Morphologic, phenotypic, and transcriptomic characterization of classically and alternatively activated canine blood-derived macrophages in vitro

Macrophages are a heterogeneous cell population playing a pivotal role in tissue homeostasis and inflammation, and their phenotype strongly depends on the micromilieu. Despite its increasing importance as a translational animal model for human diseases, there is a considerable gap of knowledge with respect to macrophage polarization in dogs. The present study comprehensively investigated the morphologic, phenotypic, and transcriptomic characteristics of unstimulated (M0), M1- (GM-CSF, LPS, IFNγ-stimulated) and M2- (M-CSF, IL-4-stimulated)-polarized canine blood-derived macrophages in vitro. Scanning electron microscopy revealed distinct morphologies of polarized macrophages with formation of multinucleated cells in M2-macrophages, while immunofluorescence employing literature-based prototype-antibodies against CD16, CD32, iNOS, MHC class II (M1-markers), CD163, CD206, and arginase-1 (M2-markers) demonstrated that only CD206 was able to discriminate M2-macrophages from both other phenotypes, highlighting this molecule as a promising marker for canine M2-macrophages. Global microarray analysis revealed profound changes in the transcriptome of polarized canine macrophages. Functional analysis pointed out that M1-polarization was associated with biological processes such as “respiratory burst”, whereas M2-polarization was associated with processes such as “mitosis”. Literature-based marker gene selection revealed only minor overlaps in the gene sets of the dog compared to prototype markers of murine and human macrophages. Biomarker selection using supervised clustering suggested latexin (LXN) and membrane-spanning 4-domains, subfamily A, member 2 (MS4A2) to be the most powerful predicting biomarkers for canine M1- and M2-macrophages, respectively. Immunofluorescence for both markers demonstrated expression of both proteins by macrophages in vitro but failed to reveal differences between canine M1 and M2-macrophages. The present study provides a solid basis for future studies upon the role of macrophage polarization in spontaneous diseases of the dog, a species that has emerging importance for translational research.


Introduction
Circulating peripheral blood mononuclear cells (PBMCs) play an important role during both the steady state and inflammation. Monocytes, which originate from hematopoietic stem cells, are capable of migrating from the blood into distinct tissues and differentiate into macrophages in order to replenish specific tissue-specific macrophage populations [1]. Functional diversity and plasticity are hallmarks of macrophages [2,3]. Together they represent a heterogeneous cell population of the mononuclear phagocyte system playing a pivotal role in tissue homeostasis, inflammation, host defense, and tissue repair [4,5]. Depending on the micromilieu, two extremes of macrophage phenotypes have been described following external or endogenous stimulation: "classically" activated M1-macrophages and "alternatively" activated M2-macrophages [6,7]. Classically activated M1-macrophages develop after exposure to proinflammatory stimuli such as interferon γ (IFNγ), lipopolysaccharide (LPS), or tumor necrosis factor (TNF). Subsequent to such stimulation, M1-macrophages release pro-inflammatory cytokines, reactive oxygen species (ROS), and nitric oxide (NO) [8]. Hence, on the functional level, M1-macrophages are characterized by an increased microbicidal, tumoricidal, and antigen presenting capacity [2,4,9]. In contrast, M2-macrophages become activated in the presence of interleukin (IL)-4, IL-10, IL-13, glucocorticoids, and transforming growth factor β (TGFβ) leading to enhanced secretion of anti-inflammatory cytokines. Accordingly, M2-macrophages are functionally associated with hypersensitivity, parasite clearance, inflammatory dampening, tissue remodeling, angiogenesis, immunoregulation, and tumor promotion [2,9,10]. However, it should be taken into consideration that the M1-/M2-paradigm is a simplified classification, representing only two extremes of phenotypes which do not fully mirror the complexity of the dynamic biological processes behind cell polarization [7]. Hence, gene expression profiling has been applied as a sophisticated technique to detect the underlying molecular mechanisms following macrophage activation in murine and human cells [11][12][13][14]. In fact, macrophage activation by any agonist involves a massive change in gene expression during the transition from one steady state to another [5]. Notably, current comparative studies strongly support the observation of marked interspecies differences and variability between mice and humans indicating that about 50% of polarization specific markers are selectively expressed in only one of both species [13,14].
Currently, spontaneous diseases in dogs play an established and increasing role as suitable animal models for human disorders including for instance demyelinating central nervous system (CNS) diseases, measles, cancer, and spinal cord injury (SCI) [15][16][17][18][19][20][21]. Conclusively, dogs represent a promising so called large animal model in the development of novel therapeutic approaches for naturally occurring diseases. However, despite the essential role of dogs as a translational animal model, there is a considerable lack of knowledge of the role of macrophage polarization in this species, and morphologic, phenotypic, and transcriptomic properties of polarized canine macrophages are enigmatic so far. However, a detailed knowledge of these basic principles doubtlessly represents a prerequisite for envisaged pharmacotherapeutic and cell transplantation studies.
Therefore, this in vitro study aimed to (i) characterize morphological differences of polarized canine macrophages, (ii) to test the capability of established murine and human prototypical markers to differentiate canine M1-and M2-macrophages using immunofluorescence, and (iii) to unravel differences in the transcriptome of canine polarized macrophages using microarray technique in order to establish unique gene signatures, which differentiate these polarization states.
The presented results provide a highly needed basis for future research upon canine spontaneous diseases such as SCI and distemper leukoencephalitis with a special emphasis upon the so far enigmatic role of macrophage polarization in these diseases.

Blood cell isolation
Blood samples were collected from a total number of 12 healthy Beagle dogs, which were kept in regulatory approved animal housing facilities of the Departments of Small Animal Medicine and Surgery and the Institute for Parasitology, University of Veterinary Medicine, Hannover, Germany. Blood collection was done by professional veterinarians from non-anesthetized dogs following the regulations of the German Animal Welfare Law and with permission provided by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit, Oldenburg, Germany (permission number: AZ33.9-42502-05-13A303). Following a clinical examination of each animal, which included auscultation of the lung and heart, and rectal measurement of the body temperature, the puncture spot was disinfected. A volume of 20 ml blood was collected from the cephalic vein of each dog. Following blood collection, local manual pressure was used to avoid bleeding. No animals were euthanized for the present study.
Peripheral blood mononuclear cells (PBMCs) were isolated via density gradient centrifugation using a routine protocol as described previously [22]. Briefly, blood samples were diluted 1:3 in phosphate buffered saline (PBS) containing 1% penicillin/streptomycin (Biochrom GmbH, Berlin, Germany) and added onto a gradient with equal volumes of histopaque 1.077 g/ml and 1.119 g/ml (Sigma Aldrich, Taufkirchen, Germany). After 30 minutes of centrifugation, cells were harvested using a Pasteur pipette, washed with PBS and 1% penicillin/streptomycin followed by depletion of erythrocyte contamination by hypotone lysis with distilled water. Subsequently, cells were resuspended in Roswell Park Memorial Institute (RPMI)medium 1640 (Biochrom GmbH, Berlin, Germany) containing 10% fetal calf serum (PAA, Cölbe, Germany) and 1% penicillin/streptomycin and seeded onto 96½-well plates (Thermo-Scientific, Waltham, MA, USA) at a density of 0.3 x 10^6 cells/well. After 24 hours of cultivation under standard conditions (5% CO 2 , 37˚C), medium was completely removed and cells were washed twice with PBS and 1% penicillin/streptomycin. Cells attached to the base of wells with strong plastic adherence were referred to as monocytes [23][24][25][26].

Cell culture and polarization
To culture canine macrophages in vitro from blood-derived monocytes, a protocol according to Durafourt et al. (2012) [27] was used with slight modifications. Briefly, canine monocytes were cultured over 7 days towards M1-/M2-macrophages by stimulation with hematopoietic growth factors over 5 days followed by activation over 2 days with distinct cytokines, purchased from companies that guarantee an endotoxin level below 0.1 EU/μg of protein. In particular, to obtain M1-macrophages, monocytes were treated with 5 ng/ml canine recombinant granulocyte macrophage colony-stimulating factor (GM-CSF; R&D Systems, Minneapolis, MN, USA), 100 ng/ml LPS (Sigma Aldrich, Taufkirchen, Germany), and 20 ng/ml recombinant canine IFNγ (Kingfisher Biotech Inc., Saint Paul, Minnesota, USA). M2-macrophages were developed by treatment with 25 ng/ml human macrophage colony-stimulating factor (M-CSF, PeproTech Inc., Rocky Hill, NJ, USA) and 20 ng/ml canine recombinant IL-4 (R&D Systems, Minneapolis, MN, USA). Serving as a control, unstimulated monocytes were cultured under the same conditions with medium change every second day (M0-macrophages). After 7 days of cultivation, morphology, phenotype, and transcriptomic changes of M0-, M1-, and M2-macrophages were assessed. Pictures were taken with an inverted fluorescence microscope (Olympus IX-70, Olympus, Optical Co GmbH, Hamburg, Germany).
The overall number of cells for each polarization was counted on at least 4 microscopic pictures (200x magnification) of three individual dogs on day 7 in culture and the mean numbers of cells per 200x field were compared using one-factorial ANOVA and post-hoc t tests applying SPSS version 21 for windows (IBM Inc., Chicago, USA).
For quantification of cellular morphology, microphotographs were taken at 200x magnification and the number of cells with a distinct morphologic appearance out of at least 100 randomly selected cells was counted for each polarization (M0, M1, M2). Here, 4 distinct morphologic types were defined. In particular, cells with a cellular diameter 10 μm and absent cytoplasmic projections on the cellular surface were defined as small/roundish cells (morphology 1); cells with a cellular diameter ! 10 μm and numerous cytoplasmic processes were defined as amoeboid (morphology 2); cells obtaining an elongated bipolar morphology were classified as spindeloid (morphology 3); and cells possessing a large, round, and flattened morphology with a cellular diameter ! 30 μm and > 1 nuclei were classified as multinucleated giant cells (MNGs; morphology 4).
The percentage of each of the four different morphologies was calculated in M0-, M1-, and M2-macrophages from 3 animals. A mixed ANOVA with post-hoc alpha adjustment (Tukey-Kramer) and significance level at p 0.05 was performed using the statistical software programme SPSS version 21 for windows (IBM Inc., Chicago, USA) in order to compare the percentages of each of the morphologies between M0-, M1-, and M2-macrophages. Graphical compilation was done with GraphPad Prism 5.0 (GraphPad Software Inc., La Jolla, USA).

Scanning electron microscopy
M0-, M1-, and M2-macrophages were prepared for representative ultrastructural characterization using scanning electron microscopy as described previously [28]. Briefly, cells were fixed in 2.5% glutaraldehyde/cacodylate buffer, post-fixed in 1% osmium tetroxide, and dehydrated in a graded series of alcohol. Subsequently, cells were dried under a Critical Point Dryer (E3000, Polaron, London, UK), and sputter-coated with gold, and examined under a digital scanning electron microscope (DSM 940, Zeiss, Oberkochen, Germany). At least 50 cells were photographed for each morphology and evaluated with regard to ultrastructural morphology.

Immunocytochemistry
Immunolabelling of canine M0-, M1-, and M2-macrophages from 5 individuals was performed using a routine protocol described previously [29,30]. A panel of different literaturebased prototype marker antibodies, reported to distinguish between M1-and M2-macrophages was applied (Table 1; [4,31]). Briefly, cells were fixed with paraformaldehyde (PFA, 4%) for 20 minutes at room temperature (RT) and permeabilized with Triton X (0.25%) diluted in PBS (PBST). Non-specific binding was blocked by treatment of cells with bovine serum albumin (BSA, 3%; Sigma Aldrich, Taufkirchen, Germany) and normal goat serum (5%) diluted in PBST for 15 minutes at RT, except for cells intended to be stained with antibodies directed against Fc-receptors, i.e. CD16 and CD32. Subsequently, primary antibodies (Table 1) were added and incubated for 2 hours at RT. All unconjugated primary antibodies were labeled with secondary goat-anti-mouse antibodies coupled to cyanine 3 (Cy3), goatanti-rat antibodies coupled to Cy2, and goat-anti-rabbit antibodies coupled to Cy3 (all received from Jackson ImmunoResearch Laboratories, Dianova, Hamburg, Germany; dilution 1:200 in PBS), respectively, and incubated for 2 hours at RT. Following appropriate washing steps with PBS and distilled water, nuclei were counterstained with bisbenzimide (H33258, 0.01% in distilled water) for 5 minutes at RT. Species-specific immunoglobulins from mouse, rat, goat, and rabbit diluted according to the immunoglobulin concentration of the primary antibodies served as appropriate negative controls. M0-, M1-, and M2-macrophages were investigated using an inverted fluorescence microscope (Olympus IX-70, Olympus, Optical Co GmbH, Hamburg, Germany). For quantification of immunopositivity, microphotographs were taken at 200x magnification and at least 100 randomly selected cells were counted for each treatment (M0, M1, M2) with respect to the number of immunopositive cells and the total number of cells. Statistical comparison of the percentage of immunopositive cells between M0-, M1-, and M2-macrophages was done using a Kruskal-Wallis-Test and pairwise Mann-Whitney-U-Tests with significance level at p 0.05 employing the statistical software programme SPSS version 21 for windows, and box plots were depicted using GraphPad Prism 5.0 (GraphPad Software Inc., La Jolla, USA).
Based on the results of the transcriptome investigations (see below), two additional antibodies targeting LXN and MS4A2 (Table 1) were applied on isolated and polarized cells of 3 dogs. Immunofluorescence was performed analogously to the above mentioned methods. For these antibodies, blocking was done with normal horse serum. Donkey-anti-goat and donkey-antirabbit antibodies were used as secondary antibodies. Due to the lower n (3 dogs), statistical comparison of the percentage of immunopositive cells between M0-, M1-, and M2-macrophages was done using parametrical tests (one-factorial ANOVA and pairwise t tests) for these both antibodies.

RNA isolation, microarray hybridization, and low level analysis
For RNA-isolation a separate and analogous in vitro experiment was performed using bloodderived cells from 6 healthy Beagles. Total RNA was isolated from 6 biological replicates of

Functional annotation and hierarchical clustering analysis
Functional enrichment analysis of the DEPs for overexpressed Gene Ontology (GO) terms of the biological process category and the Kyoto encyclopedia of genes and genomes (KEGG) was performed using Web-based Gene Set Analysis Toolkit (WebGestalt; http://bioinfo. vanderbilt.edu/webgestalt/; [37][38][39]). In order to detect the genes in the pairwise comparison of M0-, M1-, and M2-macrophages, whose expression was most severely affected, genes with a FC!50.0 or -50.0 were selected and grouped into biological categories [36].
Moreover, unsupervised hierarchical clustering analysis of the DEPs was performed on log 2 -transformed data using TM4 MultiExperimentViewer (MeV) with default settings (Euclidean distance and complete linkage) [40]. The list of DEPs for each of the resulting clusters was functionally analyzed for significant enrichment applying WebGestalt as described above.

Correlation-based marker gene selection
Based on the transcriptional profile, a subset of probe sets for polarization prediction was selected using Prophet [41], provided by Babelomics 4.3 [33,42]. Correlation-based feature selection was used to pre-select the informative subset of probe sets and K-nearest-neighbors (KNN) algorithm was used for class prediction with leave-one out error validation [42]. The informative subset of 369 probe sets derived from Prophet was further analyzed for the polarization prediction strength of each individual probe set in three independent "M0 versus M1 and M2", "M1 versus M0 and M2", and "M2 versus M0 and M1" tests using Signature evaluation tool (SET; [42,43]). The log 2 -transformed expression values of three biomarkers, suggested by Prophet, were additionally statistically compared between M0-, M1-, and M2-macrophages, employing Kruskal-Wallis-Test and subsequent pairwise Mann-Whitney-U-Tests with significance value p 0.05.
Comparative evaluation of canine M1-/M2-genes with established literature-based human and murine orthologous genes All unique gene symbols, which were up-regulated (FC!2; q 0.05) in the comparison of M1 vs. M0 and simultaneously up-regulated in the comparison of M1 vs. M2 were defined as M1-exclusive-genes (n = 404). Likewise, all unique gene symbols, which were up-regulated (FC! 2; q 0.05) in the comparison of M2 vs. M0 and simultaneously up-regulated in the comparison of M2 vs. M1 were defined as M2-exclusive-genes (n = 700). The gene sets of these exclusive M1-or M2-associated canine genes were compared with a list of established human and murine genes specifically associated with M1-or M2-macrophages. This list was generated based on peer-reviewed publications [4,27,31,44] as previously described [8,45]. This list was translated into canine orthologous gene symbols by employing MADgene (http://cardioserve. nantes.inserm.fr/madtools/madgene/; [46]) and missing orthologous canine official gene symbols were manually added with the help of the web-based "information hyperlinked over proteins" (ihop; http://www.ihop-net.org/UniPub/iHOP/; [47]) as described previously [36]. Summarized, the literature-based gene list included a total number of 65 orthologous canine genes for M1-macrophages and 58 orthologous canine genes for M2-macrophages [36]. Venn diagrams (http://bioinfogp.cnb.csic.es/tools/venny/index.html) were used to reveal the intersections between established literature-based genes and the gene sets identified by the present analysis.

Morphological characterization of polarized canine macrophages
On day 7 in culture, the overall number of cells differed between the three polarities (p = 0.02; S1 Fig). The number of cells was highest in M2-polarized cells (mean = 135 cells per field), while M0-macrophages were lowest in number (mean = 30 cells per field). M1-macrophages had a mean number of 52 cells per field. T tests revealed that the number of M2-macrophages was significantly higher than M0-macrophages (p = 0.02), while comparisons between M0and M1-macrophages and M1-and M2-macrophages failed to reach the level of significance (p = 0.30 and p = 0.06, respectively).
Morphologically, striking differences between M0-, M1-, and M2-polarized cells were observed at day 7 in culture by scanning electron microscopy and phase contrast microscopy (Fig 1). Canine M0-macrophages predominantly appeared as small and roundish cells with an average size of 8 μm in diameter and no or little cytoplasmic extensions that measured up to 1 μm in length (Fig 1). However, a noteworthy proportion of M0-macrophages (about 25%) also obtained an amoeboid morphology, which however was predominantly observed in M1 macrophages (Fig 1). The majority of M1-macrophages was amoeboid and had a mean size of 15 μm and numerous fibrillary cytoplasmic processes on the cellular surface that had a length of up to 5 μm (Fig 1). M2-macrophages appeared as a heterogeneous cell population with a mixture of 4 different morphologies. Besides roundish and amoeboid macrophages, large bipolar spindeloid macrophages were present measuring up to 35 μm that were characterized by an elongated cell body with cytoplasmic extensions with an average length of 7 μm at both poles (Fig 1). Moreover, large MNGs appeared that had a mean size of 40 μm with an extensive cytoplasm and numerous evenly distributed processes that measured up to 7 μm (Fig 1). Allover, the mean percentage of small/roundish macrophages (morphology 1) was significantly higher in untreated cultures (M0) compared to polarity 2 at days 5 and 7 (p 0.05; Fig 1E). For amoeboid macrophages (morphology 2), there was a trend towards a higher mean percentage in polarity 1 compared to polarity 0 at day 5 (p = 0.07) and day 7 (p = 0.06; Fig 1F). The mean percentage of spindeloid macrophages (morphology 3) was significantly higher in polarity 2 at day 5 compared to polarity 1 at day 5 (p 0.05; Fig 1G). MNGs (morphology 4) were exclusively present in polarity 2, as compared to the polarities 0 and 1 at day 7 (p 0.05; Fig 1H).

Phenotypical characterization of polarized canine macrophages
For the phenotypical characterization of canine macrophages, the percentage of immunopositive cells for selected literature-based M1-/M2-antigens was evaluated in 5 animals and related to the polarization of cells (polarity 0, 1, 2). Interestingly, except for MHC class II and CD206, none of the remaining tested antigens (CD16, CD32, iNOS, CD163, and arginase-1) were differently expressed between canine M0-, M1-, and M2-macrophages (p-values ranging from 0.101 to 0.691; Kruskal-Wallis-Test; Fig 2). The mean percentage of immunopositive cells expressing MHC class II was significantly higher in M2-macrophages (mean positive cells: 93.64%) when compared to untreated M0-cells (mean positive cells: 72.22%; p = 0.008). However, there was no statistical difference in the percentage of positive cells for MHC class II, when M1-macophages (mean positive cells: 89.92%) were compared to M2-macrophages and M0-macrophages, respectively (Fig 2).

Differentially expressed genes between M0-, M1-, and M2-polarized macrophage cultures
One-factorial multigroup analysis of microarray data and fold change criteria identified 6358 probe sets that were differentially expressed in at least one of the three post-hoc pairwise comparisons. The total number of DEPs was 3555 in M1 vs. M0, 4831 in M2 vs. M0, and 3141 in M2 vs. M1, respectively ( Table 2). In all pairwise comparisons, the number of up-and downregulated probe sets was nearly equally distributed (M1 vs. M0: 1699 up, 1856 down; M2 vs. M0: 2467 up, 2364 down; M2 vs. M1: 1572 up, 1569 down). Functional annotation of enriched biological processes in DEPs, up-regulated in the comparison M1 vs. M0, revealed terms for "organonitrogen compound metabolic process", "carbohydrate metabolic process", "carboxylic acid metabolic process", and "tricarboxylic acid cycle" ( Table 2). In contrast, biological terms in down-regulated DEPs in this comparison reflected terms for "positive regulation of immune response", "immune response-regulating signaling pathway", and "regulation of innate immune response". In both comparisons M2 vs. M0 and M2 vs. M1, up-regulated DEPs displayed significantly enriched gene ontology terms for "M phase of mitotic cell cycle" and "mitotic spindle organization". Additionally, in M2 vs. M0 comparison, up-regulated DEPs were associated with "oxidation-reduction process", "carboxylic acid metabolic process", and "organonitrogen compound metabolic process", whereas down-regulated DEPs were related to the biological term "immune response-activating signal transduction". In the comparisons M2 vs. M1, down-regulated DEPs were functionally associated to biological terms such as "response to other organism", "defense response", "regulation of lymphocyte activation", and "regulation of immune response" (Table 2).
Interestingly, gene expression of CD209 markedly decreased in the pairwise comparison of

Hierarchical cluster analysis
Unsupervised hierarchical clustering analysis formed 9 different clusters based on similarities and differences in the expression profile of DEPs (Fig 4). Two out of these 9 clusters had an expression profile that was visually clearly associated with either the M1-or M2-phenotype (Fig 4). Functional annotation of these clusters identified that the M1-polarization cluster (cluster 8) was significantly associated with the biological process "respiratory burst involved in defense response" (Fig 4). The M2-cluster (cluster 6) was significantly associated with multiple biological processes of mitosis such as "M phase of mitotic cell cycle" and "mitotic spindle organization" (Fig 4). The genes associated with these M1-and M2-specific clusters are listed with their particular fold changes in S1 and S2 Tables. The remaining 7 clusters were neither specific for M1-nor for M2-macrophages and were associated with biological terms like "peptidyl-lysine mono-and dimethylation" (cluster 1), "immune response-activating signal transduction" (cluster 3), "monosaccharide metabolic process", "organic substance catabolic process", "cellular catabolic process" (cluster 4), "response to other organism", "regulation of lymphocyte proliferation" (cluster 5), "tRNA aminoacylation for protein translation" (cluster 7), "cytokinesis" and "antigen receptor-mediated signaling pathway" (cluster 9, S3 Table). No significantly enriched biological terms were identified for cluster 2.

Literature-based gene signatures and marker genes
Intersections of genes exclusively up-regulated in M1 vs. M0 and down regulated in M2 vs. M1 (M1-macrophage genes), as well as genes up-regulated in the comparisons M2 vs. M0 and M2 vs. M1 (M2-macrophage genes) were selected and compared to literature-based markers that are known to distinguish between human and murine M1-and M2-macrophages ( Fig 5A).
Notably, many markers identified in the recent study to be specific for canine M1-polarization (404 unique genes in total corresponding to 565 probesets, S5 Table) did not match with the literature-based M1-markers (65 genes). Similarly, predominating numbers of canine M2markers (700 unique genes in total corresponding to 1029 probesets, S5 Table) did not match with the literature-based M2-markers (58 genes; Fig 5A). However, overlapping expression of genes reflecting M1-polarization state was present for 8 genes, i.e. BIRC3, CCR7, CD80,   5A). Conflicting results were present for the genes P2RY14, TGFBR2, and TPST2, which were expected as M2-markers based on the literature but were differentially up-regulated in canine M1-macrophages. Furthermore, PIK3CB was up-regulated by both canine M1-and M2-macrophages in the present study. The differentially expressed genes of the intersections, which are not namely mentioned in Fig 5A are listed in S4 Table. The top 50 candidate canine M1-and M2-macrophage genes, as defined above, are given in Tables 6 and 7. The full lists of canine probesets, corresponding to the 404 unique M1 genes and 700 unique M2 genes are shown in S5 Table. Biomarker selection In a hypothesis-driven approach, polarization specific prediction markers were detected with Prophet using the microarray data of unstimulated macrophages (M0), M1-, and M2-macrophages [41,42]. 369 probe sets were identified by the correlation-based feature selection algorithm of Prophet that discriminated between M0-, M1-, and M2-polarity, using the KNN algorithm with 100% correct predictions. Subsequent "one versus all" analyses using SET to create ranked lists of genes, based on their potential to serve as biomarker for M1-or M2-macrophages were performed (S6 Table; [43]). The highest scoring probe set differentiating M1from M0-and M2-macrophages was CfaAffx.14358.1.S1_at, annotated as latexin (LXN), which was up-regulated in M1-macrophages, exhibiting a prediction-accuracy for the M1-phenotype of 100%. Additionally, the probe set ID's occupying the ranks 2 to 4 (Cfa.5195.1. A1_s_at, Cfa.5195.1.A1_at, Cfa.14007.1.A1_x_at) were similarly annotated to LXN (S6 Table). Analysis of the expression data of CfaAffx.14358.1.S1_at revealed a relatively high expression in all three polarities with a significantly higher expression in M1-macrophages as compared to M0-and M2-macrophages (pairwise Mann-Whitney-U-Tests; p = 0.02, Fig 5B). For the comparison of M2 versus M0 and M1, the highest scoring probe set was Cfa.3662.1.S1_at, Characterization of polarized canine macrophages  Table). This probe set was up-regulated in M2-macrophages and displayed a prediction accuracy of 100% for this phenotype. Statistical evaluation with pairwise Mann-Whitney-U-Tests underlined a low expression of MS4A2 in M0-and M1-macrophages, but a significantly higher expression in M2-macrophages (p = 0.02, Fig 5C).

Testing of antibodies for the detection of predicted biomarkers
In order to test, whether the biomarkers predicted to distinguish between canine M0-, M1-, and M2-macrophages during the transcriptome investigations are also mirrored by altered protein expression, an additional experiment using blood from 3 dogs was performed. The cells were isolated and polarized as described and labeled with antibodies targeting MS4A2 and LXN (Table 1). There was protein expression in canine macrophages in vitro for both molecules (Fig 6). However, the percentage of cells labeled with an antibody against MS4A2 showed no differences between M0-, M1-, and M2-macrophages (p = 0.117; Fig 6). The Characterization of polarized canine macrophages percentage of positive cells for LXN was significantly higher in both M1-and M2-macrophages as compared to M0-macrophages (p = 0.02 and p = 0.01, respectively); however, there was no difference in the percentage of LXN-immunopositive cells between M1-and M2-macrophages (p = 0.06; Fig 6). A) The analysis identified 9 distinct clusters (I-IX) based on similarities as well as differences in the expression intensity of canine polarized macrophages. Two out of these 9 clusters (VIII and VI) visually displayed an expression profile that was clearly associated with either the M1-or M2-phenotype. B) Functional annotation of the M1-cluster (magnified from A) using Web-based Gene Set Analysis Toolkit (WebGestalt) identified the enriched biological process category "respiratory burst" (adjusted p-value 0.05). C) The M2-cluster (magnified from A) is associated with enriched biological GO terms such as "M phase of mitotic cell cycle" and "mitotic spindle organization" (adjusted p-value 0.05). https://doi.org/10.1371/journal.pone.0183572.g004 Characterization of polarized canine macrophages

-based human and murine orthologous genes (A) and detection of polarization-specific biomarkers employing supervised clustering with a K-nearest-neighbors algorithm and correlation-based feature selection using Prophet (B, C). A)
The Venn diagram depicts the numbers and intersections of differential and common canine M1-and M2-genes with literaturebased human and murine genes. The majority of literature-based M1-(65 genes = 57+8) and M2-markers (58 genes = 44+11+3) did not comply with the present microarray data upon canine macrophages. However, overlapping expression of 8 genes for the M1-phenotype and 11 genes for the M2-phenotype was identified. The genes of the intersections, not specifically mentioned in the figure, are listed in S4 Table B) Biomarker selection detected 4 highly specific probe sets for annotated by the mammalian carboxypeptidase inhibitor latexin (LXN) to be highly correlated to the M1-phenotype. The boxplots depict the significantly enhanced log 2 -transformed expression values for LXN in M1-macrophages as compared to both M0-and M2-macrophages (p = 0.002), as well as between M0-and M2-macrophages (p = 0.002). C) For the M2-phenotype, the most significant predicted biomarker was high affinity receptor for IgE, i.e. membrane-spanning 4-domains, subfamily A, member 2 (MS4A2). The expression data of M2-macrophages show significantly higher log 2 -transformed expression values as compared to both M0-and M1-macrophages (p = 0.002) as well as between M0-and M1-macrophages (p = 0.002). Asterisks indicate significance (Kruskal-Wallis-Test with subsequent pairwise Mann-Whitney-U-Tests). https://doi.org/10.1371/journal.pone.0183572.g005

Discussion
The current investigation is the first report upon the properties of canine polarized macrophages in vitro with a special emphasis on the establishment of unique distinctive gene signatures. Even though in vitro data on macrophage polarization doubtlessly cannot be simply extrapolated to in vivo situations and the paradigm of M1-/M2-polarization represents a simplified approach, which depicts only two extremes of macrophage heterogeneity [8,[48][49][50], this initial reductionistic in vitro approach will provide a basis for future investigations on the role of macrophage polarization in both healthy and diseased dogs. Striking morphological differences were observed between canine M0-, M1-, and M2-macrophages, which are most probably attributed to direct effects of cytokine stimulation. Similar to observations in other species [51][52][53][54], M1-macrophage cultures appeared to dominantly adapt an amoeboid morphology, although a significant proportion of M0-macrophages also obtained such an amoeboid morphology. Notably, M2-cultures were rather heterogeneous including spindeloid cells and MNGs, which is in line with previous observations in human and murine bone marrow-and blood-derived M2-macrophages [51][52][53][54]. Interestingly, stimulation of macrophages with IL-4 and IL-13 together with colony stimulation factors can lead to the formation of MNGs [55][56][57][58]. The recent observation of MNG formation, exclusively in canine M2-cultures, is consistent with reports from other species [44,57,59].
A variety of different phenotypic markers has been proposed for the differentiation of human and murine M1-and M2-macrophages [4,31]. Even though a multitude of antibodies targeting immune cells including macrophages is known to cross-react with canine antigens [60][61][62][63], commercially available antibodies explicitly designed for the use on canine cells and tissues are frequently not available. The prototypical literature-based M1-markers CD16, CD32, and iNOS and the M2-markers CD163 and arginase-1 were demonstrated to be inappropriate for the immunophenotypic discrimination of M0-, M1-, and M2-polarization states in canine macrophages in the present study. This may in part be explained by interspecies differences [64,65]. For instance, following classical stimulation with pro-inflammatory cytokines, murine macrophages produce NO, whereas human macrophages nearly lack synthesis of NO in response to classically activating stimuli [66,67]. Moreover, arginase-1 has been reported as a prototype marker for murine M2-macrophages, which is however inappropriate for the detection of human M2-macrophages [13,44,68,69]. Based on the present observations, canine macrophages may thus share closer similarities with human than murine macrophages, thus underlining the role of dogs in translational research. Similar to the immunofluorescence data, and supporting the results of the immunophenotyping, expression values for all probesets annotated with the canine genes NOS2 (2 probesets) and ARG1 (3 probesets), which encode iNOS and arginase-1, respectively, also lacked differences in their expression values between M0-, M1-, and M2-macrophages (data available under accession number: E-MTAB-5458; http://www.ebi. ac.uk/arrayexpress). MHC class II is sometimes regarded as a marker for classically activated M1 macrophages [70]; however, a substantial fundus of literature also indicates MHC class II as a pan-macrophage marker, which is expressed on both M1 and M2 macrophages in mice and Characterization of polarized canine macrophages humans without discriminating between both phenotypes [4,7,68]. Concordantly, the immunofluorescence data demonstrated that MHC class II was found to distinguish between canine M2-and M0-macrophages with a lower expression in the latter (Fig 3). In fact, it is well known that IL-4 acts as a potent activator of macrophages and induces an up-regulation of MHC class II [23, [71][72][73][74][75]. However, it must be critically considered that MHC class II appeared inappropriate for the differentiation of canine M0-and M1-macrophages on the protein level, which contradicts reported up-regulation of MHC class II on M1 macrophages [7]. Whether these differences are attributed to true species effects or whether the low number of investigated animals and the quantification method (immunofluorescence) are responsible for not reaching the level of significance remains to be determined. Conclusively, it is unlikely that a single antibody will be sufficient to specifically detect canine M1-and M2-macrophages in vivo. Due to the paucity of commercially available antibodies detecting canine antigens this will probably also involve the generation of antibodies targeting canine epitopes.
The present study thus aimed to set up a transcriptomic basis that should encourage future attempts to establish a broader antibody panel against antigens, which are predicted to potentially represent discriminating markers for canine macrophage phenotypes. Interestingly, in the present study, CD206 was among the top regulated genes of canine M2 macrophages on the transcriptome level, and the results of the immunophenotyping also validated CD206 as a promising marker for canine M2 polarized macrophages. This implies that CD206 might constitute a conserved marker, which is appropriate for labeling M2-macrophages in various species including the dog [4,27,31,72,76]. However, a limitation of the present study is that the suitability of markers identified in the transcriptome analyses either still needs to be verified on the protein level or produced partially conflicting results in the immunofluorescence investigations (i.e. CD163, LXN, and MS4A2).
For M1-associated up-regulated genes, multiple metabolic pathways such as "steroid biosynthesis", and "glycolysis/gluconeogenesis" demonstrated to be enriched in the functional annotation of microarray data. Human and murine monocyte to macrophage differentiation has previously been shown to go along with profound changes in the lipid metabolism as a prerequisite for phagocytosis [77]. Moreover, in the hierarchical clustering analysis, the M1-specific cluster was functionally annotated to biological processes involved in "respiratory burst". M1 macrophages produce a variety of pro-inflammatory mediators including ROS, whereas in contrast, IL-4 inhibits the respiratory burst [78].
Enriched terms related to the peroxisome proliferator activated receptor (PPAR) signaling pathway were retrieved for the comparison of M2 vs. M1 up-regulated genes. PPARγ is a member of the nuclear receptor superfamily with potent anti-inflammatory properties and regulatory functions in fatty acid metabolism [79][80][81][82]. Interestingly, PPARγ agonists such as rosiglitazone induce an alternative M2a-activation state in murine macrophages and have been used as neuroprotective agents [83,84]. Pharmacological approaches, designed to enhance M2-dominated immune responses may thus similarly represent a promising tool in canine diseases. Interestingly, multiple biological processes related to an enhanced cell cycle were enriched in canine M2-macrophages as compared to both M0-and M1-macrophages. IL-4 has previously been reported to induce local macrophage proliferation in the context of chronic inflammation [59]. Moreover, human monocyte to macrophage differentiation in the presence of M-CSF is associated with a dramatic regulation of multiple cell-cycle genes [44]. The Characterization of polarized canine macrophages transcriptomic link to enhanced cell cycle and proliferation is probably also reflected by the higher cell number of M2-macrophages in the present study as compared to M0-macrophages (S1 Fig). Comparison of the canine transcriptome data with published murine and human prototype markers demonstrated a relatively low overlap. This confirms recent reports on marked interspecies differences between polarized macrophages [13] and suggests that some properties of polarized macrophages are unique to the dog, demonstrating that reported literature-based markers cannot simply be transferred to another species.
Based on this observation we sought to predict novel transcriptomic markers using a hypothesis-driven approach. Using a correlation-based algorithm, the carboxypeptidase inhibitor latexin (LXN) [85][86][87][88] was retrieved for canine M1-macrophages. Though LXN-expression by murine macrophages upon pro-inflammatory stimulation has been demonstrated [85], LXN has so far not been proposed as a marker for M1-macrophages in the literature for any species. For the canine M2-phenotype, the high-affinity receptor for IgE membrane-spanning 4-domains, subfamily A, member 2 (MS4A2) was predicted to represent the most powerful biomarker. Interestingly, other members of the molecule family, i.e. MS4A4A and MS4A6A, have been previously shown to be associated with M2-polarization [4,44,89]. In an attempt to validate LXN and MS4A2 as markers for canine macrophages, we demonstrated that MS4A2 showed no differences in the percentage of immunopositive cells between the three conditions. The percentage of immunopositive cells for LXN was higher in both M1-and M2-macrophages as compared to M0-macrophages. However, similar to MS4A2, LXN failed to distinguish between canine M1-and M2-macrophages. Thus, similar to CD163, the protein data did not accurately mirror the transcriptomic prediction. This could be attributed to conflicting differences between RNA-and protein level for these molecules. However, the low number of dogs tested and the fact that low level differences in protein expression may not be detected by immunofluorescence have certainly influenced the statistical power and sensitivity, and thus future validating experiments both in canine tissues and in vitro remain to be performed. One-factorial ANOVA with group-wise t tests reveals a significantly higher number of cells in M2-polarized macrophages as compared to non-stimulated (M0)-macrophages (p 0.05; asterisk). (TIF) S1 Table. List of the genes included in the M1-associated cluster of the hierarchical clustering analysis (refer to Fig 3). Table. List of the genes included in the M2-associated cluster of the hierarchical clustering analysis (refer to Fig 3). Table. Overview on retrieved enriched gene ontology biological processes and KEGG pathways of the clusters, resulting from the hierarchical clustering analysis (refer to Fig 3). Table. List for the intersections of differentially expressed genes in the comparison of literature-based human and murine markers with canine M1-and M2-associated genes as retrieved by the present study (refer to Fig 4). Excel table. (XLSX) S5 Table. Lists of differentially expressed M1-and M2-macrophage associated probesets with fold change. Sheet 1 depicts all genes, which were upregulated in M1-vs. M0-macrophages and simultaneously downregulated in M2-vs M1-macrophages (i.e. canine M1-macrophage genes). Sheet 2 shows all genes, which were upregulated in M2-vs. M0-macrophages and simultaneously upregulated in M2-vs M1-macrophages (i.e. canine M2-macrophage genes). (XLSX) S6