Cell-Intrinsic Tumorigenic Functions of PPARγ in Bladder Urothelial Carcinoma

Bladder urothelial carcinoma (UC) is the ninth leading cause of cancer worldwide, with an estimated 81,400 new cases in 2020 and approximately 17,980 annual deaths per year in the United States alone (1). Clinically, UC is stratified into non–muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC), both contributing to disease morbidity and mortality (2). MIBC is currently treated by radical cystectomy preceded or followed by chemotherapy including methotrexate, vinblastine, doxorubicin, and cisplatin (3). Recently, large-scale genomic sequencing of hundreds of primary patient samples has revealed distinct clusters of MIBC, most broadly stratified into luminal and basal subcategories based on common transcriptional features (3, 4). The identification of molecular signatures and drivers of these programs will pave the way for improved targeted therapies, which are needed because there are currently few available to treat localized and metastatic UC.

One transcription factor reported to be a hallmark of luminal MIBC is the PPAR gamma (PPARγ; ref. 5). PPARγ is a member of the nuclear receptor family, which acts as molecular sensors to regulate gene expression based on environmental and metabolic cues (6). PPARγ binds DNA as a heterodimer with retinoid X receptor (RXR) and exists in complexes with corepressor or coactivator proteins depending on its activation state. Associations with coactivator or corepressor proteins facilitate chromatin remodeling and target gene activation or repression, respectively (7). During development, PPARγ and CCAAT-enhancer-binding proteins (C/EBP) transcription factors induce adipogenesis from mesenchymal progenitor cells (8). In differentiated postmitotic adipocytes, PPARγ regulates the expression of genes involved in lipid and glucose metabolism, as well as the expression of endocrine “adipokine” factors which affect whole-body energy homeostasis (9).

Long-term use of certain thiazolidinediones, synthetic agonists of PPARγ, Pioglitazone, but not rosiglitazone, used to manage blood glucose levels in type 2 diabetics, has been associated with increased risk of bladder cancer development (10–12). In a preclinical study employing a carcinogen-induced model of bladder cancer, rosiglitazone treatment increased incidence and size of UC in rats in a dose-dependent manner (13). Of note, rosiglitazone treatment alone did not induce UC tumors, suggesting a potential synergistic relationship between PPARγ activation and carcinogen exposure in UC development in vivo. Nevertheless, these observations have led our group and others to speculate that PPARγ might be an important factor in UC development and progression in a subset of patients. Indeed, multiple recent studies link PPARγ activity to UC, describing its ability to promote disease progression in both cell-autonomous (5, 14–16) and non–cell-autonomous manners (17). However, the direct molecular mechanisms by which PPARγ promotes tumor growth have not been described, nor have the downstream mediators of PPARγ–RXR transcriptional activity been defined. In this study, our objective was to further elucidate cell-intrinsic roles of PPARγ signaling in UC and identify candidate target genes responsible for UC progression based on genome-wide approaches. Importantly, our data demonstrated for the first time direct transcriptional regulation of the oncogene Sonic Hedgehog (SHH) by PPARγ. Moreover, we demonstrate a functional link between SHH- and PPARγ-dependent cell functions, thus providing insights into the tumorigenic nature and molecular mechanism involved in PPARγ-dependent UC tumorigenesis.

Bladder UC tissue array containing 60 cases of UC plus adjacent normal bladder tissue was purchased from US Biomax, Inc. (cat. BC12011c). The Cancer Genome Atlas (TCGA) datasets (2, 18) were analyzed for PPARG and RXRA mutational burden, copy-number variation, mRNA upregulation, and coexpression patterns using cBioPortal for Cancer Genomics (19, 20), and the Sanchez-Carbayo dataset was analyzed for mRNA upregulation (Oncomine). PPARG expression levels of various cell lines were extracted from RNA sequencing (RNA-seq) data available on the Cancer Cell Line Encyclopedia. The CERES dependency score (Fig. 3H) is based on data from a cell depletion assay. A lower score indicates a higher likelihood that the gene of interest is essential in a given cell line. A score of 0 indicates a gene is not essential; correspondingly -1 is comparable with the median of all pan-essential genes (21).

Human UC cell lines (5637, HT-1197, HT-1376, TCCSUP, and T24) were purchased from the American Type Culture Collection (ATCC), with the exception of Cal29 (DSMZ, cat. ACC-515). Cell lines were authenticated by the ATCC or DSMZ before distribution via morphology, karyotyping, and PCR-based approaches. All cell lines were cultured in RPMI 1640 medium containing l-glutamine (Thermo Fisher Scientific, cat. 11875–085) supplemented with 10% FBS (Gemini Bio-Products, cat. 900–108) and were tested regularly for mycoplasma contaminations using the Lonza MycoAlert assay. All cell lines were used for 6 weeks for eight to ten passages. Human single-guide RNAs (sgRNA) targeting PPARG #6 (cattacgaagacattccatt) and #9 (caactttgggatcagctccg) along with control gRNA targeting mouse Rosa26 locus (aagatgggcgggagtcttct) were cloned into LentiCRISPRv2 plasmid. Mature antisense human PPARG shRNA #2 sequence (clone ID: TRCN0000001672), SHH shRNA #4 sequence (Clone ID: TRCN0000033304), and shRNA #5 sequence (Clone ID: TRCN0000033305), along with scrambled (SCR) control were purchased from Dharmacon and cloned into a pLKO lentiviral plasmid. Lentivirus was prepared and cells were transduced as previously described (22).

Human SHH prepackaged lentiviral particles (Hygromycin selective) were purchased from G&P Biosciences (cat. LTV7062). All experiments were performed with cells that survived puromycin and/or hygromycin selection and displayed knockdown of PPARG, SHH, and overexpression of SHH, as assayed by Western blot. Primary bladder epithelial cells (A/T/N (ATCC PCS-420–010) were purchased from the ATCC and processed immediately for protein and RNA sampling.

Rosiglitazone (Sigma-Aldrich, cat. R2408) was used at a final concentration of 100 nmol/L in all experiments unless stated otherwise. The PPARγ inverse agonist T0070907 (Santa Cruz Biotechnology, cat. sc-203287) was used at a final concentration of 100 nmol/L in all experiments. An equivalent volume of DMSO was used as a control for all experiments involving PPARγ pharmacologic agents.

Total RNA was isolated using TRIzol reagent (Thermo Fisher Scientific, cat. 15596026) and RNeasy mini kit (QIAGEN, cat. 74104). Reverse transcription was performed using High-Capacity RNA-to-cDNA (Applied Biosystems, cat. 4387406). qRT-PCR was performed using ViiA7 Real-Time PCR system (Applied Biosystems) with TaqMan master mix (Life Technologies). TaqMan probes purchased from Thermo Fisher Scientific were used to quantitate expression of PPARG (cat. Hs01115513_m1), FABP4 (cat. Hs01086177_m1), PLIN2 (cat. Hs00605340_m1), SCD (cat. Hs01682761_m1), LPL (cat. Hs00173425_m1), PPARA (cat. Hs00947536_m1), SHH (cat. Hs00179843_m1), MMP3 (cat. Hs00968305_m1), MMP9 (Hs00957562_m1), and tissue inhibitor of metalloproteinase-1 (TIMP1, cat. Hs01092511_m1) and normalized to housekeeping gene TBP (Hs00427620_m1).

Western blot was performed as previously described (22). Primary antibodies were diluted 1:1,000 in 5% w/v nonfat milk (except GAPDH, 1:10,000), and secondary antibodies were diluted 1:2,000 in 5% w/v nonfat milk. PPARγ (cat. 2435), SHH (cat. 2207), MMP3 (cat. 14351), MMP9 (cat. 13667), TIMP1 (cat. 8946), KRT5 (cat. 25807), FOXA1 (cat. 53528), GAPDH (cat. 2118), and anti-Rabbit IgG, horseradish peroxidase–linked (cat. 7074) antibodies were purchased from Cell Signaling Technology. KRT6 (cat. MBS2526156) was purchased from MyBioSource.

Human bladder cancer tissue array was processed for IHC. Slide was baked in a dry oven for 20 minutes at 55°C, then deparaffinized by incubation in 100% xylene for 15 minutes, twice, followed by rehydration [100% ethanol (EtOH), 95% EtOH, 70% EtOH, and 100% dH₂O for 5 minutes each]. Slide was then boiled in antigen unmasking solution (1 mol/L sodium citrate solution, pH 6.0) for 20 minutes and left to cool on bench for 20 minutes. Endogenous peroxidase activity was blocked with 1% H₂O₂/H₂O solution for 30 minutes. Slide was washed in 1X TT buffer for 3 × 5 minutes and incubated in blocking buffer (5% BSA, 2% goat serum diluted in 1X TT buffer) for 1 hour at room temperature (RT). Samples were incubated overnight in PPARγ antibody (1:200) in blocking buffer at 4°C. Slide was washed in 1X TT buffer for 3 × 5 minutes and incubated in secondary antibody (1:200 biotinylated goat anti-Rabbit, diluted in 1X TT) for 1 hour at RT. Slide was further processed using Vectastain Elite ABC kit (Fisher Scientific, cat. PK6100) according to the manufacturer's instructions, and staining was visualized with diaminobenzidine (DAB) peroxidase substrate kit (Vector Laboratories, cat. SK4100). Slide was counterstained with hematoxylin and dehydrated before mounting and imaging. Where mentioned, densitometry was assessed using NIH Image J software package.

Chromatin immunoprecipitation (ChIP) was performed with whole-cell extracts isolated from 5637 parental cells using 2.5 μg PPARγ (Santa Cruz Biotechnology, cat. sc-7196), 2.5 μg RXRα/β/γ (Santa Cruz Biotechnology, cat. sc-774), or 2.5 μg Normal Rabbit IgG (Cell Signaling Technology, cat. 2729) antibodies for IP. Briefly, confluent 10 cm dishes of cells were cross-linked with 1% formaldehyde for 15 minutes at RT and quenched with 125 mmol/L glycine for 5 minutes. Cells were harvested by scraping, and pellets were resuspended in 200 μL SDS lysis buffer [50 mmol/L HEPES/NaOH, pH 7.5, 1% SDS, 10 mmol/L EDTA, 1 mmol/L PMSF, and Roche ULTRA protease inhibitor cocktail (cat. 05892791001)] on ice for 10 minutes. Sonication was performed using Bioruptor Pico (Diagenode, cat. B01060010) for 4 cycles of 30 seconds on, 30 seconds off, followed by centrifugation of lysates. Sixty-six microliter of sheared chromatin from each tube was then diluted 10X, with 5% saved as Input DNA and the rest prepared for either PPARγ, RXR, or IgG IP.

PPARγ IP, RXR IP, IgG IP, and Input libraries were prepared in duplicate from two independent biological replicates. For ChIP sequencing (ChIP-seq), sequencing data were mapped to the human genome (GRCh38) using STAR (23) with parameters appropriate for ungapped alignments. Peaks were called for each sample with input samples as background by HOMER (24). HOMER was also used for differential peak calling (PPARγ vs. Input and RXR vs. Input) and to annotate peaks to proximal genes as described in Ensembl v85 (http://www.ensembl.org/index.html). Motif enrichment analysis was performed using HOMER against the standard list of known motifs with consideration of lengths 8, 10, 12, 15, and 18 bp. For bioinformatics analyses displayed in Fig. 5A, “high-confidence sites” were defined by the following criteria: peak score ≥ 10 (≥ 1 read per million) and RXR peak called with strict overlap. Gene ontology (GO) was performed using Metascape (metascape.org). ChIP-seq data are available at GEO database (GSE166803).

Total RNA was isolated from four technical replicates of 5637 control, 100 nmol/L T0070907-treated, PPARG sgRNA 6, and PPARG sgRNA 9 cell lines as described above. NEBNext Ultra II RNA Library Prep Kit for Ilumina was used for preparing approximately 250 to 300 bp insert cDNA, nondirectional libraries according to the manufacturer's instructions at Novogene Corporation. Libraries were sequenced using the Ilumina platform with paired-end 150 bp sequencing at 20 M raw reads/sample at Novogene. Raw sequence files (fastq) for 16 samples were mapped using Salmon (25) against the human transcripts described in GENCODE (version 28, built on the human genome GRCh38.p12). Transcript counts were summarized to the gene level using tximport (26), and normalized and tested for differential expression using DESeq2 (27). Statistics for each contrast of interest were exported for further analysis. Genes used for GO met the following criteria: significantly up- or down-regulated by inverse-agonist or CRISPR treatment at least 1.5-fold with P_adj < 0.05. GO was performed using Metascape (metascape.org). RNA-seq data are available at GEO database (GSE166803).

Five thousand cells of each cell line were plated in triplicate on 6-well plates. Cells were treated for 8 days with media containing DMSO or 100 nmol/L T0070907 and then prepared using the FITC–Annexin V and PI Kit (BD Biosciences, cat. 556547) according to the manufacturer's instructions. Flow cytometry was performed using the BD Accuri C6 instrument, with viable cells represented as the double-negative population.

Twenty-five thousand cells of tested conditions were plated in triplicate on 6-well plates. The following day (day 0), cells were trypsinized and counted using the Countess Automated Cell Counter (Invitrogen, cat. C10281), as per the manufacturer's instructions with Trypan blue. Cells were then counted again at the indicated timepoints, and fresh media were added every 2 days.

Five thousand cells of each cell line were plated in triplicate on 6-well plates. The next day, media containing either 100 nmol/L T0070907 or an equivalent volume of DMSO were added to the wells in triplicate. UC cells were treated with DMSO or T0070907 for 9 days (with the exception of T24, 5 days) with fresh medium supplied every 2 days. At the end, cells were washed in PBS and stained with crystal violet solution (0.5% crystal violet in 20% methanol) for 10 minutes. Plates were then washed twice with PBS and allowed to dry. Quantitation was performed by adding 500 μL methanol to dry wells and incubating at RT with rocking for 20 minutes. Absorbance was read using a spectrophotometer at 570 nm, and data are represented as fold change relative to DMSO readings.

Five thousand cells of the 5637 cell line were plated on 6-well plates. The following day, cells were serum starved for 24 hours to synchronize the population, after which media containing DMSO and 100 nmol/L T0070907 were added. Cells were harvested after 5 days of drug treatment, fixed using ice-cold 70% ethanol, and stored at −20°C for 2 hours. Following centrifugation and removal of ethanol, cells were washed twice in FACS buffer (PBS containing 2% FBS and 1 mmol/L EDTA). For dual Ki67/propidium iodide (PI) staining to determine nonproliferative (G₀ population), samples were resuspended in 200 μL FACS buffer and stained for 30 minutes at RT in the dark with Ki67-FITC monoclonal antibody (SolA15; Life Technologies, cat. 11–5698–82) at 1:100. Cells were then washed twice with FACS buffer and resuspended in 500 μL PI staining solution (PBS containing 50 μg/mL PI, 100 μg/mL RNase A, and 2 mmol/L MgCl₂). Cells were strained through 40 μm filters and incubated at RT in the dark for 20 minutes before analysis by flow cytometry. For analysis of the cell-cycle phases, cells were stained with PI antibody only.

Cells were labeled with 10 mmol/L EdU over 4 hours, collected by trypsinization, and fixed (4% PFA). The incorporated EdU was detected by a “click-It reaction” with Alexa Fluor 488 according to the manufacturer's instructions (Thermo Fisher Scientific). Data were recorded by flow cytometry, and resultant data were analyzed with the FlowJo 10.6.2 software (https://www.flowjo.com).

Twenty-five thousand NIH-3T3 cells (ATCC) were infected with murine stem cell virus (MSCV)-retroviral empty vector or vector containing PPARG cDNA resistant to shRNA #2. When cells were confluent, induction medium was added to the plates for 48 hours (DMEM containing 10% FBS supplemented with 125 nmol/L indomethacin, 0.5 mmol/L isobutylmethylxanthine, 1 mmol/L dexamethasone, 20 nmol/L insulin, and 1 μmol/L rosiglitazone). After 48 hours, medium was changed to DMEM containing 10% FBS supplemented with 20 nmol/L insulin and 1 μmol/L rosiglitazone. Medium was changed every 2 days until plates were harvested on day 10 after induction and stained with oil red O to visualize adipogenesis. Oil red O staining was performed as previously described (22).

Cells (100,000) were plated in the upper chamber of semipermeable inserts for 24-well plates (6.5 mm Transwell with 8.0 μm Pore Polycarbonate Membrane Insert; Corning, cat. 3422) in RPMI 1640 medium containing l-glutamine, supplemented with 10% FBS. For the Invasion assay, 40 μL of Matrigel (Corning, cat. 354234) was solidified in a 37°C incubator for 15 to 30 minutes to form a thin gel layer. Cells (100,000) were seeded on top of the Matrigel coating.

Thirty minutes after seeding, growth medium in the upper chamber was carefully replaced by serum-free RPMI 1640 medium. After 20 hours of incubation at 37°C with 5% CO₂, cells on the upper side of the membrane were removed with a cotton-tipped applicator. Migrated cells were fixed in cold methanol and stained with crystal violet. The membranes were removed from the insert using a scalpel and mounted on a glass slide. Bright-field images were taken with 10X objectives on a Leica DM 5000B microscope. Images were analyzed using NIH Image J software package.

Cells (100,000) of each experimental condition were plated in 24-well plates and grown to confluence. These cells were treated overnight with 10 μg/mL Mitomycin C. A straight scratch was made using a P200 pipette tip. The cells were then washed with PBS 3 times, and initial gap sizes at T₀ were photographed. The cultures were further incubated for 8 hours (37°C and 5% CO₂) in RPMI 1640 medium containing l-glutamine, supplemented with 10% FBS, and photographed again (T₈). Gap areas were measured with NIH Image J software package, and results are expressed in percent wound closure = ratio: (gap area at T₀ – gap area at T₈)/gap area at T₀.

Subcutaneous xenograft experiments were performed using male 6- to 7-week-old BALB/c Nude Mouse (CAnN.Cg-Foxn1^nu/Crl), purchased from The Jackson Laboratory. 5637 WT/Ctl vector/or PPARG sgRNA 6 cells (1 × 10⁶), suspended in PBS, were mixed with 100 μL Matrigel, and a total of 200 μL was injected s.c. Tumor volume and body weight were measured 3 times a week. Tumor volume was calculated using the formula volume = [length × width)²]/2. Once tumors were established (∼100 mm³), animals were randomized into two groups and treated with vehicle (5% DMSO, 45% PEG300, and ddH₂O) or 5 mg/kg T0070907 daily through i.p. injection.

No inclusion or exclusion criteria parameters were used, and all animals were included in analyses. Animal well-being was monitored by certified veterinary staff. All mouse experiments were performed according to NIH guidelines and received ethical approval by the University of Pennsylvania's Institutional Animal Care and Use Committee.

Statistical analyses were performed in GraphPad Prism 8 using the Student two-tailed unpaired t test for pairwise comparisons, one-way ANOVA for multiple comparisons, or two-way ANOVA for multiple comparisons involving two independent variables. Data are presented as mean ± SEM of at least three independent experiments. A P value less than 0.05 was considered significant. Statistical significance was defined as *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; n.s., not significant.

Previous reports using available sequencing data from TCGA demonstrated that approximately 20% of UC tumors exhibit genomic PPARG amplification and/or increased PPARG mRNA levels, whereas approximately 15% display genomic amplification and/or activating mutations in RXRA, the obligate heterodimer of PPARγ, in a largely mutually exclusive pattern (5, 16). RNA-seq data from the Sanchez-Carbayo dataset also showed an increased number of both PPARG and RXRA transcripts in superficial UC (Fig. 1A and B). To validate that PPARG mRNA abundance translated to increased protein expression, we performed IHC for PPARγ on 59 independent samples of patients with UC from a commercially available tissue array (Fig. 1C–E). Surprisingly, our analysis revealed that nearly 37% of UC tissues stained strongly positive for nuclear PPARγ accumulation, whereas another 10% had more “intermediate” levels. In contrast, PPARγ levels in normal bladder epithelium were largely undetectable in these assays (Fig. 1C and D). Quantitative analysis revealed that overall PPARγ levels were significantly increased in UC samples compared with normal bladder epithelium (Fig. 1E). Interestingly, distinction between NMIBC and MIBC revealed that MIBC tissue sections have slightly higher PPARγ levels than NMIBC tissue sections (Supplementary Fig. S1A).

Multiple UC cell lines (5637, HT1197, Cal-29, HT-1375, TCCSUP, and T24) were employed to further evaluate the role of PPARγ in this article. Consistent with the patterns observed in primary patient samples, immunoblot and mRNA expression analysis showed increased but somewhat variable PPARG expression levels (Supplementary Fig. S1B and S1C). Moreover, analysis of Cancer Cell Line Encyclopedia (CCLE) data confirmed our experimentally observed PPARG expression patterns (Supplementary Fig. S1D). Across all tumor types, PPARG mRNA expression was significantly elevated in human cancer cell lines derived from the urinary tract (Fig. 1F). Genome expression studies identified two distinct molecular subtypes of bladder cancer, basal and luminal, with luminal tumors representing 75% of NMIBCs and 60% of MIBCs (5). Subtyping of these cell lines using basal markers KRT5/6 and luminal marker FOXA1 indicated that our analyzed UC cell lines are comprised of luminal, basal, and undetermined subtypes (Supplementary Fig. S1E). Although PPARG copy-number alterations are associated with luminal tumors, no clear correlation could be observed between PPARG expression levels and molecular subtypes. Overall, we concluded that PPARγ is frequently overexpressed in UC tumors compared with healthy bladder tissue, but can be detected at low, intermediate, and high levels in primary patient samples.

In agreement with previously published data (14, 15), treatment of multiple human UC cell lines (5637, HT-1197, TCCSUP, and Cal29) with the PPARγ inverse agonist, T0070907, decreased clonogenic growth over the course of 6 to 10 days, whereas clonogenic growth of two other cell lines (HT-1376 and T24) did not significantly change (Fig. 2A and B). We confirmed that treatment with T0070907 affected target gene expression by assessing the expression of three PPARγ-regulated genes, PLIN2, SCD, and FABP4, by qRT-PCR in 5637 cells (Supplementary Fig. S2A). The cell lines displaying the strongest decrease in proliferation, 5637 and HT-1197, are notable in that they are among the UC cell lines with highest PPARG expression levels (see Supplementary Fig. S1D) and that they display PPARG genomic amplification and an RXRα p.S427F activating mutation (16), respectively. We next performed flow cytometric analysis of Annexin V–PI staining to determine the effect of the PPARγ inverse agonist on cell viability. Long-term treatment of six UC cell lines with T0070907 did not decrease viability of cells relative to controls (Supplementary Fig. S2B), suggesting that reduced cell number following pharmacologic inhibition of PPARγ may result from cell-cycle arrest rather than induction of cell death (14). To test this, we performed PI staining on 5637 cells treated with T0070907 for 5 days to analyze the percentage of cells within each phase of the cell cycle. PPARγ inverse-agonist treatment significantly increased the percentage of cells in G₁ phase and reduced the G₂–M phase population relative to control (Fig. 2C; Supplementary Fig. S2C). In addition, we determined that T0070907 increased the percentage of cells in G₀ phase relative to control 4-fold (Fig. 2D) after 7 days of treatment.

Several reports demonstrated a regulatory role for PPARγ signaling in cell motility and adhesion in various cancer cell types including hepatocellular carcinoma cells (HepG2) and UC cells (28–30). Accordingly, T0070907 delayed closure of scratch wound migration assays (Fig. 2E) and impaired transwell migration and invasion (Fig. 2F; Supplementary Fig. S2D). Moreover, expression levels of MMP3 and MMP9, two matrix metalloproteases linked to tumor cell migration (31, 32), were decreased upon T0070907 treatment, whereas expression of the TIMP1 was slightly increased (Fig. 2G; Supplementary Fig. S2E). These results suggest that pharmacologic inhibition of PPARγ reduces cell-cycle progression in most of the tested UC cell lines and impairs cell migration.

To validate the effects observed by pharmacologic inhibition of PPARγ, we next depleted PPARγ protein from the panel of sensitive UC cell lines using two independent sgRNAs and the CRISPR-Cas9 gene editing system. PPARγ depletion via CRISPR also inhibited the proliferation of these UC cell lines in vitro (Fig. 3A–C), in a similar manner to that observed through pharmacologic PPARγ inhibition. In agreement, PPARG KO also resulted in a block in cell-cycle progression in the 5637 cell line (Fig. 3D; Supplementary Fig. S3A), increasing the percentage of cells in G₁ and lowering the G₂–M phase population relative to control KO. Furthermore, we employed an shRNA-resistant PPARG cDNA to validate the effects of PPARγ depletion on cell growth in UC (Supplementary Fig. S3B and S3C), and confirmed functionality of the shRNA-resistant PPARG cDNA by its ability to induce adipogenesis in NIH-3T3 cells when cultured in the presence of adipogenic stimuli (Supplementary Fig. S3D). Similar to results obtained using T0070907 treatment, PPARγ depletion impaired cell migration and invasion in all UC cell lines tested, as demonstrated by scratch wound assays (Fig. 3E), Boyden chamber migration/invasion assays (Fig. 3F; Supplementary Fig. S3E), and decreased expression of MMP3 and MMP9 while stimulating TIMP1 expression (Fig 3G; Supplementary Fig. S3F). These data confirm the sensitivity of a subset of urinary tract–derived tumor cell lines to PPARG depletion in vitro, as predicted from Achilles heel (Broad institute) shRNA and CRISPR screens (Fig. 3H; refs. 21, 33).

To investigate PPARγ occupancy genome-wide in UC, we performed ChIP-seq in the 5637 cell line for PPARγ, RXR, IgG, and Input controls (Fig. 4A, presequencing ChIP validation shown in Supplementary Fig. S4A). In addition, we confirmed PPARγ and RXR binding at several strong binding sites by ChIP-qPCR (Supplementary Fig. S4B–S4D). De novo motif analysis revealed that the majority of PPARγ–RXR binding in UC cells occurs at the direct repeat 1 (DR1) nuclear receptor motif (Fig. 4B), consistent with binding patterns in other cell types like adipocytes, macrophages (34), lung adenocarcinoma (35), and clear cell renal cell carcinoma (ccRCC; ref. 22). A previous report demonstrated that the transcription factors FOXA1, GATA3, and PPARγ are major molecular drivers of the luminal subtype of UC (36), but did not investigate whether these proteins work coordinately based on proximal DNA binding to promote a luminal versus basal state. Our data indicate that FOXA1, GATA3, and PPARγ likely regulate distinct genes in UC, based on the fact that neither FOXA1 nor GATA3 motifs were enriched at PPARγ–RXR binding sites (Fig. 4B). In agreement with this, the DR1 motif was not enriched under FOXA1-occupied regions in the luminal RT4 bladder cancer cell line (36). PPARγ–RXR heterodimers were predominantly located in intergenic and intronic regions of the genome at 46.08% of sites and 42.37% of sites, respectively (Fig. 4C). To elucidate the transcriptional programs regulated by PPARγ in UC, we annotated PPARγ–RXR binding sites to the nearest gene and performed GO for 2,581 genes: GO pathways involved in migration, adhesion, and regulation of transmembrane receptor tyrosine kinase signaling were most highly represented among this gene set (Fig. 4D).

To determine the effects of PPARγ inhibition on gene expression, we performed RNA-seq in the 5637 cell line with four experimental conditions involving either inverse-agonist treatment or gene knockout by two PPARG sgRNA. As the knockout of PPARG was incomplete at the protein level for sgRNA 9 (Fig. 4E), we proceeded to narrow our targets of interest based on overlap between inhibitor treatment and PPARG KO by sgRNA 6. Significantly altered genes by both PPARG CRISPR KO and inhibitor were enriched in pathways related to inflammatory responses (Fig. 4F), including leukocyte chemotaxis, response to TNF, and response to lipopolysaccharide. Interestingly, six chemokines whose expression increased following PPARγ inhibition (Fig. 4G) did not have PPARγ–RXR binding sites within 200 kb of the transcription start site (data not shown), suggesting that genomic binding of another transcription factor such as NF-κB mediates the inflammatory response. Of note, transrepression of NF-κB targets by PPARγ has been previously suggested (17) to explain the effects of PPARγ signaling in UC, yet a protein complex whereby PPARγ is tethered to chromatin-bound NF-κB is incompatible with the absence of PPARγ occupancy at inflammatory gene loci.

Although PPARG sgRNA 6 generated a list of 132 genes significantly downregulated at least 1.5-fold (P_adj < 0.05), T0070907 treatment generated a list of 342 genes significantly downregulated. We therefore predict that PPARγ targets of interest in regulating UC proliferation fall into a list of 30 protein-coding genes by both inhibitor treatment and PPARG KO by sgRNA 6 (Fig. 4H; Table 1). Of the 30 genes, 7 had a significant positive association with PPARG expression in RNA-seq data from 126 primary bladder UC samples (2, 18).

PPARγ–RXR heterodimers bind several locations upstream and downstream of SHH, one of the most downregulated genes among all groups in our dataset and a regulator of embryonic development and maintenance of tissue polarity (37, 38), within a 200 kb window (Fig. 5A and B). Of note, SHH-expressing cells have been previously shown to give rise to MIBC in a carcinogen-induced mouse model (39), but its connection to PPARγ signaling has not yet been described.

SHH exhibited a strong positive correlation with PPARG expression in primary patient samples according to TCGA data (Fig. 5C). Furthermore, silencing of PPARG reduced SHH expression in three UC cell lines (Fig. 5D and E), and treatment of 5637 cells with the PPARγ agonist rosiglitazone and the inverse agonist T0070907 significantly increased and decreased SHH gene expression, respectively (Supplementary Fig. S5A and S5B). Interestingly, SHH ablation using two SHH shRNAs, reducing SHH expression by approximately 80% (Supplementary Fig. S5C and S5D), decreased EdU incorporation in UC cells and cell growth over the course of 6 days (Fig. 5F; Supplementary Fig. S5E), thus phenocopying the proliferation defect observed upon PPARγ depletion. Previous studies have reported SHH to be involved in invasion and metastasis (40, 41). Accordingly, we observed decreased scratch wound closure and reduced transmembrane migration upon SHH inhibition (Fig. 5G and H). In addition, SHH inhibition impaired UC cell invasion capacities (Supplementary Fig. S5F).

To demonstrate the downstream effect of SHH on PPARγ-dependent cellular functions, we overexpressed SHH in PPARG KO cells (Fig. 5I). SHH overexpression upon PPARG inhibition resulted in a partial rescue in cell proliferation (Fig. 5J; Supplementary Fig. S5G) and migration/invasion (Fig. 5K and L; Supplementary Fig. S5H), demonstrating that the SHH functions downstream of PPARγ signaling and that PPARγ-dependent cell activities in UC cells are at least in part dependent on PPARγ-mediated SHH expression.

To assess the function of PPARγ in UC tumor growth in vivo, we implanted 5637 control and PPARG sgRNA 6 cells subcutaneously into opposing flanks of 6- to 7-week-old BALB/c nude Mice. We noticed a marked decrease in the growth of PPARG KO tumors relative to controls over a course of 28 days (Fig. 6A and B) with no reports of body weight loss (Fig. 6C). Importantly, expression analyses of tumor lysates showed that PPARγ depletion was sustained during the 28 days of tumor growth (Fig. 6D–F). In agreement with our in vitro data, PPARG KO tumors showed lower SHH expression levels than control tumors (Fig. 6D, G, and H), suggesting that PPARγ is required for UC proliferation and SHH expression in vivo. In addition, pharmacologic PPARγ inhibition as result of daily i.p. T0070907 injections also impaired UC tumor growth (Fig. 6I). No weight loss was reported during the treatment period (Fig. 6J), and tumor weight measured at day 28 after injection showed smaller tumors in T0070907-treated mice than control mice (Fig. 6K). As expected, T0070907 treatment reduced SHH expression levels in UC xenograft tumors (Fig. 6L–N). These data confirm our findings observed in in vitro experiments and demonstrate the importance of PPARγ signaling in SHH expression and UC tumor growth.

Active PPARγ–RXR signaling has been implicated as a molecular signature of luminal MIBC, yet the cell-autonomous mechanisms by which PPARγ contributes to UC growth are incompletely understood. In this study, we confirmed that multiple UC cell lines exhibit reduced growth upon treatment with PPARγ inverse agonist T0070907 in vitro as previously reported (15), and demonstrated that pharmacologic PPARγ inhibition results in reduced cell-cycle progression rather than induction of apoptosis. Moreover, T0070907 treatment impaired UC cell migration and invasion. The block in the G₁ phase of the cell cycle and the migration/invasion defect observed following treatment with inverse agonist were also replicated by PPARG genetic knockout via CRISPR-Cas9. Interestingly, patient MIBC tissue sections showed higher levels of PPARγ than NMIBC tissue sections, agreeing with our findings that UC cell migration and invasion is PPARγ dependent, which suggests that the invasive nature of MIBC tumors might be PPARγ dependent. To further elucidate the effects of genetic and pharmacologic inhibition of PPARγ signaling in UC, we performed ChIP- and RNA-seq to identify commonly deregulated pathways. Our findings corroborate recently published data showing an inverse correlation between PPARγ activity and immune infiltration in UC (17), and also reveal novel PPARγ–RXR transcriptional targets which may be responsible for promoting growth in UC.

Interestingly, lipid and glucose metabolism–related pathways were not among the most highly enriched pathways in our ChIP-seq dataset, suggesting that the PPARγ–RXR cistrome is distinct in UC relative to other cell types including adipocytes and ccRCC (22, 42). However, analyses of downregulated genes comparing control with either PPARγ inhibitor–treated or PPARG sgRNA 6–treated cells suggest that some “canonical” targets are affected in UC, as evidenced by repression of several targets within the KEGG hsa03320: PPAR signaling pathway (FABP4, PLIN2, LPL, or SCD) by both manipulations (Supplementary Fig. S5I). The absence of widespread alterations in gene expression involved in lipid and glucose metabolism pathways may be explained by the lack of C/EBP transcription factor co-occupancy with PPARγ in this cistrome. We hypothesize that the targets most important for UC proliferation are related to extracellular matrix organization (matrisome associated), adhesion, and cytokine-mediated signaling pathways (Fig. 4D and F). Future studies performing ChIP-seq in primary UC tissue would be beneficial to validate the genomic occupancy patterns observed in the 5637 cell line, and to further correlate genomic binding with PPARγ-responsive genes in vivo.

Based on our ChIP- and RNA-seq data, we propose SHH as a novel target gene of PPARγ in UC. Interactions between PPARγ and SHH have not been fully investigated. Several studies demonstrated that SHH signaling leads to PPARγ phosphorylation at Ser112 in adipocytes, and subsequent protein degradation (40, 43), but to our knowledge, no reports are available showing a regulatory role for PPARγ in SHH expression. Briefly, SHH is a signaling effector that binds extracellularly to the transmembrane receptor Patched (PTCH). The PTCH–SHH interaction inactivates the 7-pass transmembrane receptor Smoothened, which allows for the transcription factor GLI1 to translocate to the nucleus and regulate genes involved in proliferation and stem cell self-renewal (38). SHH expression has a weak, yet statistically significant positive association with PPARG expression in primary patient samples according to TCGA data, and treatment of the 5637 cell line with the PPARγ agonist rosiglitazone increased expression of this gene 4-fold relative to control. These gain- and loss-of-function experiments suggest for the first time that SHH is a direct target gene of PPARγ in in vitro cell culture models of UC (Table 1). The varied levels of natural endogenous ligands of PPARγ in vitro and in vivo certainly affect gene expression, and it is an important consideration when trying to identify targets likely to be most relevant to the human disease. Functionally, SHH, a gene known to be essential for growth and proliferation of cells during embryonic development in several tissue types such as neural tube, lungs, skin, hair, and teeth, has been implicated in normal bladder development and in the maintenance of bladder cancer stemness (44, 45). Moreover, our data demonstrate that genetic inhibition of SHH reduces cell proliferation to levels observed in PPARG-silenced UC cells. Genetic inhibition of SHH phenocopied the effect of PPARγ inhibition, resulting in impaired UC cell proliferation and migration/invasion. Importantly, SHH overexpression following PPARγ inhibition only partially rescued the PPARγ-dependent cell activity defects, suggesting that although SHH seems to be a major factor involved in PPARγ-dependent cell activities, additional factors likely contribute. SHH positivity is also associated with the luminal-papillary subclass of MIBC, the same as PPARγ (2). The role of the SHH pathway in UC is controversial, with reports indicating both protumorigenic (46) and antitumorigenic (47) roles in UC initiation and progression. Therefore, the relationship between PPARγ signaling, SHH expression, and UC proliferation and migration remains to be further clarified.

As stated above, it is unlikely that a single target gene would underlie PPARγ dependence, and other PPARγ targets need to be investigated to elucidate the complete PPARγ dependence in UC. Interestingly, in addition to SHH, our Chip-seq revealed Proline-rich 15 (PRR15) to be another putative PPARγ target with two PPARγ–RXR binding sites within a 200 kb window, which is also significantly downregulated following PPARγ inhibition in all three experimental conditions (data not shown). Although several observations link PRR15 to cell-cycle progression and tumorigenesis (48, 49), its role and expression patterns in UC have not been previously described, making the involvement of PRR15 in the PPARγ-mediated UC tumorigenesis an interesting factor for future projects.

Finally, we demonstrated that in vivo PPARγ inhibition in UC xenografts impaired tumorigenesis, suggesting that inhibition of PPARγ could be a potential therapeutic strategy for treatment of PPARγ-dependent bladder cancer.

In conclusion, our study supports a cell-autonomous, pro-oncogenic role for PPARγ signaling in luminal bladder UC and provides a candidate list of target genes responsible for these phenotypes from the intersection of in vitro genome-wide approaches and in vivo primary patient sequencing data.

No disclosures were reported.

D.J. Sanchez: Conceptualization, data curation, validation, investigation, methodology, writing–original draft. R. Missiaen: Conceptualization, data curation, validation, investigation, methodology, writing–original draft. N. Skuli: Data curation, investigation. D.J. Steger: Data curation, software, methodology. M.C. Simon: Conceptualization, supervision, writing–original draft.

The authors thank John Tobias for bioinformatics assistance on the ChIP-seq and RNA-seq datasets, and the Simon lab for helpful discussions. This work was supported by P32HD083185 and F31CA206381 (D.J. Sanchez), the Belgian American Educational Foundation (R. Missiaen), R35CA220483 (N. Skuli), R01 DK098542 (D.J. Sanchez), and R35CA220483 and NCIPO1CA104838 (M.C. Simon).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.