scRNA-Seq analysis reveals AG immune response to physiological stressors. Macrophages are remarkably diverse immune cells that play roles in maintaining tissue homeostasis and modulating immune responses (26, 27). To determine the role of AG macrophages in regulating stress responses, we performed scRNA-Seq on sorted CD45+ immune cells from AGs of male and female mice following acute or chronic stress exposure. Chronic stress was induced using an atherosclerosis hyperlipidemia model, as these animals exhibit high cholesterol, elevated inflammatory markers, and increased corticosterone production. Overnight cold housing was applied to mimic an acute stress response (Figure 1A). scRNA-Seq data were compared against our prior analysis of AG immune cells at steady state from C57BL/6 mice (14) (Figure 1A).

scRNA-Seq profiling of AG immunity. (A) scRNA-Seq analysis of AG CD45+ sorted immune cells from acute cold–challenged C57BL/6 (B6) (male, n = 16, Hashtag 9; female: n = 15, Hashtag 10) or HFD-fed Ldlr–/– (male: n = 8 male, Hashtag 11; female: n = 13, Hashtag 12) mice. Data were integrated with previously collected control B6 AG immune cells. (B) We generated 19 clusters from all cells using resolution 0.5 shown in UMAP embedding. ILC, innate lymphoid cell; NKT, natural killer T cell; T gd, γδ T cell. (C) SingleR-predicted AG immune populations shown in UMAP embedding. (D) Dot plot showing top 3 enriched features in each of 19 clusters. Dot size and color represent percentage expressed and expression level, respectively. (E) UMAP split by treatment conditions showing cluster representation. (F) Pie chart showing the proportion of monocyte cluster (red) compared to other immune cell clusters (blue), split across treatment conditions. (G) PCA of pseudobulk-transformed scRNA-Seq data. Color indicates treatment condition. Shape represents sex. (H–J) Differentially upregulated (red) or downregulated (blue) genes comparing acute (foreground) stress against steady state (background) (H), chronic against steady state (I), and acute against chronic state (J).

scRNA-Seq data from control, acute, or chronic stress challenge were integrated using Seurat and mapped in principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) dimensions (Figure 1B). SingleR (28) was used for unbiased cell type annotation (Figure 1C). A dominant pool of macrophages and monocytes was identified along with T cells, B cells, ILCs, NK cells, and DCs (Figure 1, B and C). We generated 19 individual clusters, with cluster-defining features showing that clusters 0, 3, 10, 11 (Folr2, Lyz1), 12 (Top2a, Mki67), and 16 constituted the macrophage populations as they additionally expressed complement-associated markers: C1qa, C1qb, and C1qc. This clustering approach also differentiated nonclassical monocyte cluster 5 (Ace, Treml4) from classical monocyte cluster 1 (Ly6c2, Ccr2, Chil3) (Figure 1D and Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.174746DS1). Furthermore, among 2 DC populations, cluster 7 was classified as a plasmacytoid DC cluster expressing Cd209a and Mgl2 whereas cluster 9 showed high coexpression of conventional DC1 markers Xcr1, Clec9a, and Naaa (29, 30). B cells were located in clusters 2 and 17 (Cd19, Cd79a). T cells were selectively distributed in cluster 4 (Cd3d, Cd3e, Cd3g). Cluster 4 also contained NK T cells that were positive for Nkg7. In addition to NK cluster 8 (Nkg7, Gzma, Prf1), a profound ILC subset was identified in cluster 6 and 15 featured for Nrgn, Xcl1, and Thy1 (Figure 1D and Supplemental Figure 1).

We investigated the impact of stress on AG immune cells. Of note, the proportion of monocytes dramatically increased in the chronic atherosclerosis model (Figure 1, E and F). Acute cold exposure induced an increase in the AG monocyte proportion but less robustly compared with the chronic model (Figure 1E), suggesting a higher rate of monocyte infiltration to the stressed AG niche. Immunofluorescence imaging of mouse AG sections revealed macrophages were evenly distributed in all AG zones at steady state. However, under the chronic model, macrophages tended to localize in the upper region of zona fasciculata (ZF) and the region surrounding the AG medulla (Supplemental Figure 2). This observation may suggest a role of macrophages in stress hormone modulation because of their close contact with glucocorticoid-producing cells in the ZF. In addition, our previous study showed that female AG macrophages exhibited a population of MHC-IIlo macrophages that were not present in male AG macrophages (14). We examined the sex-dimorphic phenomenon in stress setting, where macrophage cluster 3 was identified as the MHC-IIlo subset (Figure 1B and Supplemental Figure 3A). However, the proportion of MHC-IIlo macrophages between males and females did not differ following acute or chronic stress challenge (Supplemental Figure 3, A and B). In addition to MHC-II, acute or chronic stimuli did not differentially alter the transcriptional features of male or female immune cells, as PCA grouped samples by stimuli but not sex (Figure 1G).

Next, we assessed the magnitude of transcriptional changes in 8 major immune cell types following stress by quantifying the differentially expressed genes (DEGs). Comparing acute stress against steady state, neutrophils showed the most abundant DEGs, suggesting these cells recognize acute stress and react to it earlier than other immune cells. This observation is consistent with neutrophils being the first-line response in the AG following stress (31–33) (Figure 1H). In assessment of chronic atherosclerosis versus steady state, all major immune populations showed dynamic transcriptional alterations, as the quantity of DEGs was high in these populations (Figure 1I). Interestingly, comparing acute and chronic models, most immune clusters appeared to distribute DEGs beyond 1-log fold-change, suggesting differential responses to acute or chronic challenge models (Figure 1J). ILCs, T, NK, and NK T cells exhibited a proportional decrease following stress stimulations (Supplemental Figure 1B). However, the AG macrophage showed a low rate of differential reprogramming in this analysis (Figure 1J). This may indicate that the mechanism underlying macrophage stress response is conserved between acute and chronic stress states. Collectively, this scRNA-Seq analysis documented the AG immune diversity and highlighted the transcriptional modulations among AG immune cells in response to stress.

Stress promotes AG monocyte recruitment and undermines macrophage viability. In our previous study, we reported that AG macrophages depend on monocyte recruitment to sustain the local macrophage population (14), but the dynamics of AG monocytes during stress has not been investigated. To examine whether AG macrophages are monocyte dependent during stress challenge computationally, we applied pseudotime trajectory analysis to AG monocytes and macrophages. A monocyte-to-macrophage lineage was revealed, and the pattern of trajectory was unchanged by either acute or chronic stress (Figure 2A). This computational approach suggests monocytes sustain AG macrophages during stress responses. In addition, the level of monocyte chemoattractant protein-1 (Ccl2) was substantially upregulated among monocytes and macrophages during both acute and chronic stimulation (Figure 2B), supporting that monocytes may be drawn to the AG more rapidly during stress. To investigate the dynamics of this turnover in mice bearing chronic atherosclerosis burden, we conducted monocyte fate mapping experiments using the CCR2creER R26TdTomato Ldlr–/– strain. After 8 weeks of HFD feeding, tamoxifen (TAM) was administered by oral gavage to label CCR2-expressing classical monocytes. Mice were sacrificed 2 or 5 days after TAM treatment to assess early and late monocyte infiltration to the AG (Figure 2C). It is important to note that B cells, T cells, or neutrophils were not labeled in this model (Supplemental Figure 4, A and B). A greater proportion of AG macrophages were TdTomato+ in mice bearing atherosclerosis at both 2- and 5-day time points compared with their controls, suggesting monocyte recruitment is accelerated following chronic stress (Figure 2D). To investigate the macrophage turnover dynamic under acute challenge, we cold-housed CCR2creER R26TdTomato mice (Figure 2E). Interestingly, the TdTomato+ macrophage proportion was comparable in male mice, albeit a trend was observed after cold stress, whereas there was a significant increase in monocyte recruitment in female mice (Figure 2F). This observation may indicate that male mice are more resistant to acute stress than females, as female AG monocytes exhibited faster turnover, indicated by an increased percentage of TdTomato+ macrophages. However, this sex dimorphism was not observed in atherosclerotic mice. Together, we have demonstrated that mice undergo rapid AG monocyte-to-macrophage differentiation in response to stress.

Stress promotes AG macrophage turnover. (A) Monocyte-macrophage pseudotime trajectory during stress stimulations. (B) Proportion of Ccl2+ monocytes and macrophages. (C) Schematic of fate mapping under chronic stress using control (chow-fed) or atherogenic CCR2creER R26TdTomato Ldlr–/– mice. All mice received 1 dose of TAM and were sacrificed at 2 and 5 days after TAM administration. (D) Percentage of TdTomato+ macrophages (CD11b+CD64+F4/80+) at day 2 and 5 after TAM induction. D2: day 2 after TAM, control (white, CCR2creER R26TdTomato, chow) or atherogenic (black, CCR2creER R26TdTomato Ldlr–/–, HFD) mice, n = 5 in each group. D5: day 5 after TAM, control (white, CCR2creER R26TdTomato, chow, n = 5) or atherogenic (black, CCR2creER R26TdTomato Ldlr–/–, HFD, n = 6) mice. Significance determined by Student’s t test, *P < 0.05. (E) Schematic of fate mapping under acute stress using CCR2creER R26TdTomato mice. Mice fed on chow diet received 1 dose of TAM and cold housing the same day and were housed for 2 days. (F) Percentage of TdTomato+ macrophages after 2 days of cold housing or room temperature (RT). RT (white): male mice housed at RT, n = 6. Cold (white): cold-housed male mice, n = 6. RT (black): female mice housed at RT, n = 7. Cold (black): cold-housed female mice, n = 7. Significance determined by Student’s t test, ***P < 0.001. (G) Absolute AG macrophage (CD11b+CD64+MerTK+F4/80+) number at steady state or stress setting. Steady: B6 mice, n = 22. Ldlr–/–: chow-fed Ldlr–/– mice, n = 3. Cold: B6 mice cold-housed for 2 days, n = 9. Ldlr–/– HFD: Ldlr–/– mice fed 8 weeks of HFD, n = 14. Significance determined by ANOVA, *P < 0.05, ****P < 0.0001. (H) Flow cytometry measurement of AG macrophage (CD11b+CD64+MerTK+F4/80+) caspase-3 during chronic stress. Steady: B6 mice, n = 8. HFD: Ldlr–/– mice fed 8 weeks of HFD, n = 8. Significance determined by Student’s t test, **P < 0.005. Panels D and F–H are presented as mean ± SEM.

Increased monocyte recruitment would logically lead to an expanded pool of macrophages if cells persisted in tissue. To test this, we quantified AG macrophages in unstressed and acutely or chronically stimulated mice. Interestingly, the number of AG macrophages decreased in both stress settings (Figure 2G). Compared with cold stressed mice, HFD-fed atherogenic mice exhibited a more dramatic macrophage loss (Figure 2G). No sex differences were observed in these experiments. In addition, the loss of AG macrophages was not a defect of Ldlr deficiency, as unstressed Ldlr–/– mice exhibited comparable levels of macrophage numbers to C57BL/6 mice (Figure 2G). We next measured active caspase-3 to reflect apoptosis induction among AG macrophages (Figure 2H and Supplemental Figure 4C). AG macrophages from atherogenic mice showed significantly higher caspase-3 activity than controls, suggesting these cells undergo rapid apoptosis under chronic stress challenge. To test whether the localization of AG macrophage cell death was associated with regions of stress hormone production, we used a TUNEL staining assay to identify apoptotic cells in the chronically stressed mice’s AGs. As expected, we observed more dying macrophages in the ZF compared with in controls (Supplemental Figure 5A). Since stress hormones are produced in the ZF, and glucocorticoid has been shown to induce apoptosis in lymphocytes, we hypothesized that it has a similar effect on macrophages (34). Indeed, macrophages stimulated with corticosterone drastically promoted cell death (Supplemental Figure 5B). In conclusion, we interpreted the increased monocyte recruitment as an action of the AG to counteract the accelerated loss of AG macrophages because of cell death.

Classically activated AG macrophages arise during stress. To better understand cellular heterogeneity, AG macrophages were separated from the remaining scRNA-Seq data and reclustered. Five AG macrophage subpopulations were generated and presented by UMAP (Figure 3A). AG macrophage subcluster 0 was high for Ccnd1, Mgl2, and Mmp12 and thus identified as an extracellular modeling subset (Figure 3B). Subcluster 1 was named as a pro-inflammatory population because of its feature for classical activation markers, including Il1b and chemokines Cxcl2, Ccl3, and Ccl4. Subcluster 2 had high expression of Lpl and Cd9, suggesting an LAM phenotype. Subcluster 3 was recognized for high S100a10, S100a6, and Slc40a1 expression and appeared to be the MHC-IIlo population (Figure 3, B and C) that we documented in our previous study. Last, subcluster 4 could be characterized by Wfdc1 expression and may suggest an alternatively activated subset (35) (Figure 3B).

scRNA-Seq reveals AG macrophage classical activation by stress stimulation. (A) SingleR-predicted macrophages from all cells were isolated, reclustered using resolution 0.6, and shown in UMAP embedding. (B) Dot plot showing top 7 enriched DEGs in macrophage subclusters. Color represents expression level. Dot size represents percentage of cells that express the feature. (C) Violin plot showing the H2-Ab1 expression in macrophage subclusters. (D) Proportion of macrophage subclusters in each stress condition. (E) Density plot showing cluster density in each stress condition. (F) Density plot showing integrated module score of Tnf, Il1b, and Cxcl2 in UMAP space. (G) Violin plot showing expression of integrated Ccl2, Ccl3, and Ccl4 in each macrophage subcluster, split by treatment condition. (H) GSEA pathway analysis showing TNF or LPS response–related (inflammatory) pathway among top enriched pathways. Red-colored strands represent inflammatory pathways. Grayed strands are all other pathways.

We next investigated the dynamic nature of AG macrophage heterogeneity that might be influenced by stress. Interestingly, AG macrophage subcluster 1 showed expansion following both acute and chronic stimulation. In opposition, subcluster 4 exhibited a proportional decrease (Figure 3, D and E), suggesting an enhanced macrophage polarization toward classical activation driven by stress stimuli. Next, a density plot was used to show the localization and expression level of pro-inflammatory genes Il1b, Tnf, and Cxcl2 in UMAP embedding, where these features colocalized with AG macrophage subcluster 1 (Figure 3F). As noted previously (36, 37), macrophage Il-1β can be released through gasdermin-D–mediated (Gsdmd-mediated) pore formation and promote inflammasome-driven pyroptosis. We quantified Gsdmd expression among AG macrophages and found the proportion of cells that express Gsdmd was increased by nearly 3-fold (Supplemental Figure 5C). Based on these data, we propose the cause for AG macrophage loss is likely correlated with inflammasome activation because of stress hormone stimulation.

In addition to Il1b, chemokines that drive monocyte infiltration, including Ccl2, Ccl3, and Ccl4 (38), constitute a part of the pro-inflammatory signatures upregulated in AG macrophage subcluster 1 (Figure 3G). This may be evidence that subcluster 1 contributes to accelerated macrophage turnover. Next, we performed pathway analysis integrating sex, stress condition, and subclusters to determine whether the pro-inflammatory response possesses sex-dimorphic or stress-related features. We defined gene set enrichment analysis (GSEA) pathways that involve TNF signaling and response to LPS as “inflammatory” and all other pathways as “else.” Consistent with prior reports (39), all AG macrophage subclusters showed an upregulation of inflammatory pathways compared with their steady-state counterparts. In line with the PCA results (Figure 1G), GSEA pathways did not show sex-specific grouping following stress responses (Figure 3H). As expected, most inflammatory pathways were classified into either acute or chronic states, indicating a pro-inflammatory skewing of AG macrophages during stress. Collectively, this AG macrophage subclustering and pathway classification approach supported the heterogeneity and particularly emphasized the pro-inflammatory polarization of AG macrophages.

AG macrophages exhibit a lipid-associated phenotype under atherogenic conditions. Based on the differential expression analysis on AG macrophage subpopulations, AG macrophage subcluster 2 was enriched for Lpl and Cd9 expression (Figure 3B). These features are widely recognized as LAM signatures (40–42). Further DEG analysis showed that AG macrophage subcluster 2 also upregulated Mmp12 and Trem2 (Figure 4A), genes highly expressed by atherosclerotic plaque–associated foamy macrophages (40, 43). Macrophage Trem2 deletion improves atherosclerosis outcome by limiting foamy macrophage formation (44). We performed an enrichment analysis by overlaying foamy macrophage gene signatures (45) in a gene background generated by comparing subcluster 2 and other AG macrophage subclusters. A correlation between foamy macrophages and subcluster 2 was established, suggesting this cluster may be an LAM population resembling macrophages found in other chronic inflammatory settings (Supplemental Figure 6) (41, 46–48). By overlaying the lipid-associated genes in UMAP, we found that although subcluster 2 was not the only population that exhibited these individual features (Cd9, Mmp12, Lpl, Trem2, Cd36), the merged expression of these markers selectively localized in subcluster 2 (Figure 4B).

Lipid-associated AG macrophages arise after stress stimulation. (A) Volcano plot showing DEGs generated by comparing macrophage subcluster 2 against all other macrophage populations. (B) Highlighted lipid-associated genes shown in UMAP space. (C) Lipid granules (Bodipy, green) highlighted by immunofluorescence staining. Macrophages were stained red (CD68). Nuclei were labeled by DAPI (blue). (D) Flow cytometry quantification of lipid deposition in atherosclerotic or control AG macrophages. Chow: WT mice fed on chow diet, n = 5. HFD: Ldlr–/– mice fed on 8 weeks of HFD diet, n = 10. Significance determined by Student’s t test, **P < 0.005. (E) Electron microscopy showing AG LAM from atherosclerotic Ldlr–/– mouse fed on 8 weeks of HFD. Image size, 30 µm × 27 µm. (F) Peritoneal macrophages from 2 WT mice harvested, combined, and cultured in 12-well plate with or without glucocorticoid (CORT) overnight. Data collected from 2 experiments, normalized, and merged. Control: replicates cultured in media (DMEM with 5% FBS, 1% Penicillin-Streptomycin, 1% HEPES, 1% l-glutamine), n = 5 replicates. CORT: replicates cultured in CORT-conditioned (50 ng/mL) media, n = 6 replicates. Significance determined by Student’s t test. (G) Lipid-related GSEA pathways associated with AG macrophage subcluster 2 (red) or other macrophages (blue). Colored triangles represent pathway names containing keyword “Lipid.” NES, normalized enrichment score. (H) Quantification of lipid-associated pathways in macrophage subcluster 2. Panels D and F are presented as mean ± SEM.

Next, the presence of the AG LAMs was validated experimentally by imaging and flow cytometry. Immunofluorescence imaging revealed deposition of lipid in Ldlr–/– AG macrophages following 8 weeks of HFD feeding (Figure 4C). To quantify macrophage lipid content using flow cytometry, AG cells were stained with neutral lipid dye, Bodipy. Compared with C57BL/6 macrophages, AG macrophages from atherosclerotic mice showed significantly higher lipid content (Figure 4D). Serial block face scanning electron microscopy (SBFSEM) further supported the presence of LAM in the adrenal cortex and in close localization to hormone-producing cells (Figure 4E). We hypothesized that the lipid phenotype observed among AG macrophages was influenced by stress hormones, such as glucocorticoids. To test this, murine peritoneal macrophages were harvested from C57BL/6 mice and cultured in either DMEM (control) or glucocorticoid-conditioned medium overnight. Macrophages under glucocorticoid stimulation showed increased lipid deposition compared with controls, indicated by a 0.5-fold increase of Bodipy staining in cells cultured with glucocorticoids (Figure 4F). To profile LAM phenotype computationally, pathway analysis was performed. Lipid-associated pathways were identified using lipid-associated keywords. As expected, markedly more lipid pathways were classified into macrophage subcluster 2 than other subclusters (Figure 4G). In fact, the majority of statistically significant pathways in subcluster 2 were lipid associated (Figure 4H). In summary, using multiple experimental and computational approaches, a likely unique LAM population was found to develop in the AG following acute or chronic stress models.

Macrophage Trem2 regulates steroidogenesis via repression on steroidogenic acute regulatory protein. Trem2 is associated with alternative macrophage activation and plays a role in lipid balance during disease (22, 24, 49). In the AG, Trem2 is broadly expressed by macrophages (Figure 5A). We found 50% of AG macrophages were Trem2 positive at steady state, and this proportion increased to 75% in acutely or chronically stressed mice (Figure 5B). Of the Trem2+ macrophages, subcluster 2 constituted the majority of Trem2hi macrophages at steady state. However, Trem2hi macrophages were composed of clusters 1 and 2 in acute and chronic stress states (Figure 5C). This dynamic change of macrophage Trem2 may indicate a role of this molecule during stress.

Macrophage Trem2 suppresses steroidogenesis. (A) Feature plot showing Trem2 expression in macrophages across treatment groups. (B) Proportion of Trem2-expressing AG macrophages in treatment groups. (C) Histogram showing Trem2 expression distribution in macrophage subclusters. (D) Serum corticosterone in WT or Trem2–/– mice after 48 hours of cold housing. WT, n = 5. Trem2–/–, n = 6. Significance determined by Student’s t test, *P < 0.05. (E) Experimental design of atherosclerosis induction in Trem2-deficient mice fed TAM-HFD for 8 weeks. (F) Serum corticosterone in WT (n = 8), Trem2–/– (n = 7), CX3CR1creER Trem2WT Ldlr–/– (Trem2WT Athero, n = 9), or CX3CR1creER Trem2fl Ldlr–/– (MacrophageTrem2Δ Athero, n = 4) mice. Significance determined by ANOVA, **P < 0.005, ****P < 0.0001. (G) Flow cytometry quantification of AG macrophage lipid content (Bodipy). Trem2WT: atherosclerotic CX3CR1creER Trem2WT Ldlr–/– mice, n = 5. MacrophageTrem2Δ: atherosclerotic CX3CR1creER Trem2fl Ldlr–/– mice, n = 6. Significance determined by Student’s t test, **P < 0.005. (H) Serum corticosterone in CCR2creER Trem2WT mice or CCR2creER Trem2fl mice. n = 8 for each group. Significance determined by Student’s t test, *P < 0.05. (I) In vitro BV2-Y1 coculture experiment schematic. Y1 cells were assessed for StAR expression and normalized to control. n = 6 in each group, except n = 5 for Y1 ACTH–. Significance determined by Student’s t test, **P < 0.005. (J) Corticosterone concentration in coculture supernatant. Blue: Y1 cocultured with WT BV2. Green: Y1 cocultured with Trem2–/– BV2. n = 6 in each group, except n = 5 for Y1 + BV2WT ACTH–. Significance determined by 1-way ANOVA, *P < 0.05, **P < 0.005, ****P < 0.0001. Panels D and F–J are presented as mean ± SEM.

To investigate Trem2 in macrophage stress response, Trem2–/– mice were cold-housed for 48 hours, and serum corticosterone was assessed using ELISA. Corticosterone was significantly higher in Trem2-deficient mice after acute cold stress (Figure 5D). However, germline Trem2–/– mice possess defects including disrupted lipid balance and inflammatory response, making Trem2–/– mouse data difficult to interpret (41, 50, 51). To overcome this, we generated a TAM-inducible Trem2-floxed system crossed to an atherogenic background that allows targeting of AG macrophages, CX3CR1creER Trem2fl Ldlr–/– (macrophage Trem2Δ). These mice were fed a TAM-enriched high-fat diet (TAM-HFD) for 8 weeks to induce atherosclerosis while deleting Trem2 in a macrophage-restricted manner (Figure 5E). Despite no changes in body weight or serum cholesterol levels (44), macrophage Trem2 deletion led to significantly higher serum corticosterone compared with controls that were fed the same TAM-HFD (Figure 5F), suggesting that macrophage Trem2 may regulate stress hormone production. Interestingly, Trem2-deficient mice also showed reduced lipid deposition in AG macrophages compared with controls (Figure 5G). Overall, this experiment suggested that Trem2 is not only antisteroidogenic but also serves a role in macrophage lipid balance following stress.

Since the majority of AG macrophages were replaced by monocytes following stress responses, we hypothesized that Trem2 deletion in monocyte-derived cells would undermine the antistress capability of AG macrophages. To test this, CCR2creERTrem2fl (MoTrem2Δ) mice were given TAM every 3 days for 2 weeks prior to cold challenge. Importantly, this model will have reduced “off-target” deletion, such as in microglia or other CX3CR1+ macrophage subsets. Mice were then cold-housed for 48 hours (Figure 5H). Trem2-deficient mice also showed higher serum corticosterone compared with control animals after cold challenge (Figure 5H). Together, these data reveal a protective role of macrophage Trem2 that regulates acute and chronic stress by direct participation in stress hormone release.

Next, we tested the role of macrophages in the regulation of adrenal cortex stress response, so we developed an in vitro coculture approach using adrenal cortex cell line (Y1) with a macrophage cell line (BV2). Y1 cells were stimulated with ACTH, and steroidogenic acute regulatory protein (StAR) in Y1 cells was quantified using flow cytometry to reflect the rate of steroidogenesis (Figure 5I and Supplemental Figure 7) (52). StAR was detectable in Y1 cells even without the stimulation of ACTH, but it was further increased by ACTH stimulation (Figure 5I, red, and Supplemental Figure 7). Interestingly, the addition of WT BV2 potently inhibited StAR induction by ACTH in Y1 cells (Figure 5I, blue). The corticosterone level in these cultures mildly increased following ACTH stimulation, which likely represented a healthy and normal stress response (Figure 5J, blue). However, Trem2–/– BV2 cells failed to restrict StAR upregulation after ACTH stimulation (Figure 5I, green). The corticosterone concentration from these samples was drastically higher than Y1/WT BV2 cultures (Figure 5J, green), suggesting macrophage Trem2 is required to regulate steroidogenesis of Y1 cells. Collectively, using in vivo and in vitro Trem2 deletion approaches, we concluded that macrophages employ a Trem2-mediated mechanism to suppress stress hormone synthesis.

AG macrophages restrict steroidogenesis via a Trem2/TGF-β axis. To investigate the molecular mechanism underlying the antisteroidogenic effects of Trem2, we first examined genes involved in the Trem2/StAR network from a published data set of whole AG cells (53). Linear correlation between the mouse genome and Trem2 or StAR was built to infer transcriptional targets involved in networking between Trem2 and StAR (Figure 6A). Genes positively associated with Trem2 but negatively correlated with StAR were anticipated to mediate the antisteroidogenic effect of Trem2. We observed strong correlation between Trem2 and macrophage markers, Ctss, C1qa, C1qb, and C1qc, and the Trem2 adaptor protein Tyrobp (encoding DAP12) (Figure 6, A and B). Among the genes that positively associated with Trem2 and negatively associated with StAR, Tgfb1 is a well-established immunomodulatory cytokine in many disease systems (54–56) (Figure 6, A and B).

Trem2 modulates steroidogenesis through TGF-β. (A) Linear correlation analysis showing Trem2- and Star-associated features in control AG macrophages; each dot represents a gene. (B) P values of highlighted genes in A. (C) Whole AG RNA-Seq: heatmap showing Trem2, TGF-β, LTBP family, and Star in control or Trem2-deficient (CX3CR1creER Trem2fl Ldlr–/–) atherosclerotic mice. Control, n = 5; Trem2-deficient, n = 4. (D) Enrichment plot showing steroid hormone biosynthesis GSEA pathways comparing WT against Trem2-deficient atherosclerotic mouse AGs. (E) Enrichment plot showing TGF-β signaling pathways comparing WT against CX3CR1creER Trem2fl Ldlr–/– atherosclerotic mice. (F) Percentage of LAP+ WT or Trem2–/– BV2 at baseline. WT BV2, n = 5;Trem2–/– BV2, n = 5 replicates. Significance determined by Student’s t test, ****P < 0.0001. (G) Histogram showing LAP MFI. FMO, fluorescence minus one. (H) Percentage of LAP+ BV2 cells, cocultured with Y1. WT BV2, n = 6 replicates. Trem2–/– BV2, n = 6 replicates. Significance determined by Student’s t test, **P < 0.005. (I) Percentage of LAP+ BV2 cells, cocultured with Y1, with ACTH. WT BV2, n = 6 replicates. Trem2–/– BV2, n = 6 replicates. Significance determined by Student’s t test, **P < 0.005. (J) Concentration of TGF-β in cell culture supernatant. WT (red), n = 6. Trem2–/– (green), ACTH+, n = 5, ACTH–, n = 6. Significance determined by Student’s t test, **P < 0.005. (K) Immunofluorescence staining of LTBP4, CD68, and DAPI in CX3CR1creER Trem2WT Ldlr–/– or CX3CR1creER Trem2fl Ldlr–/– mice fed TAM-HFD for 12 weeks. (L) Quantification of LTBP4 by MFI of red pixels, n = 6 for each group. Significance determined by Student’s t test, ***P < 0.001. (M) Serum TGF-β in CX3CR1creER Trem2WT Ldlr–/– (Trem2WT♂ n = 3, or Trem2WT ♀ n = 4) or CX3CR1creER Trem2fl Ldlr–/– (MΦTrem2Δ♂ n = 4, or MΦTrem2Δ♀ n = 6) mice after 16 weeks of TAM-HFD. Significance determined by ANOVA, *P < 0.05, **P < 0.005. Panels F, H, I, J, L, and M are presented as mean ± SEM.

TGF-β is secreted as a multiprotein latent complex, covalently bound to the latent TGF-β binding protein (LTBP) adaptor (57). LTBP complex interacts with integrins in the extracellular matrix to release TGF-β protein for signaling through TGF-β receptors (58). In the AG, all 3 isoforms (β1, β2, and β3) of TGF-β and LTBPs 1–4 are detected, and interestingly, AG macrophages are the only immune cell type that produces TGF-β in the adrenal niche (53). Using in vitro models, TGF-β has been shown to broadly limit steroidogenesis in human hormone-producing cells, including granulosa, trophoblasts, and adrenal cortical cells (59–61). It has also been shown that TGF-β signaling can directly impact the expression of StAR (62). Therefore, we hypothesized that AG macrophage Trem2 inhibits the rate of steroidogenesis through TGF-β and StAR. To test this hypothesis, bulk RNA sequencing of whole AGs harvested from Trem2-deficient and Trem2-sufficient atherosclerotic mice was performed (Figure 6C). In line with the gene correlation analysis, Trem2 deficiency led to substantial downregulation of the TGF-β family, including Tgfb1, Tgfb2, Tgfb3, and Tgfbi. Interesting, Ltbp1, Ltbp2, Ltbp3, and Ltbp4 were also decreased at the transcriptional level in the Trem2-deficient setting (63) (Figure 6C). Pathway analysis derived from this sequencing experiment also supported the notion that the TGF-β pathway was downregulated, while steroid hormone production was upregulated in Trem2-deficient mice (Figure 6, D and E).

To test the Trem2/TGF-β correlation experimentally, the Y1 macrophage in vitro coculturing model was used (Figure 5I). Since both small and large latent forms are required for TGF-β signaling (64–66), we measured the latency-associated peptide (LAP) as a reflection of TGF-β production. In the BV2-only setting, LAP was decreased in Trem2–/– BV2 at the basal level compared with WT BV2 (Figure 6, F and G). After coculturing with Y1 cells without ACTH stimulation, both Trem2–/– and WT BV2s exhibited elevation of LAP compared with BV2-only culture. However, the pattern of reduced LAP in the Trem2–/– setting persisted (Figure 6H). Notably, after ACTH stimulation on cocultured cells, Trem2 WT BV2 further increased LAP to cope with enhanced steroidogenesis, but Trem2–/– BV2s failed to do so (Figure 6I). Consistent with LAP, free TGF-β quantification in supernatant revealed that WT BV2s enhanced TGF-β production in response to ACTH (Figure 6J, red), but Trem2–/– BV2s lost the ability to secrete TGF-β (Figure 6J, green).

Next, we sought to validate the in vitro findings using our in vivo atherosclerosis model. Immunofluorescence staining targeting LTBP4 was measured as a readout for TGF-β. Strikingly, LTBP was readily detectable in control samples but clearly diminished in Trem2-deficient mouse AG after atherosclerosis (Figure 6, K and L). Likewise, LTBP4 level was reduced in cold-housed Trem2-deficient mice compared with control (Supplemental Figure 8, A, B, and H). We then measured both serum and local adrenal TGF-β levels. In line with the LTBP4 data, TGF-β from Trem2-deficient mice was reduced in both serum and AG in chronically (Figure 6M and Supplemental Figure 8C) or acutely (Supplemental Figure 8, D and E) stressed mice. Notably, we did not observe any sex dimorphism for TGF-β production (Figure 6M and Supplemental Figure 8F). According to previous sequencing studies (67, 68) in Alzheimer’s disease, apolipoprotein E (Apoe) was suggested to downregulate TGF-β in microglia, but whether this applies to other disease contexts is unknown. Thus, we cold-challenged Apoe–/– mice and measured the LTBP4 accumulation in the AG of these mice. Surprisingly, unlike the Trem2-deficient mice, the Apoe–/– mice did not show signs of attenuated LTBP4 (Supplemental Figure 8, G and H).

TGF-β signaling blockage enhances steroidogenesis by StAR upregulation. The regulatory function of TGF-β on steroidogenesis and StAR expression has been recognized previously (61, 62, 69), but the cellular origin of TGF-β production is unclear. Our scRNA-Seq approach showed that the primary immune cells in the AG producing TGF-β are macrophages (Figure 7, A and B). To validate the steroidogenesis-suppressive function of TGF-β, we performed a TGF-β–blocking experiment in vitro. Y1 cells were cocultured with WT BV2 cells, then exposed to anti–TGF-β antibody and ACTH (Figure 7C). Interestingly, anti–TGF-β–treated cultures exhibited significant StAR upregulation to a level that resembled StAR in Y1 cultured with Trem2–/– BV2s (Figure 7D). Since macrophages are the only source for TGF-β in this coculture system, we concluded that macrophage-secreted TGF-β is sufficient to mediate the antisteroidogenic effects of Trem2.

TGF-β inhibition promotes steroidogenesis. (A) Trem2 and Tgfb1 expression level shown in UMAP. (B) Quantification of Tgfb1+ immune cells. Tgfb1 positivity determined by normalized Tgfb1 expression > 0. (C) Schematic of TGF-β neutralization in in vitro Y1-WT BV2 coculture. WT BV2 and Y1 were cultured for 3 hours. αTGF-β neutralizing antibody–conditioned (0.3%) medium was reconstituted with ACTH. Cells were further cultured for 3 hours and harvested for flow cytometry. (D) Normalized (to vehicle control) percentage of StAR+ Y1 cells cultured with or without αTGF-β antibody. n = 5–6 replicates. Significance determined by ANOVA, ***P < 0.001, ****P < 0.0001. (E) Schematic of TGF-βR antagonism by LY573636 in B6 mice. Mice received 150 μL (20 mg/mL in DMSO) LY573636 1 day before cold challenge and daily during cold housing. (F) Immunofluorescence staining showing StAR expression in AG cortex. (G) Mean intensity of StAR (green pixels) in vehicle or LY573636-treated animals. Green: DMSO vehicle control, n = 3. Red: LY573636 treated, n = 3. Significance determined by Student’s t test, **P < 0.005. (H) ELISA analysis of corticosterone concentration in AG tissue. Green: DMSO vehicle control, n = 5. Red: LY573636 treated, n = 4. Data were normalized to vehicle control. Significance determined by Student’s t test, **P < 0.005.

Although TGF-β has been shown to reduce steroidogenesis, such mechanisms have not been extended in vivo to our knowledge. TGF-β receptor (TGF-βR) blockage using LY573636 (tasisulam sodium) (70, 71) was used to inhibit TGF-β signaling in vivo. Mice were given LY573636 daily and received cold challenge for 2 days (Figure 7E). StAR was quantified using immunofluorescence staining and confocal imaging. Consistent with the in vitro results, antagonism of TGF-βR led to enhanced StAR expression compared with controls (Figure 7, F and G). Furthermore, local AG corticosterone level in LY573636-treated mice was also increased (Figure 7H), verifying the protective role of TGF-β signaling in stress regulation. Collectively, in vitro and in vivo TGF-β–blocking approaches supported that TGF-β is used by AG macrophages to restrict steroidogenesis.

Glucocorticoid receptor enhances the Trem2/TGF-β axis to reinforce macrophage steroidogenic regulation. We demonstrated that Trem2 and TGF-β regulate adrenal steroidogenesis by limiting StAR. However, the mechanism by which macrophages sense stress hormones remains unclear. Since glucocorticoid signals through glucocorticoid receptor (GR) to mediate immunoregulatory functions, we hypothesized that glucocorticoid/GR interaction may reinforce the antisteroidogenic Trem2/TGF-β pathway. Notably, according to a recent RNA-sequencing study, GR–/– macrophages exhibited substantial downregulation of Trem2 at baseline versus WT control (72) (Supplemental Figure 9A), suggesting GR may drive Trem2 transcription in AG macrophages. To test this hypothesis experimentally, RU486 (mifepristone) was used to antagonize GR in BV2s. In addition, we used dexamethasone, a synthesized glucocorticoid, to mimic stress stimulation in vitro. WT or Trem2–/– BV2s were cultured in glucocorticoid and RU486 overnight as indicated (Supplemental Figure 9B). Significant Trem2 upregulation was observed following administration of glucocorticoid in the absence of RU486. In contrast, RU486-treated WT BV2s showed no change after glucocorticoid stimulation (Supplemental Figure 9C). Trem2–/– BV2s did not respond to either glucocorticoid or RU486 stimulation (Supplemental Figure 9C). This observation supported the idea that GR activation indeed promotes macrophage Trem2 expression. Furthermore, LAP was assessed to reflect the influence of glucocorticoid and GR antagonism on TGF-β production. Interestingly, glucocorticoid stimulation alone was able to dramatically drive LAP in WT BV2s (Supplemental Figure 9D). However, this pattern was inhibited by GR antagonism in WT cells. Regarding Trem2–/– BV2s, there was no evidence that these cells can efficiently produce LAP (Supplemental Figure 9D), likely suggesting Trem2 is required for effective TGF-β production. In addition, RU486, but not glucocorticoid, seemed to inhibit the survival of BV2s, as cell death, indicated by Ghost Dye staining and flow cytometry, increased by 1-fold in treated cells (Supplemental Figure 9, E and F). Together, these data indicate that macrophages are capable of sensing stress through GR to reinforce the Trem2/TGF-β pathway.

It is well established that Trem2 modulates macrophage response through Syk-dependent pathways (73, 74), but whether Trem2/TGF-β interaction is Syk mediated remains unclear. We first examined this possibility using a published RNA-Seq data set of human macrophages (75). Interestingly, Syk inhibitor–treated macrophages showed a drastic decrease of the TGF-β and LTBP families following LPS-induced stress stimulation (Supplemental Figure 10A). To test this experimentally, we treated cultured peritoneal macrophages with Syk inhibitor, BAY 61-3606, overnight. In line with the RNA-sequencing data, LAP was significantly downregulated in BAY 61-3606–treated macrophages (Supplemental Figure 10B). In addition, we were also interested in the impact of TGF-β itself on TGF-βR signaling, which signals through Smad-dependent or -independent pathways (76). We performed immunofluorescence imaging for phosphorylated SMAD2 (p-SMAD2) in acute or chronically stressed samples. We observed a dramatic reduction in total p-SMAD2 in both settings when Trem2 was deleted (Supplemental Figure 10, C–F). These data suggest that Trem2 is required for stress-mediated TGF-βR signaling in the AG.