Constitutive enhancer activity of AcH3 peak sequences at 8q24. The DNA sequence containing each of the 15 identified AcH3 sites or a control sequence from the neighboring unacetylated region was inserted upstream of TK-luciferase reporter vector. The constructs were transfected into 5 different cell lines (LNCaP, PC3, HCT 115, COLO 205, and MCF7) along with pRL-TK Renilla luciferase plasmid for 24 h. Dual luciferase assays were conducted. The results were normalized against the internal Renilla control for each transfection. The luciferase activity of the control region was defined as 1. Relative luciferase activity values are presented as mean 6 SD of triplicate transfections. doi:10.1371/journal.pgen.1000597.g004 

Context in source publication

Context 1

... contrast, region 1 was totally devoid of transcripts in all tissues and cell lines. We did not investigate the transcripts originating in region 2 and 3 further, since their abundance in prostate tissue was not affected by risk haplotypes in the region [14]. In parallel, we generated high-resolution epigenomic profiles for the entire 5-Mb interval using ChIP-chip. For this purpose we hybridized ChIP material to the same tiling array used earlier for transcriptional profiling. Initially, we analyzed AcH3 in three cell lines representing prostate (LNCaP), breast (MCF7) and colorectal (HCT116) cancer. Because three regions independently impose prostate cancer risk, we also interrogated two prostate cancer cell lines (PC3 & LNCaP) more extensively for other key epigenetic marks at high resolution. Additional histone modifications chosen were the activation marks H3K4me1 & H3K4me3 [10], the transcription elongation mark H3K36me3 and the polycomb repressive mark H3K27me3. We also profiled RNA polymerase II (RNAPII) and patterns of androgen receptor (AR) occupied regions (ARORs). This entire multi-dimensional dataset (including cDNA profiles) was then subjected to extensive statistical analysis using spatial clustering, a new method that allows the dissection of large genomic regions into distinct clusters, each reflecting a specific combinatorial pattern of epigenetic marks in an unbiased manner [15,16]. Spatial clustering of the 5-Mb region surrounding and including the 8q24 risk loci is shown in Figure 2A and 2B. This unsupervised cluster analysis revealed domains of combinatorial histone modification and cDNA patterns, and determined the most likely type of behavior at each genomic locus. Six domain types were evident, color-coded and numbered I–VI (Figure 2A). The cancer risk regions are bordered by two distinct domains located 2-Mb apart: a 1-Mb type IV domain (located , 127 Mb), which is weakly enriched with H3K27me3 marks, and a type I domain- encompassing MYC (located , 129 Mb), which is strongly enriched with activation-associated marks and transcription (Figure 2B). The prostate cancer risk regions 1–3 were assigned to a type VI domain, indicating that the chromatin of the risk- linked domain is uniquely structured, and includes features that are distinctly different from the aforementioned flanking regions. Importantly, an additional LNCaP H3K27me3 domain (domain IV) is located downstream of MYC, with significant H3K27me3 enrichment limited to LNCaP (Figure S3). As H3K27me3 is a modification associated with polycomb-mediated repression, this suggests that in LNCaP the chromosomal architecture may group the MYC genes and the risk regions in between large repressed domains, possibly facilitating interactions between them. A higher-resolution epigenetic map of the risk regions in LNCaP is shown in Figure 3. As noted above, regions 1 and 3 were not robustly transcribed in either the normal tissues or prostate cancer cell lines. The histones in this region, however, were highly modified in LNCaP, with particular enrichment for active chromatin marks, i.e. AcH3, H3K4me1 and H3K4me3. Addi- tionally we observed significant occupancy of AR and RNAPII. Importantly, these patterns of activity were absent from PC3, which does not express the AR. The risk regions were also enriched for the elongation mark H3K36me3; however, in line with the general lack of transcription, the H3K36me3 areas were not polarized to a specific side of adjacent RNAPII peaks. Risk region 1 included, in addition, the three strongest H3K27me3 peaks in the 5-Mb region, suggesting that some polycomb dependent repression may affect region 1 activity in LNCaP cells. The epigenomic organization of the risk regions therefore reflects multiple hotspots of active chromatin, involving RNAPII, AR occupancy and activation as well as elongation marks, but without any detectable transcriptional footprints. Thus, these features may be understood as describing enhancers that regulate either dormant transcriptional units in cis or remote active transcriptional units in trans . We note that we could not rule out the possibility of small non-coding RNAs being transcribed from the region, since RNA species shorter than 200-bp were excluded from our preparation. In order to investigate the regulatory potential of the loci exhibiting active chromatin marks, we next performed a systematic series of heterologous enhancer assays, focusing initially on defined acetylation peaks contained within the cancer risk intervals (called AcP1 through AcP15, in Figure 3). We cloned approximately 1.5-kb DNA fragments, centered on AcPs from LNCaP, HCT116 or MCF7 cells, upstream of a luciferase reporter gene driven by the thymidine kinase ( TK ) minimal promoter. Enhancer activities of the fragments were determined by transient transfection and luciferase assays in LNCaP & PC3 (prostate cancer cells), HCT116 & COLO 205 (colorectal cancer cells) and MCF7 (breast cancer cells) (Figure 4). AcP6 (in the breast cancer risk region) and AcP10 (in prostate cancer/colorectal cancer risk region 3) had the most pronounced enhancer activities, whereas AcPs12 – 15 (in prostate cancer risk region 1) had activities that were lower, but clear compared to the negative control and several other AcPs. Interestingly, these active enhancers also displayed unmistakable H3K4me1 and H3K4me3 marks. The results suggest that some of the active chromatin foci we identified (Figure 3, right inset) have intrinsic enhancer activities within cellular contexts. This concept was further supported in a parallel study, in colorectal cells, which demonstrated that region 3, encompassing AcP10 and harboring SNP rs6983267, bound transcription factor T-cell factor 4 (TCF4) in an allele specific manner [17]. In the present study we did not study this region but rather analyzed region 1 further in prostate cancer cells. Risk region 1 is specifically linked to prostate cancer risk, and the three most robust acetylation peaks also exhibited strong AR binding (Figure 3). This region additionally exhibited both active marks (H3K4me1&3, found at active TSSs and enhancers) and inactive marks (H3K27me3, found throughout silenced genes and some intergenic regions), as well as occupancy of RNAPII. Further analysis of potential AR-mediated enhancers was strongly justified considering the major involvement of AR in all phases of prostate cancer development, including advanced ablation-resistant disease [18]. Consequently, we investigated the potential for androgen- dependent enhancer activities in this region. First, to verify and further characterize the AR binding at AcPs13, -14 & -15 as suggested by ChIP-chip (Figure 3), site-specific ChIP analyses were conducted using cells treated with dihydrotestosterone (DHT) or vehicle (Figure 5A). All three sites, in particular AcPs 14 & 15, revealed strong DHT-stimulated AR occupancy. Second, to test directly for androgen-dependent enhancer activities, we cloned narrower regions ( , 0.5-kb fragments) than the original AcP regions (which were , 1.5-kb in length), centered around the AR occupancy peaks in the same TK-luciferase reporter plasmid described above, named AROR13, -14 & -15, respectively. LNCaP cells, which express AR, were transfected with these plasmids and luciferase activity was measured. The results revealed robust DHT-dependent enhancer activity in AROR14 and -15, even higher than that of the PSA enhancer used as a positive control (Figure 5B), and this level of activity roughly correlated with the DHT-stimulated AR occupancies at the respective sites (Figure 5A). AROR-14 exhibited a remarkable basal activity but only a 3-fold response to DHT. In order to capture all common genetic variations in this region, we resequenced ARORs14 and -15 in prostate cancer cases of European ancestry (172 chromosomes). Through this effort we identified two SNPs in AROR15 that were strongly correlated with the risk variant rs10090154 (reported in [4]), which itself was not located within an AROR (rs11986220, r 2 = 1.0 and rs11988857, r 2 = 0.923; Figure 5C). We introduced all allelic combinations of both SNPs into the AROR15 reporter, creating 4 plasmids representative of the 4 alleles as shown in Figure 5D. In six independent experiments, using six independently constructed sets of plasmids, the DHT-dependent enhancer activity observed with the A-allele of rs11986220 was , 2-fold higher than the enhancer activity observed with the T allele, regardless of the SNP at rs11988857. Since the A allele at rs11986220 is also the allele associated with the risk allele for prostate cancer at rs10090154, these results suggest that the increased androgen-mediated activity of the enhancer may upregulate expression of an important oncogene in prostate epithelial cells. What are the mechanism(s) that govern the SNP effect on the DHT-mediated enhancer activity described above? Interestingly, the SNP at rs11986220 resides within a putative binding site for forkhead transcription factors, with the A allele better matching the consensus sequence (Figure S4). An interesting and relevant forkhead transcription factor is FoxA1, which has been implicated in augmenting responsiveness of some ARORs to androgens [12,19]. Although LNCaP cells are homozygous for the T allele at rs11986220, the physical presence of FoxA1 at the AROR15 enhancer was nevertheless demonstrated by site-specific ChIP analysis (Figure 5E). Importantly, this occupancy was enhanced by DHT treatment of the cells. In a competition electromobility shift assay (Figure 5F), an oligonucleotide centered around SNP rs11986220 competed better for FoxA1 binding to a consensus Fox oligonucleotide, when the SNP position was an A as compared to a T. Thus, the stronger DHT-responsiveness of the AROR15 enhancer observed with the A SNP at rs11986220 is attributable to higher affinity for the AR collaborator, FoxA1. Since the histone acetyl transferase and ...