Integrated analysis identifies a class of androgen-responsive genes regulated by short combinatorial long-range mechanism facilitated by CTCF

Abstract

Recently, much attention has been given to elucidate how long-range gene regulation comes into play and how histone modifications and distal transcription factor binding contribute toward this mechanism. Androgen receptor (AR), a key regulator of prostate cancer, has been shown to regulate its target genes via distal enhancers, leading to the hypothesis of global long-range gene regulation. However, despite numerous flows of newly generated data, the precise mechanism with respect to AR-mediated long-range gene regulation is still largely unknown. In this study, we carried out an integrated analysis combining several types of high-throughput data, including genome-wide distribution data of H3K4 di-methylation (H3K4me2), CCCTC binding factor (CTCF), AR and FoxA1 cistrome data as well as androgen-regulated gene expression data. We found that a subset of androgen-responsive genes was significantly enriched near AR/H3K4me2 overlapping regions and FoxA1 binding sites within the same CTCF block. Importantly, genes in this class were enriched in cancer-related pathways and were downregulated in clinical metastatic versus localized prostate cancer. Our results suggest a relatively short combinatorial long-range regulation mechanism facilitated by CTCF blocking. Under such a mechanism, H3K4me2, AR and FoxA1 within the same CTCF block combinatorially regulate a subset of distally located androgen-responsive genes involved in prostate carcinogenesis.

Figure 1.

(A) Cartoon illustration of various blocks. Bar: CTCF binding sites; rectangle: AR binding site; hexagon: FoxA1 binding site; line: H3K4me2 enrichment site. A genomic region with CTCF binding sites as borders is defined as a block. The first block from the left containing a RefSeq gene, AR binding site and H3K4me2 enrichment is both an AR block and a H3K4me2 block. The second block is referred to as FoxA1 block since it contains only FoxA1 binding site. The last block is both an AR-H3K4me2 block (AR binding site overlap with H3K4me2 enrichment) and a FoxA1 block. (B) Distribution of length of blocks in kb. In addition to the overall distribution, the figure also shows the distribution of blocks containing AR binding sites (AR blocks), blocks containing FoxA1 (FoxA1 blocks), blocks containing at least one responsive genes (responsive blocks), and blocks consist of only non-responsive genes (non-responsive genes) in LNCaP cell line 4 h after being induced by androgen.

Figure 2.

Genes with AR in the same block exhibit higher levels of gene expression fold change. (A) Illustrations of two types of blocks that are being compared in (B). AR-ARG block is defined as block with both AR binding site (purple box) and androgen-responsive gene (purple arrow). Nearby non-AR block is defined as block adjacent to AR-ARG blocks without AR binding site. (B) Expression fold-change of all genes in AR-ARG blocks (blocks with androgen-responsive gene and AR binding site) are significantly higher than genes in nearby non-AR blocks. (C) Heatmap showing the log2 fold-change of expression level of all genes in nearby non-AR blocks (black bar), genes in upAR-ARG blocks (blocks with upregulated androgen-responsive gene and AR, red bar) and genes in downAR-ARG blocks (blocks with downregulated androgen-responsive genes and AR, green bar). Color represents log₂ fold-change of expression level of genes after 4 h DHT versus 0 hr DHT (basal level). (D) Expression level of androgen-responsive genes in AR blocks are significantly higher than those in blocks without AR (non-AR blocks). (E) Silencing of CTCF decreases CTCF protein expression. LNCaP cells were transfected with siControl or siCTCF, and treated with vehicle or DHT for 4 h. Western blots were performed using antibodies indicated. (F) Silencing of CTCF decreases CTCF binding at the CTCF blocks with AR (regions 1–4) and without AR (regions 5–6). LNCaP cells were transfected with siControl or siCTCF, and stimulated with vehicle or DHT. ChIP assays were performed using antibodies against CTCF. (G) Silencing of CTCF significantly decreases expression fold changes of responsive genes in blocks with AR. siControl or siCTCF transfected LNCaP cells were stimulated with vehicle or DHT for 4 h. Total RNA was isolated and amplified with gene-specific primers.

Figure 3.

Genes in Class 1 are closer to a transcription factor AR and to FoxA1 binding sites compared to genes in Class 2. The same patterns are observed in both up- and down-regulated genes, although this distance preferential are more obvious for over-expressed genes.

Figure 4.

Comparison of genes in Class 1 and Class 2 in terms of expression level and block classification. (A) Genes in Class 1 exhibit higher change of expression level than those in Class 2. (B) 93% of genes in Class 1 have AR binding sites which overlap with H3K4me2 as well as having FoxA1 bindings in the same CTCF block. In contrast, only 22% of genes in Class 2 have AR overlap with H3K4me2 and FoxA1 in the same block.

Figure 5.

Genes in Class 1 are preferentially expressed in prostate cells compared to other tissues. Genes in Class 1 exhibit significantly higher expression level in prostate compared to other tissues (paired Wilcoxon test, P-value = 2.2 × 10⁻¹⁶). Genes in Class 2 also show significantly higher expression level in prostate cells versus the remaining tissues (paired Wilcoxon test, P-value = 2.15 × 10⁻⁸).

Figure 6.

Genes in Class 1 shows distinctive pattern in metastatic prostate cancer while Class 2 shows less distinctive markings. Color represents log₂ ratios of expression. When the expression value is not available, it is denoted by gray color.