Overweight and obesity induce Toll-like receptor 2-induced Type I IFN signaling

Abstract

Toll-like receptor 2 (TLR2) is an innate immune receptor linked to obesity primarily via NF-κB activation. Using a mouse model of overnutrition and in vitro TLR2 stimulation of human peripheral blood mononuclear cells, we show that lipids, advanced glycation end products, and low-density lipoproteins extend TLR2 signaling beyond NF-κB to promote Type I IFN production and signaling, establishing the relevance of this pathway to human obesity. This response was abolished by pharmacologic inhibition of the receptor for advanced glycation end products, which recognizes glycated proteins and lipids. In vivo dietary reversal, metformin, and tirzepatide each reduced diet-induced inflammation, but differed in their metabolic effects: dietary reversal reduced weight gain and LDL, tirzepatide reduced weight without lowering LDL, and metformin exerted anti-inflammatory effects independent of weight or LDL. These findings identify TLR2-Type I IFN signaling as a feature of diet-induced inflammation and highlight the diverse immunomodulatory effects of weight-management therapies.

Introduction

The global rise in obesity and the widespread adoption of glucagon-like peptide-1 receptor agonists (GLP-1RAs) underscore the urgent need to understand the biology of obesity. Many of its complications stem from chronic inflammation, which disrupts organ function, remodels the extracellular matrix, and alters crosstalk between adipose tissue and immune cells¹. Adipose tissue is now recognized as a central immunometabolic organ in obesity, functioning not only as a site of energy storage but also as a tissue enriched in immune cells, cytokines, and extracellular mediators that shape systemic inflammation². Toll-like receptors (TLRs) are innate immune receptors that recognize pathogen-associated molecular patterns and initiate inflammatory responses³. TLRs are expressed on both immune cells and adipocytes, positioning adipose tissue as a relevant site for TLR-driven inflammatory signaling during overnutrition. Although TLRs are central to the pathogenesis of obesity, the effects of obesity on TLR signaling remain incompletely defined.

Among the TLR family members, TLR2 has emerged as a key mediator of metabolic disease. TLR2 forms heterodimers with TLR1 or TLR6 and is broadly expressed by neutrophils, antigen-presenting cells, memory T cells, epithelial cells, and microglia. Mice lacking TLR2 are protected from high-fat diet-induced adiposity, insulin resistance, and atherosclerosis^4–6. Through NF-κB activation, TLR2 drives production of the proinflammatory cytokines characteristic of obesity. TLR2 can also induce Type I interferon (IFN) signaling in human and mouse monocytes in vitro and in a mouse model of infection with vaccinia virus or listeria^7–9. However, whether TLR2-Type I IFN signaling is engaged in human metabolic disease remains unknown.

Obesity is associated with an accumulation of extracellular lipids and glycated molecules that can engage innate immune signaling pathways, particularly through the receptor for advanced glycation end products (RAGE), a receptor known to synergize with multiple TLRs¹⁰. In adipose tissue, these extracellular cues act within a cellular environment enriched in adipocytes and tissue-resident inflammatory immune cells, making this compartment particularly relevant for immunometabolic signaling during overnutrition. Based on this scientific premise, we tested the central hypothesis that overnutrition activates a TLR2-Type I IFN pathway that contributes to diet-induced inflammation by increasing the production and sensing of Type I IFNs. To address this hypothesis, we used a mouse model of overnutrition with in vivo pharmacologic interventions complemented with in vitro studies of human peripheral blood mononuclear cells (PBMCs) stimulated with TLR2 agonists and obesity-associated extracellular factors, including lipids, advanced glycation end products, and low-density lipoproteins (LDLs). Using these approaches, we identify TLR2 as a sensor of extracellular lipid changes, delineate mechanisms regulating TLR2-Type I IFN signaling in mice and humans, and assess how dietary changes, metformin, and tirzepatide modulate this pathway.

Results

TLR2 heterodimers detect modest increases in the lipid environment

To define how overnutrition affects TLR signaling, we first identified the TLRs responsive to modest increases in the lipid environment. Human PBMCs were cultured in media supplemented with a chemically defined lipid mixture that models high-fat diet exposure^11–14, containing soluble cholesterol, polyunsaturated fatty acids (linoleic, linolenic, arachidonic), monounsaturated fatty acids (oleic), and saturated fatty acids (myristic, palmitic, stearic). Although lipid concentrations were more than 10-fold higher than in fetal bovine serum, they remained 10–100-fold lower than in human plasma, modeling a modest increase in lipid exposure^11,15,16. Among the TLR agonists, only TLR2 agonists synergized with the lipid mixture, resulting in increased secretion of IL-6 beyond that elicited by either stimulus alone (Fig. 1). In contrast, increased lipid concentrations did not enhance the secretion of TNF-α, IL-8, IL-10, IL-18, or IFN-α downstream of TLR2 or any other TLR (Supplementary Fig. 1). Thus, increased lipid availability does not broadly amplify TLR-induced cytokine production, but instead selectively augments a restricted TLR2-driven inflammatory program.

Fig. 1. TLR2 is highly sensitive to small changes in the lipid environment.

A Peripheral blood mononuclear cells from lean control individuals were stimulated for 24 h with the indicated TLR agonists in the presence or absence of a chemically defined lipid mixture (n = 17-20/group). Median values with interquartile ranges are shown. Pam3CSK4, Pam3CysSerLys4; FSL-1, Pam2CGDPKHPKSF; LPS, lipopolysaccharide; Imiquimod; ssRNA, single-stranded RNA; CpG, synthetic single-stranded DNA with CpG dinucleotides. Median values with interquartile range shown, **p < 0.01, ***p < 0.001, Mann–Whitney U test, adjusted for multiple comparisons with the Holm-Šídák method.

A short-term Western diet (WD) increases TLR2 signaling in vivo

To determine whether the lipid-mediated potentiation of TLR2 signaling extends beyond the effects elicited by 24-h in vitro lipid supplementation, we next examined TLR2 signaling using a mouse model of overweight in which mice are fed a WD for 7 days¹⁷. This model is more representative of the human experience of gradual weight gain than conventional models of obesity that rely on continuous WD feeding for 22 weeks. We previously showed that this 7-day diet (henceforth referred to as the short WD) induces ~6% weight gain and increased circulating cholesterol, triglycerides, insulin, and leptin, but does not cause hepatic steatosis¹⁷. In contrast, chronic obesity models capture later stages of metabolic disease characterized by sustained adipose expansion, end-organ injury, and broader immune remodeling. We therefore used this short-term model to identify mechanisms by which overnutrition enhances TLR2 responsiveness before the onset of advanced obesity-associated pathology. Before injection of TLR2 agonists, mice fed the standard and short-term WD had comparable numbers of peritoneal monocytes and neutrophils, with no differences in TLR2 expression on either cell population (Supplementary Fig. 2A, B). In vivo administration of intraperitoneal FSL-1 (TLR2/6 agonist) or Pam3CSK4 (TLR1/2 agonist) produced higher circulating IL-6, TNF-α, and IL-12p40 in short WD–fed mice than controls (Fig. 2B, Supplementary Fig. 2C). Transcriptomic analysis of peritoneal immune cells after intraperitoneal injection of FSL-1 revealed enrichment of differentially expressed genes in NF-κB and macrophage activation pathways (Fig. 2C), consistent with elevated plasma levels of IL-6, TNF-α, and IL-12p40. Notably, genes characteristic of IRF7 and Type I IFN signaling were even more significantly enriched than those associated with NF-κB activation (Fig. 2C). Similar findings occurred in mice with obesity after 22 weeks of WD feeding (Supplementary Fig. 3A).

Fig. 2. The short Western diet increased TLR2-driven NF-κB and Type I IFN signaling.

Mice were fed a standard or short Western diet (1 week) and injected intraperitoneally with FSL-1; plasma, peritoneal cells, and visceral adipose tissue were collected 2.5 h later. A Weight change (n = 12/group, 3 experiments). B Plasma cytokines (n = 36–42/group, 6 experiments). C Bulk RNA-seq of peritoneal cells showing enriched pathways and IRF7 heatmap (n = 8/group, 2 experiments). D scRNA-seq UMAP with Nfkb1 and Irf7 expression (n = 6/group, 2 experiments). E Differential metabolites (n = 14–16/group, 4 experiments). F Ifnb1 expression and Cd11b⁺Ly6G⁺Ly6C^hi monocytes in visceral adipose tissue (n = 4–8/group, 2 experiments). Statistical significance was determined by Mann–Whitney U or Kruskal–Wallis with Benjamini–Krieger–Yekutieli correction for panels with multiple comparisons. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

We next used single-cell RNA-sequencing to identify the cells enriched for NF-κB and IRF7 signaling. We found broad expression of Nfkb1 across monocytes, neutrophils, dendritic cells, NK cells, T cells, and B cells, whereas Irf7 expression was highest in macrophages and inflammatory monocytes from short WD–fed mice (Fig. 2D). These cells are known to accumulate in adipose tissue and drive obesity-associated inflammation^18,19. IFN-β is typically the first Type I IFN to be expressed, followed by induction of Ifna4 and increased IRF7 expression, thus leading to downstream transcription of additional IFN-α subtypes and interferon-stimulated genes (ISGs)²⁰. Despite robust IFN pathway enrichment, scRNA-seq did not detect Ifnb1 or Ifna4 expression in peritoneal CD45⁺ cells (Supplemental Fig. 3B), indicating that these cells were sensing, but not producing, Type I IFNs. Metabolomics revealed only minor changes, including increased glycerophospholipid PS (18:2/18:0), abundant in WD-derived fats (Fig. 2E).

To identify the source of Type I IFNs in this model, we next profiled peritoneal visceral adipose tissue, an organ known to contribute to the tonic inflammation characteristic of obesity²¹. We found that hematopoietic CD45⁺ cells isolated from adipose tissue in mice fed the short WD had increased Ifnb1 expression after TLR2 stimulation, and that adipose tissue was enriched in Cd11b⁺Ly6G⁺Ly6C^hi monocytes known to exhibit TLR2-Type I IFN signaling (Fig. 2F). In addition, the short WD increased expression of ISGs, together with enrichment of cholesterol biosynthesis and SREBP-dependent lipid metabolism pathways, in visceral adipose tissue (Supplemental Fig. 3C). These findings indicate that overnutrition activates the TLR2-Type I IFN axis in adipose tissue, in part by increasing the production of Type I IFNs in CD45⁺ tissue-resident immune cells.

RAGE mediates diet-induced amplification of TLR2-driven inflammatory signaling

Pathways regulating Type I IFN production and sensing in adipose tissue have been previously described²². In contrast, the mechanisms that control TLR2-Type I IFN sensing in circulating immune cells, particularly in the setting of overnutrition, remain less well-defined. Prior studies have shown that RAGE functionally interacts with TLRs by utilizing shared adapter proteins, including TIRAP and MyD88²³, rather than by physical interaction²⁴. Since the WD increases ligands for the receptor for AGEs (RAGE), which can activate both NF-κB and IRF7²⁵, we investigated whether RAGE is necessary for WD-associated TLR2 responsiveness. We administered FPS-ZM1, a highly specific and non-toxic inhibitor of RAGE^26,27, during Western diet feeding. RAGE inhibition reduced weight gain without altering cholesterol levels (Fig. 3A, B). It also blunted FSL-1-induced secretion of IL-6, IL-12p40, and TNF-α, suppressed NF-κB-associated transcriptional responses, and reduced TLR2 expression in peritoneal CD45⁺ immune cells (Fig. 3C–E). These findings are concordant with prior work showing that RAGE and TLR2 converge on TIRAP/MyD88-dependent signaling pathways that activate MAPK- and NF-kB-dependent signaling in inflammatory myeloid cells^23,28. Exposure to the short WD increased expression of IRF7 peritoneal immune CD45⁺ cells in a RAGE-dependent manner (Fig. 3E), supporting a role for RAGE in amplifying Type I IFN sensing downstream of TLR2 activation. In contrast, the secretion of CXCL1, CCL17, and CCL22 was not affected by either the Western diet or RAGE inhibition (Supplementary Fig. 4). Together, these findings suggest that the Western diet does not globally amplify TLR2-driven outputs in vivo, but instead selectively amplifies a subset of canonical proinflammatory cytokines in a RAGE-dependent manner.

Fig. 3. Diet-induced inflammatory signaling depends on RAGE.

Mice fed the short Western diet received daily intraperitoneal injections of FPS-ZM1 or PBS for 7 days, followed by FSL-1 injection. A Weight change on day 7 (n = 12–14/group, 3 experiments). B Plasma cholesterol before FSL-1 (n = 10–13/group, 3 experiments), using ANOVA with Dunnett’s T3 test. C Plasma cytokines 2.5 h after FSL-1 (n = 13–20/group, 4 experiments), using Kruskal–Wallis with Benjamini–Krieger–Yekutieli correction for multiple comparisons. D Bulk RNA-seq of peritoneal cells showing enriched pathways and heatmap of differentially expressed genes in the IRF7 signaling pathway (n = 8/group, 2 experiments). E TLR2 and IRF7 expression, expressed as log2 read counts per million (n = 8/group, 2 experiments). For all panels, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Diet reversion and metformin differentially restore immune homeostasis

Although RAGE inhibitors have been used in clinical trials for neurodegenerative diseases, they are not used for the treatment of obesity. We therefore next compared dietary reversal and metformin, both longstanding therapies for obesity and insulin resistance. After 1 week of short WD feeding, mice either reverted to standard chow or continued WD with metformin. Diet reversion normalized body weight and cholesterol, whereas metformin normalized LDL but not body weight or total cholesterol (Fig. 4A). Diet reversion restored TNF-α and IL-12p40 responses to baseline but not IL-6, whereas metformin selectively reduced TNF-α after FLS-1 injection (Fig. 4B). Despite these differences, transcriptomic profiling of CD45⁺ peritoneal cells showed both interventions broadly downregulated proinflammatory programs, including Type I IFN signaling, NF-κB, TLR signaling, Th1 and Th2 T cell activation, and nonalcoholic fatty liver disease pathways (Fig. 4C). These gene sets overlapped with those suppressed by RAGE inhibition (Figs. 3D and 4C). Because metformin reduced TLR2-driven inflammatory gene expression in the absence of weight loss, we next examined additional pathways that might contribute to this effect. We found that metformin increased expression of negative regulators of RAGE/NF-κB signaling (Mef2c, Ddit4, Il10) without altering expression of RAGE itself (Ager) (Fig. 4D, Supplemental Fig. 5). These findings show that inflammatory signaling downstream of TLR2 activation is modulated by both weight-dependent and weight-independent therapies.

Fig. 4. Effects of dietary reversion or metformin on TLR2-driven signaling.

Mice were fed the Western diet for 1 week, followed by a second week of the Western diet, the standard diet (diet reversion), or the Western diet in conjunction with metformin (WD+metformin). Controls received the standard diet for 2 weeks (standard). A Weight change and plasma cholesterol (n = 6–10/group, 2 experiments) using Kruskal–Wallis with Benjamini–Krieger–Yekutieli correction for multiple comparisons of weights and ANOVA for comparisons among cholesterol fractions. B Plasma cytokines 2.5 h after FSL-1 (n = 8–12/group, 2 experiments) using Kruskal–Wallis with Benjamini–Krieger–Yekutieli correction for multiple comparisons. C Bulk RNA-seq of peritoneal cells showing enriched pathways and IRF7 heatmap (n = 8/group, 2 experiments). D Ager (encoding RAGE) expression normalized to standard diet controls. For all panels, ns not significant; **p < 0.01, ***p < 0.001, ****p < 0.0001.

Tirzepatide suppresses diet-induced TLR2 inflammatory responses

Given the success of GLP-1RA/GIP agonists in treating obesity, we tested tirzepatide in short WD–fed mice. After 3 weeks, tirzepatide induced 6% weight loss compared to 25% weight gain in vehicle-treated mice (Fig. 5A) and lowered total and high-density lipoprotein (HDL), but not LDL, cholesterol (Fig. 5B). Tirzepatide significantly reduced IL-6, TNF-α, and IL-12p40 after FSL-1 injection (Fig. 5C). Transcriptomic profiling of peritoneal CD45⁺ cells showed enrichment of cAMP response element-binding protein (CREB) target genes (Fig. 5D), consistent with the ability of GIP-GIPR and GLP-1RA signaling to inhibit NF-κB by competing for the shared coactivator CBP/p300^29,30. Tirzepatide also downregulated T cell activation pathways (Fig. 5D), an effect attributed to CREB-mediated inhibition of protein kinase A activity³⁰, and suppressed expression of S100A8/9, an alarmin that activates RAGE (Fig. 5E). Because similar signaling effects have been observed in cell culture systems treated with tirzepatide^29,30, these findings raise the possibility that tirzepatide exerts both weight-dependent and weight-independent anti-inflammatory effects.

Fig. 5. Tirzepatide suppresses TLR2-driven inflammatory signaling.

Mice were fed a standard diet (3 weeks), a Western diet (3 weeks), or a Western diet with tirzepatide added in the last 2 weeks. A Weight change on day 21 (n = 10/group, 2 experiments) using Kruskal–Wallis with Benjamini–Krieger–Yekutieli correction. B Plasma cholesterol before FSL-1 (n = 7/group, 2 experiments) using ANOVA. C Plasma cytokines 2.5 h after FSL-1 (n = 8–10/group, 2 experiments) using Kruskal–Wallis with Benjamini–Krieger–Yekutieli correction. D Bulk RNA-seq of peritoneal cells showing enriched pathways and heatmaps of indicated signatures (n = 8/group, 2 experiments). E S100a8 and S100a9 expression normalized to Western diet controls (n = 6–7/group, 2 experiments) using Mann–Whitney U with Holm–Šídák correction. For all panels: ns not significant; **p < 0.01, ***p < 0.001, ****p < 0.0001.

Type I IFN signatures associate with metabolic dysfunction in children with obesity

To test the human relevance of these findings, we profiled whole blood from 62 children and adolescents (body mass index (BMI) 12.2–47.5 kg/m², Supplementary Table 1). Children with obesity have increased circulating levels of NF-κB-dependent proinflammatory proteins, including IL-6, C-reactive protein, and vascular endothelial growth factor^31,32. In contrast, much less is known about Type I IFN pathway activation in this pediatric population. Expression of CD169, a monocyte marker specifically induced by Type I IFNs³³, correlated with BMI, indicating enhanced systemic Type I IFN sensing (Fig. 6A). To test the hypothesis that plasma from children with obesity promotes Type I IFN sensing after TLR2 activation, we cultured PBMCs from donors with normal BMI in 10% allogenic plasma from donors with either normal BMI or obesity with prediabetes for 24 h, followed by stimulation with FSL-1 for an additional 24 h. All plasma was heat-inactivated to 56 °C for 30 min to degrade the activity of complement and circulating cytokines, including Type I IFNs^34,35. In contrast, advanced glycation end products are irreversible and heat-stable^36,37. Plasma from individuals with obesity and prediabetes induced a transcriptional program in lean donor PBMCs consistent with IRF7 signaling, proinflammatory cytokine signaling, and classical macrophage activation (Fig. 6B), alongside increased secretion of IL-6, TNF-α, and IL-1β (Fig. 6C). These findings support a model in which circulating factors can reprogram immune cells to amplify TLR2 responsiveness and IRF7 expression.

Fig. 6. Inflammatory signatures in children and adolescents with obesity.

A Correlation between BMI and the mean fluorescence intensity of CD169 on CD14+ monocytes (27 males and 35 females total, 10 experiments) using Spearman’s correlation. B PBMCs from lean donors cultured with plasma from donors with normal weight or obesity as described in the methods, then stimulated with media or FSL-1 (n = 8/group, 3–5 females and 3–5 males/group). Bulk RNA-seq shows enriched pathways and representative heatmaps. Differentially expressed genes: fold change >2 or <−2, FDR < 0.05. C Quantification of indicated cytokines from in vitro cultures described in (B). D Left, Quantification of IL-6 cytokines from in vitro cultures of PBMCs from lean donors supplemented with chemically defined lipid mix and FSL-1. Right, qPCR of IRF7 expression, expressed relative to unstimulated, from in vitro cultures of PBMCs from lean donors supplemented with chemically defined lipid mix and FSL-1 as indicated (n = 7–12/group, 4–6 females and 3–6 males/group). E Left, Quantification of indicated cytokines from in vitro cultures of PBMCs from lean donors supplemented with either AGE-BSA or BSA as a vehicle and FSL-1 as indicated (n = 6–8/group, 3–4 females and 3–4 males/group). Right, qPCR of IRF7 expression, expressed relative to unstimulated, from in vitro cultures of PBMCs from lean donors supplemented with AGE-BSA or BSA as a vehicle and FSL-1. F Left, Quantification of indicated cytokines from in vitro cultures of PBMCs from lean donors supplemented with LDL and FSL-1 as indicated (n = 6–8/group, 3–4 females and 3–4 males/group). Right, qPCR of IRF7 expression, expressed relative to unstimulated, from in vitro cultures of PBMCs from lean donors supplemented with LDL and FSL-1. For all panels: ns not significant; **p < 0.01, ***p < 0.001, ****p < 0.0001 determined using Kruskal–Wallis with Benjamini–Krieger–Yekutieli correction for multiple comparisons.

To identify candidate mediators of this effect, we tested the effect of three WD-associated plasma components on TLR2-induced cytokine secretion and IRF7 expression. We first tested the chemically defined lipid mixture used in Fig. 1, which selectively increased TLR2-driven IL-6 secretion. Although the components of this mixture are not known to bind to RAGE ligands in their native forms, oxidative or glycoxidative modifications can generate RAGE-binding lipid species^36,37. Lipids increased TLR2-induced IL-6 in a RAGE-dependent manner without altering IRF7 expression (Fig. 6D). We next tested the effects of a highly purified RAGE ligand, advanced glycation end-product bovine serum albumin (AGE-BSA), generated by reacting BSA with the monosaccharide sugar glycolaldehyde. In combination with TLR2 stimulation, AGE-BSA increased IL-6, TNF-α, and IL-1β in a RAGE-dependent manner but likewise failed to induce IRF7 expression (Fig. 6E). Finally, we examined the effect of LDLs purified from human plasma on TLR2-driven signaling. This preparation contains physiologic RAGE-binding LDL species, of which only approximately 4% in normoglycemic individuals and 8% in patients with diabetes are estimated to bind to RAGE³⁸. However, some of these modified LDL species, such as LDL oxidized by hypochlorous acid generated by myeloid cells, exhibit much greater affinity for RAGE than AGE-BSA³⁹. In contrast to the lipid mixture and AGE-BSA, plasma lipoproteins increased IRF7 expression, in addition to IL-6, IL-1β, in a RAGE-dependent manner (Fig. 6F). Notably, these components had minimal proinflammatory effects in the absence of TLR2 stimulation (Fig. 6D–F). Together, these findings show that obesity-associated plasma components differentially modulate TLR2 responses, with LDL uniquely enhancing both proinflammatory cytokine production and IRF7 induction.

Discussion

We identify TLR2 as a sensor of modest lipid excess that amplifies both NF-κB and Type I IFN signaling. Our data supports a model in which the short WD establishes an inflammatory circuit whereby TLR2 and RAGE cooperate to drive Type I IFN production in adipose tissue and enhance IFN responsiveness in circulating immune cells (Fig. 7). In this model, overnutrition increased Type I IFN production in adipose tissue-resident CD45⁺ immune cells following TLR2 stimulation. In parallel, peritoneal immune cells from overweight mice exhibited increased RAGE-dependent expression of IRF7, a master regulator of Type I IFN sensing. In humans, obesity and prediabetes were associated with increased expression of IFN-stimulated genes, and plasma from children with prediabetes and obesity induced a TLR2-Type I IFN–like signature in PBMCs from lean donors. In complementary in vitro studies, supplementation of TLR2-stimulated PBMCs with a chemically defined lipid mixture, AGE-BSA, or LDL enhanced proinflammatory cytokine secretion in a RAGE-dependent manner, whereas only LDL induced RAGE-dependent IRF7 expression. Together, these findings identify Western diet exposure as a physiologic context that engages TLR2-Type I IFN signaling, a pathway previously studied largely in infection and other inflammatory settings.

Fig. 7. Proposed model of short WD-induced TLR2-Type I IFN signaling.

The short Western diet establishes an inflammatory circuit in which TLR2 and RAGE cooperate to increase Type I IFN production in adipose tissue-resident CD45⁺ immune cells and enhance IFN responsiveness in circulating immune cells. Created in BioRender. Chou, J. (2026) https://BioRender.com/dlcuoji.

Our data further indicates that overnutrition does not globally amplify all TLR2-dependent outputs, but instead selectively enhances the responsiveness of specific downstream programs. IL-6 was induced in a RAGE-dependent manner by all the stimuli tested, including the chemically defined lipid mixture, AGE-BSA, human plasma-derived LDL species, and the short WD in mice. This pattern suggests that IL-6 may have a relatively low threshold for induction in the setting of metabolic dysfunction. In contrast, broader proinflammatory cytokine induction was observed only with in vitro cultures containing preformed RAGE ligands, including AGE-BSAs, human plasma-derived LDL species, and the short WD. Although the components of the chemically defined lipid mixture are not known to bind RAGE in their native form, oxidative or other modifications can confer RAGE-binding activity to lipids. This difference may account for the more restricted inflammatory response seen elicited by lipid supplementation. Notably, even the short WD, which acts over a longer time frame and in a more complex in vivo metabolic environment that includes adipose-resident CD45⁺ immune cells, did not enhance all cytokines downstream of TLR2 activation. In particular, CCL17 and CCL22 were unchanged by the Western diet. Because these chemokines are more closely linked to alternative macrophage activation programs involving transcriptional regulators such as IRF4, they would not be expected to increase simply as a consequence of stronger NF-kB signaling. These findings indicate that the WD exposure does not globally potentiate all TLR2-induced mediators, but instead selectively biases its downstream output toward defined proinflammatory outputs.

These findings also identify RAGE as an important contributor to the Type I IFN sensing after TLR2 activation. Prior work indicates that TLR2 ligands do not directly activate RAGE²³, and in our system, FSL-1 alone likewise failed to activate the TLR2-IRF7 axis. However, the two pathways share downstream adapters, including TIRAP and MyD88²³, providing a plausible basis for signaling convergence. This is notable because RAGE has long been implicated in obesity-associated inflammation and its complications. Small-molecule RAGE inhibitors have been shown to be safe in humans⁴⁰, and RAGE inhibition reduces hepatic inflammation in mouse models of NASH⁴¹. Our findings extend this framework by suggesting that RAGE contributes not only to inflammatory cytokine production but also to diet-induced Type I IFN amplification. Future studies will be needed to determine if RAGE inhibition can attenuate obesity-associated inflammation in humans.

The nature of the RAGE ligand further appeared to influence the qualitative output of TLR2 signaling. The chemically defined lipid mixture and AGE-BSA enhanced proinflammatory cytokine production, consistent with amplification of canonical TLR2 signaling pathways. In contrast, LDL uniquely induced IRF7 expression in addition to inflammatory cytokines. This finding is concordant with prior work showing RAGE-dependent induction of IRF7 in atherosclerotic plaques of LDL receptor-deficiency mice with constitutively elevated circulating LDL²⁷. Because LDL isolated from human plasma contains different types of modified LDL, including oxidized LDL forms that bind to RAGE more avidly than AGE-BSA³⁹, these data suggest that engagement of the TLR2-IRF7 axis requires either a higher threshold or a qualitatively distinct mode of RAGE activation than needed to augment NF-κB-dependent cytokine production. More broadly, these findings indicate that obesity-associated plasma factors do not merely amplify TLR2 signaling, but instead differentially shape its downstream output toward inflammatory vs. IRF7-driven IFN sensing. The in vitro induction of these responses in human PBMCs further indicates that WD components can potentiate inflammatory signaling even in the absence of weight gain.

These observations are likely relevant to obesity-associated complications. Both NF-κB and Type I IFNs have been implicated in metabolic pathology, as supported by studies in knockout models showing reduced weight gain and insulin resistance when these pathways are disrupted. Previous studies have shown that prolonged exposure to high-fat diets in mice and nonhuman primates induces increased tonic IFN signaling²². In contrast, we show that TLR2-Type I IFN signaling occurs after just 1 week of Western diet exposure. Activation of this inflammatory pathway, even before the onset of obesity, may help explain epidemiologic observations linking overweight to increased risk of cardiovascular disease and respiratory infections^42–44.

Therapies for obesity differentially modulated TLR2-driven inflammation. Diet reversal and tirzepatide broadly reduced cytokine secretion, whereas metformin produced a narrower effect, primarily reducing TNF-α. Despite these differences, all three interventions reversed the proinflammatory transcriptional signature induced by the Western diet. Tirzepatide exerted the most pronounced effects, suppressing IL-6, TNF-α, IL-12p40, T cell activation, and macrophage activation signatures while inducing CREB targets and downregulating IRF7 and S100A8/9. These observations are consistent with prior studies showing that GLP-1R and GIPR agonists activate CREB signaling, while GIP deficiency enhances S100A8 expression in myeloid cells^45,46. Because diet reversal and tirzepatide, but not metformin, also reduced body weight, part of their anti-inflammatory effect may reflect reduced adipose tissue burden and a corresponding decrease in adipose-resident immune cells capable of producing IFN-β. At the same time, metformin suppressed TNF-α secretion and Type I IFN sensing in the absence of weight loss, supporting the idea that obesity-directed therapies can also act through weight-independent mechanisms. This interpretation is consistent with prior studies showing that metformin inhibits NF-κB through AMPK-dependent and mitochondrial stress-related pathways⁴⁷. Although our experiments were not designed to distinguish weight-dependent from weight-independent effects of tirzepatide, prior evidence that GIP signaling limits the generation of the endogenous RAGE ligands S100A8/A9⁴⁸ is concordant with our finding that tirzepatide attenuates RAGE-dependent amplification of TLR2 signaling. Thus, tirzepatide may interrupt this inflammatory circuit through a combination of metabolic and direct immunomodulatory effects. Additional studies are needed to distinguish the weight-independent from the weight-dependent effects of tirzepatide.

This study has limitations. We used recombinant TLR2 ligands to isolate diet-specific effects on this receptor, whereas real-world pathogens and human diets activate multiple innate and adaptive pathways. In addition, our conclusions regarding RAGE involvement are based on pharmacologic inhibition rather than genetic loss-of-function approaches. Although FPS-ZM1 is a highly selective RAGE inhibitor²⁶ that produced consistent effects across multiple inflammatory and transcriptional readouts in our mouse model of overnutrition and in vitro stimulations of human PBMCs, we cannot exclude off-target effects. Future studies using genetic models will be important to confirm the causal role of RAGE in amplifying TLR2-driven inflammatory and Type I IFN responses during overnutrition. In addition, although we identified adipose tissue-resident CD45⁺ cells as a source of IFN-β, our study did not define the specific IFN-β-producing cellular subsets responsible for this response. Single-cell sequencing or related approaches focused on adipose CD45⁺ cells will be needed to precisely resolve the IFN-β producing populations and the changes induced in these cells, enabling the production of Type I IFNs. Longitudinal studies are also needed to assess the effects of diet reversion, metformin, and tirzepatide on inflammatory signaling induced by infectious pathogens and various dietary patterns.

In summary, we identify enhanced TLR2 signaling as a sensitive hallmark of diet-induced inflammation. Our data support a model in which overnutrition potentiates a TLR2-Type I IFN axis, with adipose tissue serving as a site of inducible Type I IFN production and circulating immune cells exhibiting heightened IFN sensing and inflammatory responsiveness, thereby linking this pathway to human overweight and obesity. We further demonstrate that diet reversion, metformin, and tirzepatide differentially attenuate TLR2-driven NF-κB and Type I IFN programs through both weight-dependent and weight-independent mechanisms. Together, these findings position the downstream transcriptional targets of this axis as candidate biomarkers for assessing the immunologic effects of current and future obesity therapies.

Materials and methods

Mice

All mice were bred and raised commercially at The Jackson Laboratory or in breeding colonies within Boston Children’s Hospital. C57BL/6 mice were kept in microisolator cages, in groups of two to five mice per cage, under specific pathogen-free conditions in the animal facility at Boston Children’s Hospital. All procedures were performed within the guidelines of the Institutional Animal Care and Use Committee of Boston Children’s Hospital.

Both male and female mice were included throughout the study, with 40–60% representation of each sex in all experiments except those utilizing scRNA-seq or metabolomics. Because preliminary and validation experiments showed no detectable sex-dependent differences in TLR2-driven inflammatory responses, sex was controlled in high-dimensional discovery assays to reduce biological variability. Female mice were used for scRNA-seq to ensure consistency across sequencing experiments, and male mice were used for metabolomics to minimize estrous cycle-associated metabolic variability. Findings from these single sex discovery experiments were supported by validation experiments that included both sexes, including bulk RNA-sequencing, qPCR for specified genes, and cytokine secretion. Mice were fed ad libitum with a standard chow diet (Prolab IsoPro RMH3000, 5P76) or a Western diet (Envigo, TD.88137). The standard diet contains 5% fat (1.75% saturated fat), 0.02% cholesterol, 60% carbohydrates, and 22% protein. The Western diet contains the following components by weight: 21.2% fat (12.7% saturated fat), 0.2% cholesterol, 48.5% carbohydrates (containing 34% sucrose), and 17.3% protein. Mice aged 15–17 weeks were fed the Western diet for 7 days. Body weight was measured daily for experiments lasting 7 days and weekly for experiments lasting more than 1 week.

Study participants

This study was approved by the Boston Children’s Hospital Institutional Review Board (protocol 09-04-113R). Informed consent was obtained from participants 18 years or older, or by their legal guardians, with assent by participants under 18 years of age. Inclusion criteria: age of 3–25 years of age. Exclusion criteria: any disorder requiring medical treatment, other than prediabetes. Normal BMI was defined as a BMI of 5th–85th percentile for individuals under 18 years of age⁴⁹ or, 18.5–24.9 for individuals ≥18 years; obesity was defined as a BMI percentile of >95th percentile for age and biologic sex for individuals under 18 years of age⁴⁹ or BMI ≥ 30 for individuals ≥18 years. Prediabetes was defined by a hemoglobin A1C of 5.7 to 6.4⁵⁰.

In vitro stimulation of human PBMCs

Human PBMCs from healthy control blood were freshly isolated by Ficoll density gradient centrifugation and plated at a concentration of one million cells per well. AIMV media (cat# 12055091, Thermo Fisher Scientific), bovine albumin fraction V (7.5% solution) (BSA) (100 μg/mL, cat#15260037 Thermo Fisher Scientific), bovine AGE-BSA biotinylated protein (AGEs) (100 μg/mL, cat# BT4127, R&D Systems), human plasma-derived LDL (250 mg/dL, cat# 02698, Stemcell Technologies), or lipid mix (cat# L0288-100, Millipore Sigma) was added to the appropriate wells. The lipid mix contains soluble cholesterol (0.22 mg/mL), polyunsaturated fatty acids (10 µg/mL linoleic acid and linolenic acid; 2 µg/mL arachidonic acid), monounsaturated fatty acids (10 µg/mL oleic acid), and saturated fatty acids (10 µg/mL myristic acid, palmitic acid, and stearic acid). Cells were stimulated for 24 h with indicated reagent and then stimulated with individual TLR agonists TLR1/2 (Pam3CSK4, 100 μg/mL, cat# tlrl-pms, InvivoGen), TLR2/6 (FSL-1 (Pam2CGDPKHPKSF), 1 μg/mL, cat# tlrl-fsl, InvivoGen), TLR3 (Poly I:C, 10 μg/mL, cat# tlrl-picw, InvivoGen), TLR4 (Lipopolysaccharide, 0.1 μg/mL, cat# L3755, Millipore Sigma), TLR5 (FLA-ST, 0.25 μg/mL, cat# tlrl-stfla, InvivoGen) TLR7 (Imiquimod R837, 10 μg/mL, cat# tlrl-imqs-1, InvivoGen), TLR8 (ssRNA40/LyoVec, 5 μg/mL, cat# tlrl-Irna40, InvivoGen), or TLR9 (CpG ODN2216, 2 μM, cat# tlrl-2216b, InvivoGen) for 24 h. Alternatively, in place of the chemically derived lipid mix, plasma from healthy lean and obese individuals was first heat-inactivated by being heated to 56 °C for 30 min before being diluted to 20% with AIMV media (Cat# 12055091, Thermo Fisher Scientific). Plasma from either healthy controls or obese individuals was added to control cells to make a final concentration of 10% plasma. The cells were then stimulated for 24 h at 37 °C. Post-incubation cells were stimulated with TLR2/6 (FSL-1 (Pam2CGDPKHPKSF), 1 μg/mL) for 24 h. In experiments where RAGE inhibitor was used, RAGE inhibitor FPS-ZM1 (1 μM, cat# 6237, Tocris Bioscience) was cultured with cells for 24 h before cells were stimulated with TLR2/6 (FSL-1 (Pam2CGDPKHPKSF), 1 μg/mL, cat# tlrl-fsl, InvivoGen) for 24 h. Supernatant cytokines were quantified from supernatants using LEGENDplex Human Inflammation Panel 1 (cat#740809, BioLegend). Data were analyzed via flow cytometry using an LSR Fortessa (BD Biosciences) with LEGENDplex Data Analysis Software (BioLegend). mRNA was extracted from cells using the RNeasy Mini kit (cat#: 74104, Qiagen). mRNA was retro-transcribed to cDNA using the SuperScript VILO cDNA Synthesis Kit (cat#: 11754050, Invitrogen). Real-time PCR was then used to examine IRF7 and HPRT gene expression.

In vivo stimulation with TLR agonists

After being fed either the standard diet or WD, mice were injected intraperitoneally with PAM3CSK4, used to activate TLR1/2 (1 mg/kg; cat#: tlrl-pms, InvivoGen) or FSL-1, used to activate TLR2/6 (1 mg/kg body weight; cat#: tlrl-fsl, InvivoGen). Approximately 2.5 h post-injection, mice were anesthetized with isoflurane, and blood was collected via retro-orbital sinus bleeding. The mice were then euthanized using carbon dioxide. Human TLR4 Reporter HEK293 Cells (cat# hkb-htlr4, InvivoGen) were used to screen for endotoxin contamination of TLR agonists prior to experimental use⁵¹. Where indicated, mice were treated with the RAGE inhibitor FPS-ZM1, metformin, or tirzepatide. For experiments using FPS-ZM1, mice were given the WD or standard diet for 1 week before being intraperitoneally injected every day for 7 days with FPS-ZM1 (1 µg/g body weight; cat# 623710, Tocris Bioscience) while continuing the indicated diets. For intervention experiments, mice were exposed to the short WD or standard diet for 1 week, followed by 2 weeks of either reversion to a standard diet, metformin (1.5 mg/mL, cat# 317240, Fisher Scientific) dissolved in drinking water given ad libitum, tirzepatide (10 nmol/kg, cat# P1206, Selleckchem) administered intraperitoneally three times weekly, or saline administered intraperitoneally as a vehicle control according to treatment timelines. This design was chosen to mirror the timing of the untreated short WD, while allowing sufficient time for the mice in the diet-reversion group to return to the baseline weight. At the start of the third week, all mice were injected with FSL-1 (1 µg/g body weight; cat#: tlrl-fsl, InvivoGen), followed by anesthetization and retro-orbital blood collection 2.5 h later. The mice were then euthanized using carbon dioxide, and peritoneal immune cells were collected using saline to lavage the peritoneal cavity. Cells were counted using an automated cell counter (Countess 3, Thermo Fisher Scientific). Using FACS analysis, we confirmed that >97% of these cells expressed CD45, indicating their hematopoietic origin. Visceral adipose tissue was collected from the gonadal region. Adipose tissue was collected from mice and either placed in RNAlater solution (cat# AM7021, Thermo Fisher Scientific) and frozen for RNA extraction or digested to extract immune cells. To extract immune cells, adipose tissue was finely chopped and incubated in digestion media (3% RPMI supplemented with 1 mg/mL collagenase (cat# C2139, Sigma Aldrich)) at 37 °C in a shaking incubator set to 270 rpm for 1 h. Samples were then centrifuged, and pellets were resuspended in RBC lysis buffer (cat# 00-4333-57, Thermo Fisher Scientific) and incubated at room temperature for 5 min. Post-incubation, samples were run through a 70 μm strainer and then centrifuged. The cells isolated from adipose tissue were counted using an automated cell counter (Countess 3, Thermo Fisher Scientific) and either stained for flow cytometry or frozen as dry pellets for mRNA extraction.

Serum analysis

Mice were anesthetized with isoflurane, and then blood was collected via retro-orbital sinus bleeding. Serum was collected by centrifuging whole blood at 4 °C for 10 min at max speed. Total cholesterol, HDL, and LDL/very low-density lipoprotein (VLDL) were measured by colorimetric assay (cat#: MAK045, Millipore Sigma). Serum cytokines were quantified using the LEGENDplex Mouse Macrophage/Microglia Panel (Biolegend, cat#740845). Data were analyzed via flow cytometry using an LSR Fortessa (BD Biosciences) with LEGENDplex Data Analysis Software (BioLegend).

Flow cytometry

Standard flow cytometric methods were used for the staining of cell-surface proteins. Anti-mouse monoclonal antibodies (mAbs) with the appropriate isotype-matched controls were used for staining. All flow cytometry data were collected with an LSR Fortessa (BD Biosciences) cell analyzer and analyzed with FlowJo software (BD Biosciences). Reagents and mAbs for flow cytometry FACS were used as described by the manufacturer: Fixable Viability Dye eFluor™ 506 (Thermo Fisher Scientific, cat# 65-0866-14), B220 (BioLegend, cat# 103211), CD4 (BioLegend, cat# 100428), CD8 (Thermo Fisher Scientific, cat#12-0081-82), DX5 (BioLegend, cat# 108906), CD11b (BioLegend, cat# 101225), Ly6G (BioLegend, cat# 127605), Ly6C (BioLegend, cat# 128013), F4/80 (Thermo Fisher Scientific, cat# 17-4801-82), TLR2 (BioLegend, cat# 148603) and human CD169 (BioLegend, cat# 346008), CD14 (BioLegend, cat# 982502), CD19 (BioLegend, cat# 982403), and CD3 (BioLegend, cat# 980010). The CD169 MFI ratio was calculated using an established method in which the CD169 MFI on CD14⁺ monocytes is divided by the CD169 MFI on CD3⁺ T and CD19⁺ B cells, which serve as the internal negative reference cell population⁵².

Single-cell RNA sequencing

For each sample, approximately 17,000 murine peritoneal immune cells at a concentration of 1000 cells/μL were input into a 10X Genomics Chromium Controller. Chromium Next GEM Single Cell 3ʹ kits (version 3.1) were used to generate single-cell gene expression libraries, which were subsequently sequenced using an Illumina NextSeq 500 system with 150-bp paired-end sequencing. Libraries were processed using CellRanger version 7.1 (10X Genomics, Pleasanton, Calif) with GRCm39 as the reference. We excluded genes expressed in fewer than 5 cells, as well as those encoding ribosomal structural proteins and noncoding ribosomal RNA. Low-quality cells with more than a 10% mitochondrial gene content or fewer than 500 features were also excluded. Principal component analysis was performed by using the elbow heuristics method to determine the 20 top principal components for subsequent clustering analysis using the Louvain clustering algorithm, followed by Uniform Manifold Approximation and Projection visualization. Cell type annotation was done using CellTypist^53,54, with gene markers for murine cell types⁵⁵. Visualization of Pathway analysis of differentially expressed genes within cell types was performed by using Ingenuity Pathway Analysis (Qiagen Bioinformatics, Redwood City, California).

Metabolomics

Mice were fed either the standard diet or WD for 7 days and then injected intraperitoneally with FSL-1 (1 mg/kg body weight; cat#: tlrl-fsl, InvivoGen). 2.5 h post-injection, mice were euthanized, and peritoneal immune cells were collected and purified for peritoneal macrophages using magnetic separation (cat#130-110-434, Miltenyi Biotec). Purified peritoneal macrophages were lysed with 80% MeOH, and the lysates were centrifuged at 16,000 rpm at 4 °C for 10 min. Supernatants were aliquoted to HPLC vials and dried overnight in a speed vac. All samples were stored at −80 °C and reconstituted prior to analysis in 200 µl starting eluent. Quality controls (QCs) were prepared by mixing aliquots of all samples. These QCs were prepared identically to the individual samples. Injection volume was 5 µl. Chromatographic separation was achieved using a Vanquish HPLC system (Thermo Fisher Scientific) with an installed Acquity UPLC BEH C18 column, 1.7 µm, 2.1 × 100 mm, for reverse phase (RP) mode and an Acquity UPLC BEH Amide column, 1.7 µm, 2.1 × 100 mm, for hydrophilic interaction liquid chromatography (HILIC) mode. For both columns, associated guard columns were used to prolong column lifetime. During analysis, the column temperature was held at 40 °C, and the flow rate was 350 µl/min. For both modes, i.e., RP and HILIC, a gradient elution program was used. For RP, it consisted of mobile phase A (0.1% (v/v) formic acid in H2O) and mobile phase B (0.1% (v/v) formic acid in methanol). The gradient was as follows: 0–11 min, 0.5 to 98% B; 11–15 min, 98% B, followed by a column re-calibration step using 0.5% B until 20 min. For HILIC, mobile phase A (10 mM ammonium formate in H2O containing 0.1% (v/v) formic acid) and mobile phase B (10 mM ammonium formate in acetonitrile/H2O (95:5) containing 0.1% (v/v) formic acid) were used. The gradient was applied as follows: 0–2 min 100% B, 2–10 min from 100 to 50% B, 10–15 min 50% B, followed by a column re-calibration step using 100% B until 25 min.

The Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific) was operated using Xcalibur (Thermo Fisher Scientific) software and equipped with a HESI source. The ion transfer tube temperature was set to 275 °C, the vaporizer temperature was set to 250 °C for RP and 320 °C for HILIC mode. Each sample was analyzed in HILIC and RP mode, and both in negative and positive polarization. For positive polarization, a capillary voltage was 3500 V, whereas for negative polarization, the capillary voltage was set to −3000 V. The gases were set as follows: Sheat gas 40, Aux Gas 8, Sweep gas 1. The full scan range was 70–800 amu with a resolution of 90,000 with the automatic gain control (AGC) target set to standard and a cycle time at 0.6 s (RP) and 0.8 s (HILIC). The MS2 acquisition was data dependent with a resolution of 30,000, an AGC target at standard, and a HCD Collision at 30, 50, and 80. Dynamic exclusion mode was used, whereby features were excluded for ddMS2 after being detected once for a duration of 3 s. Lock mass correction was performed using EASY-IC^TM in scan-to-scan mode. The acquired MS data sets were processed with Compound Discoverer 3.3 (Thermo Fisher Scientific). Metabolites were annotated using mzcloud™ and ChemSpider databases. After preprocessing, metabolite lists were further filtered by excluding metabolites without or with divergent annotation and retention within the void volume or re-calibration time. Peak areas were normalized using cell count. Peak lists of the four modes were finally merged, whereby for metabolites detected in multiple modes, the entry with the highest abundance was used. Pathway analyses were performed with metabolites with annotation levels 2 and 3 by MetaboAnalyst 6.0 (https://www.metaboanalyst.ca/).

Transcriptome analysis

mRNA was extracted from peritoneal immune cells using the RNeasy Mini kit (cat# 74104, Qiagen) and from human whole blood collected in PAXGene tubes using PAXgene Blood RNA Kit (IVD) (cat# 762164, Qiagen). To extract mRNA from adipose tissue, adipose tissue was homogenized and then extracted using the RNeasy lipid tissue mini kit (cat# 74804, Qiagen). mRNA was retro-transcribed to cDNA using the SuperScript VILO cDNA Synthesis Kit (cat# 11754050, Invitrogen). The Ion AmpliSeq Transcriptome Mouse Gene Expression Kit (Thermo Fisher) or Ion AmpliSeq Transcriptome Human Gene Expression Kit (Thermo Fisher) was used to prepare barcoded libraries, which were sequenced on an Ion S5 system. Differential gene expression analysis was performed using the Transcriptome Analysis Console software with the ampliSeqRNA plugin (Thermo Fisher Scientific), followed by pathway analysis using Enrichr or Ingenuity pathway analysis^56–58. Final analyses were performed by using the false discovery rate calculated with the Benjamini-Hochberg procedure.

Statistical analysis

Statistical analysis of the data with the indicated statistical tests was performed with Prism 10.0 software (GraphPad Software, La Jolla, Calif). Normality of data was determined using the Shapiro-Wilk test. For normally distributed data, the mean and standard error are shown with Student’s t-tests for univariate comparisons, or ANOVA for multiple comparisons was used. For data that did not follow a normal distribution, the median and interquartile range are shown with the Mann–Whitney U test for univariate comparisons or the Kruskal–Wallis test for multiple comparisons of non-parametric data.

Author contributions

M.E. and J.C. wrote the main manuscript text and prepared the figures. M.E. designed and performed experiments. A.A. contributed to the experiments detailed in Fig. 1. B.W. and A.B. contributed to the experiments detailed in Fig. 4. B.L., A.N., and S.H. contributed to the experiments detailed in Fig. 6. L.L. and V.T. contributed to the experiments detailed in Fig. 2. M.C., M.G., and T.O. contributed to the analysis of single-cell transcriptomics. A.F. contributed to patient study enrollment. A.F. and V.T. contributed conceptual insight to the study. All authors reviewed the manuscript. J.C. supervised and designed the study.

Data availability

All next-generation sequencing data have been deposited in Sequence Read Archives with the following accession numbers: PRJNA1271125 (mouse bulk transcriptomic data), PRJNA1271463 (mouse single-cell RNA-sequencing data), and PRJNA1270697 (human bulk transcriptomic data).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44324-026-00125-1.