Vasculature-associated fat macrophages readily adapt to inflammatory and metabolic challenges

The adipose tissue of healthy animals contains a large population of innate and adaptive immune cells, numerically dominated by macrophages (Weisberg et al., 2003; Xu et al., 2003; Lumeng et al., 2007). The functions of these cells in the healthy adipose tissue of animals in steady state are not completely understood. The proportions and numbers of these immune cell types are known to change extensively in conditions such as diet-induced obesity (DIO), genetically determined obesity, and aging, among other situations.

It has been long known that obesity is associated with a heightened inflammatory condition, in which TNFα is an important mediator of insulin resistance (Feinstein et al., 1993; Hotamisligil et al., 1993, 1995). It has been postulated that the myeloid cells found in the adipose tissue during DIO contribute to the pathogenesis of insulin resistance and the metabolic syndrome by increasing their secretion of proinflammatory cytokines (Dalmas et al., 2011; Lumeng, 2013; Wynn et al., 2013; Xu et al., 2013; Fitzgibbons and Czech, 2016; Alemán et al., 2017).

High-fat diet (HFD) induces obesity and insulin resistance in B6 mice (Surwit et al., 1988; Rebuffé-Scrive et al., 1993). Adipocytes become hypertrophic, which could be the initial signal to trigger inflammation (Sun et al., 2011; Rutkowski et al., 2015); however, cultured adipocytes made hypertrophic with oleic acid display insulin resistance without triggering an inflammatory response (Kim et al., 2015). Thus, the role of the inflammatory response in obesity-related disease is not clearly established. Moreover, wild animals endure fluctuations in food supply, in which the adipose tissue triglyceride reserves are mobilized via lipolysis or built from the diet, and little is known about the response of immune cells to these fluctuations.

In this report, we found that tissue-resident macrophages from healthy epididymal white adipose tissue (eWAT), a visceral fat depot, not only surround adipocytes but also associate tightly with blood vessels, being among the most endocytic cells in the body for blood-borne macromolecules. We refer to these macrophages as vasculature-associated adipose tissue macrophages (VAMs). Upon chronic HFD feeding, a monocyte-derived CD11c⁺CD64⁺ double-positive (DP) population of macrophages, poorly represented in animals fed a normal diet (ND), surges in the eWAT. Contrary to some predictions, these macrophages have predominantly an anti-inflammatory and repair gene signature. Importantly, eWAT macrophages from HFD-fed mice have a reduced endocytic capacity. Differently, a brief fasting period rapidly and reversibly reduces all macrophage populations, especially VAMs. Acute systemic inflammatory stress, such as that induced by LPS administration or Salmonella enterica serovar Typhimurium infection, results in a rapid and transient reduction in the main populations of VAMs; however, in this case, the decrease in VAMs is accompanied by an increase in DP macrophages. Taken all together, the macrophage populations dynamically adapt to metabolic stress and inflammation by adjusting their numbers and the expression of beneficial genes, even when pathology ends up prevailing.

We characterized the eWAT lymphoid and myeloid cell populations under steady state. We chose to focus on the visceral WAT because excess abdominal fat, unlike excess subcutaneous fat, strongly correlates with metabolic syndrome (Sharma, 2002; von Eyben et al., 2003). First, we performed an extensive flow cytometry analysis in perfused adult male mice. Flow cytometry of adipose tissue macrophages (ATMs) is made somewhat difficult because of their high autofluorescence. To minimize the impact of autofluorescence, we carefully selected the fluorochromes for each cell marker and used fluorescence minus one controls (FMOs). The gating strategy, including all markers, is shown in Fig. S1 (A–H), and the abbreviated list of surface markers used throughout this article to identify the key myeloid populations is shown in Fig. 1 A. The expression of a series of commonly used surface markers for the eWAT myeloid populations is shown in Fig. S1 I. We observed a minor Ly6C^HIGH monocyte population (Fig. S1 G), the two main populations of dendritic cells (DCs), CD11b⁻ (Fig. S1 E) and CD11b⁺ (Fig. S1 H), and four populations of macrophages, besides eosinophils (Fig. S1 F), neutrophils (Fig. S1 F), innate lymphoid cells, and lymphocytes. As rationalized below, we refer to the macrophage populations as VAM1, VAM2, PreVAMs, and DP macrophages (Fig. S1, D and H). VAM1 and 2 are similar but can be readily distinguished by their surface expression level of MHCII and Tim4 (Fig. S1 D). To gain information regarding cell morphology, we sorted the populations and observed them in Giemsa-stained cytospin preparations. The most remarkable aspect is that, in healthy mice eating an ND, VAMs and PreVAMs have lipid droplets in their cytoplasm (Fig. S1 J), a morphology that resembles ATMs from obese mice (Aouadi et al., 2014), but, to our knowledge, have not been previously reported in adipose macrophages from lean mice. The composition of all immune cells in the healthy male eWAT is summarized in Fig. S1 K. The combination of the two VAM populations outnumbers the sum of all other CD45⁺ eWAT cells. CD11b⁺ DC and eosinophils are, respectively, the third and fourth cell types by abundance. DP macrophages, which are quite rare in eWAT from mice fed ND (Fig. S1 K), express high surface levels of CD11b, CD11c, CD64, and MHCII and intermediate levels of CD206. We refer to these macrophages as DP because of their surface expression of both CD11c and CD64. PreVAMs and DP macrophages express lower level of surface CD206 than VAMs (Fig. S1 I). Thus, the myeloid populations that form the eWAT are more complex than previously described. To summarize the eWAT hematopoietic composition in a visual manner, we performed t-distributed stochastic neighbor embedding (t-SNE) analysis for CD45⁺ and CD11b⁺SiglecF⁻Ly6G⁻ eWAT cells (Fig. 1, B and C).

Given this complexity, we further used RNA sequencing (RNASeq) to analyze the transcriptome of purified eWAT monocytes, macrophages, and DCs. Principal component analysis (PCA) showed that VAM1, VAM2, and PreVAMs cluster near each other (Fig. 1 D). CD11b⁻ DCs and eWAT monocytes are more distant from the other myeloid cells analyzed. An unbiased global heat map illustrates the similarities and differences between the eWAT cell types studied (Fig. 1 E). The assessment of key transcription factor expression corroborates the existence of three major groups, one containing VAM1, VAM2, and PreVAMs; another containing the two DC populations; and the other containing monocytes and DP macrophages (Fig. 1 F).

We identified a small population of cells expressing Ly6C^HIGH, CD11b⁺, CD115⁺, CD64⁺, MHCII⁻, NK1.1⁻, Ly6G⁻, SiglecF⁻, and CD206⁻, which are markers used to identify Ly6C^HIGH monocytes (Mildner et al., 2017). However, unbiased gene expression analysis indicated that these eWAT monocytes were quite distinct from the main blood monocyte populations, Ly6C^HIGH or Ly6C^LOW (Fig. 2 A), as confirmed by PCAs and volcano plots (Fig. 2, B–E). To assess the kinetics of monocytes in the eWAT, we injected BrdU and chased the labeled cells. By 72 h, the BrdU label was becoming undetectable, a consequence of replacement with new incoming precursors, which are unlabeled, and of dilution of the BrdU⁺ label by cell division. We could observe a peak of BrdU labeling in the eWAT monocytes at 24 h, followed by a decline (Fig. 2 F). Thus, the process of eWAT monocyte differentiation appears to be rapid.

To further study the origin of eWAT myeloid cells, we performed a genetic fate mapping experiment using Cx3cr1^{Cre-ERT2-EYFP} × Rosa26^LSL-DsRed mice (Parkhurst et al., 2013), in which tamoxifen was administered to label with DsRed all cells expressing CX3CR1 at the time of administration, such as eWAT monocytes (Fig. S1 I), monocyte-DC precursors (MDPs), and common DC precursors (http://www.immgen.org). Microglia were used as a positive control because of their robust labeling by this method (Parkhurst et al., 2013; Wolf et al., 2017). There was a peak of DsRed on eWAT monocytes on day 5 after tamoxifen administration, with the label being greatly reduced by day 10 (Fig. 2 G). This result indicates that DsRed⁻ precursors, such as hematopoietic stem cells, differentiate into eWAT Ly6C^HIGH monocytes and replenish the majority of the eWAT monocyte compartment in as little as 10 d. In addition, by day 10, DsRed⁺ eWAT monocytes either differentiated into another cell type or died. The kinetics of PreVAM labeling, with a peak at day 10, is compatible with the first possibility. VAMs do not express sufficiently high CX3CR1 levels (Fig. S1 L) to induce DsRed expression upon tamoxifen administration, and they acquire only a minimal DsRed labeling at late time points, further indicating that they do not normally differentiate from DsRed⁺ cells such as monocytes or MDPs.

Finally, to study the repopulation of eWAT myeloid cells from bone marrow precursors, we irradiated WT CD45.1⁺ mice only in the hind legs and reconstituted the irradiated mice with CD45.2⁺ bone marrow. Under these conditions, the eWAT, most of the bone marrow, the brain, and all viscera are protected from irradiation, but the hind legs' exposure allows some level of chimerism. After 2 mo, we analyzed the donor-derived CD45.2⁺ cells. Virtually all myeloid populations in the eWAT were similarly reconstituted to the level observed for blood Ly6C^HIGH monocytes (5–10%) by bone marrow progenitors, with the clear exceptions of VAM1 and, especially, VAM2 (Fig. 2 H). These data are in agreement with a study showing that ATMs proliferate locally (Amano et al., 2014). Since VAMs are the most abundant eWAT macrophages, studies addressing all ATMs are likely to detect mostly their properties. Adipose monocytes, DP macrophages, and PreVAMs were, as anticipated, predominantly bone marrow derived.

In conclusion, our data indicate that once monocytes enter the eWAT, they rapidly begin to acquire a unique gene expression profile, while transiently retaining a monocyte signature. Under normal conditions, they inefficiently generate VAMs. However, as we will discuss in more detail below, monocytes generate DP macrophages and PreVAMs.

Our gene expression analysis suggests that steady state VAM1 and VAM2 macrophages in mice fed an ND have anti-inflammatory, insulin-sensitizing, repair, and detoxifying functions (Fig. 3 A). Confirming and expanding the flow cytometry data, they express high levels of genes encoding the lectin and scavenger receptors, and also genes encoding receptors for “eat me” signals in apoptotic cells (Fig. 3 A).

Unbiased analysis comparing VAM2 with the most abundant nonmacrophage myeloid cell type in the same fat pads, the CD11b⁺ DCs, showed that VAMs appear to be well equipped for endocytosis in steady state conditions (Fig. 3, B and C). They express high levels of the genes encoding clathrin heavy and α light chain (Cltc and Clta), the clathrin adaptor protein Dab2, and components of the clathrin-associated Ap2 complex, such as Ap2a2 and Ap2b1. Another endocytosis adaptor highly expressed by VAMs was Eps15, which has been implicated in both clathrin-dependent and -independent endocytosis (Carbone et al., 1997; Sigismund et al., 2005; Fig. 3 B). Interestingly, VAMs also express high levels of Endophilin B1 (Sh3glb1), involved in ultrafast, clathrin-independent endocytosis, a mechanism used by very active cells (Watanabe and Boucrot, 2017). Genes involved in macrophage macropinocytosis, such as Snx5 and Snx2 (Lim et al., 2015), are also very highly expressed by VAMs (Fig. 3 B; Kvainickas et al., 2017; Simonetti et al., 2017).

Functional enrichment analysis (gene ontology term analysis) strongly suggests that steady state VAMs have an active endocytic and filtering function (Fig. 3 C). Given the gene ontology indications, we tested the hypothesis that VAMs may be particularly apt at endocytosis of blood substances. First, we observed by confocal microscopy that VAMs maintain close contact with blood vessels, rounding them tightly (Fig. 3 D). To directly test the endocytic capacity, we injected dextran-rhodamine 70 kD i.v., and only 20 min after the i.v. injection, we sacrificed the mice for flow cytometry analysis. VAM1 and VAM2 took up dextran very efficiently, followed by PreVAMs. Neither of the two eWAT DC populations, or the eosinophils, significantly associated with dextran sulfate under these conditions (Fig. 3, E and F). Strikingly, eWAT VAMs took up more dextran-rhodamine than spleen marginal zone macrophages (Fig. 3, E and F). This indicates a special property of eWAT VAMs. Intravital microscopy showed the extremely rapid (5-min) uptake of the i.v.-injected dextran-rhodamine (Video 1 and Fig. 3 G). To confirm that the uptake of dextran-rhodamine by VAMs was not an aberrant result, we injected i.v. a different type of molecule, the protein ovalbumin (Ova). Once again, VAM1 and VAM2 strongly took up fluorescently-tagged Ova, to a higher degree than splenic marginal zone macrophages (Fig. 3 H). This association between Ova and VAMs involved endocytosis and protein processing in vivo, as proven by the fact that VAMs fed in vivo with whole Ova were able to stimulate OTII T cells in vitro (Fig. S2).

To quantify the association between VAMs and vasculature and to visualize the degree of uniformity of VAM distribution throughout the entire fat pad, we used Adipoclear, a whole adipose 3D immunolabeling and clearing method (Chi et al., 2018). To label VAMs, we injected dextran-rhodamine i.v., and 1 h later, we dissected the epididymal fat pad for tissue clearing and whole-mount imaging with a light sheet fluorescence microscope after labeling the vasculature with anti-CD31. VAMs displayed a strong and uniform perivascular localization throughout the entire pad (Video 2 and Fig. 3 I). After the analysis of 300 stacks containing 2,234 macrophages, we concluded that 90.2% of the dextran⁺ macrophages were in contact with vessels (as analyzed by Imaris software), and <1% of macrophages were at a distance >10 µm (Fig. 3 J). These results show that VAMs are indeed strongly associated with the adipose tissue vasculature. Finally, bodipy labeling of adipose droplets confirmed their presence in VAMs of lean mice (Video 3 and Fig. 3 K). This video further shows the intimate association of VAMs with the vasculature, supporting the use of the VAM nomenclature. In summary, the gene expression profile, the location in close proximity with the vasculature, and the high endocytic capacity suggest that VAMs contribute to establish and maintain a healthy adipose tissue environment.

Having established key properties of eWAT macrophages in mice fed ND, we next assessed the response of the myeloid populations to metabolic and inflammatory stress. First, we studied the myeloid cell populations in mice switched to an HFD (60% of calories from fat, 20% carbohydrates, and 20% protein) for 20 wk (Fig. 4 A). This treatment causes obesity and insulin resistance (Surwit et al., 1988; Rebuffé-Scrive et al., 1993). As expected, HFD induced dramatic changes in the eWAT macrophage populations (Fig. 4, B and C), as well as increases in the eWAT pad size (Fig. 4 D) and the cellularity of the stromal vascular fraction (SVF) and of total CD45⁺ cells (Fig. 4 E). Confirming what has long been established (Weisberg et al., 2003; Xu et al., 2003), the total number of macrophages per gram of fat was increased (Fig. 4 C). The most apparent change in myeloid cells was the dramatic surge of DP macrophages, which, as mentioned above, express high levels of CD11b, CD11c, CD64, and MHCII and intermediate levels of CD206 (Fig. 4, B and C; Nguyen et al., 2007; Shaul et al., 2010; Wentworth et al., 2010; Zeyda et al., 2010). A recent publication refers to these as CD9 macrophages (Hill et al., 2018), but we prefer to maintain the traditional CD11c-based DP nomenclature, as CD9 expression is proportionally more up-regulated than CD11c expression in VAMs by HFD.

DPs are newcomer macrophages promptly generated from blood monocytes in HFD-fed mice (Fig. 4 F). VAM1 and VAM2 also increased in number during HFD feeding (Fig. 4 C), although they were inefficiently generated by blood monocytes (Fig. 4 F), an observation consistent with the fact that VAMs are tissue resident and are maintained, to a large extent, by self-renewal (Amano et al., 2014). PreVAMs are generated from blood monocytes, although with lower efficiency than DP macrophages, but much higher efficiency than either VAM1 or VAM2 (Fig. 4 F). We performed RNAseq studies of eWAT macrophages from mice fed HFD, including DPs, PreVAMs, VAM1, and VAM2. While the characteristic VAM markers from ND-fed mice were maintained in HFD-fed mice, there was a convergence of the VAM and DP gene expression patterns (Fig. 4 G).

Our transcriptional analysis did not confirm the view that ATMs arising in HFD-fed mice are overtly proinflammatory. DP macrophages arising during HFD expressed high levels of genes encoding potent secreted anti-inflammatory molecules such as IL-10 (Il10), TGF-β1 (Tgfb1), and IL-1 receptor antagonist (Il1rn; Fig. 4, H and I). There was no expression of α, β, or γ interferons and minimal expression of IFN-κ and IL-6 in DP macrophages on VAMs before or during HFD feeding. However, Tnf, Cxcl1, pro-IL-1β, and Cd44 were moderately up-regulated. Many of the HFD-induced highly up-regulated genes in eWAT macrophages have been previously ascribed an inflammatory signal-deactivating (Tnfaip3, Rgs1) function, an insulin-sensitizing (Pparg) function, or detoxifying (Igf2r, Gpx1, Prdx1, Lrp1) functions (Fig. 4, H and I). Some of these genes, such as Lgals3 and Fabp4, have been attributed with both pro- or antidiabetogenic functions, depending on the experimental conditions. A number of beneficial genes highly up-regulated by HFD in eWAT macrophages are involved in lysosomal functions, as previously reported (Xu et al., 2013), or in facilitating digestion of excess lipids (Singh and Cuervo, 2012; Aouadi et al., 2014; Fig. 4, H and I).

In conclusion, the gene expression pattern showed that both the newcomer CD11c⁺CD64⁺CD206^INT DP macrophages, as well as the preexisting (and also expanded) CD206^HIGH VAMs, appear more involved in suppressing the negative consequences of HFD treatment than in the promotion or propagation of these negative effects.

The elongated shape of VAMs observed in mice fed ND was noticeably altered by HFD, with a predominance of rounded CD206⁺ macrophages in mice fed HFD (Fig. 4 J). Adipocytes were larger, and macrophages acquired a rounder morphology. As previously noticed (Cinti et al., 2005), the number of crown-like structures greatly increased under HFD. These morphological changes may have a correlate in the gene expression profile. Among the genes up-regulated by eWAT macrophages in mice fed HFD were several that affect the cytoskeleton of macrophages, such as Myo1E, Tpm4, Flna, and Vim, or alter the extracellular matrix, such as Mmp12 and Timp2 (Fig. 4 K). We sought to determine whether these morphological and genetic changes had functional consequences. Mice were injected with dextran-rhodamine i.v. and, after 20 min, were sacrificed to analyze rhodamine uptake. Indeed, there was a significant reduction in the uptake of rhodamine by eWAT macrophages from HFD-fed mice, compared with ND-fed mice (Fig. 4 L). To visualize the uptake in vivo, we performed intravital microscopy, which confirmed the impaired uptake of the dye, and allowed us to visualize crown-like structures in live mice (Video 4 and Fig. 4 M). In summary, our data show that the remarkable endocytic activity displayed by VAMs from ND-fed mice is highly impaired in HFD-fed mice, which could potentially affect their protective function.

Given the remarkable redistribution of VAMs and DP macrophages upon overnutrition, we assessed the impact of acute nutritional restriction (24-h fasting) on eWAT macrophages (Fig. 5 A). Upon food withdrawal, blood glucose levels decrease and lipolysis is triggered, which results in the release of nonesterified fatty acids (FFAs) into the serum. To test some metabolic parameters in mice under our conditions, we quantified serum FFAs, blood glucose, and weight loss during a 24-h fasting experiment. Serum FFA concentration peaked at ∼12–16 h of fasting, indicating a net lipolysis and release of FFAs, and thereafter serum FFA concentrations reached a plateau or diminished slightly (Fig. S3, A and B). As expected, blood glucose decreased early (Fig. S3 C). After 24 h of fasting, body weight loss was ∼10% (Fig. S3 D). Myeloid cells were dramatically reduced by 1-d fasting, with VAMs being the most affected (Fig. 5, A, B, and E). Not only VAMs were reduced; also PreVAMs, eosinophils, and both CD11b⁺ and CD11b⁻ DCs were reduced (Fig. 5 B). Unlike HFD, fasting did not provoke an increase in the DP macrophage population; instead, it dropped, albeit nonsignificantly. An overview of the CD45⁺ eWAT cell landscape upon nutritional restriction can be observed in Fig. 5 E. Interestingly, VAM decrease was almost completed after only 6 h of food deprivation (Fig. S3 E). The cells that were no longer in the eWAT could not be detected either in the spleen or in draining LNs (Fig. S3 F). To assess apoptosis, we determined the expression of active caspase-3, the main executioner of caspase during apoptosis, by flow cytometry. We did not detect a surge in active caspase-3 concomitantly with the rapid cell decrease (Fig. S3, G and H). In fact, active caspase-3 labeling of VAMs is always low, compared with other eWAT cell types. After 24 h of fasting, the weight reduction of the eWAT pads was not significantly different (Fig. 5 C), and neither was the number of SVF cells, while the total CD45⁺ cells were significantly reduced (Fig. 5 D), consistently with the decrease in all major eWAT myeloid cell populations.

In nature, mice must endure fluctuations in the food supply. Given the conspicuous reduction of VAMs that we observed after 1 d of fasting, we surmised that the number of VAMs would rebound in rapid fashion. We therefore placed a group of 1-d fasted animals back on ND and evaluated the myeloid eWAT populations after 2 d back on ND (“fasted+fed”; Fig. 5 F). To have the appropriate controls in the same experiment, we had a group of mice that were normally fed, never fasted (“fed”) and a group of mice that were fasted the last day (“fasted”; Fig. 5 F). As expected, animals feeding after fasting quickly reestablished their initial weight (Fig. S3 D). Remarkably, after 48 h of refeeding, the eWAT populations were back to the normal situation (Fig. 5 G). In these “fasting and feeding” experiments, we also did not observe significant changes in the weight of the fat pads (Fig. 5 H) or in the number of SVF cells; however, the reduction and subsequent rebound of the total number of CD45⁺ cells were statistically significant (Fig. 5 I).

To evaluate the VAM morphology after fasting, we also performed a confocal microscopy analysis. Confocal images of the eWAT showed that, compared with mice fed all along, VAMs (CD206^HIGH) from 1-d fasted mice were abnormally rounded in shape, suggesting that they partially lost their contacts with vasculature and adipocytes (Fig. 5 J, second panel from left). We chose a microscope field that contained many VAMs to allow the morphological characterization, while leaving quantitation for the flow cytometry method. The contorted vasculature, which surrounds adipocytes, allowed us to infer the expected reduction in adipocyte diameter upon fasting. VAMs from fasted+fed mice were abundant in number but had only incompletely regained their normal topology (Fig. 5 J, middle panel). Thus, the VAM depletion induced by 1-d fasting is brief and, it appears, fully reversible.

To assess whether the reconstitution of eWAT VAM numbers was due to self-division, we monitored the incorporation of BrdU over the last 48 h of the fasting and refeeding experiment described above. This gave us a fed (baseline) group, a 24-h fasted group, and a group that was fasted and had fed over the last 48 h. BrdU was poorly incorporated by VAMs in all groups evaluated (Fig. 5 K). Most likely, VAMs were generated by monocytes or PreVAMs that had divided before the BrdU pulse and were differentiating to VAMs while the BrdU pulse was ongoing. It thus appears that, to rapidly reconstitute VAMs after fasting, mice made use of monocytes or monocyte-derived PreVAMs.

Fasting causes a series of metabolic changes, of which increased lipolysis is one. It was possible that the VAM reduction was directly caused by lipolysis, with remodeling of the niche that VAMs occupy in steady state, or in response to other metabolic parameters, such as lowering of glucose or insulin. To induce lipolysis without food withdrawal, we administered a β3 adrenergic agonist, CL 316,243 (Yoshida et al., 1994). Adipocytes express β3 adrenergic receptors, which are locally triggered by sympathetic nerve activation, or systemically by adrenal catecholamines (Bartness et al., 2014). Like fasting, CL 316,243 also induced a dramatic drop in VAM populations after 12 h of a single 2-mg/kg dose (Fig. 5, L and M). Intriguingly, there was a major influx of monocytes and neutrophils into the eWAT (Fig. 5 M and Fig. S3, I–L). By confocal microscopy, we could clearly observe that CL 316,243 reduced and changed the morphology of CD206^HIGH (VAM) macrophages, and also induced atypical crown-like structures that are composed of Ly6G⁺ neutrophils, not macrophages (Fig. 5 J, rightmost panel). In sum, both fasting and a β3 adrenergic agonist caused the rapid depletion of VAMs in the eWAT, suggesting that lipolysis itself, and not the nutritional state, is the driving force in the VAM reduction.

We conclude that VAMs from fasted mice are highly sensitive to the nutritional status of the animals. Acute fasting results in an abrupt but transient decrease in VAMs, whereas chronic overnutrition results in VAMs that are partially dysfunctional.

Having studied the response of eWAT myeloid cells during two models of nutritional stress, we next studied the impact of acute inflammation on the composition of the different myeloid eWAT cell populations (Fig. 6 A). 24 h after i.v. administration of 50 µg LPS, a sublethal dose, the macrophage populations were dramatically changed in frequency and absolute numbers (Fig. 6 B). VAM1 and 2 were rapidly reduced by approximately fourfold, and the number of DP macrophages (CD64⁺CD11c⁺) more than tripled (Fig. 6, C and D). These changes were already significant at 12 h (Fig. 6 D). The number of CD11b⁻ DCs did not change upon acute LPS-driven inflammation, but the number of CD11b⁺ DCs was greatly reduced within 12 h (Fig. 6 D). Monocytes and neutrophils were increased. Almost all myeloid cell populations returned to the steady state composition by day 5 (Fig. 6 D). Despite the changes in myeloid cell populations, neither the weight of the fat pads nor the total number of SVF cells and CD45⁺ cells were significantly altered during the 5-d experiments (Fig. 6, E and F). The fact that VAMs decreased rapidly, and, at the same time, DP macrophages increased, made us investigate the possibility that LPS could induce the conversion of VAMs into DP macrophages. To investigate this possible conversion, we used a “functional fate mapping” strategy. We took advantage of the very high endocytic capacity of VAMs to label them with dextran-rhodamine and follow the fate of labeled cells. We injected dextran-rhodamine i.v.; 36 h later, injected LPS i.v.; and 24 h after LPS injection, analyzed the eWAT myeloid cells by FACS (Fig. 6, G and H). As expected, VAMs efficiently took up dextran-rhodamine, which remained in the few VAMs found after LPS treatment (Fig. 6 H). Importantly, the DP macrophages were not labeled with rhodamine, indicating that they did not derive from converted VAMs (Fig. 6 H). Most likely, they are derived from circulating monocytes, similarly to the equivalent DP population that robustly expands in the eWAT during HFD feeding. In contrast to VAMs, PreVAMs lost their rhodamine after LPS treatment (Fig. 6 H), suggesting they were quickly replaced by unlabeled cells, as we know to be the case in mice kept in steady state (Fig. 2 G).

If the rapid reduction in VAM numbers was due to a rare emigration of mature tissue macrophages from eWAT, two of the possible alternative locations would be the spleen, if VAMs emigrate via blood, or the eWAT-draining LNs, if they emigrate via lymph. We therefore determined the myeloid changes in the spleen and draining LNs 24 h after i.v. LPS injection. Macrophages with the surface profile of VAMs were not found in the spleen, while the number of splenic marginal zone macrophages was not significantly changed (Fig. S4 B). We also determined whether VAMs could emigrate to the eWAT draining LNs rapidly upon LPS treatment; however, we detected neither an increase in CD11b⁺CD206^HIGHCD64^HIGH cells in the draining LN nor an increase in dextran-rhodamine⁺ cells, an approach that would have captured emigrating VAMs even if they had suddenly changed their phenotype (Fig. S3 F). We did not observe an increase in active caspase-3 in any eWAT myeloid population (Fig. S4 C). Thus, LPS causes rapid changes in the eWAT myeloid cells, with a reduction in VAMs that we cannot ascribe to differentiation into a different cell type, emigration to the spleen or draining LNs, or caspase-3–mediated apoptosis.

As the number (and proportion) of VAMs is rapidly and dramatically reduced by LPS, it is equally remarkable that the VAM populations return to steady state numbers by 5 d after LPS administration (Fig. 6 D). Given the fact that WAT macrophages have been shown to divide locally (Amano et al., 2014), we speculated that, after LPS, the repopulation of VAMs would take place by their own cell division. We therefore monitored cell division by BrdU incorporation over the first and third days of the experiment (Fig. 6 I). Without LPS, more than half of eWAT monocytes, expectedly, become BrdU⁺ during the 24-h labeling (Fig. 6 J). VAMs incorporated a very low amount of BrdU. Among macrophages, DPs and PreVAMs, the two populations that we showed to be monocyte-derived, are the only populations with a measurable BrdU incorporation (Fig. 6 J). Upon LPS treatment, BrdU incorporation into VAMs remained very low during the first day, but increased at the third day (Fig. 6 J). This is the time in which the VAMs begin their recovery from LPS. This result is compatible with the self-division of VAMs that was discussed above, but it does not exclude that dividing monocytes took up BrdU and differentiated into VAMs.

To assess whether bacterial infections could also cause a rapid alteration in the composition of the eWAT myeloid compartment, we infected mice with the Gram-negative bacteria, S. enterica serovar Typhimurium (hereafter Salmonella; Fig. 6 K). Salmonella septicemia is a dangerous complication of Salmonella infections; therefore, to study this sequela, we injected the bacteria i.v. Similarly to LPS, systemic Salmonella infection rapidly caused a reduction of VAMs by 3 and 6 h (Fig. 6, L and O). Also similarly to LPS, Salmonella caused an increase in DP macrophages, although it was of shorter duration than the increase caused by LPS, and lower in magnitude (Fig. 6 O). Neither the weight of the fat pads nor the total number of SVF cells and CD45⁺ cells were significantly altered during the early hours after infection (Fig. 6, M and N).

Thus, the common feature of LPS and a live Gram-negative bacterial infection is that both induce a very rapid reduction in number and frequency of eWAT VAMs, with increase in DP macrophages. Alterations in eWAT myeloid cells were not, however, limited to LPS and Gram-negative bacteria. Infection with the Gram-positive bacteria Staphylococcus aureus also resulted in a decrease in the number of VAMs (Fig. S4, F–J), more markedly in VAM1 (Fig. S4 H). Analogous to LPS administration, S. aureus infection caused an increase in the proportion of eWAT monocytes (Fig. S4 H).

We conclude that LPS or systemic bacterial infections have an immediate impact on the composition of eWAT myeloid cells, in particular causing a reversible decrease in VAMs. In this regard, both acute infections and fasting affect VAMs in a similar fashion. Our results therefore show a striking adaptation of ATMs to diverse environmental challenges.

In this report, we characterized the behavior of the different murine eWAT macrophage populations in the steady state and in response to chronic and acute inflammatory and metabolic stress. By gene expression profiling, flow cytometry, and imaging, we could distinguish four populations of macrophages, which we refer to as VAM1 and VAM2, PreVAM, and CD64⁺CD11c⁺ DP macrophages. In steady state, VAM1 and VAM2 are the two dominant myeloid populations by far. Previous characterization of ATMs in steady state mice undoubtedly included these macrophages. Although steady state VAM1 and VAM2 poorly express two of the more typical M2 macrophage genes, arginase 1 (Arg1) and chitinase-like 3 (Chil3), the overall expression pattern resembles M2 macrophages, as they express high levels of CD206, CD301a, CD163, CD209, and Retnla/Fizz1. They express genes that allow the uptake of apoptotic cells, many scavenger receptors, and antioxidant and detoxifying molecules. The most previously underappreciated aspect of the steady state VAMs is their very close contact with vessels, leading to an ultrarapid endocytic capacity of diverse macromolecules present in the bloodstream. This occurs in the absence of any obvious vascular discontinuity. VAMs’ high endocytosis, gene expression, and lipid droplets suggest that, in mice fed an ND, VAM1 and VAM2 are not passively waiting for disturbance, but rather, they are active under steady state. VAM properties suggest that a possible function of these tissue-resident macrophages is to keep clean the eWAT adipose tissue environment, both by removing noxious catabolites from the adipocytes and by acting as intermediaries between the blood and the adipocytes. A correlation between phagocytosis capacity, CD206 expression, and tissue homeostasis by macrophages in different organs has recently been reported (A-Gonzalez et al., 2017).

Over the past decade, it has become apparent that many tissue-resident macrophages are not derived from bone marrow progenitors, via circulating monocytes, as had previously been thought (Schulz et al., 2012; Ginhoux and Guilliams, 2016). Typical examples of self-renewing non–bone marrow–derived macrophages in adult animals are microglia and Langerhans cells (Merad et al., 2002; Ajami et al., 2007; Chorro et al., 2009; Ginhoux et al., 2010). In the adipose tissue, self-renewal of macrophages was also shown to be the case (Amano et al., 2014). Indeed, we showed that VAM1 and VAM2 macrophages from steady state mice are not efficiently generated by bone marrow precursors, or by monocytes, although DP macrophages are.

We studied the dynamics of eWAT myeloid cells under three types of nutritional or inflammatory stresses: chronic HFD feeding, fasting, and acute systemic infections. In animals fed HFD, the most striking quantitative change was the increase in the representation of DP macrophages, which are CD11c⁺ monocyte-derived macrophages whose numbers increased by approximately one order of magnitude. The number of VAM 1 and VAM2 macrophages also increases in response to HFD, but to a lower degree. The most striking qualitative change that we observed in VAMs was their impaired endocytic capacity. The degree by which the impaired endocytic capacity of VAMs contributes to obesity-related disease remains unknown; however, it is possible that the removal of toxic catabolites becomes impaired.

We compared gene expression of all macrophage types in mice chronically fed HFD with that of mice fed ND. In general, whatever gene expression changes were induced by HFD in DP macrophages were also induced in VAMs, although the degree of the changes can be different for DP or VAMs. In fact, we know of no gene whose expression is highly up-regulated in DP macrophages and decreased in VAMs in mice fed HFD, or vice versa. Our gene expression work shows no evidence of an M2 to a classically activated/M1 shift during DIO. Gene expression of VAMs or DP macrophages cannot be defined as M1 or M1-like. Our data are more compatible with the category of “metabolically activated” macrophages (MMe), discovered by stimulating M0 macrophages in vitro with a mixture of insulin, glucose, and palmitate (Kratz et al., 2014), and subsequently validated ex vivo. The MMe activation is independent of proinflammatory signals (Kratz et al., 2014). Among the key MMe genes, Cd36 and Plin2 are highly up-regulated in our eWAT samples by HFD, and thus in good agreement with the MMe nomenclature. However, we did not observe up-regulation of Abca1, Acsl1, Bag6, Dhcr7, Far1, Plxnb2, or Ptprj. Our gene expression data on macrophages from mice fed HFD also showed activation of lysosomal metabolism and of genes encoding anti-inflammatory molecules such as A20 (Tnfaip3), the IL-1 receptor antagonist (Il1rn), IL-10, and TGF-β1, in broad agreement with other authors (Xu et al., 2013; Fitzgibbons and Czech, 2016; Coats et al., 2017).

The most striking change induced by acute 1-d fasting in the eWAT myeloid cell populations is their rapid decrease in numbers, which is more marked for VAMs. In the life of wild mice, fluctuations in dietary intake are to be expected. In this context, the rapid and reversible reduction in eWAT VAMs, eosinophils, monocytes, and DCs is remarkable. Fasting-induced lipolysis provides circulating FFAs, which can be used by other cells (e.g., skeletal muscle cells, hepatocytes) to provide energy. During this fasting period, the liver is known to activate gluconeogenesis in response to ghrelin, thus maintaining blood glycemia at acceptable levels. Lipolysis causes, naturally, adipocytes to shrink, and we speculate that this change in size removes cytoskeletal cues from the VAMs, possibly causing a number of them to undergo cell death. Thus, adipose tissue remodeling could induce the loss of VAMs. The effect of fasting on eWAT macrophages was mimicked by a β3 adrenergic agonist, CL 316,243, in conditions of food availability. These results support the view that lipolysis itself, not the nutritional state, caused the loss of VAMs. However, basal lipolysis, such as the one induced by chronic HFD feeding in the presence of hypertrophic adipocytes (Duncan et al., 2007; Zechner et al., 2009), caused an increase in VAMs.

Interestingly, in a model of adipocyte cell death caused by overexpression of an active caspase-8, Fischer-Posovszky et al. (2011) described an increase of macrophages that displayed an M2 activation profile, which we are here referring to as VAMs. Since fasting-induced lipolysis does not involve massive cell death, our results showing VAM reduction are entirely compatible.

The strong inflammatory response to a single CL 316,243 dose was not unexpected, but did not occur with acute fasting within the 24-h time frame that we studied. It was reported that one i.p. treatment with CL 316,243 resulted in an increased myeloid population in the WAT after 6 h (Mottillo et al., 2007). Chronic CL 316,243 treatment with osmotic pumps was reported to cause rapid inflammation in the adipose tissue, which was more marked in PPARγ-deficient than WT mice (Li et al., 2005). Kosteli et al. (2010) administered CL 316,243 i.p. and observed an increase in F4/80 staining by immunohistology after 18 h. Considering that F4/80 is expressed by both eWAT macrophages and monocytes, it is possible that the increase in total F4/80 cells reported by Kosteli et al. (2010) reflects the balance between the increase in monocytes that we observed and the decrease in VAMs that we also observed. It is possible that the discrepancy between the neutrophilia induced by CL 316,243 and not by 1-d fasting lies in the fact that, besides lipolysis, CL 316,243 also induces thermogenesis (Kazak et al., 2017). β3 Adrenergic agonists are currently approved by the US Food and Drug Administration for human use in overactive bladder syndrome (Takasu et al., 2007; Malik et al., 2012) and have been discussed as anti-obesity and insulin-sensitizing treatments for humans (Arch, 2008; Cypess et al., 2015). It is therefore important that the basis of the proinflammatory response caused by β3 adrenergic agonists be further studied.

Intriguingly, upon acute stress caused by i.v. LPS or bacterial infections with Salmonella or S. aureus, the number and proportion of VAMs decreases substantially and returns to normal shortly after the stressful condition is removed. Since this phenomenon was also observed after acute fasting, this dynamic fluctuation seems to be a common response to acute stress. Bacterial infections (or LPS treatment) cause metabolic changes in the host, which in turn may be associated with rapid changes in adipose myeloid cell composition. Infections induce “sickness behavior,” whereby the host reduces food consumption, triggering gluconeogenesis and lipolysis, and a higher brain consumption of ketone bodies (O’Neill and Hardie, 2013; Wang et al., 2016). LPS itself was shown to increase glucose uptake and aerobic glycolysis in macrophages (Krawczyk et al., 2010; Palsson-McDermott et al., 2015). Thus, the rapid and reversible reduction of eWAT VAMs could be part of a global response of the eWAT to acute stress, be it infectious or nutritional.

In contrast to the acute fasting, systemic bacterial infections or LPS triggered an increase in DP macrophages in the eWAT, whereas VAMs were decreased by both types of stress. Although DP macrophages induced by HFD feeding and LPS share the expression of all the key surface markers, we do not presently know the precise degree of similarity between DP macrophages elicited by systemic infections of chronic HFD feeding.

In both nutritional and inflammatory acute stresses, there is a rapid reduction and subsequent recovery of VAM numbers. In the case of LPS-induced VAM reduction, we observed BrdU incorporation in VAMs at the time of their recovery (Fig. 6 J), confirming that they are self-dividing tissue-resident macrophages. However, in the case of fasting, we did not detect BrdU incorporation in VAMs at any time point (Fig. 5 K). In this case, the most likely source of VAMs are PreVAMs, or monocytes. PreVAMs have a gene expression signature that is quite similar to that of VAMs (Fig. 1, D–F; and Fig. 3 A), and they display intermediate endocytic capacity compared with that of VAMs (Fig. 3, E and H). We showed that, unlike VAMs, PreVAMs are generated from bone marrow precursors (presumably via monocytes) in mice fed ND or HFD (Fig. 2, G and H; and Fig. 4 F). Under normal conditions, PreVAMs turn over rapidly, as shown by their basal BrdU incorporation being higher than VAMs (Fig. 5 K); their steady state numbers are one order of magnitude lower than the VAMs (Fig. S1 K). The totality of the data are therefore consistent with a model whereby PreVAMs are continuously being generated from monocytes, but they do not become VAMs unless a major VAM replacement is needed, for example in the recovery from fasting.

We analyzed the rapid reduction of VAMs in more detail. We determined that the VAMs were not converting to other eWAT cell types and that they did not emigrate to the spleen or draining LNs. Although mature tissue-resident macrophages such as VAMs are not expected to be migratory, our data do not rule out that, under acute stress, VAMs temporary emigrate to a reservoir outside of spleen and draining LNs, only to return when conditions stabilize. We did not observe signs of caspase-3–mediated apoptosis in VAMs at different time points. Due to the rapid in vivo uptake and destruction of apoptotic cells, it is sometimes difficult to detect active caspase-3 in vivo. Several caspase-3–independent programmed cell death pathways have been reported in the literature (Lee et al., 2006; O’Sullivan et al., 2007). Both necroptosis and pyroptosis occur in the absence of caspase-3 activation; however, at least in the case of fasting, these two pathways are unlikely, as they are considered proinflammatory deaths, and a brief fasting is accompanied by no signs of eWAT inflammation; instead, most inflammatory cells are decreased. It has been long known that, in vitro, LPS or LPS-containing bacteria can cause apoptosis of macrophages (Bingisser et al., 1996; Hilbi et al., 1997; Xaus et al., 2000), but it is also known that LPS is a powerful activator of macrophages, through TLR4 signaling. It has been shown that Gr-1 ILC could kill ATMs (Boulenouar et al., 2017), and that M2 macrophages, such as VAMs, are more efficient targets. It is therefore possible that Gr-1 ILC are the executioners of VAMs very rapidly after acute infectious or metabolic stress.

A major unresolved issue raised by our work is the possible adaptive value of acutely removing potentially valuable macrophages such as VAMs during inflammatory or nutritional stress. During acute infections, it is likely that the monocyte-derived DP macrophages are better equipped than VAMs to participate in pathogen removal functions. It is less clear what the adaptive value of transiently losing VAMs is during fasting, if any. Experiments to address this question directly, such as injecting VAMs to repopulate the WAT, are technically very difficult. The natural depletion of VAMs is also likely to affect the interactions with other immune cells such as effector and regulatory T cells, which abound in WAT. Thus, our characterization of the several different myeloid cell types could help the clarification of the interactions between lymphoid, myeloid, and nonhematopoietic cells in adipose tissue.

In conclusion, we showed how VAMs, which are located at the intersection between the vasculature and the adipose tissue, display an ultra-fast endocytic activity and are very sensitive to different environmental changes, yet maintain a striking capacity for rapid recovery.

Foxp3^GFP male mice in C57BL/10.PL background (Weiss et al., 2012) and C57BL/6J (000664) purchased from The Jackson Laboratory were used as WT animals throughout the study. B6 CD45.1 (564) animals were purchased from Charles River. CX3CR1-GFP mice (005582) were purchased from The Jackson Laboratory. Cx3cr1^CRE-ERT2-YFP × Rosa26^LSL-DsRed was a kind gift of the Wenbiao Gan and Dan Littman laboratories (New York University [NYU] Langone Medical Center, New York, NY; Parkhurst et al., 2013). HFD-fed male mice were treated with an HFD for ≥16 wk starting at 4–6 wk of age. Mice were fed standard chow providing 17% calories from fat (LabDiet Formulab 5008) or an HFD providing 60% calories from fat (Research Diets; D12492i). Blood glucose was determined with a One Touch Basic glucometer (Lifescan). Mice were sacrificed by CO₂ euthanasia, and serum was collected when necessary. Serum analysis of FFAs content was performed using an FFA assay kit (Abcam; ab65341). As indicated in the text, some mice were treated i.v. with 2 mg/kg of β3 receptor agonist (CL 316,243; Tocris; 1499), 50 µg of LPS (Sigma-Aldrich; L4391), 18 mg/kg of dextran-rhodamine 70 kD (Sigma-Aldrich; R9379), 2 mg/kg of OVA-A647 (Thermo Fisher Scientific; O34784), and 10 mg of tamoxifen (Sigma-Aldrich; T5648). Animals were housed at NYU Medical Center Animal Facility under specific pathogen–free conditions. All procedures were approved by NYU School of Medicine Institutional Animal Care and Use Committee.

To isolate leukocytes from adipose tissue, mice were anesthetized with a mixture containing 12.5 mg/ml ketamine, 2.5 mg/ml xylazine, and 25 mg/ml acepromazine and intracardially perfused with PBS with 5 mM EDTA. After dissection, the fat tissue was minced and incubated for 35 min at 37°C under gentle agitation (150 rpm) in complete Roswell Park Memorial Institute medium (cRPMI; 10% FBS, 1 mM of sodium pyruvate [Gibco; 11360-070], Glutamax 1× [Gibco; 35050-061], MEM nonessential amino acids [Gibco; 11140-050], Hepes [Gibco; 15630-080], penicillin/streptomycin [Gibco; 15140-122], and RPMI [Gibco; 21870-076] containing 1 mg/ml of collagenase type VIII [Sigma-Aldrich; C2139], and 100 µg/ml of DNase I [Sigma-Aldrich; 10104159001]). After digestion, the suspension was washed with ice-cold cRPMI and spun down. The floating fat layer was removed, and the leukocyte-enriched pellet was resuspended in cRPMI and filtered through a 70-µm cell strainer.

Although the RNAseq data shown in this article were obtained following the above extraction method, we also performed a comparative experiment analyzing VAM2, in which we used 2% FFA-free BSA (Thermo Fisher Scientific; BP9704100) in the digestion and staining buffers instead of FBS. RNAseq results between the two extraction methods were highly similar without displaying a single gene with significant adjusted P value. Data are deposited in the Gene Expression Omnibus repository, accession no. GSE126407.

Mice were perfused as described above. Spleens were minced and digested in PBS containing 5% FBS, 10 mM Hepes, and 160 U/ml collagenase D (Sigma-Aldrich; 11088882001) at 37°C for 35 min. After incubation, 10 mM EDTA was added for 5 min in to disrupt DC–T cell complexes. The cell suspension was then filtered through a 70-µM cell strainer.

S. enterica serovar Typhimurium (SL1344) strain (Yu et al., 2014) was used in infection experiments. Mice were injected i.v. with 10⁸ CFU and analyzed 3–6 h after infection, as indicated. Briefly, a single aliquot of Salmonella was grown in 3 ml of LB overnight at 37°C with agitation, and then the bacteria were subcultured (1:30) into 3 ml of LB for 3 h at 37°C with agitation. Bacteria were injected i.v. into recipient mice in a total volume of 100 µl. S. aureus strain Newman were cultured overnight from a single colony in 5 ml tryptic soy broth with shaking at 37°C. Cultures were diluted 1:100 in tryptic soy broth (5 ml for i.v. infection) and cultured with shaking at 37°C for an additional 3 h until optical density at 600 nm was 1.5 (10⁹ CFU/ml). I.v. infection was performed via retro-orbital vein. Mice were injected i.v. with 10⁸ CFU and analyzed 12 h after infection, as indicated.

The following antibodies were purchased from BD Biosciences: anti-active caspase-3 (clone C92-605) BUV395, 564095; anti-BrdU (clone 3D4) Percp-cy5-5, 560809; anti-CD45 (clone 30-F11) BUV 395, 564279; anti-Ly6G (clone 1A8) Percp-cy5.5, 560602; anti-SiglecF (clone E50-2440) BV421 and APC-R700, 565934 and 565183; and Streptavidin-BUV395, 564176. The following antibodies were purchased from Thermo Fisher: anti-CD8 (clone 53–6.7) APC-E780, 47–0081-82; anti-CD11b (clone M1/70) APC-E780, 47–0112-82; anti-IL33R (clone RMST2-2) PE, 12–9335-82; anti KI-67 (clone SolA15) PE, 12–5698-82; Ovalbumin-A647, O34784; Bodipy 493/503, D3922; Isolectin GS-IB4 A647, I32450; Isolectin GS-IB4 A568, I21412; Qtracker 655 vascular labels, Q21021MP; and Hoechst 33342, 62249. 4′,6-diamidino-2-phenylindole (DAPI), D3571 or live/dead fixable blue, L34961 were used to exclude dead cells. The following were purchased from BioLegend: anti-CD4 Percp-cy5.5 (clone RM4-5), 100540; anti-CD11b (clone M1/70) A488 and A647, 101217 and 101218; anti-CD11c (clone N418) A647, 117312; anti-CD19 (clone 1D3) FITC, 152404; anti-CD31 (clone 390) A647, 102415; anti-CD45 (clone 30-F11) A700, 103128; anti-CD64 (clone x54-5/7.1) APC, PE, BV421 and Biotin, 139306, 139304, 139309 and 139318; anti-CD206 (clone C068C2) A647, PE-CY7 and PE/DAZZLE594, 141711, 141720 and 141731; anti-CD301 (clone LOM-14) Percp-cy5-5, 145709; anti-F4/80 (clone BM8) BV421, 123137; anti-I-A/I-E (clone M5/114.15.2) A488, A647 and BV711, 107616, 107618 and 107643; anti-Ly6C (clone HK1.4) A488 and BV510, 128021 and 128033; anti-Ly6G (clone 1A8) PE/DAZZLE594, 127647; anti-NK1.1 (clone PK136) BV650 and PE/DAZZLE594, 108736 and 108747; anti-γδ TCR (clone GL3) FITC, 118105; anti-Thy1.2 (clone 30-H12) FITC, 105306; anti-TCRβ (clone H57-597) BV510, 109233; and Streptavidin-BV421; 405225. The following was purchased from Sigma-Aldrich: Rhodamine B isothiocyanate-dextran 70 kD; R9379.

Flow cytometry data were acquired on an LSR-II flow cytometer (Becton Dickinson) and analyzed using FlowJo software (TreeStar 8.7) and Cytobank (https://www.cytobank.org) for dimensionality reduction and visualization by t-SNE. All stainings were performed as recommended by the antibody manufacturer. All samples were preincubated with Fc block (2.4G2; Bioxcell; BE0307) for 15 min at 4°C. FACS sorting was performed using an ARIA II sorter using a 100-µm nozzle (Becton Dickinson). Flow cytometry of ATMs is made somewhat difficult because of their high autofluorescence. To minimize the impact of autofluorescence, we meticulously selected the fluorochromes for each cell marker and used FMOs.

WT mice were injected i.p. with 1 mg of BrdU as indicated in the figure legends. Spleens and fat pads were harvested and processed as described above. Cells were surface stained and intranuclearly stained for BrdU using the BD BrdU proliferation kit (552598).

To evaluate if ATM subpopulations have origin from circulating monocytes in HFD-treated animals, monocytes were isolated from the spleens and bone marrow of CD45.1 WT mice by flow cytometry sorting as CD45⁺CD90⁻CD19⁻NK1.1⁻CD11b⁺LY6G⁻MHCII⁻Ly6C^HIGH. Sorted cells (2 × 10⁶) were injected i.v. into CD45.2 WT mice. After 80 h, adipose tissue cells were isolated and analyzed by flow cytometry.

Total RNA from sorted target cell populations was isolated using TRIzol LS (Invitrogen; 10296010) followed by DNase I (Qiagen; 79254) treatment and cleanup with RNeasy Plus Micro kit (Qiagen; 74034). RNA quality was assessed using Pico Bioanalyser chips (Agilent; 5067-1513). Only RNAs displaying a RNA integrity number of 9 or higher were used for downstream procedures. RNAseq libraries were prepared using the Nugen Ovation Trio low input RNA Library Systems V2 (Nugen; 0507-08) according to the manufacturer’s instructions by the NYU Genome Technology Center. Pooled libraries were sequenced as 50-nucleotide, paired-end reads on an Illumina HiSeq 2500 using v4 chemistry.

Illumina sequencing adapters and reads with Phred quality scores <20 were removed with Trimmomatic. Trimmed reads were mapped to the Mus musculus genome Ensembl annotation release 91 using STAR v2.5.3a with default settings. The number of reads uniquely mapping to each gene feature in the corresponding annotation file was determined using featureCounts. The resulting count tables were passed to R for further analyses. RNAseq data are deposited at the Gene Expression Omnibus repository under accession no. GSE126407.

Consistency between replicates was checked using PCA and Euclidean distance–based hierarchical clustering on normalized counts. Weakly expressed genes, defined as having <2 fragment per kilobase of transcript per million mapped reads in at least one group of replicates, were removed from differential expression analysis. Counts normalization were performed using the method of trimmed mean of M-values from edgeR package. Differential expression analyses were performed using limma. Genes were considered significant when the observed log2 fold-change was >1 and the adjusted P value was <0.05. The set of significant genes for each contrast of interest was passed to Goseq for functional gene categories enrichment analyses.

Sorted cells were spun onto Shandon Cytoslides with a Thermo Shandon Cytospin 4 cytofuge (10 min at 500 rpm), fixed with methanol; and stained with Wright-Giemsa (Sigma-Aldrich; GS500). Images were acquired in a Zeiss AxioObserver microscope using a magnification of 63× N.A. 1.4 in bright field at room temperature (RT). Zen software was used for image acquisition and ImageJ software, Fiji version 1.0 (National Institutes of Health), for contrast, brightness, and pseudo-color adjustments.

WAT was stained as described (Martinez-Santibañez et al., 2014), with modifications. Mice were euthanized and slowly perfused by intracardiac injection with 20 ml of PBS/5 mM EDTA. Epididymal fat pads were excised and fixed for 1 h in PBS/1% PFA with gentle shaking at 4°C. After that, samples were blocked for 1 h in PBS/5% BSA (blocking buffer) with gentle rocking at RT. Primary antibodies were diluted in blocking buffer and added to fat samples for 18 h at 4°C. Samples were then washed three times with cold PBS and incubated with blocking buffer-diluted Bodipy (Thermo Fisher Scientific; D3922), Alexa Fluor 647–conjugated isolectin, Alexa Fluor 568–conjugated isolectin, and/or appropriate antibodies with gentle rocking at RT. Nuclei were stained with DAPI for 5 min at RT. Fat pads were imaged at RT on an inverted confocal microscope (Zeiss 710 MP; optical lenses 25× N.A. 0.8 and 63× N.A. 1.4 or a high-resolution Zeiss 880 with Airyscan fast; optical lenses 25× N.A. 0.8 and 63× N.A. 1.4) by placing the pad in Fluoromount-G (Southern Biotech; 0100-01) in a chambered coverslip. Zen software was used for image acquisition. ImageJ software, Fiji version 1.0, or Imaris/Bitplane software were used for contrast, brightness, and pseudo-color adjustments.

The method was performed as described before (Chi et al., 2018). Mice were injected with 18 mg/kg of dextran-tetramethylrhodamine lysine fixable (Thermo Fisher Scientific; D1818). After 1 h, animals were anesthetized as described above, and an intracardiac perfusion/fixation was performed with 1× PBS followed by 4% PFA. All harvested samples were postfixed in 4% PFA at 4°C overnight. Fixed samples were washed in PBS for 1 h three times at RT.

Fixed samples were washed in 20, 40, 60, and 80% methanol in H₂O/0.1% Triton X-100/0.3 M glycine (B1N buffer, pH 7), and 100% methanol for 30 min each. Sample were then delipidated with 100% dichloromethane (DCM; Sigma-Aldrich) for 30 min three times. After delipidation, samples were washed in 100% methanol for 30 min twice, then in 80, 60, 40, and 20% methanol in B1N buffer for 30 min each step. All procedures above were performed at 4°C with shaking. Samples were then washed in B1N for 30 min twice followed by PBS/0.1% Triton X-100/0.05% Tween 20/2 µg/ml heparin (PTxwH buffer) for 1 h twice before further staining procedures.

Samples were incubated in primary antibody dilutions in PTxwH for 4 d. After primary antibody incubation, samples were washed in PTxwH for 5, 10, 15, and 30 min; 1, 2, and 4 h; and overnight, and then incubated in secondary antibody dilutions in PTxwH for 4 d. Samples were finally washed in PTwH for 5, 10, 15, and 30 min; 1, 2, and 4 h; and overnight.

Samples were embedded in 1% agarose in PBS and dehydrated in 25, 50, 75, and 100% methanol/H₂O series for 30 min at each step at RT. Following dehydration, samples were washed with 100% DCM for 30 min twice, followed by an overnight clearing step in dibenzyl ether (DBE; Sigma-Aldrich). Samples were stored at RT in the dark until imaging.

All whole-tissue samples were imaged on a light-sheet microscope (Ultramicroscope II; LaVision Biotec) equipped with 1.3× (used for whole-tissue views with low-magnification) and 4× objective lenses (used for high-magnification views) and an sCMOs camera (Andor Neo). Images were acquired with InspectorPro software (LaVision BioTec). Samples were placed in an imaging reservoir filled with DBE and illuminated from the side by the laser light sheet. The samples were scanned with the 488-, 640-, and 790-nm laser channels and with a step size of 3 µm for 1.3× objective and 2.5 µm for 4× objective.

All whole-tissue images were generated using Imaris x64 software (version 8.0.1; Bitplane). 3D reconstruction was performed using the “volume rendering” function. Optical slices were obtained using the “orthoslicer” tool. Optical sections were generated using the “snapshot” tool.

WT mice were anesthetized with i.p. injection of 20 µl/g of 2.5% Avertin before surgery for intravital imaging. Anesthesia was maintained by continuous administration of 1% isofluorane and 1 liter/min oxygen mixture while imaging was performed. Mice were injected with Hoechst dye (blue) and Qtracker 655 vascular labels for visualization of cell nuclei and blood vessels. qDots have a PEG surface coating specially designed to minimize interactions and do not get taken up by VAMs or any other cell type in the adipose tissue. 10 min following induction of anesthesia, mice were placed on a custom platform heated to 37°C. Upon loss of recoil to paw compression, a small incision was made in the abdomen. One perigonadal fat pad was located, exposed, and placed onto a raised block of thermal paste covered with a wetted wipe. A coverslip was placed on top of the fat. The platform was then transferred to the FV1000MPE Twin upright multiphoton system (Olympus) heated stage. Time-lapse was ±30 s with a total acquisition time between 15 min and 1 h depending on the condition. A complete Z stack (80 µm) was made during each acquisition. Depending on the experimental condition, 450 µg of dextran-rhodamine 70 kD was injected i.v. after 15 min of video acquisition.

For each individual video, time-lapse was ±30 s with a total acquisition time varying from 15 min to 1 h. Raw data as imported from the microscope were used for all subsequent analyses. Imaris (Bitplane) software was used for preparation of individual images and videos. Some postanalysis pseudo-color adjustment was performed for individual images and videos to account for differences in autofluorescence and Hoechst labeling. For all comparison images and videos (comparing pseudo-color channels showing dextran-rhodamine labeling over time), identical settings for all image parameters were used.

Mean, SD, and SEM values were calculated with GraphPad Prism (GraphPad Software). Error bars represent ± SEM. Unpaired Student’s t test was used to compare two variables. For three or more variables, one-way ANOVA was used with a Tukey post hoc test, as indicated in each figure legend. P values <0.05 were considered significant. Statistics symbols were *, P < 0.05; **, P < 0.01; and ***, P < 0.001.

Fig. S1 shows gating strategies and further flow cytometric and morphological analysis of the myeloid cells analyzed. Fig. S2 shows that VAMs present antigens captured from blood circulation to T cells. Fig. S3 shows additional analyses of animals fasted or treated with a β3 adrenergic receptor agonist. Fig. S4 shows further analyses of animals treated with LPS or infected with S. aureus. Video 1 shows the rapid uptake of blood-borne molecules by VAMs. Video 2 reveals the close association between VAMs and adipose tissue blood vessels. Video 3 shows that VAMs display lipid droplets in their cytoplasm. Video 4 shows an impaired uptake of rhodamine by macrophages from mice fed HFD.

We thank Drs. Carey Lumeng, Peter J. Murray, P’ng Locke, and Anthony W. Ferrante for helpful discussions at different stages of this project and Maria Lafaille for valuable comments on the manuscript. We thank the kind assistance of Dr. Michael Cammer in helping with microscope acquisitions and image processing. We thank Alireza Khodadadi-Jamayran from NYU Langone Medical Center Applied Bioinformatics Laboratory for initial assembly of processed RNAseq data. We thank Rita Chan for preparing bacteria cultures for injection. We thank Dr. Gabriel Victora for kindly allowing the use of the FV1000MPE Twin upright multiphoton system (Olympus). We thank Dr. Dan Littman for generously sharing equipment, and critical review of the manuscript.

The NYU microscope core is supported by the National Center for Research Resources, grant S10 RR023704-01A1. NYU genomic core is a shared resource partially supported by the Laura and Isaac Perlmutter Cancer Center, Cancer Center Support Grant P30CA016087. H.M. Silva was supported by a fellowship from the National Council for Scientific and Technological Development and a postdoctoral fellowship from Dr. Bernard B. Levine. A. Báfica, G.F. Rodrigues-Luiz, D. Mucida, and D.S. Mansur were funded by the Computational Biology Program, Coordenação de Aperfeiçoamento de Pessoal de Nível Superior.

The authors declare no competing financial interests.

Author contributions: H.M. Silva, A. Báfica, and J.J. Lafaille conceived the study; H.M. Silva and A. Báfica supervised the study; H.M. Silva, A. Báfica, D. Mucida, and J.J. Lafaille designed experiments; H.M. Silva, A. Báfica, P.d.A. Santos, R.D.N. Arifa, and D.A.G.B. Mendes performed and analyzed experiments; J. Chi, A. Crane, and P. Cohen set up, performed, and analyzed adipo clear experiments; P. Martin, K. Cadwell, and V.J. Torres provided bacterial reagents and expertise; D.P. Hoytema van Konijnenburg and B.S. Reis helped set up adipose tissue in vivo imaging experiments; G.F. Rodrigues-Luiz and D.S. Mansur were responsible for RNAseq data analysis; H.M. Silva prepared figures; and J.J. Lafaille, H.M. Silva, and D. Mucida wrote the paper.