Dysregulation of macrophage PEPD in obesity determines adipose tissue fibro-inflammation and insulin resistance

Resulting from impaired collagen turnover, fibrosis is a hallmark of adipose tissue (AT) dysfunction and obesity-associated insulin resistance (IR). Prolidase, also known as peptidase D (PEPD), plays a vital role in collagen turnover by degrading proline-containing dipeptides but its specific functional relevance in AT is unknown. Here we show that in human and mouse obesity, PEPD expression and activity decrease in AT, and PEPD is released into the systemic circulation, which promotes fibrosis and AT IR. Loss of the enzymatic function of PEPD by genetic ablation or pharmacological inhibition causes AT fibrosis in mice. In addition to its intracellular enzymatic role, secreted extracellular PEPD protein enhances macrophage and adipocyte fibro-inflammatory responses via EGFR signalling, thereby promoting AT fibrosis and IR. We further show that decreased prolidase activity is coupled with increased systemic levels of PEPD that act as a pathogenic trigger of AT fibrosis and IR. Thus, PEPD produced by macrophages might serve as a biomarker of AT fibro-inflammation and could represent a therapeutic target for AT fibrosis and obesity-associated IR and type 2 diabetes. Obesity-associated AT fibro-inflammation and metabolic disturbances are linked to PEPD activity and PEPD extracellular levels.

A s obesity progresses, sterile inflammation and constitutive activation of pro-fibrotic cells (for example, macrophages and AT progenitors) occurs in the AT depots, leading to AT fibrosis 1,2 . Fibrosis of the AT impairs adipocyte hypertrophy and hyperplasia adaptations and associates with IR, and increased risk of type 2 diabetes (T2D) 3,4 . Following bariatric surgery, fibrosis in AT impedes weight loss due to the rigid fibrotic wall constraining the adipocytes and preventing efficient and timely AT lipid mobilization 5,6 . The degree of AT fibrosis and inflammation (fibro-inflammation) might be a more important risk factor for metabolic complications than the degree of obesity. Supporting this view, the spectrum of obesity includes patients with severe obesity who appear to be inappropriately metabolically healthy, in contrast with other mildly overweight patients who exhibit surprisingly severe metabolic complications 7 . Genome-wide association studies (GWAS) have identified DNA variants dissociating obesity from its metabolic complications helpful to investigate these clinical paradoxes. We and others identified the rs731839 variant in the locus corresponding to prolidase or PEPD, influencing lipid and glycaemic profiles 8,9 . The G allele was associated with metabolically unhealthy lipid (higher triglyceride and lower HDL cholesterol levels) and glycaemic (higher fasting insulin and lower adiponectin levels) profiles. Impaired lipid and carbohydrate metabolism were paradoxically associated with lower adiposity 10,11 .

Dysregulation of macrophage PEPD in obesity determines adipose tissue fibro-inflammation and insulin resistance
PEPD is a homodimer cytosolic protein responsible for degrading proline-containing dipeptides (Xaa-Pro or Xaa-hyp), playing a pivotal role in the final step of collagen degradation and turnover. Complete PEPD deficiency in humans is a rare genetic disorder characterized by a complex phenotype of angiopathies and defective wound healing, consistent with impairment of collagen turnover 12 . Impaired collagen turnover in PEPD-deficient patients leads to pulmonary fibrosis 13 .
Evidence from cancer and liver fibro-inflammation stages indi cates that PEPD protein is actively secreted, acting as a noncanonical ligand of the EGFR and ErBB2 receptors [14][15][16][17] . Therefore, we hypothesized that functional defects in PEPD may be pathogenically relevant in AT through impairing collagen degradation and defects in intracellular signalling mediated by its extracellular and non-enzymatic roles.
We observed decreased PEPD peptidase activity and increased PEPD protein secretion in inflammatory macrophages and AT from obese individuals and mouse models fed on a high-fat diet (HFD). Lower PEPD activity in AT and higher PEPD protein levels in serum are directly proportional to AT fibrosis, inflammation and metabolic dysfunction. Using pharmacological and genetic mouse models to target PEPD peptidase activity and/or its secretion, we show that dysfunctional PEPD intracellular dipeptidase activity induces AT fibrosis, whereas the PEPD secreted by AT is the main contributor to inflammation, IR and metabolic dysfunction. PEPD originated in inflammatory macrophages and plays an essential role in promoting fibro-inflammatory responses via activation of EGFR in macrophages and preadipocytes. Moreover, genetic ablation of Pepd in haematopoietic cells (HCs) prevented obesity-induced PEPD release and averted AT fibro-inflammation and obesity-associated metabolic dysfunction. Altogether, these data reveal that obesity-associated severity of AT fibro-inflammation and metabolic disturbances for a given fat mass depend on PEPD activity and PEPD extracellular levels.

Results
Low prolidase activity in obese adipose tissue associates with high PEPD release, insulin resistance and fibrosis. We measured PEPD mRNA expression, prolidase activity (PA) and secretion from AT of seven human cohorts spanning a broad spectrum of body mass index (BMI) values, ages and degrees of IR (Extended Data Fig. 1a). PEPD expression was lower in visceral white AT (VsW) of obese than lean individuals and lower in obese individuals with T2D than in obese normo-glycaemic participants (Fig. 1a,b). PEPD mRNA expression was lower in VsW than in subcutaneous white AT (ScW) in a paired sample cohort study (Fig. 1c), and lower PEPD expression in VsW was associated with higher BMI and collagen content in AT from obese individuals (measured as hydroxyproline (HPro) levels; Fig. 1d). Moreover, low PEPD expression in VsW was associated with lower expression of AT metabolic genes such as PLIN1, PPARG and GLUT4 (Table 1) and reduced VsW PEPD expression in obese individuals was associated with reduced PA in VsW (cohort 2e,f; Fig. 1e). Lower PA matched the higher release of PEPD from the VsW of these participants (cohort 2e,f; Fig. 1f), suggesting a causal relationship between the reduction of the intracellular enzymatic acti vity and the enzyme's release. Among obese participants, those with higher VsW PEPD release were more prone to IR and T2D (Extended Data Fig. 1b). The receiver operating characteristic (ROC) curve analysis highlighted that both the PEPD released and low PA in VsW (but not ScW) predicted T2D (cohort 2e,f; Fig. 1h and Extended Data Fig. 1c,d). Higher PEPD release from VsW was coupled with higher PEPD serum levels in obese participants than lean individuals and was associated with elevated cholesterol and aspartate transaminase (ASAT) levels (cohort 2e, Fig. 1h and Extended Data Fig. 1e). Reanalysis of serum proteomic data from an independent set of insulin-resistant and insulin-sensitive obese participants 18 further confirmed the predictive value of PEPD for IR (Fig. 1i). Moreover, serum PEPD levels were positively correlated with lipid metabolism (for example, APOA proteins), inflammation (for example, S100A12 and DPP4) and extracellular matrix (ECM)-related proteins (for example, TNC and COL4A2; Fig. 1j,k). Altogether, these human data provide evidence for the potential relevance of PEPD as a marker of fibro-inflammation and IR in obese individuals.
A similar pattern of low AT PEPD activity coupled with high circulating levels of PEPD was recapitulated in two independent mouse models of obesity (that is, mice fed 45% HFD and 58% HFD; Extended Data Fig. 1f-m). We found that HFD only affected PEPD peptidase activity in AT but not in the liver or muscle (Extended Data Fig. 1f,g). In mice fed a HFD for 28 weeks, the tissue PA in AT was reduced in the first 8 weeks, followed by the release of PEPD (Extended Data Fig. 1i-k). Despite the increased PEPD serum level, the enzymatic activity of PEPD in serum remained stable during the HFD time course, suggesting that the activity of the enzyme in serum is reduced and might not affect the development of AT fibrosis (Extended Data Fig. 1l). While the liver contributed to the overall levels of serum PEPD in mice, only the gonadal white adipose tissue (GnW) depot showed a significantly increased PEPD release in response to a HFD (Fig. 1l). Dysregulation in PEPD activity and release were associated with AT fibrosis and metabolic disturbances (for example, hyperglycaemia and liver steatosis; Fig. 1m and Extended Data Fig. 1m). Accordingly, serum levels of PEPD predicted the degree of fibrosis (peri-adipocyte collagen) and inflammation (Tnf expression) in GnW (Fig. 1n).  c). d, Pearson correlation matrix between ScW and VsW ECM remodelling markers and metabolic parameters in obese participants from cohort 2b (n = 14). e, PA in ScW and VsW of obese diabetic (D, n = 10) and non-diabetic (ND, n = 12) participants from cohort 2e,f. f, ELISA analysis of PEPD levels from VsW explants medium of obese diabetic (D, n = 9) and non-diabetic (ND, n = 12) participants from cohort 2e,f. g, Area under the receiver operating curve (AUC) values (95% confidence interval (CI)) for PEPD VsW and ScW levels from cohort 2e,f to discriminate participants with type 2 diabetes. h, ELISA analysis of PEPD levels in serum from lean (n = 9) and obese non-diabetic (ND) or diabetic (D; red dots; n = 9) participants from cohort 2d,e. i, AUC values (95% CI) for PEPD serum levels from the Choi et al. cohort 18 to discriminate participants with IR. j,k, PEPD-protein correlations for the Choi et al. cohort 18 , shown as volcano plots (j) and Gene Ontology enrichment analysis of the corresponding proteins (k). l, ELISA analysis of PEPD levels from tissue explants medium (that is, brown adipose tissue, ScW, GnW and liver) of C57BL/6 mice fed chow or 20 weeks on a HFD 58% (n = 4 biologically independent animals per group). m, Pearson correlations between PEPD serum levels and metabolic parameters in chow and HFD conditions (n = 22 biologically independent animals). n, AUC values (95% CI) for PEPD serum levels to predict a high degree of fibro-inflammation in GnW (that is, GnW peri-adipocyte (AD) collagen and GnW Tnf expression) from mice fed chow and a HFD 58% (n = 34 biologically independent animals). Data are presented as mean values ± s.e.m. Data were analysed using a two-tailed Student's t-test (a-c, f, h and l). Two-way ANOVA with Sidak's post hoc multiple-comparisons test was performed; Diab, diabetic status; X, interaction (e). NS, not significant; TG, triglyceride.
Prolidase activity inhibition induces PEPD secretion, adipose tissue fibrosis and metabolic complications in lean mice. To uncouple the effects of PEPD from obesity, we treated chow-fed lean mice with N-benzyloxycarbonyl-l-proline (CBZ-Pro), a pharmacological inhibitor of PEPD activity 19 . CBZ-Pro-treated mice showed higher levels of the dipeptide glycine-proline in serum, confirming PEPD activity inhibition, and exhibited higher PEPD serum levels than control mice (Fig. 2a,b). CBZ-Pro-treated mice had similar body weight (BW), fat mass and tissue weight than control mice (Extended Data Fig. 2b-d). However, despite similar fat mass, they had more severe AT fibro-inflammation in GnW and were more hyperglycaemic, glucose intolerant and insulin resistant than controls ( Fig. 2c-h). Of note, CBZ-Pro-treated mice showed comparable fasting insulin and free fatty acid (FFA) blood levels (Extended Data Fig. 2e,f) and no impairment to the expected insulin level increase in response to glucose compared to controls (Extended     Fig. 2g). However, the phosphorylated-AKT/total-AKT ratio in non-stimulated conditions was lower in CBZ-Pro-treated mice than controls in both GnW (associated with a slight increase in AKT total) and skeletal muscle, suggesting higher AT and muscle IR (Extended Data Fig. 2h,i). The amount of PEPD into the serum of CBZ-Pro-treated mice was positively correlated with the degree of ScW fibrosis and glycaemia in the fed state (Fig. 2i). Altogether, these results show that downregulation of the intracellular PEPD activity in AT and concomitant increase in the serum levels of PEPD-induced AT fibro-inflammation and IR. Furthermore, in line with the human GWAS data, dysregulation of PEPD uncouples fibro-inflammation and metabolic dysfunctions from obesity 8,9,11 . Given this dual role of PEPD, it remained unclear which event-that is, decreased PEPD intracellular dipeptidase activity or increased PEPD released from the cells-was primarily responsible for AT fibrosis and/or metabolic alterations. We performed an exploratory factor analysis (EFA) to investigate how serum PEPD level/PEPD activity covaried with the metabolic parameters measured in the CBZ-Pro-treated mice and controls (Fig. 2j). EFA showed that factor 1, representing a cluster of AT fibrosis-related variables, covaried with PEPD serum levels and fed glucose levels. Moreover, factor 2, the cluster of 'obesity'-related variables (adiposity, glycaemia and steatosis), also covaried with PEPD serum levels, whereas factor 3 showed that the activity of GnW PEPD covaried with fasting glucose. Therefore, EFA indicated that the serum level of PEPD might explain the 'fibrotic' profile and hyperglycaemic status observed in CBZ-Pro mice.
The link between serum PEPD, fibrosis and glycaemia was further strengthened when we included the CBZ-Pro mice fed HFD 58% data in the EFA (factor 1; Extended Data Fig. 2i). However, the covariation between PEPD serum level and 'obesity'-related variables was not sustained (Extended Data Fig. 2j). We rationalized that because HFD per se downregulates PEPD activity/increases its secretion, it was likely that HFD-fed control mice might have developed fibro-inflammation as in CBZ-Pro-treated mice, matching the pro-fibrotic effects of this PEPD pharmacological inhibitor observed in chow-fed conditions. Confirming this interpretation, untreated control and CBZ-Pro-treated mice fed HFD 58% exhibited similar severe AT collagen deposition, fasting insulin, FFA and glucose levels and insulin intolerance (Extended Data Fig. 2e,f,k-m).
Global Pepd ablation worsens obesity-associated adipose tissue fibrosis and metabolic comorbidities. To dissect the role of inhibition of PEPD activity from the extracellular action of the released PEPD, we performed phenotyping of the global Pepd heterozygote (HET, +/−) and knock-out (KO, −/−) mice against wild-type (WT, +/+) littermates. PEPD activity and serum levels were not detected in the total KO group. Interestingly, Pepd HET mice showed 50% decreased PEPD activity but no differences in PEPD serum levels compared to WT mice (Fig. 3a,b and Extended data Fig. 3a,b) and were anatomically similar (that is, length, BW, fat mass and tissue weight; Extended Data Fig. 3c-f). In contrast, Pepd KO mice exhibited a runty phenotype characterized by short length, low BW and decreased fat mass (Extended Data Fig. 3c-f), making subsequent phenotyping difficult and of limited value to assess metabolism. Therefore, Pepd HET mice were subsequently metabolically phenotyped.
Higher collagen accumulation was observed in ScW (HET/KO) and GnW (KO) compared to WT littermates fed a chow diet, supporting the role of reduced PEPD activity in promoting fibrosis (Fig. 3c), an association strengthened by correlation matrix analysis (Extended Data Fig. 3g). Moreover, EFA analysis of chow-fed Pepd mice showed that factor 1, clustering GnW or ScW fibrosisrelated variables, had strong loading of glucose blood levels and    negative loading of PEPD activity in AT depots (Fig. 3d). Additionally, correlation matrix analysis showed a positive association between AT fibrosis and IR (Extended Data Fig. 3g). Pepd HET showed similar glucose, FFA levels, and glucose and insulin tolerance than WT mice despite lower fasting insulin levels (Extended Data Fig. 3h-k). However, we found that insulin-induced AKT phosphorylation was lower in both GnW and liver from Pepd HET chow-fed mice than WTs, with no observed differences in skeletal muscle (Extended Data Fig. 3l-n).
The development of metabolic complications in Pepd HET mice (that is, IR, increased fasting FFA and liver steatosis) was exacerbated on a HFD compared to a chow diet. There were no differences in BW among genotypes following HFD feeding (Extended Data Fig. 4a). Pepd HET mice fed HFD 45% resulted in higher fed glucose levels and insulin intolerance than WT mice, despite maintaining similar fasted glucose, insulin and FFA levels (Extended Data Fig. 4b-d). When challenged with HFD 58% 20 , Pepd HET mice exhibited greater fibro-inflammation in GnW than when challenged with HFD 45% (Extended Data Fig. 4e,f) and demonstrated a trend to glucose and insulin intolerance despite showing lower fasting glucose than WT mice ( Fig. 3e-g). Pepd HET mice also showed higher FFA levels than controls (Fig. 3g). In addition, the basal AKT phosphorylation in GnW was decreased in Pepd HET mice, and the index of adipose tissue and insulin resistance (AT-IR) 21 was higher than in WT mice (Fig. 3h,i). Similarly, the liver of Pepd HET mice exhibited more steatosis than that in WT (Fig. 3j). Of note, the basal AKT phosphorylation in the gastrocnemius muscle was not significantly different in Pepd HET mice fed either chow or HFD (Extended Data Fig. 4g). Collectively, these results indicate that reduced PEPD activity in Pepd HET mice was sufficient to develop AT fibrosis in the absence of obesity and that HFD further exacerbated metabolic complications. Moreover, EFA revealed that IR and liver steatosis could be linked to BW and GnW fibrosis in Pepd HET mice fed either chow or HFD (45% and 58%), as these variables clustered and covaried together (factor 2; Fig. 3k).
To unmask the molecular mechanisms underlying the fibrogenic and pathogenic effectors driven by the downregulation of Pepd, we performed RNA-sequencing (RNA-seq) analysis for GnW and liver of Pepd HET and KO mice fed chow and HFD 45%. Of relevance, gene expression profiling of the gastrocnemius muscle did not differ in Pepd mice among genotypes and dietary interventions (Extended Data Fig. 4h). The 45% HFD nutritional challenge was selected to prevent the confounding fibro-inflammatory effect resulting from HFD 58% downregulating Pepd in WT mice (Extended Data Fig. 4i). Analysis of differentially expressed gene (DEGs) and the top differen tially regulated pathways in GnW Pepd HET and KO mice (fed 45% HFD) versus WT mice revealed that Pepd HET and KO mice had higher expression of genes involved in actin cytoskeleton and cell cycle regulation, immune system, inflammatory-related pathways and ECM/ECM organization-related proteins (Fig. 4a, Extended Data Fig. 4j and Supplementary Tables 2a-5a). Also, Pepd HET mice fed HFD 45% showed lower expression of genes involved in metabolic pathways, including fatty acids, leptin and insulin signalling. The Pepd KO mice showed higher expression of genes in the pro-diabetes-related cluster. Validation of these data using additional in vitro and ex vivo experiments confirmed that partial or total Pepd ablation results in AT dysfunction characterized by impaired adipogenesis from GnW progenitors, lipolysis in mature adipocytes and leptin secretion from GnW tissue explants ( Fig. 4b-f).
Next, we focused on analysing the DEGs related to AT function and fibro-inflammation regulated by HFD 45% on each genotype ( Fig. 4g and Supplementary Tables 6a-8a). In response to HFD 45% and compared with chow diet, GnW of the Pepd HET mice exhibited higher expression of pro-fibrotic, ECM components, ECM remodelling genes, inflammatory and immune cells and lower expression of metabolic genes.
Extending the transcriptomic analysis to the liver, we observed an enrichment of the DEGs related to 'immune system'-related pathways-mainly T cell-related pathways-were upregulated in Pepd KO and HET compared to WT in mice fed a HFD 45%. The 'diabetes' pathway was also enriched in the DEGs upregulated in Pepd KO versus WT mice. However, 'matrisome' and 'collagen-formation' pathways were downregulated in both Pepd KO and HET livers compared to WT (Fig. 4h, Extended Data Fig. 4k and Supplementary Tables 2b-5b). Further analysis of the DEGs related to fibro-inflammation and steatosis did not reveal significant differences between Pepd HET and WT mice in response to HFD ( Fig. 4i and Supplementary Tables 6b-8b). These results align with the absence of liver fibrosis in Pepd-ablated mice fed HFD 45% (Fig. 4j). However, Pepd HET mice showed a more significant upregulation of pro-steatosis markers in response to HFD (that is, beta-oxidation and cholesterol metabolism) than WT mice ( Fig. 4i and Supplementary Tables 6b-8b).
Collectively, transcriptomic analyses in GnW and liver confirmed exacerbated GnW fibro-inflammation and dysfunction in both the KO and HET mice when challenged with HFD 45%. The GnW fibro-inflammatory response to HFD was stronger in Pepd HET mice than in WT mice. Notably, fibro-inflammation was not exacerbated in the liver of Pepd-ablated mice. This supports the relevance of PEPD in ECM remodelling of the AT when challenged with HFD and suggests that hepatosteatosis observed in Pepd HET mice might not directly result from ablation of Pepd in the liver but as a consequence of AT dysfunction and fibro-inflammation.
The strong pro-inflammatory fingerprint observed in the RNA-seq data suggested that GnW macrophages and immune c, Representative images of red Sirius staining in ScW and GnW from Pepd WT (n = 6), HET (n = 8) and KO (n = 6) mice fed chow and quantification of the percentage area of peri-AD collagen. d,k, Heat map representing the four factors extracted through EFA in Pepd mice fed chow (d) and HFD 58% (k). The columns report the factor loadings of the observed variables. e,f, Blood glucose levels up to 120 min after an intraperitoneal injection of glucose in a GTT (e) or insulin (0.75 IU per kg) in an ITT (f) with the representative AUC (e) or AOC (f) in Pepd WT (n = 8), HET (n = 11) and KO (n = 9) mice fed HFD 58%. g, Fasting glucose, insulin and FFA levels in Pepd WT (n = 8), HET (n = 11) and KO (n = 9) mice fed HFD 58%. h, Representative images of blots of total and phosphorylated (S473) AKT protein and beta-actin in GnW from Pepd WT (n = 8), HET (n = 11) and KO (n = 9) mice fed HFD 58%. i, AT-IR index in Pepd WT, HET and KO mice in chow (n = 6, 8 and 6 biologically independent animals per group, respectively), HFD 45% (n = 9, 11 and 9 biologically independent animals per group, respectively) and HFD 58% (n = 8, 11 and 9 biologically independent animals per group, respectively) conditions. j, Representative images of H&E staining in liver from Pepd WT, HET and KO mice fed chow (n = 4, 6 and 4, respectively), HFD 45% (n = 7, 10 and 6 biologically independent animals per group, respectively) and HFD 58% (n = 9, 11 and 6 biologically independent animals per group, respectively) conditions and quantification of liver steatosis. k, Heat map representing the four factors extracted through EFA. The columns report the factor loadings of the observed variables. Data are presented as mean values ± s.e.m. Data were analysed using two-way ANOVA with Turkey's post hoc multiple-comparisons test; G, genotype; X, interaction (a, b, e, f, i and j). One-way ANOVA with Sidak's (e and f) or Dunnett's (c, g and h) post hoc multiple-comparisons test were used. ND, not detected.
NoraD., P < 0.0001 G, P < 0.   Fig. 5c). Reanalysis of proteomic data from human-induced pluripotent stem cell-derived macrophages (iPSDMs; PXD001953) 22 ), confirmed the high abundance of PEPD in macrophages compared to undifferentiated induced pluripotent stem cells (iPSCs; Extended Data Fig. 5c). Given that obesity is associated with an imbalance between classically activated macrophages (inflammatory) and alternatively activated macrophages (non-inflammatory), we assessed the modulation of Pepd expression by specific macrophage polarizing agents 23 . In keeping with the reduction of Pepd expression in inflamed AT of obese mice, Pepd expression was also lower in pro-inflammatory M(lipopolysaccharide (LPS)) cells, whereas it was higher in M(GC) and not modulated in M(interleukin (IL)-4)treated cells compared to unstimulated macrophages (Extended Data Fig. 5d). Additional evidence of the relevance of human macro phages was provided by RNA-seq data comparing undifferentiated human iPSCs and differentiated and analysis of the effect of LPS on human macrophages derived from monocytes (MDMs) and iPSDMs (EGAS00001000563) 24 . In line with mouse models, stem cell transcriptomic analysis confirmed that (1) PEPD expression increases during human macrophages differentiation; (2) iPS-DMs and MDMs display similar PEPD expression levels; and (3) PEPD expression decreases in response to LPS in both iPSDMs and MDMs (Fig. 5d). We then validated at the cellular level that BMDMs treated with CBZ-Pro or LPS decreased PEPD activity (Fig. 5e) and increased PEPD released to the medium compared to non-activated or activated macrophages ( Fig. 5f and Extended Data Fig. 5e-g). These results strengthen the functional coupling between the lower enzymatic activity of PEPD in AT and the higher release of PEPD to the extracellular compartment. Interestingly, higher PEPD release from LPS-activated BMDMs was not associated with higher PA in the culture medium, suggesting that released PEPD might exhibit reduced PA following post-translational modifications (Extended Data Fig. 5h). a,h, Pathway enrichment analysis of the DEGs in GnW (a) and liver (h) from Pepd HET (C; n = 10) and KO (D, n = 9) mice compared to WT mice (n = 8) fed HFD 45%, using different databases (KEGG, Reactome, Biocarta, NABA and PID). The heat maps indicate the level of significant changes (false discovery rate-adjusted P value). b,c, Representative image of confocal analysis (Bodipy staining in green, b) and quantification of lipid accumulation (c) in primary differentiated adipocytes isolated from GnW of Pepd WT, HET and KO mice (n = 4 biologically independent animals per group). d, Gene expression in primary differentiated adipocytes isolated from GnW of Pepd WT, HET and KO mice (n = 4 biologically independent animals per group). e, Lipolytic dose-response curves for mature adipocytes isolated from GnW of from Pepd WT (n = 5), HET (n = 7) and KO mice (n = 6) in chow condition in response to 2 h of treatment with noradrenaline. f, ELISA analysis of leptin secretion from ScW and GnW explants from Pepd WT (n = 4), HET (n = 5) and KO (n = 5) mice in chow condition. g,i, Heat maps of fibro-inflammatory-related (g and i), metabolism-related (g) and steatosis-related (i) DEGs in GnW (g) and liver (i) of Pepd WT, HET and KO mice in HFD 45% (n = 8, 1 and 9 biologically independent animals per group, respectively) expressed as log 2 fold-change (log 2 FC) variation over chow diet (n = 4, 8 and 5 biologically independent animals per group, respectively). ProF, pro-fibrotic; AntiF, anti-fibrotic; Rem, ECM remodelling; Sig, ECM-related signalling; mono/Mφ, monocyte/macrophage; DC, dendritic cell; Gran, granulocyte; BC, B cell; TC, T cell; NKC, natural killer cell. j, Representative images of red Sirius staining in the liver from Pepd WT (n = 6), HET (n = 8) and KO (n = 6) mice fed HFD 45%. Data are presented as mean values ± s.e.m. One-way ANOVA with Dunnett's post hoc multiple-comparisons test (c, d and f) or two-way ANOVA with Tukey's post hoc multiple-comparisons test (e) were used. *Adjusted P value < 0.1, significant DEG compared to chow diet. # q < 0.05 compared to Pepd WT (+/+). Statistical analysis of the RNA-seq data is detailed in the Methods.  t-test (b, c, m, n and p). Two-way ANOVA with Turkey's (i and o), Sidak's (j and l-n) or Dunnett's (k) multiple-comparisons test were used; G, genotype; X, interaction.   Pepd ablation in haematopoietic cells uncouples adipose tissue fibrosis from metabolic alterations. To dissect the pivotal role of PEPD secreted from macrophages driving fibro-inflammation and metabolic disturbances, we performed a bone marrow transplant (BMT) from Pepd KO into WT recipient mice to ablate Pepd exclusively in HCs (Fig. 5g,h). BMT is commonly used in the literature to target adipose tissue macrophages (ATMs) [25][26][27] as approximately 85% of the ATMs originate from the bone marrow 28 . Thus, BMT KO mice showed similar adiposity to BMT WT mice (when fed a chow diet). BMT KO mice showed reduced PA of 85% in peritoneal macrophages and 50% in the whole AT (Fig. 5i,j). As anticipated, the reduction in PEPD activity by genetic manipulation was not associated with increased PEPD levels in serum. This contrasted with the increase observed in secreted PEPD levels in CBZ-Pro-treated mice or dietary models of obesity (Fig. 5k). As previously seen in models with global decreased PEPD activity, the BMT KO also had increased collagen accumulation in AT compared to BMT WT (Fig. 5l). However, the BMT KO mice maintained carbohydrate metabolic homeostasis in contrast to the chow-fed CBZ-Pro-treated mice, in which the decreased PEPD activity was inversely associated with PEPD secretion (Extended Data Fig. 5i-n). Of note, BMT KO mice adiposity was marginally lower than that of WT mice when fed a chow diet but it was significantly lower when fed 58% HFD (Extended Data Fig. 5o-q). Furthermore, when fed 58% HFD, the BMT KO mice exhibited more fibrosis than WT mice but were resistant to obesity and associated metabolic disturbances (that is, glucose and insulin tolerance, AT IR and liver steatosis; Fig.  5l-o and Extended Data Fig. 5r,s). In addition, while GnW showed reduced fibro-inflammatory markers expression, liver and skeletal muscle did not show significant changes at the gene expression level, reinforcing the importance of macrophage PEPD of the GnW in regulating metabolic functions (Fig. 5p). Thus, the BMT KO model exhibited increased AT fibrosis uncoupled from metabolic complications such as IR and liver steatosis, and pointed to the relevance of the extracellular PEPD produced by macrophages as a critical trigger of fibro-inflammation and metabolic complications. Moreover, EFA run for BMT chow-fed mice confirmed the negative association between AT fibrosis and PEPD enzymatic activity (factor 1; Fig. 5q), and factor 3 showed that AT IR-related parameters (that is, AT-IR and FFA levels) covaried with PEPD serum levels (Fig. 5q). Thus, these results support that extracellular PEPD might be an essential trigger of AT fibro-inflammation and related metabolic dysfunctions.
We investigated whether the released PEPD per se was a determinant of macrophages polarization and could trigger inflammation. Treatment of macrophages with purified PEPD protein-induced phosphorylation of nuclear factor-kappa B (NF-κB; ser536) and the expression of inflammatory markers, such as cox2 and il-1β, while non-inflammatory markers mgl1 and mrc2 were downregulated ( Fig. 6a and Extended Data Fig. 6d). Phospho-kinase proteome analysis using a profiler array on PEPD-treated BMDMs con firmed that PEPD phosphorylated EGFR (Extended Data Fig. 6e,f) 15 . Pretreatment of macrophages with erlotinib, an EGFR-specific tyrosine kinase inhibitor, attenuated PEPD-induced NF-κB phosphorylation, as well as cox2 and il1β expression (Fig. 6b,c and Extended Data Fig. 6g). We further confirmed this association by showing that PEPD-induced Cox2 and Il1β expression was reduced in response to Egfr silencing (~75%) (Extended Data Fig. 6h-j).
Purified PEPD promotes adipose tissue fibro-inflammation and insulin resistance. Using the Tabula Muris database 29 , we observed that AT, liver and muscle were the organs with the highest Egfr expression (Extended Data Fig. 6k). We rationalized that secreted PEPD from macrophages could also signal to other cells in AT or other organs through EGFR-dependent pathways. To test the effect of PEPD at the whole-organism level, C57BL/6N chow-fed mice were injected intraperitoneally with purified PEPD protein three times a week for 6 weeks. We investigated the gene expression of Cox2, known to be induced by PEPD through EGFR-dependent pathways 15 and found that GnW showed the most significant increase in Cox2 expression (Fig. 6l). Following injection of PEPD, the tissular PA remained unchanged compared to control mice (Extended Data Fig. 6m), indicating that the effects observed in mice injected were independent of its enzymatic activity. While BW, body composition, insulin tolerance test (ITT) and glucose tolerance test (GTT) were unaltered after PEPD injection ( Fig. 6d and Extended Data Fig. 6n-r), we observed active tissue remodelling, notably in the GnW, with increased peri-adipocyte fibrosis in PEPD-injected mice compared to control (Fig. 6e). Of relevance, the collagen and ECM remodelling enzymes' mRNA expression levels were strongly downregulated, suggesting the existence of compensatory mechanisms (Fig. 6f,g). We observed higher presence of crown-like structures around adipocytes close to high expression of macrophage markers of 'M2 tissue remodelling' in PEPD-injected mice compared to controls (Fig. 6e,g). Expression of adipocyte markers was reduced in AT from PEPD-injected mice (Fig. 6h), and both basal AKT phosphorylation levels and GLUT4 protein expression were reduced in GnW, indicative of reduced AT insulin sensitivity (Fig. 6i).
In support of these in vivo results, we characterized in vitro the effect of added extracellular PEPD on differentiated and nondifferentiated preadipocytes, considered critical pro-fibrotic cellular effectors in the AT 30,31 , and on mature primary adipocytes. Supporting a direct pro-inflammatory role, purified PEPD increased Il6 expression in differentiated preadipocytes, promoted the production of collagen I, and prevented lipid accumulation in preadipocytes in part through a mechanism involving EGFR signalling (Extended Data Fig. 7a). Purified PEPD protein also induced IL-6 release from mature adipocytes (Extended Data Fig. 7b). However, while purified PEPD increased mRNA expression of inflammatory markers and deposition of collagen from hepatic stellate cells

Fig. 6 | Purified PEPD protein induces fibro-inflammation and adipose tissue insulin resistance. a, Heat map of gene expression in BMDMs after
4 h of treatment with purified PEPD protein (n = 4 biologically independent samples). b, Representative images of blots and quantification of total and phosphorylated (S536) NF-kB protein and beta-actin in BMDMs (n = 4 biologically independent samples) pretreated or not with erlotinib (5 µM) and prior treatment with purified PEPD protein (0, 10 and 60 min; 250 nM), and quantification. c, IL1B mRNA relative expression in BMDMs (n = 4 biologically independent samples) pretreated or not (PBS) with erlotinib 5 µM and prior treatment with (PEPD) or without (control) purified PEPD protein (250 nM) for 24 h. *compared to control PBS, # compared to PEPD PBS. d, BW curve in PEPD-treated mice compared to control (saline) mice (n = 11 biologically independent animals per group) fed a chow diet between 0 and 6 weeks of treatment. e, Representative images of red Sirius staining in GnW PEPD-treated mice compared to control (saline) mice (n = 11 biologically independent animals per group) fed chow diet and quantification of peri-AD collagen (represented as a percentage) and crown-like structures (CLS). f-h, mRNA expression of ECM remodelling (f), macrophage polarization markers (g) and adipocyte markers (h) in GnW (from PEPD-injected mice compared to controls; n = 11 biologically independent animals per group). i, Representative images of blots of total and phosphorylated (S473) AKT protein, GLUT4, GADPH and beta-actin in GnW from PEPD-injected mice compared to controls (n = 11 biologically independent animals per group). j, Heat map representing the four factors extracted through EFA. The columns report the factor loadings of the observed variables. Data are presented as mean values ± s.e.m. Data were analysed using a two-tailed Student's t-test (a, e-i). Two-way ANOVA with Tukey's (b and c) or Sidak's (e) multiple-comparisons test was used; erlo, erlotinib; X, interaction.
(HSCs) in vitro, we did not observe any effect on liver fibrosis or lipid accumulation, nor changes in gene expression profile (of functional, inflammatory and fibrotic genes) in the liver in vivo (Extended Data Fig. 7c-g). Similarly, we did not observe any effect on skeletal muscle fibro-inflammation, despite higher mRNA expression of Tnf compared to control mice (Extended Data Fig. 7h,i). In vitro, PEPD had a mild effect on muscle fibroblasts, inducing Cox2 and Ccnd1 expression (Extended Data Fig. 7j). These effects were reversed in response to erlotinib, supporting the role of EGFR as a mediator of PEPD effects. While PEPD did not modify collagen production in muscle, we found that it did increase αSMA protein expression suggesting activation of fibroblasts in response to exogenous PEPD. Nevertheless, this effect on αSMA staining was not reversed in the presence of erlotinib (Extended Data Fig. 7k). Altogether, these results suggest that purified PEPD treatment targeted GnW inducing severe tissue remodelling and IR. Moreover, EFA supported a causal relationship between serum PEPD levels, insulin sensitivity (factor 1) and GnW fibrosis (factor 2; Fig. 6j).
Secreted PEPD is the primary pathogenic factor in fibroinflammation. To dissect the main pathogenic factors linking specific PEPD dysregulation and metabolic and fibro-inflammatory differences, we performed a global integrative correlation matrix analysis and EFA, integrating the four PEPD in vivo experimental models (pharmacological, genetic, macrophage-specific genetic ablation and exogenous PEPD treatment). Correlation analysis, including either chow-fed mice alone or with HFD-fed animals, revealed a negative correlation between the level of AT fibrosis and PEPD activity accounting for the reduced AT PEPD activity and fibrosis in these models (except the PEPD injection experiment; Extended Data Fig. 8b,c). In agreement with this, EFA performed in chow-fed mice showed that factor 3, clustering AT fibrosis-related variables, had negative loading of AT PEPD activity and BW (Extended Data Fig. 8a). This inverse relationship was observed between AT fibrosis and PEPD activity when EFA included chow-fed and HFD-fed animals (factor 4; Fig. 7a). We then transformed the data to plot and visualize in two dimensions the relationship between the variables (Fig. 7b and Extended Data Fig. 8d) and the animal models. We observed that CBZ-Pro and BMT mice diverged in dimension 4 (corresponding to factor 4), in which fibrosis covaried negatively with both PEPD activity and serum level, that is, fibrosis could be explained by the absence of PEPD (Fig. 7c). This association is consistent with evidence that, in contrast to BMT KO mice with no PEPD released from macrophages, the CBZ-Pro-treated mice displayed a higher release of PEPD. Notably, they also differed in terms of insulin sensitivity status. Therefore, these data strengthen the role of PEPD released from macrophages in promoting IR. Factor 3, clustering ScW vascular fibrosis and glucose levels, had high loading of PEPD serum levels (Fig. 7a). Interestingly, Pepd HET mice fed a HFD 58% and PEPD-injected mice overlapped in this dimension (dim 3), suggesting similarities between these two models regarding the role of serum PEPD in promoting AT fibrosis (Fig. 7c). Of note, these two models also shared AT IR. However, Pepd HET mice fed HFD 58% and PEPD-injected mice diverged in dimension 1 (corresponding to factor 1), in which the insulin sensitivity status could be explained by adiposity (Fig. 7a,d). Indeed, while both models displayed AT IR, Pepd HET mice fed a HFD were obese, while PEPD-injected mice remained lean. Finally, Pepd KO mice and both CBZ-Pro/ PEPD-injected mice diverged on dimension 2 (corresponding to factor 2) in which GnW fibrosis and FFA levels covaried with PEPD serum levels, that is, AT fibrosis and higher FFA levels could be explained by higher PEPD serum level (Fig. 7a,d). Therefore, Pepd KO mice differed from CBZ-Pro mice and PEPD-injected mice because displaying higher PEPD serum levels exacerbated metabolic complications and fibrosis deposition in GnW.

Discussion
Departing from human GWAS identifying PEPD as a potential causal gene of IR and dyslipidaemia 8,9,11 , we focused on its role in metabolic organs in the context of obesity. In obesity, PEPD was downregulated exclusively in AT, notably in the gonadal depot, where we posited that it would play a central role, but not in skeletal muscle or liver. A critical observation was that the lower the PA in GnW (or the visceral depot in humans), the higher the PEPD levels detected in the serum of obese mice and humans. The liver appeared to be an essential organ contributing to serum levels of PEPD. However, gonadal fat was the only tissue where PEPD release is altered by HFD. We, therefore, hypothesized that the higher PEPD serum level observed in obese participants might reflect the release of PEPD from the GnW. Our in vivo and in vitro time-course HFD and LPS treatments, respectively, suggest that the release of PEPD might occur secondary to the decreased PA. In line with this hypothesis, our animal/cellular models in which PEPD was pharmacologically/nutritionally manipulated (HFD, CBZ-Pro or LPs-treated macrophages) showed that when the PA was inhibited, the PEPD release (in serum or the extracellular compartment) was increased. Conversely, in genetically manipulated mice-that is, in both PEPD total KO and macrophage-specific KO (BMT model), the Pepd gene had been ablated. Therefore, both PEPD activity and release are suppressed, either globally (totally in the KO and partially in the HET mice) or in the HCs (BMT model). These designs justify why the decreased PA in PEPD genetic models is not associated with an increased release of PEPD at the systemic level. Of note, the serum level of PEPD changed in the BMT KO mice that were fed a HFD. Collectively these data suggest that, under physiological conditions, the amount of PEPD released from/by the macrophage population might be quantitatively marginal compared to the contribution of other organs/cell types. However, under pathogenic conditions, such as a HFD, the expected increase in PEPD serum level was blunted suggesting that macrophages might play a more relevant role in contributing to the elevated serum PEPD observed in obesity. We cannot completely exclude that other cells in the haematopoietic fraction might also contribute to the phenotype of the BMT KO mice, a question that will require further investigation.
The specific routes and mechanisms regulating the release of PEPD from cells still need to be defined. PEPD has been reported in several large-scale profiling studies to be secreted in exosomes [32][33][34][35] . Inflammation (for example, obesity in vivo and LPS in vitro) could also be a possible trigger. There is some evidence supporting a role for reactive oxygen species and reactive nitrogen species regulating the secretion of PEPD and its intracellular peptidase activity 36-38 . Our results indicate that the extracellular PEPD exacerbated fibro-inflammation and IR in obesity. However, whereas PEPD serum levels were higher in obese mice, their serum PA was not modified, suggesting that the extracellular effect was independent of the peptidase activity of PEPD. While Pepd HET, BMT KO and CBZ-Pro-treated mice shared decreased intracellular PEPD enzymatic activity and increased fibrosis in AT, they diverged in other metabolic aspects. The CBZ-Pro-treated mice had a low residual enzymatic activity level and the highest increase in secreted PEPD, exhibiting the most severe metabolic phenotype in terms of AT fibro-inflammation and metabolic disorders. In contrast, when fed a chow diet, Pepd HET and BMT KO mice did not exhibit increased PEPD in the serum and showed a milder metabolic phenotype. Pepd HET mice only showed mild metabolic dysfunction when fed a 58% HFD, known to promote PEPD release from the AT further. Using purified PEPD exogenously, we showed that in vivo and in vitro PEPD treatment promotes AT fibrosis and IR. Thus, we propose that 'decreased PA' and 'increased PEPD serum level' independently induce metabolic complications.
Our results conclude that dysregulation of PEPD elicits a dual role in mediating AT fibrosis and metabolic risk through complementary mechanisms. This conclusion was obtained by integrating the phenotypical data from mouse in vivo models of dysregulated PEPD activity and/or secretion. This study provides a strong rationale for measuring serum PEPD in obese individuals to identify and stratify those at a higher metabolic risk by recognizing their susceptibility to AT fibrosis and inflammation.

Methods
Human studies. Age, BMI and glycaemic status of the different cohorts can be found in Supplementary Fig. 1a.

Cohorts 1 and 3.
In cohort 1, we analysed 84 VsW and 70 ScW samples from participants from Dr Josep Trueta University Hospital (Girona) with a wide range of adiposity (BMI between 20 and 68 kg/m 2 ). In cohort 3, 46 VsW and 36 ScW samples from morbidly obese participants (BMI > 35 kg/m 2 ) were analysed. Participants were of European ancestry with stable BW for 3 months. Participants were studied in the post-absorptive state, had no systemic disease other than obesity, and were free of any infections in the month before the study. Specific biochemical tests excluded liver diseases (tumoral disease and hepatitis C virus infection) and thyroid dysfunction. All participants gave written informed consent, which was validated and approved by the ethical committee. Samples were partially provided by the FATBANK (CIBEROBN) coordinated by the IDIBGI Biobank (Biobanc IDIBGI, B.0000872), Spanish National Biobanks Network, processed following standard operating procedures with approval of the Ethics, External Scientific and FATBANK Internal Scientific Committees. Subcutaneous and visceral AT samples were obtained from cholecystectomy, abdominal hernia and gastric bypass surgical procedures.
Cohort 2. ScW and VsW (omental) biopsy samples were obtained from severely obese (BMI > 35 kg/m 2 ) and lean participants (BMI ≤ 25 kg/m 2 ) undergoing elective surgery. Obese individuals who were candidates for bariatric surgery were studied in Paris following the Helsinki Declaration and Ethics Committee of Clinical Research (CPP Ile-de-France 1, Fibrota study, clinical trial no. NCT01655017). Patient phenotyping is described in ref. 39 . Blood sampling was performed fasted 1 month before the surgery. The same surgeon obtained paired ScW and VscW samples. Among glucose-intolerant and diabetic participants (n = 15), 7 were treated with metformin, 3 with insulin and 2 with a GLP1 agonist. Exclusion criteria included anaemia, abnormal thyroid-stimulating hormone, human immunodeficiency virus and/or hepatitis C virus infection, severe hepatic and/or renal failure and inflammatory disorders. Signed informed consents were obtained in all lean and obese individuals who agreed with ethics regulations.
Animals. Male C57BL/6J mice aged 8-10 weeks old were purchased from Charles River. Pepd KO mice were generated by the Wellcome Trust Sanger Institute Mouse Genetics Project on a C57BL/6J background by mating heterozygous mice. Pepd homozygous (null and WT) and heterozygous pups were observed in normal Mendelian ratios. In respect to the 3R, only a subpopulation of the pups born was selected for each study. Male Pepd WT, HET and KO mice aged 8-10 weeks old were used for each experiment. Male WT and leptin-deficient mice aged 8-10 weeks, Lep Ob/Ob , were on a C57BL/6 background. Details of the mouse models are provided in Supplementary Table 9. This research has been regulated under the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body under pathogen-free conditions and housed according to UK Home Office guidelines and carried out in the Disease Model Core unit. Animals were housed in groups of 3-4 mice per cage in a temperature-controlled room (22 °C) with a 12-h light-dark cycle, with 55% relative humidity and ad libitum access to food and water. A standard chow diet (DS-105, Safe Diets) was administered to all animals from weaning, consisting of 64.3% carbohydrate, 22.4% protein and 13.3% lipid of total calories. Only male mice were used for in vivo experiments, isolation of primary cells (that is, adipose cells, HSCs and muscle fibroblasts) and preparation of BMDMs.
Generation of bone marrow chimeras. C57BL6/J mice at 10 weeks of age received a sublethal dose of whole-body irradiation (9 Gy). The day after irradiation, donor Pepd KO mice were culled, and their femurs and tibias were removed aseptically. Marrow cavities were flushed in RPMI medium, and single-cell suspensions were prepared. The irradiated recipients received 1 × 10 7 bone marrow cells in 0.1 ml of PBS by tail-vein injection. During 4 weeks after BMT, Bactrim (Roche) was added to drinking water. After two additional weeks, mice were switched to a 58% HFD. Mice were culled 16 weeks later to collect blood and tissues.
Insulin injection. Male 8-to 10-week-old WT, pepd HET and Pepd KO mice were injected intraperitoneally with saline solution or insulin (2 U per kilogram BW) for 15 min.
Pharmacological inhibition of PEPD. Male 8-week-old C57BL/6J male mice were first fed with chow or HFD 58% for 10 weeks. CBZ-Pro-supplemented pellets (both chow and HFD 58%) were prepared daily by spraying and mixing a solution of CBZ-Pro (2 mM) dissolved in 70% ethanol. Mice consumed an approximatively daily dose of 60 mg per kg BW of CBZ-Pro, a dose previously used in other murine studies 19 , with treatment lasting for 6 weeks, and control mice were offered regular pellets. CBZ-Pro does not alter viability or promote toxic effects in mice 19 . CBZ-Pro-treated mice did not show evidence of hepatotoxicity/liver damage (alanine aminotransferase or ASAT levels; Extended Data Fig. 2a).
PEPD treatment. Male 8-to 10-week-old C57BL/6N mice were injected intraperitoneally with the purified PEPD protein (0.2 mg per kg BW) or saline three times a week for 6 weeks 40 . ITTs and DTTs were performed at weeks 4 and 5. PEPD treatment did not modify the ASAT levels (Extended Data Fig. 7e).

Body composition.
Fat and lean masses were calculated by time-domain nuclear magnetic resonance using a minispec Live Mice Analyzer LF50 (Bruker).
Glucose and insulin tolerance tests. For the GTT, mice were fasted overnight (16 h) with free access to drinking water. Glucose was administered intraperitoneally (2 g per kg BW), and blood glucose levels were monitored from the tip of the tail with a glucometer. For ITTs, insulin was administered intraperitoneally (0.75 mU per gram BW), and blood glucose was measured at various timepoints after injection.

Serum biochemistry.
Triglycerides and transaminases were measured on the Dimension RXL analyser (Siemens Healthcare) or Perkin Elmer DELFIA using reagents and calibrators (Siemens). FFAs were measured using the Roche Free Fatty Acid Kit (half-micro test, 11383175001). Insulin was measured using electrochemical luminescence immunoassay on the MesoScale Discovery immunoassay platform.
Explants for conditioned medium. Approximately 100 mg of freshly dissected BAT, ScW, GnW and liver pieces from 30-week-old mice in chow-fed and HFD-fed conditions, or VsW from human participants cut into fine pieces were incubated for 6 h at 37 °C in 5% CO 2 in DMEM with 5% heat-inactivated FBS, 20 mM HEPES, 100 units per ml penicillin, 100 μg ml −1 streptomycin and 20 mM l-glutamine (1 ml of medium per 100 mg of tissue).
Bone marrow-derived macrophage preparation and treatments. Femur and tibia bones from 10-to 16-week-old C57BL/6 mice or Pepd WT, HET and KO mice were isolated and cleaned, and 10 ml of RPMI 1640 was flushed through each bone. Total bone marrow cells were passed into a 100-μm cell strainer and counted using Countess II automated cell counter (Thermo Fisher). Cells were spun (400g, 5 min), resuspended in BMDM culture medium (RPMI 1640 supplemented with 20% of L929-conditioned cell medium, 10% heat-inactivated FBS and 1% penicillin-streptomycin). To make L929-conditioned cell medium, L929 cells (CCL-1, American Type Culture Collection (ATCC)) were seeded in DMEM supplemented with 10% heat-inactivated FBS, 100 U ml −1 penicillin-streptomycin and 2 mM l-glutamine (Sigma) at a density of 250,000 cells per 50 ml of medium per T175 tissue culture flask. The medium was collected after 1 week of culture, and then 50 ml of fresh DMEM supplemented with 10% heat-inactivated FBS, 100 U ml −1 penicillin-streptomycin and 2 mM l-glutamine were added onto cells and harvested 1 week later. Batches obtained after the first and second weeks of culture were mixed at a 1:1 ratio, aliquoted and stored at −20 °C.
Total bone marrow cells were seeded in 10-cm non-culture-treated plates (Falcon) at a density of 5 × 10 6 cells per plate per 10 ml of macrophage differentiation medium and cultured for 7 d at 37 °C in 5% CO 2 . On day 5 of differentiation, the medium was removed and replaced with 10 ml of fresh BMDM culture medium. On day 7, BMDMs were detached using ice-cold PBS-EDTA 1 mM, spun (400g, 5 min) and resuspended in fresh BMDM culture medium. Differentiated BMDMs were counted using a Countess II automated cell counter, and cell concentration adjusted to 5 × 10 5 cells per ml. Immediately after, cells were plated for experiments at the following densities: 100 μl per well of a 96-well plate, 500 μl per well of a 24-well plate, 1 ml per well of a 12-well plate, 2 ml per well of a 6-well plate and 10 ml for each 10-cm plate. Cells were incubated for 16-24 h after plating before conducting experiments.
Macrophage purity was routinely tested by the expression of CD11b and F4/80 by flow cytometry, and 93-97% of the cells expressed high CD11b and F4/80 levels after 7 d of differentiation.
After differentiation, BMDMs were cultured in 12-well-culture plates (5 × 10 5 cells) for 24 h in BMDM medium before 6-24 h of stimulation with LPS (100 ng ml −1 ), DEX (100 nM), IL-4 (10 ng ml −1 ), purified PEPD (250 nM), Gly-Pro (10 mM) or CBZ-Pro (6 mM) and stored at −80 °C before RNA extraction or prolidase assay. Erlotinib (5 µM) was added to the culture medium 2 h before and during the treatment with PEPD. We validated PEPD-specific effects by measuring Cox2 expression in BMDMs, reported as a PEPD target gene 15 . In addition, we discarded the potential cytotoxic effects of purified PEPD on BMDMs and its potential endotoxin contamination by boiling the purified protein and tested it on BMDMs for Cox2 induction (Extended Data Fig. 7a-c).
Cell culture. RAW264.7 macrophage culture and transfection. RAW264.7 cells (TIB-71, ATCC) were cultured in DMEM (4.5 g l −1 glucose) supplemented with 10% heat-inactivated FBS, 100 U ml −1 penicillin-streptomycin and 2 mM L-glutamine (Sigma). In total, 100,000 cells per well were seeded in a 24-well plate for knockdown experiments. The following day, cells were transfected using Lipofectamine LTX Reagent with PLUS (Invitrogen, Thermo Fisher) according to the manufacturer's instructions. ON-TARGET plus siRNA against EGFR or CTR were obtained from Dharmacon (Horizon). At 24 h after transfection, RAW264.7 cells were treated 24 h with purified PEPD 250 nM. Erlotinib (5 µM) was added to the culture medium 2 h before and during the treatment with PEPD.
Primary adipose cells. GnW from 10-week-old Pepd WT, HET and KO mice was dissociated by collagenase treatment isolating unilocular adipocytes from the SVF. Floating adipocytes were collected and used for lipolysis assay (on the day of isolation) or culture in hydrogel (48 h) 6 to measure IL-6 secretion in response to purified PEPD. Preadipocytes were isolated from the SVF and differentiated into adipocytes (day 9) accordingly to the protocol described by Lacasa et al. 42 .
Primary muscle fibroblasts. Both hind limbs from 10-to 12-week-old C57BL/6 mice were removed by dislocating hip joints and were dissected in 5 ml of digestion medium (30 mg collagenase II, 36 mg dispase II, 1 M HEPES 2.5%, DMEM high glucose and 15 ml 1% penicillin-streptomycin). Enzymatic digestion was carried out for 20 min in a 37 °C shaking bath and stopped with DMEM (20% FBS). Sequentially, the cell mixture was passed through 100-µm, 70-µm and 40-µm cell strainers and pelleted at 400g for 5 min at room temperature (RT). Cells were resuspended in 12 ml of seeding medium (Ham F12, 20% FBS, 10% horse serum, 1% l-glutamine, 1% penicillin-streptomycin and 10 ng ml −1 basic fibroblast growth factor and plated on an uncoated 10-cm petri dish for 1 h at 37 °C, 5% CO 2. The medium was changed for fresh seeding medium and renewed every 2 d until reaching 70% confluence. Fibroblasts were grown at a confluence of 50,000 cells per well in 12-well plates (for RNA) or 25,000 cells per well in 24-well plates (on coverslips, for IF). Fibroblasts were pretreated or not with erlotinib (5 µM) before treatment with purified PEPD protein (250 nM) for 24 h (RNA) or 5 d (IF).
Lipolysis assays. Primary mature adipocytes were isolated from GnW of C57BL/6 mice and incubated in Krebs-Ringer bicarbonate buffer with an increasing dose of noradrenaline (from 10 −10 to 10 −5 M) for 2 h at 37 °C. Glycerol was measured as an index of lipolysis by using free glycerol reagent (Sigma) against a glycerol standard curve. Absorbance was read at 515 nm, and proline concentration was calculated using proline standards ranging from 0.5 μg to 32 μg. Enzyme activity was reported in micromoles of proline released per minute per milligram of protein.
Gas chromatography-mass spectrometry analysis of amino acids. Plasma and GnW explants samples were analysed for free amino acid concentrations using the EZ:faast GC-MS Kit (KGO-7166 Phenomenex).
Extraction. AT (~100 mg) was weighed and amino acids were extracted into 1 ml of 75% 0.01 M HCL and 25% acetonitrile using a FastPrep (MP Biomedical). All liquid was removed to a 1.5-ml microcentrifuge tube and centrifuged at 16,000g for 10 min. Following centrifugation, the upper lipid layer was discarded and the lower acetonitrile/aqueous layer was removed to a fresh microcentrifuge tube and the centrifugation step was repeated. Samples were transferred to another fresh microcentrifuge tube and dried overnight using a SpeedVac before being resuspended in 100 µl of 0.9% saline by repeatedly pipetting up and down and vortexing (3×) for 5 s. After resuspension, samples were centrifuged at maximum speed for a further 10 min and analysed following the EZ:faast protocol.
Modified EZ:FaaST protocol. Amino acids were analysed using the EZ:faast free (physiological) amino acid analysis kit (KGO-7166) for gas chromatography-mass spectrometry (GC-MS) according to the manufacturer's instructions up to the chromatography step. Briefly, 100 µl of serum or WAT extract was mixed with 100 µl of reagent 1, which contained 0.2 mM norvaline as an internal standard. Samples were drawn into the proprietary sorbent tip and then washed by drawing a further 200 µl of reagent 2 followed by air through the tip until dry. Samples were eluted by attaching a new syringe to the tip, drawing the eluting medium (200 µl total) into the tip until the sorbent granules were wet and then ejecting the granules. Samples were then derivatized by adding 50 µl of reagent 4 using a Drummond Dialamatic Microdispenser and vortexing for 5-8 s. Samples were allowed to stand for 1 min, then vortexed again and allowed to stand for a further min. In total, 100 µl of reagent 5 was added and samples were vortexed for 5 s and allowed to stand for 1 min. The upper layer was transferred to an autosampler vial and dried under nitrogen for 10 min. Samples were resuspended in 100 µl of reagent 6 and transferred into an insert in the same autosampler vial. The samples then proceeded to chromatography.
Chromatography. GC-MS analysis was performed using an Aglient 7890B GC and an Agilent 5977A MSD using a 0.25-mm ID, 0.25-µm film thickness 15 M DB-1701 (122-0712, Agilent) column. The GC conditions were as follows: Inlet conditions: temperature, 250 °C; split 10:1 for serum, splitless for Gly-Pro determination and AT; inlet liner Liner, Phenomenex AG0-4680: 900 μl (FocusLiner)-part of kit; column flow, 2 ml min −1 ; injection volume, 1 µl. Data were acquired using MassHunter Workstation Software. Peak integration and quantification were performed using MassHunter Workstation Quantitative Analysis software (vB.07.00, Agilent) selecting ions for each amino acid specified in the EZ:faast protocol. Data were analysed by first normalizing all data points to the Norvaline internal standard. Amino acid concentrations were then determined using a three-point standard curve made using 25, 50 and 100 µl of the 0.2 mM standards provided in the EZ:faast kit. AT extracts and serum were expressed as nanomoles per 100 µl of sample.
Histological analysis. AT and liver samples were fixed in 4% paraformaldehyde for 24 h, embedded in paraffin, sectioned into 5-μm sections and processed for Sirius (fibrosis) or H&E (liver steatosis) staining. The slides were scanned (Microscopy Zeiss Axioscan Z1 Slidescanner) and processed for fibrosis (Sirius staining excluding vessels) and steatosis (vacuole percentage area) quantification using HALO image analysis software (Indica Labs).
Confocal analysis. Cells were fixed in 4% paraformaldehyde for 1 h at RT, then transferred to PBS, and stored at 4 °C until IF analysis. Samples were blocked in 1 M glycine and PBS with 3% BSA and 0.1% Triton X-100 and then incubated overnight (4 °C, in agitation) with primary antibody diluted in blocking buffer (PBS 3% Albumin, pH 7.4). Antibodies used were anti-αSMA 1A4 (1:200 dilution; Sigma, A2547) and anti-collagen I (1:200 dilution; Abcam, ab21286). After washing in PBS and blocking, cells were incubated with the secondary antibodies conjugated to Alexa Fluor dyes, either Alexa Fluor 555-conjugated anti-rabbit (1:250 dilution; Life technologies, A21428) or Alexa Fluor 488-conjugated anti-mouse (1:250 dilution; Life technologies, A21202), BODIPY 493/503 (1:2,000 dilution; Life science, D-3922), used as a stain for neutral lipids, and Phalloidin-iFluor 594 (1:200 dilution; Abcam, ab176757), used to stain the actin filaments, for 1 h at RT in blocking buffer, followed by Hoechst 33342 (1:2,000 dilution; Thermo Fisher, H3570) for 5 min, and multiple PBS washes. Samples were mounted and imaged blind using a Zeiss LSM 510 Meta Confocal microscope with LSM 3D software (Carl Zeiss). At least three images per sample for each stain were used for quantification and was performed blind. Staining quantification was performed using Fiji software (https://imagej.net/software/fiji/). A detailed list of the antibodies used for this study is presented in Supplementary Table 9.
Hydroxyproline assay. HPro measurement was done using a HPro colorimetric assay (BioVision) as previously described 45 . Briefly, frozen fat is weighted and heated in 6 N HCl at 110 °C overnight in sealed tubes, as 10 μl of HCl per mg of WAT. Ten microliters are evaporated before incubation with chloramine-T and p-dimethylaminobenzaldehyde (DMAB) at 60 °C for 90 min. The absorbance was read at 560 nm, and the concentration was determined using the standard curve created with HPro. ELISA assays. Mouse and human PEPD protein concentrations were measured using an ELISA kit for Mouse Xaa-Pro dipeptidase (PEPD) ELISA kit (CSB-EL017784MO, CUSABIO) and Human PEPD (Peptidase D) ELISA Kit (E-EL-H5575.96, Elabscience), respectively, in tissue explants (from which debris was removed by centrifugation) and serum according to the manufacturer's instructions. Leptin protein concentration in AT explant media was measured using an ELISA kit for murine leptin (R&D systems, DY498). IL-6 protein concentration in conditioned media from mature adipocytes isolated from GnW was measured using an ELISA kit for mouse IL-6 (R&D systems, DY406). A standard curve was prepared according to the manufacturer's instructions, and the value associated with an unconditioned medium blank was subtracted from that of conditioned medium.
RNA extraction and real-time PCR. RNA from cells was extracted using RNeasy Mini columns (Qiagen) according to the manufacturer's instructions. RNA was harvested from frozen tissue using RNA-STAT-60TM (AMS Bio) and purified by chloroform extraction and isopropanol precipitation. Reverse transcription was performed using the Reverse Transcriptase System (Promega) according to the manufacturer's instructions. Real-time PCR was carried out using TaqMan or Sybr Green reagents using an Abi 7900 real-time PCR machine with default thermal cycler conditions. Primer sequences are described in Supplementary  Table 9. Reactions were run in duplicate, checked for reproducibility, and then averaged. A standard curve generated from a pool of all cDNA samples was used for quantification. The expression of genes of interest was normalized using the geometric average of four housekeeping genes (18 s, 36b4, βactin and B2m), and data are expressed as arbitrary units.
Regarding human samples (cohorts 1, 2 and 3), RNA purification, gene expression procedures and analyses were performed, as previously described 46,47 . Total RNA was briefly extracted and purified using the RNeasy Lipid Tissue Mini kit, and Agilent Bioanalyzer checked the integrity. Total RNA was quantified using a spectrophotometer. The same amount of total RNA was reverse transcribed to cDNA from all samples using a High-Capacity cDNA Archive kit following the manufacturers' instructions. Gene expression was assessed using a LightCycler 480 real-time PCR system, using TaqMan and SYBRgreen technology suitable for relative gene expression quantification. The commercially available and pre-validated TaqMan primer/probe sets used are described Supplementary  Tables 9 and 10. RNA sequencing. Library preparation and sequencing. Total RNA of GnW and liver was extracted using the miRNeasy mini kit (Qiagen) according to the manufacturer's instructions. For each experiment, 4-10 independent biological repeats were used.
Total RNA was quality checked (RIN score > 7) via the Agilent Bioanalyser 2100 system, using the Agilent RNA 6000 Nano Kit. A total of 1 µg of RNA was used to construct barcoded sequencing libraries with the TruSeq Stranded mRNA HT Sample Prep Kit (Illumina) following the supplier's instruction. All the libraries were validated using the Agilent Bioanalyser DNA 12000 and then multiplexed and sequenced on two lanes of an Illumina HiSeq 4000 at SE50 at CR-UK Cambridge Institute Genomics Core Facility.
Processing of RNA-seq data. RNA-seq reads (Supplementary Table 1) were mapped to the most recent ENSEMBLE version GRCm38.p5 of the mouse reference genome sequence (GRCm38.p5) using STAR (v2.5.1b) 48 including the annotations as hints for exon-intron borders. Reads were considered mapped if the similarity was at least 95% over at least 90% of the read length as previously described 49 . FeatureCounts (v1.5) 50 was applied for the generation of count tables based on the mapping files. Customized Python scripts 49 were deployed for downstream processing, including the raw counts' normalization to the total number of assigned reads per gene (TPMs) and the combined exon length (fragments per kilobase of exon per million mapped fragments), respectively.
Gene expression and pathway enrichment analyses. Raw counts were subjected to differential gene expression analysis via DESeq2 (ref. 51 ) and different R packages (Supplementary Information). Genes that showed raw counts lower or equal to 2 in 50% of all samples were removed before the differentially expressed analysis. The Wald test was applied to extract DEGs (Supplementary Tables  2, 3 and 6-8). Obtained DEGs were annotated via customized Python scripts. Pathway enrichment analyses were performed with PIANO 52 , using the gene-set collection C2 retrieved from the Molecular Signatures Database (MSigDB) 53,54 (Supplementary Tables 4 and 5). The Bonferroni and Holm method was applied to correct for multiple testing.
Array-based detection of phosphorylated receptor tyrosine kinase. The Proteome Profiler Mouse Phospho-RTK Array Kit (R&D Systems, ARY014) was used to screen for the phosphorylation level of receptor tyrosine kinase in BMDMs in response to purified PEPD, according to the manufacturer's instructions. Briefly, the array membranes were incubated for 1 h with an array blocking buffer before incubation overnight at 4 °C on an orbital shaker with 1.5 ml of cellular extract. The membrane was then washed and incubated for 30 min with a streptavidinhorseradish peroxidase solution. Membranes were exposed digitally using the ChemiDoc MP System (Bio-Rad), and spots were quantified using Fiji software (https://imagej.net/software/fiji/). One condition corresponds to a pool of cellular extracts from four independent experiments. All the arrays were measured three times; each spot was normalized to the positive controls. The results are presented in a heat map and considered relevant when the fold variation was >1.3 or <0.6.

Cytotoxicity assays.
To determine the cytotoxic effect of compounds on macrophages, cells were seeded in RPMI 1640 medium without FBS supplemented at a density of 15,000 cells per well in wells of a 96-well plate. Cells were treated with the given compounds at the stated concentrations for 24 h, and cytotoxicity was measured using an LDH-Cytotoxicity Calorimetric Assay Kit (BioVision) according to the manufacturer's instructions.
Data analysis. The number of animals or independent experiments was determined based on pilot data and indicated in the figure legends. No statistical methods were used to predetermine the total number of animals needed for this study. Animal allocation to cages (four mice per cage) and diet were not randomized as we ensured each cage had mice from the three genotypes and each group (chow and HFD, control and CBZ-Pro) had the same average of initial BW. All data from experiments are summarized by their mean and standard error. The number of replicates is reported in the figure legends. When the pairwise comparison results are expressed as a fold-change value, it is declared in the figure legends to what value the data were normalized. Statistical analyses were performed using GraphPad Prism (v8.0.2). Comparisons between two groups were conducted using an unpaired t-test. Comparison between more than two groups was conducted using a one-way ANOVA followed by appropriate post hoc multiple-comparison tests. Comparisons between more than two groups and factors were conducted using a two-way ANOVA followed by appropriate post hoc testing. Pearson's coefficients were evaluated to estimate the correlation between data series.
Data were excluded when identified as outliers using Grubbs' test (https:// www.graphpad.com/quickcalcs/Grubbs1.cfm). The animals were randomly assigned to the different experimental settings (GTT and ITT). Areas under the ROC curves were determined for each variable to identify the predictors of AT fibro-inflammation and IR/T2D. A ROC curve is a plot of sensitivity (true positive) versus 1-specificity (false positive) showing the ability of a biomarker (PEPD level) to discriminate between true positives (for example, insulin resistant) and true negatives (for example, insulin sensitive). The best marker has a ROC curve shifted to the left with the area under the curve close to unity 55 . The Youden index was calculated (sensitivity + specificity−1) to determine the optimal cut-off values for fibro-inflammatory status or IR indices. The values for the maximum of the Youden index were considered the optimal cut-off points using the Web-tool easyROC 56 .
Exploratory factor analysis. EFA is an unbiased statistical method that establishes the cluster of biological variables with high loadings (correlation equivalent) on each specific factor in a reduced number of underlying variables ('factors') considered as 'superfamilies' of variables 57 . EFA was conducted to determine the possible latent structure of the variables (listed; for example, in Fig. 1p, rows of the depicted matrix) measured in each animal model, that is, Pepd CBZ-Pro-treated mice, BMT mice and PEPD-treated mice. Factor analysis identifies a minimum number of new variables (factors), which are linear combinations of the original (measured) ones, such that the new (fewer) variables contain most or all the information and can facilitate the interpretation of a complex multivariate scenario. First, a matrix of correlation coefficients is computed, and a set of principal components (factors) is extracted from it. The relationships between the original variables and the factors are expressed in terms of 'factor loadings' , which estimate the degree of correlation between the original variables and the factors.
We determined whether our data were appropriate for EFA using the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett's test of sphericity. Data were considered appropriate if the KMO > 0.7 and Bartlett's test was significant at P < 0.05. To determine the number of factors to retain, we decided to rule out the components associated with eigenvalues less than 1. The parallel analysis confirmed that our choice was viable. Factors extraction was computed using the maximum-likelihood-estimate method, implemented in the function 'fa' of the psych R package 58 . Plots were produced using either in-house scripts or built-in functions of the FactoMineR package in R 59 .
Principal-component analysis. Principal-component analyses of the Lukk and the own dataset were performed in R v3.1.2 using the prcomp function of the stats package 60 .
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
DEGs and pathways from the RNA-seq analysis are available in Supplementary Tables 1-10. The RNA-seq dataset is deposited in the Gene Expression Omnibus under accession number GSE198358. Source data are provided with this paper. All other raw data are available upon request.