Three tissue resident macrophage subsets coexist across organs with conserved origins and life cycles

The analysis of tissue macrophages has been focused on defining differences both between and within organ-specific subpopulations. However, what has been largely overlooked is the possibility of a common macrophage structure that is shared among tissue macrophage subpopulations at the steady state amid this known heterogeneity. We performed scRNA-seq to transcriptionally profile murine macrophages across five organs. We sorted macrophages from the adult mouse heart, liver, lung, kidney, and brain using a broad macrophage gating strategy (CD45⁺CD64^int-hi cells; Fig. 1A and fig. S1). A total of 15,567 single-cell transcriptomes were subjected to Louvain clustering, resulting in 17 transcriptionally distinct populations in our combined analysis of all five organs (fig. S2A). Six were macrophage based on known markers (Adgre1, C1qa, C1qb, C1qc, Fcgr1, and Cd14), whereas the remaining 11 were identified as monocytes, dendritic cells, lymphocytes, and numerous proliferating populations (fig. S2A and table S1). Several macrophage clusters contained cells from a single tissue, such as lung alveolar macrophages (expressing Siglecf, Plet1, and Net1), brain microglia (expressing Tmem119, Siglech, and P2ry12), and two entirely liver-derived macrophage clusters, including Kupffer cells (expressing Clec4f and Vsig4). Two macrophage clusters (MF-A and MF-B) contained cells from the heart, lung, kidney, and brain, suggesting some level of transcriptional similarity (fig. S2A).

To investigate the macrophage substructure within each organ and their transcriptional relationship across organs, we subclustered macrophages from each organ separately and excluded monocytes, dendritic cells, lymphocytes, and proliferating cells (Fig. 1B and fig. S2B). Dimensionality reduction and unbiased graph-based clustering delineated three macrophage subpopulations present in each tissue, with varying relative frequencies (Fig. 1, B and C, and fig. S2C). In the lung and brain, this macrophage substructure was present in addition to the numerically dominant alveolar macrophage (Siglecf-expressing) and microglia (Tmem119-expressing) populations, respectively (Fig. 1, B to D). The first subpopulation in each organ highly expressed Timd4, Lyve1, and Folr2 (similar to MF-A in our combined analysis) in addition to several other recurring markers (Mrc1, Igf1, F13a1, Cd163, and Ninj1; Fig. 1D, fig. S2C, and table S1). We termed this subset TLF⁺ macrophages to denote the expression of any combination of Timd4, Lyve1, and/or Folr2. The second subpopulation in each organ lacked expression of these genes and distinctly expressed Ccr2 and Cd52, in addition to high levels of major histocompatibility complex II (MHC-II) genes (H2-Eb1, H2-Aa, and H2-Ab1) and S100 genes (S100a4, S100a6, and S100a8), termed CCR2⁺ macrophages. The third macrophage subpopulation lacked Timd4, Lyve1, Folr2, and Ccr2; had high expression of MHC-II genes; and coexpressed Trem2, Apoe, and Cd14 in an organ-specific manner. Although this subset expressed an average of 88 differentially expressed genes (DEGs) (range 27 to 236) in each organ (fig. S2C), unlike the first two subpopulations, it lacked consistent markers across all five organs. Thus, this subset was most accurately defined by the lack of TLF subset genes and Ccr2 gene expression and high expression of antigen presentation genes. For simplicity and to maintain consistency with prior studies, we termed this subpopulation MHC-II^hi macrophages.

Sexual dimorphism in immune cell transcriptomes has been reported (28–30). To investigate sex-specific differences in macrophage substructure, we retrospectively assigned cells as male versus female using Y-chromosomal genes and the X-linked inactivation gene (Xist), as previously done (fig. S3A) (31). Our macrophage subpopulations were representative of both males and females in each organ, with a similar relative frequency (fig. S3B). TLF⁺ macrophages expressed the most DEGs between males and females (>200 DEGs; fig. S3C). In males, this subset was enriched in phagocytic and endocytic pathways, whereas in females, they were enriched in response to wounding, angiogenesis, and hemopoiesis (fig. S3D). MHC-II^hi and CCR2⁺ macrophages had fewer DEGs across sex, and some of these were shared with the TLF⁺ subset. These data indicate that the overall macrophage substructure is sex-independent while highlighting putative sexual dimorphic pathways in macrophages (fig. S3D).

We next interrogated the level of transcriptional similarity between analogous subpopulations across tissues. Principal components analysis of the average transcriptional signature in each subset revealed that organ-specific differences were the major driver of gene expression patterns in macrophage subpopulations (Fig. 1E), which was corroborated with hierarchical clustering (fig. S3E). To explore the degree to which analogous subpopulations across tissues were transcriptionally related, we first defined DEGs for each subset within each organ separately (e.g., heart TLF⁺ versus heart MHC-II^hi versus heart CCR2⁺), then examined the extent to which these subset-defining DEGs were shared across organs (e.g., the overlap between DEGs in TLF⁺ macrophages of the heart, liver, lung, kidney, and brain). While the majority (~75%) of DEGs remained unique to only one organ (Fig. 1F and table S2), we identified a conserved core gene signature for each macrophage subpopulation (Fig. 1G). TLF⁺ macrophages exhibited the greatest level of conservation across tissues, as 11% of DEGs were shared by at least three organs, followed by CCR2⁺ macrophages (9.3%), whereas MHC-II^hi shared fewer genes (4.8%) (Fig. 1, F and G, and table S2). We assessed the efficacy of the combinatorial use of four of the top conserved cell surface marker genes (Timd4, Lyve1, Folr2, and Ccr2) in predicting these three macrophage subpopulations using random forest modeling. At the transcriptional level, the combination of these four genes sufficiently predicted our macrophage classification with high accuracy (TLF⁺: 93%, MHC-II^hi: 88%, and CCR2⁺: 89%; Fig. 1H and table S3), and this was consistent in both male and female parsed datasets (table S3). Pathway analysis revealed that TLF⁺ macrophages were enriched in cellular transport and endocytosis pathways in all organs, whereas CCR2⁺ macrophages were enriched in cellular activation, degranulation, and immune effector processes (fig. S3F), suggesting that functional similarities might also exist. MHC-II^hi macrophages were more diverse in their functions across organs (table S4). In addition, macrophage subpopulations were enriched in pathways unique to each organ (table S4). For example, heart TLF⁺ macrophages mapped to lipoxin metabolism, whereas kidney macrophages were enriched in IGF1-R signaling (TLF⁺) and IGF transport (MHC-II^hi). CCR2⁺ macrophages mapped to hematopoietic progenitor differentiation in the liver and cytoskeleton organization in the heart. Perhaps the most tissue-specific functions were that of brain MHC-II^hi macrophages, which had pathways related to cerebral and forebrain cell migration, as well as gliogenesis (table S4). These data highlight that while organ specificity accounts for most transcriptional differences, small core gene expression patterns define macrophage subpopulations across tissues.

It has recently been described that resident interstitial macrophages are organized into a dichotomy between LYVE1 and MHC-II expressing subsets (17). LYVE1^hi macrophages defined by Chakarov et al. (17) had similar gene expression patterns to TLF⁺ macrophages observed in our analysis. However, our data suggest that the MHC-II^hi subpopulation consists of two subpopulations (MHC-II^hiCCR2⁺ and MHC-II^hiCCR2⁻; fig. S3G), indicating additional heterogeneity present in each organ. To assess the reproducibility of our macrophage substructure and its core gene profile, we probed publicly available scRNA-seq datasets in the mouse heart, liver, lung, kidney, and brain (fig. S4) (16, 18, 32–34). We focused our analyses on tissue macrophages (C1qc⁺C1qa⁺Plac8⁻Ace⁻Ly6c2⁻) and found three analogous subsets in each organ, with a reciprocal expression of Timd4, Lyve1, and Folr2 versus Ccr2 (fig. S4A). To formally assess transcriptional similarity with our data, we scored each macrophage in replicate datasets based on our core gene signatures (Fig. 1, F and G). Macrophage subsets were most enriched for the corresponding subset signature relative to other subsets, highlighting conservation of a core gene profile (fig. S4B). Random forest modeling demonstrated that the expression of Timd4, Lyve1, Folr2, and Ccr2 was sufficient to predict this macrophage classification with high accuracy in these independently generated datasets (fig. S4C). Overall, our scRNA-seq data reveal a refined set of macrophage subset genes that reproducibly track with three subpopulations in five organs.

We explored whether the observed macrophage subpopulations retained any conserved biological properties beyond the core transcriptional signature. First, we focused on the relationship between circulating blood monocytes and tissue macrophage replacement over time. To validate the ability to track these subpopulations, we performed flow cytometry on single-cell suspensions from heart, liver, lung, kidney, and brain using the key subset defining genes and general macrophage markers defined by scRNA-seq. We visualized the flow cytometric data of total macrophages (CD45⁺CD64⁺; fig. S5A) in each organ using t-distributed stochastic neighbor embedding (t-SNE) computed based on 10 parameters {8 surface antigens, 2 physical characteristics: size [forward scatter area (FSC-A)] and granularity side scatter area (SSC-A)}. We observed three recurring macrophage clusters across all organs (Fig. 2A) with the addition of a fourth numerically dominant cluster in the lung and brain, similar to our scRNA-seq data (Fig. 1B). At the protein level, cluster 1 expressed TIMD4, LYVE1, and FOLR2 (TLF⁺); had low levels of MHC-II; and lacked CCR2 expression (Fig. 2, A and B). Both cluster 2 and cluster 3 macrophages expressed high levels of MHC-II and lacked TIMD4, LYVE1, and FOLR2 but could be distinguished based on CCR2 expression. In the lung, the fourth cluster represented CD11b^lo alveolar macrophages (35), and in the brain, it represented CD45^lo microglia (Fig. 2B) (36). Thus, for future analysis of cells in clusters 1 to 3, we pregated on CD11b^hi lung interstitial macrophages and CD45^hi BAMs in the brain (fig. S5A, red dashed line). Stratifying tissue macrophages by TIMD4 and CCR2 alone (Fig. 2C) was sufficient to identify these three macrophage subpopulations, TLF⁺CCR2⁻ (TLF⁺ macrophages), TLF⁻CCR2⁺ (CCR2⁺ macrophages), and TLF⁻CCR2⁻ (MHC-II^hi macrophages), with frequencies that reflected the scRNA-seq data.

Next, we surgically joined wild-type mice (CD45.1 and CD45.2) to assess the ability of circulating chimeric monocytes to replace tissue macrophages in joined parabionts over 6 weeks. Blood monocyte chimerism was similar to prior studies (total monocytes ~25%, Ly6C^hi monocytes ~10%; Fig. 2D) (25, 37). Total macrophages in the heart, lung interstitium, and brain BAMs achieved ~2% chimerism (~20% when normalized to Ly6C^hi monocytes) and ~4% (~40% normalized) in the kidney, whereas liver and lung alveolar macrophages, as well as brain microglia, received negligible monocyte input. After stratifying into the three macrophage subpopulations, we observed that CCR2⁺ macrophages had the highest level of replacement and TLF⁺ macrophages had low to negligible replacement, whereas, in general, MHC-II^hi macrophages had an intermediate level of replacement (Fig. 2E). These data suggested that CCR2⁺ macrophages in each organ had a life cycle almost exclusively driven by blood monocyte replacement, whereas MHC-II^hi and particularly TLF⁺ macrophages were considerably less reliant on monocytes.

Previous studies have suggested that bone marrow–derived monocytes continually replace some interstitial tissue resident macrophage populations over time, yet the extent to which this occurs differs among different tissues. For example, parabiotic and fate mapping studies using the S100a4^Cre of lung interstitial macrophages (LYVE1^hi and MHC-II^hi) suggested equal replacement of both subsets by blood monocytes (17). These results differed from our observation that the lung interstitium and other organs contain three macrophage subsets, each with a distinct life cycle. Similarly, this study and others using the Cx3cr1^CreERT2 have also suggested that cardiac macrophages are replaced by blood monocytes (17, 38). These data may be explained by use of different gating strategies or observation of single time points in parabiotic studies, where the trajectory of the monocyte contribution over time could not be assessed.

While monocytes did contribute to MHC-II^hi and, to a lesser extent, to TLF⁺ macrophages (Fig. 2E), we explored whether this contribution progressed over time or reached a saturation point. We performed sequential long-term parabiotic studies by joining monocyte-deficient CD45.2 Ccr2^−/− mice to wild-type CD45.1 mice for 6 or 10 months [as done previously (17)] and assessed the contribution of CD45.1 donor cells to each macrophage subset in the CD45.2 Ccr2^−/− recipient (fig. S5B). Here, Ccr2^−/− mice were used as recipients to increase chimerism, and to focus on turnover of the CCR2 nonexpressing subsets, MHC-II^hi and TLF⁺ macrophages. As expected, blood monocyte chimerism in the CD45.2 Ccr2^−/− recipient was high (~85 to 90%; Fig. 2F). Gating on total macrophages in each organ demonstrated about 5 to 20% replacement at 6 months after pairing. We observed no difference in the percentage of CD45.1 donor cells accumulated in the recipient between 6 and 10 months, suggesting that recruitment occurred to a defined maximum (Fig. 2F). By subset, TLF⁺ macrophages in each organ had minimal monocyte contribution at either 6 or 10 months, similar to lung alveolar macrophages and brain microglia (Fig. 2G). Thus, the limited peripheral monocyte contribution into the TLF⁺ population was stable at both time points. MHC-II^hi macrophages had higher chimerism in the heart, liver, lung, and brain (~20 to 40% chimerism at 6 months). However, similar to TLF⁺ macrophages, chimerism of MHC-II^hi macrophages remained stable between 6 and 10 months, except within the liver, where MHC-II^hi macrophage chimerism progressively increased. Our data agree with a prior study that observed that MHC-II–expressing capsular liver macrophages were replaced by blood monocytes (21). As expected, all CCR2⁺ macrophages present in the CD45.2 Ccr2^−/− host were of donor origin (fig. S5C). We confirmed that replacing TIMD4 (as the TLF subset marker) with either LYVE1 or FOLR2 identified a very similar, monocyte-independent macrophage subset in each organ (fig. S5D). Thus, the use of TIMD4 and CCR2 to stratify organ macrophage subpopulations reveals previously unrecognized differences in their life cycles. TLF⁺ macrophages represent a distinct subset of macrophages that persists through self-renewal with minimal monocyte input in adult animals. MHC-II^hi macrophages, while receiving modest monocyte contribution, are not continually replaced—rather, monocyte-derived macrophages only contribute until they reach a defined upper limit after which they do not outcompete the pre-existing MHC-II^hi resident macrophages.

To determine whether these three macrophage subpopulations were broadly conserved beyond the organs we initially investigated, we expanded our study to 12 additional tissues, including muscle tissues (hindlimb, diaphragm, tongue, and uterus), adipose tissues (brown adipose tissue, epididymal white adipose tissue, inguinal white adipose tissue), digestive tissues (esophagus, pancreas, small intestine, and colon), and skin. Each contained CD64⁺CD11b⁺ macrophages (fig. S6A) that could be further stratified into three subpopulations using TIMD4 and CCR2 by flow cytometric analysis (fig. S6B).

Next, we used the Cx3cr1^CreERT2 inducible fate mapping mouse model crossed to a Rosa26^Td reporter (termed Cx3cr1^CreER:R26^Td) as a complementary tool to broadly assess life cycle properties. Mice were fed tamoxifen chow at P21 (21 days) for 10 days to label all monocyte and macrophage populations with the Td reporter and then were assessed immediately after cessation of tamoxifen chow (initial; P31) or after 4 weeks (P61; chase) (Fig. 3A). Microglia were entirely composed of Td⁺ cells at both time points, defining the labeling efficiency. In contrast, blood monocytes were highly labeled initially (total monocytes ~80%, Ly6C^hi monocytes ~70%) but lost the label entirely after 4 weeks. Alveolar and Kupffer cells achieved low labeling at both time points as expected (Fig. 3B). For the three macrophage subsets in each tissue or organ, we computed a ratio of %Td⁺ cells at 4 weeks to %Td⁺ cells at the initial time point, which we termed %Td retention. Thus, a %Td retention of more than 100% indicates an increase in the proportion of Td⁺ cells after 4 weeks relative to the initial label. Conversely, a %Td retention of less than 100% indicates a decrease of Td⁺ cells over that same time. Across all tissues and organs examined, TLF⁺ macrophages generally had a %Td retention of 100% or greater (Fig. 3C). CCR2⁺ macrophages generally showed less than 50% Td retention across all tissues analyzed while MHC-II^hi macrophages showed intermediate turnover. Cross-comparison of the three macrophage subsets within each tissue (Fig. 3D) or aggregated across all tissues (Fig. 3E) indicated that TLF⁺ macrophages have a greater capacity for self-renewal, whereas MHC-II^hi, and, to a greater degree, CCR2⁺ macrophages have increased turnover. This could be attributed to internal competition between Td⁺ and Td⁻ macrophages or from peripheral recruitment of Td⁻ blood monocytes.

To specifically assess monocyte dependency, we joined wild-type (CD45.1 and CD45.2) parabionts for 6 weeks as before (Fig. 2, D and E). A similar pattern of monocyte replacement was generally present across all tissues, in which CCR2⁺ macrophages were highly replaced, TLF⁺ macrophages were minimally replaced, and MHC-II^hi macrophages had intermediate levels of replacement (fig. S7, A and B). Cross-comparison of the three macrophage subsets in each tissue or organ (fig. S7C) or aggregated (fig. S7D) indicated a subset-specific monocyte dependence complementary to the life cycles that we had observed using the Cx3cr1^CreER system (Fig. 3, D and E). CCR2⁺ macrophages, regardless of tissue of residence, were primarily of donor origin after 6 weeks of parabiosis, whereas host MHC-II^hi and TLF⁺ macrophages were maintained with less reliance on peripheral recruitment. TLF⁺ macrophages in particular received less monocyte input, suggesting that they are mainly composed of self-renewing macrophages in all tissues. The Cx3cr1-based system labeled monocytes highly. Although it is plausible that these Td⁺ monocytes explain the above 100% Td⁺ retention observed in TLF⁺ macrophages, parabiosis data demonstrate that this is likely not the case. Thus, the combinatorial use of TIMD4 and CCR2 alone was sufficient to identify tissue macrophage subsets with three distinct monocyte-dependence and life cycles across 17 murine tissues.

We then explored in adult animals whether newly recruited, monocyte-derived macrophages would mirror established resident subpopulations, in both relative frequencies and transcriptional character, and what effect prolonged tissue residence plays on both characteristics. To address these questions, we performed short-term (5 weeks) and long-term (25 weeks) parabiotic studies by joining wild-type CD45.1 to CD45.2 mice. We separately sorted macrophages (CD45⁺CD64^int-hiCD11b^hi) composed of either host- or donor-derived cells from the heart and lung and performed scRNA-seq (fig. S8A). This system’s lack of prior depletion represents an unperturbed look at the single-cell transcriptional fate of recruited cells. We analyzed a total of 5893 cells in the heart and 1371 cells in the lung, of which donor macrophages represented a small fraction. The combined analysis of host and donor macrophages at 5 and 25 weeks revealed three subpopulations in the heart and lung (fig. S8B). The donor macrophage populations were similar to our initial analyses and had reciprocal expression of Timd4, Lyve1, and Folr2 versus Ccr2. The relative frequency of chimeric, donor monocyte–derived macrophages (ratio of TLF⁺ versus MHC-II^hi versus CCR2⁺) mirrored that of host macrophages present in the tissue at each time point (fig. S8D), suggesting that the tissue instructs the subset compositional fate of recruited monocytes during homeostasis.

Monocytes entering the heart and lung exhibited distinct temporal differentiation patterns. At 5 weeks after pairing, donor (chimeric) macrophage subsets in the heart were transcriptionally different from their corresponding host macrophage subsets with a range of 200 to 400 DEGs in each subset (fig. S8E and table S5). By 25 weeks, the transcriptional profile of donor monocyte–derived macrophages was nearly identical to the corresponding host macrophage subset. In contrast, monocytes entering the lung at 5 weeks after pairing that became TLF⁺ or MHC-II^hi macrophages were transcriptionally indistinguishable from their corresponding host macrophage subset (fig. S8E and table S5). Monocytes that became lung CCR2⁺ macrophages behaved similar to the heart, acquiring the host signature with time, albeit the difference between chimeric macrophages at 5 weeks was more modest. The smaller magnitude difference suggests a faster adoption of the resident signature in the lung and slower maturation of cardiac macrophages. The heart had the greatest number of DEGs between host and donor macrophages with Ccr2 and antigen presentation genes (H2-Eb1, H2-Ab1, and H2-Aa) generally expressed at higher levels in donor macrophages, whereas host macrophages expressed higher levels of core resident macrophage genes at 5 weeks, such as Folr2, Lyve1, Igf1, Ninj1, Cd163, F13a1, and Gas6 (table S5). Together, these data indicate the remarkable capacity of recruited monocytes to adopt each of the three macrophage cell fates and the ability to do so in a tissue-specific temporal fashion.

TLF⁺ macrophage maintenance is driven almost entirely through self-renewal in the adult—next, we sought to explore their origin. During development, tissue resident macrophages emerge from multiple sources, the earliest in development being primitive hematopoiesis within the yolk sac (39). We observed that the vast majority (85%) of macrophages isolated from the mouse yolk sac at an early time point (E14.5) coexpressed TIMD4, LYVE1, and FOLR2 and lacked CCR2 and MHC-II (Fig. 4A). Therefore, we use the name TLF⁻CCR2⁻ macrophages in development and MHC-II^hi (TLF⁻CCR2⁻) at time points after birth where MHC-II is up-regulated. Consistent with an early embryonic origin, TLF⁺ macrophages were present in embryonic tissue (E14.5) in the five main mouse organs analyzed as the dominant population, while the other two populations were present at lower frequencies (Fig. 4, B and C). Because of the immature nature of alveolar macrophages (and lack of their functional niche—the alveoli) (11), we could not differentiate lung alveolar from lung interstitial macrophages by flow cytometry using conventional markers in early prenatal tissue at E14.5. From E19.5 onward, however, interstitial macrophages could be separated from alveolar macrophages by high CD11b expression before gating on interstitial subpopulations (fig. S9A). Microglia and BAMs were distinguished as outlined (fig. S9B) with a large proportion of BAMs (predominately TIMD4⁺) noted during fetal development, as recently described (40).

After birth, the relative frequency of the three macrophage subsets within each organ changed over time in a tissue-dependent manner, with TLF⁺ macrophages remaining the dominant population in the heart and liver up to 1 year of age (Fig. 4C). The lung, brain, and, in particular, the kidney had a reduction in the proportion of TLF⁺ macrophages after birth, with the kidney demonstrating a nearly complete loss of TLF⁺ macrophages (Fig. 4C). The low frequency of kidney TLF⁺ macrophages in the adult is consistent with previous studies (41). Despite what we observed in the adult, all macrophage populations in embryonic tissue had very low expression of MHC-II (Fig. 4D). However, we observed a gradual up-regulation of MHC-II after birth in a tissue-dependent manner, first on CCR2⁺ macrophages, followed by TLF⁻CCR2⁻ (MHC-II^hi) macrophages, and lastly on TLF⁺ macrophages. In general, expression of MHC-II was relatively lower and more variable in TLF⁺ macrophages across tissues (Fig. 4D). These data highlight that MHC-II protein expression itself was not a reliable marker of subset identification, particularly during development, and support our scRNA-seq analysis that lack of TIMD4, LYVE1, FOLR2, and CCR2 within the CD64⁺ macrophage gate is most accurate to identify this third subset.

To determine a developmental relationship between TLF⁺ macrophages in the yolk sac and those in the tissue at E19.5, we used the Cx3cr1^CreER:R26^Td fate mapping mouse model (42) (Fig. 5A). Pregnant Cx3cr1^CreER:R26^Td mice were administered tamoxifen by oral gavage at E8.5 to label yolk sac–derived progenitor cells with the Td reporter (1); tissue was harvested at E19.5, just before birth. We observed ~40% labeling of microglia (fig. S10A), a subset entirely yolk sac–derived (13), defining the labeling efficiency. By gating on Td⁺ macrophages and then stratifying them based on TIMD4 and CCR2 expression, we observed the majority of E8.5 yolk sac–derived cells preferentially differentiated into TLF⁺ macrophages across organs (heart, 73.8 ± 4.5%; liver, 98.8 ± 0.4%; lung int., 64.0 ± 4%; kidney, 75.5 ± 2.9%; BAMs, 46.8 ± 4.9%), whereas a minority fell within the TLF⁻CCR2⁻ gate (analogous to MHC-II^hi macrophages) (Fig. 5B). We then gated on the different macrophage subsets based on TIMD4 and CCR2 expression and calculated the proportion composed of E8.5-derived (Td⁺) cells. We observed that brain BAMs (both TLF⁺ and TLF⁻CCR2⁻) had a similar contribution to that of microglia, suggesting that their origin is entirely from the yolk sac (fig. S10A). Similar to prior studies (11), we observed that alveolar macrophages were not derived from yolk sac progenitors at E8.5 and were entirely Td⁻ (fig. S10B). In the heart and liver, E8.5-derived cells had a greater contribution to TLF⁺ macrophages (~10 to 30% of total macrophages, normalized to microglia) when compared with TLF⁻CCR2⁻ macrophages (~5 to 12%) (fig. S10C). Virtually, all CCR2⁺ macrophages were Td⁻ (fig. S10C), consistent with their fetal monocyte origin (11). Collectively, these data demonstrate that TLF⁺ macrophages exist in the mouse yolk sac, and macrophages that arise from yolk sac hematopoiesis in the mouse primarily give rise to TLF⁺ macrophages and, to a lesser extent, TLF⁻CCR2⁻ macrophages across organs.

Next, we sought to determine whether fetal monocytes could also contribute to TLF⁺ and TLF⁻CCR2⁻ macrophages during development. In our current and previous work (25), CCR2 expression tracked closely with monocyte origin. Here, we observed that, at E14.5, CD64⁺ macrophages in each organ could be partitioned into an apparent monocyte-derived population (CCR2^hiCD64^intCD11b^hi) and a more mature macrophage population (CCR2⁻CD64^hiCD11b^int; fig. S11A, top). The more mature macrophages were TLF⁺ and TLF⁻CCR2⁻, whereas the monocyte-derived macrophages were CCR2⁺ (fig. S11A, bottom). To assess the contribution of monocytes during development, we used the Ccr2^CreERT2 inducible fate mapping system crossed to an R26^TdT reporter (termed Ccr2^CreER:R26^Td; Fig. 5C). As a control, tamoxifen administration to these mice at P21 labeled >90% of blood monocytes, with rapid loss of labeling first in Ly6C^hi, and later Ly6C^lo monocytes (fig. S11B).

To label fetal monocytes, tamoxifen was administered by oral gavage to pregnant Ccr2^CreER:R26^Td female mice at E14.5. By E19.5, the majority of Td⁺ cells were CD11b⁺CD64⁺ macrophages (~80%; Fig. 5C). We assessed the fate of E14.5-labeled cells by gating on Td⁺ macrophages and then stratifying them based on TIMD4 and CCR2 expression. We found that E14.5-derived Ccr2-expressing cells primarily gave rise to TLF⁺ macrophages in the liver, kidney, and brain (BAMs), whereas in the heart and lung interstitial populations, they preferentially gave rise to TLF⁻CCR2⁻ macrophages (Fig. 5D). This represented a narrow window of engraftment, with the highest degree of labeling observed in alveolar macrophages (~15 to 20% labeled) consistent with their fetal origin (11), and a complete absence of Td signal in blood monocytes by E19.5 (fig. S11C, left). Few Td⁺CCR2⁺ macrophages were observed (Fig. 5D), indicating that the majority of fetal monocyte–derived macrophages from E14.5 had down-regulated CCR2 expression by E19.5, and contributed to the TLF⁺ and TLF⁻CCR2⁻ subsets. When gating first on individual macrophage subsets and assessing the proportion of Td⁺ cells in each population, we observed that E14.5-derived cells contributed most to the CCR2⁺ subset and least to the TLF⁺ macrophage subset in the heart and lung. The kidney, liver, and brain had similar contributions to each subset (fig. S11C, right).

Overall, these data suggest that across organs, yolk sac–derived macrophages contribute to TLF⁺ and TLF⁻CCR2⁻ tissue macrophage subsets but not the CCR2⁺ subset, while fetal monocytes contribute to all three macrophage populations and are the primary source of CCR2⁺ macrophages. In corroboration with previous findings (43), our data suggest that fetal monocytes contribute to tissue macrophage subpopulations and we define that contribution to the three subpopulations we have identified here. Moreover, fetal monocyte contribution is tissue-dependent and likely changes at different stages of development as tissue niches become available.

To examine whether early fetal monocyte–derived macrophages could gain long-term self-renewal capacity or would be replaced by monocytes after birth, we tracked Td⁺ cells in the Ccr2^CreER:R26^Td reporter mice after tamoxifen labeling at E14.5. We compared labeled cells at E19.5 versus P21 and observed a net numerical expansion of Td⁺ cells in the heart and liver (per milligram of tissue), while numbers of lung interstitial and alveolar macrophages, as well as BAMs in the brain, remained stable (Fig. 5, E and F, and fig. S11D, left). There was a loss of Td⁺ macrophages in the kidney, indicating that kidney macrophages had decreased from E14.5. When tamoxifen was administered to Ccr2^CreER:R26^Td mice early after birth (P7) to label neonatal Ccr2-expressing cells, we found an absolute loss, or trend toward loss, of labeled Td⁺ macrophages from 3 to 15 weeks generally across tissues (Fig. 5G and fig. S11D, right). This suggests that fetal monocytes have a greater capacity to differentiate into self-renewing tissue macrophages compared with neonatal monocytes.

We then examined the fate of E14.5-labeled Ccr2-expressing cells for TIMD4 or CCR2 expression, and we observed that the increase in total Td⁺ macrophages from E19.5 to P21 was driven primarily by increased TLF⁺ macrophages in the heart, liver, and lung (Fig. 5H), suggesting increasing competitiveness of E14.5-labeled fetal monocytes in comparison with other populations, which preferentially differentiate into TLF⁺ macrophages. TLF⁻CCR2⁻ macrophages behaved in a tissue-specific manner, with increased numbers seen in the heart, stability seen in the kidney and lung, and little labeling in the liver and brain. CCR2⁺ macrophages had negligible labeled cells across all organs, which together indicate at least two differentiation trajectories for fetal monocyte–derived macrophages—those that become short-lived CCR2⁺ macrophages, and those that down-regulate CCR2 expression and become self-renewing TLF⁺ or TLF⁻CCR2⁻ macrophages. While these data are consistent with previous reports indicating a dual origin of tissue resident macrophages (10, 44), we extend those observations and demonstrate that yolk sac–derived macrophages and early fetal monocytes have a greater propensity to become self-renewing TLF⁺ and MHC-II^hi macrophages compared with postnatal (P7) monocytes. Thus, TLF⁺ macrophages derived from either source can become a long-lived subset within tissues.

Next, we determined whether TLF⁺ macrophages are also present in early human development. Publicly available scRNA-seq data from human yolk sac (membrane and contents) at 4 post-conception weeks (PCW) (45) demonstrated that all macrophages (C1QC^hi) present at this stage expressed high levels of TIMD4, LYVE1, and FOLR2 (Fig. 6A), similar to murine yolk sac macrophages. Human yolk sac macrophages also highly expressed CD163, IGF1, F13A1, and SPP1, which were also expressed in murine tissue TLF⁺ macrophages (clusters 1 to 5; fig. S12A and table S6). In addition, we observed a small cluster that coexpressed macrophage (C1QC and HLA-DQA1) and monocytic genes (FCN1, S100A8; cluster 6), suggesting the emergence of early fetal monocytes or myeloblasts. On the basis of our observations that some murine TLF⁺ macrophages were yolk sac–derived, we assessed whether human yolk sac macrophages are transcriptionally related to murine macrophage subpopulations that we have defined (Fig. 1) using the machine-learning algorithm Garnett (46). First, we defined a human yolk sac macrophage gene signature (top 75 DEGs) and trained the “Garnett classifier” using this list. We tested this classifier against the human yolk sac single-cell dataset and observed a high classification accuracy (84.6% of cells classified as yolk sac macrophages; fig. S12B). We next performed a cross-species macrophage comparison to determine which murine adult macrophage subsets we defined by scRNA-seq (Fig. 1) were most similar to human yolk sac macrophages. In all organs assessed, murine TLF⁺ macrophages had the highest transcriptional similarity to human yolk sac macrophages (fig. S12C). In the mouse, brain TLF⁺ and MHC-II^hi macrophages demonstrated comparable similarity to human yolk sac macrophages (fig. S12C), consistent with our fate-mapping data that suggested these were both equally derived from E8.5. These data demonstrated that human yolk sac macrophages (TLF expressing) are transcriptionally similar to adult murine TLF⁺ macrophages.

Next, we explored whether human fetal TLF-expressing tissue macrophages develop similarly to murine TLF⁺ macrophages by investigating publicly available scRNA-seq data of the human fetal liver, kidney, and heart (45, 47, 48). At 7 PCW, a time point that corresponds to initial seeding of embryonic macrophages, we observed two transcriptionally distinct populations of macrophages expressing LYVE1, FOLR2, and TIMD4 in the human fetal liver and kidney (Fig. 6, B to G, and table S6). Trajectory analysis using Monocle 3 (49, 50) revealed that these two macrophage subsets occupied two separate transcriptional partitions, suggesting independent trajectories. This includes a differentiated “TLF⁺ macrophage” partition (high C1QC expression and absence of the monocyte gene FCN1), and a more heterogeneous “monocyte-derived TLF⁺ macrophage” partition (Fig. 6, B, C, E, and F). Closer assessment of the monocyte-derived TLF⁺ macrophage partition revealed a gradual transition from monocytes to macrophages (coordinated up-regulation of C1QC, LYVE1, TIMD4, and FOLR2, with reciprocal down-regulation of the monocyte genes FCN1, CCR2, and S100A8; Fig. 6, D and G). We pooled single-cell transcriptomes from early fetal heart data (5 to 9 PCW) (47), and despite lower cell numbers, we observed that all macrophages expressed LYVE1, FOLR2, and most expressed TIMD4 (fig. S13A). Later in fetal development (20 to 24 PCW) (47), cardiac macrophages appeared similar to the fetal liver and kidney at 7 PCW, where TLF⁺ macrophages occupied two separate partitions, including a “monocyte-derived macrophage” partition, suggesting the emergence of a second source of TLF⁺ in the heart.

To determine whether macrophages in human fetal tissue were transcriptionally related to yolk sac macrophages, we applied our Garnett human yolk sac classifier to human fetal tissue macrophages. Macrophages in the “TLF⁺ macrophage” partition were more transcriptionally similar to yolk sac macrophages in comparison with TLF⁺ macrophages from the monocyte-derived partition in the fetal liver and kidney (fig. S13B). These data indicate that TLF⁺ macrophages are the earliest state of macrophage development in humans within the yolk sac. As human organs develop, a second fetal monocyte–derived TLF⁺ macrophage population emerges, similar to our findings in the mouse, where we also demonstrated that both the yolk sac and fetal monocytes give rise to TLF⁺ macrophages.

Next, we examined whether the murine macrophage substructure extends to human tissue after birth and the degree of transcriptional relatedness. We performed scRNA-seq on human cardiac immune cells from a 12-year-old patient pooled with our published dataset of neonatal cardiac immune cells (Fig. 7, A and B, fig. S14A, and table S7) (51) and leveraged publicly available scRNA-seq data for the human liver (52), lung (53), and kidney (Fig. 7, C to E) (54). In datasets that were generated from several donors (heart, n = 2; liver, n = 5; and lung, n = 2), we applied batch correction to minimize nonbiological variation [Harmony (55); fig. S14B]. We excluded all nonmyeloid cells and dendritic cells from our analysis with the additional removal of the numerically dominant alveolar macrophages from the lung to focus on the interstitial subpopulations. Three transcriptionally distinct macrophage (C1QC^hi and CD14^hi) and several monocyte (FCN1^hi) subpopulations emerged in each organ (Fig. 7, A to E, and fig. S14C).

The first macrophage subset (MF1) in each human organ expressed LYVE1, FOLR2, and several recurring markers (MRC1, F13A1, and CD163), which comprised part of the core transcriptional signature of murine TLF⁺ macrophages (Figs. 1G and 7, A to E, and table S7). The remaining two subsets (MF2 and MF3) similarly shared individual genes that correlated to those observed in mice. MF2 lacked expression of LYVE1, FOLR2, and CCR2 and expressed TREM2, FABP5, and CD9 in an organ-dependent fashion, whereas MF3 expressed varying levels of CCR2 and high levels of antigen presentation genes (HLA-DQA1, HLA-DPA1, and CD74). While TIMD4 was easily detected in mouse studies across ages, its expression was highly restricted to TIMD4⁺LYVE1⁺ macrophages in early human life (Fig. 6), followed by a notable decrease in both human heart and liver through development into adulthood (fig. S14D). Furthermore, we confirmed that three macrophage subpopulations stratified by LYVE1 and CCR2 were clearly detectable by flow cytometry in the human lung and heart (fig. S14E).

To understand the transcriptional relationship between adult mouse and human macrophage subpopulations, we generated a murine macrophage signature for each subset in each mouse organ and scored individual macrophages in the corresponding human tissue for this signature. We observed that, in all human organs assessed, human MF1 was the most enriched for the murine TLF⁺ macrophage gene signature (Fig. 7F and fig. S15A). Human MF2 and MF3 were enriched for the mouse CCR2⁺ macrophage signature, whereas mouse MHC-II^hi macrophages did not consistently map to human macrophage subpopulations, highlighting the narrow focus of conservation between mice and humans. Human MF1 had the greatest conservation of DEGs, with ~34% (range, 14 to 55%) of genes overlapping with mouse TLF⁺ macrophages across all organs (fig. S15B and table S8), which was unexpectedly robust, given the known lack of transcriptional conservation between mouse and human macrophages (56). In contrast, conservation among MHC-II^hi and CCR2⁺ macrophages was much lower (averages: 8.3 and 17.4%, respectively; table S8). Testing using our Garnett classifier revealed that adult human MF1 was transcriptionally the most similar to human yolk sac macrophages in each organ, highlighting a potential origin of this subset (fig. S15C). Together, these data highlight multiple macrophage subpopulations in human organs and identify a human subset that transcriptionally resembles murine TLF⁺ macrophages, which also appears to be the earliest macrophage subset identified in the human yolk sac and developing organs.

This study assesses macrophage subpopulations across tissues in mice and human with a goal of establishing a common framework informed by biological commonalities. We performed scRNA-seq analysis of macrophages sorted from five organs to assess transcriptional heterogeneity in healthy adult mice. Bioinformatics analyses guided our characterization of macrophage subsets with the use of fate-mapping and parabiosis. For each experiment, n values are specified in corresponding figure legends and in table S9 and were predetermined for each experiment based on expected biological and technical variability. In murine experiments, males and females were used in equal proportions. To make comparisons between mouse and human data, we collected human heart and lung tissues according to research ethics board (REB) approval and used publicly available datasets. Information pertaining to scRNA-seq data and statistical analyses can be found within Supplementary Methods.

Wild-type C57BL/6J mice (Jackson Laboratory, #000664) and CCR2^GFP/+ mice (Jackson Laboratory, #027619) of various ages were used for this study. Rosa^Td reporter mice (Jackson Laboratory, #007914) were crossed with Cx3cr1^CreERT2 (Jackson Laboratory, #020940) or Ccr2^CreERT2 mice, a gift from B. Becher. Parabiosis experiments used CD45.2 Ccr2^+/+ mice (Jackson Laboratory, #000664), CD45.2 Ccr2^−/− mice (Jackson Laboratory, #004999), and CD45.1 Ccr2^+/+ mice (Jackson Laboratory, #002014). All mice were bred in our animal facility at the University Health Network. All protocols were approved by the Animal Resource Centre, University Health Network (AUP#4054), Toronto, Ontario, Canada.

After gating on CD45⁺ cells, doublets were excluded, and live cells were analyzed using FSC vs. SSC live/dead exclusion. Detailed methods including antibodies can be found in Supplementary Materials and Methods. Analysis software (FlowJo) was used to analyze flow cytometric data. Full-gating strategies of tissue macrophages are shown in fig. S1 and described in Supplementary Materials and Methods. For t-SNE analysis (Fig. 2), FlowJo was used to calculate the mean fluorescence intensity for each parameter. Heatmaps (Figs. 2 and 3 and fig. S7) were created using Morpheus software (https://software.broadinstitute.org/morpheus/).

Macrophages (DAPI⁻CD45⁺CD64^int-hi) of murine heart, liver, lung, kidney, and brain were sorted for scRNA-seq (two males and two females pooled for each tissue). Single-cell suspensions were prepared according to the Chromium Single Cell 3’ Reagent Kits User Guide (v2 Chemistry). Samples were loaded onto the 10x Chromium to produce sequencing libraries, which were processed according to methods provided by 10x Genomics. Immune cells from human cardiac tissue (CD45⁺) were processed using v3 Chemistry. Cells were sequenced and processed to generate expression matrices using Cell Ranger (10x Genomics). Raw base call (BCL) files from HiSeq2500 sequencer were demultiplexed into FASTQ files. The FASTQ files were aligned (STAR) and filtered, followed by barcode and UMI counting to generate the counts table. The scRNA-seq package Seurat (v3.1) (70, 71) was used for all downstream analyses using R 3.6.1. Details are described in the Supplemental Materials.

All data are presented as means ± SEM. For comparisons of means between two experimental groups, two-tailed Student’s t test was used. For comparisons of means between three experimental groups, two-way analysis of variance (ANOVA) or one-way ANOVA was used as appropriate. Significant differences were defined at P < 0.05 or as indicated. For multiple comparisons, Bonferroni correction was used. Statistical analyses were done using GraphPad Prism software.