OBJECTIVE—To establish the mechanism of the phenotypic switch of adipose tissue macrophages (ATMs) from an alternatively activated (M2a) to a classically activated (M1) phenotype with obesity.
RESEARCH DESIGN AND METHODS—ATMs from lean and obese (high-fat diet–fed) C57Bl/6 mice were analyzed by a combination of flow cytometry, immunofluorescence, and expression analysis for M2a and M1 genes. Pulse labeling of ATMs with PKH26 assessed the recruitment rate of ATMs to spatially distinct regions.
RESULTS—Resident ATMs in lean mice express the M2a marker macrophage galactose N-acetyl-galactosamine specific lectin 1 (MGL1) and localize to interstitial spaces between adipocytes independent of CCR2 and CCL2. With diet-induced obesity, MGL1+ ATMs remain in interstitial spaces, whereas a population of MGL1−CCR2+ ATMs with high M1 and low M2a gene expression is recruited to clusters surrounding necrotic adipocytes. Pulse labeling showed that the rate of recruitment of new macrophages to MGL1− ATM clusters is significantly faster than that of interstitial MGL1+ ATMs. This recruitment is attenuated in Ccr2−/− mice. M2a- and M1-polarized macrophages produced different effects on adipogenesis and adipocyte insulin sensitivity in vitro.
CONCLUSIONS—The shift in the M2a/M1 ATM balance is generated by spatial and temporal differences in the recruitment of distinct ATM subtypes. The obesity-induced switch in ATM activation state is coupled to the localized recruitment of an inflammatory ATM subtype to macrophage clusters from the circulation and not to the conversion of resident M2a macrophages to M1 ATMs in situ.
Obesity activates inflammatory pathways in leukocytes that contribute to the pathogenesis of obesity-associated diseases, such as type 2 diabetes and atherosclerosis. In particular, adipose tissue macrophages (ATMs) have been identified as the primary source of inflammatory cytokine production in adipose tissue and a key component in the progression to insulin resistance with obesity. ATMs are increased in visceral fat depots with obesity and correlate with measures of insulin resistance (1). A range of mouse models with loss-of-function mutations in genes important in macrophage recruitment (Ccr2), inflammatory cytokine production (Tnfα), and proinflammatory activation (Ikkβ) have demonstrated protection from high-fat diet–induced insulin resistance (2–4).
Despite these advances, the function of ATMs in lean and obese states is poorly understood. It has been suggested that ATMs influence a range of processes in adipose tissue, including adipogenesis, angiogenesis, and the response to hypoxia (5–7). One potential clue to understanding these diverse functions emerges from the evidence that macrophages can exist in different activation states with distinct properties (8). On stimulation with lipopolysaccharide (LPS) and interferon-γ, macrophages assume a proinflammatory classical activation profile also known as M1. However, under the influence of TH2 cytokines interleukin (IL)-4 or IL-13, macrophages assume an alternative activation state (M2a) and produce immunosuppressive factors, such as IL-10, IL-1RA, and arginase (9). We have previously demonstrated that ATMs in lean mice are polarized toward an alternatively activated state (M2a), whereas ATMs in obese mice have an M1 profile (10). The importance of this M1/M2a balance in glucose metabolism has been illustrated in macrophage-specific Pparg−/− mice that have deficient M2a polarization, increased adipose tissue inflammation, and worse insulin resistance (11,12).
How does obesity switch ATMs from an M2a to an M1 activation state? One hypothesis is that resident M2a ATMs are uniformly converted to an M1 state. A second hypothesis is that M1 ATMs are generated by the obesity-induced trafficking of a subset of inflammatory CCR2+Ly6GhiCX3CR1low monocytes to fat similar to what is seen in atherosclerotic lesions (13). Currently, there is insufficient information to differentiate between these two models. In this study, we provide evidence for the latter model by characterizing subtypes of ATMs in lean and obese mice. We identified a population of resident ATMs expressing the M2a marker macrophage galactose N-acetyl-galactosamine specific lectin 1 (MGL1/CD301) that localizes to interstitial spaces between adipocytes. High-fat diet feeding does not alter the presence of MGL1+ ATMs but instead induces the spatially restricted accumulation of M1-polarized MGL1− ATMs in clusters surrounding dead adipocytes. Kinetic studies show that the rate of recruitment of these subtypes into adipose tissue differs significantly and support a model in which the M2a-to-M1 switch in ATMs is coupled to differential monocyte/macrophage recruitment.
RESEARCH DESIGN AND METHODS
Animal care and use.
Male C57Bl/6 mice or Ccr2−/− mice were fed a normal chow (LabDiet 5001) or high-fat diet (45% kcal from fat, Research Diets) ad libitum for 16–20 weeks from 8 weeks of age. Ccl2−/− mice were provided by D. Eitzmann (University of Michigan). Animals were housed in a specific pathogen–free facility and given free access to food and water. All animal use was in compliance with the Institute of Laboratory Animal Research Guide for the Care and Use of Laboratory Animals and approved by the University Committee on Use and Care of Animals at the University of Michigan.
Mice were euthanized and slowly perfused by intracardiac injection with 10 ml of 1% paraformaldehyde diluted in PBS. Small fat pad samples were excised and blocked for 1 h in 5% BSA in PBS with gentle rocking at room temperature. For detection of intracellular antigens, blocking and subsequent incubations were done in 5% BSA in PBS with 0.3% Triton X-100. Primary antibodies were diluted in blocking buffer to 0.5–1 μg/ml and added to fat samples for 2 h at room temperature or overnight at 4°C. After three washes, fluorochrome-conjugated secondary antibodies were added for 1 h at room temperature, and nuclei were stained with TOPRO-3. Alexa 647–conjugated isolectin (Molecular Probes) was used to identify vascular structures. Fat pads were imaged on an inverted confocal microscope (Olympus Fluoview 300) by placing the pad in 95% glycerol in a chambered coverslip. Anti-mouse antibodies used were against TLR4 and CCR2 (Santa Cruz Biotechnology); CD11c, CD11b, and Caveolin1 (BD Biosciences); and F4/80 and MGL1 (clone ER-MP23; Abcam).
Macrophage labeling with PKH26.
PKH26 (Sigma) was prepared for phagocytic labeling per the manufacturer's instructions. One milliliter of 1 μmol/l PKH26 was injected intraperitoneally into mice fed high-fat diet for 12 weeks. Clusters were imaged by taking 5- to 10-μm z-sections, and TOPRO3+F4/80+ cells were scored for PKH26 intracellular staining.
Flow cytometry and gene expression analysis.
Flow cytometry on stromal vascular fraction (SVF) cells and real-time RT-PCR analysis were performed as previously described (10). Anti-mouse antibodies used were F4/80-PE-Cy5 (eBioscience) and CD301/MGL1 (ABDSerotec).
Macrophage conditioned media and adipocytes.
Differentiation, 2-deoxyglucose uptake assays, and immunoblots of 3T3-L1 adipocytes were performed as described previously (14). Bone marrow macrophages were derived by culturing bone marrow suspensions in Dulbecco's modified Eagle's medium with 10% heat-inactivated fetal bovine serum in the presence of 20% L929 conditioned medium for 7 days.
Resident ATMs reside between adipocytes and express the M2a marker MGL1.
We first examined the properties of resident ATMs in lean mice. Consistent with previous studies (10), we observed that ATMs in multiple fat depots uniformly express MGL1, an established marker of alternative M2a macrophage activation (Fig. 1A) (15,16). MGL1+ ATMs were uniformly F4/80+ and CD11b+ and resided in interstitial spaces between adipocytes in lean mice (Fig. 1B). These ATMs were occasionally associated with blood vessels but were not restricted to perivascular regions (Fig. 1C). Reinforcing their M2a polarization, MGL1+ resident ATMs also express IL-10 (Fig. 1D).
To understand the mechanism of recruitment of MGL1+ ATMs, we examined the expression of chemokine receptors previously implicated in macrophage migration. MGL1+ ATMs expressed CCR2 (Fig. 1E); however, the localization and quantity of the MGL1+ ATMs were not altered in visceral fat pads from Ccr2−/− and Ccl2−/− mice (Fig. 1F). This demonstrates that MGL1+ ATMs are recruited to fat via CCR2- and CCL2-independent mechanisms.
Obesity induces the accumulation of MGL1− ATMs in adipose tissue while retaining MGL1+ interstitial ATMs.
The M2a marker MGL1 permitted the examination of how obesity biases ATMs toward an M1 activation state. If obesity alters the phenotype of all resident ATMs, we predicted that MGL1 expression might be decreased in all ATMs with high-fat diet. However, we observed that MGL1 expression was retained in interstitial ATMs in high-fat diet–fed mice, despite adipocyte hypertrophy and ongoing adipocyte death (Fig. 2A). In contrast, ATMs in clusters (“crown-like structures”) were uniformly MGL1−. Occasional MGL1+ ATMs were observed toward the outer rim of the clusters. To see if this heterogeneity of MGL1 expression is found in other models of obesity, ATMs in db/db mice were examined and found to have an identical distribution of MGL1+ ATMs in interstitial spaces and MGL1− ATMs confined to clusters (supplementary Fig. 1, available in an online appendix at http://dx.doi.org/10.2337/db08-0872).
The MGL1− ATMs in clusters were F4/80+ and CD11b+ and expressed high levels of inflammatory markers CD11c and TLR4, although these were also expressed to some degree in MGL1+ ATMs (Fig. 2B–D). In contrast, MGL1− ATMs had high affinity for isolectin, a previously identified marker for inflammatory ATMs (5), whereas MGL1+ ATMs did not bind isolectin (Fig. 2E). MGL1− ATMs in clusters also had high CCR2 expression, whereas the interstitial MGL1+ ATMs had low CCR2 expression (Fig. 2F). Conversely, IL-10 expression was prominent in interstitial ATMs and downregulated in the ATM clusters (Fig. 2G). Overall, these results suggest that M1 activation of ATMs is spatially restricted to ATM clusters and is superimposed on the MGL1+ ATM population seen in lean mice.
ATM clusters were closely associated with vascular structures (Fig. 2E) and surrounded lipid droplets from dead adipocytes that were caveolin negative. Some adipocytes were identified with a discontinuous distribution of caveolin at the membrane and were surrounded by a thin layer of macrophages, suggesting that this is an early lesion (Fig. 2H). Uniformly, lipid droplets surrounded by multiple layers of macrophages were caveolin negative, consistent with coordinated regulation of M1 ATM clustering and adipocyte death. This suggests that loss of adipocyte membrane integrity is the likely initiating event for MGL1− ATM accumulation.
To assess the importance of CCR2 for MGL1− ATM accumulation, obese Ccr2−/− mice were analyzed. Compared with Ccr2+/+ mice, Ccr2−/− mice had fewer ATM clusters (Fig. 2I). However, the ATMs in these clusters were similar to those seen in Ccr2+/+ mice in that they were CD11c+ and MGL1− (Fig. 2J). In these mice, all F4/80+ ATMs were also CD11c+ (supplementary Fig. 2, available in the online appendix). This suggests that CCR2-independent mechanisms exist to recruit MGL1− ATMs to clusters.
Diet-induced obesity induces the recruitment of M1-polarized MGL1− ATMs into visceral fat.
The imaging data suggested that the primary difference between ATMs in lean and obese mice is the accumulation of MGL1− ATMs. To confirm this, we analyzed the SVF from adipose tissue from lean and obese mice by flow cytometry. Of ATMs in both epididymal and subcutaneous fat pads from lean mice, >80% were MGL1+ (Fig. 3A). In high-fat diet–fed mice, there was a decrease in the percentage of MGL1+ ATMs in epididymal fat (Fig. 3B). This decrease was accompanied by a 2.5-fold increase in the percentage of MGL1− ATMs in obese mice with high-fat diet. These changes resulted in a shift in the ratio of M2a to M1 ATMs from 4:1 in normal diet–fed mice to 1.2:1 in high-fat diet–fed mice. We next examined ATMs from subcutaneous fat and found that the percentage of MGL1+ ATMs was not changed with high-fat diet in this fat depot. Like visceral fat, high-fat diet induced the accumulation of MGL1− ATMs in subcutaneous fat, but the accumulation of MGL1− ATMs in subcutaneous fat was less than what was observed for visceral fat.
To verify that MGL1+ and MGL1− ATMs have different activation profiles, gene expression was analyzed in the two ATM populations isolated from obese mice. Compared with MGL1− ATMs, MGL1+ ATMs had increased expression of M2a macrophage genes Il10, Arg1, and Pgc1b (Fig. 3C). In contrast, MGL1− ATMs had significantly higher expression of inflammatory genes Nos2 and Il1b, consistent with a proinflammatory activation state. Examination of other proinflammatory genes demonstrated trends of higher expression of Ccl3 and Ccl4 in MGL1− ATMs relative to MGL1+ ATMs.
MGL1− ATMs are recruited to clusters at a faster rate than MGL1+ interstitial ATMs.
The spatial separation between MGL1− and MGL1+ ATMs in obesity suggests that these subtypes are recruited to adipose tissue via different mechanisms. To evaluate this, we used PKH26 labeling to identify newly recruited ATMs. Mice were injected with PKH26 that label ATMs with near 100% efficiency (17). Because this inert dye is taken up by macrophages and not monocytes (18), ATMs recruited to adipose tissue after the time of injection can be identified as F4/80+PKH26− cells. Because this experiment is dependent on similar rates of retention of the PKH26 dye in M1- and M2-polarized macrophages, we examined PKH26 dye staining in macrophages stimulated with vehicle, LPS, or IL-4 (supplementary Fig. 3, available in the online appendix). This demonstrated that the intracellular retention of PKH26 dye was not altered by macrophage activation with LPS or IL-4 for up to 7 days.
After 12 weeks of high-fat diet feeding, mice were injected with PKH26 and were analyzed at different times after injection for F4/80+PKH26− ATMs. Three days after injection, almost all ATMs in both clusters and interstitial spaces were PKH26+ (Fig. 4A). With increasing time after injection, the percentage of PKH26− ATMs in the clusters steadily increased, which suggests that new ATMs were recruited specifically to the clusters (Fig. 4B). In contrast, the proportion of PKH26− ATMs in interstitial spaces remained low throughout the time course, suggesting that ATMs traffic to interstitial areas at a significantly slower rate than they traffic to clusters.
These results suggested that MGL1 expression defines the differentially recruited ATM subtypes identified by PKH26 labeling. By confocal microscopy, the majority of the PKH26− ATMs in clusters were MGL1−, and interstitial cells with retained PKH26 label were MGL1+ (Fig. 4C). To confirm this, ATMs were analyzed by flow cytometry 28 days after PKH26 injection (Fig. 4D). This showed that PKH26+F4/80+ ATMs were MGL1+, whereas PKH26−F4/80+ had low MGL1 expression.
To assess how CCR2 deficiency alters these recruitment dynamics, we injected high-fat diet–fed Ccr2−/− mice with PKH26 and analyzed the appearance rate of ATMs in clusters and in interstitial spaces. The percentage of PKH26−F4/80+ cells in clusters 28 days after injection from Ccr2−/− mice was decreased compared with Ccr2+/+ mice, which suggests that ATMs are recruited to clusters at a slower rate in the absence of CCR2 (Fig. 4E). The accumulation of new interstitial ATMs was not altered in Ccr2−/− mice, consistent with our previous results (Fig. 1F).
M1- and M2a-activated macrophages have different effects on adipogenesis and glucose uptake.
The differences between ATM subsets suggest that M1 and M2a macrophages may have different effects on adipocyte function. To test this, we used conditioned media from untreated, LPS-stimulated (M1), and IL-4–stimulated (M2a) macrophages in assays with 3T3-L1 adipocytes. These activation conditions generated the expected macrophage phenotypes: LPS stimulated tumor necrosis factor-α and IL-12 production, whereas IL-4 induced arginase and MGL1 (data not shown). Insulin-stimulated glucose uptake was inhibited by M1 conditioned medium but not with conditioned media from M2a or unstimulated macrophages (Fig. 5A). Additionally, when preadipocytes were differentiated in the presence of conditioned media conditions, we observed similarly that whereas M1 conditioned medium effectively inhibited adipogenesis, conditioned media from control and M2a macrophages were permissive to adipogenesis even when added at high concentrations (Fig. 5B and C).
Our studies address several gaps in what is known about ATMs. Our data demonstrate that ATMs in lean mice extensively localize to interstitial spaces between adipocytes and reinforce the concept that ATMs are a fundamental component of adipose tissue even in lean states. These “resident ATMs” uniformly express markers of M2a alternative activation, MGL1 and IL-10. MGL1 is strongly induced by IL-4/IL-13 in macrophages in vitro but is expressed in dermal macrophages independent of IL-4 or IL-13, suggesting that multiple inputs regulate MGL1 expression (15). MGL1 is expressed in a subtype of macrophages involved in granuloma formation, and Mgl1−/− mice demonstrate defective tissue repair after injury, consistent with the proposed role for M2a macrophages in tissue remodeling (19).
MGL1+ ATMs are poised to communicate directly with adipocytes, and our in vitro experiments suggest that macrophage activation state is an important variable in this communication. The permissive nature of M2a-polarized conditioned medium on glucose uptake and adipogenesis is consistent with the notion that in lean settings, MGL1+ ATMs promote an environment favorable for new fat cell formation. The inhibition of these processes by M1 macrophages may instead ultimately promote adipocyte hypertrophy and death, which has been closely linked to ATM accumulation (20).
Obesity does not significantly alter the localization of MGL1+ ATMs in adipose tissue but instead superimposes a new population of M1-polarized MGL1−CCR2+ ATMs on these resident cells. A decrease in the quantity of M2a ATMs combined with the appearance of these “recruited ATMs” ultimately shifts the balance of M1/M2a ATMs to create a more proinflammatory environment in adipose tissue. PKH26 labeling shows that this recruitment is highly spatially restricted to clusters and is consistent with intravital microscopy data that show increased leukocyte trafficking into adipose tissue with obesity (21). We note that it has been reported that isolectin does not label inflammatory clusters of macrophage in the crown-like structures (22). In our experiments, we did not identify heterogeneity in isolectin binding to the clusters because nearly 100% of all clusters are isolectin positive. This may be due to technical differences in the staining approach or differences in the models used (db/db in ref. 22 and diet-induced obesity in our study).
Although it has been suggested that macrophages can be converted between M1 and M2a states (23), our results do not support this notion for ATMs. The restricted distribution of M1 and M2a ATM subtypes instead support a model where obesity induces the specific migration of inflammatory CCR2+Ly6GhiCX3CR1low monocytes from the circulation to ATM clusters. This is similar to what has been proposed to mediate monocyte/macrophage recruitment to atherosclerotic plaques (24,25). Another hypothesis that we considered was that resident MGL1+ ATMs migrate from their interstitial position to the clusters and are subsequently converted to an M1 state. Pulse-labeling ATMs with PKH26 did not support this. If resident ATMs migrate from their interstitial position toward clusters over time, we would expect to observe an increase in PKH26+ ATMs in the clusters after labeling and a gradual replacement of interstitial ATMs with PKH26− ATMs. This is not what we observed. Although it is possible that PKH26 is metabolized differently in M1 and M2a ATMs, which would confound our results, in vitro experiments did not show that M1 activation per se leads to the loss of the PKH26 dye from macrophages.
Our studies suggest that specifically targeting MGL1− ATMs can potentially decrease high-fat diet–induced inflammation in adipose tissue. Blockade of CCR2 has been proposed as one way to accomplish this (2). In this regard, PKH26 labeling showed that Ccr2−/− mice have a decreased rate of accumulation of ATMs in clusters. This is consistent with the evidence that Ccr2−/− mice have few circulating CCR2+7/4+Ly6G− monocytes because of defective migration out of the bone marrow (13). However, ATM cluster formation is not completely abolished in the absence of CCR2, and these ATMs are still CD11c+MGL1−F4/80+. Therefore, it appears that MGL1− ATMs can be recruited to clusters via CCR2-independent signaling pathways, although CCR2 deficiency may limit this process. This may explain some of the conflicting results published on ATMs in obese Ccl2−/− mice (26,27).
Studies have yet to conclusively support or refute the validity of the M1/M2a model in human ATMs. Consistent with our studies, Bourlier et al. (28) identified ATM subtypes in subcutaneous fat with differential expression of the M2a markers CD206 and CD16 and found that adipose tissue expansion correlated with the accumulation of CD206+CD16− ATMs. In contrast, Zeyda et al. (29) failed to identify subsets of ATMs and argued that the ATMs were homogeneous. Both studies failed to identify ATM subtypes with a strict M1 or M2a bias. These studies emphasize that although the identification of M1 and M2a macrophages is fairly mature in mice, it is still unclear how this paradigm applies to human diseases.
Overall, we provide evidence that subtypes of ATMs account for the M2a-to-M1 switch in obesity and can be distinguished by differences in kinetics of accumulation in fat, spatial localization, MGL1 expression, and inflammatory gene expression. This adds to our understanding of ATM biology and provides specific targets for potential intervention. The goal of future studies will be to dissect the molecular signals that govern the migration and recruitment of MGL1+ and MGL1− ATMs and to validate this heterogeneity in human adipose tissue.
Published ahead of print at http://diabetes.diabetesjournals.org on 1 October 2008.
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.org/licenses/by-nc-nd/3.0/ for details.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
C.N.L. has received National Institutes of Health (NIH) Grants DK-078851 and HD-028820. A.R.S. has received NIH Grant DK-060591. This work was additionally support by the Michigan Diabetes Research Training Center Cell and Molecular Biology Core Grant P60-DK-020572 and the Ferrantino Young Investigator Award from the Department of Pediatrics, University of Michigan, to C.N.L.