An Alternatively Translated Isoform of PPARG Suggests AF-1 Domain Inhibition as an Insulin Sensitization Target

Insulin resistance is a major driver of the epidemic metabolic diseases that challenge global health (1). Thiazolidinediones (TZDs) comprise a class of drugs that decrease insulin resistance by agonizing proliferator–activated receptor γ (PPARγ) (2), a nuclear hormone receptor that contains an autonomous activation function 1 (AF-1) domain, DNA binding domain (DBD), and ligand binding domain (LBD) (Fig. 1A) (3). TZDs have demonstrated clinical efficacy in treating type 2 diabetes and cardiovascular disease (4), but their use has been limited by serious adverse effects, including weight gain and fluid retention (2). Much pharmacologic effort has focused on the development of selective PPARγ modulators that retain the benefits of TZDs without the attendant adverse effects, but successful compounds have not reached the clinic (5). Therefore, the need for alternative approaches to therapeutically activate PPARγ without PPAR-mediated adverse effects remains unmet.

Loss-of-function (LOF) mutations in PPARG occurring in the DBD and LBD have been shown to cause familial partial lipodystrophy type 3 (FPLD3), a Mendelian genetic syndrome characterized by insulin resistance, metabolic syndrome, and gluteofemoral fat loss (6,7). These pathogenic FPLD3 mutations establish the clinical significance of the PPARγ DBD and LBD, but they represent only the tip of the iceberg in human protein–coding variants found in PPARG (8). In previous work, we identified hundreds of protein-coding variants in PPARG occurring in all protein domains, including the AF-1 domain (9). As with FPLD3, those that cause LOF in the DBD and LBD increase insulin resistance and type 2 diabetes risk. We observed no apparent clinical impact of LOF variants in the AF-1 domain, leaving in question its function in human metabolic health.

In this article, we report a novel protein isoform of PPARγ (PPARγ M135), which lacks the AF-1 domain and is generated from an alternative translational start site. Through biochemical and transcriptomic profiling, we found that PPARγ M135 demonstrated enhanced ligand-inducible transcriptional and functional activity compared with wild type (WT), leading us to a model of derepression by loss of AF-1. To evaluate the clinical consequence of this model, we identified and analyzed human carriers of PPARG variants that impair AF-1 function, finding that these variants increased PPARγ function and decreased metabolic syndrome severity in individuals who carry them. Taken together, our study findings propose AF-1 domain inhibition as a new targetable mechanism to activate PPARγ.

Human monocytic leukemia cells (THP-1; TIB-202; American Type Culture Collection) and human preadipocyte cells (Simpson-Golabi-Behmel syndrome [SGBS]) were cultured, differentiated, and stimulated to activate PPARγ as described in the Supplementary Methods. Statistical analysis did not include sex, because all cells originated from the same male cell lines.

Guide RNAs (sgRNAs) to target PPARG (Supplementary Table 1) were cloned into lentiCRISPRv2 (ID 52961; Addgene), and the vectors were pooled for virus production (2304; Mirus Bio). THP-1 cells were infected at a multiplicity of infection of 0.3, and edited cells were selected for using puromycin (P8833; Sigma-Aldrich). To assess the functional impact of PPARG insertions and deletions (indels), cells were differentiated, stimulated, and FACS sorted based on CD36 expression (n = 5 independent sorts). Enrichment scores (ESs) were calculated as the log2 ratio of CD36⁺/CD36⁻ normalized counts. The impact of indels at each codon of PPARG was calculated based on previously published methods (9).

The endonuclease Cas9 and guides Int-sgRNA, Ex1-sgRNA, Ex1-sgRNA2, and Ex3-sgRNA (Supplementary Table 3) were introduced into THP-1 and SGBS cells by lentiviral transduction. The transduced THP-1 cells were sorted at one cell per well into 96-well plates (BD FACSAria II) and expanded. Infected SGBS cells were differentiated 7 days after infection as previously published (10). Genomic edits and zygosity were confirmed by Sanger sequencing.

Proteins were extracted, quantified, loaded into 4–12% Bis-Tris gels (NP0336; Invitrogen), and transferred onto 0.45-μm nitrocellulose membranes (1620115; Bio-Rad Laboratories). Antibodies used were Cell Signaling Technology cat. nos. 2435 (PPARγ N-terminus), 2443 (PPARγ C-terminus), 43603 (cyclophilin B), 41185 (actin), 2920 (Akt), and 4060 (phospho-Akt Ser473), and fluorescent secondary antibodies used were cat. nos. 5366, 5151, and 5470. Imaging was performed on the Odyssey CLx imager (LI-COR).

To evaluate the complementation of PPARγ using WT and M135 PPARγ isoforms, in vitro transcription was performed as previously described (11) with the following primers: WT forward, 5′-GAATTTAATACGACTCACTATAAGGAAATACGCCACCATGGGTGAAACTCTGGGAGAT-3′; M135 forward, 5′-GAATTTAATACGACTCACTATAAGGAAATACGCCACCATGGCAATTGAATGTCGTGTCT-3′; and reverse, 5′-CTAGGACATCGCAGTCTGCACCTAGTACAAGTCCTTGTAGATCTCCTG-3′.

The transcripts were electroporated into Pγ^−/− THP-1 cells in a 4-mm cuvette with one 400-V 5-ms square wave pulse (Xcell; Bio-Rad Laboratories). To match PPARγ protein expression, 2 μg M135 mRNA and 8 μg WT mRNA were used for each electroporation of 3 million cells (n = 5). Each sample was split into three aliquots for protein collection and differentiation with or without rosiglitazone treatment. RNA was extracted (R1050; Zymo Research) and sent for library preparation (Illumina Stranded mRNA Prep) and 100-bp paired-end sequencing (25 million reads per sample on the NovaSeq S4). Analysis was performed in R 4.1.3 using edgeR 3.36.0, limma v3.50.1, UpSetR v1.4.0, and fgsea v1.20.0 (12–15).

PPARG cDNA was synthesized (Twist Bioscience) with synonymous mutations (CCDS2609 c.C594T, G603C, T610A, C611G) to eliminate the Ex3-sgRNA recognition sequence and PCR amplified to generate cDNA encoding PPARγ2 and PPARγ M135. These sequences were cloned into doxycycline-inducible pCW (ID 184708; Addgene). Virus was produced, and SGBS Pγ^−/− cells were infected to create the SGBS Pγ^−/− + WT and SGBS Pγ^−/− + M135 cell lines. Insulin stimulation was performed after 24-h serum starvation with 100 nmol/L insulin (I9278; Sigma-Aldrich).

For all exome-sequenced cohorts, variants within the genomic coordinates of PPARG (chr3: 12287368–12434356 hg38) were extracted, and variant annotation was performed using SnpEff v4.3 (16). Nomenclature used for missense variants is for the canonical PPARG transcript ENST00000287820.10; protein ENSP00000287820.6. Function scores were obtained from the PPARG saturation mutagenesis (9). Serum HDL cholesterol (field 30760), waist circumference (field 48), serum triglycerides (TG; field 30870), systolic blood pressure (SBP; field 4080), and glycated hemoglobin (HbA_1c; field 30750) were extracted for all UK Biobank participants. SBP values were corrected for individuals reported to be taking blood pressure medication by adding 15 mm Hg (17), and TG values were log normalized. The TG to HDL (TG/HDL) ratio was log transformed and z normalized across UK Biobank. Regressions were adjusted for the covariates of age, age squared, sex, and the first 10 principal components of genetic ancestry.

All biobank data used in this study are accessible through applications to the respective databases. Data and resources are available on request from the corresponding author.

In our initial experiments, the endogenous PPARG locus was systematically disrupted by inducing indels in each coding PPARG exon in a human macrophage cell line (THP-1), a tractable model suitable for large-scale genetic perturbation that phenocopies PPARG-related transcriptional responses in adipocytes (9,18). Findings from THP-1s were confirmed and extended in human adipocytes models, the physiologically relevant cell type for metabolic disease. The effect of PPARG disruptions was measured by quantifying the ability of the resulting cells to transactivate CD36, a direct PPARγ transcriptional target (19). A custom lentiviral library of CRISPR/Cas9 constructs with sgRNAs targeting all coding exons, untranslated regions, and introns (n = 95) (Supplementary Table 1) was introduced into THP-1 monocytes at one construct per cell. The resulting population of genome-edited cells was differentiated into macrophages, stimulated with 1 μmol/L PPARγ agonist rosiglitazone, and sorted by FACS according to the expression of CD36 (Fig. 1B). The CD36+ and CD36− populations were sequenced to recover the identities of the sgRNAs, and an enrichment score (ES) was calculated based on the counts of each sgRNA in the CD36⁺/CD36⁻ pools (Fig. 1B). Intron-targeting sgRNAs introduced as controls had an ES of 0.487 ± 0.025. As expected, targeting sgRNAs to exon B of PPARG, specific to the PPARγ2 isoform, did not reduce CD36 activity (ES 0.622 ± 0.035), and most sgRNAs targeted to exons downstream of the PPARγ1 start site caused severe loss of CD36 transactivation (ES −1.33 ± 0.147). Intriguingly, five sgRNAs targeting exon 1 of PPARG, downstream of the PPARγ1 start site, which would be predicted to maximally disrupt the protein sequence, had little effect on CD36 transactivation (ES 0.597 ± 0.084), suggesting an intact PPARγ response in the cells that harbored them (Fig. 1C).

To further understand this unexpected finding, we analyzed data generated from a previously conducted saturation mutagenesis study of PPARG that contained indels at every codon of the PPARγ2 cDNA (CCDS2609) and calculated function scores (FSs) such that WT PPARG had an FS of 0 (Supplementary Table 2) (9). Most indels that caused frameshift terminations in PPARG completely inhibited CD36 transactivation, as shown by negative FSs. However, frameshifting indels in the 5′ region of the cDNA, predicted to cause early termination of protein translation, paradoxically retained cellular PPARγ transactivation (FS 0.0322 ± 0.0186). This concurred with our finding of tolerated exon 1 disruptions at the endogenous PPARG locus (Fig. 1C) and suggested a possible postsplicing mechanism for retained PPARγ activity. The tolerance to early frameshifting indels was observed until c.A403, after which frameshifting indels induced significant dysfunction (FS −1.16 ± 0.0074). These findings were replicated in experiments using prostaglandin J2, a putatively endogenous PPARγ ligand (9) (Supplementary Fig. 1A). C.A403-405 encodes a methionine, leading us to hypothesize an alternative translation initiation site, which would explain the preservation of PPARγ transactivation functions in transcripts with frameshift and nonsense-inducing indels before c.A403.

A translation initiation site at CCDS2609 c.A403 (hg38:chr3:12381414; ENSP00000287820 p.M135) would lead to a protein isoform (PPARγ M135) shorter than PPARγ2 by 134 amino acids with a predicted molecular weight of ∼40 kDa. To evaluate this hypothesis, we engineered clonal THP-1 monocytes with disruptions in exon 1, exon 3, and a PPARG intron using CRISPR/Cas9 (Ex1-sgRNA chr3:12379745, Ex3-sgRNA chr3:12392733, Int-sgRNA chr3:12363492) (Fig. 1A). Two independent cell lines were derived for each of the following genotypes: Int-sgRNA^+/+, Ex1-sgRNA^+/−, Ex1-sgRNA^−/−, and Ex3-sgRNA^−/− (Supplementary Table 3). Immunoblotting with PPARγ antibodies targeting N-terminal (p.Asp69) and C-terminal (p.His494) epitopes was performed on differentiated THP-1s. The N-terminal blot showed a 53-kDa band for full-length PPARγ1 in control (Int-sgRNA) and heterozygous Ex1-sgRNA samples, whereas no PPARγ bands were detected in homozygous Ex1-sgRNA or Ex3-sgRNA cells (Fig. 1E). The C-terminal PPARγ blot corroborated the PPARγ1 detection and identified a band of ∼40 kDa in Ex1-sgRNA cells, matching the predicted size of PPARγ M135, with higher intensity in Ex1-sgRNA^−/− cells (Fig. 1F). Additional smaller bands may reflect alternative translation initiation sites upstream of p.M135. No PPARγ protein was detected in Ex3-sgRNA^−/− cells, indicating complete loss of PPARγ (Fig. 1F).

To compare the transcriptional profiles of cells expressing WT and PPARγ M135, the Int-sgRNA^+/+ and Ex1-sgRNA^−/− clonal cell lines were differentiated into macrophages, stimulated with 1 μmol/L rosiglitazone and assessed for gene expression at several canonical PPARγ target genes (Fig. 1G). On agonist induction, the Ex1-sgRNA^−/− cells expressed significantly higher levels of ANGPTL4 (20) (P = 0.001), PDK4 (21) (P = 6.2e−6), and PLIN2 (22) (P = 5e−4) than the Int-sgRNA^+/+ cells. These PPARγ M135–expressing cells also showed nonsignificant increases in CD36 and FABP4 expression. Because PPARγ plays a role in monocyte-to-macrophage differentiation, we measured ITGAM (CD11b) (23) and CD68 (24) to assess phorbol 12-myristate 13-acetate (PMA)–induced differentiation as a potential confounder. After PMA treatment, CD11b and CD68 expression increased similarly in Ex1-sgRNA^−/− and Int-sgRNA^+/+ cells (Supplementary Fig. 1B).

We subsequently isolated and compared the transactivation potentials of WT and PPARγ M135. A PPRE-driven luciferase reporter (25) and WT PPARγ or PPARγ M135 mRNA were transfected into HEK293s, which have minimal endogenous PPARγ activity (26). In this system, PPARγ M135 activated transcriptional activity more potently than WT when induced with rosiglitazone (Supplementary Fig. 1C). We further compared the stability of PPARγ M135 with that of WT by performing a cycloheximide chase (27) in the heterozygous Ex1-sgRNA cells that generate both isoforms (Supplementary Fig. 1D). The M135 isoform degraded more slowly than PPARγ1 (Supplementary Fig. 1E), which could contribute to its enhanced transactivation potential.

We next sought to isolate the activity of PPARγ M135 and evaluate its effect on global transcriptional profiles in comparison with full-length WT PPARγ. To compare the direct transcriptional responses of M135 and WT, we performed RNA sequencing on PPARγ-null (Pγ^−/−) THP-1 monocytes electroporated with in vitro–transcribed mRNA of each of the two PPARγ isoforms; EGFP mRNA was also electroporated as a process control. In preliminary experiments, mRNA amounts for each isoform were titrated to express similar amounts of protein at the point of harvest (Supplementary Fig. 2A and B), such that 2 μg PPARγ M135 mRNA and 8 μg PPARγ WT were used for each electroporation. The electroporated cells were differentiated into macrophages to mimic the cellular context in which PPARγ is active (28) and treated with 0 (−) and 1 μmol/L (+) rosiglitazone. As a positive control, WT (Pγ^+/+) THP-1 cells with intact PPARG were also treated with or without rosiglitazone and transcriptionally profiled (Fig. 2A).

After filtering for low expression, 16,732 transcripts were retained for analysis across all samples. As expected, rosiglitazone treatment increased gene expression of canonical PPARγ target genes, including CD36, FABP4 (3), and PLIN2 (22) (Fig. 2B). Remarkably, PPARγ M135–electroporated (Pγ^−/− + M135) cells exhibited greater agonist-induced transcriptional responses for some of these targets compared with Pγ^−/− + WT or Pγ^+/+ THP-1 macrophages, despite comparable PPARγ protein levels (log2 fold change [log2FC]: CD36: Pγ^−/− + M135 4.22, Pγ^+/+ 2.94, and Pγ^−/− + WT 1.62; PLIN2: Pγ^−/− + M135 3.35, Pγ^+/+ 3.24, and Pγ^−/− + WT 1.50).

To comprehensively evaluate if M135 generated a stronger agonist-induced transcriptional response than WT in Pγ^−/− cells, we performed a series of differential expression analyses. We first identified the top-ranked PPARγ target genes, defined as the 50 most significant differentially expressed genes (DEGs) in Pγ^+/+ cells with or without rosiglitazone, and then compared the log2FC of the same genes across Pγ^−/− + WT and Pγ^−/− + M135 with or without rosiglitazone (Fig. 2C). Of the top 50 Pγ^+/+ DEGs, 46 were also differentially expressed in Pγ^−/− + M135 (P < 0.05 [same sign log2FC]), whereas only 33 were differentially expressed in Pγ^−/− + WT. Moreover, in several key genes, including PDK4, DYSF, ANGPTL4, ALOX5AP, and CYBB (29–31), Pγ^−/− + M135 had a greater activation or repression than Pγ^+/+ cells (Fig. 2C), despite transient and lower PPARγ protein expression per cell (Supplementary Fig. 2A and B). Across all the Pγ^+/+ DEGs (n = 1,779), the magnitude of gene expression change was more similar for Pγ^−/− + M135 cells (slope 0.85) than for Pγ^−/− + WT cells (slope 0.36), indicating greater potency of M135 in mediating agonist-induced PPARγ gene expression response than WT (Fig. 2D).

We then queried for DEGs specific to PPARγ M135 to assess if the lack of AF-1 domain in M135 resulted in transactivation/repression of genes not regulated by WT PPARγ. In response to rosiglitazone treatment, Pγ^−/− + M135 had the greatest number of DEGs (n = 4,247) (Fig. 2E), as compared with Pγ^−/− + WT (n = 1,794) or Pγ^+/+ (n = 1,779). Of the 4,247 DEGs, 2,313 were exclusive to Pγ^−/− + M135 (Fig. 2E). To understand the gene expression programs captured by these putative M135-specific genes, we performed gene set overrepresentation analysis (32,33) among the Gene Ontology Biological Process pathways (34,35) and found 28 overrepresented pathways that were confirmed to have been altered by rosiglitazone treatment in M135-complemented cells (P < 0.05) (Supplementary Table 4). Among these, 21 of 28 were similarly altered, although to a lesser degree, by rosiglitazone treatment in either WT-electroporated cells or Pγ^+/+ THP-1s (Fig. 2F). Taken together, these analyses suggest that PPARγ M135 regulates similar gene expression programs as WT, but more potently when induced by rosiglitazone, perhaps because of derepression from the loss of the N-terminal AF-1 domain (36).

Because many of the major metabolic effects of PPARG on human physiology occur in adipocytes (8), we evaluated whether adipocytes could also generate PPARγ M135 and to what functional consequence. We targeted a human preadipocyte cell line (SGBS (37)) with disruptions in exon 1, exon 3, and a PPARG intron (Ex1-sgRNA2 cut site at hg38 chr3:12379716, Ex3-sgRNA chr3:12392733, and Int-sgRNA chr3:12363492) (Fig. 3A and Supplementary Table 3). The edited cells were treated with inducers of adipocyte differentiation and examined for PPARγ protein expression, target gene expression, and adipocyte differentiation efficiency. After 4 days of differentiation, we were able to detect both PPARγ WT and M135 in exon 1–targeted cells, whereas exon 3–targeted cells expressed no PPARγ, and control cells expressed only WT PPARγ (Fig. 3B and C). These results indicated that like THP-1 macrophages, SGBS adipocytes are capable of alternatively generating PPARγ M135 in response to disruptive mutations targeted to exon 1.

To evaluate the ability of preadipocytes expressing PPARγ M135 to activate PPARγ target genes, we queried the gene expression of several targets during early adipocyte differentiation (Fig. 3D). SGBS preadipocytes that expressed PPARγ M135 increased expression of CD36 (P = 0.022) and PDK4 (P = 5.2e−5) to a significantly greater extent than PPARγ WT–expressing cells, similar to macrophages (Fig. 1G) and showed a trend toward increased expression of FABP4 (P = 0.22). Furthermore, we examined adiponectin (ADIPOQ), an adipokine and PPARγ target specific to adipocytes (38), and found it also to be significantly upregulated by PPARγ M135 expressing–adipocytes (P = 0.017) compared with WT (Fig. 3D). Cells targeted at exon 3, which expressed no PPARγ, did not upregulate the expression of any of these genes.

Finally, we characterized the ability of PPARγ M135–expressing SGBS cells to mature into adipocytes and accumulate lipids during differentiation (Fig. 3E and F). Exon 1–targeted PPARγ M135–expressing SGBS cells differentiated and accumulated lipids at the same rate as control PPARγ WT–expressing cells (P = 0.933), whereas exon 3–targeted cells had significantly reduced lipid accumulation (P = 2e−16). These analyses show that like macrophages, human preadipocytes can generate the PPARγ M135 isoform, and the truncated isoform is fully functional in driving differentiation into mature, lipid-laden adipocytes.

To isolate the effect of PPARγ M135, we transduced Pγ^−/− SBGS cells (i.e., Ex3-sgRNA targeted cells) with doxycycline-inducible WT PPARγ2 cDNA (SGBS Pγ^−/− + WT) or PPARγ M135 cDNA (SGBS Pγ^−/− + M135) transgenes and evaluated differentiation/lipid accumulation and insulin response (Fig. 4A). These cells only express PPARγ (WT or M135) when treated with doxycycline (Fig. 4B) and differentiate to a similar degree only when PPARγ is induced (P = 0.56) (Fig. 4C and D).

To assess insulin response, control (intronic), SGBS Pγ^−/− + WT, and SGBS Pγ^−/− + M135 cells were differentiated, stimulated with insulin, and immunoblotted for phosphorylated Akt (S473) and total Akt. In response to insulin, all cell lines phosphorylated Akt, and the response was augmented in doxycycline-treated SGBS Pγ^−/− + WT and SGBS Pγ^−/− + M135 (ANOVA P = 1.33e−13; Tukey honest significant difference [HSD] P = 3.67e−8 for WT and P = 3.43e−11 for M135) (Fig. 4E and F). Notably, PPARγ M135–expressing adipocytes showed increased insulin-stimulated Akt phosphorylation compared with both WT (P = 0.025) and control SGBS (P = 0.045), indicating an enhanced insulin response.

Next, we sought to evaluate the potential in vivo consequence of nonsense mutations in the PPARG sequence before chr3:12381414 by identifying human carriers of such mutations and performing genotype:phenotype correlation under the hypothesis that carriers would not exhibit insulin resistance given the enhanced molecular activity of PPARγ M135 resulting from AF-1 domain deletion. Across biobanks and databases comprising >1.2 million individuals with sequencing at the PPARG locus, we found only seven carriers of nonsense mutations before chr3:12381414 (Supplementary Table 5). Among these, two had no evidence of metabolic syndrome or insulin resistance past 50 years of age, one had type 2 diabetes but no ascertainment of insulin resistance or metabolic syndrome, and four had no available phenotypic information.

Because the number of human pre-M135 nonsense mutation carriers was insufficient to make robust inferences, we turned to carriers of PPARG missense variants to test the hypothesis that genetic variants abrogating AF-1 domain function would enhance PPARγ activity and thereby increase insulin sensitivity in vivo (Fig. 5A). Although missense variants are not equal to having the M135 isoform, they can model how disruptions to the AF-1 domain affect PPARγ activity in vivo. We rationalized this hypothesis based on recent data showing that the AF-1 domain intramolecularly binds to the LBD of PPARγ, and this interaction inhibits ligand-dependent activity (36). We identified all carriers of rare (minor allele frequency <0.001) protein-coding variants in PPARG in the UK Biobank (n = 454,787) (39) and analyzed the cellular function and amino acid position of the variants carried in relation to the insulin sensitivity–related phenotypes of the individuals carrying them. To quantify PPARγ activity, we leveraged the PPARγ FS derived from our previously published deep mutational scan (9), in which every possible missense variant was scored by its transactivation of CD36 (9). We found 1,250 carriers of 260 unique rare, protein-coding PPARG variants and partitioned them by pre-/post-M135 and by the BLOSUM62 substitution matrix, which quantifies the tolerance of amino acids to substitution across evolutionary distance (40). Variants were categorized as conservative (BLOSUM62 >0) or nonconservative (BLOSUM62 <0). Nonconservative substitutions in AF-1 (median FS 2.21) had higher PPARγ FSs than conservative substitutions (median FS 1.56), whereas nonconservative substitutions post-M135 in the DBD and LBD (median FS −0.158) showed decreased PPARγ FSs relative to conservative mutations (median FS −0.376; ANOVA P < 2e−16; adjusted P < 0.005 for each pairwise comparison by Tukey HSD) (Fig. 5B). These data support the hypothesis that missense variants disrupting AF-1 domain function increase the transcriptional activity of PPARγ.

To quantify insulin sensitivity in these PPARG missense variant carriers, we computed a per-individual metabolic syndrome severity score (METSS), a measure of insulin sensitivity determined from waist circumference, SBP, serum TGs, HDL cholesterol, and serum glucose (HbA_1c), using methods analogous to those previously published (41–43). We calculated METSS for the 368,911 individuals in the UK Biobank who had all five measurements ascertained, including 908 of the 1,250 carriers of rare protein-coding PPARG variants (Supplementary Fig. 3A and B). Of the rare PPARG missense variant carriers in the UK Biobank, 340 pre-M135 and 370 post-M135 carriers had computable METSS values. As with the PPARγ function score analysis above (Fig. 5B), we partitioned each group according to conservative and nonconservative BLOSUM62 to examine the effect of each variant category on METSS (Fig. 5C). Under the model that amino acid substitutions that abrogate AF-1 domain function would increase PPARγ activity and thereby decrease METSS, we hypothesized that nonconservative missense variants in AF-1 (pre-M135) would confer lower METSS as compared with conservative amino acid substitutions that would preserve AF-1 function. Conversely, we expected that nonconservative mutations post-M135 in the DBD or LBD would increase METSS, as is the case for lipodystrophy (44). As a positive control, we identified carriers of post-M135 disruptive (i.e., nonsense and frameshift causing) PPARG variants (n = 14) in our cohort and found their METSS to be significantly elevated (median METSS 1.5) (Fig. 5D) compared with other PPARG variant carriers (ANOVA P = 4.1e−6; Tukey P < 8.6e−4) and the general UK Biobank population (Welch t test P = 1.1e−4). We observed an ordinal trend, with nonconservative pre-M135 variant carriers having the lowest METSS (median METSS −0.079), followed by conservative pre-M135 (median METSS 0.0064), conservative post-M135 (median METSS 0.24), and finally nonconservative post-M135 (median METSS 0.36) carriers. The difference between pre- and post-M135 nonconservative variant carriers was significant (Welch t test P = 0.016).

We performed a similar analysis alternatively using the ratio of serum TGs to HDL cholesterol (TG/HDL) as a surrogate measure of insulin sensitivity (45,46). The trends observed with METSS were consistent in the TG/HDL results; carriers of post-M135 disruptive (i.e., nonsense and frameshift causing) PPARG variants (n = 15) had the highest values compared with other PPARG variant carriers (median TG/HDL 1.65; ANOVA P = 7.31e−7; Tukey HSD P < 1.1e−4) (Fig. 5D), and the nonconservative pre-M135 variant carriers had the lowest TG/HDL (n = 108; median TG/HDL −0.080). The difference between pre- and post-M135 nonconservative variant carriers was significant (Welch t test P = 0.017).

We separately analyzed the well-known PPARG p.P12A variant (rs1801282; minor allele frequency 0.1050) that is associated with decreased type 2 diabetes risk (47) and occurs frequently in the general population. Under the above partitioning scheme, PPARG p.P12A would be classified as pre-M135 nonconservative (FS 1.3; BLOSUM62 score −1). Carriers of the p.P12A allele (n = 80,882) in the UK Biobank had significantly decreased METSS (per allele effect size −0.039; P = 2e−16) (Supplementary Fig. 3C and Supplementary Table 6). These data are suggestive of a model in which AF-1–disrupting variants (both common and rare) can improve insulin sensitivity in vivo.

Here, we characterize a novel isoform of PPARγ, termed PPARγ M135, which lacks the AF-1 domain and can be generated from an alternative translational start site. In both macrophages and adipocytes, PPARγ M135 is transcriptionally active, ligand inducible, and more potent than WT PPARγ, likely because of derepression from the loss of the AF-1 domain. We also assess insulin sensitivity in human carriers of PPARG variants, demonstrating that variants impairing the AF-1 domain may protect carriers from insulin resistance. Our data support a model for in vivo derepression of PPARγ in humans that is metabolically beneficial.

Our findings that PPARγ M135 enhances transactivation and improves metabolic health align with studies of naturally occurring and synthetic PPARγ variations. Previous investigations demonstrated that deleting the PPARγ N-terminus increases transcriptional potency compared with WT in NIH-3T3 cells (48,49), and a MAPK phosphorylation site at PPARγ p.S112 inhibits PPARγ transactivation. PPARγ p.S112A, which lacks the phosphorylation site, is more transcriptionally active (50). Additionally, PPARγ2 p.P12A (rs1801282), which is associated with a reduced risk of type 2 diabetes (47), weakens the interaction between PPARγ and its corepressor NCoR, resulting in increased expression of PPARγ target genes and improved insulin sensitivity in mice (51). Our data corroborate these mechanisms, because METSS decreased per p.P12A allele. Furthermore, SUMOylation at p.K107 inhibits ligand-induced transactivation of PPARγ targets (52), and removing that modification increases insulin sensitivity without increasing adiposity in mice (53). Together, these studies illustrate that impairing the AF-1 domain increases PPARγ activity and insulin sensitivity.

Regarding therapeutic development, our study proposes a new method to activate PPARγ distinct from TZDs, which target the LBD. We propose the AF-1 domain as a therapeutic target that is mechanistically distinct from TZDs and selective PPARγ modulators, because removing AF-1 would derepress rather than activate PPARγ. Accordingly, our data show that adipocytes engineered to produce PPARγ M135 expressed higher levels of adiponectin, an insulin-sensitizing adipokine (54), and had increased Akt phosphorylation in response to insulin stimulation. Further supporting this proposal are the murine models of human and synthetic PPARG variants that increase PPARγ activity via impairing AF-1 (p.P12A (51), p.S112A (55), and p.K107 (53)), showing enhanced insulin sensitivity compared with WT littermates.

Limitations of our study include the use of in vitro cell models, the number of human carriers with analyzable PPARG protein–coding variants, and the generalizability of the UK Biobank population. The THP-1 monocyte and SGBS preadipocyte cell lines, although human, do not fully replicate in vivo conditions; however, they have shown consistent results in prior PPARG variant studies (8,9,18). Furthermore, PPARγ is active in other tissues, including muscle and liver, which may have additional metabolic consequences (56,57). These could be the subjects of future investigations to fully dissect the metabolic consequences of PPARγ M135. In addition, the number of human carriers of AF-1 domain nonconservative missense variants (n = 94) limited our statistical power to detect changes in metabolic syndrome severity in this group. Moreover, the UK Biobank represents a relatively healthy middle-aged population of largely British ancestry, which is not representative of global populations (58). In the future, our approach could be easily reapplied to larger cohorts and multiethnic samples to corroborate and strengthen our findings as they become available to investigators. Another future direction would be to validate the therapeutic hypothesis of generating PPARγ M135 in vivo using transgenic murine models and evaluate tissue specificity.

In summary, we present PPARγ M135, a novel isoform of PPARG arising from an alternative translational start site, as a more potent transactivator than full-length PPARγ. This work points to a new mechanism of activating PPARγ by inhibiting the AF-1 domain, which could lead to more effective treatments for insulin resistance–related disorders.

Acknowledgments. The authors thank the All of Us participants, because the All of Us Research Program would not be possible without their partnership. The authors are grateful to Angela Liu for technical assistance. Microscopy was performed at the Nikon Imaging Center at University of California San Diego with the support of P. Guo and R. Sanchez. This research was conducted using the UK Biobank resource under applications 41189 and 51436. The authors acknowledge the participants of and the contributions made by UK Biobank (www.ukbiobank.ac.uk) in providing the data and resources used in this study.

Funding. This work was supported by grants 1R01DK123422 and R01HL159760 (A.R.M.) and 1R01DK125490 (J.F., A.R.M.) from the National Institute of Diabetes and Digestive and Kidney Diseases and University of California (UC) San Diego/UC Los Angeles Pilot and Feasibility grant P30 DK063491 (A.R.M.). This publication includes data generated at the UC San Diego Institute for Genomic Medicine Genomics Center using an Illumina NovaSeq 6000 that was purchased with funding from National Institutes of Health Shared Instrument grant S10 OD026929. The All of Us Research Program is supported by the Office of the Director, National Institutes of Health, through regional medical centers (grants 1 OT2 OD026549, 1 OT2 OD026554, 1 OT2 OD026557, 1 OT2 OD026556, 1 OT2 OD026550, 1 OT2 OD 026552, 1 OT2 OD026553, 1 OT2 OD026548, 1 OT2 OD026551, and 1 OT2 OD026555), an interagency agreement (AOD 16037), federally qualified health centers (grant HHSN 263201600085U), a data and research center (grant 5 U2C OD023196), a biobank (grant 1 U24 OD023121), a participant center (grant U24 OD023176), a participant technology systems center (grant 1 U24 OD023163), communications and engagement (grants 3 OT2 OD023205 and 3 OT2 OD023206), and community partners (grants 1 OT2 OD025277, 3 OT2 OD025315, 1 OT2 OD025337, and 1 OT2 OD025276).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. X.D., K.M.-L., S. Hu, R.D., G.B., R.H., J.F., S. He., and A.R.M. were involved in data interpretation. X.D., K.M.-L., and A.R.M. designed the experiments and drafted the initial manuscript. A.R.M. conceived the study. All authors generated/acquired and analyzed the data and were involved in manuscript revision. A.R.M. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.