ARID1A-deficient bladder cancer is dependent on PI3K signaling and sensitive to EZH2 and PI3K inhibitors

Metastatic urothelial carcinoma is generally incurable with current systemic therapies. Chromatin modifiers are frequently mutated in bladder cancer, with ARID1A-inactivating mutations present in about 20% of tumors. EZH2, a histone methyltransferase, acts as an oncogene that functionally opposes ARID1A. In addition, PI3K signaling is activated in more than 20% of bladder cancers. Using a combination of in vitro and in vivo data, including patient-derived xenografts, we show that ARID1A-mutant tumors were more sensitive to EZH2 inhibition than ARID1A WT tumors. Mechanistic studies revealed that (a) ARID1A deficiency results in a dependency on PI3K/AKT/mTOR signaling via upregulation of a noncanonical PI3K regulatory subunit, PIK3R3, and downregulation of MAPK signaling and (b) EZH2 inhibitor sensitivity is due to upregulation of PIK3IP1, a protein inhibitor of PI3K signaling. We show that PIK3IP1 inhibited PI3K signaling by inducing proteasomal degradation of PIK3R3. Furthermore, ARID1A-deficient bladder cancer was sensitive to combination therapies with EZH2 and PI3K inhibitors in a synergistic manner. Thus, our studies suggest that bladder cancers with ARID1A mutations can be treated with inhibitors of EZH2 and/or PI3K and revealed mechanistic insights into the role of noncanonical PI3K constituents in bladder cancer biology.

Cell Viability (%) Cell Viability (%) Cell Viability (%) E VMCUB-1 (ARID1Amut/empty vector)     Expression value from each sample is shown in addition to probe identifiers, log 2 fold change and t-test p-value. A 2 fold change was considered significant.

Supplementary Methodology
In silico data analysis Using cBioPortal, the mutation profiles of ARID1A and EZH2 in muscle-invasive and non-muscle invasive bladder cancer were obtained (1). cBioPortal provides user-friendly graphical interface to analyze whole exome sequencing datasets from The Cancer Genome Atlas Project and other published studies (2)(3)(4). To study EZH2 gene expression profile in invasive bladder carcinoma patients, TCGA level 3 RNA-seq data (including "raw_read_count" and "scaled_estimate" for each sample) was downloaded for all primary tumor and normal samples using TCGA-Assembler (5). Transcript per million values for each gene was obtained by multiplying scaled estimate by 1,000,000. Using patient ID from cBioPortal, primary tumors were categorized based on ARID1A mutation status. Boxplot was generated using R (https://cran.r-project.org/).  (6)(7)(8)(9)(10)(11). Differential expression analysis between ARID1A mutated/altered and control samples was performed for each study separately. Profile graphs, log2 fold change and p-value was obtained for probes related to PIK3R3.

Colony formation assay
Bladder cancer cells were seeded at 800 cells per well of 6-well plates (triplicate) and incubated at 37 °C with 5% CO2 for 7-10 days while treating with GSK126 or pictilisib at the doses indicated every two days. Here both untreated and DMSO treated cells served as controls. Colonies were fixed with 10% (v/v) ethanol for 30 min and stained with crystal violet (Sigma-Aldrich, St Louis, MO, USA) for 20 min. Then, the photographs of the colonies were taken using Amersham Imager 600RGB (GE Healthcare Life Sciences, Pittsburgh, PA, USA). Colony quantification was carried out using ImageQuant TL Colony v.8.1 software (GE Healthcare Life Sciences).

qRT-PCR analysis
RNA from cultured cells was extracted with Direct-zol RNA miniprep kit (Zymo Research). For qRT-PCR, cDNA was generated using a High-Capacity cDNA Reverse Transcription kit (Applied Biosystems). qRT-PCR analysis was performed using the Taqman Gene Expression Master reagent mixed with Taqman primers and analyzed with the QuantStudioTM 6K Flex Real-Time PCR System (Applied Biosystems). mRNA expression levels were normalized to human TATAbinding protein (TBP) or GAPDH, and the normalized cycle threshold (Ct) values were quantified using the double delta Ct analysis. qRT-PCR data represent relative expression. In general, controls in experimental data are normalized to a value of 1. Indicated Taqman primers were predesigned from Applied Biosystems as follows: PIK3R1 (Hs00933163_m1), PIK3R3 (Hs01103591_m1), GAPDH (Hs02786624_g1), TBP (Hs00427620_m1).

Protein extraction and western blotting of Xenograft tissues
The mice were euthanized, and the xenograft tumors were excised and weighed. Next, fragments of xenograft tumors were lysed in RIPA lysis buffer (Thermo Scientific, IL, USA) and boiled for 10 min at 90 °C. Protein concentrations were measured using the BCA assay. Approximately, 50 μg of protein extract from fresh surgical xenograft tissues were separated by 10% SDSpolyacrylamide gel electrophoresis (SDS-PAGE). The gels were then electrotransferred onto polyvinylidene difluoride (PVDF) membranes (EMD Millipore, Billerica, MA, USA).

CUT&RUN library preparation
RT112 cells were harvested using Accutase (Gibco) and slow frozen in aliquots containing 500,000 cells in serum-containing DMEM+10% DMSO. For CUT&RUN, cell aliquots were rapidly thawed at 37°C and washed as directed. Samples were processed in a single batch using the CUTANA CUT&RUN Kit (Epicypher) according to the manufacturer's instructions and using

CUT&RUN data processing and analysis
Sequencing reads were demultiplexed and converted to FASTQ format using bcl2fastq2 (Illumina). Adapter sequences were trimmed (cutadapt) and reads were aligned to both the human (GRCh38) and E. Coli K-12 MG1655 genomes (bwa-mem2). PCR duplicates were removed (Picard) and non-uniquely mapping reads were filtered out (samtools). Replicates were scaled to each other using the ratio of E. coli reads to total reads and scaled BigWig files were generated for data visualization (bedtools bamcoverage). Alignment files were converted to paired-end BED files and then full-fragment BED files (bedtools bamtobed, awk, cut) and scaled BedGraph files were generated using the same scaling strategy as for the BigWig files (deeptools genomecoverage). Peaks were called from the scaled BedGraph files using SEACR (12) in relaxed mode to facilitate broad discovery of potential peaks; the top 0.001% of peaks were retained for analysis. Replicate peaks were merged (bedtools merge), annotated (ChIPpeakAnno) for the closest TSS, and visualized in IGV. The CUT&RUN data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE203033.