The mechanisms underlying ETS-driven prostate cancer initiation and progression remain poorly understood due to a lack of model systems that recapitulate this phenotype. We generated a genetically engineered mouse with prostate-specific expression of the ETS factor, ETV4, at lower and higher protein dosages through mutation of its degron. Lower-level expression of ETV4 caused mild luminal cell expansion without histologic abnormalities and higher-level expression of stabilized ETV4 caused prostatic intraepithelial neoplasia (mPIN) with 100% penetrance within 1 week. Tumor progression was limited by p53-mediated senescence and Trp53 deletion cooperated with stabilized ETV4. The neoplastic cells expressed differentiation markers such as Nkx3.1 recapitulating luminal gene expression features of untreated human prostate cancer. Single-cell and bulk RNA-sequencing showed stabilized ETV4 induced a novel luminal-derived expression cluster with signatures of the cell cycle, senescence, and epithelial to mesenchymal transition. These data suggest that ETS overexpression alone, at sufficient dosage, can initiate prostate neoplasia.

Mouse prostate digestion: Intraperitoneal injection of tamoxifen was administered in 8-week-old mice. 2 weeks after tamoxifen treatment, the mouse prostate was digested 1 hour with Collagenase/Hyaluronidase (STEMCELL, #07912), and then 30 minutes with TrypLE^TM Express Enzyme (Thermo Fischer, # 12605028) at 37°C to isolate single prostate cells. The prostate cells were stained with PE/Cy7 conjugated anti-mouse CD326 (EpCAM) antibody (BioLegend, 118216) and then, CD326 and EYFP double positive cells were sorted out by flow cytometry, which are luminal cells mainly from the anterior prostate and dorsal prostate. The mRNA or genomic DNA were extracted from these double-positive cells and then were used for ATAC-sequencing and RNA-sequencing analysis.

ATAC-seq and primary data processing: ATAC-seq was performed as previously described. Primary data processing and peak calling were performed using ENCODE ATAC-seq pipeline (https://github.com/kundajelab/atac_dnase_pipelines). Briefly, paired-end reads were trimmed, filtered, and aligned against mm9 using Bowtie2. PCR duplicates and reads mapped to mitochondrial chromosome or repeated regions were removed. Mapped reads were shifted +4/-5 to correct for the Tn5 transposase insertion. Peak calling was performed using MACS2, with p-value < 0.01 as the cutoff. Reproducible peaks from two biological replicates were defined as peaks that overlapped by more than 50%. On average 25 million uniquely mapped pairs of reads were remained after filtering. The distribution of inserted fragment length shows a typical nucleosome banding pattern, and the TSS enrichment score (reads that are enriched around TSS against background) ranges between 28 and 33, suggesting the libraries have high quality and were able to capture the majority of regions of interest.

Differential peak accessibility: Reads aligned to peak regions were counted using R package GenomicAlignments_v1.12.2. Read count normalization and differential accessible peaks were called with DESeq2_v1.16.1 in R 3.4.1. Differential peaks were defined as peaks with adjusted p-value < 0.01 and |log2(FC)| > 2.  For visualization, coverage bigwig files were generated using bamCoverage command from deepTools2, normalizing using the size factor generated by DESeq2. The differential ATAC-seq peak density plot was generated with deepTools2, using regions that were significantly more or less accessible in ETV4^AAA samples relative to EYFP samples.

Motif analysis: Enriched motif was performed using MEME-ChIP 5.0.0 with differentially accessible regions in ETV4^AAA relative to EYFP. ATAC-seq footprinting was performed using TOBIAS. First, ACACCorrect was run to correct Tn5 bias, followed by ScoreBigwig to calculate footprint score, and finally BindDetect to generate differential footprint across regions.

RNA-seq analysis: The extracted RNA was processed for RNA-sequencing by the Integrated Genomics Core Facility at MSKCC.  The libraries were sequenced on an Illumina HiSeq-2500 platform with 51 bp paired-end reads to obtain a minimum yield of 40 million reads per sample. The sequenced data were aligned using STAR v2.3 with GRCm38.p6 as annotation. DESeq2_v1.16.1 was subsequently applied on read counts for normalization and the identification of differentially expressed genes between ETV4^AAA and EYFP groups, with an adjusted p-value < 0.05 as the threshold. Genes were ranked by sign(log2(FC)) * (-log(p-value)) as input for GSEA analysis using ‘Run GSEA Pre-ranked’ with 1000 permutations (48). The custom gene sets used in GSEA analysis are shown in Table S2.

Unsupervised hierarchical clustering: To get an overall sample clustering as part of QC, hierarchical clustering was performed using pheatmap_v1.0.10 package in R on normalized ATAC-seq or RNA-seq data. It was done using all peaks or all genes, with Spearman or Pearson correlation as the distance metric. To have an overview of the differential gene expression from the RNA-seq data, unsupervised clustering was also performed on a matrix with all samples as columns and scaled normalized read counts of differentially expressed genes between ETV4^AAA and EYFP as rows.

Integrative analysis of ATAC-seq, RNA-seq, and ChIP-seq data: ERG ChIP-seq peaks were called using MACS 2.1, with an FDR cutoff of q < 10^-3 and the removal of peaks mapped to blacklist regions. Reproducible peaks between two biological replicates were identified as ETV4^AAA ATAC-seq peaks. ERG ChIP-seq peaks and ETV4^AAA ATAC-seq peaks were considered as overlap if peak summits were within 250bp. To determine whether the overlap was significant, enrichment analysis was done using regioneR_v1.8.1 in R, which counted the number of overlapped peaks between a set of randomly selected regions in the genome (excluding blacklist regions) and the ERG-ChIP seq peaks or ETV4^AAA ATAC-seq peaks. A null distribution was formed using 1000 permutation tests to compute the p-value and z-score of the original evaluation.

To assign ATAC-seq peaks to genes, ChIPseeker_v1.12.1 in R was used. Each peak was unambiguously assigned to one gene with a TSS or 3’ end closest to that peak. Differential gene expression between ETV4^AAA and EYFP was evaluated using log2(FC) calculated by DESeq2. p-values were estimated with Wilcoxon rank t-test and Student t-test.

scRNA-sequencing: Tmprss2-CreER^T2, EYFP; Tmprss2-CreER^T2, ETV4^WT; Tmprss2-CreER^T2, ETV4^AAA; and Tmprss2-CreER^T2, ETV4^AAA; Trp53^L/L mice were euthanized 2 weeks or 4 months after tamoxifen treatment (n=3 mice for each genotype and time point). After euthanasia, the prostates were dissected out and minced with scalpel, and then processed for 1h digestion with collagenase/hyaluronidase (#07912, STEMCELL Technologies) and 30min digestion with TrypLE (#12605010, Gibco). Live single prostate cells were sorted out by flow cytometry as DAPI-. For each mouse, 5,000 cells were directly processed with 10X genomics Chromium Single Cell 3’ GEM, Library & Gel Bead Kit v3 according to manufacturer’s specifications. For each sample, 200 million reads were acquired on NovaSeq platform S4 flow cell.

Reads obtained from the 10x Genomics scRNAseq platform were mapped to mouse genome (mm9) using the Cell Ranger package (10X Genomics). True cells are distinguished from empty droplets using scCB2 package. The levels of mitochondrial reads and numbers of unique molecular identifiers (UMIs) were similar among the samples, which indicates that there were no systematic biases in the libraries from mice with different genotypes. Cells were removed if they expressed fewer than 600 unique genes, less than 1,500 total counts, more than 50,000 total counts, or greater than 20% mitochondrial reads. Genes detected in less than 10 cells and all mitochondrial genes were removed for subsequent analyses. Putative doublets were removed using the Doublet Detection package. The average gene detection in each cell type was similar among the samples. Combining samples in the entire cohort yielded a filtered count matrix of 48,926 cells by 19,854 genes, with a median of 6,944 counts and a median of 1,973 genes per cell, and a median of 2,039 cells per sample. The count matrix was then normalized to CPM (counts per million), and log₂(X+1) transformed for analysis of the combined dataset. The top 1000 highly variable genes were found using SCANPY (version 1.6.1) (77). Principal Component Analysis (PCA) was performed on the 1,000 most variable genes with the top 50 principal components (PCs) retained with 29% variance explained.

To visualize single cells of the global atlas, we used UMAP projections (https://arxiv.org/abs/1802.03426). We then performed Leiden clustering. Marker genes for each cluster were found with scanpy.tl.rank_genes_groups. Cell types were determined using the SCSA package, an automatic tool, based on a score annotation model combining differentially expressed genes (DEGs) and confidence levels of cell markers from both known and user-defined information. Heat-map were performed for single cells based on log-normalized and scaled expression values of marker genes curated from literature or identified as highly differentially expressed.

Differentially expressed genes between different clusters were found using MAST package, which were shown in heat-map. The logFC of MAST output was used for the ranked gene list in GSEA analysis (48). The custom gene sets used in GSEA analysis are shown in Table S2.

Gene imputation was performed using MAGIC (Markov affinity-based graph imputation of cells) package, and imputated gene expression were used in the heatmap.

Analysis of public human gene expression datasets: To analyze TP53 RNA expression in human prostate cancer samples, we obtained normalized RNA-seq data from prostate cancer TCGA (www.firebrowse.org) (3). To assess the role of TP53 loss on progression free and overall survival of ETS-positive prostate cancer, we used cBioportal. We queried MSKCC (queried on 10/12/2022), TCGA, and Stand-Up-To-Cancer cohorts, we selected patients with fusions of ERG, ETV1, ETV4, FL1, or TMPRSS2 and compared the overall survival between TP53 wild-type and TP53 mutant subsets.

Statistics: All statistical comparisons between the two groups were performed by Graphpad Prism 7.0 software using a two-tailed unpaired t-test.

Please see the README document ("README_Dataset-ETV4_mediates_dosage-dependent_prostate_tumor_initiation.txt") and the accompanying published article: Dan Li, Yu Zhan, Naitao Wang, Fanying Tang, Cindy J. Lee, Gabriella Bayshtok, Amanda R. Moore, Elissa W.P. Wong, Mohini R. Pachai, Yuanyuan Xie, Jessica Sher, Jimmy L. Zhao, Makhzuna Khudoynazarova, Anuradha Gopalan, Joseph Chan, Ekta Khurana, Peter Shepherd, Nora M. Navone, Ping Chi, Yu Chen, ETV4 mediates dosage-dependent prostate tumor initiation and cooperates with p53 loss to generate prostate cancer. Science Advances 2023. Accepted. DOI: 10.5061/dryad.v41ns1s0s