high levels of F4/80 and low levels of class II molecules of the major histocompatibility complex (MHC). LPMs appear to be originated from embryogenic precursors, and their maintenance in PerC is regulated by expression of specific transcription factors and tissue-derived signals. Conversely, SPMs, a minor subset in unstimulated PerC, have a F4/80lowMHC-IIhigh phenotype and are generated from bone-marrow-derived myeloid precursors. In response to infectious or inflammatory stimuli, the cellular composition of PerC is dramatically altered, where LPMs disappear and SPMs become the prevalent population together with their precursor, the inflammatory monocyte. SPMs appear to be the major source of inflammatory mediators in PerC during infection, whereas LPMs contribute for gut-associated lymphoid tissueindependent and retinoic acid-dependent IgA production by peritoneal B-1 cells. In the previous years, considerable efforts have been made to broaden our understanding of LPM and SPM origin, transcriptional regulation, and functional profile. This review addresses these issues, focusing on the impact of tissue-derived signals and external stimulation in the complex dynamics of peritoneal macrophage populations.

#### **Keywords: peritoneal macrophages, peritoneal cavity, LPM, SPM, origin**

# **Introduction**

Macrophages are resident cells found in almost all tissues of the body, where they assume specific phenotypes and develop distinct functions. Tissue macrophages are considered as immune sentinels because of their strategic localization and their ability to initiate and modulate immune responses Cassado et al. Peritoneal macrophage heterogeneity

during pathogenic infection or tissue injury and to contribute to the maintenance of tissue homeostasis (1–3). Macrophages were first identified in the late 19th century by Élie Metchnikoff (1845–1916) and designated as large phagocytes (4, 5). Based on their phagocytic activity, macrophages were first classified as cells from the reticuloendothelial system, which also comprised endothelial cells, fibroblasts, spleen and lymphoid reticular cells, Kupffer cells, splenocytes, and monocytes (6). However, because endocytosis performed by endothelial cells is a process that is distinct from phagocytosis, by the late 1960s a new classification system for mononuclear phagocytic cells as cells from "mononuclear phagocytic system" (MPS) was proposed (7). The MPS was defined as a group of phagocytic cells sharing morphological and functional similarities, including pro-monocytes, monocytes, macrophages, dendritic cells (DCs), and their bone marrow (BM) progenitors (7–12). Although the phagocytic cells play similar roles in orchestrating the immune response and maintaining tissue homeostasis (11), they represent cell populations that are extremely heterogeneous (13), and the general classification of mononuclear cells in a unique system is currently under intense discussion (12, 14). In this context, Guilliams et al. suggested a classification of MPS cells based primarily on their ontogeny and secondary on their location, function, and phenotype, promoting a better classification under both steady state and inflammatory conditions (14).

In the last few years, a complex scenario to describe macrophage origins has been developed (15–19), replacing the simplistic view of myeloid precursors giving rise to blood monocytes that, in turn, originate tissue macrophages (20– 22). For example, resident macrophages from brain, lung, liver, peritoneum, and spleen are not differentiated from monocytes; instead, they are derived from an embryonic precursor and maintained by self-renewal (23–27). In addition to resident macrophages, infiltrating monocytes are also found in injured tissues, where they can differentiate into inflammatory macrophages or TNF-α- and inducible nitric oxide synthase (iNOS)-producing (Tip)-DCs (28). Currently, it is accepted that inflammatory macrophages and tissue-resident macrophages comprise developmentally and functionally distinct populations (3, 14, 17, 18, 29).

Under steady state conditions, some tissues and serous cavities, including lung, spleen, and the peritoneal cavity (PerC), present distinct resident macrophage subpopulations. In the spleen, at least three macrophage subsets are found: red pulp, metalophilic, and marginal zone macrophages (30). In the PerC, two peritoneal macrophage subsets have been described: large peritoneal macrophage (LPM) and small peritoneal macrophage (SPM) (31). Mouse peritoneal macrophages are among the best-studied macrophage populations in terms of cell biology, development, and inflammatory responses (24, 31–42). Peritoneal macrophages play key roles in the control of infections and inflammatory pathologies (43, 44), as well as in the maintenance of immune response robustness (40). Therefore, this review will discuss recent advances in our understanding of peritoneal macrophage subsets characterization, origin and functions, and the accurate experimental approaches to analyze them.

# **Identification of Peritoneal Macrophages**

Cohn and collaborators introduced the study of peritoneal macrophages (45–48). Indeed, a representative portion of the current knowledge regarding macrophage biology, such as their function, specialization, and development stems from studies performed using peritoneal macrophages as a cellular source. However, the existence of two resident macrophage subsets present in the PerC was described recently (31). These macrophage subsets were designated LPM and SPM according to their size. LPMs and SPMs were initially identified based on their differential expression of F4/80 and CD11b, where LPMs express high levels of F4/80 and CD11b while SPMs show F4/80lowCD11blow phenotype (**Table 1**). CD11b is an integrin that, together with CD18, forms the CR3 heterodimer (13, 30, 49), but is not exclusively expressed on macrophages and is found on several others cell types, including polymorphonuclear cells (50, 51), DCs (52), and at low levels on B lymphocytes (53, 54). F4/80, a 160 kD glycoprotein from the epidermal growth factor (EGF)-transmembrane 7 (TM7) family, is expressed by macrophages in several organs, such as the kidney (55), BM (56), epithelium (57), lung (58, 59), lymphoid organs (60), and among others (61, 62), and it is not found on fibroblasts, polymorphonuclear cells, and lymphocytes (63). However, peritoneal eosinophils show low levels of F4/80 (31) and some macrophage subpopulations exhibit low levels or do not express F4/80, such as white pulp and marginal zone splenic macrophages (30). Therefore, F4/80 expression levels distinguish macrophage subpopulations, including those residing in the same tissue, such as subsets found in the spleen and PerC (30, 31, 35). In this sense, the great majority (approximately 90%) of F4/80+CD11b<sup>+</sup> cells present in the PerC from several mouse strains, including BALB/c, C57BL/6, 129/S6, FVB/N, SJL/J, and RAG*−*/*−*, express high levels of these molecules and correspond to the LPM subset, whereas the minor SPM subset expresses low levels of these markers (31).

An accurate evaluation of SPMs and LPMs by flow cytometry and optical microscopy revealed that in addition to the differential expression of CD11b and F4/80, SPMs and LPMs display unique morphologies and phenotypes. LPMs assume the

#### **TABLE 1 | Phenotypic profile of SPMs and LPMs**.


classical morphology described for macrophages after adherence, exhibiting prominent vacuolization and abundant cytoplasm, whereas SPMs display a polarized morphology in culture, presenting dendrites similar to DCs (35). Moreover, the analysis of a complex panel of cell surface molecules (**Table 1**) demonstrated that SPMs express higher levels of MHC-II (IA<sup>b</sup> ), dectin-1, and DC-sign endocytic receptors than LPMs. Moreover, half of SPM subset expresses high levels CD62L (31, 35, 36). Conversely, LPMs express higher levels of toll like receptor (TLR)-4 and co-stimulatory molecules in comparison to SPMs (31, 35, 36).

Given that PerC is a singular compartment where specialized immune cells reside and interact, including macrophages, B cells, DCs, eosinophils, mast cells, neutrophils, T cells, natural killer (NK), and invariant NKT cells (31, 32, 35, 36, 64), the identification of myeloid cells from PerC based on cell surface molecules is still a complex matter, particularly in terms of distinguishing macrophage subsets from DCs and inflammatory monocytes. The expression of 12/15-lipoxygenase (LOX), Tim4, and Ly6B has also been examined to discriminate heterogeneous macrophage subsets in PerC under steady state conditions and during peritonitis (24, 37, 38, 42). The high expression of 12/15-LOX and Tim4 was observed in peritoneal macrophages, which also express high levels of F4/80 and CD11b, correlating with the phenotype and frequencies observed for LPMs (24, 31, 37, 38, 42). Conversely, 12/15-LOXcells and SPM share the same CD11b+F4/80lowMHCIIhigh phenotype; however, 12/15- LOXcells express high levels of CD11c and co-stimulatory molecules, suggesting that 12/15-LOXcells and SPMs are, at least in part, distinct populations (31, 35, 37). Despite similarities in cell morphology and MHC-II expression presented by SPMs and DCs, the possibility that SPMs may be part of the peritoneal DC pool is excluded by the smaller size, the distinct and lack of the CD11b and F4/80 expression presented by DCs and, primarily, by the lower expression of CD11c (HL3 or N418 clones of monoclonal anti-CD11c) on SPMs compared with LPMs or typical peritoneal DCs (31, 35).

Given the cell complexity present in PerC and the importance of the development of efficient strategies to correctly identify macrophage subsets as well as to avoid contamination by other cell populations and misinterpretation of peritoneal macrophage studies, our group has proposed a simple way to identify peritoneal macrophage subsets using a four-color flow cytometry staining panel. From doublet, CD19high and CD11chigh discarded selected cell populations; the analysis of F4/80<sup>+</sup> cells based on MHCII expression defines three distinct subpopulations, F4/80highIAb-neg, F4/80lowIAb-high, and F4/80lowIAb-neg, which correspond, respectively, to LPMs, SPMs, and granulocytes (35).

# **Origin and Development of LPM and SPM**

The theories that explain the origin of macrophages have been completely reformulated in the last few years. The differentiation process of monocytes, macrophages, and DCs that occurs in the BM starts with the earliest progenitor, the hematopoietic stem cell (HSC), and follows the common myeloid progenitor (CMP) and the granulocyte and macrophage progenitor (GMP) (16). The clonotypic BM-resident precursor differentiated from GMP, termed the macrophage-DC precursor (MDP), expresses high levels of the fractalkine receptor CX3CR1, c-kit, and CD115, and gives rise to circulating blood monocytes, some macrophage populations and a common DC precursor (CDP), but does not originate granulocytes (15, 65, 66). The recruitment of monocyte subsets under steady state or inflammatory and pathological conditions depends on particular chemokines and the expression of their counterpart's receptors. The Ly6C<sup>+</sup> monocyte subset migrates via a CCR2-dependent pathway, whereas Ly6Cappears to migrate in response to CX3CR1 signaling (67). Under steady state conditions, extravasated monocytes do not contribute to the pool of resident macrophages in many tissues (3, 15, 16). In inflammatory settings, the Ly6C<sup>+</sup> monocyte subset differentiates into inflammatory macrophages and monocyte-derived DCs, such as Tip-DCs (15, 16).

Recent accumulating evidence supports the prenatal origin of tissue-resident macrophages and the idea that they are maintained locally by self-renewal throughout adult life, both in the steady state and after cell turnover, which is predominantly independent of hematopoiesis (17, 18, 23–27, 29, 68, 69). Microglia, Langerhans cells, Kupffer cells, red pulp splenic macrophages, lung, and peritoneal macrophages are originated from embryogenic precursor and proliferative cells maintained by self-renewal (23–27, 69– 71). Fetal-liver monocytes or primitive macrophages found in the yolk sac, an extraembryonic tissue, have been related with the origin of tissue-resident macrophages. In this context, recent date using yolk sac macrophages depletion and fate-mapping models demonstrated that yolk sac macrophages, which are generated from early erythro-myeloid progenitors (EMPs), are important for development of macrophages in mid-gestation; however in adulthood, only microglia is maintained by these embryogenic precursor (69). In contrast, fetal monocytes that are derived from late EMPs give rise to tissue-resident macrophages from liver, lung, skin, kidney and spleen (69). The exception to the origin of resident macrophages is intestinal macrophages, which are continuously repopulated by circulating monocytes (72).

Understanding the dynamics of maintenance and recruitment of peritoneal macrophages is of particular interest since these cells are involved in physiological as well as pathological processes, such as peritonitis, tumors, and pancreatitis (40, 43, 44). Early studies demonstrated that peritoneal macrophages are maintained in PerC through self-renewal in the steady state or under inflammatory conditions (73–76). The omentum, a fat tissue that connects the abdominal organs, is also involved in peritoneal macrophage development through the proliferative capacities of omental macrophages (75, 76). The combination of these early observations, which were acquired recently, with the technical advances to correctly identify the peritoneal macrophage subsets has permitted the ontogeny of the peritoneal macrophage subsets to be elucidated (24, 31, 36, 39, 40, 42).

Under steady state conditions, LPMs appear to be maintained by self-renewal and independent of hematopoiesis (26, 36), whereas SPMs are originated from circulating monocytes (31, 36, 40) (**Figure 1**). Dates from Schulz et al. suggest that, in general, F4/80 expression by tissue macrophages correlated with yolk sac (F4/80high) and not hematopoietic (F4/80low) progenitors (25). In the CX3CR1GFP/WT mice, Cain et al. (36) showed the presence of

hematopoietic progenitors (69). Local proliferation of LPMs ensures

homeostatic maintenance by self-renewal (36).

GFP<sup>+</sup> cells in DC and SPM pool, but not in the LPM population. Conversely, in the CX3CR1CreRosa26R-FGFP mice, which show the active and past expression of CX3CR1, the presence of GFP<sup>+</sup> cells was found within DC, SPM, and LPM populations. These data indicate that SPMs are short-lived cells, whereas LPMs have a more distant ontogenic relationship with a CX3CR1<sup>+</sup> progenitor, corroborating the idea that they originate from the yolk sac (36). However, in chimeric C57BL/6 mice reconstituted with C57BL/6- CD45.1 BM, around 80% of SPMs and more than 70% of LPMs are CD45.1-expressing cells, demonstrating that both peritoneal macrophage subsets differentiate from BM precursors after ablation of peritoneal macrophages induced by irradiation (36). Data from our group suggest that PerC recruited Ly6C<sup>+</sup> monocytes could give rise to SPMs during inflammatory conditions (31). Confirming that SPMs are generated via the differentiation of inflammatory monocytes recruited to PerC, reduced numbers of SPMs are found in the PerC of CCR2*−*/*<sup>−</sup>* mice (40).

The analysis of Ki67 and phosphorylated histone H3 (pHH3 at a discrete stage of mitosis) staining and the quantification of cell cycle and basal DNA content revealed that the number of proliferating F4/80highCD11bhigh cells decreases in 12-week-old mice compared with proliferation capacity of this population in newborn mice (15 days to 4 weeks) (24). After 12–16 weeks, the number of F4/80highCD11bhighcells in PerC is maintained under a low rate of proliferation, which suggests that the number of F4/80highCD11bhigh peritoneal cells increases during mouse development until PerC acquires sufficient homeostatic cell numbers (24). Indeed, BrdU-labeled LPM frequencies after a single BrdU pulse were 7 and 15-fold lower than those found in HSC and GMP, respectively. Moreover, the presence of BrdU<sup>+</sup> LPMs was detectable 14 days after BrdU pulse, suggesting that they are a long-lived population, i.e., maintained at low levels of proliferation (36). Conversely, the detection of low numbers of proliferating SPMs at 6–10 days after one pulse of BrdU suggests that these cells have a low proliferation rate under steady state conditions and are short-lived cells (36).

Studies with mice deficient in CCAAT/enhancer binding protein (C/EBP)b also support the notion that LPMs and SPMs represent distinct ontogenies, because in the absence of this transcription factor, PerC did not contain LPMs and exhibited increased numbers of SPMs (36). Interestingly, adoptively transferred SPMs differentiated into LPMs in Cebpb*−*/*<sup>−</sup>* mice. However, in control mice that have normal numbers of LPMs, only a small frequency of transferred SPMs acquired the F4/80hiMHCIIlowCD93<sup>+</sup> phenotype of LPMs. Based on these results, the authors proposed that under physiological conditions, SPMs appear to contribute in only a small way to generate LPMs, but SPMs could be involved in the maintenance of LPMs in situations where this pool has been greatly reduced, such as under inflammatory conditions or following radiation ablation (36). These data are consistent with the findings of Yona et al. (26), which demonstrated the presence of monocyte-derived cells in the LPM compartment 8 weeks after the i.p. injection of thioglycollate. Together with LPMs, a subset of proliferating BM-derived inflammatory macrophage has also been associated with selfrenewal mechanisms during the resolution of peritonitis induced by zymosan and thioglycollate (42). Conversely, LPMs do not seem to contribute to the SPM pool, even during inflammation. Our group demonstrated that adoptively transferred CFDA-SElabeled LPMs 1 h after LPS stimulation retained its phenotype, and no CFDA-SE<sup>+</sup> cells were found in the SPM compartment until 2 days after stimulation (31).

In the last year, a great advance in the understanding of the transcriptional control of peritoneal macrophages provided novel insights into this scenario (39, 40). The zinc finger transcription factor GATA-binding protein 6 (GATA6) appears to regulate the majority of peritoneal macrophage-specific genes (PMSGs). Of note, GATA6 is selectively expressed by LPMs (40). Accordingly, the number of LPMs were greatly reduced in peritoneal lavages from GATA6-KOmye and Mac-GATA6 KO mice, which have a GATA6 deficiency in all myeloid cells or only in the macrophage lineages, respectively (39, 40). Interestingly, retinoic acid (RA) is the extracellular factor that regulates GATA-6-specific gene expression in LPMs, because vitamin A depleted (VAD; the RA precursor) mice exhibited a decrease in GATA6 expression and LPM numbers (40). Moreover, the stimulation of peritoneal macrophages from VAD mice with all-trans RA restored the expression of GATA-6 and many PMSGs at levels found in peritoneal macrophages from control mice. In addition to the regulation of gene expression profiling in peritoneal macrophages, GATA-6 appears to be involved in the control of the proliferation, survival, and metabolism of these cells (39, 77). GATA-6-deficient macrophages demonstrate an altered proliferation state during peritonitis (39). Moreover, Lyz2-Cre *<sup>×</sup>* GATA6(flox/flox) mice also exhibit reduced numbers of peritoneal macrophages, which could be explained by the perturbation in their metabolism, culminating in the high frequency of cell death found in this compartment (77). Despite great contributions to our understanding in the involvement of GATA-6 in peritoneal macrophage development, metabolism, self-maintenance, and survival, the existence of distinct pathways that could govern the transcriptional regulation of SPMs remains largely unknown.

In addition to transcriptional regulation, signaling factors derived from the microenvironment also play an essential role in promoting the development and phenotype of tissue-resident macrophages. For example, TGF-β1 signaling is required for the development of the microglia population and to regulate a microglia expression program through the Smad tissue factors (78–80). Heme has been shown to induce Spi-c, a transcription factor important for red pulp macrophage development (81, 82). Finally, in PerC, omentum-derived RA promotes the expression of GATA-6 in the LPM subset, determining its localization and functions (40), even if the factors that maintain the SPM pool under steady state conditions still remain to be elucidated.

# **Dynamics and Function of Peritoneal Macrophage Subsets**

Mouse PerC is a compartment where many cell types co-habitat and interact, similar to the secondary lymphoid organs. In addition, PerC is a unique body compartment that contains B-1 cells (83). Under steady state conditions, the peritoneal cells comprise LPMs, SPMs, B-1 cells, conventional B-2 cells, T cells, NK cells, DCs, and granulocytes (mostly eosinophils) (31, 35). B1 cells constitute the majority of the PerC cell population, whereas the SPM and LPM frequencies represent 30–35% of total peritoneal cells (31, 35) (**Figure 2A**). However, after inflammatory or infectious stimuli, there is a dramatic alteration in cell numbers and the frequencies of each of PerC cell subpopulation. With regard to the myeloid compartment, modifications in PerC cell composition include the disappearance of LPMs, increases in SPM frequency and numbers, and a massive recruitment of inflammatory monocytes (24, 31, 35, 36, 40) (**Figure 2B**).

The "macrophage disappearance reaction" (MDR) in PerC has been extensively described during delayed-type hypersensitivity (DTH) and acute inflammatory processes (84). MDR has been associated with cell death, emigration to draining lymph nodes, or adherence of macrophages to structural tissues. LPMs are the unique peritoneal macrophage subset that disappears from PerC, which is attributed not to cell death but rather to their migration to the omentum (31, 40). LPM disappearance in response to inflammatory stimuli is accompanied by an increase in SPM and inflammatory monocyte numbers (24, 31, 35, 36, 40) (**Figure 2B**), and has been correlated with the renewal and improvement of immune conditions of the PerC (35). Adherent peritoneal cells from naive mice, which are composed primarily of LPM, exhibit

#### **FIGURE 2 | Summary of the dynamic of peritoneal macrophage subsets**. **(A)** Under homeostatic conditions, peritoneal macrophages comprise two subsets LPMs and SPMs (31). LPMs, which are the major peritoneal macrophage population, appear to be responsible for phagocytosis of apoptotic cell and tissue repair (36). **(B)** At the outset of inflammation, the myeloid compartment is modified in general by disappearance of LPMs, increase of SPMs numbers, and monocytes influx (31, 35, 36, 40). The changes in the myeloid cells from zymosan, *T. cruzi*, and LPS stimulated or thioglicollate-elicited PerC result in the gain of immune state (35, 36). SPMs

from zymosan and *T. cruzi* stimulated mice contribute to effector function of PerC through secretion of high levels of NO and presence of IL-12-producing cells (35). In response to LPS *in vivo*, SPMs produce several inflammatory cytokines, such as IL-12, MIP-1α, TNF-α, and RANTES, whereas LPMs produce enhanced amounts of G-CSF, GM-CSF, and KC (36). LPMs, which migrate to omentum by a retinoic acid and GATA-6-dependent way in response to *in vivo* LPS stimulation or vitamin-A deprivation, return to PerC and appear to be correlated with GALT-independent and TGF-β2-dependent IgA production by B-1 cells in the intestine (40).

a high frequency of cells stained for β-galactosamine (β-gal), a senescence marker (85–87). These cells are unable to secrete NO in response to LPS challenge (35). In contrast, adherent peritoneal cells from *Trypanosoma cruzi* or zymosan-stimulated mice in which the main cell population constitutes SPMs and monocytes (F4/80lowMHCIIintLy-6C+), respectively, display a significant reduction in the frequency of β-gal-positive cells and secrete high levels of NO in response to LPS (35). The frequency of IL-12-producing cells after *in vitro* LPS plus IFN-γ stimulation was also higher within myelo-monocytic cells from mice exposed to zymosan and *T. cruzi* than the frequencies of IL-12-producing cells found in unstimulated mice (35). In response to *Staphylococcus epidermidis* cell-free (SES) supernatant *in vivo* stimulation, F4/80lowCD11b<sup>+</sup> cells (consisting of SPMs and DCs) produced enhanced levels of IL-1β, IL-1α, TNF-α, and IL-12 in the presence or absence of subsequent SES treatment (37). In contrast, the supernatants of adherent cells from naïve mice treated with SES were found to contain high levels of MCP-1, MCP-1α, MIP-1β, and G-CSF (37). It is important to note that 4 days after thioglycollate injection, peritoneal cells, an extensively studied cell population (88–91), also consist primarily of SPMs and inflammatory monocytes (31, 40). The increase in SPM numbers and the influx of inflammatory monocytes that will give rise to SPMs greatly contribute to the improvement of the capacity of PerC to deal with inflammatory stimuli. Indeed, although neither SPMs nor LPMs produce significant levels of pro- or anti-inflammatory cytokines under steady state conditions (35–37), SPMs appear to develop a pro-inflammatory profile in response to *in vitro* stimuli. SPMs produced high levels of TNF-α, MIP-1α, and RANTES in response to LPS, whereas LPMs were the unique population that produced abundant levels of G-CSF, GM-CSF, and KC in response to the same stimulus (36) (**Figure 2B**).

The NO secretion and pro-inflammatory cytokine production are the most important functions of activated macrophages by inflammatory stimulation and assigns the M1 profile (13, 34, 92–97). The functional profile of peritoneal macrophages was previous studied by our group and others (33, 34). Peritoneal macrophages from Th1-prone mouse strains (C57BL/6 and B10.A) are easily activated to produce NO in response to rIFN-γ or LPS, characterizing the M1 profile. In contrast, macrophages from Th2-prone mouse strains (BALB/c and DBA/2) exhibit a weak NO response as a consequence of high levels of spontaneously secreted TGF-β1 (34). Moreover, the cells from C57BL/6 IL-12p40-deficient mice have a bias toward the M2 profile, indicating that IL-12 is required for M1 polarization of peritoneal macrophages (33). Although LPMs from naïve mice can produce NO after *in vitro* LPS stimulation, SPMs produce higher levels of NO than LPMs following *in vivo* LPS stimulation. The NO secretion by LPMs was also detected by flow cytometry in *Escherichia coli* inoculated mice (31), whereas nitrite was not produced *in vitro* by LPS-stimulated adherent peritoneal cells from control mice, which is composed mainly by LPMs (35). In addition, adherent cells obtained 48 h after *T. cruzi* infection, which are mostly composed by SPMs, were the unique source of NO without *in vitro* subsequent challenge with LPS (35). In resume, the SPM and LPM subsets cannot be accommodated in the M1/M2 framework considering the NO secretion. However, considering phagocytic assays, SPMs appear to develop an efficient profile to control infections as M1 macrophages, whereas LPMs assume a role in the maintenance of PerC physiological conditions as M2 or alternative macrophages. Despite the preserved phagocytic ability of LPMs, higher numbers of zymosan and *E. coli* were found inside of SPMs at early time points after i.p. injection (31, 35). Conversely, at 1 h after challenge, LPMs appear to present a higher phagocytic index of apoptotic thymocytes in comparison to SPMs (36) (**Figure 2A**).

In addition, it was recently demonstrated that LPMs have a unique ability to induce gut-associated lymphoid tissue (GALT)-independent IgA production by peritoneal B-1 cells (40) (**Figure 2B**). RA and TGF-β2 are the most critical factors to induce IgA class switching, and the production of TGF-β2 is regulated by the *Tgfb2* and *Ltbp1* genes, which are expressed by LPMs in a GATA-6-dependent manner. This process is regulated by the abundant presence of RA in the omentum, which is responsible for the induction of GATA-6 expression in LPMs that migrates to this tissue. The dynamic of LPM migration between the PerC and the omentum after the stimulation of PerC is correlated with their disappearance and the return to basal numbers of LPMs later after stimulation with LPS, zymosan, and thioglycollate (24, 31, 35, 36, 39, 40). This observation suggests that LPMs can return to PerC to resolve an infectious or inflammatory process. Therefore, the presence of two specialized macrophage subsets in PerC is crucial to maintain the health of this compartment under different situations.

# **Concluding Remarks**

Peritoneal macrophages represent one of the most studied macrophage populations. However, the existence of two phenotypically and functionally distinct subsets, LPMs and SPMs, residing in the PerC was recognized recently (31). In the last year, great advances in our understanding of the transcriptional regulation of peritoneal macrophages have brought novel insights into the identification of LPMs and SPMs (39, 40). GATA-6, an LPMrestricted transcription factor, regulates many PMSGs, including those related to the maintenance of LPMs in PerC (40) and those that determine their function (40), metabolism, proliferation, and cell survival (39, 77). Under steady state conditions, LPMs appear to originate independently from hematopoietic precursors and retained the ability to proliferate *in situ*, maintaining physiological numbers (26, 36). Conversely, SPMs appear to originate from circulating monocytes (31, 36, 40), and their numbers increase remarkably under inflammatory conditions. Of note, SPMs together with their precursor, the inflammatory monocyte population, are the major myeloid populations present in elicited PerC, and are an excellent resource to study the biology of inflammatory macrophages. SPMs and LPMs exhibit specialized functions in the PerC, where SPMs present a pro-inflammatory functional profile, and LPMs appear to have a role in the maintenance of PerC physiological conditions. Moreover, the particular interactions between macrophage subsets and other peritoneal cell populations appear to play crucial roles in PerC immune state. Although the consequences of the crosstalk between SPMs and peritoneal T and B lymphocytes remain to be clarified, LPMs are required for GALT-independent and RA-dependent IgA production by peritoneal B-1 cells (40). Finally, the elucidation of the influence of soluble factors and the microbiota on the maintenance of LPM/SPM ratios in PerC, and the role of these subsets in the systemic immune response are the future challenges for this field.

# **References**


# **Acknowledgments**

This work was supported by Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP – Brazil) Proc 2013/16010- 5 and 2013/07140-2, Brazilian Research Council (CNPq-Brazil), CAPES, and INCTV.


**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 Cassado, D'Império Lima and Bortoluci. 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 complex myeloid network of the liver with diverse functional capacity at steady state and in inflammation

#### **Christoph Eckert, Niklas Klein, Miroslaw Kornek and Veronika Lukacs-Kornek \***

Department of Medicine II, Saarland University Medical Center, Homburg, Germany

#### **Edited by:**

Florent Ginhoux, Singapore Immunology Network, Singapore

#### **Reviewed by:**

Frank Tacke, University Hospital Aachen, Germany Charles-Antoine Dutertre, Duke-NUS Graduate Medical School, Singapore

#### **\*Correspondence:**

Veronika Lukacs-Kornek, Department of Medicine II, Saarland University Medical Center, Kirrbergerstr. 100, Homburg 66424, Germany e-mail: veronika.lukacs-kornek@ uniklinikum-saarland.de

In recent years, it has been an explosion of information regarding the role of various myeloid cells in liver pathology. Macrophages and dendritic cell (DC) play crucial roles in multiple chronic liver diseases such as fibrosis and non-alcoholic fatty liver disease (NAFLD). The complexity of myeloid cell populations and the missing exclusive marker combination make the interpretation of the data often extremely difficult. The current review aims to summarize the multiple roles of macrophages and DCs in chronic liver diseases, especially pointing out how these cells influence liver immune and parenchymal cells thereby altering liver function and pathology. Moreover, the review outlines the currently known marker combinations for the identification of these cell populations for the study of their role in liver immunology.

**Keywords: dendritic cells, Kupffer cells, liver fibrosis, NASH, inflammatory monocytes**

#### **LIVER AS AN IMMUNE MILIEU**

The liver functions as a metabolic center to ensure the proper processing of nutrients and the clearance of toxins; yet, plays multiple roles in systemic immune responses and in immune surveillance. The liver receives blood from both the systemic circulation and the intestine that mixes within the liver sinusoids (1). Approximately, two-third of the hepatic blood flow procures from the oxygen rich arteria hepatica and one-third is from the vena porta carrying microbial and food-derived antigens and molecules (1, 2). The mixed blood travels through the sinusoids that are specialized blood vessels lined by the liver sinusoidal endothelial cells (LSECs). LSECs assemble a discontinuous endothelium that is in contact with various passenger and organ-resident immune cells (3). Besides LSECs, the liver contains other parenchymal cells such as hepatocytes and hepatic stellate cells (HSCs). The activation status and extracellular matrix production of HSCs are critical for the progression of multiple liver diseases (4, 5). Importantly, these liver parenchymal cells interact with the variety of immune cells, influence memory T cells, respond to danger signals, and additionally take on the role of antigen presenting cells (APCs) within the liver (6, 7). As APCs, they present antigens

in the context of immunosuppressive cytokines and inhibitory surface molecules resulting largely in tolerance (6, 7). The liver also encompasses large populations of hematopoetic cells such as innate lymphocytes (NK, NKT cells, and γδT cells) and myeloid cells [dendritic cells (DCs) and macrophages] (6). Multiple crosstalks exist between hematopoetic cells and liver parenchymal cells at steady state and during injury. This review focuses on the physiological and pathological roles of liver DCs and macrophages paying special attention to chronic liver diseases such as fibrosis and non-alcoholic fatty liver disease (NAFLD).

#### **LIVER DENDRITIC CELLS**

Dendritic cells are present in all tissues and represent the major APCS within the body (8). They constantly sense their environment and capable of recognizing pathogens and various danger signals. Activation of DCs results in their maturation toward several functionally distinct "effector DCs" (9) that drive T cell responses, such as T helper cell differentiation, induction of CTL, and T cell tolerance (9). Additionally, DCs communicate with innate lymphocytes (e.g., NK, NKT cells), therefore, can influence both innate and adaptive immune responses (8).

Murine liver DC population, similarly as in most nonlymphoid organs (except the lamina propria and dermis), consists of three types of DCs (**Table 1**): the cDC1s (classical type 1 DCs), the cDC2s (classical type 2 DCs), and pDCs (10, 11). Despite of this categorization, in most liver studies, DCs are evaluated as either CD11c<sup>+</sup> or MHCII<sup>+</sup> cells. Although neither of the molecules pinpoint exclusively DCs, using these markers liver DCs are primarily located within the portal area and rarely scattered within the parenchyma (6). The cDC1 cells resemble lymphoid tissue CD8<sup>+</sup> DCs, show migratory capacity in various non-lymphoid organs, and can efficiently cross-present cell-associated antigens (10, 11). Although the role of DC migration in liver pathology has not been explored in details, antigen injected or targeted to the

**Abbreviations:** APC, antigen presenting cell; BDL, bile duct ligation; BM, bone marrow; DCs, dendritic cells; Batf3, basic leucine zipper transcription factor, ATFlike 3; CCL4, carbon tetrachloride; CCR9, chemokine (C–C motif) receptor 9; CTL, cytotoxic T lymphocyte; DNGR-1, DC NK lectin group receptor-1; DT, diphtheria toxin; DTR, diphtheria toxin receptor; ECM, extracellular matrix; FA, fatty acid; Flt3L, FMS-like tyrosine kinase 3 ligand; GM-CSF, granulocyte macrophage colonystimulating factor; HSC, hepatic stellate cells; HL-DC, high-lipid liver dendritic cell; Id2, inhibitor of DNA binding 2; IL, interleukin; KC, Kupffer cells; LL-DC, low-lipid liver dendritic cell; LSEC, liver sinusoidal endothelial cells; LPS, lipopolysaccharide; MCD, methionine choline deficient; MCP-1, monocyte chemotactic protein-1; MMP-9, matrix metalloproteinase-9; MMP13, matrix metalloproteinase-13; NASH, non-alcoholic steatohepatitis; pDC, plasmacytoid dendritic cell; PDL-1, programed cell death 1 ligand; TNFα, tumor necrosis factor-alpha; TAA, thioacetamide; TLR, toll like receptor; zbtb46, zinc finger and BTB domain containing.



DCs and macrophages are classified according to the recently suggested nomenclature based on the ontogeny of these cells (15).The M1 type monocyte/macrophage population present in liver injury and the restorative macrophages during resolution could not be incorporated in this nomenclature as their origin and relation to monocytes and resident liver macrophages (KCs) need further clarification in the future. DCs, dendritic cells; KCs, Kupffer cells; pDCs, plasmacytoid DCs; pre-DCs, precursor DCs.

liver reaches the draining LN and induces T cell activation (12, 13). Additionally, migratory DCs could be identified within the portal lymphatic vessels in electronmicroscopy analyses (1, 14).

The cDC2s within non-lymphoid organs are heterogeneous and partially monocyte lineage derived (10, 11). Their specific role is less understood in non-lymphoid organs, involving the liver as well. While the development of this subset depends in most non-lymphoid organs on the presence of FLT3L and M-CSF, in the liver, these cells are not exclusively dependent on these growth factors but yet on an unidentified molecule (10). The liver, similarly to other non-lymphoid tissues, contains not only fully differentiated DCs but precursor DC population as well. From these pre-DCs, either FLT3L or GM-CSF can induce liver DC development bestowing DC homeostasis *in situ* (29, 30).

Functionally, CD11c<sup>+</sup> cells isolated from healthy mouse liver are less mature, have lower capacity to endocytose antigen, and induce less efficient allogenic T cell activation as secondary lymphoid organ (SLO)-derived DCs (31, 32). The inhibitory/tolerogenic capacity of liver DCs could be attributed to the specific microenvironment provided by parenchymal cells of the liver. Fibroblastic and VCAM<sup>+</sup> cells derived from the liver could induce hematopoetic progenitor cells to differentiate toward tolerogenic DCs *in vitro* that can inhibit experimental autoimmune hepatitis (33). It is assumed that circulatory DCs during their translocation within the liver sinusoids toward the lymphatics receive such tolerogenic education from liver parenchymal cells (14, 34). Yet, its *in vivo* relevance needs to be elucidated.

Freshly isolated murine liver CD11c<sup>+</sup> cells promote Th2 rather than Th1 T cell differentiation and via interacting with NK cells induce regulatory T cell (Treg) development (35, 36). Moreover, liver DCs produce increased amount of IL-10, IL-27 but less IL-12 upon LPS stimuli (37, 38). This hyporesponsive behavior toward TLR stimuli, known as endotoxin tolerance, involves LPS/TLR4

but also extends toward other TLRs (6). This is especially important, as the liver is constantly exposed to gut derived microbial products. The breakdown in this tolerance could be observed in colitis where pro-inflammatory DC/macrophage population expands within the liver due to the increased amount of bacterial products present in the portal blood. This creates an inflammatory environment in the liver despite the absence of direct liver damage (39). The tolerant state toward TLRs is an active process and involves the action of various negative regulators of the TLR signaling pathway (6). Interestingly, under steady state, liver DCs rather respond to ECM stimuli (collagen-type I, laminin, fibronectin) that induces MHC-II upregulation and maturation of GM-CSF expanded liver DCs *in vitro* (40).

In humans, the cDC2 cells (CD11c<sup>+</sup> BDCA1+) are the most abundant in the liver and they exhibit similar immature, tolerogenic capacity as their murine counterpart (16, 41) (**Table 1**). The cDC1 cell population that expresses CD141<sup>+</sup> has been recently identified as a counterpart of murine CD8α <sup>+</sup> cells (42). These cells induce pro-inflammatory allogeneic MLRs, resulting in IFNγ and IL-17 production by activated T cells (17). Importantly, as opposite to cDC2s and pDCs, cDC1s (identified in the study as CD141<sup>+</sup> cells) were markedly decreased during liver diseases but among the DC-subsets produced the highest level of IFN-λ (17). It is possible that functional differences are reflected among the DC subsets and each subset represents different aspects of liver immunity and tolerance. In line with this, a classification of murine liver DCs according to their lipid content distinguishes between immunogenic and tolerogenic liver DCs. Due to their acetyl-CoA carboxylase activity, HL-DCs (high lipid DCs) mount strong immunogenic CTLs while the LL-DCs (low lipid DCs) with low lipid content are tolerogenic (43). Notably, the marker combinations used for this study showed that both HL-DCs and LL-DCs include multiple DC-subsets distinguished by currently known

surface markers and were not restricted to one specific subset. Novel surface molecules are needed to specifically explore their functional diversity.

pDCs are the major source of type-I IFN, regulate NK cell activity, and play important role in the induction of antiviral immunity (44, 45). The murine liver is especially rich in pDCs; yet, the human counterpart contains a smaller proportion of this population among all DCs (17) (**Table 1**). Under steady state condition, pDCs express low level of costimulatory molecules, are weak T cell stimulators, and induce apoptosis in activated T cells in a Treg dependent manner (46). Later could indicate a cellular interplay between pDCs and Tregs in the liver microenvironment in order to maintain the tolerogenic milieu. Accordingly, pDCs can induce efficient CD4 and CD8 T cell tolerance to orally administered antigens (47).

Microbial products, such as muramyl dipeptide present in the portal blood, upregulate PDL-1 in pDCs and reduce their response to TLR9 stimuli (48). This is another example for the TLR-mediated hyporesponsiveness ("endotoxin tolerance") in the liver. Strikingly, upon FLT3L treatment, the expanded liver pDCs display strong immunostimulatory properties (49). It is unclear whether this could be due to the expansion of a specific subpopulation of pDCs, or their modified interaction with Tregs, or the result of the complete rearrangement in the myeloid cell compartment, and the consequent imbalance in the tolerogenic milieu.

Taken together, the multiple DC subsets within the liver participate in guarding the tolerogenic environment and primarily skewed toward suppressing T cell responses and toward induction of Tregs. While DCs are the main APCs and inducers of T cell immunity in SLOs, within the liver environment the question still remains: how immunity can be induced by DCs in such suppressive microenvironment? Induction of immunity might be attributed to special DC subpopultaions such as the CD141<sup>+</sup> cDC1s in humans (17) and the CD103<sup>+</sup> cDC1s in mice (50). Moreover, the appearance of novel DC population, such as monocyte-derived DCs present in iMATEs (intrahepatic myeloid-cell aggregates for T cell population expansion), participates in efficient CLT expansion within the liver (51). Alternatively, immunity is induced by migratory DCs reaching the draining LN, thus, outside of the liver suppressive environment. In line with this, antigen specifically expressed in draining LN results in hepatitis inducing CD8 T cell activation, while the same antigen within the liver induces tolerance (52). Additionally, the liver can provide newly formed structures for T cell activation resulting in immunity. Portal tract associated lymphatic structures (PALPs) during *Propionibacterium acnes* granuloma formation and tertiary lymphoid structures in biliary cirrhosis represent locations where possible T and B cell activation takes place, respectively (53, 54).

#### **KUPFFER CELLS – RESIDENT MACROPHAGE POPULATION OF THE LIVER**

Kupffer cells (KCs) are tissue resident macrophages and they represent the largest hematopoetic cell population within the liver. They arise from yolk sac during fetal development (55), adjust themselves to the local microenvironment (56, 57), and renew their population at steady state locally throughout adult life with no or minimal contribution of hematopoetic progenitors or blood monocytes (58–60). In mice, KCs can be distinguished from monocytes among the F4/80<sup>+</sup> cells as Ly6Clow CD11blow cell population (20, 21) (**Table 1**) and possess functional specifications according to their positioning within the sinusoid (61). Recent study could distinguish two KC functional groups: the one with higher phagocytosis capacity and the one with preference toward cytokine production (61, 62). Additionally, macrophages are functionally grouped into two classes M1 and M2. While such plain classification is questionable and often overstated, still provide a simple but distinguishable concept for functional categorization of these cells. M1 (termed classically activated) macrophages are pro-inflammatory, while the M2 (termed alternatively activated) macrophages are suppressive and involved in cellular repair (63). According to this, KCs belong to the M2 type of cells and play fundamental role in homeostasis, immune surveillance, and tissue repair (63).

Their importance as tolerogenic APCs in the liver microenvironment is demonstrated in liver transplantation where they prolong allograft survival (31). At steady state, they inhibit DC mediated T cell activation within the sinusoids and presentation of high affinity peptide by KCs results in deletional CD8 T cell tolerance (6,64). Furthermore, they promote the suppressive capacity of Tregs toward hepatic antigens (65, 66).

As all tissue resident macrophages, KCs express a wide repertoire of receptors for the recognition of pathogens and danger signals such as Toll-like receptors, members of the inflammasome, and scavenger receptors (31). In the presence of TLR ligands such as LPS and CpG, KCs become immunogenic, and can induce CD8 T cell activation, and the generation of efficient CLT response (67, 68). Thus, during liver infection, they support the development of antimicrobial T cell responses. Unfortunately, KCs induce efficient CTL against antigens from the systemic circulation such as the case in influenza infection (69). This CTL response results in bystander hepatitis, often accompanying systemic viral infections. Besides CD8 T cell responses, recent study describes naive CD4 T cell activation in the murine liver by antigens expressed in hepatocytes. This process is independent from lymphoid tissue but dependent on clodronate-sensitive liver APC population possibly involving KCs as well (70). Thus, KCs participate in the generation of both CD4 and CD8 T cell responses.

Using their scavenger receptor repertoire, KCs are involved in the clearance of apoptotic cell debris and central to iron homeostasis (71). KCs interact with multiple immune cells within the sinusoids, such as Tregs, DCs, DC precursors, and innate lymphocytes (7, 53, 72, 73). After recognizing any danger signals, KCs primarily drive the influx of inflammatory leukocytes such as neutrophils and monocytes (63). Thus, KCs function as sentinels and central orchestrators of cellular processes in healthy and injured liver. Additionally, while they support the tolerogenic milieu within the liver, their presence also ensures the protection of the liver during pathogen invasion.

#### **TOOLS TO STUDY LIVER DCs AND MACROPHAGES**

In order to characterize the specific physiological and pathological roles of DCs and macrophages, various animal models, tools have been developed (**Table 2**) Among these models, there are mouse lines deficient in transcription factors that are responsible

#### **Table 2 | Summary of the available models to study liver macrophages and DCs**.


ND, non-determined; DCs, dendritic cells; MM, marginal zone.

for the development of one or multiple subsets of myeloid cells. Due to the multiple cell types affected in these models, the broader impact of each of these genes makes it difficult to unequivocally pinpoint subset specific functions. Nevertheless, these transgenic animals helped significantly to establish broader understanding of macrophage and DC development and their role under steady state and inflammation (63, 74). However, just few of these models have been evaluated so far in fibrosis and non-alcoholic steatohepatitis (NASH) models (**Table 2**) but also in liver immunology. This might extend in the future as genetic model lacking cDC1s have just recently demonstrated that cDC1s respond to hepatotropic viral infection and are keys in the induction of anti-viral CD8 T cell response *in situ* (50).

The most frequently used cell-depleting tools in liver immunology are the clodronate liposome mediated depletion of mononuclear cells and the CD11c-/CD11b-DTR (diphtheria toxin receptor) transgenic system (87, 89). Clodronate-encapsulated liposomes are taken up by mononuclear cells and the clodronate bisphosphonate within the cell induces apoptosis that results in depletion of the phagocytic cell population. Multiple phagocytic cell types are affected using this depletion method such as KCs, macrophages, and some members of the DC population as well. Since more than one cell types are affected, the effects can be

extrapolated to a group of cells and not to individual subtypes (89). Additionally, the release of inflammatory mediators (such as TNF) has been associated with this type of cell depletion further complicating the interpretation of experimental results (90).

The other widely used tool for liver biology is the CD11c-DTR-based depletion system. Here, the human diphtheria toxin receptor is expressed under the CD11c promoter and administration of diphtheria toxin results in the depletion of CD11c<sup>+</sup> cells. This model is used to dissect the role of conventional DCs. The major disadvantage in this system is that multiple cell types are affected such as marginal zone macrophages, monocytes, activated CD8 T cells, NK cells, and plasmacytoid DCs (89). Two different CD11c-DTR mouse lines have been generated: the one encompassing only a short piece of the CD11c promoter (3) and the one with the full-length promoter (91). Although, they differ in the list of affected cell types, they gave important insights in the role of CD11c<sup>+</sup> cells in liver immunology (**Table 2**). Novel DTR tools have been developed in recent years that aim to restrict the expression of DTR more specifically to DCs. The zbtb-46-DTR model uses the transcription factor zbtb46 that is exclusively expressed by DCs and DC-committed precursors (92). Unfortunately, zbtb46 is upregulated in monocytes stimulated with GM-CSF and IL-4, suggesting some limitations to this promoter (93, 94). Another promising promoter is the DC NK lectin group receptor-1 (DNGR-1) that seems to be highly restricted to the DC linage (95). Not only for DCs but also for the study of macrophages, the perfect targeting tool still needs to be developed. The primary tool for analyzing macrophages is the CD11b-DTR system. However, CD11b is a widely expressed marker among multiple immune cell types causing caveat for the interpretation of cell types using this model for understanding liver diseases (**Table 2**) (87).

To follow myeloid cells *in situ* using *in vivo* imaging, flow cytometry or microscopy multiple reporter mouse models have been developed. The promoters from different molecules such as e.g., CD11c, Csfr1, CCR2, MHC-II, or CX3CR1 were used to generate these animal models (88) (**Table 2**). These models have their own limitations according to their expression profile that have been reviewed elsewhere (88) (**Table 2**). Some of the promoters are also utilized to express Cre recombinase (**Table 2**). Crossing these Cre expressing lines with animals carrying a flox-ed gene allows the analyses of the cell specific depletion of the gene of interest. Certainly, the specificity and limitation of the models are determined by the expression pattern of the promoter used for Cre expression (87). Despite the availability of these models, only limited have been exploited for understanding specifically liver fibrosis and NASH (**Table 2**).

Taken together, multiple models are available to answer liver immunological questions. While each of the available models has its own limitation, the combination of these models with each other can still pinpoint important contribution of DCs and macrophages in liver pathology.

#### **THE ROLE OF DENDRITIC CELLS AND MACROPHAGES IN CHRONIC LIVER DISEASES**

#### **LIVER FIBROSIS PROGRESSION**

Liver fibrosis is a common endpoint of many chronic liver diseases such as viral hepatitis, primary biliary cirrhosis, alcoholic and NASH, or autoimmune liver disorder (96, 97). To investigate liver fibrogenesis, several rodent models have been developed inducing toxic (CCL4), biliary (bile duct ligation), oxidative (TAA-induced), or metabolic (MCD/methionine choline deficient diet induced) liver damage (96, 97). The MCD diet contains high sucrose and fat (usually 40% sucrose and 10% fat) but lacks the amino acid methionine and the small molecule choline that are essential for hepatic β-oxidation and production/secretion of very low density lipoprotein (VLDL). As a result, lipids are deposited in the liver and steatosis, and NASH develops in these animals (98).

Remarkably, even though the molecular mechanisms leading to hepatic cell death are very different, the process of fibrogenesis and the cellular components involved share common hallmarks. Such common components, that have established their role in liver fibrosis, are the macrophages and the recruited inflammatory monocytes.

Major evidence for the involvement of macrophages in liver fibrosis is demonstrated in *in vivo* depletion studies using the CD11b-DTR system and the clodronate-liposome mediated cell depletion. In CCL<sup>4</sup> induced liver injury, the progression of fibrosis was attenuated in the absence of CD11b<sup>+</sup> cells and the number of HSC-derived myofibroblasts was greatly reduced (79). The administration of clodronate liposomes similarly suggested that macrophages are pro-fibrogenic and affect the survival of HSCs via TNF and IL-1 induced NF-kb signaling (84, 99). Liver macrophage populations change during liver injury. One of the major changes is the recruitment of inflammatory monocytes to the injured liver and their differentiation toward tissue macrophages (24, 26, 28, 100). Resident KCs in liver injury rapidly secrete pro-inflammatory cytokines such as IL-1β, TNF, CCL2, and CCL5 resulting in recruitment of multiple immune cells involving monocytes as well. The accumulation of circulating Ly6Chi monocytes within the liver is greatly dependent on CCR2/CCL2 and CCL1/CCR8 axis (100). The monocyte recruiting chemokines, however, not only originate from KCs but also from TLRactivated HSCs (101). Moreover, senescent hepatocytes and NFkb-inducing kinase (NIK) activation in hepatocytes lead to the release of numerous chemokines (86, 102). These chemokines can influence the migration or activation state of macrophages that in turn induce hepatocyte apoptosis. Accordingly, hepatocytespecific expression of the NIK *in vivo* triggers massive liver inflammation and hepatocyte apoptosis leading to liver fibrosis (86). Thus, the macrophage–hepatocyte cross-talk seems to greatly influence cell recruitment and the activation state of macrophages, thereby affecting the progression of liver injury. The fact that in the above study KC/macrophage depletion using clodronate reversed NIK-induced damage, also strongly suggests this.

Monocyte recruitment to the injured liver can be observed early within 24 h after the induction of CCL<sup>4</sup> damage (25). These early recruited cells are CCR9+, colocalize and interact with CCR9<sup>+</sup> HSCs (27). Furthermore, these monocyte-derived macrophages are characterized as CD11b+F4/80+iNOS<sup>+</sup> cells that exhibit profibrogenic properties via promoting HSC activation, Th1 cell differentiation, and TGFβ release (25, 26). In addition to this, profibrogenic Ly6Chi macrophages express PDGF, IL-13 and IL-4 that directly act on HSC derived myofibroblasts and induce ECM production (25, 26). Macrophages produce various lectins among them galectin-3 is required for TGFβ mediated myofibroblast activation and matrix production that further underline their profibrogenic capacity (103).

Another chemokine that affects the infiltrating monocytes is the fractalkine receptor (CX3CR1). Fractalkine is released by hepatocytes and HSCs during liver injury. It ensures the survival of infiltrating monocytes and influences their differentiation toward tissue macrophages (25). In the absence of CX3CR1, infiltrating monocytes develop into highly inflammatory macrophages that die early via apoptosis. This perpetuates further inflammation and recruitment of Ly6Chi cells. Additionally, CX3CR1 on KCs increase their IL-10 expression and reduces their TNF and TGFβ (104). Thus, fractalkine represent a negative feedback on the extension of liver inflammation through affecting KCs and the presence and destiny of Ly6Chi cells at least in the murine system. It requires future research to clarify how the changes in monocyte and macrophage subsets observed in mice are reflected in humans.

Regarding the M1–M2 classification of macrophages, during the progression phase of liver fibrosis and during fibrosis resolution, both types of cells are present in the liver side by side (105). Interestingly, based on histological analyses these M1 and M2 macrophages localize near to the fibrotic septa and could indicate further undiscovered cross-talk among these cells in liver pathology. Of note, transcriptional analyses of macrophages that are present in the resolution phase display a profile that cannot be classified according to the M1/M2 nomenclature (28).

Accumulation of macrophages within the injured liver caused just partially by the recruited monocytes and their differentiation toward tissue macrophages. There is some evidence that local multiplication of resident and monocyte-derived macrophage population contribute to this process. Ki67 staining during CCL<sup>4</sup> mediated liver injury demonstrated the presence of proliferating KCs and monocyte-derived macrophages (28, 62). In most recent study, Listeria infection of the liver resulted in monocyte-derived macrophage proliferation via IL-4 and IL-33 (106). Whether these cytokines are also involved in this process during other types of liver injury and in humans as well remain to be elucidated.

Multiple animal studies reported that the number of dendritic cells, pre-DCs, and pDCs increase during the progression phase of liver fibrosis (76, 107). This raised the assumption that DCs might contribute to fibrosis progression. Using the CD11c-DTR model, it has been demonstrated that CD11c<sup>+</sup> cells provide a proinflammatory milieu by producing IL-1β and TNF during injury (76). Moreover, isolated cells contribute to HSC survival *in vitro* suggesting a clear profibrogenic capacity of these cells (37). This phenomenon, despite of the relatively broad CD11c expression among other myeloid cells as discussed above (**Tables 1** and **2**), was attributed to DCs.

Another study determined using the same CD11c-DTR system that DC depletion accelerates the development of fibrosis due to their influence on angiogenesis. DCs seem to be the source of the anti-angiogenic VEGF receptor 1 (also known as sFlt-1) and thus influence the bioavailability of VEGF during fibrogenesis (78). Notably, recent study has demonstrated that VEGF<sup>+</sup> inflammatory monocytes/monocyte derived macrophages colocalize with newly formed vessels in injured liver and pharmacological inhibition of CCL2 mediated recruitment of inflammatory monocytes reduces fibrosis-induced angiogenesis without affecting fibrosis progression (108). Thus, recruited monocytes/macrophages seem to counterbalance the anti-angiogenic property of DCs during fibrosis progression. Whether classical DCs or pDCs truly contribute to fibrosis progression or play other role during liver injury still remain to be clarified in the future.

#### **LIVER FIBROSIS REGRESSION**

During liver fibrosis,the increased production of ECM is accompanied by high expression of MMPs and the presence of collagenase activity, suggesting alterations and adjustments in the fibrotic ECM. In fact, the fibrotic ECM seems to be different biochemically than ECM produced during a steady state turnover (109). In lung fibrosis, the pathological ECM activates fibroblastic cells to build further matrix indicating a positive cross-talk between fibroblast and matrix components (109). Also during liver fibrosis, heavily cross-linked, modified ECM could be identified (110); however, it remains to be elucidated whether similar regulatory loop as in the lung operates in liver fibrosis as well.

Importantly, after removal of the noxious agents causing liver damage, fibrotic scars degrade and normal liver architecture can be restored. This process is called resolution. While this functions well in various animal models, in humans this seems to be a point of no return where fibrosis and cirrhosis progresses nonetheless (97).

In resolution, the role of macrophages has been demonstrated in multiple animal studies. Depletion of CD11b<sup>+</sup> cells during fibrosis progression, as above discussed, reduced scarring while during fibrosis resolution led to a failure in matrix degradation (79, 80). This strongly suggests the dominant presence of two functionally different macrophage populations. According to this, Ramachandran et al. have identified a subset of Ly6Clow "restorative" macrophages during resolution (28). These cells originated from Ly6Chi recruited monocytes expressed MMPs including MMP9,MMP12,and phagocytosis related genes. Importantly, based on gene expression profiling, they could not be fit in the M1/M2 macrophage classification. Moreover, phagocytosis of liposomes or cellular debris by liver macrophages could recapitulate this type of restorative phenotype (28). In addition to this, recent study demonstrated that scar associated myeloid cells attract endothelial cells to the scar tissue via VEGF and that genetic ablation of VEGF in myeloid cells resulted in the increase of MMP2 and MMP13 and decrease of TIMP1 in the liver. While macrophages have not been unequivocally identified as myeloid cells in this study, the results indicate that the myeloid cells induced angiogenesis gears the balance toward fibrolysis (111). This is in line with recent findings that demonstrated that VEGF signaling plays key role during liver fibrosis resolution. Anti-VEGF antibody treatment during resolution led to impaired tissue repair. Mechanistically, VEGF regulated endothelial permeability, monocyte recruitment, and affected the CXCL9 and MMP13 expression of scar-associated macrophages. Importantly, depletion of Cfsr1<sup>+</sup> cells (including macrophages, monocytes, and DCs **Table 2**) impaired fibrosis resolution (112).

Based on these findings, macrophages can be grouped in profibrogenic and restorative macrophage populations beyond the M1/M2 scheme, a classification that might be much more beneficial for finding new targets for fibrosis therapy. However, multiple open questions remain concerning the balance of the heterogeneous population in liver diseases and the relation to each other. One molecule could provide a better understanding to the problem, the chemokine CX3CL1. Ramachandran et al. showed a higher expression of CX3CR1 within the restorative macrophage population, then in the profibrotic subset (28). Consistent with these findings, HSC and hepatocyte-derived fractalkine led to the induction of Arginase 1 in a mixed Kupffer-cell/macrophage cell population, a marker that has been associated with the fibrolytic macrophage subset (28, 104). Thus, an intriguing possibility is the progressive class switch between macrophage populations during fibrosis progression and regression. This possibility is underlined by the fact that the overall number of profibrogenic Ly6Chi macrophages strongly decreases in resolution despite the presence of their strong proliferation activity at early time points of fibrosis regression. At the same time, the number of Ly6Clo macrophages increases (28). Along this line, blocking CCL2 dependent liver infiltration by Ly6Chi monocytes during fibrosis regression leads to a higher relative amount of Ly6Clo macrophages (113). Moreover, the Ly6Clo macrophages could be shown to be postphagocytic and seem to appear in the phase of reduced hepatocyte death, further supporting the switch concept (28). It remains to be clarified in the future how the macrophage populations interact and relate to each other. Similarly as the murine restorative macrophages in humans, this population is likely represented by the CD14<sup>+</sup> CD16<sup>+</sup> cells (24, 26). They display phagocytic activity but as opposite to the murine cells express a variety of pro-inflammatory and pro-fibrogenic molecules as well.

Besides macrophages, DCs have also been implicated in liver fibrosis resolution. Jiao et al. have demonstrated that depletion of CD11c<sup>+</sup> cells leads to delay in fibrosis resolution and delayed clearance of activated HSCs. To more precisely pinpoint DCs in this process, adoptive transfer of purified DCs or expansion of endogenous DCs using FLT3L could accelerate regression. Moreover, DCs were the source of MMP9 and therefore seem to complement restorative macrophages in this process (75).

#### **NON-ALCOHOLIC FATTY LIVER DISEASE**

Non-alcoholic fatty liver disease (NAFLD) is the hepatic manifestation of metabolic syndrome that includes hypertension, hyperlipidemia, insulin resistance, and visceral adiposity, and shows a worldwide increasing tendency among chronic liver diseases (114, 115). In most cases, the liver steatosis is mild. However, up to onefifth of the cases progresses toward NASH that is characterized by intrahepatic inflammation, increased steatosis with hepatocellular ballooning, and often accompanied by progressive fibrosis (114, 115). NASH is prone to the development of cirrhosis and liver cancer (115). While the precise cellular and molecular mechanisms of NASH are not yet fully understood, multiple studies have investigated macrophages and DCs in this disease.

Similarly as during liver fibrogenesis, in NASH, the two main components that show alterations are the response of macrophages/KCs and the recruited inflammatory monocytes. The key role of macrophages/KCs in NASH has been demonstrated in studies where these cells were specifically depleted using gadolinium chloride or clodronate liposomes (81–83, 85). In the absence of KCs, the steatohepatitis was markedly reduced. In addition to this, KCs display an M1 TNF expressing pro-inflammatory phenotype and increase triglyceride accumulation, decrease fatty acid oxidation and insulin responsiveness of hepatocytes (82, 83). KC-derived TNF production seems to be central in NASH development, as silencing liver TNF or using TNFR1/2 deficient animals attenuate liver steatosis compared with control wild-type animals (85, 116).

Multiple triggers have been identified for KC activation and for the induction of their pro-inflammatory cytokine production in NASH. TLR4 deficient animals showed reduced liver damage and KC depletion prevented the increase in TLR4 expression during MCD diet (81). Bacterial product induced KC activation is in accordance with clinical data that demonstrate bacterial translocation in NASH patients (117). Notably, TLR4 can be triggered not only by LPS but also by free fatty acids and high mobility group box-1 protein (HMGB1) that is increased in obesity and during hepatocyte injury. Not only LPS but also translocated nucleic acids have been implicated in the development of NASH via triggering TLR9 mediated KC activation and IL-1β release (118).

Lipidomics and mass spectrometry profiling revealed that KCs accumulate toxic lipids due to the dysregulation of lipid metabolism during high fat diet. Moreover, these lipid-loaded KCs produce pro-inflammatory cytokines and chemokines (119). The balance between the M1 and M2 type of KCs seem to be a key for NASH progression. Mice fed with high-fat diet displayed a predominant M2 KC polarization, the apoptosis of M1 KCs and resistance to hepatocyte steatosis. *In vitro* experiments demonstrated that M2 macrophages release IL-10 that in return increase the sensitivity of M1 macrophages to undergo apoptosis (120).

The other hallmark of NASH is the increased monocyte recruitment to the injured liver. Activated KCs upregulate their MCP-1 expression that is the major chemokine involved in the recruitment of Ly6Chi cells. These Ly6Chi cells are pro-inflammatory and further perturb hepatic inflammation (85). Consequently, CCR2 deficient animals show decreased steatosis (113, 121). On the other hand, CCR2 signaling, when MCD diet is replaced with control diet, counteracts tissue resolution by perpetuating inflammation (113). This is a similar phenomenon as observed in fibrosis resolution (113).

Non-alcoholic steatohepatitis is associated with increased number of hepatic DCs identified by MHCII<sup>+</sup> CD11c<sup>+</sup> cells (77). Depletion of these cells using CD11c-DTR mouse model exacerbates hepatic inflammation whereas during the resolution phase delays the reconstitution of normal tissue homeostasis. Importantly, these cells take up apoptotic cells, inhibit TLR expression, T cell expansion, and cytokine production by innate cells (77). This strongly suggests DCs as an important negative regulator of NASH inflammation. As opposite to this, another study has classified CD11c<sup>+</sup> cells during MCD-diet based on their lipid content (43). It remains to be clarified whether the tolerogenic LL-DC (low lipid DC) population is equivalent with the same immunoregulatory DCs in NASH as described by Henning et al. Of note, based on the surface marker expression profile of HL-DCs and LL-DCs, they rather seemed to be a part of a heterogeneous population, despite that all cells expressed various level of CD11c (43).

#### **SUMMARY AND CONCLUSION**

Taken together, the liver represents a unique immunological niche within the body. Its parenchymal and non-parenchymal cells guard its tolerogenic and suppressive microenvironment while supporting its sentinel task of the portal and systemic circulation (**Figure 1**). Most liver injuries trigger the activation of resident KC/macrophage population that rapidly releases proinflammatory mediators such as TNF and IL-1β. This is followed by a chain of events that seem to be commonly shared by many injuries causing NASH and leading to liver fibrosis. The response involves the alterations within the myeloid cell composition primarily affecting macrophages. Importantly, other immune cells such as DCs, neutrophils, innate cells, and activated T cell are also recruited to the injured liver and play various roles in disease progression (6, 97). The exact role of liver DCs during chronic liver injury is yet to be determined. Nevertheless, they seem to be similarly pro-inflammatory as the Ly6Chi recruited monocytes. This feature is shared with infectious liver diseases; thus, it supposes to induce liver protecting immunity (2, 7). During chronic liver diseases, the overwhelming presence of pro-inflammatory immune cells together with liver damaging noxious agents eventually lead to extensive cell death and scar formation, a common outcome for chronic liver disorders. While KC activation alarms other immune

of various immune cells involving inflammatory monocytes and DCs. The Ly6Chi monocytes differentiate into M1 CCR9<sup>+</sup> iNOS<sup>+</sup> macrophages, and cells to travel to the liver, it influences metabolic processes and

the differentiation of monocytes to inflammatory macrophages could be a possible target in the future. Additionally, DCs and restorative macrophages could become novel objectives for inducing fibrolysis and reversing liver damage. Notably, autologous transfer of expanded mononuclear cells to chronic viral hepatitisassociated fibrotic patients showed improved outcome as indicated by reduced Child-Pugh score (122), suggesting a great potential of myeloid cell transfer-based therapeutic procedures in the future.

cells; LL-DC, low lipid containing DCs; LSEC, liver sinusoidal endothelial cells.

### **AUTHOR CONTRIBUTIONS**

CE, NK contributed to the writing of the manuscript, prepared the tables. MK supervised the students and critically read the manuscript. VL-K developed the concept of the manuscript, supervised, and wrote the manuscript.

#### **ACKNOWLEDGMENTS**

This work was supported by the Alexander von Humboldt Foundation, Sofja Kovalevskaja Award to VL-K by a Deutsche Krebshilfe grant #111184 to MK.

#### **REFERENCES**


survival of hepatocytes. During disease progression, Ly6Chi cells seem to develop into Ly6Clo restorative macrophages. These cells, if the harmful agent vanishes, lead to resolution and can restore normal tissue architecture (**Figure 1**). Especially, in this process, DCs are complementing the macrophage population. In infection, recent report demonstrated that necroptosis of KCs was necessary to induce the Th2 mediated tissue repair (106) that remains to be tested to affect fibrosis resolution in the future. Equally important is the more detailed understanding of the factors involved in the switch from the pro-inflammatory to the restorative macrophage population.

Despite of the significant amount of data available in mice, we have just limited understanding about the course of events in human liver diseases. It will need future studies to analyze DC, monocyte, and macrophage populations within human liver samples not only phenotypically and functionally but also on genomic level in comparison with their murine counterparts. This can lead to better understanding of liver diseases but also for identifying novel therapeutic targets. A promising clinical perspective is to target chemokines in the early phase of the liver response to avoid inflammatory cell recruitment and further inflammation. One possibility is affecting the CCL2 axis. Currently, Cenicriviroc, an inhibitor of CCR2, is tested (Centaur study, phase 2 clinical trial, NCT:022117475) to attenuate fibrosis progression in NASH patients. Along this line, other chemokines that could affect


CD8(+) T cell accrual. *Immunity* (2010) **33**:955–66. doi:10.1016/j.immuni. 2010.11.020


**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: 27 February 2015; accepted: 30 March 2015; published online: 20 April 2015. Citation: Eckert C, Klein N, Kornek M and Lukacs-Kornek V (2015) The complex myeloid network of the liver with diverse functional capacity at steady state and in inflammation. Front. Immunol. 6:179. doi: 10.3389/fimmu.2015.00179*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology.*

*Copyright © 2015 Eckert , Klein, Kornek and Lukacs-Kornek. 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.*

# Human and mouse mononuclear phagocyte networks: a tale of two species?

#### *Gary Reynolds1,2 and Muzlifah Haniffa1 \**

*1Human Dendritic Cell Laboratory, Institute of Cellular Medicine, Newcastle University, Newcastle upon Tyne, UK, 2Musculoskeletal Research Group, Institute of Cellular Medicine, Newcastle University, Newcastle upon Tyne, UK*

Dendritic cells (DCs), monocytes, and macrophages are a heterogeneous population of mononuclear phagocytes that are involved in antigen processing and presentation to initiate and regulate immune responses to pathogens, vaccines, tumor, and tolerance to self. In addition to their afferent sentinel function, DCs and macrophages are also critical as effectors and coordinators of inflammation and homeostasis in peripheral tissues. Harnessing DCs and macrophages for therapeutic purposes has major implications for infectious disease, vaccination, transplantation, tolerance induction, inflammation, and cancer immunotherapy. There has been a paradigm shift in our understanding of the developmental origin and function of the cellular constituents of the mononuclear phagocyte system. Significant progress has been made in tandem in both human and mouse mononuclear phagocyte biology. This progress has been accelerated by comparative biology analysis between mouse and human, which has proved to be an exceptionally fruitful strategy to harmonize findings across species. Such analyses have provided unexpected insights and facilitated productive reciprocal and iterative processes to inform our understanding of human and mouse mononuclear phagocytes. In this review, we discuss the strategies, power, and utility of comparative biology approaches to integrate recent advances in human and mouse mononuclear phagocyte biology and its potential to drive forward clinical translation of this knowledge. We also present a functional framework on the parallel organization of human and mouse mononuclear phagocyte networks.

Keywords: mononuclear phagocyte system, dendritic cells, macrophages, monocytes, comparative genomics

# Introduction

The mononuclear phagocyte system (MPS) is a branch of the immune system comprising dendritic cells (DCs), macrophages, and monocytes (1–3). The many functions of the MPS include tissue maintenance and healing, innate immunity and pathogen clearance, and the induction of adaptive immune responses (1–3). Manipulating these functions could lead to clinical benefit, such as modulating DCs to develop antigen-specific anti-tumor immunity or suppressing peripheral autoreactive T cell responses in autoimmunity (4, 5). Several factors need to be considered in designing immunotherapy targeting the MPS, including cellular or pathway target choice and the relevant disease and tissue context. Diversity and plasticity of the MPS, two core features that are paramount for directing the quantity and quality of specific immune responses, have frustrated attempts to develop successful focused therapies. The additional variable of local tissue environment, which also heavily influences the

#### *Edited by:*

*Shalin Naik, Walter and Eliza Hall Institute, Australia*

#### *Reviewed by:*

*Kristen J. Radford, University of Queensland, Australia Simon Yona, University College London, UK*

#### *\*Correspondence:*

 *Muzlifah Haniffa, Human Dendritic Cell Laboratory, Institute of Cellular Medicine, The Medical School, Newcastle University, Framlington Place, Newcastle upon Tyne NE2 4HH, UK m.a.haniffa@ncl.ac.uk*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 11 May 2015 Accepted: 12 June 2015 Published: 25 June 2015*

#### *Citation:*

*Reynolds G and Haniffa M (2015) Human and mouse mononuclear phagocyte networks: a tale of two species? Front. Immunol. 6:330. doi: 10.3389/fimmu.2015.00330*

composition and function of resident and infiltrating mononuclear phagocytes (MPs), also requires careful consideration (1–3).

The MPS was conceived in the 1960s by van Furth to encompass a family of phagocytic mononuclear leukocytes regarded as functional variations of monocytes (6). DCs were embraced as members of the MPS several years later (7). The revolutionary discovery that human monocytes and CD34<sup>+</sup> hematopoietic stem cells (HSCs) could be differentiated into DC (mo-DC) and macrophage-like (mo-Mac) cells provided a convenient *in vitro* model to study human MP biology (8–10). However, murine studies have demonstrated the independence of many DCs, macrophages, and Langerhans cells (LCs) from blood monocytes questioning the accuracy of human *in vitro* monocyte-derived cells in recapitulating *in vivo* populations (11–16). Conventional DCs arise from HSCs along a lineage that does not go through a monocyte stage and are dependent on the growth factor receptor FLT3 (11). In contrast, the majority of tissue macrophages arise from prenatally seeded precursors that can survive into adulthood and are dependent on CSF1-R (12–16).

The constituents of MPS share overlapping surface markers, which poses a challenge in parsing functionally distinct populations. A rewarding approach to unravel this complexity has been comparative biology analysis (17–28). In essence, comparative biology relies on the concept that core developmental programs and functions such as differential CD4 and CD8 T cell priming, cross-presentation, migration, and cytokine production are likely to be non-redundant and conserved between species. In support of this, around 99% of murine genes have human analogs and around 96% are syntenic, despite the two species having 80 million years of divergent evolution (29). Comparative transcriptomic mapping has revealed conserved gene expression profiles in the two species allowing parallels to be drawn between DC and macrophage subsets (17–28). This approach places comparative analysis as the central fulcrum facilitating the integration of fundamental immunology to fertilize clinical translational strands (**Figure 1**). Integrating this workflow with

cutting-edge technologies including single-cell genomics and proteomics approaches has the potential to accelerate discovery in basic MP biology and its clinical applicability (**Figure 1**). Comparative biology has revealed further insights into the origin and function of human and mouse mononuclear phagocyte populations (17–28) and generated new hypotheses to be tested in both species.

The concept of functional specialization as an inherent property imprinted by MP ontogeny and tissue anatomy has been well demonstrated in many murine studies [reviewed in Ref. (1, 3, 30)]. However, the MPS possesses an additional layer of complexity in the form of dynamic mobility, plasticity, and adaptability to tissue/local microenvironment both in steady state and in inflammation (1, 3, 31). These issues have been particularly difficult to dissect in human, where the temporal resolution to observe these kinetics is constrained by snapshot analysis during inflammation and disease without adequate recourse to their onset and evolution (**Figure 2**). Snapshot observations during inflammation may be confounded by temporal variations in MPS composition and function resulting in highly variable biological data. This variability may account for the biological noise inherently observed with outbred humans in contrast to inbred mice in specific pathogen free (SPF) facilities.

Mononuclear phagocytes and their progenitors are in dynamic equilibrium between peripheral tissue, blood, and bone marrow (1, 3, 31, 32). The distinction between MPs within peripheral interstitial tissue and blood can be difficult to establish in highly vascularized organs such as liver and spleen, where large sinusoids are present adjacent to discontinuous endothelial lining that enables greater mobility of leukocytes within these organs. In addition, inflammatory perturbations affect the dynamic equilibrium between tissue, blood, and bone marrow compartments favoring the relative expansion and egress of specific lineages in response to distinct stimuli (33–35). Expansion of monocyte-derived cells dominates the response to inflammatory stimuli in tissue but little is known regarding their fate upon resolution of inflammation (35). Peripheral tissue DCs migrate to the lymph node where they mediate their potent functions upon inflammatory stimuli. Whether they play a prominent role in local tissue immune regulation and how migratory DCs are repopulated during inflammation and its resolution has been poorly characterized.

will accelerate discovery.

# Comparative Biology to Interrogate Human and Mouse MP Networks

Identifying homology between mice and humans in other hematopoietic cells such as T and B cells has been relatively simple at phenotype and practical levels because of shared lymphocyte surface markers (CD3/CD4/CD8 and CD19, respectively) as well as the relative ease of isolating lymphocytes, which form 90% of human peripheral blood mononuclear cells (c.f. <1% being DCs). Nevertheless, there are functional differences in lymphocytes between the two species, such as differentiation requirements for IL-17 (36) and GM-CSF (37, 38) secreting CD4<sup>+</sup> T cells, the specificity of granzyme and FOXP3 expression to define natural Tregs (39), the distinct classes of immunoglobulin (40) and human CD1a, 1b, and 1c-restricted responses to lipid molecules (41). Unfortunately, components of the human and mouse MPS lack overlapping phenotypic markers, hampering initial progress in identifying homologous populations between species.

A range of –omics technologies such as transcriptomics, metabolomics, proteomics, and epigenomics could potentially be employed to assess proximity between species. Of these approaches, transcriptomics is technically most tractable and generates enough complexity to achieve good definition between populations (*n*-dimensions where *n* is the number of genes analyzed) (42, 43). Transcriptome-based comparison of various hematopoietic lineages between human and mouse shows broad conservation but also highlighted specific differences and transcriptional divergence due to gene duplication (43).

#### Transcriptomics

The hypothesis underlying comparative transcriptomics is that the identified MP populations were present in a shared ancestor and that these same subsets are present in modern animals. Furthermore, despite divergent evolution over time, cells from each subset will have a conserved transcriptomic signature similar to that of its equivalent in the other species. Two approaches are generally used to measuring this similarity: (1) unsupervised hierarchical clustering and principal component analysis (PCA), which assigns samples a point in *n*-dimensional space (*n* corresponding to the number of genes analyzed) and applying a distance metric with greater proximity suggesting a developmental relationship, or (2) supervised assessment of defined transcriptome signature enrichment between populations of interest exemplified by gene set enrichment analysis (GSEA) (44) and its later variations (45).

In hierarchical clustering, the Euclidean distance is calculated between samples. In PCA, the same Euclidean metric is used after the *n*-dimensional data are projected on to the two or three dimensions over which the most variation occurs. This approach has the disadvantages inherent in using large sets of gene data, large number of variables/genes, and high inter-sample variability when testing a limited number of samples. The consistent finding that tissue-specific genes predominate in DC microarray transcriptomes highlights the first point. As a result, microarray data of DC subsets from the same tissue tend to cluster together rather than with their equivalent in blood or another tissue (46). This can be corrected for by techniques such as excluding genes that are differentially expressed between pooled cells from each tissue (and classifying these "tissue-specific") (23, 26) or through using an abbreviated gene panel that is enriched for genes that are known to give good definition between DC subsets (17). An important corollary of this finding is that, while the relative contribution of ontogeny and environment to DC function remains to be determined, the list of genes that define ontogeny is a small fraction of the genes that are modulated by the environment and highlights a potential drawback of using blood DCs as a proxy for tissue DCs.

The use of GSEA derives from large-scale microarray data in which it was recognized that groups of co-regulated functionally linked genes may be more relevant than the few genes that are most significantly differentially regulated but functionally unrelated. This approach is dependent upon an *a priori* understanding of gene function and this can introduce bias. When GSEA has been used in aligning DC subsets between species, a "query signature" is produced that defines the subset of interest. Samples in the test population can then be interrogated for whether they are enriched for this query signature. The underlying analysis is based on the non-parametric goodness-of-fit Kolmogorov–Smirnov test statistic with the reference probability distribution that of the query signature. GSEA and its later variant connectivity map analysis (CMAP) have been successfully used to identify homologous MP populations between species and the developmental origin of human inflammatory DCs (17, 23, 25, 26, 47). Steady state homologous MP populations in human and mouse blood, lymph node, and peripheral tissues are illustrated in **Figure 3**.

Most transcriptomics studies thus far on MPs have involved ensemble or bulk-population analysis. This introduces an inherent bias, as cell populations have to be defined *a priori* based on expression of specific markers. More recently, the application of single-cell RNA-sequencing (sc-RNA-Seq) with unbiased analysis potential has been successfully used to interrogate cellular heterogeneity to

uncover new cell populations, functional immune states, and to establish cellular lineage hierarchies and lymphocyte differentiation programs (48–53). These technical advances combined with novel computational approaches have the potential to revolutionize our understanding of MPS biology by unraveling predicted and unexpected functional heterogeneity, which underpins the dynamic repertoire of our immune system in health and disease.

#### Proteomics

Proteomics analysis has revealed differences in viral sensing pathways between murine splenic DC subsets (54) and identified the murine common monocyte progenitor (cMOP), an intermediate cell-type between the monocyte/macrophage and DC precursor (MDP) and monocyte (55). However, current large-scale proteomics approaches require high cell numbers for robust analysis and are impractical for rare populations, especially from limited human tissue material. Protein expression on a more limited scale has been the mainstay of conventional flow cytometry to define populations and assess MP functions at single-cell resolution. Although the number of parameters that can be analyzed simultaneously is limited (17–18 parameters using commercial instruments), the application of new unbiased probabilistic analysis to define populations could reveal new insights to MP heterogeneity (56). Mass cytometry (CyTOF) provides additional parameters (up to 100) and combined with unbiased population assignment has enormous discovery potential. This combined analysis on mouse myeloid cell populations has revealed far greater population heterogeneity than previously appreciated (57).

### Functional Validation

Comparative functional analysis between mouse and human MPs has resulted in variable findings [reviewed in Ref. (30, 58)]. It is unknown if this is due to true biological differences or experimental factors which are not comparable within and between species, including the common use of murine *in vivo* models in contrast to human *in vitro* assays to assess MP functions. Conserved functions are detailed in **Figure 4**.

#### Lineage Analysis

The power and utility of comparative biology to identify homologous MP populations is beginning to be applied to MP lineage analysis. The recent identification of the successive downstream progenies of human MDP; the Common DC precursor (CDP) and precursor of myeloid DCs (pre-cDCs) exploited the conserved dependency on growth factors and cytokines between human and mouse DC precursors (27, 28). Similarly, comparative analysis suggested the monocyte-origin of human dermal CD14+ cells (25) and inflammatory DCs (47). The preservation of LCs and dermal macrophages in GATA2 and biallelic IRF8 deficiencies show that they are bone marrow independent in the steady state and similar to their murine counterparts, also arise from prenatally seeded precursors (59, 60).

# DC, Monocyte, and Macrophage Subsets in Mice and Humans

This approach of using ontogeny and by extension transcription factor dependence to define MPS populations was formalized

Figure 4 | Conserved specializations between human and mouse mononuclear phagocytes.

recently in a proposed nomenclature (61). In this scheme, four adult HSC-derived MP populations are described in mice: two conventional/classical DC subsets (cDC1 and cDC2), plasmacytoid DCs (pDCs), and monocyte-derived cells (61). Both cDCs and pDCs are derived from murine CDP (62, 63). The CDP-derived cells are defined by their dependence on specific transcription factors (TFs): cDC1 are Batf3-dependent, cDC2 are Irf4-dependent, and pDC are E2-2-dependent (61). This definition is unambiguous and avoids using surface markers that can vary between tissues and in inflammation. While the ontogeny approach aids definition of murine populations, it cannot be easily transferred to human DC nomenclature, due to inherent logistical difficulties of human ontogeny studies. However, with the aid of comparative biology approaches, homologous populations between human and mouse MP subsets can be identified and inferences between species on ontogeny and function can be made (**Figures 3** and **4**).

# cDC1

#### Phenotype

This subset is identified in mouse by the expression of CD8α in the spleen and CD103 in non-lymphoid tissues (NLT). Its human equivalent in blood and NLT were initially defined by their high expression of CD141 (thrombomodulin, BDCA-3) (19–23). However, this antigen can be upregulated on blood monocytes and expressed promiscuously by other DC subsets in human tissue (23). The cell adhesion molecule CADM1 (NECL2), C-type lectin, CLEC9A (which recognizes damaged cells), and the chemokine receptor XCR1 are expressed on human and mouse cDC1 (19, 22, 64–66). However, CADM1 expression is not restricted to leukocytes and CLEC9A is also expressed on murine DC precursors (67, 68). Although cDC1 is the only leukocyte expressing XCR1, a commercial antibody against it is currently unavailable. Notably, langerin is expressed on murine but not human cDC1 (23, 69, 70).

#### Homology

Homology between human (XCR1<sup>+</sup>CD141<sup>+</sup>DCs) and mouse (CD8+/CD103+) cDC1 was demonstrated by comparative transcriptomics, phenotype, and functional analyses (17, 19–21, 23, 71). Furthermore, blood and skin CD141<sup>+</sup> DCs cluster together separately from CD1c<sup>+</sup> DCs, CD14<sup>+</sup> and CD16<sup>+</sup> monocytes, and pDCs, suggesting that skin XCR1<sup>+</sup>CD141<sup>+</sup> DCs are the tissue equivalents of blood XCR1<sup>+</sup>CD141<sup>+</sup> DCs (23).

#### Transcription Factors

In addition to Batf3 (72), murine cDC1 differentiation requires Irf8 (73), Id2 (74, 75), and NFIL3 (76). In human, shRNA knockdown of *BATF3* in cord blood HSCs inhibits their differentiation into cDC1 *in vitro* (22). However, cDC1 were detectable in humanized mice reconstituted with *BATF3* knockdown CD34<sup>+</sup> HSCs (22). A possible explanation for this seeming contradiction was shown in mice, where in inflammatory conditions (specifically in the presence of IL-12), other members of the Batf family of TFs appear to be able to compensate for loss of Batf3 (77). *ID2* mRNA is expressed at low amounts in human CD34<sup>+</sup> HSCs but upregulated during DC differentiation in the presence of GM-CSF and IL-4 (74). Its role is potentially in suppressing B cell differentiation from a common precursor. Definitive evidence for the requirement of ID2 and NFIL3 in cDC1 development in humans is lacking and highlights the potential difficulties of translating TF-based definitions of DC subsets from mice to humans.

#### Function

The cDC1 subset is thought to be able to efficiently prime CD8<sup>+</sup> T cells through functional specializations such as cross-presentation of antigens and the production of IL-12p70 (78–80). This process is important in the induction of tumor immunity and the control of viral and bacterial infections when DCs are not the malignant cells or directly infected. The expression of Clec9A and XCR1 by both murine and human cDC1s supports this notion. cDC1s express a more limited TLR profile than cDC2s with high expression of TLR3 and TLR10 but without TLR4, –5, –7 and –9 (54, 81). TLR3 senses viral dsRNA but the role of TLR10 is currently unknown. Human cDC1s do not produce large amounts of IL-12p70 in response to TLR ligands alone but do following the combination of TLR ligands and CD40-CD40L signaling through activated T cells (71), in common with the finding in mice (82). IFN-λ is produced by murine and human cDC1 upon stimulation with the TLR3 agonist, poly I:C (83).

Murine cDC1s have an advantage over other subsets at crosspresentation of antigens by being able to (1) maintain optimal phagosomal pH for antigen processing (84) and (2) enhance the transfer of proteins from the endosome in to the cytosol so they can be loaded on to MHC Class I (85). This advantage is apparent when assessing cross-presentation of dead cell-derived antigens and upon stimulation with TLR3. However, recent data showed that murine cDC2 are also able to cross-present and cross-prime antigen upon stimulation with R848, a TLR7/8 agonist (86). In human, cDC1 appears to be superior at cross-presenting cell-derived antigen, particularly upon polyI:C stimulation (19–21, 71) and when antigens are delivered to late endosomes and lysosomes (87). However, in common with mice, cDC2 are also able to cross-present soluble antigen and long-peptide particularly upon R848 stimulation (88, 89). The variable findings reported may also be due to type of antigens used in the cross-presentation assays and the validity of comparing murine *in vivo* models with human *in vitro* assays.

In mouse, cDC1 preferentially induce Th1 immune response through IL-12p70 production (90, 91), although Th2 induction has also been reported (92). In human, both cDC1 and cDC2 have been shown to induce Th1 and Th2 responses (93). cDC1s were also shown to promote enhanced Th2 differentiation in response to TSLP in an influenza infection humanized mouse model (94). As most human experiments are performed using blood DCs and *in vitro*, it has been logistically difficult to establish pathogen and tissue-specific effects relevant for driving Th priming *in vivo*.

### cDC2

#### Phenotype

cDC2s in mice are lin– MHCIIhiCD11c<sup>+</sup>CD11b<sup>+</sup>. However, this fraction also includes monocyte-derived cells and macrophages (95). This is demonstrated by the variable depletion of cells from this fraction in Flt3 or Csf1r KO mice suggesting contamination by Flt3-independent cells (75). This is in contrast to the near complete absence of cDC1 in Flt3 KO mice (74).

Genetic tracing using Clec9A-reporter mouse to identify all CDP-derived cells demonstrated near-complete labeling of cDC1s but variable labeling of CD11b<sup>+</sup> DCs in NLT (68). Although this is in keeping with the presence of monocyte-derived cells and macrophages within CD11b<sup>+</sup> cells, it does not exclude the possibility of an alternative DC differentiation program that does not undergo a monocyte or CDP intermediate stage. Splenic CD11b<sup>+</sup> DCs are divided into an ESAMhi population that requires Notch2-, Flt3, and LTβ-signaling for its development and a monocyte-like ESAMloClec12A<sup>+</sup>CX3CR1<sup>+</sup> population that is Flt3-independent and expresses high levels of CD14, TNFα, CCR2, and Lyz2 (96). In murine lung, it has been possible to divide the MHCII<sup>+</sup>CD11c<sup>+</sup>CD11b<sup>+</sup> fraction into CD11b<sup>+</sup>CD64<sup>+</sup> monocyte-macrophage cells and CD11b<sup>+</sup>CD24<sup>+</sup> cDC2s (24). In murine skin, the MHCII<sup>+</sup>CD11b<sup>+</sup>Langerin– fraction comprises cDC2, monocyte-derived cells, and macrophages (97).

There is evidence that similar heterogeneity may be present within human cDC2. Only 170 genes characterized human cDC2, in comparison to 1020 for cDC1 and 1065 genes for pDCs (23). This limited list of differentially expressed genes predicts heterogeneity within the boundaries of the phenotype parameters used to define human cDC2, specifically a subpopulation derived from or closely related to another mononuclear phagocyte such as CD14<sup>+</sup> monocytes.

Human cDC2 (CD1c<sup>+</sup> DCs) are defined as lin– MHCII<sup>+</sup>CD14– CD16– CD11c<sup>+</sup>CD1c<sup>+</sup> cells, a definition they share with *in vitro* monocyte-derived DCs. Although human peripheral blood and murine cDC2 additionally express CD11b, CX3CR1, and SIRPα, these antigens do not distinguish them from monocyte-derived cells (24, 98). Uniquely in the small intestine, cDC2s co-express CD103 and SIRPα (24, 26). *In vitro* human mo-DCs express CD206/MMR and CD1a but peripheral blood cDC2 do not (47, 99). However, tissue CD1c<sup>+</sup> DCs express CD206 and CD1a (100, 101). In addition, some tissue CD1c<sup>+</sup> DCs co-express CD14 particularly during inflammation (47).

#### Homology

The transcriptional signatures of human blood CD1c<sup>+</sup> DCs are enriched with that of mouse spleen CD4<sup>+</sup>/CD11b<sup>+</sup> DCs (17, 23). In NLT, the transcriptional signatures of human small intestine CD103<sup>+</sup>SIRPα+ DCs and dermal CD1c<sup>+</sup> DCs are enriched with that of murine spleen and mesenteric lymph node CD11b<sup>+</sup> DCs and dermal CD11b+ DCs, respectively (25, 26). A similar relationship was also observed between murine lung CD11b<sup>+</sup> DCs with human blood CD1c<sup>+</sup> DCs (24).

#### Transcription Factors

cDC2 development has been shown to be dependent on the TFs Irf4, PU.1, RelB, and RBPJ (24, 96, 102–108). Irf4 directly supports MHC class II antigen presentation to promote CD4<sup>+</sup> T cell responses (109). In humans, CD1c<sup>+</sup> DCs express high amounts of IRF4 (24). Interestingly, IRF4 is also required for mo-DC differentiation, suggesting a shared differentiation program between cDC2 and mo-DC. PU.1 interacts with Irf4 but also upregulates Flt3 expression critical for early DC differentiation in mice (110, 111). The PU.1 binding site in the Flt3 promoter is conserved in mice and humans, and so it is thought to be similarly required for DC differentiation in humans (111). Administration of Flt3 results in expansion of DC subsets in lymphoid and non-lymphoid tissue (112). PU.1 mutations in humans and mice are associated with myeloid leukemias (113). Biallelic human IRF8 K108E mutation resulted in complete loss of monocytes, pDCs, cDC1, and cDC2 in the peripheral blood (60). Surprisingly, human autosomal dominant IRF8 T108A mutation results in selective loss of the cDC2 subset and IL-12 production (60). It is now apparent from studies on Irf8R294C(BXH2) and Irf8-/- mice that in addition to cDC1, pDCs and monocytes are also dependent on IRF8 (73, 75, 114–116). However, cDC2 frequency in mice with Irf8-/- and the hypomorphic mutation Irf8R294C are unaffected, in contrast to the findings in humans (73, 75, 114).

### Function

The transcriptome of cDC2s is enriched for genes related to antigen processing such as LAMP1, LAMP2, and cathepsins (117). Murine cDC2s have been shown to be able to promote Th17, Th2, and regulatory T cell responses depending upon the pathogen and antigen stimulus (24, 108, 118–120). This may be a consequence of their innate plasticity but could also relate to unresolved heterogeneity within murine cDC2. In human, cDC2 have been shown to induce Th17 differentiation (24).

Both human and mouse cDC2 share many transcriptional and functional similarities with monocyte-derived cells (24, 25, 47, 97). Both cDC2 and mo-DC are capable of promoting naïve CD4<sup>+</sup> and CD8<sup>+</sup> T cell proliferation and in mice cDC2 appear to be superior at trafficking to lymph nodes (97, 98), leading to the hypothesis that mo-DCs specialize in activating tissue-tropic T cells. Mo-DCs also produce higher levels of monocyte-attracting chemokines (CCL2, CCL7, CCL12) than cDC2s (98). Human blood cDC2 have a TLR expression profile that is close to murine lymphoid cDC2 with significantly higher levels of TLRs 2, 4, and 5 than other DC subsets (81), a profile it also shares with *in vitro* mo-DCs (121). The pathogenic role of cDC2 in human disease is not clear but they have been shown to accumulate in conditions such as RA (122), chronic kidney disease (123), and atopic airway inflammation (124), although their distinction from inflammatory mo-DCs is unclear. Human cDC2 are also implicated in the accumulation of CD103<sup>+</sup>CD8<sup>+</sup> mucosal T cells in the lung and promote fibrosis in the kidney through production of TGFβ (123, 125). Finally, human and mouse cDC2 share a similar cytokine production profile which includes IL-6, IL-23, and IL-1β (24, 81, 126, 127). In addition, unlike murine cDC2, human blood cDC2 can secrete high amounts of IL-12p70 upon in vitro stimulation with R848 and LPS, which was augmented in the presence of IFNγ and CD40L (89).

#### Plasmacytoid DCs (pDCs) Phenotype

Plasmacytoid DCs are specialized IFNα producing cells that were first described in human peripheral blood and tonsil (128–131). In blood, their morphology resembles that of lymphocytes but upon *in vitro* culture with IL-3 and CD40L, they acquire dendrites resembling myeloid DCs (129). pDCs are identified in mice by expression of CD11cintCD11b– B220<sup>+</sup> in combination with markers such as SiglecH and CD317 (BST2) to exclude a subset of NK cells and precursors of cDCs (132). In humans, they are identified by expression of CD123, CD303, and CD304. CD123 is the IL-3 receptor alpha chain and is also expressed on precursor cells, basophils, and eosinophils (133, 134). CD303 (BDCA-2) is a C-type lectin that is specifically expressed by human pDCs (135). Functionally, it has a role in antigen capture and when ligated it inhibits IFNα production (136). CD304 (BDCA-4) is uniquely expressed by pDC in peripheral blood but is also expressed by other cells such as endothelial cells (137).

### Homology

The relative distance of the pDC transcriptome from other leukocyte subsets and its conservation directly aligns murine and human pDCs (17). However, a subset of murine pDCs also appears to have cDC differentiation potential (138, 139), which has not been observed in human.

### Transcription Factors

Plasmacytoid DC development in humans and mice is dependent on the transcription factor E2-2 (140). E2-2 opposes default differentiation of precursors into cDCs and controls expression of a range of pDC-associated TFs, including SpiB, Irf7, and Irf8 (140, 141). In humans, haploinsufficiency of E2-2 results in Pitt-Hopkins syndrome, a condition with a range of features including developmental delay and characteristic facial features but without known clinical immunodeficiency (142). A population of CD45RA<sup>+</sup>CD123<sup>+</sup> cells is present in the blood of patients with Pitt-Hopkins syndrome but these cells fail to express CD303 and have severely reduced expression of IFNα, indicating that loss of E2-2 blocks full pDC differentiation (140). The transcription factor SpiB is required for IFNα production by pDCs in mice (143). SpiBknockdown in human CD34<sup>+</sup> HSCs inhibits pDC differentiation *in vitro* (144).

#### Function

Plasmacytoid DCs have a functional program that is wellconserved between mice and humans (145). In contrast to cDCs, pDCs express a narrow range of pattern recognition (146). Both mouse and human pDCs express TLR7 and TLR9 (146). TLR8 is expressed at very low amounts if any by human pDCs (81, 147, 148) and appears to have a different function in mice (146, 149, 150). pDCs in both mice and humans are specialized in the production of IFNα and thought to be important in viral immunity but also human autoimmunity such as SLE (151, 152).

### Monocytes and Monocyte-Derived Cells Phenotype

Two subsets of monocytes exist in mice and can be distinguished by the differential expression of Ly6C, CCR2, and CX3CR1. Similarly in humans, there are two monocyte subsets in peripheral blood identified by expression of CD14 and CD16 (CD14++CD16– and CD14<sup>+</sup>CD16<sup>+</sup>) as well as an intermediate phenotype (CD14++CD16<sup>+</sup>). In addition to these antigens, human monocytes are also heterogeneous for the expression of the angiopoietin receptor, Tie2, and 6-sulfoLanNAc(Slan), a carbohydrate modification of the P-selectin glycoprotein ligand-1 (PSGL-1) (153, 154).

#### Homology

Homology between peripheral blood monocyte subsets has been demonstrated by the extensive transcript enrichment between Ly6ChiCX3CR1lo and CD14++CD16 monocytes and between Ly6CloCX3CR1hi and CD16<sup>+</sup> monocytes (18, 23, 155).

#### Transcription Factors

The TFs that regulate the sequential differentiation of HSCs into MDP in mice include PU.1, Irf8, and Klf4 [reviewed in Ref. (156)]. PU.1 is required at each developmental bifurcation including HSC maintenance (157) and the generation of early myeloid progenitors (16, 158–160). Similarly in humans, PU.1 is required for monocyte differentiation from CD34<sup>+</sup> cord blood precursors (161). In murine monopoiesis, Irf8 and Klf4 act together to skew differentiation toward monocytes by antagonizing the granulocyte-supporting TF C/EBPα (115, 162). Consistent with this, human autosomal recessive Irf8 deficiency results in complete loss of circulating monocytes and DCs in the presence of neutrophilia (60).

The TFs that control cell-fate decisions downstream of MDP are less well defined. In mice, Irf5 and TCFEB are implicated during MDP to CMoP differentiation (55). The TF Nur77 has been implicated in Ly6CloCX3CR1hi monocyte generation (163).

PU.1 and MafB act antagonistically to support human monocyte differentiation into mo-DC and mo-Mac, respectively *in vitro* (164). Irf4 was also implicated in human *in vitro* mo-DC differentiation (165). Irf5 promotes the differentiation of classical/M1 macrophages from human monocytes *in vitro* (166). In contrast, Irf4 activates transcription of the alternative/M2 macrophage markers in mice (167) and humans (168, 169).

#### Function

CD14<sup>+</sup> human and Ly6ChiCX3CR1lo murine monocytes can exhibit considerable functional plasticity as demonstrated by their acquisition of DC-like and macrophage-like characteristics *in vitro* and *in vivo*. Recent fate mapping studies have demonstrated that monocytes do not contribute to tissue-resident macrophages in the steady state (12, 14, 15), with the notable exception of gut and dermal macrophages (14, 97, 170). However, monocytes can give rise to tissue macrophage-like cells in inflammation (35, 171). Monocytes can also differentiate into DC-like cells in the steady state in mucosal tissues and skin (97, 172). This process is enhanced during inflammation (97, 98, 173), including infections with Leishmania (34), Influenza (174), Trypanosoma (175), Listeria (33), and pulmonary *Aspergillus* (176). Alternatively, rather than DC-like or macrophage-like differentiation, monocytes may remain as tissue monocytes upon extravasation (177).

CD14<sup>+</sup>CD16<sup>+</sup> intermediate and CD16<sup>+</sup> non-classical monocytes are expanded in multiple disease, infection, and inflammatory states (178). CD16<sup>+</sup> monocytes "patrol" the endothelium *in vivo*, are weak phagocytes, and sense nucleic acids and viruses via TLR7 and 8 receptors (155). Additional heterogeneity has been reported within human monocytes. Tie2<sup>+</sup> monocytes are associated with angiogenesis and Slan<sup>+</sup>CD16<sup>+</sup> cells, which are also present in inflamed skin, are potent producers of TNF α, IL-1β, and IL-12 (179, 180). Monocyte-derived dermal CD14<sup>+</sup> cells express IL-1α (25) have been shown to induce differentiation of follicular helper T cells (126) and provide direct B cell help (181).

#### Langerhans Cells

Langerhans cells are located in epidermal surfaces such as skin and are characterized by the presence of cytoplasmic organelles containing Langerin called Birbeck granules (182). The function of these organelles is unclear but their absence does not affect their capacity to process and present antigen (183). LCs form a dynamic network with adjacent keratinocytes and protrude dendrites through tight junctions to pick up antigens that have passed the stratum corneum barrier (184). The easy accessibility of LCs and their functional plasticity has generated significant interest in targeting them for vaccination strategies (185).

In the steady state, LCs are maintained independently of the bone marrow through local self-renewal (186–188). Human LCs can proliferate *in situ* and have been shown to remain donor in origin up to 10 years after limb transplant (189–191). During inflammation, LCs can be replaced by circulating precursors. The identity of the circulating LC precursor remains unclear. In mice, there appears to be two waves of replenishment with monocytes in the first wave giving rise to short-term LCs that retain some monocyte features and an as yet unknown CD34<sup>+</sup> HSC-derived precursor that gives rise to long-term LCs (186, 187). In humans, CD1c<sup>+</sup> DCs are able to upregulate langerin and CD1a, a phenotype resembling LCs, upon *in vitro* culture with TSLP and TGFβ or GM-CSF and BMP7, but the relevance of this to *in vivo* LC differentiation is uncertain (192, 193). Although human LCs can self-renew locally after BMT, they are replaced by donor-derived cells, even after non-myeloablative transplant conditioning (194–196).

Langerhans cells are developmentally independent of Flt3 but dependent on Csf1r. However, it is IL-34 signaling through Csf1r, rather than Csf1, that is critical for LC development and maintenance (197). IL-34 is also expressed in human skin but the dependence of human LCs on this cytokine remains untested.

#### Phenotype

Human and murine LCs are CD11clo, langerinhi, EPCAM<sup>+</sup>, and also characterized by the presence of cytoplasmic Birbeck granules (198). In human, LCs are additionally CD1ahi and CD1c<sup>+</sup> (23, 199).

#### Homology

The homology between LCs in humans and mouse is obvious given their exclusive anatomical occupancy and shared expression of langerin, EPCAM, and presence of Birbeck granules. Comparative transcriptomic analysis of human and mouse LCs has never been performed.

#### Transcription Factors

Langerhans cell development is dependent on PU.1, Runx3, and Id2, although the latter may be dispensable for bone marrowderived LCs (74, 188, 200, 201).

#### Function

Langerhans cells are able to induce different immune responses depending on the context. Depletion of murine LCs can either exacerbate or suppress contact hypersensitivity immune response [reviewed in Ref. (202)]. In a mouse model of graft versus host disease (GVHD), LCs neither primed CD8<sup>+</sup> T cells nor programed their homing to the epidermis but were required for their effector function *in situ* (203). This is consistent with their inability to cross-present antigen *in vivo* (80, 204), although cross-presentation has been reported using *in vitro* assays (205). In mice, LCs appear to be critical for Th17 response against the yeast form of *Candida albicans* in the epidermis through engagement of Dectin-1 and their subsequent production of IL-6 (206). In humans, failure to generate effective Th17 responses (as a result of a range of mutations in, for example, IL-17RA, IL-17F, STAT1 genes) can result in chronic mucocutaneous candidiasis (CMC) (207). However, it is unclear if immunity against *Candida* infections in the skin in healthy individuals is dependent upon LCs. Notably, human LCs do not appear to express Dectin-1, which is important for *Candida* recognition (208). *In vitro* human LCs appear versatile and are capable of generating Th1, Th2 (209), Th17 (210), Th22 (211), and Treg (212) responses depending on the experimental conditions used.

#### Macrophages

Macrophages are a diverse population of tissue-resident cells with roles in inflammation, tissue homeostasis, and repair. Macrophage identity and function can be influenced by three variables: (1) resident tissue environment; (2) exposure to activation signals; and (3) ontogeny (monocyte- vs. prenatal precursor-derived) [reviewed in Ref. (3)].

The nomenclature of macrophages is based upon their tissue of origin [for example, Kupffer cells (liver), osteoclasts (bone), and microglia (CNS)]. This is in recognition of the central influence of environment on their phenotype and function. Examples of these functional specializations include breakdown of RBCs (Kupffer cells and splenic macrophages), bone resorption (osteoclasts), gut peristalsis (muscularis macrophages), and neural network development and maintenance (microglia) (213–215). Although macrophages in the vast majority of tissues, except dermis and the lamina propria, are prenatally derived, their preservation into adulthood by self-renewal is variable by site and in the presence of inflammation [(15, 216, 217) and reviewed in Ref. (218)]. The relative preservation of dermal macrophages and LCs in patients lacking circulating blood monocytes and DCs due to heterozygote GATA2 and biallelic IRF8 deficiencies supports a prenatal origin of some human macrophages (59, 60).

Microarray transcriptome analysis has identified several thousand transcripts with greater than twofold difference in expression between macrophages from different sites in mice (219), supporting unique local microenvironment-related characteristics. These tissue specific transcripts are more prominent within macrophages than DCs (219) and may reflect the tissue-resident nature of macrophages. The impact and underlying mechanisms of environmental regulation on macrophages was elegantly demonstrated by the unique epigenetic modulation of macrophage in distinct tissues and the ability of macrophages from one environment to develop the characteristics of their counterparts in another tissue (220, 221).

#### Phenotype

Murine macrophages express the antigens CD11b, CD68, CSF1R, and F4/80 (215). With the exception of F4/80 which is predominantly expressed on eosinophils (222), these antigens are also expressed on human macrophages (223). Furthermore, human alveolar macrophages were shown to express many antigens, which are conserved at transcript level with murine bone marrow-derived macrophages (163).

#### Homology

Comparative analysis between human and mouse macrophage populations has been poorly studied. In skin, homologous monocyte-derived dermal macrophage populations have been identified (25) but the murine counterparts of human dermal macrophages containing melanin-granules (melanophages) remain uncertain. While a range of transcriptional analyses of human macrophage populations in health and disease have been performed, comparisons between human tissues and across species have not been rigorously undertaken (224).

#### Transcription Factors

The transcriptional requirements of murine YS-derived macrophages differ to those of HSC-derived macrophages. YS-derived microglia require PU.1 and Irf8 but are independent of Myb, Id2, Batf3, and Klf4 (2, 12, 225). Consistent with macrophage tissue specializations, additional TFs such NFATc1 and Spi-C have been shown to be required for osteoclasts and splenic and bone marrow macrophage differentiation, respectively (226–228).

#### Function

The M1/M2 paradigm has been described to model the diverse programs of macrophage activation but has largely relied on

*in vitro* generated macrophages. This has provided a useful tool to examine macrophage activation in the absence of tissue-specific effects. More recently, a spectrum of responses, with M1 and M2 being two poles of a continuum that is transcriptionally apparent, were identified (229). It is unclear how closely human and murine macrophages are aligned in response to a similar range of stimuli. There are inter-species differences in the response to a single stimulus (LPS) between human and mouse *in vitro* derived macrophages; INOS transcript is preferentially induced in mouse but human macrophages characteristically upregulate CCL20, CXCL13, IL-7R, P2RX7, and STAT4 (230).

#### Mononuclear Phagocytes in Inflammation

Classical Ly6ChiCX3CR1lo monocytes infiltrate inflamed tissues where they can acquire either DC or macrophage properties (33, 231). This *in vivo* process (thought to be analogous to *in vitro* mo-Mac and mo-DC differentiation) can be influenced by local microbiota (97, 98, 170, 171). In infection and disease, monocytederived cells accumulate in greater numbers in a broad range of tissues [reviewed in Ref. (232)]. In many such models of infection, they are non-redundant and required for clearance of pathogens by promoting protective Th1 and Th17 responses (34, 233, 234). This suggests that despite shared functions with resident conventional DCs, there are important differences that require the presence of monocyte-derived cells to overcome infection. In murine experimental autoimmune encephalomyelitis, monocytes infiltrate the CNS but are not long-lived and following resolution do not contribute to the microglial pool (231). Analysis of murine Kupffer cells suggests functional heterogeneity between resident and recruited populations (235).

Snapshot analysis of inflamed human tissue similarly reveals additional subsets that are not present in health [(47, 99, 179, 236, 237) and reviewed in Ref. (31)]. These include inflammatory dendritic epidermal cells (IDECs) found in atopic dermatitis, TNF, and iNOS producing DCs (Tip DCs) and slan DCs, found in psoriasis (99, 179, 236, 237). In rheumatoid arthritis synovial

### References


fluid and malignant ascites, there is an accumulation of cells that express overlapping markers with blood CD1c<sup>+</sup> DCs but additionally express CD1a, CD206, SIRPα, and CD14 (47). Monocytes can acquire DC characteristics when cultured with *ex vivo* GM-CSF-primed synovial T cells, which potentially suggests a mechanism for their generation (238). Histiocytes are pathological MPs expressing CD68 and CD163. It is unknown if these cells, often found in granulomas, arise from resident macrophages or are monocyte-derived. Further studies are required to establish the *in vivo* differentiation requirements of inflammatory MP populations and how they contribute to disease.

# Conclusion

In this review, we have discussed the parallel organization of the MPS between humans and mice. We demonstrate the use of comparative biology approaches as both a validation and discovery tool to dissect the development and functional heterogeneity of mononuclear phagocytes in a reciprocal manner across the two species. The incorporation of high-dimensional unbiased single-cell genomics and proteomics technologies will facilitate the interrogation of functionally relevant populations with indiscrete phenotypes and validate current definitions of cell-types based on limited antigen expression profile particularly during inflammation. This combined strategy will accelerate the translation of fundamental MPS biology to clinical benefit through enhanced understanding of the pathomechanisms of disease and facilitate the development of novel approaches in vaccination and cancer immunotherapy.

# Acknowledgments

MH and GR are funded by The Wellcome Trust, UK [Intermediate Clinical Fellowship; WT088555A (MH) and Clinical Research Training Fellowship; WT098914MA (GR)]. The authors wish to acknowledge Matthew Collin for assistance with illustration.


with distinct cytokine profiles. *Am J Pathol* (2008) **172**:940–50. doi:10.2353/ ajpath.2008.070703


tissues and transport antigen to lymph nodes. *Immunity* (2013) **39**:599–610. doi:10.1016/j.immuni.2013.08.007


in primary human versus mouse macrophages. *Proc Natl Acad Sci U S A* (2012) **109**:E944–53. doi:10.1073/pnas.1110156109


**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 Reynolds and Haniffa. 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.*

*Edited by:* 

*Shalin Naik, Walter and Eliza Hall Institute, Australia*

*Reviewed by:* 

*Christophe Jean Desmet, University of Liege, Belgium Richard A. Kroczek, Robert Koch-Institute, Germany*

#### *\*Correspondence:*

*Thien-Phong Vu Manh and Marc Dalod, Centre d'Immunologie de Marseille-Luminy, Parc Scientifique et Technologique de Luminy, Case 906, Marseille Cedex 9 F-13288, France vumanh@ciml.univ-mrs.fr; dalod@ciml.univ-mr.fr Isabelle Schwartz-Cornil, UR892, Virologie et Immunologie Moléculaires, INRA, Domaine de Vilvert, Jouy-en-Josas Cedex 78352, France isabelle.schwartz@jouy.inra.fr*

#### *† Present address:*

*Jamila Elhmouzi-Younes, CEA, Division of Immuno-Virology, IDMIT Center, Institute for Emerging Diseases and Innovative Therapies (iMETI), DSV, Fontenay-aux-Roses, France*

*‡ Thien-Phong Vu Manh and Jamila Elhmouzi-Younes have contributed equally to this work. § Senior co-authorship*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 20 March 2015 Accepted: 25 May 2015 Published: 19 June 2015*

#### *Citation:*

*Vu Manh T-P, Elhmouzi-Younes J, Urien C, Ruscanu S, Jouneau L, Bourge M, Moroldo M, Foucras G, Salmon H, Marty H, Quéré P, Bertho N, Boudinot P, Dalod M and Schwartz-Cornil I (2015) Defining mononuclear phagocyte subset homology across several distant warm-blooded vertebrates through comparative transcriptomics. Front. Immunol. 6:299. doi: 10.3389/fimmu.2015.00299*

# Defining mononuclear phagocyte subset homology across several distant warm-blooded vertebrates through comparative transcriptomics

*Thien-Phong Vu Manh1,2,3 \* ‡ , Jamila Elhmouzi-Younes4 †‡ , Céline Urien4 , Suzana Ruscanu4 , Luc Jouneau4 , Mickaël Bourge5 , Marco Moroldo6 , Gilles Foucras7,8 , Henri Salmon9,10, Hélène Marty 9,10, Pascale Quéré9,10, Nicolas Bertho4 , Pierre Boudinot <sup>4</sup> , Marc Dalod1,2,3 \*§ and Isabelle Schwartz-Cornil4 \* §*

*1UM2, Centre d'Immunologie de Marseille-Luminy, Aix Marseille Université, Marseille, France, 2U1104, INSERM, Marseille, France, 3UMR7280, CNRS, Marseille, France, 4UR892, Virologie et Immunologie Moléculaires, INRA, Domaine de Vilvert, Jouy-en-Josas, France, 5 IFR87 La Plante et son Environnement, IMAGIF CNRS, Gif-sur-Yvette, France, 6CRB GADIE, Génétique Animale et Biologie Intégrative, INRA, Domaine de Vilvert, Jouy-en-Josas, France, 7UMR1225, Université de Toulouse, INPT, ENVT, Toulouse, France, 8UMR1225, Interactions Hôtes-Agents Pathogènes, INRA, Toulouse, France, 9UMR1282, Infectiologie et Santé Publique, INRA, Nouzilly, France, 10UMR1282, Université François Rabelais de Tours, Tours, France*

Mononuclear phagocytes are organized in a complex system of ontogenetically and functionally distinct subsets, that has been best described in mouse and to some extent in human. Identification of homologous mononuclear phagocyte subsets in other vertebrate species of biomedical, economic, and environmental interest is needed to improve our knowledge in physiologic and physio-pathologic processes, and to design intervention strategies against a variety of diseases, including zoonotic infections. We developed a streamlined approach combining refined cell sorting and integrated comparative transcriptomics analyses which revealed conservation of the mononuclear phagocyte organization across human, mouse, sheep, pigs and, in some respect, chicken. This strategy should help democratizing the use of omics analyses for the identification and study of cell types across tissues and species. Moreover, we identified conserved gene signatures that enable robust identification and universal definition of these cell types. We identified new evolutionarily conserved gene candidates and gene interaction networks for the molecular regulation of the development or functions of these cell types, as well as conserved surface candidates for refined subset phenotyping throughout species. A phylogenetic analysis revealed that orthologous genes of the conserved signatures exist in teleost fishes and apparently not in Lamprey.

Keywords: comparative biology, immunology, dendritic cells, monocytes, macrophages, genomic and bio-informatic methods

# Introduction

Reaching the global health objective requires to improve disease prevention and treatments in humans and in a wide variety of animal species. To achieve that goal, knowledge of the immune system, and particularly of the mononuclear phagocyte system that orchestrates the immune response, needs to be translated across species in order to develop better vaccines and immune response-targeting therapies in relevant species.

The mononuclear phagocytes encompass three main functional cell types: monocytes (Mo), macrophages (MP), and DC. The main functions of Mo are to patrol the body to detect infections and to produce microbicidal compounds including TNF, superoxide, or nitric oxide intermediates, or to differentiate into MP. The main function of MP is to preserve tissue homeostasis through trophic and scavenger functions. DCs are professional antigen-presenting cells that are key instructors of immunity, controlling tolerance to self and immune defense against pathogens. However, beyond these generic definitions, each of these mononuclear phagocyte category encompasses a complex array of different subtypes with distinct ontogeny and functions, as described extensively in mice and to some extent in humans. Mo include at least two main subsets, classical Mo (cMo) and non-classical Mo (ncMo) (1), that express different innate immune recognition receptors and mediate distinct functions, with ncMo showing the original property of patrolling blood vessels (2). Adult MP are derived either from embryonic precursors and self-renew in tissues, or in some cases are replenished from circulating Mo (2–6). The MP subtypes populating different tissues show distinct molecular and functional characteristics which are in a large part determined by their anatomical microenvironment (7, 8). Two cell types with morphologic and functional features of DC derive from the Mo/MP lineage, namely monocyte-derived DC (MoDC) and Langerhans cells (9). MoDC are generated (i) upon inflammatory stimuli *in vivo* (10), (ii) at steady-state in the skin (3), and (iii) upon culture of purified Mo or of total bone marrow cells with GM-CSF ± IL-4 *in vitro* (11, 12). Langerhans cells derive from embryonic monocytic precursors upon IL-34 signaling and populate the outer layer of epithelia (13). Finally, three types of *bona fide* DC exist, the plasmacytoid DC (pDC) and the conventional DC (cDC) cDC1 and cDC2 types which derive from a bone marrow common DC precursor and are present both in lymphoid organs and as interstitial DC in the parenchyma of non-lymphoid tissues such as skin, lung, gut, and liver (14). Comparative transcriptomic analyses pioneered by us and used by other groups, as well as functional studies, have demonstrated the existence of similar mononuclear phagocytes and DC subsets between human and mice (15–20). DC subset candidates have also been described in other mammals such as in ruminants and pigs. However, no systematic study has demonstrated the existence of a framework of homologous DC subsets throughout distant species [for review see Ref. (21)]. Overall, it remains unknown whether a similar diversity in mononuclear phagocyte subsets exists across distant mammals and vertebrates, and when during evolution this complex organization of the mononuclear phagocyte system arose.

The combination of phenotypic, functional, and ontogenic studies used in the mouse model cannot be used to define cell subsets in most other species of interest due to technical, financial, or ethical limitations. As the ontogeny and functions of cell types are instructed by specific gene-expression modules, cell type identity can be defined by its molecular fingerprinting (22). We thus reasoned that mononuclear phagocyte subset identity could be defined by gene-expression profiling, whatever the species. In addition, cell types that are homologous between species must exhibit closer molecular fingerprints and gene-expression programs than non-homologous cell types, based on the definition of homologous cell types as "those cells that evolved from the same precursor cell type in the last common ancestor" (23).

In this paper, we developed a streamlined approach (see Figure S1 in Supplementary Material) to identify homologous mononuclear phagocyte subsets in distant species with reference to the mouse, consisting in (i) designing antibody panels for sorting candidate cell subsets to high marker-based purity, (ii) generating genome expression profiling of the sorted cell subsets, and (iii) performing computational transcriptomic analyses to establish gene signatures and compare them to the transcriptomic fingerprints of the well-characterized immune cell types of the mouse referent species. Our analysis was extended to chicken cell subsets, showing that it is amenable to establish mononuclear phagocyte subset homology throughout vertebrates. We also derived gene-expression signatures and gene interaction networks that are selectively expressed in mononuclear phagocyte subsets in a conserved manner throughout distant mammals and that can be used to identify homologous subsets throughout species. The conserved gene-expression signatures and networks not only encompassed genes with known functions in mononuclear phagocyte subsets but also pointed out novel candidate genes likely involved in the ontogeny or functional specialization of these cell types. Finally, we conducted a phylogenetic analysis to examine the presence in bony fishes and in Lamprey of orthologs of genes from the transcriptomic signatures identified in mammals.

# Materials and Methods

# Pigs and Sheep for Blood Collection

All animal experiments were carried out under licenses issued by the Direction of the Veterinary Services of Versailles (accreditation numbers B78-93) and under approval of the Committee on the Ethics of Animal Experiments of AgroParisTech and INRA-Jouyen-Josas (COMETHEA, authorization number 00604.01). The eight pigs (blood) used in this study (four males, four females) were around 2 years old and weighted between 60 and 85 kg. Down-sized pigs were kept at the Centre d'Imagerie Interventionnelle (Jouyen-Josas). «Prealpe» female sheep (total 37, 50–80 kg), originate from and were raised in the «Unité Commune d'Expérimentation Animale» in Jouy-en-Josas, France. Blood (<400 ml/animal) was collected by venous puncture on sodium citrate.

# Isolation of DC Subset Candidates, B Lymphocytes, and Mo from Pig Blood

PBMC were obtained from pig peripheral blood buffy coat samples by 1.076 g/ml density Percoll (GE Healthcare) gradient centrifugation (24). For B cell sorting, PBMC were surface-labeled with 2 μg/ml primary monoclonal antibody (mAb) against IgL (K139 3E1, IgG2a) followed by Alexa647-conjugated goat anti-mouse IgG2a antibodies (Invitrogen). For pDC sorting, PBMC were surface-labeled with 2 μg/ml primary mAb anti-pig CD4 (PT90A, IgG2a), CD3 (8E6, IgG1), CD14 (CAM36, IgG1), and CD172A (74-22-15, IgG2b) followed by Alexa488, phycoerythrin (PE), or Alexa647-conjugated goat anti-mouse isotype-specific antibodies (Invitrogen). Blood pDC candidates were sorted as CD3<sup>−</sup> CD14<sup>−</sup> CD4<sup>+</sup> CD172int cells, based on previously published indicative data (25). For cDC candidates and Mo sorting, PBMC were surfacelabeled with 2 μg/ml mAb anti-pig IgL (K139 3E1, IgG2a), anti-pig IgG (K138 4C2, IgM), anti-pig IgM (PG145A, IgM), anti-pig CD4 (PT90A, IgG2a), anti-human and pig cross-reacting CD14 (TUK4, IgG2a), anti-pig CD172A (74-22-15, IgG1), anti-artiodactyl MHC class II (Th21A, IgG2b), and chicken anti-human and artiodactyl cross-reacting CADM1 (3E1, IgY). The primary antibodies were revealed with Alexa488, PE, or Alexa647-conjugated goat antimouse isotype-specific antibodies and with donkey anti-chicken IgY Peridinin Chlorophyll Protein Complex (PerCP)-conjugated IgG. The cDC2 candidates were isolated as FSChi IgL<sup>−</sup> IgG<sup>−</sup> IgM<sup>−</sup> CD4<sup>−</sup> CD14<sup>−</sup> MHC class II<sup>+</sup> CADM1<sup>−</sup> CD172hi or CD172int cells. The cDC1 candidates were isolated as FSChi IgL<sup>−</sup> IgG<sup>−</sup> IgM<sup>−</sup> CD4<sup>−</sup> CD14<sup>−</sup> MHC class II<sup>+</sup> CADM1<sup>+</sup> CD172lo cells. Mo candidates were sorted as MHC class II<sup>−</sup> CD172hi cells. Non-relevant antibodies (IgG1, IgG2a, IgG2b, and IgM) were systematically used as controls to measure the level of non-specific background signal caused by primary antibodies. The cell subsets were sorted by flow cytometry on the ImaGif Cytometry platform using the analyzer-sorter MoFlo XDP cytometer and the Summit 5.2 software from Beckman Coulter (cytometric assessment of post-sort purity >98%). The numbers of DCs that were collected per pig lay between 2 and 3 × 105 for pDC, 25 and 47 × 103 for cDC1, 20 and 40 × 105 for cDC2 candidates.

#### Isolation of DC Subset Candidates from Sheep Blood and B Lymphocytes and Macrophages from Sheep Spleen

Sheep PBMC were loaded on 1.065 density iodixanol gradient (Optiprep, Nycomed Pharma) to isolate low density cells from blood. Sheep pDC candidates were isolated by flow cytometry as previously described (26). For isolating sheep cDC candidates, the low density PBMC from several sheep were reacted with anti-CD11c mAb (2 μg/ml, OM1 clone, IgG1) followed by a saturating concentration of pacific blue-labeled anti-mouse IgG donkey Fab (50 μg/ml). After extensive wash, cells were further incubated anti-CD172A mAb (2 μg/ml, ILA24, IgG1) followed by a saturating concentration of Alexa488-labeled anti-mouse IgG donkey Fab (50 μg/ml). After extensive wash, cells were incubated with 2 μg/ml primary mAbs anti-ruminant B cells (DU-204, IgM), CD11b (ILA130, IgG2a), TCR1γ/δ receptor (CC15, IgG2a), CD45RB (CC76, IgG1), and chicken anti-human and artiodactyl cross-reacting CADM1 (3E1, IgY). The IgM and IgG2a primary antibodies were revealed with PE-conjugated goat anti-mouse isotype-specific antibodies, the IgG1 anti-CD45RB primary antibody was revealed with Alexa647-conjugated goat anti-mouse IgG1 antibody, and the anti-CADM1 with anti-IgY PerCP-conjugated IgG. The cDC2 candidates were isolated by flow cytometry as B<sup>−</sup> CD11b<sup>−</sup> TCR1<sup>−</sup> CD45RB<sup>−</sup> CD11c<sup>+</sup> CADM1lo CD172hi FSChi cells. The cDC1 candidates were isolated by flow cytometry as B<sup>−</sup> CD11b<sup>−</sup> TCR1<sup>−</sup> CD45RB<sup>−</sup> CD11c<sup>+</sup> CADM1hi CD172lo FSChi cells. The numbers of DCs that were collected per sheep lay between 1 and 2 × 105 for pDC, around 600 for cDC1, and around 4000 for cDC2. The far lower amounts of collected blood cDCs from sheep as compared to pig may probably originate from the multiple staining steps due to the necessity to separately identify several IgG1 as primary antibodies. B cells and MP were sorted by flow cytometry from isolated sheep spleen cells using the anti-ruminant B cell (DU-204, IgM) and anti-CD14 (CAM36, IgG1), respectively.

# Production of Sheep MoDC

Three independent cultures of sheep MoDC were produced with GM-CSF as previously described (27).

# RNA Extraction and Hybridization on Microarrays

Total RNA from subsets was extracted using the Arcturus PicoPure RNA Isolation Kit (Arcturus Life Technologies) and checked for quality with an Agilent 2100 Bioanalyzer using RNA 6000 Nano or Pico Kits (Agilent Technologies). All RNA samples had an RNA integrity number (RIN) above 8.5. When insufficient total RNA amounts for hydridization were obtained (<25 ng for sheep DNA chips, <50 ng for pig DNA chips), the RNAs from the sorted subsets of distinct animal were mixed. RNA amplification and labeling was performed using the onecolor Low Input Quick Amp Labeling kit (Agilent Technologies) following the manufacturer's recommendations. Each RNA sample (25 ng for sheep and 50 ng for pig) was amplified and cyanin 3 (Cy3) labeled, and subsequently the complementary RNA (cRNA) was checked for quality on a Nanodrop and on an Agilent 2100 Bioanalyzer. The cRNAs (600 ng) were fragmented and used for hybridization on custom-designed Agilent ovine and porcine arrays. Our arrays for sheep and pig were customdesigned based on the commercial ovine Agilent arrays for these two species, as previously described (28, 29). In brief, the commercial probes with poor Sigreannot scores (30) were replaced with new probes designed using the e-array software from Agilent Technologies and including ovine or porcine orthologs of genes known to be selectively expressed in human and mouse DC subsets (15). After hybridization of the cRNAs on the custom-designed ovine array, the chips were washed according to the manufacturer's protocol and scanned using a G2565CA scanner (Agilent Technologies) at the resolution of 3 μm. The resulting .tiff images were extracted using the Feature Extraction software v10.7.3.1 (Agilent Technologies), using the GE1\_107\_Sep09 protocol. All the protocols used can be obtained by contacting the CRB GADIE facility1 . The transcriptomic data from the chicken immune subsets were obtained from a previous study (31). All microarray data have been deposited in the Gene Expression Omnibus (GEO) database under reference numbers GSE9810, GSE53500, GSE55642 which have already been released and GSE66311 which is under embargo until publication of the present study.

<sup>1</sup>http://crb-gadie.inra.fr/

# Computational Pipeline to Assess Cell Subset Homology Across Species

We have designed a computational pipeline in order to define cell subset homology across species, based on the analysis of gene-expression microarray data. In the current study, it has been applied to identify homology relationships between mononuclear phagocyte cells in mammalian species and then extended to the comparison with a more distant species (chicken). However, it can be applied to any cell type and to any species, provided that the annotations of the genes for each species are sufficiently well documented to allow the retrieval of the orthologous genes. In order to perform the comparison of expression profiles of cells coming from different species, thus from different platforms, we have designed two independent procedures. The first procedure (Figure S2A in Supplementary Material) is based on the assessment of the conservation of cell-specific fingerprints/signatures, as assessed by performing gene set enrichment analyses (GSEA, see below) between pairs of cell types. The second procedure (Figure S2B in Supplementary Material, see below) consists in cross-normalizing the expression datasets coming from the different species, in order to simultaneously examine the relationships between all cell types together.

## Cross-Species Transcriptome Comparison by Pairwise Gene Set Enrichment Analyses

The methodological pipeline is depicted in Figure S2A in Supplementary Material, based on an example with comparison of three different species (A, B, and C). Species A is the reference species, i.e., the species for which the cell types are the most accurately described and generally also for which gene orthologous relationships can be retrieved from (mouse or human here). Species B and C are the test species, i.e., the species for which the identity of the cell types has to be established. Coming from three different platforms, the expression datasets have different numbers of probes, illustrated by boxes of different size. In brief, the strategy is to examine by GSEA whether the transcriptomic fingerprint of a given cell type (X) from the referent species A is enriched in one cell type (Y) of a test species (B for example) as compared to all other cell types of the same species. If this is the case, this would support the hypothesis that the cell type Y from the test species B is homologous to the cell type X of the referent species A. To perform these high-throughput GSEA in a processive way that could be easily reproduced and interpreted by other researchers devoid of bio-informatics expertise, we designed and implemented a dedicated software, called Bubble GUM (manuscript in preparation in which an extensive description of the software will be provided)2 . Bubble GUM encompasses two main modules, GeneSign and BubbleMap, respectively, dedicated to the generation of gene sets and to their use for GSEA applied to multiple pairwise comparisons of samples integrated together into a simple graphical output that helps in the interpretation of the results. The first step consists in extracting from the reference species the transcriptomic fingerprints of each cell type. A cell-specific transcriptomic fingerprint can be defined as the list of genes that are more highly expressed in the cell type of interest than in all other cell types. These fingerprints were extracted using the "Min (test) vs. Max (ref)" method [(minimum expression among all replicates for all samples for which the transcriptomic fingerprint is defined/maximum expression among all replicates used as reference) ≥1.5-fold] (15, 32), using the GeneSign module of Bubble GUM. These transcriptomic fingerprints, in gene symbol format, will be assessed for enrichment on the expression datasets of species B and C. Thus, it is necessary to convert the probe annotations from the arrays of species B and C into the gene symbol of their orthologous counterparts in species A. For this purpose, we used the orthology relationships defined by the Sigenae pipeline which annotated the pig and sheep genes with their human and mouse orthologous gene symbols (30). The genes present on the gene chips of species B and C that were not associated to an orthologous counterpart in species A remained annotated with the gene symbol corresponding to their species of origin. The statistical enrichment of the cell-specific transcriptomic fingerprints extracted from the reference species A were then calculated between pairwise comparisons of cell types from species B or C with the GSEA methodology, using gene set permutations for computing the *p*-values and false-discovery rates (FDR) (33). This was achieved, and the results graphically represented, by using the BubbleMap module of Bubble GUM.

## Cross-Normalization of the Species-Specific Expression Datasets

Using the same starting expression datasets, this is an alternative strategy which is complementary to the pairwise GSEA of the species-specific expression datasets, since it allows clustering all cell types together based on the overall evaluation of the proximity of their expression patterns of hundreds to thousands of orthologous genes. The first step (Orthology Filter) consists in aligning the genes across the species (A, B, and C). It requires retaining only one representative probe per gene for each species/platform. This is needed since, in microarray designs, many genes are often each represented several times by a number of individual probes having each a different signal-to-noise ratio. However, probes have no equivalence across species, whereas genes do. In our experience, the signal-to-noise ratio is generally better for probes that have the strongest signal in positive control samples, while certain probes that have a low signal-to-noise ratio can give misleading high fold changes across conditions when using a limited number of replicates. Hence, we computed for each probe in each platform the sum of normalized expression values across all samples and kept for each gene the probe that had the highest computed value. Then, for the genes of species B and C, we retrieved the gene symbol of their orthologous counterparts in species A (reference species). In the example illustrated in Figure S2B in Supplementary Material, species A is the reference: the genes of species B and C are thus annotated using the gene symbol of the orthologous genes in the species A. The genes not represented in each of the gene chip platform used were removed from the analysis. This Orthology Filter yielded a filtered expression dataset for each species, where the number of genes and their associated symbols were similar between all species, as illustrated by boxes of the same size (Figure S2B in Supplementary Material). In order to be able

<sup>2</sup>http://www.ciml.univ-mrs.fr/applications/BubbleGUM/index.html

to rigorously merge the different datasets together, the dynamic ranges of expression values for each gene across all species must be homogenized by setting for each dataset and for each gene across all samples the mean expression to 0 and the variance to 1, a process called data centering and reduction. To prevent this mathematical transformation of the expression data to introduce noise by forcing artifactual expression changes for genes that were unregulated in the initial datasets, it is mandatory to remove all the genes that are not regulated in at least one of the datasets. This thus requires keeping only the genes that are differentially expressed between at least two cell types for each of the species studied. This was achieved in the second and third steps of the data processing. The second step (Differentially Expressed Gene, DEG, Filter) consisted in identifying independently in each dataset the genes that are differentially expressed between at least two cell types. The identification of DEG was performed by calculating the minimal ratio between each pairwise comparison of cell types and by selecting only the genes for which this minimal ratio was higher than twofold. The third step (DEG intersection) consisted in keeping only the genes that were common to all filtered DEG lists, i.e., the orthologous genes which expression was modulated across samples in each of the species studied. The fourth step consisted in data centering and reduction for each dataset, which was performed using the R statistical environment. This step consists in setting, for each dataset, the mean to 0 and the variance to 1, so that all datasets are comparable. In the fifth and final step, the different datasets were merged together simply by aligning their rows based on the common gene symbol extracted from species A. The final cross-normalized expression dataset including the data for all species was then used to perform canonical analyses for classification of samples, namely here hierarchical clustering.

## Generation of Conserved Cross-Species Cell Type-Specific Signatures

For each species (human, mouse, sheep, and pig), the transcriptomic fingerprint of each cell type was generated by selecting the genes more highly expressed in the cell type of interest, as compared to all other studied cell types of the same species in the case of "absolute" transcriptomic signatures, or as compared to selected cell populations of the same species in the case of "relative" transcriptomic signatures, using the "Min (test) vs. Max (ref) ≥1-fold" method. Once the fingerprints had been obtained for each species for a given cell type, the gene identifiers were all converted into their corresponding official human gene symbol using BioMart and we selected the intersection of these four lists as the final conserved cross-species transcriptomic signature specific of that cell type, with the following exceptions. First, for certain cell types such as MoDC, data were available from only three, and not four, species. Second, in order to avoid removing putatively relevant signature genes, we kept in these signatures the genes found in all species but one, when their absence in the signature of that given species was due to absence or non-functionality of corresponding ProbeSets on the array of that species.

### Real-Time PCR

For relative quantitation of gene expression in subsets, RNA was reverse transcribed using random primers and the Multiscribe reverse transcriptase (Applied Biosystems). Real-time PCR (qPCR) was carried out with 300 nM primers in a final reaction volume of 25 μl of 1 X SYBR Green PCR Master Mix (Applied Biosystems). The primers used to amplify ovine and porcine cDNA were designed with the Primer Express software (v2.0) using publically available GenBank sequences (Table S1 in Supplementary Material). PCR cycling conditions were 95°C for 10 min, linked to 40 cycles of 95°C for 15 s and 60°C for 1 min. Real-time qPCR data were collected by the Mastercycler® e0p realplex-Eppendorf system and 2−ΔCt calculations for the relative expression of the different genes (arbitrary units) were performed with the Realplex software using GAPDH for normalization. All qPCR reactions showed >95% efficacy.

# Results

#### Isolation of Mononuclear Phagocyte Subset Candidates from Artiodactyl Blood or Spleen Using a Set of Surface Markers

In order to establish a framework of homologous mononuclear phagocyte subsets across different species, we selected two mammalian artiodactyl species, sheep and pig, belonging to the Laurasatherians, a phylogenetically distant order from the Euarchontoglires that include the human and mouse species (**Figure 1**). A set of available antibodies exist to isolate cell subsets in these species of interest as food animals, hosts of zoonotic diseases, and biomedical models. We focused on blood or spleen immune subsets, because (i) a large source of transcriptomic data is available from this compartment in the human and mouse reference species, (ii) they are readily accessible with a minimum of technical biases in all species, and (iii) their gene-expression profiles are not expected to be influenced by peripheral tissue imprinting. We designed antibody panels to sort the subsets. In human and mice, cDC lack expression of T and B lymphocyte and Mo/MP markers and they abundantly express CD11c and MHC class II. Independent groups identified SIRPα as a conserved marker suitable to distinguish cDC2 from cDC1 across species (17, 34). Whereas XCR1 stands as the best marker for identifying cDC1 (34–41), appropriate reagents are not yet available in species outside human and mouse, and CADM1, whose sequence is highly conserved in evolution (42), can be used as an alternative (43, 44). cDC1 and cDC2 candidates were isolated from sheep and pig low density blood cells after exclusion of irrelevant cells (**Figure 2A** for sheep and **Figure 2B** for pigs, see Material and Methods section). The «candidate» nature of a sorted cell subset is marked by a star before the considered subset name in this paper. Due to restricted reagent availability, CD11c and MHC2 class II markers were used to isolate sheep and pig cDC, respectively.

In the case of pig, two populations being CADM1<sup>−</sup> CD172<sup>+</sup> or CD172int were identified and selected as potential candidates and designated as \*cDC2 and \*\*cDC2, respectively (**Figure 2B**). We previously published the marker phenotype, morphology, and type I IFN production properties of sheep lymph and blood \*pDC as CD45RB<sup>+</sup> FSChigh TCRγ/δ− B<sup>−</sup> CD11b<sup>−</sup> cells (26, 48). The sorted cells were very potent at type I IFN production upon viral-type stimulation, demonstrating at the functional level that they were highly enriched in pDC. Moreover, the sorted cells

had the expected size and plasmacytoid morphology, indicating that they were not contaminated by other types of myeloid cells (48). Blood \*pDC from pigs were sorted as CD3<sup>−</sup> CD4<sup>+</sup> CD172dim cells based on marker phenotype, morphology, and type I IFN production properties established by others (25, 49). Pig \*Mo were sorted as CD172high MHC class II<sup>−</sup> cells and sheep splenic \*MP as CD14<sup>+</sup> cells.

To decrease the risk of improper identification of sorted cell subsets, we performed a quality control consisting in examining the expression of a few control genes by qRT-PCR (**Figure 2C**) prior to performing genome-wide transcriptomic analyses. Control genes were chosen based on their high selective expression in a given subset of mononuclear phagocytes in a conserved manner between mouse and human (15, 36) and encompassed *TCF4* for pDC, *CD14* for Mo/MP, *FLT3* for cDC and pDC, *ZBTB46* for cDC, *BATF3* and *XCR1* for cDC1. As expected, *TCF4* was expressed to much higher levels in sheep (26) and pig \*pDC as compared to all other cell types examined except for pig \*\*cDC2. *CD14* was expressed at much higher levels in sheep and pig \*mono/MP as compared to all other cell types examined except one of the two replicates of pig \*cDC2. *FLT3* was expressed at much higher levels in sheep and pig \*cDC1 and in sheep \*cDC2 as compared to all other cell types examined. *BATF3* and *XCR1* were expressed at higher levels in sheep and pig \*cDC1 as compared to all other cell types examined. Importantly, these control analyses have allowed us to improve our initial strategy for sheep \*cDC1 and \*cDC2 sorting. In fact, in our initial sorting (Figure S3 in Supplementary Material), the CD45RB<sup>+</sup> cells were not excluded to sort cDC candidates, and the \*cDC1 were found to express high levels of *TCF4* mRNA, leading us to refine the sheep cDC sorting as presented in **Figure 2**. Thus, overall these control analyses validated our strategy for phenotypic identification and flow cytometry purification of sheep and pig \*pDC, \*Mo/MP and \*cDC1 DC, and of sheep \*cDC2. In the case of pig cell subsets, the nature of \*\*cDC2 and \*cDC2 was not clear since the former expressed high levels of *TCF4* and *XCR1*, and the latter expressed relatively high levels of *CD14* in one out of two replicates. Because pig \*\*cDC2 presented a relatively high expression level of both *TCF4* and *XCR1*, we concluded that they were significantly contaminated by pDC and cDC1. Therefore, we excluded these cells from further analyses and assumed that pig \*cDC2 cells were the proper candidate.

### Use of Pairwise Gene Set Enrichment Analyses for Assessment of the Similarity Between Mononuclear Phagocyte Subsets Across Distant Mammal Species

As a first approach to establish mononuclear phagocyte subset homology across species, we determined the level of similarity between artiodactyl, mouse, and human mononuclear phagocytes using pairwise GSEA, as previously performed to characterize

FIGURE 2 | Sorting of B cells, DC subset candidates, and Mo/MP candidates from pig and sheep blood or spleen and analysis of their expression of control genes. (A) Sheep cell subset sorting from blood and spleen. For sorting of blood cDC subset candidates, low density blood cells were gated on FSChi CD11c+ B− CD11b− TCR1− CD45RB− cells and analyzed for CADM1 and CD172 expression, based on isotype control references for each staining. The CADM1hi CD172lo (\*cDC1) and CADM1lo CD172+ (\*cDC2) cells were sorted. Blood pDC candidates (\*pDC) were sorted as low density FSChi B− CD11b− TCR1− CD8− CD11c− CD45RB<sup>+</sup> cells. Splenic candidate \*Mo/MP were sorted as CD14+ cells. Splenic B cells were identified as DU-2-104+ cells. (B) Pig cell subset sorting from blood. For cDC candidate sorting, low density PBMC were gated on FSChi MHC class II+ B− CD14− CD4− cells and analyzed for CADM1 and CD172 expression. One cDC1 candidate population was identified and sorted, as

CADM1+ CD172lo (\*cDC1). Two cDC2 candidate populations were identified and sorted, as CADM1− CD172hi (\*cDC2) and CADM1− CD172int (\*\*cDC2). Candidate Mo were sorted as CD172+ MHC2− cells (\*Mo). Candidate pDC were sorted as CD3− CD14− CD4+ CD172int cells (\*pDC). B cells were identified and sorted as IgL+ cells. (C) qPCR analysis of the expression of control genes in sorted candidates from one or two animals. RNA from candidate cell subsets (left, sheep; right, pig) were subjected to detection of control transcripts by qPCR. Control transcripts were chosen based on their high selective expression in specific subsets of mononuclear phagocytes in a conserved manner between mouse and man, i.e., *TCF4* for pDC, *FLT3*, *BATF3* and *ZBTB46* for cDC, *XCR1* for cDC1, and CD14 for Mo/MP. Data are represented as relative expression levels normalized to maximal expression across cell types, each bar corresponding to a distinct animal.

human immune cell subsets (19) and chicken cDC (31). To that aim, we used publicly available transcriptomic data from a selection of human and mouse immune cell types (Data Sheet S1 in Supplementary Material). We established human and mouse transcriptomic fingerprints for B cells, pDC, cDC1, Mo/MP, MoDC, cMo, and ncMo as the list of genes that are expressed at least 1.5-fold higher in the index cell population than in a large number of other immune cell types (Data Sheet S2 in Supplementary Material). B lymphocytes were chosen in all species as a reference cell subset, because their phenotypic identification in each species and their homology across species are already well established, and because they are expected to share with mononuclear phagocytes a genetic program underlying their common function of antigen-presenting cells. We generated a common fingerprint for Mo and tissue MP because their gene program is very close in the mouse (9), even though tissue MP generally derive from embryonic precursors rather than from circulating blood Mo. We could not establish a human or mouse cDC2 transcriptomic fingerprint with a sufficiently large number of genes for subsequent reliable statistical analysis. We also defined relative transcriptomic signatures for cDC vs. Mo/MP as the list of genes that are 1.5-fold higher in all cDC relatively to Mo and MP from different tissues, and reciprocally (Data Sheet S2 in Supplementary Material). Finally, we identified transcriptomic fingerprints from human and mouse MoDC (12). We then tested whether the transcriptomic signatures of mouse and human immune cell types were enriched between sheep or pig candidate cell subsets using GSEA (33) (**Figures 3** and **4**). As control, since homologies between mouse and human cell subsets have been previously demonstrated by other methods of transcriptional analyses (15, 16, 18), mouse fingerprints were also used for GSEA analysis on human cells (Figure S4 in Supplementary Material) and reciprocally (Figure S5 in Supplementary Material).

As expected, sheep B cells were significantly enriched for the expression of both human and mouse B cell transcriptomic fingerprints as compared to all other sheep cell subsets examined (**Figure 3**, ). The sheep \*pDC were enriched for the human and mouse pDC fingerprints in most comparisons (**Figure 3**, ), suggesting that sheep \*pDC correspond to homologs of human and mouse pDC. However, both mouse and human pDC fingerprints were not significantly enriched in the comparison of sheep \*pDC with \*cDC2 (NES = 1.29 and 1.24, and FDR = 1.0 and 1.0, respectively), indicating that sheep pDC probably contaminate sheep \*cDC2 despite exclusion of CD45RB<sup>+</sup> cells for their purification. The sheep \*pDC were also enriched for the human B cell fingerprint in most comparisons (except with sheep B cells), what can be partly explained by the known overlap between the gene-expression program of pDC and B (15, 50–52); however, the human pDC fingerprint is not enriched in the sheep \*pDC comparison with B cells and the extent of the human B cell fingerprint enrichment in sheep \*pDC is above the expectations provided by similar analyses in the human and mouse reference species (Figures S4 and S5 in Supplementary Material), all of this indicating that B cells are likely to contaminate sheep \*pDC despite exclusion with a pan-B cell marker for \*pDC selection. Finally, \*cDC2 did not show a clear enrichment for any human and mouse signatures (**Figure 3**, ). However, it is also the case when examining enrichment of mouse cell subset fingerprints in human cDC2 (Figure S4 in Supplementary Material, ) and reciprocally (Figure S5 in Supplementary Material, ). Hence, this GSEA approach is not very informative for identification of cDC2, due to the lack of robust human or mouse fingerprints that are specific of this cell type as mentioned earlier. Sheep \*cDC1 were significantly enriched in the human and mouse cDC1 fingerprints in all comparisons (**Figure 3**, ). They were also enriched systematically in the mouse cDC vs. Mo/MP fingerprints. This suggested that sheep \*cDC1 correspond to true homologs to human and mouse cDC1. Sheep splenic \*Mo/MP were strongly enriched for the human and mouse Mo/MP vs. cDC fingerprints except when compared to MoDC, and not for the human and mouse fingerprints of B lymphocytes, pDC, or cDC (**Figure 3**, ). This confirmed that sheep splenic \*Mo/MP belong to the monocytic lineage and not to the B nor DC lineages. However, their precise identity remained unclear as they were enriched for the mouse cMo fingerprint but not for the human cMo or ncMo fingerprints. When mouse fingerprints were applied on human immune cell subsets comparisons and vice versa, there was also no consistent alignment of ncMo between the two species (Figures S4 and S5 in Supplementary Material, highlights and ). Finally, sheep \*MoDC that were derived from bone marrow cells in GM-CSF (27), were systematically and strongly enriched in the human and mouse MoDC signatures (**Figure 3**, ), confirming the homology between these three populations.

A similar analysis for pig candidate cell subsets also clearly established similarities with their putative human and mouse equivalents for B cells (**Figure 4**, ), pDC (**Figure 4**, ), and cDC1 (**Figure 4**, ) but not for cDC2 (**Figure 4**, ). Pig \*Mo were clearly enriched for human and mouse fingerprints of cells of the monocytic lineage, and not for human and mouse signatures of B lymphocytes, pDC, or cDC (**Figure 4**, ).

Thus, altogether, GSEA analysis of the sheep and pig data for the fingerprints of human and mouse immune cell subsets gave results as informative as those obtained when comparing together human and mouse cell types, and clearly established similarities between sheep and pig cell subset candidates and their putative human and mouse equivalents for B cells, pDC, cDC1, and MoDC. Further analyses are necessary to precisely identify the nature of sheep and pig \*cDC2 and \*Mo/MP subsets.

#### Confirmation and Extension of the Conclusions on the Similarity Between Mononuclear Phagocyte Subsets Through Global and Simultaneous Analysis of the Gene-Expression Profiling of All Cell Types from Mammalian Species Using Hierarchical Clustering

In order to confirm the identification of homologous mononuclear phagocytes across species as deduced from GSEA analyses, and to potentially gain more insights into the exact nature of pig and sheep \*cDC2 and \*Mo/MP, we next processed all the data together for global analysis by hierarchical clustering (**Figure 5**). Only the genes that showed significant variation in their expression across subsets in each species were selected


FIGURE 3 | GSEA-based assessment of the identity of sheep cell subset candidates by comparison with well-defined human and mouse mononuclear phagocytes. Candidate sheep cell subsets were compared to one another for their relative enrichment in transcriptomic fingerprints (GeneSets) specific of human (A) or mouse (B) mononuclear phagocyte subsets, using GSEA through the Bubble GUM software. The human and mouse GeneSets were defined through the same approach based on pre-existing knowledge of equivalency between human and mouse mononuclear phagocytes. A GeneSet specific for B cells was included as a control for the methodology, since the identity of this cell type is clearly established in all species and its homology across species is undisputed. The GeneSets used were named and defined as follows. The transcriptomic fingerprints "B cell," "pDC," "cDC1," "cMo," "ncMo," and "MoDC" consisted in the lists of human/mouse genes showing a high selective expression in the eponym human/mouse cell subset as compared to many other leukocytes

(see Materials and Methods for further details, Data Sheet S2 in Supplementary Material). The transcriptomic fingerprints "cDC vs. Mo/MP" and "Mo/MP vs. cDC" consisted in the lists of human/mouse genes expressed in cDC to higher levels than in Mo/MP, and reciprocally (Data Sheet S2 in Supplementary Material). All possible pairwise comparisons between sheep cell subsets were performed to assess their respective expression of the transcriptomic fingerprints of human and mouse mononuclear phagocyte subsets, using the Bubble GUM software for calculations and graphical output. Results are represented as bubbles, in a color matching that of the cell subset in which the GeneSet was enriched. Stronger and more significant enrichments are represented by bigger and darker bubbles, as illustrated in the legend box of the figure. Specifically, the surface area of bubbles is proportional to the absolute value of the normalized enrichment score (NES). The color intensity of dots is indicative of the false-discovery rate (FDR) statistical value.

and the resulting datasets were normalized across species. All B cells from the four mammalian species grouped together in a specific branch of the tree, rather than each with other immune cells of the same species. This finding validates hierarchical clustering as an alternative method for identifying homologous mononuclear phagocytes across species. A closer examination of the dendrogram shows that the different cell types grouped in two major branches. The first one encompassed all the known and candidate cells of the monocytic lineages and pig \*cDC2, and split further into two subgroups, one including all the identified or candidate MoDC, and the other one including all the identified or candidate Mo/MP and pig \*cDC2. The second branch encompassed all the other cell types known or hypothesized not to belong to the monocytic lineage. This branch further split into two sub-branches, one constituted of the group of B cells and of the group of identified or candidate pDC, and the other

candidates were analyzed exactly as described in Figure 3 for sheep cell

constituted of identified or candidate cDC subsets except pig \*cDC2. The common clustering of B and pDC transcriptome can be explained by the shared gene-expression program between B and pDC as mentioned above. Hence, this analysis confirmed the conclusion already drawn from the GSEA analyses, namely the monocytic nature of sheep and pig \*Mo/MP and \*MoDC, as well as the homology between pig, sheep, mouse, and human \*pDC/ pDC. Moreover, the hierarchical clustering analysis allowed to better define the nature of sheep and pig \*cDC2. Specifically, it confirmed the hypothesis that sheep \*cDC2 belong to the cDC family, while, on the contrary to our *a priori* assignment, it shows that pig \*cDC2 rather resemble Mo than cDC. However, within the branch of monocytic cells, this analysis grouped Mo/MP by species of origin rather than by cMo vs. ncMo subsets. Similarly, this analysis grouped cDC by species rather than by cDC1 vs. cDC2 subsets.

through the Bubble GUM software.

## Identification of Similarity Between Subsets of DC and of Mo Across Species Through Hierarchical Clustering Analyses Focused on These Cell Types

The expression patterns of genes outside of the cell types of interest may mask similarity between cDC or Mo subsets, as previously reported (15). Hence, we further evaluated the similarities between subsets of cDC on the one hand, and of Mo/ MP on the other hand, by re-analyzing their gene-expression profiles focusing only on the genes that showed significant variation in their expression across DC subsets (**Figure 6**) or Mo/MP (**Figure 7**) in each species. Pig data were not used in the analysis focused on cDC, because, pig \*cDC2 belonged to the monocytic branch and not to the DC branch of **Figure 5**. Sheep data were not used in the analysis focused on Mo/MP, because only one subset of sheep Mo/MP had been purified. Remarkably, these focused analyses grouped samples by cell types rather than by species. The cDC-focused hierarchical clustering confirmed the conclusion drawn from GSEA that sheep, mouse, and human cDC1/\*cDC1 are homologs, and refined our understanding of the identity of sheep \*cDC2 by showing their homology to mouse and human cDC2 (**Figure 6**). The Mo/MP-focused hierarchical clustering allowed to newly identify pig homologs to mouse and human cMo vs. ncMo (**Figure 7**). Pig \*cDC2 correspond to ncMo and pig \*Mo correspond to cMo. In a complementary phenotypic FACS analysis, we confirmed that likewise human ncMo as compared cMo, pig \*cDC2 express higher membrane levels of CD16 and CD163 as compared to pig \*Mo (Figure S6 in Supplementary Material).

Altogether, our comparative analyses of the gene-expression profiles of mononuclear phagocyte subsets across mammals

hierarchical clustering analysis was performed as described in Figure 5, but focused only on cDC subsets. The corresponding filtered dataset included 868 unique orthologous genes found regulated between cDC1 and cDC2 from human (Hs), mouse (Mm), and sheep (Oa). Pig cDC could not be included in this analysis due to the lack of data on proper pig cDC2.

analysis due to lack of data on subsets of sheep monocytes. indicated that the complex specialization of these cells into distinct subsets is conserved across mammals for both DC and Mo. Subset grouping did not indicate existence of a relationship between transcriptomic proximity of subsets and phylogenetic closeness of species. The conserved organization across distant mammals suggests that the mononuclear phagocyte complexity arose in a common mammalian ancestor and

that the different subsets can be considered as homologous

orthologous genes found regulated between cMo and ncMo from human (Hs), mouse (Mm), and pig (Ss). Sheep data could not be included in this

## Evidences for Homologous cDC and Mo/MP Lineages Across Warm-Blooded Vertebrates

subsets across mammals.

We recently generated the transcriptomic profile of MP, total cDC, and B cells from chicken spleen and found similarities with human and mouse corresponding immune cell subsets by GSEA (31). In order to extend our subset homology analysis to non-mammalian vertebrates, we normalized and processed the transcriptomic data in a hierarchical clustering analysis as described above, using mammalian and chicken Mo/MP, B cells, and cDC subsets (**Figure 8**). There again, a tree consisting of two main branches was obtained, corresponding to a split between Mo/MP and B cells/DC. In the cDC branch, the cDC1 subset clustered together and included the chicken total cDC. The chicken MP grouped with the mammalian Mo/MP. Whereas this analysis is still partial due to limited knowledge and availability on marker sets for sorting immune cell subsets in chicken, it shows that our transcriptomic comparative approach can be used to define subset homology throughout vertebrates. It also further supports that separation of mononuclear phagocytes into Mo/MP and cDC occurred early during vertebrate evolution and must already have been in place in the common ancestor of reptiles (including birds) and mammals.

### Identification of Mononuclear Phagocyte Gene-Expression Signatures Across Mammals

Taking advantage of our multi-species microarray data, we sought to identify core gene-expression signatures that should universally define at the molecular level each of the mononuclear phagocyte subset and that should hold biological relevance based on their selective and conserved expression in homologous subsets throughout mammalian evolution. Absolute signatures ["Min (test) vs. Max (ref)" method, see Materials and methods] encompassed all genes selectively expressed at higher levels in the cell subset of interest (index population) as compared to all the other cell subsets studied (comparator populations), in all species studied. An absolute signature was computed for B cells in order to validate the approach by comparison of the gene list obtained with the advanced knowledge available on the biology of this lymphocyte population. Absolute signatures were also found for pDC, cDC1, and MoDC. Relative signatures encompassed genes selectively expressed to higher levels in one or several cell subsets of interest (index population) as compared to a selection of other cell subsets (comparator populations). The choice of index and comparator populations was largely based on the branching of different cell subsets in hierarchical clustering (**Figure 5**), or on known sharing of specific functions between cell subsets in mouse or human. The conserved absolute and relative gene-expression signatures in mononuclear phagocyte subsets are listed in **Table 1** and Data Sheet S3 in Supplementary Material. In several instances, Ingenuity Pathway Analysis (IPA) mapped a high proportion of the genes to gene interaction networks (**Figure 9** for the DC lineage subsets, **Figure 10** for the monocytic lineage subsets and Figure S7 in Supplementary Material), and revealed predicted upstream regulators (**Figure 11A**) and canonical pathways and functions (**Figure 11B**) that are described thereafter for B cells, DC lineage subsets, and Mo/MP categories. Although certain functions or pathways were enriched in several gene signatures, the genes responsible for the enrichments differed (Data Sheet S4 in Supplementary Material) and pointed out to different, complementary contributions of the distinct cell types to the corresponding functions or pathways.

The conserved B cell signature that we use as our reference subset (**Table 1**) includes a regulatory gene network directed to immunoglobulin production (Figure S7 in Supplementary Material), with PAX5 as an upstream regulator (*p* = 10<sup>−</sup>5.8) (**Figure 11A**). SOX11 (*p* = 10<sup>−</sup><sup>8</sup> ) and FOXO1 (*p* = 10<sup>−</sup><sup>7</sup> ) are predicted to be other upstream regulators in the conserved B signature (**Figure 11A**), in agreement with existing knowledge. As expected, this signature is associated to B lymphocyte ontogeny and functions [e.g., "development of B lymphocytes" *p* = 10<sup>−</sup>10.4, "antibody response" (*p*= 10<sup>−</sup>7.5), "proliferation of B lymphocytes" (*p* = 10<sup>−</sup>10.5), and "morphology of B lymphocytes" (*p* = 10<sup>−</sup>7.5) as well as to the "B cell receptor signaling" pathway (*p* = 10<sup>−</sup>7.3)] (**Figure 11B**). The B cell signature also pinpoints to genes without any known function in B cells yet, such as the cell cycle gene

FIGURE 8 | Unsupervised cross-species hierarchical clustering including a chicken dataset demonstrates a conserved organization of vertebrate mononuclear phagocytes in the two main lineages of Mo/MP vs. cDC. An unsupervised cross-species hierarchical clustering analysis was performed as described in Figure 5, but including geneexpression data from chicken (Gg prefix for *Gallus gallus*) and focused only on the cell types commonly sorted in all five vertebrate species, i.e., B cells, Mo/MP, and cDC. The corresponding filtered dataset included 388 unique orthologous genes found regulated across cell subsets in each species.

*RAD17* (53) or the *SP140* gene that encodes a nuclear body protein (54) (**Table 1**). Altogether, the results of the functional analysis of the conserved signature of B cells support the biological relevance of the conserved gene signatures generated by our approach.

In the conserved signatures corresponding to the DC lineage, the pDC signature is restricted to few genes including *RUNX2*, which encodes for a major known regulator of pDC development (56) and other genes whose role is not yet known in this subset, with three of them coding for potential cell surface markers or targeting molecules, i.e., the low density lipoprotein receptorrelated protein 8 (*LRP8*), tetraspanin 13 (*TSPAN13*), and a zinc-family transporter protein member (*SLC30A5*) (**Table 1**). These genes, except *SLC30A5*, map to a common network (**Figure 9A**). No functional annotation was found significantly enriched in the pDC absolute signature due to the low number of associated genes. Interestingly, the pDC vs. cDC relative signature includes genes belonging to a regulatory network pointing to IFN −α/β production (**Figure 9B**) and retrieves as a major putative upstream regulator X-box binding protein 1 (*XBP1*) (*p* = 10<sup>−</sup>15) (**Figure 11A**), a transcription factor involved in mouse DC development (57). The pDC vs. cDC relative signature was also enriched for "proliferation of B lymphocytes" (*p* = 10<sup>−</sup><sup>4</sup> ), "morphology of B lymphocytes" (*p* = 10<sup>−</sup><sup>5</sup> ), and "B cell receptor signaling" pathway (*p* = 10<sup>−</sup>2.9), similarly to the conserved B cell signature (**Figure 11B**). These observations are consistent with the known usage downstream of mouse and human pDC endocytic receptors of a signaling pathway akin to that of the B cell receptor (58). This known pDC signaling pathway involves the products of *SYK*, *BLNK*, and *PIK3AP1*, three of the six genes responsible for the enrichment of the "B cell receptor signaling" pathway in the conserved pDC vs. cDC gene signature (Data Sheet S4 in Supplementary Material), as well as *CARD11* which contributes to the enrichment for the annotation "proliferation of B lymphocytes" in the pDC vs. cDC signature. This strongly suggests that this signaling pathway is conserved in pDC of all mammalian species. Beside *TCF4* which encodes for a major known regulator of both B and pDC development (52), several other genes associated to B cell biology are found in the pDC vs. cDC relative signature (**Table 1**), namely *CD79B*, *PTPRCAP*, *SEMA4D*, *CTCF*, *IFR1*, and *MEF2C*. This suggests that additional biological processes shared between B cells and pDC remain to be identified.

No absolute signature could be generated for cDC but interesting informations were obtained with relative signatures, i.e., the cDC vs. Mo/MP and cDC vs. pDC. The cDC vs. Mo/MP signature includes *FLT3*, a key gene in mouse DC development (59) as well as many genes of a regulatory network including *BCL11A*, *HLA-DOA*, *HLA-DRA*, *HLA-DMB*, *HLA-DOB*, *CD74*, the axone guidance neuron navigator *NAV1* and the MHC class 2 transcription regulator *RFX5* (**Figure 9C**). In relation to this network, *IL27* (*p* = 10<sup>−</sup>4.4), *IFNG* (*p* = 10<sup>−</sup>2.6), and NFkB (10<sup>−</sup>2.8) were retrieved as putative upstream regulators (**Figure 11A**). The cDC vs. Mo/MP signature was enriched for canonical pathways such as "antigen presentation" (*p* = 10<sup>−</sup>9.6), "DC maturation" (*p* = 10<sup>−</sup>4.6), and "T helper cell differentiation" (*p* = 10<sup>−</sup>6.2) (**Figure 11B**). The cDC vs. pDC signature includes a main regulatory network encompassing *PIK3CB*, *ICAM1*, *CLEC7A*, *HLA-DRA*, *IL1B*, and *LGALS3* (**Figure 9D**) and is enriched for "functions of antigen-presenting cells" (*p* = 10<sup>−</sup>11.1), "inflammatory response" (*p* = 10<sup>−</sup>8.9), "bacterial infection" (*p* = 10<sup>−</sup>7.9), "migration of cells" (*p* = 10<sup>−</sup>11.4), and "clathrin-mediated endocytosis signaling" pathway (*p* = 10<sup>−</sup>4.6) (**Figure 11B**). TNF (*p* = 10<sup>−</sup><sup>9</sup> ), RELA (10−5.8), NFKB1 (10−6.1),


#### TABLE 1 | Conserved gene signatures for mammalian mononuclear phagocytic cell subsets.

(*Continued*)

#### TABLE 1 | Continued


*a Genes conserved in 4/4 species, or in 3/3 species for cDC2, cMo, and ncMo since only three species could contribute to the analysis.*

*b Genes conserved in 3/4 species or 2/3 species.*

*c Genes in bold were previously demonstrated to play a significant role in the development or functions of the population of interest.*

*d Underlined genes were annotated as located in "plasma membrane" according to Ingenuity Pathway Analysis.*

*e Genes highlighted in gray have been previously identified as signatures genes for the corresponding mouse and human cell populations in our earlier study (15).*

*f Signature genes of the relative cMo vs. ncMo and of the ncMo vs. cMo signatures were provided only for the 3/3 species selection since the gene lists for the 2/3 species selection encompassed hundreds of genes.*

P38 MAPK (10<sup>−</sup>5.5), IFNG (10−5.4), and to a lesser extent STAT3 (10<sup>−</sup>4.7) and CSF2 (10−4.3) are predicted as putative upstream regulators in this signature (**Figure 11A**). In addition, this cDC vs. pDC signature includes *BATF3*, a gene highly expressed in cDC that is key in cDC1 development in mouse and human (60), as well as *ARHGAP22*, a gene involved in actin cytoskeleton regulation (61), that was initially described as a top gene of the absolute cDC signature common to human and mouse (15). Altogether, the relative gene signatures of cDC emphasize their nature of highly endocytic, motile, and expert antigen-presenting cells throughout species.

The conserved cDC1 signature encompasses genes with known contribution in the biology of this lineage, such as *XCR1*, *FLT3*, and *CADM1* (59), as well as additional genes which biological function in this subset remains enigmatic, such as the germinal center B-cell-expressed transcript 2 protein (*GCET2*),

FIGURE 9 | IPA gene interaction networks of the conserved signatures in subsets of the DC lineage. The conserved signatures of subsets of the DC lineage were analyzed in Ingenuity Pathway Analysis which generates networks based on the connectivity of the genes in each signature (in boldface) but also on their connectivity with genes not belonging to the signature (in plain characters). The identified networks are displayed as graphs showing the molecular relationships between genes/gene products. Genes are represented

as nodes, and the biological relationship between two nodes is represented as an edge (line). The edges can represent direct (continuous) or indirect (dashed) relationships between nodes. Selected networks generated by IPA and covering parts of conserved cell-specific signatures are displayed: (A) pDC signature network, (B) pDC vs. cDC signature network, (C) cDC vs. Mo/MP signature network, (D) cDC vs. pDC signature network, (E) cDC1 vs. (pDC and cDC2) signature network, (F) cDC2 vs. (pDC and cDC1) signature network.

pathways (B). A promoter sequence analysis of the conserved cell type gene signatures performed using IPA is displayed as a heatmap of the *p*-value [upstream regulators, (A)]. A biological function and canonical pathway analysis

> The absolute cDC2 conserved signature was empty. Many genes of the relative cDC2 vs. (pDC and cDC1) signature belong to a network that includes *SIRP*α (*CD172A*), a selective marker of cDC2 within the DC lineage (44), together with *CSFR1*, *TREM1*, *CLEC4A* (also known as *DCIR*), *CD1B*, and *RELB* which is known to control mouse cDC2 differentiation (65) (**Figure 9F**). A second network includes *ITGAM* (*CD11b*), a marker used to identify mouse cDC2, *CLEC6A*, and *IL1B* (Figure 7 in Supplementary Material). SPI1 (*p* = 10<sup>−</sup>9.8), CEBPD (10−6.2), and CEBPB (10−3.6) are predicted upstream regulators, as well as CSF2 (10<sup>−</sup>3.8), STAT3 (10−4.4), and IFNG (10−<sup>4</sup> ) which were already enriched in the cDC vs. pDC signature (**Figure 11A**). This cDC2 relative signature also includes *IFI30*, also known as *GILT*, a lysosomal thiol reductase important in MHC class II and class I antigen processing (66, 67) (**Table 1**). This relative signature is enriched for "function of

based on the average linkage metrics. Enrichments were considered significant

when supported by at least three genes and by a *p*-value ≤0.05.

the WDFY family member 4 (*WDFY4*) whose polymorphism is associated to autoimmune diseases (62), and two intracellular trafficking proteins, a formin-binding protein (*FNBP1*) (63) and Sorting Nexin-22 (*SNX22*) (64) (**Table 1**). The conserved cDC1 vs. (pDC and cDC2) relative signature provides a longer list of genes belonging to an interaction network that includes *BCL6*, a transcriptional repressor that was recently found involved in the specification of cDC1 (17) as well as *XCR1* and *CALM1* (**Figure 9E**). IPA did not retrieve significant annotations for the cDC1 absolute or relative gene signatures. This emphasizes how little is currently known on the molecular regulation of the functions specific to cDC1s, such as cross-presentation. Future studies investigating in mouse cDC1s the functional role of the genes identified here as being part of the conserved cDC1 signatures will advance our understanding of the functions of these cells and their molecular regulation.

antigen-presenting cells" (*p* = 10<sup>−</sup>6.2), "inflammatory response" (*p* = 10<sup>−</sup>6.9), and for the pathways "TREM1 Signaling" (10−2.5), "Toll-like receptor canonical signaling" (10<sup>−</sup>2.5), and "role of pattern recognition receptors in recognition of bacteria and viruses" (*p* = 10<sup>−</sup>3.9) (**Figure 11B**). Other genes were uncovered that may be important regulators of the function of cDC2s or which product could be used to identify or target these cells, including the genes coding for plasma membrane proteins such as glycoprotein *CD300A*, the sialic binding lectin *SIGLEC8*, and the paired immunoglobin-like type 2 receptor *PILRA*. This conserved relative signature shows that within the DC lineage throughout species, cDC2 express specific networks of genes related to pathogen sensing, antigen presentation, IL-1β production, and inflammation.

In the conserved signatures corresponding to the monocytic lineage, the absolute MoDC and relative MoDC vs. Mo/MP signatures are enriched for "inflammation of organ" (*p* = 10<sup>−</sup>6.9 and *p* = 10<sup>−</sup>3.3), "function of antigen-presenting cells" (*p* = 10<sup>−</sup>5.1 and *p* = 10<sup>−</sup>3.6), and "migration of antigen-presenting cells" (*p* = 10<sup>−</sup>3.3 and *p* = 10<sup>−</sup>4.3) (**Figure 11B**) and encompasses *NOS2*, *CCL17*, *VEGFA*, and *FCGR2B* that map to a common major network (**Figure 10A**). IL-4 (*p* = 10<sup>−</sup>5.7) and IL5 (*p* = 10<sup>−</sup>5.9) are predicted regulators (**Figure 11A**). Among other genes of interest that had not yet been associated to MoDC are the triose phosphate isomerase *TPI1*, the NADH dehydrogenase flavoprotein *NDUFV2*, the aldolase *ALDOA*, and the *CD200R1* gene that encodes for an inhibitory cell surface receptor of MP functions (68) (**Table 1**). The relative MoDC vs. Mo/MP signature encompasses additional genes that participate in "migration of cells" (*p*= 10<sup>−</sup>2.3, with *S1PR3*, *CCL17*, and *SLC2A1*), "bacterial infection" (*p*= 10<sup>−</sup>3.2, with *CD1B*, *CD209*, and *FCGR2B*), and "synthesis of nitric oxide" (10<sup>−</sup>2.5, with *PLAU*, *IL1R2*, and *NOS2*) (**Figure 11B** and Data Sheet S4 in Supplementary Material). The conserved MoDC signatures indicate a dominant association of this subset to inflammation, as well as to DC functional properties when compared to Mo/ MP across species.

Most of the genes in the Mo/MP vs. cDC conserved signature had been previously identified as overexpressed in murine MP, such as *TLR4*, *CEBPB*, *C5AR1*, and *SOD2* (9, 15) (**Table 1** and **Figure 10B**). A significant proportion of the genes within this signature are related to "inflammation of organ" (*p* = 10<sup>−</sup>2.7), "production of reactive oxygen species" (*p* = 10<sup>−</sup>4.4), "synthesis of nitric oxide" (*p* = 10<sup>−</sup>3.9), "bacterial infection" (*p* = 10<sup>−</sup>4.6), "role of pattern recognition receptor in recognition of bacteria and viruses" (*p* = 10<sup>−</sup>3.3), and "acute phase response signaling" (*p* = 10<sup>−</sup>2.9) (**Figure 11B**). Putative upstream regulators are NPC1 (*p* = 10<sup>−</sup>6.8), APOE (*p* = 10<sup>−</sup>6.8), IFNG (*p* = 10<sup>−</sup>6.4), SPI1 (*p* = 10<sup>−</sup>5.2), and NFkB (*p* = 10<sup>−</sup>5.1) (**Figure 11A**). Additional proteins are potential transcriptional regulators of importance in Mo/MP, such as the RNA-binding protein RBMS1 and the cell cycle progression factor CCPG1. The Mo/MP vs. MoDC signature includes a gene network centered on *NFkB* and *MAP3K5* (**Figure 10C**). Overall, the conserved Mo/MP relative signatures support the association of Mo/MP to inflammation and oxidative stress across species.

The conserved comparative signature of cMo vs. ncMo retrieved genes belonging to a network with *IL1*, fibronectin (*FN*), *S100A8*, and *XBP1* (**Figure 10D**), the latter being proposed as an upstream regulator (10<sup>−</sup>2.5) together with *NPC1* (*p* = 10<sup>−</sup><sup>4</sup> ) (**Figure 11A**), and is strongly associated to "inflammatory response" (*p* = 10<sup>−</sup>8.4) (**Figure 11B**). The reciprocal ncMo vs. cMo signature includes a gene network with *IL17A*, *CSFR1*, *NFKBIA*, and *serpinA1* (**Figure 10E**), and is significantly associated to the "glucocorticoid receptor signaling pathway" (*p* = 10<sup>−</sup>2.5) and to some extent to the "inflammation of organ" (*p* = 10<sup>−</sup>2.6) (**Figure 11B**). These relative signatures indicate that cMo have a conserved gene program directed to strong inflammation, whereas ncMo, a poorly understood subset, might be exquisitely regulated by glucocorticoids as suggested in the literature (69, 70).

Altogether, the mononuclear phagocyte system from distantly related mammals is composed of a diversity of subsets that belong to the DC or to the Mo/MP lineage and express discriminating gene signatures involved in distinct regulatory networks and biological functions conserved through mammalian evolution. In most instances, the subset signatures also point to several unexpected genes and upstream regulators that are likely to be important in the subset biology since their selective expression pattern across subsets of mononuclear cells is conserved across species.

### Phylogenetic Evidences for the Existence of a Gene Repertoire for Mononuclear Phagocyte Subsets in Birds and Bony Fishes

The existence of orthologous genes of the conserved mononuclear phagocyte subset signatures in reptile/birds, fishes, and agnathans, would indicate that the genetic equipment for mononuclear phagocyte subset diversity is available in vertebrate species distant from mammals. In the case of birds and reptiles, it remains unknown whether they have pDC, cDC1, and cDC2 subsets homologous to mammals. An orthology analysis of selected genes from conserved subset signatures revealed that most genes possess a unique ortholog in birds and reptiles, with conserved synteny with human, for instance *XCR1*, *BATF3*, *RUNX2*, *TSPAN13*, and *CSF1R* (**Table 2**). Furthermore, these same genes also possess one or more orthologs in fish. Multiple orthologs in fish are often due to the whole genome duplication that occurred during the evolution of teleosts, and to further local duplications. Importantly, fish co-orthologs of mononuclear phagocyte subset genes are generally supported by conserved synteny. Genes duplicated in fish may have been subjected to sub-functionalization, as it is the case for many immune genes duplicated in this group of vertebrates; however, some markers have a unique counterpart in fish genomes (like *BATF3*, *RFX5*, and *CIITA*), with copy loss possibly due to detrimental effects of duplication. The case of *MHC class II* is particular: although fish *MHC class II* genes are not always considered as true orthologs of human *MHC class II* genes, their sequences show the hallmarks of *bona fide* class II antigen-presenting receptors and they likely have similar functions. For c-type lectin-like (*CLEC*) molecules, no true orthologs can be identified in fish nor in birds/reptiles, as each branch of vertebrates – even each group of mammals – shows its own set of expanded *CLEC* genes. Altogether these data show that a repertoire of conserved genes for mononuclear phagocyte subsets exists in bony fishes and reptiles, which constitutes a list of candidates for relevant markers.


TABLE 2 | Search for the existence of orthologs in reptile/birds, fishes, and agnathans for selected genes of the conserved mononuclear phagocyte subset signatures.

*a Birds have one co-ortholog of human C5RA1 and C5RA2; bonly in the lizard Anolis, not in available bird genomes; c fish generally have one co-ortholog of human C5RA1 and C5RA2; donly in some species: zebrafish, catfish, and salmonids; e see Ref. (36); f for all CLEC, no true ortholog, each deep branch of vertebrates has its own set of expanded CLEC; g the neighborhood is not conserved but zebrafish CD79B is close to Arhgap27 and Plekhm1 that are on the same human chromosome (chr17) as CD79B but at 20 megabases; CD79B genes often are not annotated in fish genomes. In zebrafish, CD79B is ENSDARG00000088902; h a lamprey gene ortholog to PAX5 has been identified and was selectively expressed in lamprey VLRB*+ *cells which resemble B lymphocytes (55); however, this gene is not identified in the current publicly available assembly of the lamprey genome; i duplicated in zebrafish and cavefish; j a lamprey gene is a co-ortholog to all vertebrate CSF1R, PDGFR, KIT, FLT3, etc.; k in birds species, several genes are co-orthologs of all mammalian SIRPs including SIRPA; l bird TREM-like genes are more closely related to TREM2 rather than to TREM1; mco-ortholog of CLNK, BLNK, and other related genes.*

The presence of *BATF3* and *XCR1* are hints at possible existence of cDC1 in these species, as *BATF3* specifically controls cDC1 development in mice (71) and *XCR1* expression is strictly associated to cDC1 in several mammals (34, 36, 40, 41). In contrast, the lamprey does not have identified orthologs for many of the genes selected from the transcriptomic fingerprints of the subsets of mammalian mononuclear phagocytes (**Table 2**). Agnathans, including lampreys and myxines, harbor three adaptive immune cell types, each expressing a specific class of variable lymphocyte receptors, VLRC, VLRA, and VLRB, and showing transcriptomic and functional commonalities with gnathostome γδ T lymphocytes, αβ T lymphocytes, and B lymphocytes, respectively (55, 72). However, it is uncertain whether or not the activation of agnathan lymphocytes requires APCs, and if so, to which extent these cells could resemble gnathostome APCs (72). Contrary to the situation in birds and fishes, our observations do not support the existence in the lamprey of gene sets similar to those defining the transcriptomic fingerprints of the mononuclear phagocytes of mammals. Although incomplete assembly and annotation of the genome of the lamprey do not allow drawing definitive conclusions, our observations are consistent with the lack in agnathans of MHC functional homologs and of the particular proteasome machinery used by mammalian APCs for antigen processing (72). Altogether, this phylogenetic study shows that the repertoire of key genes characterizing the diversity of the mononuclear phagocytes in mammals were already present in the common ancestor of tetrapods and fishes but might be largely absent in agnathans.

# Discussion

Our computational transcriptomic meta-analysis indicates that the complex organization of the mononuclear phagocyte system shows conservation throughout distantly related mammals, a finding that appears to extend to chicken, a non-mammalian vertebrate. In the present work, by using GSEA and hierarchical

clustering for unbiased pan-genomic analysis of the molecular identity of immune cell subsets across four vertebrate species, we convincingly established the existence of strong homologies between these cell types across mammals, beyond the already known existence of B cells in all species. Specifically, we could align across mammals cDC1, cDC2, pDC, MoDC, Mo/MP, and cMo vs. ncMo. In addition, we found that many of the genes that we showed to be selectively expressed in distinct mononuclear phagocyte subsets in mammals have existing orthologs in bony fishes while this appears not to be the case in lamprey. Thus, our study suggests that conserved mononuclear phagocyte subsets might exist in all gnathostomes but not in agnathans. However, this hypothesis will require to be tested experimentally, by re-examining the presence of orthologous genes in lamprey upon completion of the genome assembly and its annotation, by identifying and studying candidate mononuclear phagocyte subsets in bony fishes, and by determining whether similar cells exist in sharks, rays, and lamprey. For example, orthologous genes of the conserved mononuclear phagocyte signatures (**Table 2**) could be targeted by the CRISPR/Cas9 technology with a reporter gene marker in order to identify and characterize mononuclear phagocyte subsets in bony fishes (73), with for certain genes the need to test several putative orthologs in fish due to genome duplication.

The two methodologies that we used to assess subset homologies across species, i.e., hierarchical clustering and GSEA, display complementary functionalities. Hierarchical clustering on filtered, centered, reduced, and aggregated datasets has the advantage of integrating all samples together into a single analysis and of providing a global overview of the homologies between cell subsets of various species (15, 17, 18, 74, 75). However, the integration of distinct datasets requires a cross-normalization procedure which consists in a rather profound mathematical transformation of the data. The normalization procedure artificially increases the variance for genes with only small differences in their initial signal intensities between the different cell types studied. Conversely, it comparatively decreases the variance for genes with high differences in their initial signal intensities between the different cell types studied. To limit the biases that this normalization introduces, it is thus necessary to select only the orthologous genes that vary strongly in their expression across the cell types examined within each species. Another corollary is that this analysis can only be applied to genes that have known orthologs in all species. If one ortholog is missing in only one species, the gene must be removed from the analysis. Hence, this method should be used with caution, only under conditions where dataset normalization does not yield too strong biases in gene-expression profiles. It is also not appropriate when the structures of the different datasets are too different (i.e., the number and potential identities of cell types vary too much across datasets), because the dynamic ranges of gene expression between datasets are not expected to be the same and should therefore not be forced to similarity. Even under conditions where the experimental design is favorable to the use of hierarchical clustering, GSEA ensures of the robustness of interpretation. GSEA has been used by us and others to perform cross-species comparisons (5, 19, 29, 42, 76–78). GSEA notably displays advantages and drawbacks distinct from those of hierarchical clustering. First, it is easier to perform GSEA since dedicated ready-to-use stand-alone programs are available which do not require bio-informatics expertise. Second, GSEA is more sensitive, notably to detect overlaps of common functions/gene networks between cell populations or cellular contaminations, as exemplified with sheep \*pDC enriched in human and mouse B cell fingerprints. This higher sensitivity is linked to (i) the fact that GSEA can detect coordinate regulation of gene modules (geneset-based approach) and thus does not rely on the strong regulation of few single genes (single gene-based approach), (ii) the fact that GSEA, when applied to multiple species, takes into account all genes that have orthologous counterparts in the considered species and is not restricted only to highly variable genes. Third, GSEA can perform cross-platform comparison without any cross-normalization thus without any supplementary artificial manipulation of the expression data. Finally, it can be performed on multiple datasets, even if their structures are different. However, GSEA presents the limitation of performing pairwise comparisons whose results can be integrated and visualized with our Bubble GUM software, but it nevertheless does not provide a global trans-species overview of subset homology. Overall, in order to increase confidence in the interpretation of the results, it is important to combine both approaches and verify that they both lead to consistent conclusions.

Our subset assignment methodology demonstrates similarity or proximity between subsets across species but not strict identity. Besides possible intrinsic transcriptomic differences between species, one of the reasons that explain this limitation is the process of subset identification itself, which makes use of different surface markers. Whenever possible, similar marker combinations were used such as CADM1 and CD172 that are known to be conserved markers across human, mouse, and sheep cDC subsets (42). However, mAb anti-CD11c did not exist for the initial gating in pig and the mAbs in the exclusion pool were not the same in pig and sheep. Moreover, existing marker combinations are not always specific and can lead to cross-contamination between different cell subsets. Indeed, the GSEA of the sheep \*cDC2 revealed that they may have been contaminated by pDC, despite our attempt to avoid this problem through exclusion of CD45RB-expressing cells. It remains possible that pDC expressing minimal levels of CD45RB were still present in the sorted \*cDC2 population, and not in the sheep cDC1 subset. However, since sheep \*cDC2 were found in the correct cDC branch of the hierarchical clustering, their contamination by pDC is likely to have been limited. Similarly, it is likely that the sorted sheep \*pDC include residual B cells, explaining the enrichment for the human B cell fingerprint at a level above expectation: indeed after exclusion of B cells with a pan-B cell marker, sheep \*pDC were selected with a mAb directed to CD45RB, which may react with residual B cells that have escaped the pan-B cell exclusion. Yet, sheep \*pDC still cluster with other species pDC, separately from B cells. In the case of pig, pDC were selected using markers not expressed by B cells and they displayed an enrichment for B cell fingerprints at a level encountered in GSEA analyses of mouse pDC (Figure 4 in Supplementary Material). Finally, our approach was able to demonstrate that *a priori* assignment of subset identity based on the expression of a few membrane markers could be wrong, like in the case of the pig \*cDC2. Moreover, our approach had the power to properly re-assign cell subset identity, demonstrating that pig \*cDC2 were actually homologous to mouse and human ncMo. Another laboratory analyzed the transcriptome of similar pig cells sorted as CD14low CD163high cells, but they could not assign them to classical nor to non-classical human Mo, due to differences in bio-informatics approaches in this study (79) and in ours.

Our study will help improving in the near future the toolbox available in each species for rigorous and consistent phenotypic identification of cell subsets, thanks to our identification of novel, conserved, and specific, combinations of surface markers for each cell subset, which should allow generating more appropriate staining reagents. For instance, fluorescently labeled recombinant XCL1 could theoretically be used in any species to rigorously identify and sort cDC1 (38, 41). In addition, cell surface proteins encoded by genes shown here to be selectively expressed in a conserved manner in specific subsets of mononuclear phagocytes represent new candidate markers to refine and homogenize phenotypic identification of these cells across species, such as LRP8, TSPAN13, NRP1, and SLC30A5 for pDC, FCGR2B, and CD200R1 for MoDC, SIGLEC8 and IGSF6 for cDC2, and CSF1R, TLR4, and C5AR1 for Mo/MP (**Table 3**). However, these potential new markers for subset identification need to be validated at the protein level.

The subset-specific signatures that are conserved throughout distant mammals included variable number of genes that were sometimes far lower than the numbers of genes in the human/ mouse common signatures. There are several explanations to this finding. There is a contribution of the very high stringency of the "Min (test) vs. Max (ref)" ≥1x method that we used to establish the signatures, since any gene which was not consistently found overexpressed in all the replicates of all the species was excluded. As an example, the gene *DNAJC7*, identified as specific of pDC in our previous work (15) was removed from the human pDC signature because its "Min (test) vs. Max (ref)" ratio was equal to 0.933, due to a single lower human pDC replicate compared to a single replicate found with a higher signal in human MoDC. There is also a contribution of incomplete mapping of the genome of some of the species studied, leading to an underestimation of the number of orthologous genes that could be queried across all species. For example, *POU2F2*, more highly expressed in human and murine B cells as compared to many other immune cells, has not been mapped yet to the pig genome while it has been mapped to the genome of more distant species such as the spotted gar with a 1-to-1 orthology relationship. Another prominent cause is linked to technical limitations of the microarray approach, such as lack of ProbeSets against certain genes in certain species. This is notably the case for the gene *CLEC9A*, known to be specific of cDC1 but for which no ProbeSet exists in the human Affymetrix HG U133 plus2 gene chip. Sometimes, low signal-to-noise ratio for certain ProbeSets can also be responsible for the loss of putative interesting signature genes, such as *ZNF521* (*Zfp521* in mouse) found to be highly specific of pDC in mouse and human while the pig and sheep orthologous ProbeSet remains at the background level whatever the cell type considered. Recent technological advances now allow performing high throughput RNA sequencing at single cell levels with high sensitivity and processivity, which could solve most of the above issues; indeed, all expressed genes should be detected without any bias and analysis at the single cell level should alleviate any issue of cross-contamination between cell types. Therefore, the generation of gene-expression data for many individual cells of the same type should increase statistical power to define genes co-expressed at the single cell level and defining cell type-specific transcriptomic modules (22). Single cell gene-expression profiling recently allowed the unbiased and *de novo* identification of the different cell types of spleen (80) and central nervous system (81, 82) via the description of their molecular identity, starting from the bulk population of all the cells that could be extracted from the organ, without any prior enrichment procedure, based on the use of potentially confounding phenotypic marker combinations. However, this strategy is



*a Exclusion with anti-CD3, anti-NK cells, and anti-B cell markers is desirable when appropriate tools are available.*

*b A combination of known markers including FLT3, MHC-II, CD11c, SIRP*α*, and CADM1 allows a first step of identification of subset candidates but is at risk of contamination by sister cell types, or may be incomplete due to non-availability of one of the marker. FLT3 labeling may be performed by using recombinant His-tag FLT3L generated for the relevant species as recently proposed in a review (21).*

*c New additional candidate markers for refinement of subset identification are derived from the identification of genes encoding cell surface molecules from the conserved cell subset gene signatures.*

still extremely difficult to apply to species which genome has not yet been completely assembled, as well as to very rare cell types recovered upon prior phenotype-based enrichment. Moreover, to obtain information of sufficient completeness on functionally important genes for which few mRNA are expressed per cell, it is necessary to sequence at a sufficient depth of about one million reads per cell, which today still represents a very high cost when multiplied by the number of individual cells and conditions. Finally, the interpretation of the RNA-seq data on single cells is still largely based on the transcriptomic/molecular identity of cell types that are deduced from microarray analysis of purified cell pools. Hence, our work constitutes a major advancement in the field and is a necessary step before an eventual, later, refinement of the definition of cell subsets and their associated molecular signatures using single cell RNA-seq. The canonical gene-expression signatures that we generated can be used to distinguish and identify cell subsets in other vertebrate species. The cDC1 signature and the cDC2 vs. cDC1 signatures could be evaluated in chicken cDC sorted as single cells to determine whether this population includes only cDC1, as suggested by the trans-vertebrate hierarchical clustering, or a mixture of cDC1 and cDC2.

The conservation of gene signatures and interacting gene networks in homologous cell subsets throughout evolution is likely to bear strong biological meaning. Indeed, many genes of the conserved signatures were already known for their functions in these cells, validating the biological relevance of our signatures. In several instances, the same functional annotations were enriched in distinct subset signatures, but the genes responsible for the enrichments differed. For example, the genes responsible for the enrichment of the pathway "role of pattern recognition receptor in recognition of bacteria and viruses" were *TLR4*, *TLR8*, and *C5AR1* for the Mo/MP vs. cDC signature, *TLR2*, *CLEC7A*, *IL1B*, and *PIK3CB* for the cDC vs. pDC signature, and *TLR8*, *CLEC6A*, *DDX58*, *OAS2*, and *IL1B* for the cDC2 vs. (pDC and cDC1) signature. This analysis shows that cDC and MP express different sets of pattern recognition receptors for detection of viruses and bacteria, and that, within DC, cDC2 are also equipped differently from cDC1 and pDC for sensing of viruses and bacteria. These observations extended to other mammalian species the previous reports that human and mouse cDC2 are preferentially equipped with PRR targeting bacteria or involved in cytosolic sensing of viral infection (83, 84), and that TLR4 is very weakly expressed on pDC and cDC as compared to Mo/MP (83, 85). Similarly, different subset signatures were all enriched for "inflammatory response," "inflammation of organs," and "bacterial infection" but due to different genes. Altogether, this analysis indicates that different mononuclear phagocyte subsets express distinct and specific gene-expression modules which can sometimes contribute in a complementary way to the same general biological process in a conserved manner throughout evolution. Within the conserved gene-expression programs in mononuclear phagocyte subsets, we identified novel candidate genes and putative upstream regulators which likely contribute to the control of the ontogeny or functions of the corresponding cell type. For instance, the *FNBP1* and *SNX22* encoded proteins may be involved in the specific intracellular trafficking properties promoting antigen cross-presentation by cDC1, *ARHGAP22* and *NAV1* could modulate the organization of the cytoskeleton of cDC to control their mobility or antigen presentation functions, and the transcription regulators *BCL11A* and *MSL2A* may control specific gene networks in cDC. *BLC11A* is known to be key in murine pDC development (50) but it may have a specific role in cDC homeostasis, as inferred from a previous study (86). Our study thus opens the way for deciphering the sets of genes encoding functional cellular modules and their specifying transcription factors in subsets of mononuclear cells, in order to further improve and connect together the molecular and functional definitions of these cell types across species (22, 23).

# Conclusion

Our meta-analysis that combines cell sorting and comparative transcriptomic analysis was implemented as a methodology pipeline that could be used by biologists with minimal training in bio-informatics for subsequent extension to other species and to other complex cellular systems. Our study should lead to the identification of homologous mononuclear phagocyte subsets in species other than sheep and pigs, and which are of importance for biomedical investigations, such as bats, rabbits, ferrets, guinea pigs, possibly zebrafishes, and in species of veterinary importance including pets and animals of the food economy. The characterization of mononuclear phagocyte subsets in these species will allow manipulating their immune responses against diseases for the sustainability of our environment.

# Author Contributions

ISC and MD directed research and wrote the paper with input from TPVM. TPVM carried out bio-informatics analyses with input from MD. JEY performed most of the cell purification experiments and analyzed data, with input from CU (microarray hybridization, blood processing), SR (pDC isolation), MB (cell sorting), MM (microarray hybridization and analysis), HM (chicken array data), PQ (chicken array data), NB (pig cell phenotyping). PB performed phylogenetic analyses. GF and HS provided key cell types (MoDC) and reagents (unique mAb, non-commercially available). LJ performed array annotations.

# Acknowledgments

The authors thank Chantal Kang and Michel Bonneau from the Centre d'Imagerie Interventionnelle for providing pig blood and the Unité Commune d'Expérimentation Animale du Centre de Jouyen-Josas for providing sheep blood and spleen. We thank Christelle Thibault-Carpentier from the Plate-forme Biopuces (Strasbourg, France) for performing one of the microarray experiments (http:// www-microarrays.u-strasbg.fr). This work was supported by institutional funding from Agence Nationale de la Recherche (ANR) PhyloGenDC, ANR-09-BLAN-0073-02. We thank the European Commission 7th Framework Programme (Initial Training Network FishforPharma, PITNGA- 2011-289209). TPVM was supported through the PhyloGenDC ANR grant and the European Research Council (FP7/2007-2013 Grant Agreement no. 281225 to MD for the Systems Dendritic project).

# References


# Supplementary Material

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fimmu. 2015.00299

of intestinal DC differentiation in humans and mice. *Nat Immunol* (2014) **15**(1):98–108. doi:10.1038/ni.2768


**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 Vu Manh, Elhmouzi-Younes, Urien, Ruscanu, Jouneau, Bourge, Moroldo, Foucras, Salmon, Marty, Quéré, Bertho, Boudinot, Dalod and Schwartz-Cornil. 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.*

# **Investigating evolutionary conservation of dendritic cell subset identity and functions**

*Thien-Phong Vu Manh1,2,3 , Nicolas Bertho<sup>4</sup> , Anne Hosmalin5,6,7,8 , Isabelle Schwartz-Cornil <sup>4</sup> and Marc Dalod1,2,3 \**

*<sup>1</sup> UM2, Centre d'Immunologie de Marseille-Luminy (CIML), Aix-Marseille University, Marseille, France, <sup>2</sup> U1104, Institut National de la Santé et de la Recherche Médicale (INSERM), Marseille, France, <sup>3</sup> UMR7280, Centre National de la Recherche Scientifique (CNRS), Marseille, France, <sup>4</sup> Virologie et Immunologie Moléculaires UR892, Institut National de la Recherche Agronomique, Jouy-en-Josas, France, <sup>5</sup> INSERM U1016, Institut Cochin, Paris, France, <sup>6</sup> CNRS UMR8104, Paris, France, <sup>7</sup> Université Paris Descartes, Paris, France, <sup>8</sup> Assistance Publique-Hôpitaux de Paris (AP-HP), Hôpital Cochin, Paris, France*

#### *Edited by:*

*Florent Ginhoux, Singapore Immunology Network, Singapore*

#### *Reviewed by:*

*Richard A. Kroczek, Robert Koch-Institute, Germany Muzlifah Aisha Haniffa, Newcastle University, UK*

#### *\*Correspondence:*

*Marc Dalod, Centre d'Immunologie de Marseille-Luminy (CIML), Parc Scientifique et Technologique de Luminy, Case 906, Marseille Cedex 09 F-13288, France dalod@ciml.univ-mrs.fr*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 27 February 2015 Accepted: 11 May 2015 Published: 02 June 2015*

#### *Citation:*

*Vu Manh T-P, Bertho N, Hosmalin A, Schwartz-Cornil I and Dalod M (2015) Investigating evolutionary conservation of dendritic cell subset identity and functions. Front. Immunol. 6:260. doi: 10.3389/fimmu.2015.00260* Dendritic cells (DCs) were initially defined as mononuclear phagocytes with a dendritic morphology and an exquisite efficiency for naïve T-cell activation. DC encompass several subsets initially identified by their expression of specific cell surface molecules and later shown to excel in distinct functions and to develop under the instruction of different transcription factors or cytokines. Very few cell surface molecules are expressed in a specific manner on any immune cell type. Hence, to identify cell types, the sole use of a small number of cell surface markers in classical flow cytometry can be deceiving. Moreover, the markers currently used to define mononuclear phagocyte subsets vary depending on the tissue and animal species studied and even between laboratories. This has led to confusion in the definition of DC subset identity and in their attribution of specific functions. There is a strong need to identify a rigorous and consensus way to define mononuclear phagocyte subsets, with precise guidelines potentially applicable throughout tissues and species. We will discuss the advantages, drawbacks, and complementarities of different methodologies: cell surface phenotyping, ontogeny, functional characterization, and molecular profiling. We will advocate that gene expression profiling is a very rigorous, largely unbiased and accessible method to define the identity of mononuclear phagocyte subsets, which strengthens and refines surface phenotyping. It is uniquely powerful to yield new, experimentally testable, hypotheses on the ontogeny or functions of mononuclear phagocyte subsets, their molecular regulation, and their evolutionary conservation. We propose defining cell populations based on a combination of cell surface phenotyping, expression analysis of hallmark genes, and robust functional assays, in order to reach a consensus and integrate faster the huge but scattered knowledge accumulated by different laboratories on different cell types, organs, and species.

**Keywords: mononuclear phagocytes, comparative genomics, human, non-human primates, mouse, pig, sheep, chicken**

# **Introduction**

The immune system includes a large variety of myeloid and lymphoid cell types which develop through distinct ontogenic pathways, express specific phenotypes, and exert specialized functions. The mononuclear phagocytes form a complex group of myeloid cells that encompass three major cell types, i.e., monocytes, macrophages, and dendritic cells (DC), together with their proximal progenitors. These three cell types contribute to maintain host integrity by shaping the innate and adaptive immune defense, a generic function related to their common phagocytic properties and their capacity to present antigen to T cells. These functions are also shared by other types of professional antigen-presenting cells (APCs), in particular B lymphocytes. However, different types of APCs are primarily devoted to distinct functions (**Figure 1**). B cells produce antibodies. Monocytes patrol the organism for the detection of pathogens and dominantly display inflammatory and oxidative stress response. Macrophages mainly perform microbicidal, scavenging, and tissue trophic/maintenance functions. DC are uniquely efficient for antigen-specific activation of naïve T lymphocytes, a process called T-cell priming. Indeed, DC were initially defined by their dendritic morphology and their exquisite capacity for T-cell priming. DC include two main cell types, the plasmacytoid DC (pDC) that are expert in type I interferon synthesis upon viral stimulation and the conventional DC (cDC) that are specialized in antigen capture, processing, and presentation for T-cell priming. Two cDC subsets can be distinguished based on a further segregation of functions. XCR1<sup>+</sup> cDC1 are particularly efficient in CD8<sup>+</sup> T-cell activation and crosspresentation, at least in mice. XCR1*−* cDC2 are most efficient for T helper cell priming, in particular polarization toward Th2 or Th17, and for the promotion of humoral immunity. Importantly, an additional layer of complexity is generated by the plasticity of the different mononuclear cell types, which display modified phenotypes and functions contingent to the anatomical microenvironment where they reside or when exposed to pathogens or inflammation. For instance, monocytes adopt a dendritic morphology and antigen-presentation functions in inflammatory settings (1–3) as well as when located in the dermis (4–6), leading to their designation as monocyte-derived DC (MoDC). Langerhans cells, long considered to be DC due to their morphology and antigen-presentation function, are now known as a type of tissue macrophages (7–13). More generally, the gene expression programs, phenotypes, and functional properties of macrophages are strongly influenced by their tissue of residence. Finally, not only XCR1<sup>+</sup> cDC but also other DC subsets including pDC and XCR1*<sup>−</sup>* cDC can also efficiently cross-present antigens to CD8<sup>+</sup>

T cells when appropriately stimulated (14–22). Thus, the plasticity of the mononuclear phagocyte responses superimposes onto the segregation of phenotypes and functions attributed to subsets (**Figure 2**), which can lead to confusion in the definition of the different cell types if only based on functional assays. Hence, morphologic, phenotypic, and functional criteria are not sufficient to rigorously define mononuclear phagocyte subsets, and to properly discriminate what are distinct cell types as opposed to different developmental or activation states of a given cell type. Complementary or robust alternative criteria are needed to rigorously define the identity of the mononuclear phagocyte subsets.

Mononuclear phagocyte subsets were recently shown to develop from distinct progenitors and/or under the instruction of different transcription factors or cytokines. cDC and pDC derive from a dedicated bone marrow precursor, the common DC progenitor, with a differentiation potential strictly restricted to this hematopoietic lineage. pDC and cDC homeostasis exquisitely depends on the growth factor FLT3-L. pDC development strictly depends on the transcription factors TCF4 (E2-2) and SPIB both in mouse and human, XCR1<sup>+</sup> cDC development on the master transcription factor IRF8 at least in mice, and XCR1*−* cDC development on IRF4. Macrophages derive from a monocytic precursor, either of embryonic origin as in the case of Langerhans cells and microglia, or at least in part from circulating blood monocytes as in the case of gut macrophages. Egress of classical monocytes from the bone marrow into the blood strictly depends on the chemokine receptor CCR2. As a consequence, in competitive mixed bone marrow reconstitution experiments in mice, all cell types derived from circulating blood monocytes are primarily reconstituted from wild-type cells and not from CCR2 deficient cells. Hence, it has been proposed that the study of their developmental pathway, in other words ontogeny, was the best way to classify mononuclear phagocyte cell types, at least in the mouse model where the knowledge in DC subset properties is also the most advanced. Indeed, in this model, genetically modified animals unambiguously permit to track the development of cell types and to dissect their phenotypes and functions, in different contexts *in vivo*. However, the identity and functions of the different mononuclear phagocyte subsets need to be established outside of the mouse model, in animal species where ontogenic studies cannot be easily conducted, in order to accelerate translation of our advanced knowledge on the functioning of the mouse immune

**FIGURE 1 | Different types of APCs are specialized in distinct primary functions**. cDC are uniquely efficient for the priming and functional polarization of T cells. Although other APCs also contribute to this process, this does not represent their primary functions. Hence, cDC play a central and non-redundant role in the orchestration of adaptive immunity.

system toward clinical and/or economical applications to sustain global human health. Very promising vaccine and immunomodulatory strategies have been developed in mouse models based on DC subset targeting (23–35). The translation of these strategies to human and other species has not yet reached the expected success, likely due to insufficient knowledge in the identity and function of homologous DC subsets across species. This knowledge is needed in biomedical model species, primarily in non-human primates, and also in alternative models such as pigs that share physiological and anatomical similarities with humans – for instance skin and lung structural properties – and that present sensitivity to human pathogens of great importance for public health such as influenza. In addition, this knowledge is needed for companion and sport animals, and for animals of the agro-economy, such as ruminants, pigs, poultry, and fishes, with the goal to improve vaccination strategies against pathogens responsible for major economic losses, to decrease antibiotic use and to ameliorate animal welfare. These species, as well as wild animals, are also targets or reservoirs for major zoonotic pathogens whose control could thus benefit from new vaccine strategies targeting DC subsets in these animal species. This raises the question how to best define DC subset identity and functions in a way that can be extrapolated from mouse to human and other species, for clinical applications as well as for a better understanding of the evolution of the immune system.

# **Different Methodologies to Define the Identity of Immune Cell Types, with Their Advantages and Drawbacks**

Several methodologies have been proposed to define cell types. They include cell surface phenotyping and morphology, ontogeny, functional characterization, molecular profiling at population level, and molecular profiling at single cell level. We will discuss the specific drawbacks and advantages of each of these approaches (**Table 1**).

# **Cell Surface Phenotyping and Morphology**

Cell surface phenotyping generally is a mandatory first step for all other proposed methodologies aiming at defining DC subsets. It may be skipped only for particular experiments of molecular profiling at single cell level and perhaps for functional tests based on validated protocols for specific depletion of the targeted cell subset *in vivo*. Indeed, phenotypic characterization/identification of DC subsets is necessary either to purify them for morphological analysis, functional assays, or molecular profiling, or to compare their characteristics in tissues or bulk cell suspensions (expression of lineage reporters in cell fate mapping experiments, anatomical location, maturation status, cytokine production, interactions with T cells*. . .*). Phenotypic characterization through cell surface phenotyping by flow cytometry is the method of DC subset identification the easiest to perform and the most frequently used. No single cell surface marker has been found to be sufficient for identification of a given DC subset, except for XCR1 expression on mouse and human XCR1<sup>+</sup> cDC (18, 36–42) and maybe BDCA2 or LILRA4 expression on human pDC (43–46). Thus, to rigorously identify any given DC subset in any species with a limited risk of contamination by another cell type, most of the time complex combinations of multiple markers are required, often including the use of exclusion marker to ensure lack of contamination of the cell population targeted by other cell types sharing with it many positive markers. For example, the CD8α <sup>+</sup> subset of mouse pDC can heavily contaminate mouse lymphoid organresident XCR1<sup>+</sup> cDC when defined phenotypically as Lineage*<sup>−</sup>* CD11c+CD8α <sup>+</sup> (47–49). This problem can be solved by exclusion of SiglecH<sup>+</sup> or CCR9<sup>+</sup> cells or by using XCR1 as a positive marker. Similarly, other cells including MoDC or activated CD1c (BDCA1)<sup>+</sup> XCR1*<sup>−</sup>* cDC can heavily contaminate human XCR1<sup>+</sup> cDC when defined phenotypically as Lineage*<sup>−</sup>* HLA-DR+CD141 (BDCA3)<sup>+</sup> (41, 50, 51). This problem can be solved by using CADM1 or XCR1 as additional positive markers (41, 52). Rigorous phenotypic identification of XCR1*<sup>−</sup>* cDC (mouse CD11b<sup>+</sup> cDC and human CD1c<sup>+</sup> cDC) can be much more challenging, since these cells can be difficult to discriminate from MoDC, in particular under inflammation settings (53, 54). Identification of DC based on oligoparameter phenotyping is even more at risk of inaccuracy in other species, due to the limited panel of available antibodies directed to surface markers and to the poor knowledge in surface marker expression selectivity in non-DC cell types. However, major advances have recently been made to refine strategies for DC subset identification by cell surface phenotyping, in part based on novel knowledge gained through ontogeny and molecular profiling studies as will be discussed below. Hence, protocols for DC subset identification by cell surface phenotyping might soon become standardized, at least in mouse and human. This would allow better comparison of data across laboratories and limit the risk of use of inappropriate protocols leading to improper data interpretation. Special attention should be given to enzymatic dissociation that can strongly modify cell surface marker detection. Ideally, universal phenotyping protocols could be designed, allowing to considerably simplify the current nomenclatures for DC subsets by using the same name and similar marker combinations to identify homologous cell types irrespective of their tissues and species of origin (55–57). Moreover, the markers used to define and name DC subsets could be chosen based on their relevance to the biology of these cells, contrary to the current situation where the markers used were discovered fortuitously/empirically and may not be linked to the biology of the eponymous cells, as is the case for CD8α and CD141 for mouse and human XCR1<sup>+</sup> cDC, respectively. However, when identifying a potentially new subset of DC or studying in a novel context a potentially known DC subset, a number of precautions need to be taken for data interpretation, including confirmation of conclusions by complementary methods such as ontogeny, functional, or molecular profiling studies.

# **Ontogeny**

Ontogeny studies in mice, in particular studies on the dependence of DC subset development on transcription factors, have been instrumental in identifying the homologies between lymphoid tissue-resident CD8α <sup>+</sup> cDC and the CD103+CD11b*<sup>−</sup>* cDC present in non-lymphoid tissues and migrating into the draining lymph nodes once activated (58). These studies, together with gene expression profiling analyses (9, 40), ultimately allowed grouping mouse CD8α <sup>+</sup> cDC and CD103+CD11b*<sup>−</sup>* cDC together under the umbrella of the XCR1<sup>+</sup> cDC subset (38, 40, 59, 60). The recent discrimination of mouse CD11b<sup>+</sup> cDC from MoDC has also been largely based on the analysis of the role of specific chemokine or growth factor receptors on cell type development *in vivo*, namely CCR2 dependence as a characteristic of monocytic origin and FLT3 dependence as a proof of cDC identity (2, 3, 6, 61). In addition, mouse CD11b<sup>+</sup> cDC development was shown to selectively depend on the IRF4 transcription factor (62, 63). Moreover, the establishment of the concept that mouse *bona fide* DC constitute a separate hematopoietic lineage, and the discrimination between mouse CD11b<sup>+</sup> cDC and MoDC, were confirmed using mutant animals allowing to track natural precursor–progeny relationships *in vivo* through irreversible fluorescent tagging of all daughter cells of a given type of hematopoietic progenitor, based on Cre-mediated conditional activation

#### **TABLE 1 | Different methodologies to define DC subsets with their advantages and drawbacks<sup>a</sup>** .


*<sup>a</sup>Advantages are indicated in bold font and drawbacks in plain font.*

of a floxed reporter gene under the control of the constitutive Rosa26 promoter, an experimental strategy-coined fate mapping (64). Based on the important contribution of ontogenic studies for rigorous delineation of the identity of mouse DC subsets and of their lineage relationships, it has been proposed to use ontogeny as a primary methodology for the classification of mononuclear cell subsets in all species (57). Recent methodological progress has now made rigorous ontogenic studies applicable to human DC subsets, by using surrogate models of DC development from human CD34<sup>+</sup> hematopoietic progenitors, either *in vitro* (41, 65, 66) or *in vivo* in alymphoid mice (66–68). Such approaches have allowed demonstrating remarkable similarities in the ontogeny of mouse and human DC subsets. For example, knock-down experiments performed by transducing human CD34<sup>+</sup> hematopoietic progenitors with shRNA-expressing lentiviral vectors allowed to show that human pDC development critically depends on the transcription factor SPIB including *in vivo* in humanized mice (67), and that human XCR1<sup>+</sup> cDC development depends on the transcription factor BATF3 *in vitro* but not *in vivo* in humanized mice (68). Moreover, the pathway for the development of human pDC, XCR1<sup>+</sup> cDC, and XCR1*<sup>−</sup>* cDC was very recently demonstrated to be similar to that described for mouse DC subsets, with the identification of the human homologs to the mouse common DC progenitor and pre-cDC (66, 69). The role of candidate genes susceptible to affect DC development can even be assessed *in vivo* in humans in the rare cases where patients have been identified with primary immune deficiencies resulting from natural mutations in such genes (70). Strategies are being developed to actively search for human primary immunodeficiencies affecting DC development as experiments of nature allowing deciphering the molecular mechanisms regulating this biological process (71). However, ontogenic studies will often not be applicable in human for rigorous assessment of the identity of DC subsets, for example when studying a potentially known DC subset in a novel physiopathological context, including characterization of the DC subsets present in steady-state non-lymphoid tissues (50) or infiltrating tumors and their draining lymph nodes (72, 73) or isolated from infected/inflamed tissues. In addition, rigorous ontogenic studies will be very difficult to perform in many species, because (i) precursor/progeny relationships remain very difficult to evaluate *in vivo* through cell fate mapping or cell transfer experiments, (ii) *in vivo* analysis of cell subset development dependence on growth factors or transcription factors cannot be reasonably done due to operational and/or financial reasons, and (iii) *in vitro* models of *bona fide* DC development are currently lacking (74). Hence, the use of other methodologies will be necessary to prove DC subset identity in these various conditions.

# **Functional Characterization**

Ideally, cell types should be defined based on the array of functions they can exert, because this definition links identity to function and is hence the most relevant to understand the functioning of the immune system and to harness the biology of DC subsets for improving health care of humans and of other species. In addition, cell type definitions based on their functional specialization could be the most universal across tissues and species. However, functional assays are often the hardest to perform experimentally and can be the most subject to variations depending on assays and experimental conditions. This is especially the case for assays aiming at comparing the ability of different DC subsets to activate T cells. If one aims at precisely comparing the cell-intrinsic ability of different DC subsets to process and present antigens, a number of potentially confounding factors must be taken into account to design the experiment in order to reduce the risk of inappropriate interpretation of results. Adequate steps must be taken to preserve the viability of DC subsets and control for it. This implies adding to each isolated DC subset the appropriate cytokines or growth factors necessary for their survival, for example GM-CSF for cDC and IL-3 for human pDC. For instance, sorted XCR1<sup>+</sup> cDC show a lower *ex vivo* survival as compared to XCR1*−* cDC in mice and sheep (75, 76). Sorting of DC subset by positive selections may affect DC subset responses due to antibody-mediated receptor stimulation (43, 77–79). This also implies including a positive control consisting in DC subsets pulsed with optimal epitopic peptides, to assess on antigen-specific T-cell priming by DC the impact of other factors than DC subset-intrinsic differences in antigen processing and presentation, not only differences in DC subset viability but also in delivery of co-stimulation or cytokine signals. In this regard, for a fair comparison between DC subsets, they should each be matured by stimulation with an appropriate adjuvant. PolyI:C is much more efficient than LPS for the activation of human XCR1<sup>+</sup> DC while it is the reverse for the activation of human MoDC. TLR7 or TLR9 ligands, but not TLR3 or TLR8 ligands, are potent activators of human pDC. Another layer of complexity is due to fundamental differences in the design of experiments in different species. While the gold standard for antigen processing and presentation assays in mice is the measurement of the activation of TcR-transgenic naïve T cells, this is not possible in other species where various surrogate readouts are used including antigen-specific re-activation of antigen-experienced T-cell clones or polyclonal T-cell lines or even proliferation of allogeneic T cells. It is known that significant differences exist in mice in the signals required for naïve T-cell priming, antigen-experienced T-cell re-activation, or allogeneic T-cell proliferation induction. Therefore, the same exact function is not fairly tested in different species. Furthermore, in species outside mice and humans, the use of epitopic peptide control requires to have accurate MHC typing and knowledge of the corresponding optimal peptides, which are generally unavailable. In addition, for accessibility reasons, the DC subsets used generally derived from different anatomic compartments depending on the species. For example, spleen DC subsets are often used in mice, blood, or tonsil DC in humans and lymph DC in sheep, which can further confound rigorous interpretation of the results when differences are observed between species. Finally, while inbred mice with defined sanitary status are generally used to limit the variability of the responses between individuals, this is not the case for other species including humans where the considerable heterogeneity in the genotypes, environments, and immune histories of individuals contribute to the strong variability of their responses (80). Hence, even for mouse experiments, there is a strong need for standardization of functional assays assessing the ability of DC subsets to process and present antigens and to functionally polarize T cells. Moreover, when attempting to compare DC subset functional specialization across two species, efforts should be made to use comparable experimental designs in both species. Thus, while functional characterization is highly desirable when identifying a potentially new subset of DC or studying in a novel context a potentially known DC subset, the identity of DC subsets must first be studied through alternative approaches measuring cell type-specific parameters that are less strongly influenced by the tissue microenvironment and the genetic or immune history of populations, and for which experimental protocols are relatively well standardized.

### **Molecular Profiling at the Population Level**

As the ontogeny and functions of cell types are instructed by specific gene expression modules, cell type identity can be defined by its molecular fingerprinting, including through gene expression profiling (81, 82). Homologous cell types between species can be defined as "those cells that evolved from the same precursor cell type in the last common ancestor" (82). This implies that homologous cell types must exhibit closer molecular fingerprints and gene expression programs than non-homologous cell types. Thus, it should be possible to decipher the identity of immune cell types of virtually all vertebrate species, by establishing their gene signatures and comparing them to the transcriptomic fingerprints of the well-characterized immune cell types of the mouse referent species. This is indeed an approach we pioneered to compare mouse spleen and human blood DC subsets (39) and later extended to comparison with sheep lymph cDC subsets (76), mouse DC subsets across tissues (40), as well as chicken spleen and pig skin mononuclear phagocyte subsets (83, 84). This approach allowed us to rigorously demonstrate for the first time to the best of our knowledge that human CD1c<sup>+</sup> cDC and CD141<sup>+</sup> cDC were homologous to mouse CD11b<sup>+</sup> cDC and CD8α <sup>+</sup> cDC, respectively (39, 85). This was later confirmed by us and others based on phenotypic, functional, and ontogeny studies (18, 37, 50, 65, 86). In addition, this approach permitted to show that cDC split into XCR1<sup>+</sup> and XCR1*<sup>−</sup>* subsets in migrating skin lymph DC in sheep, a species belonging to the Laurasiatherians, which is a mammalian order distant from the mouse and human Euarchontoglires (76). This approach also provided the first compelling evidence for existence of *bona fide* cDC and macrophages in chicken, showing that diversification in mononuclear phagocyte cell types appeared in a common ancestor to mammals and reptiles (83). Comparative transcriptomics also led to recognize CADM1 and SIRPα as surface molecules whose conserved expression throughout distant species can be used as a first phenotyping step to identify XCR1<sup>+</sup> and XCR1*<sup>−</sup>* cDC subsets in any mammal (76). Notably, CADM1 is a highly conserved molecule, presenting about 90% identity across mammalian orthologs, thus allowing using commercial anti-human CADM1 antibodies for cellular staining in distant species (76, 84). We found the *Xcr1* gene among genes specifically expressed in mouse spleen CD8α <sup>+</sup> DC when compared to a number of other immune cell types [see Supplementary Material "Additional file 5; gb-2008-9-1-r17-s5.xls" from Robbins et al. (39), specifically in the "CD8a\_DC\_gene\_signature" established from our microarray data and confirmed from our own re-analysis of the microarray dataset independently generated by Dudziak et al. (87)]. Specific expression of the Xcr1 protein on mouse lymphoid tissue-resident CD8α <sup>+</sup> DC and its functions were first unveiled in the pioneering report from the group of Kroczek (36), who showed that CD8<sup>+</sup> T-cell cross-priming depends on their ability to secrete the Xcr1 ligand Xcl1 in experimental models where either the OVA coupled to an anti-CD205 Ab or OVA-expressing allogeneic pre-B cells are administrated *in vivo*. Xcr1 expression on CD8α <sup>+</sup> DCs was also found to be critical for the optimal induction of CD8<sup>+</sup> Tcell responses upon *Listeria monocytogenes* infection (18). Importantly, comparative transcriptomics revealed XCR1 as a specific and universal marker for XCR1<sup>+</sup> cDC across tissues and species. This was initially shown in human, mice, and sheep (18, 37, 76) and subsequently in non-human primates and pigs (18, 37, 38, 40, 52, 59, 60). Altogether, these studies were critical for the current proposal of cDC subset classification into XCR1<sup>+</sup> and XCR1*<sup>−</sup>* cDC (38, 40). Many other recent studies have demonstrated the power of gene expression profiling to determine with a high degree of certainty the identity of mononuclear phagocyte subsets in a tissue where they had not been rigorously studied before or to identify homologous subsets of mononuclear phagocytes across species (5, 6, 8, 9, 50, 88–90). Importantly, standardized protocols for generation and analysis of gene expression data are routinely performed in many laboratories, platforms, or commercial companies in many countries. The corresponding costs have strongly decreased over the last decade and continue to go down. Hence, gene expression profiling at the population level is a very robust and reproducible methodology that is feasible in virtually all species where tools are available or can be developed to phenotypically identify and purify candidate cell subsets. However, potentially confounding factors must be taken into account to design experiments in order to reduce the risk of inappropriate interpretation of results (**Figure 3**). First and foremost, great care and rigor must be exerted in designing the experimental sampling protocol for cell subset purification, inasmuch as minor contamination by another cell type can dramatically impact the gene expression profile obtained. Hence, it is critical to carefully design the marker combination used to purify the different cell populations to be studied, and to control cell purity prior to the generation of the gene expression data. Second, to allow proper analysis of the gene expression profiles of the targeted cell type, appropriate cell type controls must be included, encompassing sister cell types as well as cell types that could be potential contaminants due to their expression of several of the markers used for positive selection of the targeted cell type. These controls are critical to allow assessing the risk of contamination by another cell type (49).

# **Molecular Profiling at the Single Cell Level**

Recent technological advances now allow performing high throughput RNA sequencing at single cell levels with high sensitivity and processivity. Transcriptomic analyses at the single cell level could solve most of the issues raised in the previous section for molecular profiling at the population level. Indeed, because it alleviates the necessity to purify cells on imperfect and potentially confounding phenotypic marker combinations, analysis at the single cell level should allow unbiased identification of potentially all cell types and their associated transcriptomic signatures. It also solves the issue of cross-contamination between cell types, since the identity of each single cell is established *a*

*posteriori* based on the analysis of its gene expression program. In addition, the generation of gene expression data for many individual cells of the same type should increase statistical power to define genes co-expressed at the single cell level and to define cell type-specific transcriptomic modules (81). As a proof-ofprinciple, single cell gene expression profiling recently allowed the unbiased and *de novo* identification of the different cell types of spleen (91) and central nervous system (92, 93) via the description of their molecular identity, starting from the bulk population of all the cells that could be extracted from the organ without any prior enrichment procedure. However, molecular profiling at the single cell level cannot be used without prior phenotype-based enrichment for very rare cell types, and it is difficult to apply to species in which genome has not yet been completely assembled. To obtain complete information, including on functionally important genes for which few mRNA are expressed per cell, it is necessary to sequence at a sufficient depth of about one million reads per cell, which today still represents a very high cost when multiplied by the number of individual cells and conditions. This is all the more the case since, likewise for molecular profiling at population level, correct interpretation of the data requires that sister cell types as well as cell types that could be potential contaminants are included in the experimental design. Moreover, the technology for single cell RNA sequencing is not yet democratized, since it is challenging both for sample preparation and for data analysis. For standardization of high quality sample preparation, commercial solutions exist but are very expensive. For data analysis, there is no consensus yet on how the data should be mathematically modeled for adequate removal of background signal and for discrimination of false negative signal due to sampling bias in the pool of the cell mRNA as opposed to true lack of gene expression. In addition, the interpretation of the RNA-seq data on single cells is still largely based on the transcriptomic/molecular identity of cell types that are deduced from microarray analysis of purified cell pools (91). Hence, molecular profiling at the population level currently represents a more sustainable strategy for most laboratories.

# **Recent Advances Brought by Comparative Transcriptomics at the Population Level for Defining the Identity and Functions of Mononuclear Phagocyte Subsets and Their Molecular Regulation**

In this section, we will review major advances brought forward by comparative transcriptomics at the population level for defining the identity and functions of mononuclear phagocyte subsets and their molecular regulation.

Gene expression profiling of cell types with apparent ambiguous phenotype or functions allowed to rigorously establish their identity, which could be achieved properly strictly contingent to their comparison with all candidate sister cell subsets as well as more distantly related cell types. Hence, we and others showed that human blood Lineage-CD16<sup>+</sup> cells are non-classical monocytes (39, 88) and not DC as was sometimes claimed (94–96). Similarly, analysis of human skin CD14<sup>+</sup> cell expression of the transcriptomic fingerprints independently established for cDC, monocytes, and macrophages provided critical evidence that these cells are monocyte-derived macrophages (5) while they were previously designated as DC (4). Transcriptomic analyses were also instrumental to demonstrate the homology of this human dermal cell type with the murine CD11b+Ly6C*−*CD64lo–hi (6) and pig CD163<sup>+</sup> (84) skin subsets. We were also able to show that cell populations claimed to correspond to novel cell types actually corresponded to a distinct differentiation or activation state of an already known cell type, for example establishing that the so-called interferon killer DC correspond to a particular activation state of NK cells (39). Furthermore, we showed that, upon many types of *in vivo* or *in vitro* stimulation, human and murine pDC and cDC undergo a remodeling of their gene expression program related to their plasticity, including induction of NFκB and IFN target genes, but still keep the canonical gene expression associated to their subset identity (41, 97). In particular, contrary to what other researchers hypothesized (98), gene expression profiling showed that activated pDC are not undergoing a cell fate conversion into a novel type of cDC (97).

Gene expression profiling also allowed aligning subsets of mononuclear cells across tissues (6, 8, 9, 40, 55, 99), establishing cell type homologies across species (5, 39, 50, 76, 83–85, 88, 89, 100), and rigorously examining the proximity of *in vitro*-derived subsets of mononuclear cells with those naturally existing *in vivo* (39, 41, 66, 101). These studies allowed significantly advancing the ontogeny and functional characterization of mononuclear phagocyte subsets based on the novel hypotheses that can be inferred from the analysis of the gene expression programs of the cells and from their comparison with other well-characterized cell types.

The study of the functional specialization of human DC subsets was strongly boosted by the demonstration of their transcriptomic homologies with mouse DC subsets (39, 85) which was recognized as a major breakthrough in the field (37, 53, 102–104) and acknowledged to have been impossible to draw from studies based on a limited set of molecular markers (105). In particular, this led to test whether human XCR1<sup>+</sup> cDC could be more efficient for cross-presentation than other human DC subsets. Even though the extent to which human XCR1<sup>+</sup> cDC are more efficient for cross-presentation than other human DC subsets is debated, the results from the functional studies performed independently by many teams concurrently demonstrate that these cells excel at cross-presentation of cell-associated antigens (18, 19, 37, 41, 86, 106) and of particulate antigens delivered through FcγR, through late endosomal targeting (21, 107) or upon polyI:C stimulation (18, 41, 86, 108). In addition, in sheep, the skin lymph migrating XCR1<sup>+</sup> cDC spontaneously displayed a higher efficiency of soluble antigen-presentation to specific CD8<sup>+</sup> T cells, as compared to XCR1*−* cDC (76).

Based on the demonstration of the striking transcriptomic similarities between mouse and human subsets of mononuclear cells, and on knowledge on the ontogeny of these cells in the mouse (109, 110), we proposed that, similar to their mouse counterparts, human pDC and cDC constitute a specific family of cells within the hematopoietic tree, should derive from a common progenitor with a DC-restricted differentiation potential, and could be derived *in vitro* from human CD34<sup>+</sup> progenitor cells in part under the instruction of the FLT3-L growth factor (39, 85), all of which was later confirmed experimentally (41, 65, 66, 69, 111, 112).

Very importantly, comparative genomics of immune cell subsets yielded conserved transcriptomic fingerprints for each of these cell types (39), a novel knowledge which considerably accelerated the deciphering of the molecular mechanisms regulating the development and functions of leukocytes as reviewed in **Table 2** (18, 36, 59, 100, 113–127). Finally, this approach uniquely allowed identifying conserved and biologically relevant cell surface markers for each subset of mononuclear cells which could enable considerably simplifying the nomenclature for DC subsets by using the same name and similar marker combinations to identify homologous cell types irrespective of their tissues and species of origin (55–57).

# **Conclusion and Perspectives**

While it might be the case in the future for single cell RNAseq, currently no single method is sufficient to allow the best possible classification of DC. Hence, ideally, all available methods (cell surface phenotyping, gene expression profiling, functional analyses, and ontogeny) should be combined together to define DC subset identity. However, such a combination of approaches cannot be used to define cell subsets in many instances due to technical, financial, or ethical limitations. Taking these limitations into consideration, the data reviewed here show that comparative transcriptomics at the population level is currently the most robust

**TABLE 2 | Genes which selective expression pattern in immune cell types was uncovered through comparative genomics and which functions in these cells were deciphered later**.


*<sup>a</sup>Transcriptomic signatures conserved between mouse and human unless specified otherwise, first reported in Robbins et al. (39), and encompassing the genes listed in this table.*

and feasible way to define the identity of cell types. Indeed, because the ontogeny and functions of cell types are instructed by specific gene expression modules, cell type identity can be defined in a universal and unbiased way by its molecular fingerprinting, including through gene expression profiling (81). However, due to its dependency on pre-selection of cell populations based on their expression patterns of a few cell surface molecules, gene expression profiling at the cell population level is imperfect and may require iterative steps of refined cell type isolation and gene expression profiling as illustrated in **Figure 3**. Hence, it is all the more important that each step of the procedure is performed and rigorously quality controlled according to the best standards in the field.

Cell purity is fundamental. It is important to design a sampling method specific for each study, through identification of the most robust criteria available in the current state of the art for purification of the target cell type based on phenotypic, morphologic, or anatomical characteristics. Cell enrichment is necessary for rare cell types among bulk populations. It relies on the depletion of other populations (MACS or EasySep™ for instance). The marker combination for negative selection must not unwillingly remove a population of interest. For instance, some antibody cocktails for human DC enrichment use anti-CD16 monoclonal antibodies, so as to deplete NK cells, but this should be proscribed for the study of non-classical, CD16<sup>+</sup> monocytes. Positive selection by magnetic or flow cytometry sorting is most often required after cell enrichment. Antibody labeling must be clear-cut, with separate peaks and/or selection of the events with the highest labeling and the lowest potential contamination by other populations. This selection implies the use of marker combinations specific for the population of interest, since specific markers are rarely available. XCR1 is a rare instance of a conserved marker so far only expressed on a discrete DC population. To the best of our knowledge, reliable commercial reagent are available for XCR1 staining only for mouse and rat, but XCR1 staining can also be achieved with fluorescently labeled recombinant XCL1 (40, 41, 52), a strategy that is amenable to many species in which XCL1 sequence is known. CLEC4C alias BDCA2 and LILRA4 alias ILT7 are specific markers for human pDC, but their engagement induces inhibitory signals which for instance reduce pDC production of type I interferons after stimulation (43, 77–79, 131). Although selectively expressed at high levels on human pDC in the blood or lymphoid organs under steady-state conditions, NRP1 alias BDCA4 can be induced on activated cDC and is also expressed on other cell types including neurons, endothelial cells, and tumor cells (132, 133). CD123 is a good marker to help identifying pDC in non-human primates, but it also labels mastocytes which are present in the blood or in lymphoid organs (134). Cell purity must be controlled in each experiment, by flow cytometry re-analysis just after sorting, and as one of the first step of transcriptomic analysis by examining the expression of negative and positive control genes (expression of genes that should be expressed only on other populations including potential contaminants, and expression of genes characteristic for the population of interest including but not restricted to genes coding for the molecules used for positive selection) (**Figure 3**).

The quality and quantity of mRNA must be adequate, even when cell numbers are low. RNA extraction kits adapted to low cell number samples may be required. mRNA quality must be controlled by electrophoresis. A linear amplification protocol must be used, that has been validated for yielding results from low input RNA showing a strong correlation with the results obtained with higher RNA input and a classical amplification procedure.

For bioinformatics analyses, the dataset must include sister cell types as well as the cell types the most likely to contaminate the cell type of interest, or at least be compatible for integrative analysis with a reference dataset including these control populations. Several independent methods for data analysis should be used, to ensure robustness of interpretation. Beyond relative classification of the cell types of the dataset by classical approaches computing the overall distance between their gene expression programs as performed by hierarchical clustering or principal component analysis, the identity of cell types can also be reliably inferred from the analysis of their relative expression of robust cell type-specific gene signatures established from re-analysis of public gene chip databases and/or from published articles.

Novel advances are being brought through molecular profiling of subsets of mononuclear cells. In addition to steady-state conditions, populations can be analyzed after stimulation to identify the specific activation pathways elicited in pure cell populations or upon interaction between different cell types (41, 97, 135–137). In addition to unbiased analysis of the cellular composition of different organs (91, 93), transcriptomic profiling at the single cell level will allow studying heterogeneity in gene expression within one cell type with the hope to link it to functional heterogeneity (138) and eventually with the former history/epigenetic imprinting of each cell. Comparative transcriptomic studies allowed us and others to identify in humans, non-human primates, pig, sheep, and chicken cDC subsets homologous to those well described in mice (5, 18, 39, 50, 52, 62, 76, 83–85). These studies suggest that similar cDC subsets already existed in the last common ancestor of birds and mammals. Conserved gene modules appear during evolution to elicit new functions (81, 82). For instance, regarding T helper lineage diversification during evolution, contrary to bony fishes, the elephant shark, a cartilaginous fish, has been reported to lack genes encoding for critical transcription factors or cytokines instructing the development or involved in the functions of Th2, Th17, and Treg cells, such as RORC and FOXP3, IL-4, IL-21, IL-23, and IL-2 (139). This suggests that the genes required for the development of the different T helper lineages might have appeared progressively as modules during evolution starting in bony fishes and with late development of the Treg and Th17 lineages (81). Comparative genomics of mononuclear phagocyte subsets and single cell gene expression profiling will critically help identifying novel gene modules and their associated immune functions. In pDC, evolutionarily conserved co-expression of *TCF4*, *RUNX2*, *TLR7*, *TLR9*, *UNC93B1*, *MYD88*, *IRAK4*, *IRF7*, and *PACSIN1* might represent part of a gene module instructing the functional specialization of this cell type in high level production of type I interferon in response to sensing of oligonucleotide sequences of viral or autologous origin. In XCR1<sup>+</sup> cDC, evolutionarily conserved co-expression of *CLEC9A*, *SYK*, *RAB11A*, *RAB7B*, *SEPT3*, *SNX22*, *TLR3*, *CADM1*, and *XCR1* might represent part of a gene module instructing the functional specialization of this cell type in CD8<sup>+</sup> T-cell activation and specifically in cross-presentation of cell-associated antigens. In any case, the discovery of the sets of genes that are tightly co-expressed in DC subsets across various tissues and species, not only at the population level but also at the single cell level, should allow identifying the gene modules instructing DC subset functions. Characterization of the members of these gene modules which role in DC is unknown yet should strongly contribute to increase our knowledge on DC subset functional specialization and their molecular regulation. Of note, not all of these gene modules might harbor the same differential pattern of expression between DC subsets in different animal species. Some functions have gained or lost expression in specific cell subsets in some species which should correlate with similar changes in the expression patterns of the corresponding gene modules. For instance, IL-12 is produced both by pDC and cDC in mice, but only by cDC in humans, while antigen cross-presentation appears to be more strongly associated with XCR1<sup>+</sup> cDC in mice than in humans (18, 19, 22). Isolation and comparison of mononuclear phagocyte subsets from homologous organs in different species may help understand how the anatomical compartmentalization of these cells is established and affects their functions, including local interaction with specific cell types and chemokines. Dating when during evolution pDC as well as classical and non-classical monocyte subsets appeared, and in which anatomical compartments they reside in the species the most distant to humans and mice, may give novel insights into the core functions of these populations.

*In vivo* manipulation of DC can promote and orient immune responses based on the intrinsic functional properties of the DC subset targeted and can be advantageously used for prophylactic vaccination or immunotherapy against cancer or infections. This strategy can benefit from the knowledge gained from the expression profiling of DC subsets and their alignment across species. Notably, based on their homology with mouse XCR1<sup>+</sup> cDC, human XCR1<sup>+</sup> cDC can be considered as a promising target when cross-presentation is desirable, in particular for fighting cancer or infections by intracellular pathogens (23, 24, 29, 72, 73, 140– 143). Moreover, because it is specifically expressed in XCR1<sup>+</sup> cDC in a conserved manner in evolution, and it has been successfully used for *in vivo* delivery of antigens specifically to XCR1<sup>+</sup> cDC to vaccinate mice (23, 24), XCR1 can be considered for a universal DC targeting strategy in potentially all vertebrate species. Interestingly, the targeting of XCR1 can be achieved with targeting units composed of recombinant XCL1 fused to protective antigens in the form of vaccibodies (24), a strategy that is amenable to many species in which the XCL1 sequence is known. Although more broadly expressed in the DC lineage at least in mice, CLEC9A is also an interesting target since it directly promotes crosspresentation of the material it binds, probably by delivering it into appropriate endosomes (144, 145), and because it is selectively expressed to high levels on XCR1<sup>+</sup> cDC in humans, sheep, and mice (25, 32, 76, 146) although it may not be the case in some other species such as pig. Arguments in favor or against the targeting of XCR1<sup>+</sup> cDC in the clinic are summarized in **Table 3**. The identification of XCR1<sup>+</sup> cDC in companion and sport animals, and in animals of the agro-economy, such as ruminants, pigs, poultry, and fishes, will allow designing better vaccines to protect



*<sup>a</sup>More details and bibliographical references can be found in the main text of this review.*


**TABLE 4 | Practical guidelines for consistent definition of DC subsets across mouse and human tissues with potential applicability to other mammals**.

*<sup>a</sup>Master transcription factors critical for cell subset development are indicated in bold font.*

them against infections in order to ameliorate animal welfare and to prevent pandemics causing severe economic losses. It will also contribute to a global public health strategy because some of these animal species as well as wild animals are targets or reservoirs for major zoonotic pathogens. The identification of XCR1<sup>+</sup> cDC in rhesus macaques and in pigs opens the way to preclinical vaccination studies in these species which are close to humans. Vaccibodies based on XCL1 dimers coupled to influenza or SIV proteins are planned to be used for vaccination of pigs or rhesus macaques, respectively, and induction of immune responses and protection against infection. pDC targeting could also be considered as an interesting alternative for vaccination against viruses or tumors (20, 147, 148), or for the induction of cross-tolerance to treat autoimmune diseases or food allergies (149, 150).

A synthetic list of phenotypic, transcriptomic, and functional hallmarks which have already allowed conserved identification of different DC and monocyte subsets in humans and mice is presented in **Table 4**. The present Special Issue and future workshop on DC nomenclature will help reach a consensus panel for practical definition of the populations, in order to integrate faster the huge, but scattered knowledge accumulated by different laboratories in different cell types, species, and organs.

# **References**


# **Acknowledgments**

The studies performed in the laboratories are supported by funding from INSERM, CNRS, INRA, a CNRS-AP-HP collaboration, Aix-Marseille University, Université Paris Descartes, Université Paris Diderot, ANRS, the Agence Nationale de la Recherche (ANR) (PhyloGenDC, ANR-09-BLAN-0073-02 and DCskin-VacFlu, ANR-11-ISV3-0001), the "Integrative Biology of Emerging Infectious Diseases" Labex (ANR-10-LABX-62-IBEID), the DCBIOL Labex (ANR-11-LABEX-0043, grant ANR-10-IDEX-0001-02 PSL\*), and the A\*MIDEX project (ANR-11-IDEX-0001-02) funded by the French Government's "Investissements d'Avenir" program managed by the ANR, as well as the European Community's Seventh Framework Programme FP7/2007–2013 (European Research Council Starting Grant Agreement number 281225 for MD including salary support to TPVM). We thank past and present members of our laboratories and Dr. Rémi Cheynier for their contribution to studies on DC and for discussions. Schematic representations of cells are adapted from Servier Medical Art Powerpoint image bank (smart.servier.fr/servier-medicalart). We apologize for not quoting certain studies because of space limitations.

dendritic cells initiate and maintain T helper 2 cell-mediated immunity to house dust mite allergen. *Immunity* (2013) **38**(2):322–35. doi:10.1016/j. immuni.2012.10.016

3. Langlet C, Tamoutounour S, Henri S, Luche H, Ardouin L, Gregoire C, et al. CD64 expression distinguishes monocyte-derived and conventional dendritic cells and reveals their distinct role during intramuscular immunization. *J Immunol* (2012) **188**(4):1751–60. doi:10.4049/jimmunol. 1102744


phenotype and function but modified by donor health status.*J Immunol*(2011) **186**(11):6207–17. doi:10.4049/jimmunol.1002632


specialization and a common molecular signature in distant mammalian species. *J Immunol* (2010) **185**(6):3313–25. doi:10.4049/jimmunol.1000824


effector T cells via the cytokine TGF-beta. *Immunity* (2013) **38**(4):818–30. doi:10.1016/j.immuni.2013.03.004


**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 Vu Manh, Bertho, Hosmalin, Schwartz-Cornil and Dalod. 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.*

# **Blood monocytes and their subsets: established features and open questions**

#### *Loems Ziegler-Heitbrock\**

*Helmholtz Zentrum München, Asklepios Fachkliniken München-Gauting, Gauting, Germany*

In contrast to the past reliance on morphology, the identification and enumeration of blood monocytes are nowadays done with monoclonal antibodies and flow cytometry and this allows for subdivision into classical, intermediate, and non-classical monocytes. Using specific cell surface markers, dendritic cells in blood can be segregated from these monocytes. While in the past, changes in monocyte numbers as determined in standard hematology counters have not had any relevant clinical impact, the subset analysis now has uncovered informative changes that may be used in management of disease.

#### *Edited by:*

*Florent Ginhoux, Singapore Immunology Network, Singapore*

#### *Reviewed by:*

*Anne Hosmalin, Cochin Institute, France Muzlifah Aisha Haniffa, Newcastle University, UK Claudia Jakubzick, National Jewish Health, USA*

#### *\*Correspondence:*

*Loems Ziegler-Heitbrock, Helmholtz Zentrum München, Asklepios Fachkliniken München-Gauting, Enzianstr.3, 82211 Herrsching, Germany lzh@monocyte.eu*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 28 May 2015 Accepted: 03 August 2015 Published: 17 August 2015*

#### *Citation:*

*Ziegler-Heitbrock L (2015) Blood monocytes and their subsets: established features and open questions. Front. Immunol. 6:423. doi: 10.3389/fimmu.2015.00423* **Keywords: monocyte subsets, nomenclature, classical monocytes, intermediate monocytes, non-classical monocytes**

# **The Definition of Monocytes**

The term monocyte is used for blood cells of a lineage called monocytes/macrophages or mononuclear phagocytes. These blood monocytes are bone marrow-derived leukocytes that are functionally characterized by the ability to phagocytose, to produce cytokines, and to present antigen. In early studies, they had been identified based on glass adherence and morphology (1). Also, cytochemistry for specific enzymes like monocyte-specific esterase (2, 3) has been employed, while the standard approach in clinical hematology relies on physical properties of these cells including light scatter.

In bone marrow, the monocytes derive from myelo-monocytic stem cells, which give rise to more direct precursors like monoblasts and pro-monocytes. These cells earlier were identified based on morphology (4) such that the monoblast was an ill-defined cell type. More recently in the mouse model, a Ly6C+ CD115+ CD117+ monoblast-type cell, termed common monocyte progenitor (cMoP), was identified in bone marrow and spleen and this cell is able to proliferate and give rise to the different monocyte subsets (5). A cMoP monoblast type of cell remains to be identified for man and other species.

The number of circulating blood monocytes in man can strongly increase within minutes by stress or exercise followed by a rapid return to baseline levels. These recruited cells are thought to come from what is called the marginal pool (6). This compartment describes areas of reduced blood velocity close to the endothelium of venules and here cells can loosely adhere and can be mobilized in a catecholamine-dependent fashion (7). These marginal pool monocytes can have an adhesion molecule pattern distinct from monocytes found in blood at rest.

In addition, CD11bhigh (CD90, B220, CD49b, NK1.1, Ly-6G, F4/80, I-Ab, CD11c)low cells are mobilized from the spleen after severe injury (8). These cells have monocyte morphology and their transcriptome matches with that of blood monocytes. Furthermore, CD11b+ Ly6Chi monocytes can be mobilized from bone marrow to blood in infectious disease models (9), and adoptively transferred monocytes were shown to return to the bone marrow (10) in the mouse. What remains to be determined is whether the spleen and bone marrow compartments also contribute to the pool of monocytes that can be mobilized by stress and exercise.

When under homeostatic or inflammatory conditions, the monocytes have migrated into tissue; then by definition, these cells are called macrophages. Cells newly emigrated into the lung have been termed monocytes in some studies [e.g., Ref. (11)]. Since monocytes, once they have arrived in tissue, will start to transform into larger cells and rapidly lose their monocyte characteristics, others have called these recently emigrated cells "small macrophages" (12).

More detailed studies in the mouse have demonstrated tissue cells with characteristics close to blood monocytes (13, 14). However, these cells in the lymph node show a gene expression pattern that distinguishes them from the blood cells (14) and in the skin they show increased expression of lysozyme and CD68, markers typical of mature macrophages (13). Therefore, more data are required in the mouse model and obviously also in man before a consensus can be reached whether we use the term tissue monocyte or whether we continue to call these cells macrophages. Until these issues have been resolved, the term monocyte should be restricted to cells in the blood compartment and the bone marrow and spleen reservoirs that can replenish the blood monocyte pool.

# **Definition of Blood Monocytes Based on Cells Surface Markers**

As explained above, monocytes initially had been identified by function and morphology and these criteria have been misleading especially when disease processes altered these features. Therefore, attempts have been made to define unequivocal criteria for monocytes. Here, monoclonal antibodies against cell surface molecules have been proposed. In man, CD14 has been used as a marker (15), and in the mouse, CD115 is often employed (16). CD115 identifies the M-CSF receptor and has the main drawback that in the mouse, it is downregulated on blood monocytes with inflammation (17). Also, the question is whether such markers are sufficiently specific and do not react with other cell types like dendritic cells (DCs). In fact, part of the CD1c+ blood DCs in man can express low-level CD14 (18) and also human B cells have been reported to express some CD14 (19). Therefore, monocytes can be identified with markers like CD14 and CD115, but this should be supported by additional markers and by functional studies. Interestingly, when searching for macrophage-specific transcripts in the mouse, CD64 and MerTK have emerged (20). While CD64 is absent from non-classical monocytes in man, MerTK is a molecule that might prove informative for blood monocytes in different species. In addition, staining for CD16, which is used for monocyte subset definition (see below), will at the same time help to exclude DCs in human blood.

# **Dissection of Monocytes from Dendritic Cells**

Dendritic cells were first described by Steinman and Cohn as stellate cells isolated from mouse spleen (21). Over the years, there have been debates as to whether these cells are a distinct lineage or part of the mononuclear phagocyte system. A common precursor for monocytes and DCs was described in the mouse (22), but the existence of this cell was later disputed (23) suggesting that DCs and monocytes may diverge at an earlier multi-potent progenitor stage (24).

However, the demonstration that monocytes can be used to generate DCs *in vitro* by adding GM-CSF and IL-4 suggested a close relationship between monocytes and DCs (25). Later, transcriptome analysis demonstrated that such monocyte-derived DCs rather resemble macrophages than DCs from lymphoid tissue (26). Therefore, these *in vitro* generated monocyte-derived cells are potent antigen-presenting cells, but they do not represent *bona fide* DCs; they rather belong to the monocyte/macrophage lineage. Still not resolved is the question whether in tissue the monocyte-derived cells with high levels of class II expression and with high antigen-presenting capacity should be termed monocyte-derived DC (13, 27, 28) or activated macrophages.

In addition to DCs in tissue, cells with DC properties have been described in blood based on the expression of CD68, CD1c, or CD141 (29, 30). Transcriptome analysis has demonstrated that these cells and the monocytes belong to different clusters (26, 31). These data suggest that blood DCs can be segregated from monocytes and macrophages as a separate lineage.

The data also demonstrate the power of transcriptomic analysis in defining and dissecting leukocyte populations like monocytes and DCs. Ontogeny can help in such a definition, but in men, adoptive transfer is limited to strategies like transfer of bone marrow stem cells, and informative mutations are rare. Also, the ontogeny approach needs to be used with caution since a defined progenitor cell can give rise to clearly distinct cell populations. An informative example is the megakaryocyte–erythrocyte progenitor (MEP) cell, which gives rise to either megakaryocytes and their platelet progeny or to erythroblasts and their red blood cell progeny (32, 33). Megakaryocytes and erythroblasts have a distinct transcriptome (34), and they are involved in distinct functions, i.e., in blood clotting and oxygen transport, respectively. Therefore, although having a common ontogeny, these cells belong to clearly separate lineages. This example illustrates that ontogeny can provide a framework, but a comprehensive analyses like transcriptomics and the analysis of cell function are required for dissecting cell types and for developing a nomenclature. Therefore, in order to assign a novel leukocyte population in blood or tissue to either monocytes or DCs, a straight-forward approach is to analyze the transcriptome (and other omics like the proteome, lipidome, glycome, or metabolome) of these cells in comparison to typical monocytes and DCs and to then ask whether the novel cell type co-clusters with either prototypic monocytes or DCs (26).

# **Monocyte Subpopulations**

Evidence for monocyte subpopulations has come from experiments using differential flotation in counter-current elutriation (35) and from differential binding to antibody-coated red blood cells, which has defined populations with different functions (36). With the use of monoclonal antibodies and flow cytometry, tools have become available to clearly define, enumerate, and isolate monocyte subsets based on the differential expression of CD14 and CD16 cell-surface markers (37).

In 2010, an international consortium under the auspices of the IUIS and the WHO has proposed a nomenclature for monocyte

subpopulations (38). The proposal defined the major population of CD14high cells found in human blood as classical monocytes and the minor population of cells with low CD14 and high CD16 as non-classical monocytes. A population in between these two subsets was termed intermediate monocytes (see **Figure 1**).

While an unequivocal approach to defining the intermediate monocytes has not been developed, as yet (39), a host of studies on intermediate monocytes has been published since the 2010 proposal. In fact, a search for the term "intermediate monocyte" under Google Scholar has revealed more than 100 studies on these cells since 2010. These reports have described an expansion of intermediate monocytes in various inflammatory diseases and these cells have been shown to be of prognostic relevance in cardiovascular disease (40). The use of additional markers for delineation of intermediate monocytes has been suggested (41) and it remains to be shown whether markers, such as CCR2 or slan, will improve the definition of these cells.

The same nomenclature as proposed for man can be used in other species [reviewed in Ref. (42)]. The respective cells can be very similar to men as seen for non-human primates (43, 44). In species like the mouse, the classical and non-classical monocyte subsets can be identified as well, but different markers like CD115, Ly6C, and CD43 are used (16, 45). Also in species like rat, pig, cow, and horse, classical and non-classical monocytes can be defined and even intermediate monocytes have been described in some animals (42). It is predicted that the nomenclature of monocyte subsets will be applicable to all mammalian species.

In human blood, a population of slan-positive cells has been described as DCs, but phenotypic analysis has shown that these

# **References**


cells are CD14-low and CD16-high (46), functional studies demonstrated a high capacity to produce TNF (47), and clinical studies showed that these cells are depleted by glucocorticoid treatment (48). These features are identical to what has been reported as characteristics of non-classical CD14+CD16++ monocytes (37, 49, 50). Also, the increased absolute numbers of slan-positive monocytes and of non-classical monocytes show a clear correlation in HIV-infected patients (51), and part of the non-classical monocytes has been shown to be slan-positive (52–54). Collectively, these findings suggest that the slan-positive cells belong to the non-classical monocytes.

There may be additional monocyte subsets including Fcepsilon-RI-positive cells (55), which were found with a median of 2.5% among CD14-positive blood monocytes in a pediatric cohort (56) and these cells may be involved in IgE clearance (57). Also, proliferating monocytes have been described (58) as well as precursors for fibrocytes (59) and osteoclasts (60). For all of these cell types, further characterization is awaited.

# **Clinical Implications of Monocyte Numbers**

Monocyte numbers as defined in the hematology lab using light scatter properties have not contributed much to diagnosis and monitoring of disease, but with the definition of monocyte subsets by flow cytometry, informative patterns have emerged. For example, severe infection will increase the number of non-classical and intermediate monocytes (61–63). Here, it remains to be analyzed whether such an increase can predict prognosis, as has been suggested (64). Furthermore, therapy with glucocorticoids leads to a decrease of non-classical monocytes, which appears to be due to a selective induction of apoptosis in the non-classical monocytes while classical monocytes even increase in number under glucocorticoids (50, 65). Also, blockade of the M-CSF pathway can lead to depletion of non-classical monocytes (66–68). A likely explanation is that M-CSF signaling via the CD115 M-CSF receptor is required for the classical monocytes to mature into non-classical monocytes. Again still to be determined is whether such a druginduced depletion can be used to predict therapeutic response in inflammatory diseases. Still unresolved is the mechanism of depletion of non-classical monocytes in three siblings within one family (69). Here, more families with this type of defect need to be analyzed in order to identify the gene and the mechanisms involved. Finally, the absolute count of intermediate monocytes was shown to predict cardiovascular events (70, 71). Hence, analysis of monocyte subsets by flow cytometry now provides clinically useful parameters in various settings. What remains to be established in this context is an unequivocal dissection of the non-classical and the intermediate monocytes.


<sup>4.</sup> Goud TJ, Schotte C, van Furth R. Identification and characterization of the monoblast in mononuclear phagocyte colonies grown in vitro. *J Exp Med* (1975) **142**:1180–99. doi:10.1084/jem.142.5.1180

molecule expression and interaction with the peripheral blood pool. *Scand J Immunol* (1996) **44**:461–9. doi:10.1046/j.1365-3083.1996.d01-334.x


**Conflict of Interest Statement:** The author declares 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 Ziegler-Heitbrock. 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.*

## **Mélanie Durand1,2 and Elodie Segura1,2\***

<sup>1</sup> Centre de Recherche, Institut Curie, Paris, France

2 INSERM U932, Paris, France

#### **Edited by:**

Florent Ginhoux, Singapore Immunology Network, Singapore

#### **Reviewed by:**

Francesca Granucci, University of Milano-Bicocca, Italy Kristen J. Radford, University of Queensland, Australia

#### **\*Correspondence:**

Elodie Segura, INSERM U932, Institut Curie, Pavillon Pasteur, 26 rue d'Ulm, Paris Cedex 05 75248, France e-mail: elodie.segura@curie.fr

Dendritic cells (DCs) initiate and orient immune responses and comprise several subsets that display distinct phenotypes and properties. Most of our knowledge of DC subsets biology is based on mouse studies. In the past few years, the alignment of the human DC network with the mouse DC network has been the focus of much attention. Although comparative phenotypic and transcriptomic analysis have shown a high level of homology between mouse and human DC subsets, significant differences in phenotype and function have also been evidenced. Here, we review recent advances in our understanding of the human DC network and discuss some remaining gaps and future challenges of the human DC field.

**Keywords: human, dendritic cells, antigen presentation, DC subsets, ontogeny**

### **INTRODUCTION**

Dendritic cells (DCs) have long been known to be the most efficient antigen-presenting cells. It is now well established that DCs are a heterogeneous population composed of several subsets that can be distinguished by their phenotype, location, and functional properties (1). Due to their remarkable ability to stimulate T cells, DCs have become in the past decade attractive therapeutic targets. However, most of our knowledge of DC subsets biology was gained from mouse studies, and cross-species differences could hinder the successful translation to humans of major discoveries made in the mouse. In the past few years, a number of studies have tackled the analysis of human DC subsets. In this review, we summarize recent advances and highlight some of the outstanding questions that remain to be addressed.

#### **HOW TO DEFINE DC SUBSETS IN HUMANS?**

Historically, human DC subsets have been defined based on a small number of phenotypic markers, within the population of MHC class II<sup>+</sup> lineage-negative cells. In blood, DCs have been divided into two main groups: plasmacytoid DCs (pDC) and "myeloid" or "classical" DCs (cDCs). cDCs can be further separated into two subsets that are usually referred to as BDCA1/CD1c<sup>+</sup> DCs and BDCA3/CD141<sup>+</sup> DCs (2). These three DC populations are also found in all lymphoid organs and represent resident DCs (3–6). In skin, liver, lung, and intestine, two main populations of CD1c+CD1a<sup>+</sup> DCs and CD141+Clec9A<sup>+</sup> DCs have been identified (7–12). Tissue DCs can migrate through the lymph to the draining lymph nodes where these migratory DCs display a mature phenotype (4, 13, 14). Additional DC subsets have been described in mucosal tissues: Langerhans cells (LCs) and CD14<sup>+</sup> DCs (15, 16) in skin and vaginal mucosa, and CD103−CD172a<sup>+</sup> DCs in the intestine (10). Finally, a population of "inflammatory" DCs with a distinct phenotype can also be found in inflamed tissues (17, 18).

Although surface markers are useful for the characterization of DC subsets (**Table 1**), phenotypic analysis has proven insufficient on its own to define DC subsets. Indeed, some phenotypic markers are not specific of a given DC subset or their expression can change upon activation, potentially leading to misinterpretation. For instance, CD141 is upregulated upon activation on pDC and CD1c<sup>+</sup> DCs (19) and is also expressed by tissue CD14<sup>+</sup> DCs (20). Clec9A, which is restricted to CD141<sup>+</sup> DCs, is downregulated rapidly during DC maturation (21). Another hurdle is the promiscuous expression of some markers on macrophages and monocytes, such as CD14 or CD64. Recently, CD14+CD1clow cells in the skin were re-defined as macrophages (22). However, the identity of tissue CD14+CD1chigh cells remains uncertain, we refer to these cells as CD14<sup>+</sup> DCs throughout this review.

The analysis of key DC properties can help assessing the DC identity of a potential subset. Hallmark properties include dendritic morphology, migratory capacity, and ability to stimulate naive T cells. These properties have been used to distinguish macrophages from DCs in the skin (22, 23) and inflammatory fluids (17), or monocytes from DCs in the blood (24).

Finally, gene expression signatures have emerged from transcriptomic studies and can be a useful tool to confirm DC identity, to assign a population to a known DC subset, or to define a new one. Lineage-negative CD16<sup>+</sup> blood cells were initially termed CD16<sup>+</sup> DCs, but transcriptomic analysis showed that they are a subset of monocytes (25). Similarly, 6-sulfo LacNac/Slan<sup>+</sup> blood cells were termed Slan<sup>+</sup> DCs, but comparative transcriptomic analysis identified them as a subpopulation of CD16<sup>+</sup> monocytes (24). Recently,dermal CD14+CD1clow cells were found to be closer in their gene expression to macrophages than to DCs (22). Transcriptomic analysis has also been used to assess the proximity of tissue DC subsets with their blood counterparts for skin (8) or intestinal DCs (10).

#### **WHAT IS THE ONTOGENY OF HUMAN DC SUBSETS?**

Addressing human DC ontogeny is challenging, but *in vitro* culture models, clinical observations, and comparative transcriptomic analysis have provided substantial insight. Human DCs are constantly replenished from bone marrow precursors as shown by the replacement of dermal DCs after hematopoietic stem cell transplantation (23) and the loss of blood DCs after bone


#### **Table 1 | Phenotypic markers for human DC subsets**.

+/−, reported in some tissues. ?, not reported.

marrow suppression induced by preparative cytotoxic therapy (22). Patients carrying mutations in *GATA2* or *IRF8* lack all blood DC subsets, consistent with a common origin (26, 27). Several lines of evidence indicate that Flt3-L is required for the generation and/or maintenance of most DC subsets: injection of Flt3-L to human volunteers increases the number of blood DC subsets (28, 29); pDCs, CD1c<sup>+</sup> DCs, and CD141<sup>+</sup> DCs equivalents can be derived *in vitro* by culturing CD34<sup>+</sup> hematopoietic precursors with Flt3-L (30–32); levels of serum Flt3-L are elevated in patients affected by mutations in *GATA2* or *IRF8* (26, 27). The importance of other cytokines in DC differentiation or homeostasis *in vivo* is unclear. Recently, a committed DC progenitor (CDP) has been identified in bone marrow and cord blood, but was absent from adult blood and tonsils (33). In an *in vitro* culture model, these CDP give rise only to pDC and cDCs, via an intermediate precursor restricted to CD1c<sup>+</sup> DCs and CD141<sup>+</sup> DCs (29, 33). This pre-cDC is present in adult bone marrow, blood, and tonsils (29). Whether pre-cDC differentiate into cDCs in the blood or lymphoid organs and tissues remains to be addressed.

The ontogeny of migratory DCs also remains to be better characterized. Of note, patients affected with a mutation in *GATA2* retain normal numbers of epidermal LC (27), showing that LC represent a distinct lineage from pDCs and cDCs. The observations that LC remained of donor origin 10 years after hand allograft and that they could proliferate *in situ* indicate that LC can self-renew in tissues (34). In addition, transcriptomic analysis shows that intestinal CD103−CD172a<sup>+</sup> DCs (10) and inflammatory DCs (17) express monocyte gene signatures, suggesting that these DC subsets derive from monocytes rather than a common DC precursor.

Cross-species comparative transcriptomic analysis suggest that pDCs, CD1c<sup>+</sup> DCs, and CD141<sup>+</sup> DCs represent distinct *bona fide* lineages, as homologies have been evidenced with the welldefined mouse DC subsets pDCs, CD11b<sup>+</sup> DCs, and CD8<sup>+</sup> DCs, respectively (8, 10, 25). Regarding the molecular ontogeny, *in vitro* culture models indicate that the transcription factors E2-2 and Batf3 drive the differentiation of pDCs and CD141<sup>+</sup> DCs, respectively (35–37). Of note, Batf3 silencing in humanized mice was not sufficient to inhibit CD141<sup>+</sup> DC differentiation (37), which might be due to molecular compensation by related transcription factors as shown in Batf3-deficient mice (38). It has been proposed that CD1c<sup>+</sup> DCs depend on IRF4 based on its preferential expression in CD1c<sup>+</sup> DCs (10, 39), however this remains to be formally proven.

Another unresolved matter is the relationship of blood cDCs and their lymphoid organ and tissue counterparts. It has been suggested that blood CD1c<sup>+</sup> DCs and CD141<sup>+</sup> DCs represent a precursor form of cDC subsets (4, 8, 40), but a direct precursor–progeny relationship remains unclear. Consistent with the idea that they are not terminally differentiated, blood CD1c<sup>+</sup> DCs and CD141<sup>+</sup> DCs become competent for cross-presentation only after activation, whereas lymphoid organ DCs cross-present without the need for activation (41). Moreover, blood CD1c<sup>+</sup> DCs retain some plasticity as they can differentiate *ex vivo* into LC-like cells, while tonsil CD1c<sup>+</sup> DCs cannot (42, 43).

#### **IS THERE A FUNCTIONAL SPECIALIZATION OF HUMAN DC SUBSETS?**

#### **PATHOGEN RECOGNITION**

Among the variety of pathogen-recognition receptors,TLR expression by DC subsets (either mRNA or protein expression) has been the most studied. pDC express TLR1, TLR6, TLR7, TLR9, and TLR10, resident CD1c<sup>+</sup> DCs express TLR1, TLR2, TLR4, TLR5, TLR6, and TLR8, and resident CD141<sup>+</sup> DCs express TLR1, TLR3, TLR6, TLR8, and TLR10 (3, 5, 44–46). Skin LC express TLR1, TLR2, TLR3, TLR6, TLR7 and vaginal mucosa LC express TLR8 in addition (15, 47, 48), skin and vaginal mucosa CD1a<sup>+</sup> DCs express TLR6 and TLR8 while the expression of other TLR is less clear, and skin and vaginal mucosa CD14<sup>+</sup> DCs express TLR1, TLR2, TLR4, TLR6, and TLR8 (15, 49). C-type lectin receptors are also important pathogen-recognition receptors, some of which have been reported to be differentially expressed by DC subsets by transcriptomic analysis (17, 49–52). Receptors whose differential expression among DC subsets has been confirmed at protein level include Clec9A on CD141<sup>+</sup> DCs, BDCA2/CD303 on pDC, ClecSF14/CD301 on CD1c<sup>+</sup> DCs, Langerin/CD207 on LC, and Clec10a and LOX-1 on CD14<sup>+</sup> DCs (5, 19, 49, 53, 54).

Differential expression of pathogen-recognition receptors can confer functional specialization to DC subsets for the response to pathogens (46, 55, 56) or vaccines (57).

Much work remains to be done to characterize the expression pattern of intracellular pathogen-recognition receptors in resident and migratory DC subsets. It has been reported so far that vaginal mucosa and skin LC, CD1a<sup>+</sup> DCs, and CD14<sup>+</sup> DCs, all express MDA-5, while only CD14<sup>+</sup> DCs express RIG-I (15, 49).

#### **CYTOKINE SECRETION**

Blood and lymphoid organ pDC have long been known to be the best producers of type I interferon (58–60). CD141<sup>+</sup> DCs from blood and from humanized mice spleen have also been reported to be the most potent for type I interferon production after TLR3 stimulation (5, 61). More recently, blood and liver CD141<sup>+</sup> DCs were shown to selectively secrete type III interferon after activation with TLR3 ligand or Hepatitis C virus (56, 62, 63).

Because cytokine secretion by a given DC subset vary depending on the stimulus used (45), it can be difficult to determine *bona fide* specialization for cytokine secretion. Accumulating evidence indicates that blood CD1c<sup>+</sup> DCs are the best producers of IL-12p70, as shown by stimulation with TLR2, TLR3, and TLR8 ligands (40, 45, 46). Whether CD1c<sup>+</sup> DCs from tissues are also specialized for IL-12p70 secretion needs to be confirmed. Indeed, no IL-12p70 secretion could be detected after stimulation of skin DC subsets (8) or intestinal CD1c<sup>+</sup> DCs (11) with several TLRligands. Intestinal and lung CD1c<sup>+</sup> DCs are also the best producers of IL-23 after TLR8 stimulation or *Aspergillus fumigatus* exposure, respectively (11, 39). CD1c<sup>+</sup> DCs from skin, intestine, and blood are also the most potent producers of IL-10 in response to several TLR-ligands (8, 11, 64). Skin LC and CD1a<sup>+</sup> DCs have been reported to be better producers of IL-15 than skin CD14<sup>+</sup> DCs (53, 65), but IL-15 secretion by other DC subsets has not been analyzed yet.

#### **CROSS-PRESENTATION AND CD8 T-CELL RESPONSES**

Numerous studies have shown that blood and lymphoid organ DC subsets can all cross-present efficiently various forms of antigen (66). Spleen, lymph node, and tonsil CD1c<sup>+</sup> and CD141<sup>+</sup> DC subsets are equally potent for cross-presenting soluble antigens, without the need for activation (4, 41, 44). When stimulated with TLR-ligands that can activate both subsets, blood CD1c<sup>+</sup> and CD141<sup>+</sup> DCs also display similar efficiency for crosspresentation (40, 44, 67). However, lymphoid organ and activated blood CD141<sup>+</sup> DCs appear to be more efficient for the crosspresentation of dead cell-derived antigen (5, 41, 68), which might be due to their selective expression of necrotic cell receptors such as Clec9A. Blood CD141<sup>+</sup> DCs were also more efficient than CD1c<sup>+</sup> DCs for cross-presentation of antigens delivered to late endocytic compartments via CD205 targeting, but were equally potent after antigen delivery to early endocytic compartments via CD40 (69).

Blood and lymphoid organ pDCs cross-present efficiently soluble (41, 44, 67, 70, 71), viral (71–74), cell-associated antigen (67, 75), or antigen targeted to surface receptors such as CD40, DCIR, CD205, BDCA2/CD303, or CD32 (67, 69, 76).

The ability of tissue DCs to cross-present is less well characterized. Skin CD1a<sup>+</sup> DCs and LC have been shown to cross-present when purified from skin or skin-draining lymph nodes (4, 77), however, a subsequent study reported that skin CD141+Clec9A<sup>+</sup> DCs are the most efficient for cross-presentation compared to other skin DC subsets (8). Skin LC also cross-present antigen targeted through DCIR (76). Both CD1c<sup>+</sup> DCs and CD141<sup>+</sup> DCs from the lung of humanized mice can cross-present (9), but these results need to be confirmed with DCs directly purified from human lung. The cross-presentation capacity of migratory DCs from other tissues and of inflammatory DCs remains to be analyzed. In addition, which DC subsets cross-present *in vivo* in a physiological situation is a challenging question that is still unaddressed.

Skin LC and CD1a<sup>+</sup> DCs induce the differentiation of cytotoxic T lymphocytes (CTL) more efficiently than skin CD14<sup>+</sup> DCs, through the secretion of IL-15 (53, 65). Activated LC also express higher levels of CD70, which promotes CTL differentiation (55, 77). Blood-activated CD1c<sup>+</sup> DCs induce higher expression of granzymes B and K by CTL than activated CD141<sup>+</sup> DCs, due to the selective secretion of IL-12p70 (40). Whether this specialization also applies to lymphoid organ and tissue CD1c<sup>+</sup> DCs remains to be confirmed.

#### **CD4 T CELLS RESPONSES**

The vast majority of studies have analyzed the ability of isolated DC subsets to stimulate and polarize allogeneic naive CD4 T cells. Blood, lymph node, or lung CD1c<sup>+</sup> DCs and CD141<sup>+</sup> DCs are equally competent for Th1 polarization, either without activation (4, 78) or after exposure to influenza virus (78) or *A. fumigatus* (39). By contrast, blood and lung CD141<sup>+</sup> DCs have been found to be more potent inducers of Th2 polarization compared to CD1c<sup>+</sup> DCs, with or without activation, due to the selective expression of OX40-L (78). Lung CD1c<sup>+</sup> DCs exposed to *A. fumigatus* are more potent than CD141<sup>+</sup> DCs for Th-17 polarization due to their secretion of IL-23 (39), however both intestinal CD1c<sup>+</sup> DCs and CD141<sup>+</sup> DCs are equally able to induce Th-17 polarization (10). Blood pDC can induce Th1 polarization after activation with CD40-L, influenza virus, or Sendai virus (60, 79), but induce Th2 polarization through OX40-L after activation with IL3 (79). Whether these observations also apply to lymphoid organ pDC, and whether pDC can induce Th-17 polarization when adequately activated remains to be addressed.

Skin DC subsets have been proposed to be specialized for CD4 T-cell polarization, LC, and CD1a<sup>+</sup> DCs being especially potent for Th2 polarization while CD14<sup>+</sup> DCs mainly induce T follicular helper (Tfh) cells (53). This specialization is conserved after skin DC migration to draining lymph nodes (4). The molecular mechanism underlying this functional specialization remains unclear. Moreover, vaginal mucosa LC and CD1a<sup>+</sup> DCs preferentially induce Th2 polarization, while vaginal mucosa CD14<sup>+</sup> DCs are better inducers of Th1 (15). The ability of vaginal mucosa DCs to induce Tfh has not been analyzed. In addition, skin LC have

DC subset.

The ability to present antigens to CD8 or CD4 T cells is represented with

been found to be more potent than other skin DCs for the induction of IL22-secreting CD4 T cells (80, 81), while both vaginal mucosa LC and CD1a<sup>+</sup> DCs are equally competent (15).

Inflammatory DCs isolated from rheumatoid arthritis synovial fluid and from tumor ascites preferentially induce Th-17 polarization through the secretion of Th-17 polarizing cytokines IL-6, IL-23, TGFβ, and IL1-β (17). Inflammatory DCs from tumor ascites also efficiently stimulate autologous effector CD4 T cells to secrete IL-17 (17). The CD4 T-cell responses induced by inflammatory DCs from other inflammatory environments remain to be investigated.

Finally, several DC subsets were shown to induce Treg: dermal CD14<sup>+</sup> DCs (20), intestinal CD1c<sup>+</sup> DCs and CD103−CD172a<sup>+</sup> DCs (10), tonsil pDC activated with IL3 or TLR-ligands (82), and bacteria-exposed skin LC (83). In addition, liver DCs (7) and TLR4-activated oral mucosa DCs (84) were proposed to promote Treg induction through the secretion of IL-10, but whether one subset is more potent for Treg induction has not been analyzed. Blood *Escherichia coli*-activated CD1c<sup>+</sup> DCs have also been proposed to inhibit CD4 T-cell proliferation through IL-10 secretion (64). In addition, skin LC, but not dermal CD1a<sup>+</sup> DCs and CD14<sup>+</sup> DCs, have been shown to stimulate the proliferation of autologous skin-resident memory Treg (85).

Collectively, these results suggest that some CD4 T-cell responses are the consequence of subset-intrinsic specialization, while others are more dependent on signals from the environment or on tissue imprinting (**Figure 1**). The clearer observations so far are the specialization of CD141<sup>+</sup> DCs for Th2 polarization and of CD14<sup>+</sup> DCs for Tfh polarization (both findings would need to be confirmed with DCs from other tissues), and of skin and vaginal mucosal LC and CD1a<sup>+</sup> DCs for Th2 and Th22 polarization. However, this specialization might be tissuedependent as lung CD1c+CD1a<sup>+</sup> DCs are not potent inducers of Th2 polarization.

#### **CONCLUSION**

Despite the technical challenges inherent to human DC work, significant progress has been made in the past few years in the characterization of human DC subsets. Important issues that will need further exploration include the ability of DC subsets to stimulate effector and memory T cells, the interplay between DC subsets, and the *in vivo* confirmation of functional specializations observed *ex vivo*. These could be achieved by the use of humanized mice models, the analysis of relevant pathological situations, or the study of patients with mutations in relevant genes.

This knowledge will be instrumental in the design of novel vaccines and DC-based immunotherapies.

#### **ACKNOWLEDGMENTS**

Our work is supported by INSERM, Institut Curie, Association pour la Recherche contre le Cancer.

#### **REFERENCES**

1. Merad M, Sathe P, Helft J, Miller J, Mortha A. The dendritic cell lineage: ontogeny and function of dendritic cells and their subsets in the steady state and the inflamed setting. *Annu Rev Immunol* (2013) **31**:563–604. doi:10.1146/annurevimmunol-020711-074950


control mucosal IL-17 cytokine responses. *Immunity* (2013) **38**(5):970–83. doi:10.1016/j.immuni.2013.04.011


for driving T-helper type 1 responses in human neonatal settings. *J Infect Dis* (2014) **210**(3):424–34. doi:10.1093/infdis/jiu103


**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: 14 January 2015; accepted: 09 March 2015; published online: 23 March 2015. Citation: Durand M and Segura E (2015) The known unknowns of the human dendritic cell network. Front. Immunol. 6:129. doi: 10.3389/fimmu.2015.00129*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology.*

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

# **A systematic approach to identify markers of distinctly activated human macrophages**

*Bayan Sudan<sup>1</sup> , Mark A. Wacker <sup>1</sup> , Mary E. Wilson1,2 and Joel W. Graff 1,2 \**

*1 Infectious Diseases Division, Department of Internal Medicine, University of Iowa, Iowa City, IA, USA, <sup>2</sup> Iowa City VA Medical Center, Iowa City, IA, USA*

Polarization has been a useful concept for describing activated macrophage phenotypes and gene expression profiles. However, macrophage activation status within tumors and other settings are often inferred based on only a few markers. Complicating matters for relevance to human biology, many macrophage activation markers have been best characterized in mice and sometimes are not similarly regulated in human macrophages. To identify novel markers of activated human macrophages, gene expression profiles for human macrophages of a single donor subjected to 33 distinct activating conditions were obtained and a set of putative activation markers were subsequently evaluated in macrophages from multiple donors using integrated fluidic circuit (IFC)-based RT-PCR. Using unsupervised hierarchical clustering of the microarray screen, highly altered transcripts (*>*4-fold change in expression) sorted the macrophage transcription profiles into two major and 13 minor clusters. Among the 1874 highly altered transcripts, over 100 were uniquely altered in one major or two related minor clusters. IFC PCR-derived data confirmed the microarray results and determined the kinetics of expression of potential macrophage activation markers. Transcripts encoding chemokines, cytokines, and cell surface were prominent in our analyses. The activation markers identified by this study could be used to better characterize tumor-associated macrophages from biopsies as well as other macrophage populations collected from human clinical samples.

**Keywords: human macrophages, activation markers, microarray, integrated fluidic circuit RT-PCR, macrophage polarization**

# **Introduction**

Macrophages assume critical roles in almost every tissue and disease state through their ability to assume distinct functional capacities in different microenvironments. Macrophages respond to a variety of external stimuli to assume different polarized activation states. Distinctly polarized macrophages, modeled *in vitro* using specific activating conditions, can be defined by functional attributes such as microbicidal activity, and by unique gene expression profiles. An early study contrasting functional and gene expression differences between IFNγ- and IL-4-treated macrophages proposed that the latter phenotype be described as alternative activation (1), a very different macrophage phenotype from IFNγ- or classically activated macrophages. Since that time, many additional polarized macrophage types, induced by different stimuli, have been proposed.

#### *Edited by:*

*Martin Guilliams, VIB Inflammation Research Center, Belgium*

#### *Reviewed by:*

*Elodie Segura, Institut Curie, France Sam Basta, Queen's University, Canada*

#### *\*Correspondence:*

*Joel W. Graff, Infectious Diseases Division, Department of Internal Medicine, University of Iowa, 400 EMRB, Iowa City, IA 52242, USA joel-graff@uiowa.edu*

#### *Specialty section:*

*This article was submitted to Antigen Presenting Cell Biology, a section of the journal Frontiers in Immunology*

> *Received: 04 March 2015 Accepted: 07 May 2015 Published: 27 May 2015*

#### *Citation:*

*Sudan B, Wacker MA, Wilson ME and Graff JW (2015) A systematic approach to identify markers of distinctly activated human macrophages. Front. Immunol. 6:253. doi: 10.3389/fimmu.2015.00253*

Several competing systems have been proposed in an attempt to provide a framework that describes the complexity of macrophage polarization. The first system describes macrophage phenotypes as a linear continuum with M1 (classically activated) and M2 (alternatively activated) macrophages at opposite ends (2, 3). The second system describes macrophage phenotypes as a spectrum akin to a color wheel, with classically activated, wound healing, and regulatory macrophages used as examples of unique polarized phenotypes that do not fit well within a linear continuum (4). A modified version of the M1–M2 system acknowledged the diversity of macrophage phenotypes with descriptions such as M1a, M1b, M2a, M2b, and M2c (5, 6). Additions to the M1–M2 nomenclature system have proposed naming macrophages differentiated in the presence of CXCL4 as "M4" (7) and IL-17-treated macrophages "M17" (8). To standardize the burgeoning descriptions of polarized macrophage types, it has been suggested that the activation condition be defined in the name of the polarized macrophage [M(IL-4), M(IL-10), M(LPS), M(IFNγ), and so forth (9)]. To preserve clarity, we have employed this descriptive nomenclature system to describe the activated macrophages in the current report (**Table 1**).

Macrophages are often very abundant within tumors (12, 13). There is evidence that macrophages can promote tumorigenesis, tumor growth, and metastasis (14). Despite macrophage pro-tumor activities, tumor-associated macrophages (TAMs) display a wide range of phenotypic diversity within a tumor due to ontogeny, activation signals, and localization (15). The plasticity of macrophage phenotypes is well known (16, 17) and this characteristic has provided a therapeutic target whereby macrophages are encouraged to switch functionally from pro-tumor to anti-tumor. Clinical approaches that modify macrophage activation in this way include blockade of M-CSF, low-dose irradiation, and combinational therapies (18–21). What is lacking is a thoroughly characterized and reliable set of macrophage activation markers that would allow for improved characterization of activation patterns, and monitoring of the therapeutic efficacy of macrophage-targeted treatments.

Gene expression profiles using microarrays have been used to analyze activation of primary human monocytes and monocytederived macrophages (MDMs) (7, 22–32). Until very recently (33), most transcriptome-based approaches to characterize polarized macrophages contrasted two macrophage-activating conditions in each study. Using a blood sample from a single human donor, we surveyed gene expression profiles in primary macrophages activated with 33 different activating conditions. This data set served as a rich resource for identifying putative human macrophage activation markers. As a followup approach, integrated fluidic circuit (IFC)-based RT-PCR was used to examine a panel of transcripts to verify the reproducibility of the gene expression changes from multiple donors. This latter assay was also used to determine the expression kinetics of previously described markers of human macrophages as well as novel markers identified by the microarray-based screen.

**Current**

#### **TABLE 1 | Macrophage-activating conditions and nomenclature used in this study**.



*<sup>a</sup>Treatments with chicken ovalbumin and with Curdlan likely had endotoxin contamination due to the extraction processes used to obtain these reagents (10, 11).*

# **Materials and Methods**

**Combinational treatments Previous**

## **Human Subjects**

Human subject protocols were approved by Institutional Review Boards of the University of Iowa and the Iowa City Veterans Affairs Medical Center. Peripheral blood samples from anonymous, healthy donors were acquired through the DeGowin Blood Bank at the University of Iowa.

## **Integrated Fluidic Circuit-Based RT-PCR**

RNA purified from MDMs using TRIzol was reverse transcribed in random hexamer-primed reactions with SuperScript III RT (Invitrogen). The cDNA was pre-amplified for 14 PCR cycles in reactions primed by a master mix of 48 TaqMan Gene Expression Assays (Applied Biosystems) using PreAmp Master Mix (Applied Biosystems) with a modified protocol according to recommendations by Fluidigm. Following a 1:5 dilution of pre-amplified product in water, 48 samples and 48 TaqMan Gene Expression Assays were loaded onto 48.48 Dynamic Array IFC plates (Fluidigm) using the 48.48 MX IFC Controller (Fluidigm). Realtime PCR was performed using the BioMark System for Genetic Analysis (Fluidigm). Cycle threshold (Ct) values were determined using real-time PCR analysis v3.1.3 software (Fluidigm). Ct values corresponding to transcripts encoding ACTB, B2M, and TBP were used as endogenous controls. Changes in transcript expression were calculated using the ΔΔCt method and converted to log<sup>2</sup> scale using Excel 2010. Line graphs of time course experiments were generated using Prism 6 (GraphPad). Heat maps were generated using Partek Genomic Suite software.

## **Cell Purification and Culture**

Peripheral blood mononuclear cells (PBMCs) were isolated from the blood by density sedimentation using Ficoll-Paque PLUS (GE Healthcare) and maintained in Petri dishes at a density of 5e7 cells/dish in 10 ml RP-10 medium [RPMI 1640 medium (Gibco) supplemented with 10% fetal bovine serum (Gibco), 100 U/ml penicillin, 50 μg/ml gentamicin, and 5 ng/ml M-CSF (eBioscience)]. After 10 days, non-adherent cells were removed by rinsing and the adherent MDMs were dislodged with cell scraping following incubation at 37°C for 10 min in 0.25% Trypsin/1 mM EDTA solution (Gibco). MDMs were seeded in 12-well tissue culture-treated plates (Corning) at 5e5 cells/well in 2 ml RP-10 and allowed to rest for 2 days at 37°C. Before treatment with macrophage-activating conditions, the culture medium was replaced with 1 ml fresh RP-10 per well. At 24 h posttreatment, RNA was purified from MDMs using TRIzol Reagent (Invitrogen).

# **Macrophage-Activating Stimuli**

All stock solutions were stored at *−*80°C unless otherwise noted. The sources of human recombinant cytokines were as follows: IL-1β (eBioscience), IL-4 (PeproTech), IL-10 (R&D Systems), IFNβ (PeproTech), IFNγ (PeproTech), GM-CSF (eBioscience), TNFα (PeproTech), and TGFβ (R&D Systems). These cytokines were stored at concentrations recommended by the manufacturers and were subjected to no more than two freeze–thaw cycles. Dexamethasone powder (Sigma-Aldrich) was suspended in 1 part ethanol and subsequently diluted in 49 parts medium to a stock concentration of 50 μM. Phenol-extracted *Escherichia coli* 055:B5 LPS (Sigma-Aldrich) and polyinosinic:polycytidylic acid sodium salt (PolyI:C) (Sigma-Aldrich) were stored at a stock concentration of 1 mg/ml in RP-10. Adenosine (Sigma-Aldrich) was suspended in RP-10 at a stock concentration of 10 mM.

Chicken ovalbumin (MP Biomedicals) was suspended at 2 mg/ml in PBS lacking Ca++ or Mg++ (Gibco) and goat antichicken ovalbumin (MP Biomedicals) was suspended in water at 16 mg/ml. Immune complexes (IC) were prepared fresh for each experiment by combining ~10:1M excess of antibody to antigen and incubating with end-over-end rotation at room temperature for 30 min. Curdlan (InvivoGen) was also freshly prepared for each experiment by suspension in RP-10 at a concentration of 1 mg/ml.

Trehalose-6,6-dibehenate (TDB) (InvivoGen) was suspended at a concentration of 10 mg/ml in DMSO and heated to 60°C for 30 s. After vortexing, the TDB/DMSO solution was diluted to 1 mg/ml by the addition of PBS. This stock solution was heated to 60°C for 15 min and stored at 4°C.

#### **Microarrays**

RNA sample preparation for microarrays and the subsequent hybridization to the Illumina beadchips were performed at the University of Iowa DNA Facility. Three Human HT-12 v4 BeadChips (Illumina) were processed individually in this experiment with 1 sample from an untreated control and 11 samples from polarized macrophages loaded onto each array. Briefly, 100 ng total RNA from each of the 36 samples was amplified and converted to biotin-cRNA using the Epicenter TargetAmp-Nano Labeling Kit for Illumina Expression Bead-Chip (Illumina). The biotin-aRNA product was purified using the RNeasy MinElute Cleanup Kit (Qiagen) according to modifications from Epicenter. Seven hundred fifty nanograms of this product were mixed with Illumina hybridization buffer, placed onto each beadchip array, and incubated with rocking at 58°C for 17 h in an Illumina Hybridization Oven. Following hybridization, the arrays were washed, blocked, and stained with streptavidin-Cy3 using the Whole-Genome Gene Expression Direct Hybridization Assay (Illumina). Beadchip arrays were scanned with the iScan System (Illumina) and data were collected using the GenomeStudio software v2011.1 (Illumina). The expression data has been deposited in NCBI Geo repository (GSE68854).

## **Transcript Expression Analysis**

Partek Genomic Suite v6.5 (Partek) was used to perform robust multi-array averaging and to calculate gene expression changes. A data set comprising of 1874 transcripts with changes in expression of more than fourfold relative to untreated controls was submitted to unsupervised hierarchical clustering and principal components analysis using default settings in Partek Genomic Suite software. Briefly, for unsupervised hierarchical clustering, agglomerative clustering was used to determine Euclidean dissimilarity distances using an average linkage method. For principal components analysis, a dispersion matrix based on correlations was normalized using Eigenvector scaling. Contribution of individual transcripts to each of the principal components was determined using the FactoMineR package in R. After principal component analysis was completed the contribution of each transcript to each of the components was extracted and ranked using Excel 2010.

Correlation coefficients were calculated for each pairwise combination of the 33 activated macrophage expression profiles for the 1874 regulated transcripts using the corandPvalue function of the WGCNA package in R. The data were then converted to heat maps using Excel (Microsoft).

For gene ontology (GO) analysis, the STRING database (version 9.05; string-db.org) was used to identify the 1615 protein coding RNAs in our set of 1874 regulated transcripts. Also, within the STRING database website, the GO categories enriched in the set of 1615 regulated transcripts identified as protein coding RNAs were determined.

# **Results**

## **Survey of Proposed Human Macrophage Activation Marker Expression in MDMs Responding to Six Distinct Activation Conditions**

untreated MDM controls as well as M(IFNβ), M(IFNγ), M(Dex), M(IL-4),

Transcripts used as markers of polarized human macrophages should change expression in response to one stimulus or a limited number of related activation stimuli. Additionally, macrophage activation markers should have sustained, rather than transient, changes in expression. In primary human macrophages responding to a variety of activation conditions, we evaluated the expression kinetics of transcripts that encode 11 proposed activation markers (4, 6, 9) over a 24-h period (**Figure 1**). Several observations from this survey were notable. First, some commonly for each indicated transcript relative to the untreated MDM controls was determined at each time point for the six types of activated MDMs (*N* = 1).

assessed transcripts, TNF and IL-10, were rapidly induced in M(LPS) and M(Curdlan) but returned to near basal expression by the 24-h time point. Second, although many genes were similarly regulated in M(IFNβ) and M(IFNγ), the expression patterns of CD163 and CXCL9 were distinct in response to these two interferon types. Third, most markers have been noted because of their increased expression in response to macrophage activation conditions but many transcripts in this panel showed a remarkable reduction in expression. Finally, the expression level of many activation marker transcripts was either continuing to change or was sustained at high levels at the 24-h time point. Together, these observations revealed there is a need for a systematic attempt to identify reliable activation markers whose expression was either up- or down-regulated in human macrophages, and which

#### **FIGURE 2 | Continued**

**Hierarchical clustering of gene expression profiles from activated human MDMs separated into 2 major clusters and 13 minor clusters**. Microarrays were performed using RNA collected from MDMs at 24 h post-treatment with 33 distinct activation conditions (*N* = 1). **(A)** A set of 1874 regulated transcripts defined as having *>*4-fold change in expression levels relative to untreated controls was compiled and displayed as a heat map (log<sup>2</sup> scale). Gene expression profiles were sorted according to unsupervised

exhibited sustained expression level changes in response to an array of activation conditions.

# **Expression Profiling of a Diverse Array of Activated Human Macrophages**

To screen for transcripts representing putative human macrophage activation markers, microarrays were performed using samples collected from human MDMs derived from a single donor, subjected to 33 unique activating conditions (**Table 1**) for 24 h. Sixteen of the conditions were composed of a single activating stimulus. Eight cytokines comprised the largest category of macrophage-activating stimuli used in this study and represent a spectrum of pro- and anti-inflammatory molecules that are abundantly expressed in sites where MDMs would be recruited such as infections or wounds. Pathogenassociated molecular patterns (PAMPs) recognized by C-type lectin receptors (CLRs) or toll-like receptors (TLRs) were the second largest set of macrophage-activating stimuli in this study and consisted of Curdlan (dectin-1 agonist), TDB (trehalose-6,6-dibehenate; mincle agonist), polyI:C (TLR3 agonist), or one of two concentrations of LPS (TLR4 agonist). Another set of stimuli, IgG–OVA IC and adenosine, were selected for their ability to reprogram inflammatory macrophages to become non-inflammatory (34, 35). Finally, we selected the glucocorticoid, dexamethasone, as an immunosuppressive stimulus. The remaining 17 conditions consisted of pairs of the above macrophage-activating stimuli (**Table 1**). The macrophageactivating conditions were selected with the expectation that they would lead to diverse gene expression profiles providing insights into the potential diversity of macrophage gene expression programs.

We first focused our attention on regulated transcripts that had changes in abundance of over fourfold relative to untreated controls changes. A data set of 1874 regulated transcripts that were differentially expressed in MDMs responding to one or more of the macrophage-activating conditions was compiled. Unsupervised hierarchical clustering was performed to evaluate the expression profiles of the regulated transcripts; this is summarized in a heat map that includes a dendrogram indicating relative dissimilarity distances between gene expression profiles of each polarized macrophage type (**Figure 2A**). Official gene names of the regulated transcripts and calculated expression changes are provided as supplemental material (Table S1 in Supplementary Material).

We considered whether the clustering analysis results separated the gene expression profiles corresponding to previously studied macrophage activation states as denoted in **Table 1**. Consistent with the previous reports (29, 33), gene hierarchical clustering of genes and treatments. Dissimilarity distances between gene expression profiles are displayed using a color-coded dendrogram to indicate 13 hierarchical clusters. See Section "Results" for dissimilarity distance cut-off rationale. Arranged in the same order as shown here, transcript names and quantitation of expression level changes are available in Table S1 in Supplementary Material. **(B)** Number of upregulated and down-regulated transcripts within each gene expression profile. Potent and mild macrophage activation conditions are indicated.

expression profiles of M(LPS + IFNγ) (previously named "M1") macrophages were quite different from that of M(IL-4) (previously named "M2a") macrophages. By contrast, the profile of M(LPS + IC) macrophages (previously named "M2b") was very similar to the profiles of M(LPS + IFNγ), separated only by the profile of M(PolyI:C). Since M(LPS + IFNγ) and M(LPS + IC) are known to have different biological activities (6), we divided the 33 macrophage expression profiles into 13 clusters, the lowest dissimilarity distance cut-off that successfully separated these profiles (**Figure 2A**).

# **Microarray Results were Confirmed Using IFC Arrays**

The IFC array-based real-time RT-PCR platform provided a highthroughput mechanism to accurately verify the expression of a large set of transcripts in samples from multiple human donors. We used several strategies to select a panel of transcripts with diverse expression patterns out of the 1874 regulated transcripts, which were re-assessed on multiple samples using IFC arrays. First, we included the 11 transcripts analyzed in **Figure 1**. Next, we used the STRING database (version 9.05) to identify enriched GOs for the 1615 protein coding RNAs in our set of 1874 regulated transcripts. Among the GOs categories that were enriched in our data set, we chose to focus on chemokine activity, cell surface, and cytokine activity because these GO categories were highly enriched (Table S2 in Supplementary Material). Finally, we selected transcripts that were uniquely regulated in one or two minor clusters. The final panel included a combination of transcripts that represented changes occurring in each of the 33 macrophage-activating conditions. We also mined the data set for reliable endogenous controls to include in the panel. Among the potential endogenous controls we considered, the expression levels of TBP (define) and B2M (define) transcripts appeared to be the least affected by the 33 macrophage-activating conditions (Table S1 in Supplementary Material).

The samples obtained from activated MDMs of a single donor that were analyzed by microarray were re-assessed using IFC arrays. Approximately 10 transcripts were not detected when using a pre-established Ct cut-off. Strong linear correlation for 15 representative transcripts was observed when comparing expression levels determined by microarray and by IFC arrays (Figure S1 in Supplementary Material). The remaining detectable transcripts in our panel had expression levels that also showed strong linear correlation when comparing microarray and RT-PCR results (data not shown). Overall, these results confirmed the microarray measurements using an independent approach and provide convincing evidence that IFC arrays was a dependable method for measuring transcript expression.

#### **FIGURE 3 | Continued**

**Comparing correlation coefficients supported the separation of gene expression profiles into two clusters, which can be modeled as a "split spectrum."** Correlation coefficients were calculated for using the 1874 regulated transcript data set (*N* = 1). **(A)** Each pairwise combination of the 33 gene expression profiles are displayed as a heat map with a range of

# **Macrophage-Activating Conditions can be Categorized as "Mild" or "Potent" Based on the Number of Transcripts regulated in Response to the Stimuli**

There was a large range in the number of highly regulated transcripts in each activated macrophage expression profile (**Figure 2B**). Specifically, the 20 activated macrophage types within clusters 1–7 had relatively few regulated transcripts (93 *±* 14), whereas the 13 activated macrophage types within clusters 8–13 had large numbers of regulated transcripts (564 *±* 31). We propose that macrophage-activating conditions can be categorized as "mild" or "potent" based on the number of transcripts the treatment alters.

When considering the mild and potent clusters of the unsupervised hierarchical clustering, we noted that the gene expression profiles did not segregate along previously described M1–M2 divisions (**Table 1**). Polarized macrophage types, M(IFNγ) and M(LPS), which have each been considered "M1" macrophage types sorted into the mild and potent clusters, respectively. Similarly, macrophage types formerly named "M2," M(IL-4), and M(LPS + IC) were categorized as mild and potent, respectively.

We considered the possibility that the wide discrepancy in the number of transcripts regulated in MDMs responding to mild and potent activating conditions was due to suboptimal concentrations of the "mild" stimulus. To address this, MDMs were treated with each of the 11 single treatment macrophage-activating conditions that were categorized as mild at concentrations ranging from 4-fold higher to 16-fold lower those used in the microarray-based experiments. In general, modest dose responses were observed. In response to the majority of the mild stimuli tested (IFNβ, IFNγ, IL-1β, IL-4, IL-10, and TNFα), the amplitude of change in expression for any given transcript was routinely <4-fold between the lowest and highest concentrations for the activating stimulus tested (Figure S2 in Supplementary Material). This suggests that the window of activity is wide for these stimuli and further suggests that use of higher concentrations of these stimuli would be unlikely to revise their macrophage-activating categorization from "mild" to "potent."

## **Evaluation of Correlation Coefficients Between Activated MDM Gene Expression Profiles Supports Conclusions Drawn from Hierarchical Clustering Analyses**

Correlation coefficients were determined for each pairwise combination of activated MDM gene expression profiles in the set of 1874 regulated transcripts (**Figure 3**). This analysis further substantiated the categorization of gene expression profiles into mild and potent categories as shown by unsupervised hierarchical clustering (**Figure 2**). As an example, there was a consistently higher coefficients of *−*0.2 to 1.0. **(B)** Each pairwise combination of the 13 "potent" gene expression profiles in clusters 8–13 are displayed as a heat map with a restricted range of coefficients from 0.7 to 1.0. **(C)** A "Split Spectrum" model of macrophage activation can be used to emphasize the high degree of correlation between treatments with at least one potent macrophage-activating stimulus.

gene expression profile correlation when the profiles of potently activated macrophages (clusters 8–13) were paired with profiles from potently activated (clusters 1–7), macrophages (**Figure 3A**). Also, we note that, when using a different color scale (**Figure 3B**), correlations between profiles in clusters 8–13 were noticeable and supported the division of the gene expression profiles of the potently activated macrophage gene expression profiles into minor clusters.

In a recent microarray-based study (33), at least 9 clusters of activated macrophages in a data set derived from human MDMs activated with 28 distinct stimuli. In agreement with the level of clustering as the previous study, we now show using unsupervised hierarchical clustering and correlation coefficient analyses that human MDMs activated with the 33 macrophage activation conditions used in this study form at least 13 clusters. Both studies support a spectrum model of macrophage activation. Because of the strong "mild" and "potent" macrophage-activating condition categories described here, we propose that macrophage activation may best be described using a "split spectrum" model (**Figure 3C**).

# **Verifying Transcripts that Serve as Markers for the "Potent" Macrophage Activation Conditions**

The first principal component (PC1) explains (42.9%) of the variance in the data set of 1874 regulated genes while PC2 and PC3 each contributed to ~10% of the variance and the remaining principal components each accounted for *<*5% of the variance (**Figure 4A** and data not shown). A scatterplot of regulated gene expression profiles based on the first two principal components segregated profiles in clusters 1–7 from those in clusters 8–13 along the PC1 axis (**Figure 4B**). The expression profiles of the 50 transcripts that contributed the most to PC1 were subjected to unsupervised hierarchical clustering and displayed as a heat map (**Figure 4C**). There was an obvious distinction between gene expression responses between the profiles within the mild and potent major clusters; the transcripts robustly regulated by potent macrophage-activating conditions and relatively unaltered by mild macrophage-activating conditions underlie PC1 and account for the major source of variance for the diverse spectrum of polarized macrophage gene expression profiles in this study.

There were many transcripts in addition to the 50 noted in **Figure 4C** that contributed to PC1. Using samples collected for analysis in **Figure 1**, we monitored the change in expression of four transcripts (CCL5, IRG1, MT1G, and S100A8) that contributed to PC1 over 24 h (**Figure 4D**). CCL5 and IRG1 transcripts showed immediate increased expression levels that were sustained through the 24-h time point. By contrast, delayed increases were seen for the expression levels of MT1G and S100A8 transcripts. These four transcripts, in addition to IL1B, IL6, and IL23A that were previously seen to have sustained high expression levels in

#### **FIGURE 4 | Continued**

**Transcripts universally regulated by "potent" macrophage-activating conditions in clusters 8–13 were the largest source of variance in the polarized MDM gene expression profiles**. Principal components analysis was performed using the data set of 1874 regulated transcripts from the 33 gene expression profiles. **(A)** The contribution of PC1–PC5 to the variance is shown. **(B)** Scatterplot displays gene expression profiles according to PC1 and PC2 scores with color coding based on "mild" (clusters 1–7) and

M(LPS) and M(Curdlan) macrophages, suggesting that numerous transcripts that can be reliably used as markers of "potent" activation conditions.

## **Evaluating the Use of Chemokine Transcripts as Macrophage Activation Markers**

Chemokines not only play an important functional role in macrophage activity but also include some of the earliest proposed markers of macrophage polarization (2, 5). We generated a heat map of transcript expression changes for chemokines from the C–C and C–X–C subfamilies from the 1874 transcripts (**Figure 5A**). Since IL-4 treated macrophages have been well characterized, the chemokines were sorted according to their average expression in the two activated macrophage types that form cluster 1, M(IL-4) and M(TDB + IL-4). Among the remaining 12 clusters, the chemokine expression profiles from macrophages in cluster 3, comprised of M(TDB), M(TGFβ + GM-CSF), and M(GM-CSF) macrophages, appeared to have the most similar trend in chemokine expression. The overall chemokine expression patterns from all other profiles shared little resemblance to those in cluster 1.

Transcripts for two chemokines, CCL13 and CCL22, accumulated in macrophages treated with IL-4 for 24 h (**Figure 5A**). Interestingly, the upregulation of these chemokines in response to IL-4 was delayed relative to other treatments that induced transient upregulation: interferons for CCL13 (**Figure 1**) and PAMPs for CCL22 (**Figure 5B**). These observations suggests that CCL13 and CCL22 can be used as specific markers for M(IL-4) as long as enough time has elapsed since the activation occurred.

We noted that nearly all chemokines had reduced expression in M(Dex) macrophages according to the microarray results (**Figure 5A**). This observation was confirmed when monitoring the kinetics of expression for five chemokine transcripts described above (CCL5, CCL13, CCL18, CXCL9, and CXCL10) (**Figures 1** and **4D**) and in four additional chemokines (CCL2, CCL3, CCL22, and CXCL5) (**Figure 5B**). The general trend of repressing chemokine production in M(Dex) macrophages may hint at a mechanism by which dexamethasone acts as an immunosuppressive molecule.

## **Donor-to-Donor Variability in Gene Expression Regulation was Minimal in Most Circumstances but was Occasionally seen in some Minor Clusters**

A caveat to the results described until this point is that they were based on MDMs derived from two human donors: one donor for monitoring transcript expression kinetics and one donor for transcriptional profiling. Since donor-to-donor variability among "potent" (clusters 8–13) categorization. **(C)** The 50 transcripts that contributed the most to PC1 were sorted according to unsupervised hierarchical clustering results. Changes in transcript expression levels relative to untreated MDMs are depicted as a heat map (log<sup>2</sup> scale). **(D)** Using IFC-based RT-PCR and samples from **Figure 1**, the changes in expression for each indicated transcript relative to the untreated MDM controls was determined at the 1, 3, 6, and 24 h time points for the six types of activated MDMs (*N* = 1).

human MDM responses was a concern, the expression profiles for many transcripts was determined in samples derived from the microarray experiment and from two additional donors whose MDMs were treated with all 33 macrophage activation conditions (**Figure 6**).

The strong correlation between the microarray results and the IFC PCR results for the first donor was discussed above (Figure S1 in Supplementary Material). Importantly, the last two rows for each transcript, which show the results for the two additional donors, indicated that the MDM responses were, in general, similar to those of the first donor (**Figure 6**). There were a few transcripts (CCL22, CXCL10, IL10, ITGB7, and TGM2) that had strong opposing changes in expression from one donor to the next (**Figure 6**). It is noteworthy that in these instances, the difference in expression was restricted to a limited number of clusters. For example, CCL22 expression regulation tended to be similar in response to all 33 macrophages activation conditions for all 3 donors; the notable exception was seen in the 5 macrophages activation conditions within cluster 12 for the second donor (**Figure 6**). This result is unlikely due to the polyIC and LPS treatments being suboptimal in the experiment involving MDMs from donor 2 since other transcripts, such as CCL5, were regulated similarly in all three donors for the macrophage-activating conditions that make up cluster 12.

A recent mass cytometry-based study produced a highdimension data set from a panel of 38 antibodies to effectively identify signature expression patterns of myeloid cell populations in mice from a number of tissues (36). Since the dimensionality of data sets produced by mass cytometry and IFC PCR are similar, we tested whether the 13 clusters originally defined by unsupervised hierarchical clustering of the 1874 regulated transcripts (**Figure 2**) could be effectively identified using IFC PCR results (**Figure 7A**). The majority of the 13 clusters remained clusters for each of the three donors (**Figure 7B**). Even the "clusters" composed of a single type of activated macrophage type [i.e., M(IFNβ)] maintained their distinctness relative to the other activated macrophage types. We conclude that gene expression platforms such as IFC PCR monitor a large enough set of macrophage activation marker transcripts to identify an overall macrophage population's type/cluster while still allowing for detection of subtle donor-to-donor differences.

## **Putative Activation Markers were Identified for Specific Clusters of Polarized Human Macrophages**

Macrophage activation markers would ideally have large expression changes in a single cluster or polarized macrophage type. We therefore queried the gene expression profiles in the current

study to identify activation markers specific to each of the 13 clusters formed by the unsupervised hierarchical clustering analysis. Many putative activation markers were identified in macrophages activated with IL-4, dexamethasone, or IFNβ (**Figures 8**–**10**).

IL-4 was used as an activation condition for gene expression profiles in clusters 1 and 10. We identified transcripts that were strongly upregulated only within cluster 1, within both cluster 1 and cluster 10, or only within cluster 10. Examples of transcripts that fit these gene expression profiles were readily detected within our data set (**Figure 8A**). Analysis of the kinetics of expression for three of the transcripts identified by this screening approach showed that while ALOX15 and CD1B each appear to be good markers for M(IL-4), although the increase in CD1B was delayed until the 24 h time point, FABP4 was not robustly induce in

M(IL-4) but could still be a valuable marker as this transcript was potently down-regulated in response to several macrophageactivating conditions (**Figure 8B**). This latter observation was consistent with the microarray data (**Figure 8A**).

Given the relative ease of finding IL-4-associated activation markers in our data set, we switched our attention to identifying additional activation markers. Dexamethasone-associated activation markers were identified that were specifically upregulated in macrophage-activating conditions from only within cluster 2, within both clusters 2 and 13, and only within cluster 13 (**Figure 9A**). The expression kinetics was determined for three of the transcripts identified by the microarray screen as dexamethasone responsive (**Figure 9B**). Of these, ALOX15B and MFGE8 appear to be a markers for M(Dex) at early and late time points, respectively.

Next, potential activation markers or IFNβ-treated macrophages were identified within cluster 7 (**Figure 10A**). Further analysis showed that AXL, IFIT, and ZBP1 were all induced rapidly in M(IFNβ) and with delayed kinetics in M(LPS) (**Figure 10B**). This observation may be explained by indirect induction of these genes by LPS-induced IFNβ production.

# **Discussion**

Characterization of TAMs has shifted from quantifying macrophage density in and around tumors to evaluating markers of activation (15, 37). It is important to note that macrophage activation markers have been used to categorize macrophage activation, typically using the M1–M2 nomenclature, yet the regulation patterns of these markers in macrophages responding to a wide variety of activation conditions are not well understood. Using a combined microarray- and IFC array-based approach in this study, previously proposed markers of macrophage activation were better characterized and novel markers of macrophage activation were identified.

In the earliest report using M1–M2 nomenclature, the authors stated that "M-1 and M-2, while useful for conceptualizing

immune responses, certainly could be an oversimplification" and that "there may be a continuum of phenotypes between M-1 and M-2 macrophages" (3). A recently proposed framework argued against using the M1–M2 nomenclature yet upheld the linear model concept that suggested M(IFNγ) and M(IL-4) to represent the polar extremes (9). However, both the results of the current study and those reported by Xue et al. (33) support a spectrum model of macrophage activation rather than a linear model.

Unsupervised hierarchical clustering, correlation coefficient analysis, and principal components analysis of the regulated transcripts each support the concept that macrophage polarized states in this study can be sorted into two major clusters. We designated these clusters "mild" and "potent" to convey the number transcripts altered in response to each specific macrophage-activating condition. It is important to note that, although we have evaluated more macrophage activation conditions in a macrophage

activation study that has previously been published, there could be activation conditions that will have an intermediate number of regulated transcripts making our split spectrum model potentially incorrect. Indeed, Xue et al. (33) studied macrophage responses to 28 activation conditions and we found that the free fatty acid conditions from their study may represent an "intermediate"

cluster (analysis not shown). While our "split spectrum" model may not represent the entirety of the spectrum, it raises the idea that strength of macrophage activation may be worth considering in future attempts to accurately describe macrophage activation/polarization.

In the analysis of the principal components, special attention was warranted for PC1 because it accounted for four times more of the variance than any other principal component. The single treatment macrophage-activating conditions that contributed to PC1 were immune complexes, Curdlan, polyIC, and LPS.

All combinational treatments that contributed to PC1 contained one or two of these potent stimuli. Treatment of macrophages with immune complexes and Curdlan initially signal through Fcγ receptor/Syk/Card9 pathways while treatment with polyIC and LPS signal through TRIF and/or MyD88 pathways. Despite these initial differences, there is substantial overlap triggered by the potent stimuli further downstream pathway signaling. For example, activation of pathways such as NF-κB and MAPK may directly and indirectly account for the regulated expression of many transcripts that contributed to PC1. Importantly, as noted in **Table 1**, chicken ovalbumin and Curdlan are often contaminated with substantial levels of endotoxin, so the "potent" activation conditions may be mostly or in part a consequence of TLRinitiated signaling (10, 11). Future studies will assess the extent that TLR signaling may have contributed to the alterations in the M(Curdlan) and M(IC) macrophage gene expression profiles.

There was substantial evidence, both gene expression and functional, that the mild and potent polarized macrophage types of our data set should be divided into smaller clusters. To define these clusters, we chose to separate our gene expression profiles based on known differences that occur in response macrophageactivating conditions rather than using a statistically based dissimilarity cut-off in the unsupervised hierarchical clustering. Specifically, we noted that M(LPS + IFNγ) and M(LPS + IC) were situated close to each other according to unsupervised hierarchical clustering analysis (**Figure 2**). Important functional differences in macrophages treated with these two distinct activating conditions such as cytokine production (IL-12 vs. IL-10) and ability to skew CD4<sup>+</sup> T cell responses (Th1 vs. Th2) (35, 38–40) supported the segregation of these gene expression profiles into separate clusters. Therefore, the dissimilarity distance between these two gene expression profiles served as our cut-off to rationally sort the 33 gene expression profiles into 13 clusters.

It is notable that if the gene expression profiles had been segregated based on dissimilarity distances into 14 clusters instead of 13, the 5 gene expression profiles currently grouped within "cluster 4" would have been split into 2 clusters. Furthermore, correlation coefficients within cluster 4 were markedly higher when comparing gene expression profiles from MDMs activated with conditions that included IL-10 (**Figure 3**). Finally, hierarchical clustering based on IFC PCR results (**Figure 7**) failed to retain the integrity of cluster 4 in any of the three donors. These observations suggest that subdividing the 33 gene expression profiles into more than 13 clusters may have been warranted starting with subdividing cluster 4. Future functional studies will be useful for supporting or modifying our current classification of 13 clusters for these 33 macrophage-activating conditions.

In order for macrophage activation markers to be useful, it is critical to know whether each marker is regulated by a wide variety or a limited number of stimuli. In our initial time course analysis survey of previously proposed macrophage activation markers, few of the 11 transcripts were found to be highly specific for a specific type of activated macrophage. Therefore, microarrays were performed and then surveyed to identify novel macrophage activation markers. This approach proved to be useful for identifying markers differentially expressed by activated macrophages in all the potent conditions used in this study (**Figure 4**) and in many of the minor clusters (**Figures 7**–**9**).

Our approach of screening for macrophage activation markers by surveying microarray results of a single donor's macrophage

## **References**

1. Stein M, Keshav S, Harris N, Gordon S. Interleukin 4 potently enhances murine macrophage mannose receptor activity: a marker of alternative responses to 33 different activation conditions and following up with IFC arrays proved effective. Also, use of unbiased, bottomup analyses of the microarray results argue against previously proposed top-down linear frameworks describing macrophage activation states, such as the M1–M2 system (3, 9). We note that our results are in line with the spectrum model proposed by Xue et al. from their microarray data set (33). There are likely to be more clusters of activated macrophages than the 13 described here and the 9 described by Xue et al. (33). Taken together, we conclude that measuring the expression changes in a panel of wellcharacterized markers would provide a useful tool to accurately differentiate various activation states associated with functional activity of TAMs or other macrophage populations.

# **Acknowledgments**

We would like to thank Dr. Fayyaz Sutterwala and Dr. John Janczy for advice on preparing immune complexes and Dr. Robert Philibert for use of the Fluidigm BioMark IFC real-time PCR machine. These studies were supported in part by grants 1IK2BX001627 (JG), 5I01BX001983 (MEW), and 5I01BX000536 (MEW) from the Department of Veterans Affairs, Veterans Health Administration, Office of Research and Development, Biomedical Laboratory Research and Development and grants AI076233 and AI045540 from the National Institutes of Health (MEW). Studies were conducted during a period of support for MAW on NIH training grant AI007511 (MW). The content of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the granting agencies.

# **Supplementary Material**

The Supplementary Material for this article can be found online at http://journal.frontiersin.org/article/10.3389/fimmu.2015.00253

**Table S1 | Gene expression profiles in the 33 activated macrophage types for the 1874 regulated transcripts**.

**Table S2 | GO enrichment for 1615 gene names recognized by STRING v. 9.05**.

**Figure S1 | IFC PCR-calculated transcript expression changes correlated well with results from the microarrays**. Scatterplots of gene expression level changes of the indicated transcripts as determined by microarray and by Fluidigm IFC-based RT-PCR. RNA samples collected at 24 h post-treatment from activated MDMs of a single donor were used as template for both assays.

**Figure S2 | Dose-dependent changes in MDM transcript expression levels were minimal across a broad range of concentrations for most of the mild, single stimulus treatments. IFC-based RT-PCR was used to monitor the expression of 48 transcripts in MDMs from a single donor treated with four different concentrations of 11 indicated mild treatment stimuli**. The concentrations tested were 4*×*, 1*×*, 1/4*×*, and 1/16*×* relative to the concentration described for each stimulus in **Table 1**. For each treatment, transcripts that had at least a fourfold change in expression (*>*2 or *<−*2 on log<sup>2</sup> scale) in any of the four tested concentrations were selected for display.

immunologic macrophage activation. *J Exp Med* (1992) **176**:287–92. doi:10. 1084/jem.176.1.287

2. Mantovani A, Sozzani S, Locati M, Allavena P, Sica A. Macrophage polarization: tumor-associated macrophages as a paradigm for polarized M2 mononuclear phagocytes. *Trends Immunol* (2002) **23**:549–55. doi:10.1016/S1471-4906(02) 02302-5


**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 Sudan, Wacker, Wilson and Graff . 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.*