Gene expression profiles and transcriptional regulatory pathways underlying mouse tissue macrophage identity and diversity

Tissue macrophage diversity

As part of ImmGen, we sorted several tissue macrophages populations from C57BL/6J mice according to strict, standardized procedures and analyzed these populations using whole-mouse genome Affymetrix Mouse Gene 1.0 ST Arrays. Sorting strategies for these populations can be found on the ImmGen website (http://immgen.org), and gene expression data are deposited in the Gene Expression Omnibus (GEO, accession # GSE15907).

We began our analysis by examining the gene expression profiles of resting macrophage populations that have historically been highly characterized and accepted as bona fide resident tissue macrophages¹⁶. Though some classical macrophages, such as Kupffer cells of the liver and metallophilic as well marginal zone macrophages of the spleen proved elusive for definitive identification and/or isolation by flow cytometric cell sorting, four resting macrophage populations submitted to Immgen met the criteria of bona fide macrophage populations: peritoneal macrophages, red pulp splenic macrophages, lung macrophages, and microglia (brain macrophages). We thus focused our initial analysis on these four key macrophage populations. Principal component analysis (PCA) of all genes expressed by the four sorted populations and several DC populations revealed a relatively greater distance between the different macrophages compared with DCs (Fig. 1a). Pearson correlation values were high for replicates within a given DC or macrophage population as per the quality control standards of Immgen; variability within replicates for a single population varied from 0.908 ± 0.048 for microglia to 0.995 ± 0.001 for peritoneal macrophages. Pearson correlations in gene expression profiles between different populations of DCs yielded coefficients ranging from 0.877 (liver CD11b⁺ versus spleen CD8⁺ DCs) to 0.966 (spleen CD4⁺CD11b⁺ versus spleen CD8⁺ DCs) (mean of all DC populations 0.931), whereas the correlation coefficients between different tissue macrophages ranged from 0.784 (peritoneal versus splenic red pulp) to 0.863 (peritoneal versus lung) with a mean of 0.812 (Fig. 1b). Several thousand mRNA transcripts were differentially expressed by at least 2-fold when, for example, lung macrophages were compared to red pulp splenic macrophages (Fig. 1c). This degree of diversity was greater than that observed when DCs from different subsets (CD103⁺ versus CD11b⁺) were compared from different organs (Fig. 1c). Finally, a dendrogram applied to the various populations showed that DCs clustered more closely than did macrophages (Fig. 1d), and this was true whether we considered all gene transcripts in the array (data not shown) or only the top 15% ranked by cross-population max/min ratio or coefficient of variation (Fig. 1d). Overall, these comparisons indicate a pronounced diversity among tissue macrophage populations.

Distinct molecular signatures among tissue macrophages

The diversity among these four classical macrophage populations extended to gene families previously associated with macrophage function - chemokine receptors, Toll-like receptors, C-type lectins, and efferocytic receptors. For example, at least one distinct chemokine receptor observed in each population was prominently expressed above the others (Supplementary Fig. 1a). Diversity among Toll-like receptors, C-type lectin domain members and efferocytic receptors was also remarkable (Supplementary Fig. 1b–d). Indeed, only a few of the mRNA transcripts profiled in these categories, including mRNA encoding the Mer tyrosine kinase receptor (MerTK) involved in phagocytosis of apoptotic cells¹⁷, and toll-like receptors Tlr4, Tlr7, Tlr8, and Tlr13, showed relatively uniform expression across all macrophages compared. Hundreds of mRNA transcripts were selectively increased or decreased by at least 2-fold in only one of the macrophage populations (Fig. 2a), and microglia in particular displayed low expression of hundreds of transcripts that were expressed in other macrophage populations (Fig. 2a). Using Ingenuity pathway analysis tools, we found each specific signature to be enriched in groups of transcripts with predicted specific functions, including those with oxidative metabolism in brain macrophages, lipid metabolism in lung macrophages, eicosanoid signaling in peritoneal macrophages, and readiness for interferon responsiveness in red pulp macrophages (Supplementary Table 1). Considering that the gene expression profiles of four macrophage populations were simultaneously compared, the number of transcripts that were increased or decreased ≥ 5-fold in only one macrophage population relative to all three of the others was striking (Fig. 2b). We also noted that many transcripts were especially strongly reduced in only one population compared to the others (Supplementary Fig. 2). Several transcription factors were markedly increased in just one of the four macrophage populations (Fig. 2c). For example, Spic was restricted to splenic red pulp macrophages, fitting with previous work revealing that this transcription factor plays a critical role in controlling the development of these cells¹⁵. Diversity at the gene expression level translated to the protein level. For example, CD11a and EPCAM were detected on lung macrophages but not microglia, spleen or peritoneal macrophages; VCAM-1 and CD31 were selectively displayed by spleen macrophages; CD93 and ICAM-2 were expressed by peritoneal macrophages but not the others macrophages; and CX3CR1 and SiglecH were selectively found in microglia (Fig. 2d). All together, these data indicate that macrophage populations in different organs express many unique mRNA transcripts that equip them for specialized local functions.

Identification of core transcripts expressed in tissue macrophages but low to absent in classical DCs

In the midst of the rather vast diversity among macrophages from different organs, we next wondered if we could identify a core gene expression profile that generally unified macrophages over other types of immune cells. Among all hematopoietic cells, the cells anticipated to be most similar to macrophages are DCs⁵. To search for mRNA transcripts that distinguished macrophages from DCs, we compared the four selected prototypical macrophage populations to the most well-defined classical DC populations, including resting CD8⁺ and CD4⁺CD11b⁺ splenic DCs, CD103⁺ tissue DCs and various populations of lymph node MHC-II^hi CD11c⁺ migratory DCs¹⁸. Because tissue CD11b⁺ DCs may be contaminated with macrophages¹⁹, tissue CD11b⁺ DCs were initially excluded from the comparison. This comparison revealed only 14 transcripts that were expressed in all 4 macrophages but not expressed in DCs (Table 1, upper left column, bolded gene names). These included few of those anticipated to be highly expressed in macrophages, including Fcgr1 (CD64) and Tlr4. Two of these molecules, G-CSF receptor (Csf3r) and the MHC-I-related gene Mr1 involved in activation of mucosal-associated invariant T (MAIT) cells²⁰, function at least partly at the cell surface. In agreement with the pattern of mRNA expression, we found MR1 protein on spleen and lung macrophages but not classical DCs (Supplementary Fig. 3), suggesting that MR1 on macrophages rather than DCs may drive MAIT cell activation. Other transcripts encode proteins involved in signal transduction, such as Fert2 encoding the fms/pfs-related protein kinase, or in metabolism and lipid homeostasis such as peroxisomal trans-2-enoyl-CoA reductase (Pecr) and alkyl glycerol monooxygenase (Tmem195), the latter being the only enzyme that cleaves the O-alkyl bond of ether lipids like platelet-activating factor that has been shown to be actively catabolized in association with macrophage differentiation in vitro²¹.

To this small number of mRNA transcripts, we added probe sets that were not absent in expression by DCs, but were at least 2-fold lower in signal intensity in all single DC populations than the lowest intensity of that same probe set in each macrophage population. Thus, we were able to add 25 more transcripts to this “macrophage core” list (Table 1, lower left column, non-bolded gene names), including those known to be associated with macrophages like Cd14, Mertk, Fcrg3 (CD16) and Ctsd (Cathepsin D).

F4/80, encoded by the gene Emr1, has served as the most definitive marker of macrophages to date^16,22. However, in order to identify additional mRNA transcripts widely associated with macrophages with the core list of macrophage-associated genes, including Emr1 (F4/80), Mafb, and Cebpb, we found it necessary to adjust the criteria of the above approach to include transcripts expressed in only 3 of 4 macrophage populations because, for instance, Emr1 mRNA was low in microglia. Making this adjustment expanded the list of mRNA transcripts associated with macrophages, adding another 93 genes (Table 1). Additional macrophage-associated genes like Mrc1 (CD206, mannose receptor), Marco and Pparg were not identified until we loosened the criteria such that only 2 out of 4 prototypic macrophages needed to express a given transcript that was otherwise absent or low on DCs (Table 2). The mRNA encoding CD68 (Cd68), widely used to identify tissue macrophages, was expressed at similar levels in DCs and macrophages and so excluded from the list. However, at the protein level, it was still several orders of magnitude higher in macrophages than in DCs of the spleen (Supplemental Fig. 4), an organ where mRNA levels were scarcely different. In summary, numerous transcripts, 366 altogether (Tables 1 and 2), were absent or markedly reduced in classical DCs relative to macrophages. However, because of the great diversity among macrophages, only 39 of these transcripts are shared by all tissue macrophages we compared.

Co-expressed gene modules in macrophages and predicted transcriptional regulators

The computational biology groups of the ImmGen project analyzed the transcriptional program of the entire large database generated in the ImmGen project (Jojic et al, in preparation; Supplementary Note 1). First, mRNA transcripts were clustered into 334 fine modules based on patterns of co-expression. Then a novel algorithm termed Ontogenet, developed for the ImmGen dataset, was applied in order to find a regulatory program for each fine module, based on its expression pattern, the expression pattern of regulators and the position of the cells on the hematopoietic lineage tree. ImmGen modules, including the gene lists in each module, and regulatory program metadata are available online (http://www.immgen.org/ModsRegs/modules.html), and the numbering of the modules reported on the website is used herein.

When the list of the 366 mRNA transcripts associated with macrophages was mapped according to the their placement into various fine modules, 14 modules were significantly enriched for the macrophage-associated gene signature we identified (Fig. 3a). In particular, the 11 genes that comprise module 161 were significantly induced in all 4 macrophages used to generate the list of macrophage-associated genes (Fig. 3a). Other modules, such as module 165, contained genes significantly induced in several specific groups of macrophages, but not in all (Fig. 3a). The 11 genes that comprise module 161 (A930039a15Rik, Akr1b10, Blvrb, Camk1, Glul, Myo7a, Nln, Pcyox1, Pla2g15, Pon3, Slc48a) are involved in redox regulation, heme biology, lipid metabolism, and vesicular trafficking (Supplementary Table 3). Beyond the comparison to DCs, genes in module 161, expressed in all macrophages, were not expressed by any other hematopoietic cell types including granulocytes (GN) or any of the blood monocyte (MO) subsets (Fig. 3b), importantly indicating that this list of genes is selectively associated with mature macrophage differentiation in the hematopoietic system.

As a framework for future studies on the transcriptional control of macrophage development, maintenance and function, we examined the predicted activators Ontogenet assigned to the modules associated with the macrophage core genes. As a specific example, the activators predicted by Ontogenet algorithm to control the expression of the 11 gene transcripts that form module 161 are listed in Fig. 3b. Overall, a highly overlapping set of 22 regulators emerged in the 14 macrophage-associated modules (Fig. 3c). In particular, TCFE3, C/EBPα and BACH1 were predicted activators in a majority of these modules (>75%) and especially novel regulators like CREG-1 also came up prominently. Among the 22 regulators associated with the 14 modules, 18 of them are predicted using Ingenuity pathway tools to interact in a regulatory network based on known protein-protein interactions or mutual transcriptional regulation (Fig. 3d). These regulators represent 5 main families of transcriptional factors as depicted in Fig. 3d. The statistical evaluation score generated for this network revealed a p-value ≤ 10⁻³⁵.

Beyond modules of genes that unified the 4 tissue macrophage populations we studied, several modules were selectively associated with a single macrophage population (Supplementary Table 4). In these specific modules, predicted regulators included Spi-C for red pulp macrophages, confirming a regulation that is already known¹⁵ and thus supporting the predictive power of the algorithm, and GATA6 as a regulator of peritoneal macrophages (Supplementary Table 4).

Use of the macrophage core signature to evaluate identity and heterogeneity of less characterized macrophage and DC populations

Finally, we utilized the resting macrophage core signature defined above to assess mononuclear phagocyte populations that we earlier excluded from our core analysis due to low levels of information on a given population or controversial discussions in the literature about origins or functional properties, including whether they should be classified as DCs or macrophages. In the ImmGen database (http://www.immgen.org), each population was assigned a classification as DCs or macrophage (Mac) a priori. The names of these populations will be used below and in Figure 4 for clarity and consistency with the database. These populations included resting and thioglycollate-elicited (Thio) mononuclear phagocytes that express CD11c and MHC II (Supplementary Fig. 5), skin Langerhans cells, bone marrow macrophages²³, and putative CD11b⁺ tissue DCs including those in the liver and gut. All thioglycollate-elicited cells from the peritoneal cavity, even those co-expressing CD11c and MHC II, strongly expressed genes in the macrophage core, including module 161 itself, similar to the prototypic macrophage populations used to generate the core (Fig. 4a and 4b), indicating that these cells are indeed macrophages despite co-expression of CD11c and MHC II. On the other hand, Langerhans cells, and CD11c⁺ MHC-II⁺ CD11b⁺ cells from the liver (CD11b⁺ liver DCs in the ImmGen database), did not robustly express the macrophage core signature or module 161 alone, nor did bone marrow macrophages (Fig. 4a and 4b). CD11c⁺ MHC-II⁺ CD11b⁺CD103⁻ cells from the intestinal lamina propria and CD11c^lo MHC-II⁺ CD11b⁺ cells from the serosa, though previously called DCs in many studies, expressed macrophage core genes including those from module 161, suggesting a strong relationship to macrophages (Fig. 4a and 4b). Accordingly, these cells are now called CD11b⁺ gut macrophages and CD11c^lo serosal macrophages herein and on the Immgen website. We clustered these mononuclear phagocytes based on their expression of the 39-gene macrophage core to model their relatedness to each other (Fig. 4c). Langerhans cells of the skin and bone marrow macrophages were positioned at the interface between DCs and macrophages, with a distal relationship to classical DCs but failing to cluster with macrophages (Fig. 4c).

As mentioned earlier, non-lymphoid tissue CD11b⁺ DCs have been argued to be heterogeneous¹⁹. Thus, we reasoned that the use of antibodies to cell surface proteins identified as macrophage-specific from our gene expression analysis may discriminate macrophage “contaminants” in a heterogeneous population. Furthermore, we aimed to determine if the same cell surface markers may also prove valuable in identifying macrophages universally, including in organs beyond those we initially analyzed and/or where F4/80 has not proved sufficiently definitive. We homed in CD14, FcγRI, and MerTK as cell surface proteins in the group of 39 mRNA transcripts deemed to be low or absent in DCs but present in all macrophages, and to which quality mAbs have been generated. Indeed, all of these proteins were expressed on all of the 4 resident macrophage populations used in our primary analysis (Fig. 5a), with lower CD14 levels compared with FcγRI and MerTK (Fig. 5a). Two of these tissues, spleen and lung, have significant DC populations. In the spleen, MerTK, FcγRI, and CD14 did not stain CD8⁺ or CD11b⁺ DCs (Fig. 5a). However, in the lung where interstitial pulmonary macrophages are CD11b⁻, there may still be an underlying heterogeneity in lung CD11b⁺ DCs that includes a subset of CD11b⁺ macrophages^19,24{Satpathy, 2012 #3482}. Indeed, CD14, FCγR1 and MerTK were expressed by a portion of lung CD11b⁺ DCs, but not by CD103⁺ DCs (Fig. 5a). Gating on MerTK⁺ FCγR1⁺ cells revealed the vast majority of such cells were SiglecF⁺ lung macrophages, but a small proportion of MerTK⁺FCγR1⁺ cells in the lung were SiglecF⁻ cells expressing high levels of MHC II (Fig. 5b). Gating on DCs (Fig. 5c) revealed that CD11b⁺ DCs could be divided into CD11b⁺ CD24⁺ FCγR1^lo MerTK⁻ CD14^int and CD11b⁺ CD24^lo FCγR1⁺ MerTK⁺ CD14^hi cells (Fig. 5d). Thus, the latter likely comprises a population of macrophages that cosegregates with DCs using many markers, but are not DCs. Indeed, the CD11b⁺ DCs were segregated within Immgen on the basis of CD24 expression based on the likelihood that those expressing CD24 were true DCs, but those without CD24 were not. Our findings suggest that this possibility is highly likely and points to the utility of using markers like MerTK and FCγR1 as a panel to facilitate the identification of macrophages from DCs.

We next turned to two tissues—liver and adipose tissue--that were not analyzed in Immgen with respect to gene expression profiling in macrophages to determine if the use of MerTK and FcγRI staining would facilitate the identification of macrophages in those organs and distinguish them from DCs. In liver, we started with a classical approach of plotting F4/80 versus CD11c. Eosinophils are now recognized as high side-scatter, F4/80⁺ cells that express Siglec F universally²⁵. Indeed, among macrophages, Siglec F is observed only on macrophages in the lung^26,27 (as used to identify lung macrophages here). In the liver, the level of F4/80 on eosinophils overlaid with that of another population of low side-scatter, F4/80⁺ cells that were CD11c^lo in liver (Fig. 5e, left plot), so that even after excluding Siglec F⁺, high side-scatter eosinophils, four gates of cells expressing varying levels of F4/80 and CD11c were found (Fig. 5e). MerTK and FcγRI was highly expressed in two of these gates, suggesting that the cells with the highest level of F4/80 (gate 2) and many that expressed lower F4/80 (in gate 3) were two populations of F4/80^hi and F4/80^lo liver macrophages, corresponding to the two types of macrophages believed to be present in many organs¹⁶. The liver CD45⁺ cells with highest CD11c were MerTK⁻FcγRI⁻, suggesting they were liver DCs (Fig. 5e). Reverse gating revealed that all MerTK⁺FcγRI⁺ cells fell into one of the two putative macrophage gates (Fig. 5e). A relatively similar picture was seen in adipose tissue (Fig. 5f), where the cells with highest F4/80 were MerTK⁺FcγRI⁺ and those with higher CD11c and lower F4/80 were MerTK⁻FcγRI⁻. In both liver and adipose tissue, MHC II was high on macrophages and DCs (Fig. 5e, f). Because F4/80 and CD11c are both expressed by many tissue macrophages and DCs, albeit at levels that are somewhat different, distinguishing macrophages and DCs based on these traditional markers can be difficult. MerTK and FcγRI staining offers the advantage of sharp differences in the magnitude of expression between macrophages and DCs. Thus, we propose that MerTK and FcγRI costaining provides a powerful approach to identifying macrophages universally and selectively in mouse tissues.

Mice

Six-week-old male C57BL/6J mice purchased from Jackson Laboratory were used for sorting and validation. CX3CR1-GFP knock in mice were from Jackson Laboratories, and Mr1 knockout mice²⁰ were generated, bred, maintained at the Washington University School of Medicine. Mice were housed in specific pathogen–free facilities at the Mount Sinai School of Medicine or Washington University School of Medicine and experimental procedures were performed in accordance with the animal use oversight committees at these respective institutions. Most of the populations in the study were sorted from resting mice. However, for thioglycollate-elicited peritoneal macrophages, macrophages were harvested from the peritoneal cavity 5 days after instilling 1 ml of 3% thioglycollate.

Cell identification and isolation

All cells were purified using the sorting protocol and antibodies listed on http://www.immgen.org. Cells were directly sorted from mouse tissues and were processed from tissue procurement to a second round of sorting into Trizol within 4 h using a Beckton-Dickinson Aria II instrument. Resting red pulp macrophages from the spleen were sorted after nonenzymatic disaggregation of the spleen and were identified as F4/80^hi cells that lacked B220 and high levels of CD11c and MHC II^34,35; macrophages from the resting peritoneum were collected in a peritoneal lavage and stained to identify CD115^hi cells that were F4/80^hi MHC II⁻; resting pulmonary macrophages were isolated from Liberase III-digested lungs (15 min. digest) and macrophages were identified as SiglecF⁺ CD11c⁺ cells with low levels of MHC II^26,27; and resting brain microglial macrophages were sorted from Liberase III-digested, Percoll-gradient separated cells that were CD11b⁺ CD45^lo F4/80^lo11. The Data Browser in the Immgen website is a resource for pdf files showing FACS dot plots that depict the purification strategies and purity after isolation of these and all other populations. A list of abbreviations used in the Immgen database relevant to macrophages and DCs can be found in Supplementary Note 1.

Microarray analysis, normalization, and dataset analysis

RNA was amplified and hybridized on the Affymetrix Mouse Gene 1.0 ST array by the Immgen consortium using double-sorted cell populations sorted directly into TRIzol. These procedures followed a highly standardized protocol for data generation and QC documentation (pdf documents found under “Protocols,” available on http://www.immgen.org; Supplementary Notes 2–4). A table listing QC data, replicate information, and batch information for each sample is also available on the website. All datasets have been deposited at National Center for Biotechnology Information/Gene Expression Omnibus under accession number GSE15907. Data analysis utilized GenePattern analysis software. Raw data were normalized using the robust multi-array algorithm, returning linear values between 10 to 20,000. A common threshold for positive expression at 95% confidence across the dataset was determined to be 120 (Supplementary Note 4). Differential gene expression signatures were identified and visualized using the “Multiplot” module of GenePattern (http://www.broadinstitute.org/cancer/software/genepattern/). Differentially expressed probesets were considered as those with a coefficient of variation less than 0.5 and a p value ≥ 0.05 (Student’s T-test). Signature transcripts were clustered (mean centered) using the “Hierarchical Clustering” module of GenePattern, employing Pearson’s correlation as a metric, and data were visualized using the “Hierarchical Clustering Viewer” heat map module. Clustering analyses performed across ImmGen centered on the most variable gene sets (objectively defined as the top 15% genes ranked by cross-population max/min ratio), to avoid noise from genes at background variation. Pearson correlation plots of gene expression profiles between different cell populations were generated using Express Matrix software. Pathway analysis as well as transcription factor network construction were performed using Ingenuity Systems Pathway Analysis (IPA) software. This software calculates a significance score for each network. The score is generated using a p-value indicative of the likelihood that the assembly of a set of focus genes in a network could be explained by random chance alone.

Principle component analysis (PCA) analysis

Only the 4417 genes whose variance of expression across all samples from the ten cell types was at least within the 80^th percentile of variance were considered for the PCA analysis. The RMA normalized and log2 transformed expression levels were used. PCA was conducted using MATLAB.

Generation of the core macrophage signature

A macrophage core signature was generated by comparing brain, lung, peritoneal, and red pulp macrophages to populations deemed not to be macrophages, but authentic DCs. These DCs included CD103⁺ DCs from lung and liver, CD8⁺ DCs from spleen and thymus, CD4⁺CD11b⁺ DCs from spleen, CD4⁻CD8⁻CD11b⁺ DCs from spleen, and all DC populations (resident and migratory) isolated from skin-draining lymph nodes. A first list of possible genes defining macrophages was generated using the median value in the group of macrophages or DCs for each probe set in order to generate a list of probe sets with a differential median expression that was ≥ 2-fold higher in the group of macrophages with a statistical significance of p < 0.05 (Student’s T-test) and a coefficient of variation less than 0.5. Then this list of probe sets was filtered to remove any probe sets that did not show ≥ 120 (the threshold for positive expression) normalized intensity value in at least 2 macrophage populations. From the remaining probe sets, we compared the mean expression values of each macrophage and DC population, filtering to identify the lowest mean value in any single macrophage population to the highest mean value in any of the DCs. The probe sets that were at least 2.0-fold higher in expression in the lowest expressing macrophage compared with the highest expressing DC comprised the core genes were retained (Table 1, left column). To account for genes observed in only some macrophages, but still not expressed in DCs, we also generated lists wherein one or two macrophages were allowed to be excluded from consideration, but the criteria for comparing the remaining macrophages to the DCs was otherwise carried forward as described.

Generation of gene modules and prediction of module regulators

The gene modules, Ontogenet algorithm and regulatory programs are described in (Jojic et al, in preparation; Method found in Supplementary Note 5). Briefly, the expression data normalization was done as part of the ImmGen pipeline, March 2011 release. Data was log2 transformed. For gene symbols represented on the array with more than one probeset, only the probeset with the highest mean expression was retained. Of those, only the 7996 probesets displaying a standard deviation higher than 0.5 across the entire dataset were used for the clustering. Clustering was performed by Super Paramagnetic Clustering³⁶ with default parameters, resulting in 80 stable coarse modules of co-expressed genes. Each coarse module was further clustered by hierarchical clustering into more fine modules, resulting in 334 fine modules.

A novel algorithm termed Ontogenet was developed for the ImmGen dataset (Jojic et al, in preparation, Supplementary Note 5). Ontogenet finds a regulatory program for each coarse and fine module, based on regulators expression and the structure of the lineage tree. The regulatory program uses a form of regularized linear regression, in which each cell type can have its own regulatory program, but regulatory programs of related cells are encouraged to be similar. This allows switching in the regulatory program but still allows robust fitting given the available data. To visualize the expression of a module on the lineage tree, the expression of each gene was standardized by subtraction of the mean and division by its standard deviation across all dataset. Replicates were averaged. Mean expression of each module was projected on the tree. Expression values are color coded from minimal (blue) to maximal (red).

Association between the macrophage core signature and gene modules

Hypergeometric test for two groups was used to estimate the enrichment of all ImmGen fine modules for the 11 gene signatures listed in Tables 1 and 2. Benjamini Hochberg FDR <= 0.05 was applied to the p-value table of all 11 signatures across all 334 fine modules.

Antibodies used for validation studies

Anti-mouse CD11c (N418), CD11b (M1/70), CD24a (30-F1), CD45 (30-F11), CD14 (Sa2-8) MHC-II (M5/114.15.2), F4/80 (BM8), CD8a (53-6.7), CD103 (2E7), CD11a (M17/4), EPCAM (G8.8), VCAM1 (429), CD31 (390), CD93 (AA4.1), ICAM2 (3C4 mIC2/4), CD68 (FA-11), and isotype control mAbs were from Ebioscience or Biolegend. Anti-mouse MERTK (BAF591) was from R&D Systems. Anti-mouse FCGR1 (X54-5/7.1) and SiglecF (E50-2440) were from BD Bioscience. Anti-mouse Mr1 antibody was previously described²⁰. Anti-SiglecH antibody was a king gift from Marco Colonna (Washington University School of Medicine).