br Figure ARv Negatively Regulates H K ac and
Figure 5. ARv7 Negatively Regulates H3K27ac and FOXA1 Chromatin Binding
(D) Violin plots of the differentially expressed (adjusted p value <0.05) ARv7-target (blue) and ARfl-target (red) genes (within 10 kb of an AR site). The number of up-and downregulated genes and median values (black dot) are shown. The Fisher’s exact test between clusters has a p = 0.074 for shARv7 and p = 1 for shARfl.
(E) Histogram of the fraction of ChIP-seq peaks in clusters I and II from (A) that contain an AR-binding (left) or FOXA1-binding motif (right).
(F) Heatmaps (left) and bar graphs (right) for FOXA1 ChIP-seq in indicated cell lines and clusters, as in (A) to (C).
See also Figure S5 and Table S4.
(legend on next page)
a lesser extent ARfl) reprogram the FOXA1 cistrome. Since this phenotype is observed at both positively and negatively regu-lated ARv7 sites (i.e., clusters I and II), it is likely an important downstream effect of AR isoform depletion. In summary, these data suggest that the ARv7 repressive function stems from bind-ing NCOR corepressors, which results in negative regulation of H3K27ac.
ARv7 Represses Negative Regulators of Tumor Cell Proliferation
To better understand the clinical relevance of the ARv7 repressor function in PCa, we compared the LNCaP95-derived ARv7 gene signature with gene expression data from CRPC biopsies. The ARv7 gene signature was defined as shARv7-dysregulated genes (DEseq; p < 0.05) with an ARv7-binding site within 50 kb of their transcription start site; this resulted in 118 ARv7-acti-vated and 157 ARv7-repressed genes (Table S5). ARv7 status in patient biopsies was determined by immunohistochemistry (IHC) (Figures S6A–S6C) and classified into low or high ARv7 expression. However, only the highest and lowest quartiles, based on ARv7 IHC score, were considered for subsequent an-alyses. Significance analysis of microarrays (SAM) (Tusher et al., 2001) was used to define gene expression differences between patient groups (using unpaired, two-sample t tests) and controlled for multiple testing by q value estimation, using the false discovery rate (FDR) method. Genes were then ranked ac-cording to their t test score and compared with the cell line-derived ARv7 gene signatures using GSEA (Subramanian et al., 2005). We detected a significant negative distribution of the ARv7-repressed signature within the entire dataset (p = 0.003), but no significant correlation with the ARv7-activated signature (p = 0.193; Figures 6A and 6B; Table S5). This suggests that ARv7-dependent repression observed in 606-58-6 is also present in CRPC cases (with high levels of ARv7). As a control, we also per-formed a GSEA-based comparison between the ARfl gene signatures (shARfl-dysregulated genes within 50 kb of an ARfl-binding site; DEseq: p < 0.05; fold change >1.5) and the patient gene expression profiles. As expected, only ARfl-activated (n = 82; p = 0), but not ARfl-repressed (n = 55; p = 0.705), genes were significantly associated with gene expression in CRPC (Fig-ure 6A and Table S5). Taken together, these analyses suggest
that both ARfl-dependent gene activation and ARv7-dependent gene repression are prominent features of CRPC.
To elucidate the function of ARv7 repression, we focused on the subset of 57 genes at the core of the enrichment in our GSEA analysis (‘‘leading-edge’’; Figure S6D). As expected, ARv7 KD in the cells led to an upregulation of all ‘‘leading-edge’’ genes (Figure 6C), consistent with an ARv7-repressive function. To elucidate the function of these genes, we next compared them with positively selected genes from a genome-wide CRISPR knockout (KO) screen in LNCaP95 cells (MAGeCK; p < 0.05, Figure 6D and Table S6). We detected four ARv7-repressed genes with a negative effect on CRPC cell proliferation (SLC30A7, B4GALT1, HIF1A, and SNX14) (Figure 6D), indicating that these genes may have anti-tumor functions. However, no unequivocal correlation between the expression of the four genes and genetic aberrations in AR or PTEN, as visualized with cbioportal (Cerami et al., 2012; Kumar et al., 2016), could be identified (Figure S6E). In support of the potential clinical importance of this finding, we observed in the Taylor (Taylor et al., 2010) and expanded Decipher-GRID cohorts (Benzon et al., 2017; Boormans et al., 2013; Den et al., 2014; Erho et al., 2013; Karnes et al., 2013; Klein et al., 2015; Ross et al., 2016; Taylor et al., 2010) that PCa patients with low expres-sion of the four genes (Figures 6E and S6F) are at greater risk of disease recurrence than patients with high expression of all four genes (Figures 6F and S6G). Moreover, in the Taylor dataset, this finding was observed for patients with primary and metastatic disease (Figure S6G), and patients with primary disease only (Figure S6H). In addition, expression of the four genes was lower in metastatic than in primary disease (Figure S6I). Similarly, B4GALT1, SLC30A7, SNX14, and HIF1A expression was nega-tively correlated with metastasis development and PCa-specific mortality (Figures 6G and 6H). Taken together, these results sug-gest that ARv7 promotes CRPC progression by repressing genes that negatively regulate tumor growth, and are associated with poor PCa prognosis.