The cistrome and gene signature of androgen receptor splice variants in castration resistant prostate cancer cells

Abstract

Purpose: Spliced variant forms of androgen receptor were recently identified in castration resistant prostate cancer cell lines and clinical samples. We identified the cistrome and gene signature of androgen receptor splice variants in castration resistant prostate cancer cell lines and determined the clinical significance of androgen receptor splice variant regulated genes.

Materials and methods: The castration resistant prostate cancer cell line 22Rv1, which expresses full-length androgen receptor and androgen receptor splice variants endogenously, was used as the research model. We established 22Rv1-ARFL(-)/ARV(+) and 22Rv1-ARFL(-)/ARV(-) through RNA interference. Chromatin immunoprecipitation coupled with next generation sequencing and microarray techniques were used to identify the cistrome and gene expression profiles of androgen receptor splice variants in the absence of androgen.

Results: Androgen receptor splice variant binding sites were identified in 22Rv1-ARFL(-)/ARV(+). A gene set was regulated uniquely by androgen receptor splice variants but not by full-length androgen receptor in the absence of androgen. Integrated analysis revealed that some genes were directly modulated by androgen receptor splice variants. Unsupervised clustering analysis showed that the androgen receptor splice variant gene signature differentiated benign from malignant prostate tissue as well as localized prostate cancer from metastatic castration resistant prostate cancer specimens. Some genes that were modulated uniquely by androgen receptor splice variants also correlated with histological grade and biochemical failure.

Conclusions: Androgen receptor splice variants can bind to DNA independent of full-length androgen receptor in the absence of androgen and modulate a unique set of genes that is not regulated by full-length androgen receptor. The androgen receptor splice variant gene signature correlates with disease progression. It distinguishes primary cancer from castration resistant prostate cancer specimens and benign from malignant prostate specimens.

Keywords: alternative splicing; androgen; castration-resistant; disease progression; prostatic neoplasms; receptors; transcriptome.

Figure 1. ChIP-seq data analyses

(A) Qualitative comparison of AR cistrome in ARFL⁺/ARVs⁺ group and ARFL⁻/ARVs⁺ group with Venn diagram. (B) Quantitative comparison of AR binding in ARFL⁺/ARVs⁺ and ARFL⁻/ARVs⁺. The MACS called peaks from two groups were consolidated. Each dot represents a binding site. The signal intensities (reads coverage) of peaks were measured by wigsum. Black dashed line indicates diagonal line, red dashed line indicates linear regression line. (C) Examples of AR-binding events around KLK3 and PMEPA1 genes loci in ARFL+/ARVs+, ARFL−/ARVs+, and ARFL−/ARVs− groups. (D) ChIP-qPCR validation for KLK3 enhancer and PMEPA1 enhancer.

Figure 2. Microarray data analyses

(A) A Venn diagram indicates the number of probe sets differentially expressed in common or unique between shRNA conditions identified by microarray. (B) Heat map representing differential expression of AR-Vs unique upregulated genes (141 probe sets; left) and downregulated genes (156 probe sets; right) between shRNA conditions (fold change > 1.3; P<0.05). Each row on the heatmap represents a probe set; each column represents an individual sample. Gene expression on each probe set was standardized to the mean of samples where red color is higher than the mean and green color is lower than the mean. (C) Percentage of AR-Vs upregulated, downregulated and all genes within 100kb of AR-Vs bindings.

Figure 3

Unsupervised clustering of localized prostate cancer (LPC) and castration-resistant prostate cancer (CRPC) by AR-Vs gene signature. Unsupervised clustering using 297 probe sets (259 genes) AR-Vs gene signature separates LPC and CRPC. Tissue-type designation N, L and C refers to normal, localized prostate cancer and castration-resistant prostate cancer, respectively.

Figure 4

Unsupervised clustering of prostate cancer specimens by AR-Vs gene signature. Unsupervised clustering using 297 probe sets (231 genes) AR-Vs gene signature separates benign and malignant prostate specimens. Tissue-type designation N, B, P, C, L refers to normal epithelium, benign prostatic hyperplasia, prostatic intraepithelial neoplasia, localized prostate cancer and lymph node metastasis, respectively. Malignant specimens are marked as “Y” and benign samples are marked as “N”.

Figure 5. AR-Vs gene signature correlates with prostate cancer disease progression

(A) Differentially expressed AR-Vs regulated probe sets in normal epithelium, localized cancer, and metastatic lymph node (LN) metastases. Pairwise comparisons were made between Gleason Pattern (GP)3, GP4, GP5, and LN metastases and normal epithelium, respectively, using the limma package implemented in R to identify differentially expressed probe sets (one-way ANOVA; p<0.05). Expression patterns of probe sets that are common or unique in GP3, GP4, GP5, and LN metastatic specimens are summarized in a 4-way Venn diagram. A detailed description of probe set tissue distribution is listed in Supplementary Table S5. (B) Expression of AR-Vs gene signature correlates with Gleason pattern and metastasis. A Spearman rank linear regression model assesses the correlation of probe set expression with Gleason Pattern and presence of lymph node metastases. The y-axis represents a Log2 scale of relative probe set expression. The 11 most significantly positively correlated probe sets (p<0.05) are plotted in a stacked line graph (upper panel). The 11 most significantly negatively correlated probe sets (p<0.05) are plotted in a stacked line graph (lower panel). Probe sets that associate with disease progression are listed in Supplementary Table S6.

Figure 6

Expression of AR-Vs gene signature correlates with biochemical recurrence. Follow-up data was available for 56 patients, and univariate association was conducted for time to biochemical recurrence (p<0.05). Probe set expression of the four significantly associated genes was extracted, and the median expression was used to dichotomize patients into “high” (red) and “low” (green) for Kaplan-Meier plots. Log rank test was used to determine differences in time to biochemical recurrence.