The androgen receptor cistrome is extensively reprogrammed in human prostate tumorigenesis

Abstract

Master transcription factors interact with DNA to establish cell type identity and to regulate gene expression in mammalian cells. The genome-wide map of these transcription factor binding sites has been termed the cistrome. Here we show that the androgen receptor (AR) cistrome undergoes extensive reprogramming during prostate epithelial transformation in man. Using human prostate tissue, we observed a core set of AR binding sites that are consistently reprogrammed in tumors. FOXA1 and HOXB13 colocalized at the reprogrammed AR binding sites in human tumor tissue. Introduction of FOXA1 and HOXB13 into an immortalized prostate cell line reprogrammed the AR cistrome to resemble that of a prostate tumor, functionally linking these specific factors to AR cistrome reprogramming. These findings offer mechanistic insights into a key set of events that drive normal prostate epithelium toward transformation and establish the centrality of epigenetic reprogramming in human prostate tumorigenesis.

Fig. 1. Genome-wide androgen receptor (AR) binding in normal prostate epithelium and tumor tissue

A. Representative 17 kilobase (kb) area. Each track depicts ChIP-seq AR binding intensity for a given sample. This locus demonstrates common AR binding sites as well as tissue-specific sites. B. Unique AR sites by tissue-type across all cohort samples and LNCaP prostate cancer cell line. C. Unsupervised pair-wise correlation of AR cistromes between all specimens as well as prostate cell lines. Cell lines include one LNCaP (LNCaP₁) and one LHSAR dataset generated for this analysis, and four publicly available datasets - three LNCaP^,, and one VCaP. Hierarchical clustering demonstrates relatedness of each AR cistrome. Black diagonal denotes the line of identity.

Fig. 2. Tissue-specific AR binding sites

A. Each horizontal line represents an AR site with significantly differential binding intensity across tissue-types (RPM, reads per million). Each column depicts binding intensity for an individual specimen (T-ARBS, tissue-specific AR binding sites; N-ARBS, normal prostate epithelium AR binding sites). B. Box and whisker plots depicting the median, 25^th–75th percentile and extremes of AR binding intensity across all specimens in tissue-specific sites. C. 3743 genes from the TCGA dataset with tissue-specific increased expression were ranked by fold-change and binned into groups of 300 (x-axis). The y-axis shows the average of normalized AR tags within 50kb of these genes.

Fig. 3. FOXA1 and HOXB13 co-localize with AR at tumor-specific AR sites

A. DNA binding motif analysis in subsets of the overall cohort - normal prostate tissue (blue circle), tumor tissue (red circle), and the T-ARBS (white dashed circle). B. At the set of LNCaP AR binding sites that overlap with the set of human T-ARBS, the HOXB13 binding motif emerges as statistically significant. The HOXB13 binding motif did not achieve significance when interrogating the entire LNCaP AR cistrome. At right, enrichment of HOXB13 motif across six conditions relative to background. Background is calculated as HOXB13 motifs in DNaseI hypersensitivity sites derived from 80 cell types included in the ENCODE project. C. HOXB13 and FOXA1 ChIP-seq in human prostate tumor specimens showed that these transcription factors (transcription factors) co-localize with T-ARBS. D. Representative core showing nuclear co-localization of AR, FOXA1 and HOXB13 protein by multi-plex immunohistochemistry staining in prostate cancer tissue: green, AR expression (1); orange, FOXA1 expression (2); red, HOXB13 expression (3); Multispectral image for AR, FoxA1 and HOXB13, with DAPI counterstained for nuclear masking, demonstrating AR, FOXA1 and HOXB13 nuclear co-localization in prostate cancer cells (4). Scale bars represent 100 μM.

Fig. 4. HOXB13 and FOXA1 are sufficient for reprogramming AR in LHSAR cells and are essential for prostate cancer cell survival

A. AR binding intensity in LHSAR cells transduced with HOXB13 and FOXA1, HOXB13 alone, FOXA1 alone or with LacZ at T-ARBS and N-ARBS. The joint effects of FOXA1 and HOXB13 recapitulate AR binding patterns in tumor. B. Cluster analysis of all human specimens and LHSAR-modified cell lines supervised by T-ARBS and N-ARBS. C. Top, dependency of LNCaP on HOXB13 and FOXA1 as measured by proliferation upon knockdown of HOXB13 and FOXA1 shRNAs. ShRNAs targeting GFP (shGFP) were used as negative controls. All knockdowns were performed in triplicate. Error bars represent 95% confidence intervals (at Day 6: shHOXB13 vs. controls, p=2.2×10^–7; shFOXA1 vs. controls, p=3.8×10⁻⁷). Knockdown of the transcription factors were confirmed by Western blot analysis (lower graphs). D. Dependency of 102 cell lines of multiple cancer types on HOXB13 (top) and FOXA1 (below). Lower ATARiS values represent increased dependency. The LNCaP cell line is in red. Data are part of the Broad Institute Project Achilles.