Obesity-linked PPARγ S273 phosphorylation promotes insulin resistance through Growth Differentiation Factor 3

Summary

The thiazolidinediones (TZDs) are ligands of PPARγ that improve insulin sensitivity, but their use is limited by significant side effects. Recently, we demonstrated a mechanism wherein TZDs improve insulin sensitivity distinct from receptor agonism and adipogenesis: reversal of obesity-linked phosphorylation of PPARγ at Serine 273. However, the role of this modification hasn’t been tested genetically. Here we demonstrate that mice encoding an allele of PPARγ which cannot be phosphorylated at S273 are protected from insulin resistance, without exhibiting differences in body weight or TZD-associated side effects. Indeed, hyperinsulinemic-euglycemic clamp experiments confirm insulin sensitivity. RNA-seq in these mice reveals reduced expression of Gdf3, a BMP family member. Ectopic expression of Gdf3 is sufficient to induce insulin resistance in lean, healthy mice. We find Gdf3 inhibits BMP signaling and insulin signaling in vitro. Together, these results highlight the diabetogenic role of PPARγ S273 phosphorylation and focuses attention on a putative target, Gdf3.

eTOC blurb

Hall et al. show that mice lacking phosphorylation at Serine 273 of PPARγ are protected from developing insulin resistance in response to high fat diet feeding, associated with dramatically reduced levels of Gdf3. Gdf3 in turn is sufficient to cause impaired insulin signaling both in vitro and in vivo.

Graphical Abstract

Introduction

Type 2 diabetes (T2D) is one of obesity’s most tightly linked comorbidities, and its prevalence is rising in parallel with the global obesity epidemic (Boden, 1997). T2D results from elevated glycemia consequent to insulin resistance and beta cell failure (Stratton et al., 2000). Although there are several available glucose-lowering agents, few address the underlying pathophysiology of this disease. The thiazolidinediones (TZDs) are a class of insulin-sensitizing anti-diabetic drugs capable of delaying or even preventing the onset of diabetes (DeFronzo et al., 2011). Despite potent insulin sensitization, the clinical use of TZDs has declined due to adverse side effects (Soccio et al., 2014). As ligands for the nuclear receptor peroxisome proliferator activated receptor-γ (PPARγ), the therapeutic actions of TZDs have been assumed to result from their classical agonism towards PPARγ. This agonism drives adipocyte differentiation and lipogenesis in fat tissues (Lehmann et al., 1995). PPARγ is most highly expressed in adipose tissue, where it was first characterized as a master regulator of adipocyte differentiation and function (Tontonoz and Spiegelman, 2008). Nonetheless, activation of PPARγ improves insulin sensitivity owing to its diverse roles in both adipocyte and non-adipocyte cells, including macrophages, T-cells, myocytes, and other cell types (Bouhlel et al., 2007; Cipolletta et al., 2012; Hevener et al., 2007; Odegaard et al., 2007). PPARγ is intimately linked to insulin sensitivity, and genetic variants in the PPARG locus associate with diabetes risk (Altshuler et al., 2000; Gouda et al., 2010; Majithia et al., 2014).

We have recently shown that, in addition to PPARγ agonism, TZDs have a second biochemical role, blocking phosphorylation on PPARγ at serine 273 (Choi et al., 2010). Importantly, phosphorylation of this residue in adipose tissue increases in murine models of obesity, and its reversal correlates with the anti-diabetic effects of TZD administration in obese patients and rodents (Choi et al., 2010; Mayoral et al., 2015). Moreover, synthetic ligands have been generated that prevent PPARγ S273 phosphorylation in the absence of PPARγ agonism. These non-agonist PPARγ ligands exhibit profound glucose-lowering effects with reduced side effects, such as weight gain and fluid retention (Amato et al., 2012; Choi et al., 2011; Choi et al., 2014). Likewise, inhibition of the major protein kinase responsible for PPARγ S273 phosphorylation, extracellular signal-regulated kinase (ERK), improves insulin sensitivity in obese animals (Banks et al., 2015). Understanding the specific mechanisms linking insulin sensitivity with reversal of obesity-linked PPARγ S273 phosphorylation is thus attractive as a new therapeutic strategy for type 2 diabetes, as well as other conditions linked to insulin-resistance such as Polycystic Ovarian Syndrome.

Here, we specifically address the contribution of PPARγ S273 phosphorylation to the development of insulin resistance, using a genetically modified mouse model where PPARγ cannot be phosphorylated at S273. These data demonstrate that genetic prevention of PPARγ S273 phosphorylation is an insulin-sensitizing event and is associated with decreased levels of Gdf3 in adipose tissues and skeletal muscle. Gdf3 is therefore a potential biomarker of PPARγ S273 phosphorylation and insulin resistance. Neutralization of Gdf3 may provide an insulin-sensitizing therapeutic approach.

Results

Improved insulin sensitivity in mice with non-phosphorylatable PPARγ S273

Reversal of obesity-linked PPARγ S273 phosphorylation with small-molecule PPARγ ligands correlates with anti-diabetic properties in humans and mice. However, it is unknown whether genetic modification of this site will result in chronic improvements in insulin sensitivity. To test this, we have generated a genetic mouse model in which PPARγ cannot be phosphorylated at S273. Using homologous recombination, a single nucleotide was changed to cause a serine to alanine substitution at codon 273 of PPARγ (Suppl. Fig. 1A, 1B). Homozygous knock-in mice (PPARγ^S273A/S273A, hereafter referred to as PPARγ^A/A) were generated using C57BL/6 ES cells and were born at the expected Mendelian frequencies. Whereas wildtype (WT) mice made obese by exposure to a high fat diet contain PPARγ that is phosphorylated at S273 in adipose tissue, PPARγ from PPARγ^A/A mice is exclusively unphosphorylated (Suppl. Fig. 1C). PPARγ S273 phosphorylation increases with progressive obesity (Choi et al., 2010). Consequently, WT mice are predicted to have little PPARγ phosphorylation in the lean state, rendering them phenotypically identical to lean PPARγ^A/A mice. On a standard chow diet, PPARγ^A/A mice gain weight similarly, have comparable body compositions, and respond similarly to WT controls in glucose tolerance and insulin tolerance tests (Suppl. Fig. 2A–D). Accordingly, in the absence of stimuli to promote PPARγ phosphorylation, differentiation of primary adipocytes is unchanged between genotypes (Suppl. Fig. 3A). Gene expression of WT and PPARγ^A/A primary adipocytes revealed similar mRNA levels of the differentiation markers Pparg2, Fabp4, Cebpa, Lpl, and Cd36 (Suppl. Fig. 3B). These data confirm earlier work that modulation of PPARγ S273 phosphorylation is neither necessary nor sufficient to affect adipogenesis. We also examined baseline differences in expression of a gene set previously identified using PPARγ^−/− cells reconstituted with viruses expressing WT or S273A PPARγ (Choi et al., 2010). Surprisingly, expression of these genes was not significantly different in primary adipocytes differentiated ex vivo from WT and PPARγ^A/A mice (Suppl. Fig. 3C).

Given that PPARγ S273 is phosphorylated in the setting of high-fat feeding (Choi et al., 2010), we maintained PPARγ^A/A and WT littermate mice on a high-fat diet (HFD) to promote obesity and insulin resistance. PPARγ^A/A and WT mice experienced equivalent weight gain (Fig. 1A) with similar fat and lean body mass composition (Fig. 1B). As PPARγ activation with TZDs is associated with weight gain (Fonseca, 2003; Lu et al., 2011), we assessed energy balance by indirect calorimetry experiments in PPARγ^A/A mice. There were no significant differences observed for VO₂, VCO₂, calculated energy expenditure, food intake, body temperature, overall energy balance, locomotor activity, or ambulatory activity between genotypes (Suppl. Fig. 4A–4B). We did observe a small but significant decrease in RER in the PPARγ^A/A animals, which is suggestive of increased fatty acid oxidation (Suppl. Fig. 4A).

Figure 1. Improved insulin sensitivity in mice with non-phosphorylatable PPARγ S273.

Male wildtype (+/+) and PPARγ^A/A (A/A) mice were fed high-fat diet (HFD) for 16 weeks. A. Body weight of mice during HFD feeding (n = 6 +/+, 5 A/A). B. Body composition of mice at 16 weeks of HFD feeding. C. Glucose tolerance test (1 g glucose/kg body weight) after 15 weeks on HFD. D. Insulin tolerance test (2 U insulin/kg body weight) after 16 weeks on HFD. C-D. Right, area under the curve (AUC) for the respective test. Results are representative of three independent experiments. E. Fasting insulin levels in male mice on HFD for 16 weeks and fasted overnight (n = 6 +/+, 5 A/A). F. Glucose (3 g glucose/kg body weight) stimulated insulin levels at indicated timepoints for male mice on HFD for 17 weeks (n = 6 +/+, 5 A/A). G-L. Hyperinsulinemic-euglycemic clamps in HFD-fed mice (n = 4 +/+, 5 A/A). G. Glucose infusion rate (GIR). H. Whole body glucose uptake (GU). I. Glucose uptake into skeletal muscle. J. Glucose uptake into epididymal white adipose tissue. K. Percentage suppression of endogenous glucose production (EGP). L. Percentage suppression of free fatty acids. M-N. Insulin sensitivity as determined by insulin-stimulated Akt phosphorylation (Ser 473) in tissues of HFD-fed +/+ and A/A mice. M. Immunoblot analyses of skeletal muscle, liver, inguinal white adipose tissue (iWAT), and epididymal white adipose tissue (eWAT). N. Quantification of blots in M. Results are representative of at least two independent experiments. Data are presented as mean ± SEM; *, P < 0.05; and **, P < 0.01 by Student’s t test (A-L) and one-way ANOVA with Newman-Keuls Multiple Comparison test (N).

We next assessed the effect of PPARγ phosphorylation on glucose homeostasis in HFD-fed mice. Similar to effects observed in chow-fed mice, glucose tolerance was comparable between genotypes (Fig. 1C and Suppl Fig. 2C). In contrast, insulin tolerance testing in HFD-fed mice revealed PPARγ^A/A mice to be more responsive to insulin, as evidenced by a significantly enhanced glucose lowering response (Fig. 1D). To eliminate the possibility of pancreatic beta cell dysfunction, we assessed both fasting insulin levels and glucose-stimulated insulin levels in HFD fed mice and found no difference between genotypes (Fig. 1E and 1F). To better understand the insulin sensitivity phenotype, we performed hyperinsulinemic-euglycemic clamps on HFD-fed mice. We observed increases in the glucose infusion rate (GIR) required to maintain euglycemia in mutant mice compared to controls, indicating improved insulin sensitivity per se (Fig. 1G). The primary factor contributing to the increased GIR is elevated whole-body glucose utilization, with pronounced effects observed in skeletal muscle but not white adipose tissue (WAT) (Fig. 1H–1J). Of note, there was no significant difference in the ability of insulin to suppress endogenous glucose production (EGP) (Fig. 1K). These findings suggest that the improved insulin sensitivity is due to greater glucose uptake into tissues and is not due to decreased glucose production. The improved glucose uptake in muscle rather than fat is highly suggestive of tissue cross-talk. Importantly, we observed a three-fold increase in the ability of insulin to suppress non-esterified fatty acid (NEFA) levels in PPARγ^A/A mice (Fig. 1L).

We also examined biochemical indicators of insulin sensitivity and insulin signaling. PPARγ^A/A mice exhibited a modest but significant increase in insulin-stimulated AKT phosphorylation within skeletal muscle but not liver, inguinal WAT (iWAT), or epididymal WAT (eWAT) (Fig. 1M–1N). These results strongly support the notion that phosphorylation of PPARγ at S273 is sufficient to promote obesity-induced insulin resistance.

PPARγ^A/A and WT mice respond similarly to PPARγ agonist treatment

The TZD class of insulin sensitizing drugs serve both as agonists and to prevent phosphorylation of PPARγ S273. However, these drugs also have adverse effects, such as weight gain, fluid retention, and bone loss (Chilcott et al., 2001). Whether these side effects are completely separable from insulin sensitization is a key unanswered question. Multiple small-molecule non-agonist ligands have successfully separated beneficial from adverse effects (Amato et al., 2012; Choi et al., 2014; Stechschulte et al., 2016). To test whether this phenomenon holds true in a genetic model, we assessed PPARγ^A/A mice for TZD-associated side effects. First, adipocytes in eWAT of PPARγ^A/A mice had similar cell surface area to cells from control mice (Fig. 2A). In addition, PPARγ^A/A mice fed HFD showed no decrease in hematocrit, as an indicator of fluid retention (Fig. 2B), compared to controls. Importantly, using microquantitative computed tomography (microCT) we also observed no change in tissue mineral density (TMD) or trabecular bone volume fraction among genotypes (Fig. 2C–2D). Cortical bone parameters, including thickness, mineralization and periosteal perimeter, were similarly unchanged (data not shown). Thus, while mice harboring the S273A knock-in allele recapitulate the PPARγ agonist-like improvements in insulin sensitivity, they avoid the adverse effects typically associated with TZDs. Treatment of HFD-fed PPARγ^A/A mice with the full agonist of PPARγ, rosiglitazone, revealed a further improvement in insulin sensitivity (Fig. 2E). These effects included the adverse effects, weight gain and increased adiposity (Fig. 2F–G). These results indicate that while PPARγ^A/A mice show an improvement in glucose homeostasis, these animals are still susceptible to certain beneficial effects of TZD treatment and to the full adverse effects of these drugs.

Figure 2. PPARγ^A/A and wildtype mice respond similarly to PPARγ agonist treatment.

A-D. PPARγ^A/A (A/A) mice fed HFD for 16 weeks, do not innately exhibit the adverse effects associated with PPARγ agonism. A. Average cell area of adipocytes from epididymal adipose tissue after HFD feeding (n = 3). B. Hematocrit (packed cell volume, PCV; n = 6 wildtype (+/+), 5 A/A). C-D. Microquantitative computed tomography (microCT) analysis of trabecular bone from femurs of 6-month old mice maintained on HFD (n = 6 +/+, 5 A/A). C. Representative image of microCT. D. Parameters from microCT analysis, including tissue mineral density (TMD, left) and trabecular bone volume fraction (BV/TV, right). E-G. Both +/+ and A/A mice respond to treatment with PPARγ agonist rosiglitazone. After 19 weeks of HFD feeding, mice were treated with either vehicle or rosiglitazone (rosi) for 10 days (8 mg/kg body weight; n = 7 per genotype). E. Insulin tolerance test (1.5 U insulin/kg body weight); right, area under the curve (AUC). F. Body weight gained over course of treatment. G. Total fat mass at end of treatment. Data are presented as mean ± SEM; *, P < 0.05; **, P < 0.01; and ***, P < 0.001 by one-way ANOVA with Newman-Keuls Multiple Comparison (E-F) and two-way ANOVA (G).

Expression of Gdf3 is associated with PPARγ S273 phosphorylation

As PPARγ is phosphorylated in obese states and is highly expressed in adipose tissue, we examined transcriptional differences between control and littermate PPARγ^A/A eWAT from mice on HFD. Differential gene expression analysis following RNA-sequencing in eWAT revealed one striking result in the mutant animals: a strong and significant (p < 10 ⁻¹²) downregulation of Gdf3, a secreted protein and member of the TGF-β superfamily (Fig. 3A). Using qPCR, we confirmed Gdf3 was expressed at lower levels in both PPARγ^A/A iWAT and eWAT from HFD-fed mice when compared to littermate controls (Fig. 3B). Consistent with previously published data, eWAT Gdf3 levels increase in obesity (Wang et al., 2004). Gdf3 expression shows a 35-fold increase in the eWAT of WT mice maintained on HFD for 25 weeks compared to chow-fed controls. In contrast, the induction of Gdf3 by HFD was reduced 80 percent in the eWAT of PPARγ^A/A mice. No significant difference was observed between genotypes on a standard chow diet (Fig. 3C). We also saw a significant decrease in Gdf3 mRNA levels in the skeletal muscle (quadriceps) of HFD fed PPARγ^A/A mice (Fig. 3D). We next investigated Gdf3 regulation by non-genetic modulation of PPARγ S273 phosphorylation using small-molecules that can limit S273 phosphorylation of PPARγ. Mice were treated with rosiglitazone, a PPARγ agonist or with MEK inhibitors, GSK1120212 (Trametinib) or PD0325901 to reduce PPARγ phosphorylation. Gdf3 mRNA levels decreased significantly in eWAT of these mice in response to rosiglitazone, Trametinib, and PD0325901 by 60%, 68%, and 50% respectively (Fig. 3E).

Figure 3. Expression of Gdf3 is associated with PPARγ S273 phosphorylation.

A-B. PPARγ^A/A (A/A) and wildtype (+/+) mice fed HFD for 37 weeks. A. Volcano plot from RNA-seq analysis on epididymal adipose tissue (eWAT), with differentially expressed gene Gdf3 highlighted. B. qPCR validation of decreased Gdf3 mRNA levels in eWAT and inguinal adipose tissue (iWAT) of HFD-fed A/A mice. C. Gdf3 mRNA levels in eWAT from mice on chow or HFD for 25 weeks. D. Gdf3 mRNA levels in quadriceps from mice fed HFD for 17 weeks. E. Gdf3 mRNA levels in obese adipose tissue also decrease following treatment with known inhibitors of PPARγ S273 phosphorylation (rosiglitazone or either of two MEK inhibitors, PD0325901 or Trametinib; n = 4–6). F. Fractionated adipose tissue indicates Gdf3 is highly expressed in both the SVF and adipocyte fractions of eWAT from mice fed HFD for 10 weeks. G-H. Single Cell RNA-sequencing of SVF from HFD-fed +/+ mice shows detectable expression mainly in the macrophage/monocyte population. I-K. Generation of bone marrow chimeras reveals dominant role of hematopoietic cells in decreased Gdf3 expression of A/A mice. I. Scheme used for bone marrow transplantation study, including use of congenic C57BL/6 strains carrying functionally equivalent alleles of the pan leukocyte marker CD45. J. Quantification of donor CD45⁺ cells in recipient eWAT tissue after HFD treatment of chimeric mice (n = 7–8). K. Gdf3 mRNA levels in eWAT after HFD treatment of chimeric mice (n = 7–8). L-M. Thioglycollate-elicited peritoneal macrophages from chow fed A/A mice exhibit decreased Gdf3 (L) mRNA and (M) protein levels. Data are presented as mean ± SEM; *, P < 0.05; **, P < 0.01; and ***, P < 0.001 by Student’s t test (B-D, per tissue/conditition; J-L) and one-way ANOVA with Newman-Keuls Multiple Comparison test (E-F).

Gdf3 is most highly expressed in adipose tissue macrophages from HFD mice

Due to the heterogeneous composition of adipose tissue, we examined whether Gdf3 levels were found within a population of fractionated adipocytes or the stromal vascular fraction (SVF). Although HFD treatment increased Gdf3 expression in both adipocyte and SVF fractions, higher Gdf3 mRNA levels were observed in the SVF (Fig. 3F). Single-cell RNA-seq of mouse SVF from HFD fed mice (Jaitin et al., 2019) suggests that Gdf3 is localized to macrophage/ monocyte cells within the SVF (Fig. 3G–H). To assess the contribution of macrophage-specific PPARγ S273 phosphorylation to obesogenic Gdf3 expression, we generated bone marrow chimeras and asked whether obese WT mice with a PPARγ^A/A immune system will have altered Gdf3 expression (Fig. 3I). We found no differences in numbers of CD45⁺ cells infiltrating recipient eWAT, irrespective of the donor genotype (Fig 3J). Strikingly, mice with a hematopoietic compartment reconstituted from PPARγ^A/A donors exhibited lower Gdf3 expression in eWAT regardless of recipient genotype (Fig. 3K). Thus, transplantation with PPARγ^A/A bone marrow recapitulated the decreased Gdf3 levels observed with whole-body PPARγ^A/A animals. Moreover, peritoneal macrophages from PPARγ^A/A mice revealed a 65% reduction in the levels of Gdf3 mRNA and an 80% reduction in protein levels (Fig. 3L–3M). Together, these data provide strong evidence for the regulation of Gdf3 by PPARγ S273 phosphorylation.

To further determine whether the differential levels of Gdf3 levels in adipose tissue were due to decreased numbers of Gdf3 expressing cells or to specific modulation of gene expression, we performed detailed immunophenotyping on eWAT from WT and PPARγ^A/A mice maintained on HFD for 16 weeks. None of the macrophage or T cell populations examined showed significant differences when either the cell number or the percentage of gated cells was examined between genotypes. Macrophage populations were gated according to the strategy outlined in Suppl. Fig. 5A. No differences were observed in total macrophage number or percent of gated live, hematopoietic cells (CD45⁺) that were not eosinophils (Siglec F⁻) (Suppl. Fig. 5B). Nor were differences in macrophage percent of cells expressing an M1 marker (CD11c⁺) or an M2 marker (CD301⁺) (Suppl. Fig. 5C). PPARγ is highly expressed in T regulatory cells (Treg), a cell type which can control tissue inflammation including macrophage recruitment (Cipolletta et al., 2015; Cipolletta et al., 2012). Treg populations were gated according to the strategy outlined in Suppl. Fig. 5D. No differences were observed in total Treg number or percent of gated live, hematopoietic cells (CD45⁺) that expressed the T cell receptor, CD4 co-receptor, and transcription factor Foxp3 (TCRb⁺ CD4⁺ Foxp3⁺) (Suppl. Fig. 5E). Similarly, tissue-resident Tregs (KLRG1⁺ ST2⁺), had no significant differences (Suppl. Fig. 5F). Coupled with the lack of cell type specific gene regulation, these data support the conclusion that similar levels of hematopoietic cells are present in PPARγ^A/A adipose tissue. These findings strongly suggest the differences in Gdf3 expression levels in PPARγ^A/A adipose tissue reflect specific transcriptional control by PPARγ S273 phosphorylation.

GDF3 limits BMP signaling

Members of the TGF-β superfamily, including BMPs, GDFs, and TGF-β proteins can affect insulin sensitivity, cellular differentiation, and obesity (Chattopadhyay et al., 2017; Chen et al., 2003; Pauk et al., 2019; Schreiber et al., 2017). These proteins typically work either through BMP-type receptors or TGF-β−type receptors. Relatively little is known of mammalian Gdf3 function on glucose homeostasis in vivo. Two seemingly incompatible models have been proposed to explain how extracellular Gdf3 signaling is propagated. One model suggests that Gdf3 can inhibit BMP-receptor signaling, possibly by competing with other ligands for receptor binding (Levine and Brivanlou, 2006; Levine et al., 2009). In the other model, Gdf3 activates TGF-β type receptors (ALK7/Cripto), especially at super-physiological concentrations of 300–500 ng/ml (Andersson et al., 2008; Bu et al., 2018; Chen et al., 2006). We examined whether Gdf3 can affect either signaling system using a WT full-length Gdf3 cDNA expressed in human HEK 293T cells. As reported previously Activin, a TGF-β receptor ligand, activates Smad2/3 phosphorylation. In contrast, Gdf3 had little impact on TGF-β signaling. Activin neither promotes nor inhibits BMP-receptor signaling as measured by phosphorylation of SMAD1/5/8. However, cells expressing Gdf3 demonstrate a strong reduction of SMAD1/5/8 phosphorylation (Fig. 4A).

Figure 4. GDF3 limits BMP signaling, affects glucose uptake in vitro and is sufficient to induce insulin resistance in vivo.

A. Immunoblot analysis of HEK 293T cells transfected with Gdf3 or Myc-Activin constructs. Cells were deprived of serum for 16 h prior to whole-cell extract. B. Luciferase assay using a BMP-responsive element (BRE) driving luciferase reporters in serum-starved HEK 293T cells. C. BRE luciferase assay in C2C12 myoblasts that have been deprived of serum. Overnight treatment of 100 ng/mL rBMP4 demonstrates reporter activation. D-E. C2C12 myotubes transduced with AAV-GFP or AAV-Gdf3. D. [³H] 2-deoxy-D-glucose (2-DG) uptake assays, n = 6 wells per group. E. Immunoblot analyses. F-L. As a gain-of function model, 13 week old male mice were administered AAV-GFP or AAV-Gdf3, n = 10 mice per group. F. Immunoblots from eWAT and quad from two mice from each group, 9 weeks post injections. G. Body weights of mice; week 0 represents the values on the day of AAV injections. H. Body composition at 6 weeks post injections. I. Glucose tolerance test (2 g glucose/kg body weight) 4 weeks post injections. J. Insulin tolerance test (0.75 U insulin/kg body weight) 3 weeks post injections. Right, area under the curve (AUC) for the respective test (I-J). K-L. Relative Id2 mRNA levels from eWAT (K) and quadriceps (L). M. Model illustrating novel role for PPARγ S273 phosphorylation and Gdf3 in impaired insulin action with obesity. Data are presented as relative activity after normalization to Renilla luciferase (B-C) and as mean ± SEM (B-D, G-L); ns, not significant; *, P < 0.05; **, P < 0.01; and ***, P < 0.001 by Student’s t test (B-C, G-L) and two-way ANOVA with Sidak’s Multiple Comparison test (D).

To understand the integrated effects of Gdf3 over time, we performed luciferase reporter assays to quantify Gdf3 transcriptional consequences. In these reporter assays, BMP signaling promotes SMAD 1/5/8 binding to the BMP responsive element (BRE) of the Id1 promoter. We observed an 80% reduction of luciferase activity with Gdf3 overexpression compared to control cells expressing an empty vector (Fig. 4B). We further confirmed these results in a second cell type, C2C12 myoblasts. As in HEK 293T cells, Gdf3 decreases BRE-Luc by 50% compared to control cells. As a positive control, recombinant BMP4 activates BMP receptor signaling and stimulates Smad1/5/8 activity (Fig. 4C). These data suggest that Gdf3 is a BMP inhibitor in a class similar to Noggin or Gremlin (Daluiski et al., 2001; Lim et al., 2000).

GDF3 affects glucose uptake in vitro

Based on the increased skeletal muscle glucose uptake in HFD fed PPARγ^A/A mice, and the luciferase reporter assay in C2C12 myoblasts, we wanted to determine if regulating GDF3 levels in muscle cells could affect insulin-dependent glucose uptake. To test this, we transduced differentiated C2C12 myotubes with either an adeno-associated virus (AAV) expressing GFP (AAV-GFP), or an AAV expressing Gdf3 (AAV-Gdf3). GDF3 overexpression in these cells led to decreased glucose uptake in the presence of insulin, compared to GFP expressing cells. No differences between groups were observed under baseline (no insulin), (Fig. 4D). Immunoblot analyses of these cells under similar conditions revealed a decrease in insulin-stimulated AKT phosphorylation in GDF3 overexpressing cells (Fig. 4E). Together these results suggest that GDF3 overexpression is sufficient to inhibit glucose uptake and reduce insulin signaling in muscle cells in vitro.

GDF3 is sufficient to induce insulin resistance in vivo

To investigate the effect of GDF3 in vivo, we administered AAV-GFP or Gdf3 into the visceral fat pads (eWAT) and skeletal muscles (quadriceps and gastrocnemius) of lean WT mice resulting in a detectable increase in protein levels of Gdf3 (Fig. 4F). Forced GDF3 expression did not affect body weight, body composition, energy expenditure, or food intake of mice fed a chow diet (Fig. 4G, 4H and data not shown). Importantly, mice overexpressing GDF3 had impairments in both glucose tolerance and insulin tolerance tests (Fig. 4I and 4J). These data indicate that increased levels of GDF3 in these tissues are sufficient to negatively impact glucose homeostasis, independently of changes in body weight. The inhibitors of differentiation (Id) genes, including Id2 are canonical targets of BMPR signaling (Izumi et al., 2006). We examined the effect of AAV Gdf3 expression on Id2 mRNA levels in eWAT and quadriceps. In both tissues, we observed significant downregulation (Fig. 4K and 4L), consistent with inhibition of BMPR signaling in vivo. These effects of GDF3 to promote impaired glucose homeostasis in healthy, lean mice suggest that the elevated GDF3 levels observed in obesity may contribute to insulin resistance.

Discussion

The link between PPARγ biology and insulin sensitivity is both robust and complex. Patients treated with TZD PPARγ agonists experience durable reduced glycemia and insulin sensitivity. These effects were long thought to depend on PPARγ-driven adipogenesis—the generation of new, healthy adipocytes from precursor cells. The properties of PPARγ agonists to promote adipocyte differentiation cause weight gain, but also decrease bone formation leading to increased risk of fractures. The demonstration that blocking PPARγ S273 phosphorylation can improve insulin sensitivity independently of receptor agonism or effects on bone, suggests new therapeutic approaches to insulin sensitizers.

Increased PPARγ S273 phosphorylation induced by ERK kinase in obesity contributes to insulin resistance. Our study shows that preventing S273 phosphorylation genetically was sufficient to maintain insulin sensitivity in obesity, as manifested by improved insulin tolerance and increased whole body glucose utilization (Fig. 1). Consistent with this interpretation, in lean wild type chow-fed mice in which the receptor was not phosphorylated, we did not observe differences in glucose homeostasis from littermate PPARγ^A/A mice. (Suppl. Fig. 2). This suggests that the observed phenotypic changes were due to the contribution of obesity, rather than developmental effects. Moreover, as PPARγ S273 phosphorylation is independent of receptor agonism, we did not observe TZD-like side effects in PPARγ^A/A mice (Fig. 2). Interestingly, our insulin sensitive phenotype in PPARγ^A/A mice was not accompanied by increased glucose tolerance or changes in insulin secretion. The hyperinsulinemic-euglycemic clamp experiment also showed increased muscle glucose uptake in PPARγ^A/A mice with no significant changes in adipose glucose uptake or insulin-dependent suppression of endogenous glucose production. Together these results were unexpected but seem to indicate that the effects of suppression of PPARγ S273 phosphorylation might be affect only a subset of PPARγ-expressing tissues.

Differentiated primary adipocytes from WT and PPARγ^A/A mice did not show differences in the expression of a previously established gene set regulated by PPARγ S273 phosphorylation. This lack of differential gene expression could be related to the absence of ERK signaling or inflammation in our experimental conditions, environments that promote PPARγ S273 phosphorylation. Alternatively this could be due to the difference between a chronic genetic mutation versus acute treatments of PPARγ agonists or retroviral PPARγ^A/A expression ((Choi et al., 2010) and Suppl. Fig. 3).

We identified Gdf3 as a transcriptional target of PPARγ S273 phosphorylation in obese adipose tissue as well as skeletal muscle. Gdf3 was expressed in adipose tissue in both the SVF and adipocyte fractions, with highest levels observed in macrophages (Fig. 3). We also show that bone marrow derived macrophages transplanted from PPARγ^A/A mice can lead to lower Gdf3 expression in the adipose tissue of recipient mice and WT macrophages could rescue Gdf3 levels in PPARγ^A/A recipients (Fig. 3). Previous studies of macrophage-specific PPARγ deficiency identified strongly reduced Gdf3 mRNA levels in the absence of PPARγ, along with the binding of PPARγ and RXR to potential enhancer regions of the Gdf3 locus (Varga et al., 2016). While macrophage PPARγ-deficient mice expressed low Gdf3 levels which would predict insulin sensitivity, they nonetheless exhibited insulin resistance due to increased secretion of multiple pro-inflammatory cytokines (Chawla et al., 2001; Hevener et al., 2007; Morán-Salvador et al., 2011; Odegaard et al., 2007; Varga et al., 2016). In contrast, PPARγ^A/A mice did not have significant differences in macrophage number, relative polarization, or cytokine expression levels (Supp. Fig. 5).

Relatively little is known of mammalian Gdf3 function on glucose homeostasis in vivo. Gdf3 is highly expressed in white adipose tissue with levels increasing with age and obesity (Bu et al., 2018; Camell et al., 2017; Wang et al., 2004). Our in vitro studies in C2C12 myotubes with AAV mediated Gdf3 expression showed that Gdf3 was sufficient to reduce insulin signaling and insulin mediated glucose uptake in muscle cells (Fig 4D and E). Although cells in culture cannot be directly extrapolated to in vivo functions, we also saw from our in vivo AAV mediated Gdf3 expression in the eWAT and skeletal muscle of mice, that Gdf3 overexpression was sufficient to cause insulin resistance and glucose intolerance in chow fed lean mice (Fig 4F, I and J). The mechanisms of these effects are still unclear but our data support a model in which Gdf3 blocks beneficial BMP receptor signaling, decreasing phosphorylation of SMAD 1/5/8, and inhibiting Id2, a SMAD target gene (Fig. 4).

Our work is consistent with studies demonstrating improvements in glucose homeostasis by increasing levels of BMP4 or BMP signaling. Conversely, BMP4 KO mice exhibit impaired glucose tolerance (Hoffmann et al., 2017; Qian et al., 2013). Heterozygous deletion of the type 1 BMP receptor, Bmpr1a causes impaired glucose tolerance (Scott et al., 2009), while adipocyte-specific receptor deletion does not (Schulz et al., 2016). Similar salutary effects of glucose homeostasis have been observed in mice treated with BMP6, BMP7, and BMP9 (Chattopadhyay et al., 2017; Chen et al., 2003; Pauk et al., 2019). BMPs can drive cellular differentiation including potentiation of brown and beige fat activation (Hoffmann et al., 2017; Qian et al., 2013; Schulz et al., 2013). In this context, tight control of BMP signaling is maintained by inhibitory proteins including Gdf3, Noggin and Gremlin (Hsu et al., 1998; Levine and Brivanlou, 2006).

An emerging literature is developing on the contribution of macrophages to catecholamine production and degradation. Macrophages expressing Gdf3 have been implicated in controlling catecholamine degradation and rates of lipolysis (Camell et al., 2017). Mice with global deletion of Gdf3 from birth maintain lean body masses when challenged with HFD (Andersson et al., 2008; Shen et al., 2009). Conversely, Gdf3 has been reported to induce severe and persistent obesity in an adenovirus overexpression model in response to HFD feeding (Wang et al., 2004). We observed no differences in energy balance compared to control mice in our models of decreased Gdf3 levels (PPARγ^A/A mice) when challenged with HFD or with exogenous Gdf3 expression using AAV under chow fed conditions (Fig. 4). These differences may be due to the role of Gdf3 in embryonic development (Chen et al., 2006; Levine and Brivanlou, 2006) versus the late timing of Gdf3 expression changes driven by differential obesity-mediated phosphorylation of PPARγ S273 or also on the late timing and tissue specificity of AAV-mediated Gdf3 expression.

Future studies will determine whether Gdf3 inhibits BMP-receptor signaling by binding to the BMP receptor or through inhibitory binding to other extracellular BMP receptor ligands and if these interactions are sufficient to cause the insulin resistance phenotype of Gdf3 overexpression. We find that Gdf3 is a factor that mediates the adverse consequences of obesity-mediated PPARγ S273 phosphorylation. Further study into the mechanisms of Gdf3 action will be beneficial to understand whether this is an appropriate future therapeutic target.

Limitations of Study

While our study demonstrates a causal role for PPARγ S273 phosphorylation in insulin resistance, the use of a whole-body knock-in model does not allow us identification of primary tissues driving this effect. In addition, although Gdf3 has previously been identified as a PPARγ target gene, our study does not show a direct mechanism for how PPARγ S273 phosphorylation alters Gdf3 expression, nor why the regulation of this gene is so selective. We demonstrate that Gdf3 levels correlate with insulin resistance, yet we do not have an elucidated mechanism to understand how Gdf3 promotes the development of insulin resistance. These efforts are needed as considerable controversy exists addressing many of the fundamental questions of Gdf3 biology including dimerization partners, receptor identity, and signaling cascade. Our data indicate macrophages are likely to be the primary source of the elevated Gdf3 levels induced with obesity, however further studies are required to confirm this finding.

STAR Methods

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Alexander S Banks (asbanks@bidmc.harvard.edu).

Materials Availability

All unique/stable reagents generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

Data and Code Availability

RNA-seq data from eWAT of WT and A/A mice have been deposited in the Gene Expression Omnibus (GEO) repository under the accession number GSE151030.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mouse model

Generation of PPARγ^S273A mice

We obtained a C57BL/6 bacterial artificial chromosome (BAC) containing the Ppar locus (Invitrogen/RPCI). An 8.9 kb fragment preceding and including exon 5 was amplified by PCR (Phusion DNA Polymerase, NEB) and cloned into a plasmid encoding both a floxed neomycin resistance cassette and a diphtheria toxin negative selection marker. Site-directed mutagenesis was used to alter the codon for serine 273 to alanine (TCA to GCA). Lastly, a 1.6 kb 3′ homology arm was similarly amplified and cloned into the plasmid. C57BL/6 ES cells were electroporated with the linearized targeting construct. Using positive and negative selection, 8 ES cell clones were isolated and two were injected into Balb/c blastocysts. The resulting progeny gave rise to highly chimeric offspring. These chimeras were mated to C57BL/6J mice to obtain germline transmission of the mutant allele. To remove the floxed neomycin cassette, these germline offspring were crossed to mice expressing Cre in the female germline (Zp3-Cre). The progeny from these crosses were found to have deleted the neomycin cassette successfully. These heterozygous mice (PPARγ^S273A/+) were mated to generate homozygous knock-in mice PPARγ^S273A/S273A or PPARγ^A/A. Mice were backcrossed onto C57BL/6J for 6 generations, after which mice heterozygous for the PPARγ^S273A allele (PPARγ^S273A/+) were bred to generate experimental cohorts. The PPARγ^S273A allele was detected by PCR genotyping with primers 5′-CAGGAGGCTGAGCAGGTGTGTT-3′, 5′-TCCAGACTGCCTTGGGAAAA-3′, and 5′-AGCACACATGTACCCAACAT-3′ (34 cycles, 96°C, 30s; 65°C, 45s; 72°C, 45s), where the WT and knock-in alleles produced PCR products of 319 bp and 217 bp, respectively. To minimize phenotypic variability from a mixed 6J/6N background, PPARγ^S273A/+ mice were genotyped for single nucleotide polymorphisms (SNPs) that distinguish between C57BL6/J and C57BL6/N alleles (Fontaine and Davis, 2016); mice homozygous for at least 9 SNPs specific to the C57BL/6J substrain on distinct chromosomes, including the 6/J null allele encoding nicotinamide nucleotide transhydrogenase (Nnt) (Nicholson et al., 2010), were used to establish cohorts for subsequent experiments.

Cell lines

HEK 293T cells (human female in origin) were cultured every 3 to 4 days using 0.05 % Trypsin EDTA and maintained in DMEM (Gibco, 11965118) supplemented with 10% heat inactivated FBS (FBS) and 1% Penicillin/Streptomycin. C2C12 myoblasts (ECACC, 91031101, C3H female mouse in origin) were cultured every 3 days using 0.25% Trypsin-EDTA and maintained in DMEM (Gibco, 11965118) supplemented with 20% FBS and 1% Penicillin/Streptomycin. All cell lines were maintained at 37°C at 5% CO₂. Confluent C2C12 myoblast cells were induced to differentiate with serum withdrawal to 2% horse serum in collagen coated 6 well plates for 5 days. Differentiated C2C12 myotubes were returned to complete medium with 20% FBS for two days prior to subsequent experiments. These cell lines were not authenticated.

Primary cell culture

The stromal vascular fraction (SVF) containing preadipocytes and mononuclear cells were extracted from adipose tissue of WT or A/A mice (male) and cultured at 37°C with 5% CO2. Cells were maintained in DMEM (Gibco, 11965118) with 10% FBS and 1% Penicillin/Streptomycin until confluency. Two days post confluency, adipocyte differentiation was induced by treating cells (adipose tissue SVF) with 5 μM dexamethasone, 250 μM isobutylmethylxanthine, 1 μM rosiglitazone, and 900 nM insulin in DMEM with 10% FBS and 1% Penicillin/Streptomycin for 48 h. Cells were maintained in culture medium containing FBS, Penicillin/Streptomycin, insulin, and rosiglitazone for another 2 days. Thereafter, cells were cultured in medium containing FBS and Penicillin/Streptomycin only.

METHOD DETAILS

Animal experiments

All animal experiments were performed with approval from the Institutional Animal Care and Use Committees of The Harvard Center for Comparative Medicine, Brigham and Women’s Hospital, and Beth Israel Deaconess Medical Center. Mice were maintained at 12 h/12 h light/dark cycles, 22 ± 2°C room temperature and 30%–70% humidity with ad libitum access to food and water, in individually ventilated cages. Cages and bedding were changed once every two weeks and mice were monitored regularly for their health status by animal technicians with the support of veterinary care and remained free of any adventitious infections for the entire duration of this study. Mice were fed a standard chow diet (13% kcal fat, LabDiet, no. 5053) or a HFD (60% kcal fat, Research Diets, no. D12492i) for the indicated durations. All experiments were performed in adult male mice (8 to 43 weeks of age). For HFD experiments, WT and AA mice were put on HFD at 6 weeks of age. Body composition of conscious mice were assessed with an EchoMRI 3-in-1 (Echo Medical Systems). At the end of experiments, mice were euthanized using CO₂, blood samples were collected via cardiac puncture in EDTA treated tubes on ice unless otherwise specified, and tissues were harvested and snap-frozen in liquid nitrogen and stored at −80°C until processing. Alternatively, samples of adipose tissue were fixed in formaldehyde, embedded in paraffin, sectioned and stained with H&E for determination of adipocyte area. Briefly, H&E stained slides were visualized and non-overlapping fields were captured for each sample using a Zeiss optical microscope and the AxioCam camera. Adipocyte area were calculated using ImageJ software (Schneider et al., 2012) using freely available macros described in (Parlee et al., 2014). Adipocytes touching the edge of each image were excluded as well as cells < 350 μm² as these cells may be a mixture of adipocytes and stromal vascular cells. For plasma collection, the blood samples in EDTA treated tubes were centrifuged at 2000g at 4°C for 15 minutes, and the resulting supernatant was collected as plasma and stored at −80°C until processing. For measures of hematocrit, a small volume of whole blood was centrifuged in a microhematocrit capillary tube, and the height of the column of packed cells was expressed as a percentage of the height of the total blood column samples with visible hemolysis were excluded).

In vivo AAV experiments

12 week old male chow fed C57BL6/J mice were purchased from Jackson Labs and allowed to acclimate to our animal facility for 1 week. Each AAV was injected bilaterally into the quadriceps and gastrocnemius muscles at a dose of 1.5 × 10⁸ vg in 30 μL of PBS per injection and into the epididymal fat depots at 2.5 × 10⁸ vg in 50 μL of PBS per injection. These viral injections were performed under isoflurane anesthesia and mice were given Buprenorphine SR Lab as the perioperative and postoperative analgesic prior to surgery at a dose of 0.5 mg/kg. These mice were maintained on chow and euthanized 9 weeks post injections.

Glucose and insulin tolerance tests

Tests were performed on mice fasted for 16 and 4 hours, respectively with ad libitum access to water. Following an initial blood glucose measurement post-fast, tolerance to an intraperitoneal (IP) injection of glucose (1–3 g/kg) or insulin (0.75–2 U/kg) was assessed by change in blood glucose over a period of 120 minutes. Glycemia was measured by tail vein bleeds at the indicated times using a Contour Next EZ glucometer (Bayer). Plasma insulin levels were measured using the Ultra-Sensitive Mouse Insulin ELISA Kit (Crystal Chem, Inc.).

For PPARγ ligand treatment prior to ITT, diet-induced obese mice were IP injected daily for 10 days with rosiglitazone (Cayman Chemical; 8 mg/kg) before insulin tolerance testing. Rosiglitazone was dissolved in dimethylsulfoxide (DMSO) and diluted into saline with 2% Tween-80 for injection. Administration of MEK inhibitors Trametinib and PD0325901 (Selleckchem) were as previously described (Banks et al., 2015). Mainly, Trametinib (3 mg/kg) or PD0325901 (10 mg/kg) were administered by daily oral gavage for 5 days. Compounds were dissolved in DMSO and diluted into an aqueous 250 μl dose containing 0.5% hypromellose and 2% Tween-80.

Hyperinsulinemic-euglycemic clamp

WT and A/A mice fed HFD for 16 weeks were subjected to a hyperinsulinemic-euglycemic clamp study as previously described with minor modifications (Jurczak et al., 2012). Briefly, mice were allowed to recover for one week after an indwelling catheter was surgically implanted in the right jugular. Following an overnight fast of approximately 14 h mice were infused with a fixed amount of insulin [4 mU/(kg-min)] and a variable amount of 20% dextrose to maintain euglycemia (~120 mg/dL). [3-³H]glucose (PerkinElmer) was included in the infusate at a rate of 0.05 μCi/min for 2 h to allow for the calculation of whole-body rates of glucose metabolism. A 10-μCi bolus injection of 2-deoxy-D-[1-¹⁴C]glucose (10 μCi/mouse; PerkinElmer) was given at 90 min to determine tissue-specific glucose uptake. Blood was collected via tail massage for plasma glucose, insulin, and tracer levels at set time points during the 140-min infusion. For biochemical analysis of insulin signaling, obese mice of both genotypes were fasted for 4 hours prior to IP injection of insulin (10 U/kg). Ten minutes after insulin injection, white adipose tissue, liver, and gastrocnemius were collected for subsequent protein analysis.

Indirect calorimetry

For measurements of metabolic rate and food intake, we placed approximately 35 week old male mice fed HFD for 29 weeks within the CLAMS (Columbus Instruments) indirect calorimeter. Data from two independent cohorts was combined. One week prior to monitoring, mice were surgically implanted with wireless body temperature probes (Starr Scientific). Mice were acclimated for 13 h followed by 24 h measurement of VO₂, VCO₂, RER, locomotor and ambulatory activity, food intake at 23 ± 0.1°C while on HFD. Energy expenditure was calculated with the Weir equation (Weir, 1949). Calories consumed were calculated by multiplying hourly food intake by the 5.21 kcal/g caloric value of the 60% HFD. Energy balance was calculated by subtracting hourly food intake from hourly energy expenditure. Data from two separate experimental runs were combined. Results for locomotor activity (n=2) and body temperature (n=3) were removed due to equipment malfunction; other data from these animals were included. Analysis with ANOVA and ANCOVA were performed using CalR with the remove outliers feature activated (Mina et al., 2018).

Microquantitative computed tomography

For microCT analysis, a Scanco Medical μCT 35 system with an isotropic voxel size of 7 μm was used to image the femur. Scans were conducted in 70% ethanol using an X-ray tube potential of 55 kVp, an X-ray intensity of 0.145 mA and an integration time of 600 ms. Analysis was performed according to published guidelines (Bouxsein et al., 2010). From the scans of the femur, a region beginning 0.28 mm proximal to the growth plate and extending 1mm proximally was selected for trabecular bone analysis. A second region 0.6 mm in length and centered at the midpoint of the femur was used to calculate diaphyseal parameters. A semi-automated contouring approach was used to distinguish cortical and trabecular bone. The region of interest was thresholded using a global threshold that set the bone/marrow cut-off at 452.65 mg HA/cm³ for trabecular bone and 579.86 mg HA/cm³ for cortical bone. 3-D microstructural properties of bone, including TMD and bone volume fraction (BV/TV), were calculated using software supplied by the manufacturer.

RNA isolation and gene expression analysis by qPCR, RNA-seq, or single cell RNA-seq

RNA Isolation

For experiments utilizing RNA-seq or qPCR, total RNA was isolated from cells or tissues using TRIzol reagent (Thermo Fisher) followed by a Direct-zol RNA MiniPrep kit (Zymo Research) according to the manufacturer’s instructions. The RNA was reverse-transcribed using Multiscribe Reverse Transcriptase and a high-capacity cDNA reverse transcription kit (Applied Biosystems).

Quantitative real-time PCR (qPCR) was performed using synthesized cDNA with SYBR Select Master Mix (Applied Biosystems) on a Light Cycler 480 II (Roche) or the QuantStudio 5 Real-Time PCR System (Applied Biosystems). Relative mRNA expression was determined by the ΔΔ-C_t method normalized to TATA-binding protein (TBP) levels. The list of primers used in this study is included in Supplementary Table S1.

RNA-sequencing (RNA-seq)

400ng of total RNA was used as input for RNA-seq Library construction, which consisted of rRNA removal via the NEBNext rRNA depletion kit and conversion to cDNA using the NEBNext Ultra II Non-Directional RNA Second Strand Synthesis Module, both per manufacturer’s instructions. cDNA was diluted 1:15 and subsequently tagmented and amplified for 12 cycles using the Nextera XT DNA Library Preparation Kit (Illumina). Individual libraries were QC’d by Qubit and Agilent Bioanalyzer, pooled, and sequenced at a final loading concentration of 1.8 pM on a NextSeq500 (36bp x 36bp paired-end reads). Sequencing reads were demultiplexed using bcl2fastq (v2.20.0) and aligned to the mm10 mouse genome by with HISAT2 (v. 2.0.5 (Kim et al., 2015). PCR duplicates and low-quality reads were filtered by Picard (https://broadinstitute.github.io/picard). Filtered reads were assigned to the annotated transcriptome and quantified using featureCounts (Liao et al., 2014). Normalization and differential expression analysis were performed using EdgeR (v 2.9.2) (Robinson et al., 2010).

Single cell RNA-seq

Identification of Gdf3-expressing cells in HFD-fed mice was performed using single cell RNA-sequencing from SVF fraction of eWAT of mice by (Jaitin et al., 2019). These data were clustered de novo using the SCTransform seurat pipeline (Hafemeister and Satija, 2019; Stuart et al., 2019) using the following parameters: 2000 variable genes, 25 UMAP and SNN dimensions, and clustering resolution of 0.4. Excluded from analysis were a list of mitochondrial and ribosomal genes identified by (Jaitin et al., 2019).

Preparation of cell or tissue lysates and immunoblotting

Whole-cell extracts were prepared in RIPA lysis buffer containing 50 mM Tris pH 7.5, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, and 1× HALT protease and phosphatase inhibitor cocktail (Thermo Fisher). For tissue lysates, frozen tissue was pulverized in liquid nitrogen followed by homogenization in lysis buffer using stainless steel ball bearings and a TissueLyser (Qiagen). Protein concentration was determined by Pierce BCA protein assay (Thermo Fisher) and lysates separated by SDS-PAGE with Mini-PROTEAN TGX gels (Bio-Rad), transferred to PVDF membranes using the Trans-Blot Turbo transfer system (Bio-Rad), and blotted according to manufacturer’s recommendations for the indicated antibodies. Primary antibodies used in this study include: rabbit anti-phospho-Akt Ser473, rabbit anti-Akt, rabbit anti-β-Tubulin, rat anti-Gdf3, mouse anti-β-Actin, rabbit anti-phospho-Smad2 Ser465/467/Smad3 Ser423/425, rabbit anti-Smad2, rabbit anti-phospho-Smad1 Ser463/465/Smad5 Ser463/465/Smad9 Ser465/467, rabbit anti-Smad1, rabbit anti-Myc-Tag, rabbit anti-phospho-PPARγ Ser273, and mouse anti-PPARγ. Horseradish peroxidase-conjugated secondary antibodies were obtained from Cell Signaling Technologies. Chemiluminescence assays were performed using SuperSignal West Dura extended duration substrates (Thermo Fisher) and detected with a ChemiDoc XRS+ or ChemiDoc Touch Imaging System (Bio-Rad). Image analyses were performed using Image Lab software (Bio-Rad).

DNA constructs and AAVs

A non-tagged construct of mouse Gdf3 was generated using the commercial vector pCMV6-Gdf3-Myc-Flag (OriGene Technologies, MR222967), containing a Myc-Flag-tagged Gdf3, and PCR-based mutagenesis to introduce a STOP codon upstream of the C-terminal Myc-Flag tag. The BMP responsive element luciferase reporter, derived from the ID1 luciferase reporter (Korchynskyi and ten Dijke, 2002) was provided by C. Brown. A Myc-Flag-tagged mouse Activin expression plasmid was purchased from Origene Technologies (MR225191). AAV Gdf3 was generated by subcloning non-tagged Gdf3 into the AAV2/1 vector backbone of AAV-CAG-GFP, a gift from Karel Svoboda via Addgene plasmid 28014 (Mao et al., 2011). Viral packaging into the AAV-DJ serotype and titer determination were performed by the Boston Children’s Hospital Viral Core. Premade control AAV2/DJ-CAG-H2BGFP virus (AAV-GFP) was purchased from the same core facility.

Lipid accumulation assay

Lipid accumulation in adipocytes was detected by Oil Red O staining as described (Ramirez-Zacarias et al., 1992).

Glucose uptake assays

C2C12 myoblast cells were differentiated and transduced with either AAV-GFP or AAV-Gdf3 (1.2 × 10¹⁰ vg per well in a 48 well plate) and (1.2 × 10¹¹ vg per well in 6 well plates). Approximately 36 hours post transduction, the cells from the 6 well plates were incubated in serum-free DMEM for 3 h at 37°C, incubated with or without 200 nM insulin for 20 min washed twice in ice cold PBS and lysed with RIPA lysis buffer for western blot analysis as described above. The cells from the 48 well plate were used for glucose uptake assays using the protocol described in (Yore et al., 2014) with minor modifications. Briefly, cells were washed twice with serum-free DMEM and incubated in serum-free DMEM for 3 h at 37°C. Cells were then washed 3 times with Krebs-Ringer-HEPES (KRH) buffer (50 mM HEPES pH 7.4, 137 mM NaCl, 1.25 mM CaCl2, 4.7 mM KCl, 5.0 mM, 1.25 mM MgSO4). Cells were incubated in KRH for 30 min then incubated in KRH with or without 200 nM insulin for 20 min before the addition of a deoxy-D-glucose, 2-[1,2–3H (N)]/2-deoxy-D-glucose solution, yielding final assay concentrations of 1 μCi of [³H] 2-deoxy-D-glucose and 100 μM 2-deoxy-D-glucose per well. After 5 min the KRH buffer was removed immediately by aspiration. Cells were washed 3 times in ice-cold PBS to inhibit further glucose uptake and wash away unincorporated radiolabeled [³H] 2-deoxy-D-glucose. Cells were solubilized with 0.5% SDS. Half of the lysates from each well were used to assay [³H] disintegration levels by liquid scintillation counting and the other half was used for normalization to protein concentrations.

Luciferase reporter assays

C2C12 myoblasts were transfected without myogenic differentiation. HEK 293T or C2C12 cells were transfected with a firefly luciferase reporter construct and a fixed amount of DNA, as normalized with empty vector plasmid using Lipofectamine 2000 and 3000 transfection reagents (Invitrogen), respectively, according to the manufacturer’s recommendations. Luciferase activity was measured 24–48 h after transfection using the Dual-Luciferase Reporter Assay system (Promega), with quantification of luminescence on a FLUOstar OPTIMA plate reader (BMG Labtech). Co-transfection of CMV-driven Renilla luciferase vector was used to normalize for transfection efficiency. Where indicated, cells were rinsed with PBS and switched to DMEM lacking serum prior to luciferase measurement. As a positive control for BMP signaling, recombinant protein of mouse BMP4 (R&D Systems) was added to the medium 16 h before analysis.

Isolation of macrophages

Peritoneal macrophages were harvested from chow fed, 15 to 20 week old mice as previously described (Ray and Dittel, 2010). Briefly, WT or A/A mice were given an IP injection of 3% thioglycolate solution (3 mL per mouse). 4 days post injection, mice were euthanized and their peritoneal cavity was carefully injected with 5 ml of ice cold PBS (with 3% FBS) using a 27G needle without puncturing any organs. The peritoneum was gently massaged to dislodge any attached cells into the PBS solution which was then aspirated using a 25G needle attached to a 5mL syringe. These peritoneal cells were collected into ice cold 50mL polypropelene tubes, erythrocytes were lysed using ACK lysis buffer (Gibco), resuspended in ice cold PBS (with 3% FBS) and centrifuged for 8 min at 400g at 4°C. The pelleted cells were resuspended in DMEM containing 10% FBS and allowed to adhere for 2 h in 10cm cell culture dishes in a 37°C incubator qith 5% CO₂. Nonadherent cells were washed away with PBS, and adherent cells were scraped and used as peritoneal macrophages for further experiments.

Isolation of cells from SVF

The stromal vascular fraction (SVF) containing preadipocytes and mononuclear cells was extracted from adipose tissue. For isolation of preadipocytes, inguinal adipose tissue was digested in a PBS buffer containing 10 mg/mL collagenase D (Roche), 2.4 U/mL dispase II, and 10 mM CaCl₂ for 45 min shaking at 37°C. Digestion was neutralized with culture medium consisting of DMEM/F12 supplemented with 10% FBS and penicillin/streptomycin. Cell suspensions were filtered through 100 μm and 40 μm cell strainers, and the SVF cells resuspended in growth medium for expansion. For isolation of mononuclear cells and subsequent immunophenotyping, epididymal and inguinal adipose tissues were digested for 20–30 min at 37°C with collagenase type II (Sigma). Cell suspensions were filtered through a sieve, and the SVF fraction was collected after red blood cell lysis (ACK Lysing Buffer, Gibco) and filtration through a 40 μm cell strainer.

Immunophenotyping and flow cytometry

Adipose tissue SVF cells were resuspended in FACS buffer (phenol red-free DMEM supplemented with 10 mM HEPES and 2% FCS) and incubated for 30–45 min at 4°C with Fc block and the indicated fluorescence-labeled antibodies for surface molecule staining. For myeloid cell analysis, cells were stained with anti-CD45.2 (clone 104), -CD45.1 (A20), -CD11b (M1/70), -F4/80 (BM8), and -CD301 (LOM-14) antibodies from BioLegend; -CD11c (N418) antibody from eBioscience; and -SiglecF (E50–2440) antibody from BD Biosciences. For lymphoid cell analysis, cells were stained with anti-CD45.2, -CD45.1, -CD19 (6D5), -TCR-β (H57–597), and -CD8 (53–6.7) antibodies from BioLegend; -CD4 (RM4–5), -KLRG1 (2F1), and -IL33R/ST2 (RMST2–2) antibodies from eBioscience; and were fixed, permeabilized and intracellularly stained for Foxp3 (FJK-16s) according to the manufacturer’s instructions (eBioscience). Dead cells were excluded by staining with a fixable LIVE/DEAD dye (ThermoFisher). Cells were analyzed using an LSRII instrument (BD Bioscience) and FlowJo software.

Bone marrow transplantation

Bone marrow progenitor cells were isolated by flushing femurs and tibiae of 10 to 12 week old donor animals with PBS, and T-cells depleted from BM by magnetic beads. Recipient mice between 8 to 12 weeks of age were irradiated with 1000 rads prior to reconstitution with 3–5 × 10⁶ donor cells by retro-orbital injection. Irradiated mice were kept in sterile cages with standard chow diet and antibiotic-treated water for 2 weeks. After 4 weeks recovery, mice were bled to monitor reconstitution and started on HFD treatment for an additional 20 weeks. To distinguish chimerism, transplantation studies made use of congenic C57BL/6 strains carrying functionally equivalent alleles of the pan leukocyte marker CD45. For studies with PPARγ^+/+ recipient mice, mice harboring the CD45.1 allele were purchased from Jackson (B6.SJL-Ptprc^aPepc^b/BoyJ) and transplanted with bone marrow from PPARγ^+/+ or PPARγ^A/A mice, which carry the WT allele for CD45 (CD45.2). For studies with PPARγ^A/A recipient mice, PPARγ^A/A mice were transplanted with bone marrow from PPARγ^+/+ or PPARγ^A/A mice heterozygous for CD45.1.

QUANTIFICATION AND STATISTICAL ANALYSIS

Data were analyzed using Prism software (GraphPad Software, Inc) and are expressed as mean ± SEM. All data were assumed to have a normal distribution and two-tailed Student’s t tests, one-way ANOVA with Newman-Keuls Multiple Comparison test, or two-way ANOVA were used to compare means between groups as indicated in the figure legends; P < 0.05 was considered significant.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
rabbit anti-phospho-Akt Ser473	Cell Signaling Technology	4060S
rabbit anti-Akt	Cell Signaling Technology	9272S
rabbit anti-β-Tubulin	Cell Signaling Technology	2146
rat anti-Gdf3	R&D Systems	MAB953
mouse anti-β-Actin	Sigma Aldrich	A1978
rabbit anti-phospho-Smad2 Ser465/467/ Smad3 Ser423/425	Cell Signaling Technology	8828S
rabbit anti-Smad2	Cell Signaling Technology	5339S
rabbit anti-phospho-Smad1 Ser463/465/ Smad5 Ser463/465/ Smad9 Ser465/467	Cell Signaling Technology	13820S
rabbit anti-Smad1	Cell Signaling Technology	6944S
rabbit anti-Myc-Tag	Cell Signaling Technology	2278S
rabbit anti-phospho-PPARγ Ser273	previously described (Choi et al., 2010)	N/A
mouse anti-PPARγ	Santa Cruz	sc-7273
anti-rabbit IgG, Horseradish peroxidase-conjugated	Cell Signaling Technology	7074S
anti-mouse IgG, Horseradish peroxidase-conjugated	Cell Signaling Technology	7076S
anti-rat IgG, Horseradish peroxidase-conjugated	Cell Signaling Technology	7077S
anti-CD45.2 (clone 104)	BioLegend	109820
anti-CD45.1 (A20)	BioLegend	110706
anti-CD11b (M1/70)	BioLegend	141704
anti-F4/80 (BM8)	BioLegend	123114
anti-CD301 (LOM-14)	BioLegend	145709
anti-CD11c (N418)	eBioscience	61-0114-82
anti-SiglecF (E50-2440)	BD Biosciences	552126
anti-CD19 (6D5)	BioLegend	115508
anti-TCR-β (H57-597)	BioLegend	109222
anti-CD8 (53-6.7)	BioLegend	100730
anti -CD4 (RM4-5)	eBioscience	61-0042-82
anti -KLRG1 (2F1)	eBioscience	11-5893-82
anti -IL33R/ST2 (RMST2-2)	eBioscience	46-9335-82
Bacterial and Virus Strains
AAV2/DJ-CAG-H2BGFP	Boston Children’s Hospital viral core	Not Applicable (N/A)
AAV2/DJ-CAG-Gdf3	This manuscript	N/A
Chemicals, Peptides, and Recombinant Proteins
Rosiglitazone, Cayman Chemical	Fisher Scientific	NC9560589
PD0325901	Selleckchem	S1036
Trametinib (GSK1120212)	Selleckchem	S2673
Recombinant mouse BMP4	R&D Systems	5020-BP-010
Critical Commercial Assays
Ultra-Sensitive Mouse Insulin ELISA Kit	Crystal Chem	90080
NEBNext rRNA Depletion Kit	New England Biolabs	E6310X
NEBNext Ultra II Non-Directional RNA Second Strand Synthesis Module	New England Biolabs	E6111L
Nextera XT DNA Library Preparation Kit	Illumina	FC-131
Dual-Luciferase Reporter Assay System	Promega	E1960
High Capacity cDNA Reverse Transcription Kit, Applied Biosystems	Thermo Fisher Scientific	4368813
Foxp3 / Transcription Factor Staining Buffer Set, eBioscience	Thermo Fisher Scientific	00-5523-00
Deposited Data
RNA-seq data	GEO repository	GSE151030
Experimental Models: Cell Lines
Human: HEK 293T	Spiegelman lab, Dana Farber Cancer Institute	N/A
Mouse: C2C12 C3H muscle myoblast	Sigma Aldrich	91031101
Experimental Models: Organisms/Strains
Mouse: C57/Bl6J	The Jackson Laboratory	000664
Mouse: Bl6J PPARγS273A/+	This manuscript	N/A
Mouse: B6.SJL-Ptprca Pepcb/BoyJ	The Jackson Laboratory	002014
Oligonucleotides
Primers for RT-qPCR analysis, see Table S1	This manuscript	N/A
Recombinant DNA
pCMV6-Gdf3	This manuscript	N/A
pCMV6-Gdf3-Myc-Flag	OriGene Technologies	MR222967
pCMV6-Activin(Inhbb)-Myc-DDK	Origene Technologies	MR225191
pAAV-CAG-GFP	Addgene	28014
Software and Algorithms
CalR	Mina et al., 2018	https://calrapp.org/ RRID:SCR_015849
ImageJ	Schneider et al., 2012	https://imagej.nih.gov/ij/
Prism 8	GraphPad Software	https://www.graphpad.com/scientific-software/prism/
FlowJo	FlowJo, LLC	https://www.flowjo.com/
Other
CLAMS indirect calorimeter	Columbus Instruments	http://www.colinst.com/

Highlights.

• Blocking PPARγS273 phosphorylation protects mice from insulin resistance in obesity

• These mice do not show the side effects associated with TZD based PPARγ agonism

• These mice have reduced expression of Gdf3 mainly in macrophages

• Gdf3 is sufficient to cause impaired glucose homeostasis in vivo and in vitro

Context and Significance.

Previous studies have revealed a mechanism wherein blocking PPARγ phosphorylation at Serine 273 could improve insulin sensitivity without causing the side effects of PPARγ agonism. This paper tests this mechanism genetically by using mice that cannot be phosphorylated at PPARγ Ser273 (A/A). They find that A/A mice are protected from developing insulin resistance without the side effects of PPARγ agonism. They implicate the regulation of TGFβ family member Gdf3, by PPARγ, in the mechanism of action, and show that GDF3 is sufficient to cause insulin resistance in lean mice. This study highlights the diabetogenic potential of PPARγ Ser273 phosphorylation and GDF3 which could be potential targets for developing better type-2 diabetes therapeutics.