Repression of Transcription Factor AP-2 Alpha by Peroxisome Proliferator Activated Receptor Gamma Reveals a Novel Transcriptional Circuit in basal-squamous Bladder Cancer

The discovery of bladder cancer transcriptional subtypes provides an opportunity to identify high risk patients, and tailor disease management. Recent studies suggest tumor heterogeneity contributes to “plasticity” of molecular subtype during progression and following treatment. Nonetheless, the transcriptional drivers of the aggressive basal-squamous subtype remain unidentified. As PPARγ has been repeatedly implicated in the luminal subtype of bladder cancer, we hypothesized inactivation of this transcriptional master regulator during progression results in increased expression of basal-squamous specific transcription factors (TFs) which act to drive aggressive behavior. We initiated a pharmacologic and RNA-seq-based screen to identify PPARγ-repressed, basal-squamous specific TFs. Hierarchical clustering of RNA-seq data following treatment of a panel of human bladder cancer cell lines with a PPARγ agonist identified a number of TFs regulated by PPARγ activation, several of which are implicated in urothelial and squamous differentiation. One PPARγ-repressed TF implicated in squamous differentiation identified is Transcription Factor Activating Protein 2 alpha (TFAP2A). We show TFAP2A and its paralog TFAP2C are overexpressed in basal-squamous bladder cancer and in squamous areas of cystectomy samples, and that overexpression is associated with increased lymph node metastasis and distant recurrence, respectively. Biochemical analysis confirmed the ability of PPARγ activation to repress TFAP2A, while PPARγ antagonist studies indicate the requirement of a functional receptor. In vivo tissue recombination studies show TFAP2A and TFAP2C promote tumor growth in line with the aggressive nature of basal-squamous bladder cancer. Our findings suggest PPARγ inactivation, as well as TFAP2A and TFAP2C overexpression cooperate with other TFs to promote the basal-squamous transition.


Introduction
While the most commonly diagnosed type of bladder cancer (BC) is morphologically defined as urothelial carcinoma, the existence of morphologic variants of BC and their association with clinical outcomes has been recognized for decades. Molecular studies show that variant morphologies in BC exhibit unique gene expression patterns, which may contribute to differing oncologic outcomes in these patients (1,2). Moreover, recent studies have identified a striking degree of intra-tumoral, morphologic and molecular heterogeneity in advanced BC (3). If advanced BC with intratumoral heterogeneity is largely clonal in nature, the fact that the vast majority of carcinoma in situ (considered the precursor for the majority of advanced BC) lesions are luminal (4) strongly suggests molecular subtype is "plastic" and can evolve over time. This perspective is substantiated by the fact that areas of variant morphology exhibit significant differences in gene expression subtype within a single tumor, yet harbor a large number of identical genetic alterations (5).
While the exact temporal sequence of genetic alterations in BC and how these alterations directly contribute to tumor heterogeneity is unknown, several lines of evidence implicate the steroid hormone receptor peroxisome proliferator active receptor gamma (PPARƔ) in morphologic and molecular plasticity.
For example, activation of this nuclear hormone receptor has been shown to oppose squamous differentiation (SqD) in vitro (6), while inactivation of both PPARƔ and PTEN expression drive squamous changes in vivo (7). Moreover, PPARƔ is amplified at the genomic level in the luminal BC subtype where it is consistently overexpressed (8)(9)(10), and activation of PPARƔ cooperates with overexpression of FOXA1 and GATA3 to "reprogram" the basal-squamous cell line 5637 to exhibit a luminal expression pattern (11). While these observations suggest PPARƔ is a master regulator of luminal BC cell fate, as well as a potential therapeutic target (8,(12)(13)(14), the transcriptional mediators of basal-squamous BC remain unidentified.
Accordingly, we hypothesized PPARƔ actively represses transcription factors (TFs) that drive basal-squamous gene expression in human BC, and by extension, inactivation of PPARƔ drives expansion of basal-squamous clones by upregulating these TFs. We tested the initial component of this hypothesis in the current study by performing a pharmacologic and RNA-seq based screen to identify PPARƔ-repressed TFs operative in driving the basal-squamous subtype. In doing so, we provide the first evidence identifying members of the Transcription Factor Activator Protein 2 (TFAP2) family as markers of basal-squamous BC that play a direct role in mediating the phenotype of this aggressive subtype of disease.

Cell culture
The UMUC1 cell line was purchased from the European Collection of Authenticated Culture Collection (Salisbury, UK). Additional lines were purchased from the American Type Culture Collection (ATCC; Manassas, VA). Cell origin was confirmed by short tandem repeat (STR) analysis (11). Culture

PPARɣ agonist and antagonist treatment
One day before transfection, 5637, UMUC1, SW780 BC cells were plated (4x10 5 cells/well) in 6 well plates (Corning Inc., Corning, NY) in complete culture medium containing 10% FBS and allowed to attach overnight. On the following day, culture medium were removed and attached cells were washed one time with serum-free medium (serum-free RPMI1640 medium for 5637 and SW780 cells, serum-free MEM/EBSS medium for UMUC1 cells) and then replaced with serum-free medium and incubated for an additional 24 hours. After 24 hours of serum starvation, medium was replaced with serum free medium containing either Dimethyl sulfoxide (DMSO; Sigma) vehicle control or rosiglitazone (TZD; 1μM; TOCRIS; Bristol UK). Cells were treated in the presence of DMSO or TZD for 48 additional hours, and RNA and protein were harvested via routine approaches as described below in the pertinent methods sections. For experiments utilizing the PPARɣ antagonist GW9662 (TOCRIS), cells were pretreated with GW9662 at a final concentration of 1 or 5 μM for 1 hour before the addition of TZD.

RNA-sequencing and computational analysis
cDNA libraries were prepared using the NEXTflex™ Illumina Rapid Directional RNA-Seq Library Prep Kit (Bio Scientific) as per the manufacturer's instructions. Denatured libraries were diluted to 10 pM by pre-chilled hybridization buffer and loaded onto a TruSeq SR v3 flow cells on an Illumina HiSeq 2500 for 50 cycles using a single-read recipe (TrueSeq SBS Kit v3) and run for 50 cycles using a singleread recipe according to the manufacturer's instructions. De-multiplexed and adapter-trimmed sequencing reads were generated using Illumina bcl2fastq (released version 2.18.0.12) allowing no mismatches in the index read. The sequencing reads were subjected to quality filtering used FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit) to keep only reads that have at least 80% of bases with a quality score of 20 or more (conducted by fastq_quality_filter function) and reads left with > 10 bases after being trimmed with reads with a quality score of < 20 (conducted by fastq_quality_trimmer function). Filtered reads were mapped to the human reference genome (GRCh38) using TopHat (version 2.0.9) (15) supplied by Ensembl annotation file; GRCh38.78.gtf. After normalization was performed via the median of the geometric means of fragment counts across all libraries, differential expression was determined using the Cuffdiff tool which is available in Cufflinks version 2.2.1 (16). All genes passing FDR criteria of 0.05 were considered differentially expressed genes. Venn diagrams and heatmaps were generated using limma in R. GO analysis was performed using DAVID Bioinformatics Resources 6.8 (https://david.ncifcrf.gov). For computational analysis of TCGA data for TFAP2A and TFAP2C expression, RNA-seq expression data was obtained from the Genomic Data Commons (https://portal.gdc.cancer.gov/). Expression was log2 normalized and clustered with the genes listed using the pheatmap package in R. pheatmap: Pretty Heatmaps. R package version 1.0.10 (https://CRAN.R-project.org/package=pheatmap) and correlation distance. Annotation data including expression subtype were obtained from the supplemental data available with the TCGA bladder cancer project (9).

RNA extraction and quantitative real time PCR (qRT-PCR)
Total RNA was extracted using the RNeasy approach (Qiagen; Hilden, Germany) according to manufacturer protocol. For cDNA synthesis, reverse transcription was performed using M-MLV reverse transcriptase (Thermo fisher) via manufacturer instructions. qRT-PCR was performed using QuantaStudio7 Real-Time PCR System (Applied Biosystems; Foster City, CA). Taqman probes used in this study were as follows. TFAP2A (Hs01029413_m1), TFAP2C (Hs00231476_m1), FABP4 (Hs01086177_m1). Relative gene expression change was calculated by deltadeltaCt method. 18S ribosomal RNA was used as an endogenous reference.

Generation of stable cell lines
One day before transfection, 4x10 6 Lenti-X 293T cells in 8ml of medium were seeded into 100 mm dishes. 7 µg of pLVX-IRES-TFAP2A or pLVX-EF1alpha-IRES-TFAP2C plasmid were diluted with 600 µl of sterile water and mixed with Lenti-X Packaging Single Shots (VSV-G) (Clontech). This mixture was added to medium in a 100 mm dish and incubated for 4 hours at 37 °C. After 4 hours, an additional 6 ml fresh medium was added and incubated for 48 hours. After 48 hours, medium was collected and virus titer was measured by Lenti-X GoStix TM (Clontech). Collected medium containing Lentivirus was filtered through a 0.45 µm polyethersulfone membrane (GE Healthcare, Chicago IL). For lentivirus infection, T24 or UMUC3 cells were seeded onto 6 well plates to reach approximately 70% confluency.
siRNA and lipofectamine 3000 were added to separate aliquots of OPTi-MEM medium, respectively and incubated for 5 minutes. Subsequently, siRNA and lipofectamine 3000 in OPTi-MEM were mixed and incubated for 20 minutes. After incubation, siRNA-lipofectamine 3000 complex was added to 2 ml cell suspension containing 4x10 5 target cells per well in a 6 well plate.

Immunohistochemistry of human bladder cancer tissue
Immunohistochemistry (IHC) on our previously described human BC TMA (17) was performed via established methods (18). Primary antibodies are referenced in Supplementary Table S1.
Nuclear expression was quantified by Allred score, which combines a measure of expression intensity (range 0 to 3, no expression to highest expression) to a measure of expression area (range 0 to 5, no expression to diffuse expression) to give a score ranging form 0 to 8 (19). in 20% methanol (Sigma) for 20 minutes, and residual cells that had not moved through the transwell were removed by gentle swabbing with a Q-tip (Unilever, Trumbull, CT). Cell numbers were counted in microscopic field (x100 magnification). All experiments were repeated at least 3 times in triplicate.

Tissue recombination xenografting
All animal experiments were performed in accordance with approved protocols from the Institutional Animal Care and Use Committee (IACUC) of Pennsylvania State University. Isolation of embryonic bladder mesenchyme (eBLM), preparation of tissue recombinants, and kidney capsule surgeries were performed as described previously (7,11,20 are expressed as the mean ± S.D. p<0.05 was considered as a statistically significant.

PPARɣ is a master regulator of luminal gene expression in bladder cancer cells
In a previous study (11), we reported that overexpression of FOXA1 and GATA3 cooperated with PPARɣ activation to "reprogram" a basal-squamous BC cell line (5637) as PPARɣ-responsive ( Figure 1B). These 3 cell lines were subsequently used for RNA-Seq studies to identify PPARɣ-regulated TFs (see Figure 1C for experimental design). While this approach identified a number of cell line-specific TZD-regulated genes following treatment, we additionally identified a total of 26 and 10 genes coordinately upregulated and downregulated amongst 5637, UMUC1 and SW780 following PPARɣ activation, respectively ( Figure 1D; see Supplementary Table S2)

PPARɣ signaling represses TFAP2A expression in bladder cancer cells
We next investigated the regulatory relationship between PPARɣ and TFAP2A. Treatment of serum-starved 5637, UMUC1 and SW780 with 1 µM TZD for 48 hours followed by Q-RT-PCR for FABP4 confirmed responsiveness of these lines to PPARɣ activation ( Figure 2A). In addition, Q-RT-PCR and western blotting confirmed the ability of PPARɣ activation to repress TFAP2A expression at the mRNA ( Figure 2B) and protein ( Figure 2C) levels, respectively. To confirm a role for a functional PPARɣ receptor, we performed identical TZD treatments of 5637, UMUC1 and SW780 alone and in conjunction with the PPARɣ antagonist GW9666. Q-RT-PCR results show that while TZD treatment significantly increased FABP4 expression and decreased TFAP2A expression, these significant changes were abolished in the presence of GW9662 co-treatment ( Figure 3A and 3B). In addition, while TZD treatment significantly reduced TFAP2A protein levels, this was prevented following co-treatment with GW9662 treatment ( Figure 3C). These results suggests that TZD-induced repression of TFAP2A mRNA and protein requires a functional PPARɣ receptor in BC cells.

TFAP2A and TFAP2C expression are markers of basal-squamous bladder cancer
basal-squamous BC is significantly enriched for SqD, and several studies have implicated TFAP2A and its paralog TFAP2C in development and differentiation of normal squamous epithelium (24)(25)(26)(27). Therefore, our identification of TFAP2A as a PPARɣ-repressed transcriptional regulator suggested a role for members of the TFAP2 family in basal-squamous BC. We therefore utilized qRT-     Figure 6C and D). Because UMUC3 cells express relatively low levels of TFAP2A, and no detectable TFAP2C (See Figure 4D), we additionally established UMUC3 cells stably overexpressing TFAP2A (UMUC3-TFAP2A) and TFAP2C (UMUC3-TFAP2C). Western blotting ( Figure 6E) and Q-RT-PCR ( Figure 6F) confirmed stable overexpression of TFAP2A and TFAP2C in UMUC3 at the mRNA and protein levels, respectively. Migration and invasion assays using these stable cells showed that increased TFAP2A or TFAP2C in UMUC3 cells significantly enhanced migration and invasion ( Figure 6G and H). Thus, these results suggest TFAP2A and TFAP2C regulate gene expression important for the control phenotypic aggressiveness associated with basalsquamous BC.

Overexpression of TFAP2A or TFAP2C in bladder cancer cells promotes tumorigenenicity following tissue recombination xenografting
We next utilized the tissue recombination system to investigate the impact of TFAP2A and TFAP2C overexpression on tumorigenicity and the ability to drive morphologic changes, such as SqD.
For these experiments, we utilized UMUC3 cells stably overexpressing TFAP2A or TFAP2C (see Figure 6E), as well as T24 cells stably overexpressing TFAP2A or TFAP2C (See Supplementary Figure   S3). These cells were chosen because of their relatively low expression of TFAP2A and TFAP2C (See However, we failed to detect SqD in any of our tissue recombinants. These observations identify TFAP2A and TFAP2C overexpression as drivers of tumorigenicity, but also suggest their individual overexpression is not sufficient to drive SqD.

Discussion
Although the presence of morphologic heterogeneity in BC and its relation to clinical outcomes has been recognized for decades, a series of recent studies definitively demonstrate the existence of intratumoral molecular heterogeneity in this common malignancy (3,5), as well as the potentially plastic nature of this heterogeneity. In keeping with specific tenants of the master regulator hypothesis (21), our in vivo and in vitro experimental studies (7,11,18,20,28), as well as additional foundational studies from other groups (6,9,10,22,(29)(30)(31)(32)(33)(34)(35)(36)(37) have identified a series of TFs apparently responsible for maintaining urothelial cell fate and establishing a luminal gene expression pattern in malignant disease.
Taken together, these studies suggest PPARɣ is a transcriptional master regulator of urothelial cell fate and the luminal gene expression pattern in BC. However, the positive drivers of the basal-squamous molecular subtype, as well as other molecular subtypes have remained largely unidentified.
Based on previous work suggesting PPARɣ activity is critical for maintaining a luminal gene expression program (10,11), we undertook a pharmacologic screen to identify PPARɣ-repressed transcription factors that act as potential master regulators of SqD and/or the basal-squamous molecular subtype (Figure 1). This screen identified TFAP2A as one such PPARɣ-repressed TF that is overexpressed in human basal-squamous disease. TFAP2A belongs to the transcription factor activator protein 2 (TFAP2) family, which consists of five members (TFAP2A, 2B, 2C, 2D, 2E) .
Structurally, these TFs contain a helix-span-helix domain important for homo/heterodimerization, which additionally cooperates with a basic domain for DNA binding. Furthermore, a proline and glutamine rich domain serves a trans activating function. In addition to TFAP2 family members being essential for neural crest development (38) and estrogen receptor binding and subsequent long-range chromatin interactions in breast cells (39), both TFAP2A and TFAP2C are implicated in keratinocyte differentiation (27,40,41) and squamous cancers (42) independent of anatomic site. At the molecular level in normal keratinocytes, TP63 (itself implicated in the basal subtype of BC (43)) activates TFAP2C expression to promote normal skin differentiation (25,44), and cooperates with TFAP2A and TFAP2C to regulate TP63 target gene expression (26). Also, Tfap2a and Tfap2c knockout in mice produces pathologic skin disease, potentially by impacting epidermal growth factor receptor signaling which is also implicated in basal BC (41). Therefore, TFAP2A and TFAP2C play a central role in squamous-specific gene expression.
Pharmacologic approaches in our current study indicate TFAP2A regulation following TZD treatment requires functional PPARɣ (Figures 2 and 3). Ligand-dependent increases in the expression of PPARƔ target genes involves release of corepressor complexes including Nuclear Receptor Corepressor 1 (NCoR1) and 2 (NCoR2/SMRT) and other factors including histone deacetylases (reviewed in (45)). This process results in the recruitment of general transcription machinery and increased gene expression. However, the mechanisms of ligand-dependent repression of PPARƔ target genes are less clear. The fact that TFAP2A is repressed by PPARƔ activation in UMUC1, SW780 and 5637, and our observation that co-treatment with a PPARƔ antagonists abrogates TZD-induced TFAP2A regulation in all 3 models suggests the existence of a general and shared mechanism.
However, the exact mechanism remains to be identified. While PPARƔ and other PPARs can directly interfere with the ability of other transcription factors to regulate gene expression via a process referred to as transrepression (45), it is not clear if this mechanism is responsible for ligand-dependent repression of TFAP2A by PPARƔ. As PPARɣ plays an important role in the BC cell autonomous regulation of cytokines and response to immunotherapy (8,46), further studies are required.
In addition to being highly expressed in in vitro models of basal-squamous BC (Figure 4), we additionally report here that both TFAP2A and TFAP2C are expressed at high levels in basal-squamous BC, as well as in areas of SqD ( Figure 5). Importantly, the fact that TFAP2A and TFAP2C promote in vitro surrogate measures of aggressive behavior typically associated with basal-squamous BC ( Figure   6), as well as in vivo tumorigenesis is in agreement with their association with lymph node metastasis and distant recurrence, and further suggests an important role for these factors in BC. In addition to the fact that SqD and the basal-squamous subtype has been associated with enhanced overall survival following neoadjuvant chemotherapy and cystectomy (47), TFAP2A was previously identified as an independent predictor of good response to cisplatin treatment in BC patients (48). While these observations further link TFAP2 family member expression to variant morphology in BC, we did not detect SqD in our tissue recombinants. However, this is not surprising as TFAP2 family members undoubtedly require the action of additional combinatorial factors (49,50) to mediate a squamous cell fate in vivo. Nonetheless, this study directly implicates TFAP2A and TFAP2C in the Basal molecular subtype of BC, as well as associated SqD. By additionally identifying TFAP2A as a PPARƔ-repressed gene, this study has also revealed a novel transcriptional circuit which may be involved in the plasticity of molecular subtype and morphologic differentiation in this common malignancy.