Prostate cancer risk locus at 8q24 as a regulatory hub by physical interactions with multiple genomic loci across the genome

Abstract

Chromosome 8q24 locus contains regulatory variants that modulate genetic risk to various cancers including prostate cancer (PC). However, the biological mechanism underlying this regulation is not well understood. Here, we developed a chromosome conformation capture (3C)-based multi-target sequencing technology and systematically examined three PC risk regions at the 8q24 locus and their potential regulatory targets across human genome in six cell lines. We observed frequent physical contacts of this risk locus with multiple genomic regions, in particular, inter-chromosomal interaction with CD96 at 3q13 and intra-chromosomal interaction with MYC at 8q24. We identified at least five interaction hot spots within the predicted functional regulatory elements at the 8q24 risk locus. We also found intra-chromosomal interaction genes PVT1, FAM84B and GSDMC and inter-chromosomal interaction gene CXorf36 in most of the six cell lines. Other gene regions appeared to be cell line-specific, such as RRP12 in LNCaP, USP14 in DU-145 and SMIN3 in lymphoblastoid cell line. We further found that the 8q24 functional domains more likely interacted with genomic regions containing genes enriched in critical pathways such as Wnt signaling and promoter motifs such as E2F1 and TCF3. This result suggests that the risk locus may function as a regulatory hub by physical interactions with multiple genes important for prostate carcinogenesis. Further understanding genetic effect and biological mechanism of these chromatin interactions will shed light on the newly discovered regulatory role of the risk locus in PC etiology and progression.

INTRODUCTION

Chromosome 8q24 contains a locus conferring an increased risk for multiple cancers including prostate cancer (PC) (1–8). The variants at this locus spread across three genomic regions (four blocks), contributing independently to PC risk (1,2,8–10) (Fig. 1). These three regions lie in a gene desert with only a few predicted non-coding genes. The closest annotated protein-coding gene, proto-oncogene MYC, is over 200 kb downstream from the nearest PC risk variant. Multiple lines of evidence indicate that the 8q24 risk locus exhibits minimal RNA transcriptional output (11,12) and contains regulatory elements, especially enhancers (13,14). Therefore, the risk alleles may increase disease risk by affecting distant target genes. By applying chromosome conformation capture (3C) technology, previous studies show that the PC risk regions physically interact with MYC in a colorectal cancer cell line, and MYC or PVT1 in PC cell lines (14–16), suggesting the potential role of these 8q24 variants in the regulation of MYC or PVT1 expression. However, due to technical limitations, these studies investigated the two pre-defined genomic regions only, leaving a significant knowledge gap in whether the risk locus has any significant effects on other genomic regions.

Figure 1.

Overview of 8q24 risk locus. (A) Risk locus boundaries and nearby genes at 8q24. (B) Three PC-related LD blocks and risk SNPs. LD is based on r², using the HapMap CEU data. The three blocks refer to the blocks analyzed in this paper. (C) EcoRI sites and PC-risk regions. E represents EcoRI, number represents different cutting site.

3C was originally developed to study the conformation of a complete chromosome in yeast (17) and was subsequently adapted to investigate the folding of complex gene loci in mammalian cells (18). It has become an invaluable tool for studying the relationship between nuclear organization and gene expression (19,20). The basic principle of the 3C assay is to determine the contact frequency between any two loci in the genome (21). Based on this, other technologies have been developed to increase throughput, including 4C (22,23), 5C (24), ChIA-PET (25), Hi-C (26) and tethered 3C (27). However, current 3C and its derivatives are not sufficient to simultaneously examine multiple risk regions and their unknown target regions with high resolution. For example, 4C (22,23), a 3C derivative, is able to identify unknown target regions only when a small genomic region containing specific regulatory elements is pre-defined. Hi-C (26), another 3C derivative, is able to reveal global chromatin interactions but lacks the needed resolution. For a given risk locus where regulatory domains and target genes are often unknown, it is necessary to examine the entire risk linkage disequilibrium (LD) block at high resolution for its distant targets across the genome.

To address these technical barriers and identify genomic regions that may interact with the 8q24 risk locus, we developed a 3C-based multi-target sequencing (3C-MTS) technology by integrating target capture sequencing into the 3C assay. Unlike previously reported 3C-based assays, this new approach allows for a high-resolution survey of the whole genome for potential interactions with multiple regions of interest simultaneously. In this study, we applied this new technology to test five prostate-derived cell lines and one lymphoblastoid cell line (LCL) for any genomic regions that may interact with the PC risk locus. Our results demonstrated convincing evidence showing multiple cis- and trans-effects of the 8q24 risk locus on other gene regions, a step closer to functionally characterizing the role of the most common risk locus in cancer development.

RESULTS

Genome-wide survey of chromatin interactions with 8q24 risk locus

To obtain a global view of the chromatin interactions with the 8q24 risk locus, we developed 3C-MTS technology, a new genome-wide adaptation of the 4C assay by incorporating two well-established technologies—multi-target capture sequencing and 3C assay. An outline of the 3C-MTS procedure is given in Figure 2. For each 3C library preparation, we tested EcoRI digestion efficiency and library quality by agarose gel electrophoresis (Supplementary Material, Fig. S1) and quantitative polymerase chain reaction (qPCR) (Supplementary Material, Table S1 for primers). Overall digestion efficiency was ∼80% when compared with the undigested control DNA for each corresponding cell line (Supplementary Material, Fig. S1). To test the target enrichment efficiency of the final 3C-MTS libraries, we performed qPCR using primers on the 8q24 risk locus (near EcoRI sites E12, E28, E29 and E60, see ‘Materials and Methods’ section) as target fragments, and primers on other chromosomes (2q31, 7p22, 12q13, and 16p11) as non-target fragments (Supplementary Material, Table S2). When compared with the pre-captured 3C-sequenicng libraries, the final 3C-MTS libraries showed ∼2300-fold (ranged from 1500 to 3400) increase of target DNA. For the non-target regions, however, the fold change was negligible with an average of ∼0.56, indicating the successful enrichment of target DNA (Supplementary Material, Fig. S2).

Figure 2.

Outline of the 3C-MTS technology. (A) 3C-MTS overview. Fragmented and size-selected 3C library fragments are ligated with sequencing adaptors. Prepared 3C sequencing library is denatured and hybridized with biotin-labeled target probes. The resulting probe-3C DNA hybrids are pulled down by streptavidin-beads. The target-enriched DNA fragments are amplified by additional PCR cycles and submitted for paired-end sequencing. The colored lines represent different genomic fragments. The vertical black bars represent EcoRI cutting sites. The lines with two different colors and a vertical black bar in the middle are hybrid fragments from different genomic locations, including intra- and inter-chromosome interactions. The yellow and blue lines represent target fragments. The purple and red lines represent unknown fragments. The blue dot is biotin, and the gray disk is streptavidin-bead. (B) Schematic representation of the location of the target enrichment probes. (C) Main steps of 3C-MTS data analysis.

To map the sequencing data, we first trimmed adaptor and low quality sequences and then submitted ∼86 (50–132) million read pairs per library for sequence alignments. Mappable read pairs accounted for ∼97% of all submitted sequences. Among these, ∼17.82% (∼14.71 million paired reads/library) was mapped to target regions of the 8q24 risk locus including chr8:128077370-128138344 for risk region 2, and chr8:128310265-128551675 for risk regions 3 and 1 (Fig. 1). These on-target sequences were defined as paired sequence reads with at least one end mapped to ≤250 bp to the nearest EcoRI site at 8q24 risk regions of interest. From these on-target sequences, we further removed these pairs with at least one end mapped to multiple genomic regions, which accounted for ∼37.07% of all on-target pairs (∼5.45 million read pairs/library). The remaining 62.93% (9.25 million pairs/library) was the sequence pairs with both reads mapped to unique genomic regions. Among the read pairs with unique mapping, 68.07% was the ligations within target regions or ±25 kb from target regions, including same fragment ligations, adjacent ligations and any other ligations within target boundaries ±25 kb. For the purpose of this study, we further excluded the ligations listed above, which were predominant in all tested 3C-MTS libraries. Finally, ∼2.95 million read pairs (long-range interactions) were left in each library, accounting for 31.93% of uniquely mapped on-target sequence pairs with one end on non-target regions. Specifically, we defined the long-range interaction as any sequence pair with one end (for example, R1) on the 8q24 risk locus and the other end (for example, R2) either on chr8 that was ≥25 kb apart from the pre-defined 8q24 risk locus boundaries or on different chromosomes. The sequence mapping data for the six cell lines is summarized in Table 1.

Table 1.

Summary of 3C-MTS mapping results in the six cell lines

Paired sequence categories	PC-3	LNCAP	DU-145	RWPE-1	BPH-1	LCL	Average
Unmapped sequence pairs	885 187	1 532 801	677 738	819 090	721 831	9 519 853	2 359 417
Mapped sequence pairs	85 473 966	109 431 597	48 888 354	70 678 998	64 478 445	122 518 203	83 578 261
Outside target regiona	62 892 939	97 164 239	39 725 876	57 619 887	52 451 193	103 385 995	68 873 355
On target regionsa	22 581 027	12 267 358	9 162 478	13 059 111	12 027 252	19 132 208	14 704 906
(% on targets)	26.15	11.06	18.49	18.27	18.45	14.49	17.82
At least one read end mapped to multiple regions	4 507 724	8 580 136	3 044 570	4 008 954	3 451 289	9 116 589	5 451 544
Both paired ends mapped to unique regions	18 073 303	3 687 222	6 117 908	9 050 157	8 575 963	10 015 619	9 253 363
Both ends on target regions or one end on target region and other end near target region (±25 kb)	12 427 690	2 780 000	4 052 713	5 541 854	6 611 604	6 380 388	6 299 042
(% of within/nearby ligations)	68.76	75.40	66.24	61.23	77.09	63.70	68.07
One end on target, the other end mapped to one unique non-target regionb	5 645 613	907 222	2 065 195	3 508 303	1 964 359	3 635 231	2 954 321
(% of long-range interactions)	31.24	24.60	33.76	38.77	22.91	36.30	31.93
Total number of sequence reads (in pairs)	86 359 153	110 964 398	49 566 092	71 498 088	65 200 276	132 038 056	85 937 677

^aTarget region: sequences mapped to ≤250 bp to nearest EcoRI site at chr8:128069997-128140428 or 128306795-128551795.

^bLong-range interaction: one end mapped to target region and the other end either on chr8, that is ≥25 kb from the pre-defined 8q24 risk locus boundaries or on different chromosomes.

From the long-range interactions, we further filtered out all duplicated reads (both ends had identical sequences) to avoid polymerase chain reaction (PCR)-related bias during sequencing library preparation. To correct for potential differences in probe capture efficiency, we first calculated normalization ratios as the sum of captured sequence reads for each probe divided by the sum of captured reads for a reference probe (E67D). We then normalized the filtered unique long interaction counts for each probe to the ratio. To be comparable among different cell lines, we scaled the normalized interaction counts to a reference cell line (PC-3). Finally, we binned the normalized read counts into 10 kb windows and plotted the counts across the human genome (Fig. 3). This genome-wide plot showed frequent long-range interactions of 8q24 locus with numerous chromosomal regions. Due to pre-exclusion of ligations within and ±25 kb from target regions (∼6.30 million sequence pairs/library, Table 1), the expected high frequency of ligation counts at 8q24 locus was dramatically reduced.

Figure 3.

Genome view of frequent long-range interaction plot between 8q24 target regions and other genomic loci. The interactions within and ±25 kb from 8q24 target region 2: 128077370-128138344, region 3–region 1: 128310265 128551675 were excluded. The Y-axis shows counts of normalized unique interactions. The X-axis represents 23 different chromosomes.

Overall, we observed a significant fraction of long-range interactions around the 8q24 or in centromere regions (1p11.2, 6p11.1, 7q11.21, 10q11.21, 18q11.1 and 19q11). Surprisingly, the strongest long-range interaction was not at intra-chromosomal locus 8q24 (after exclusion of ligations within and near target regions) but at inter-chromosomal locus 3q13. This interaction was common in all cell lines tested, but the three cancer cell lines (DU-145, LNCaP and PC-3) showed 5–10-fold higher interaction frequencies than non-malignant cell lines (BPH-1, RWPE-1 and LCL). The second most common interaction was at the MYC locus and demonstrated a strong cell line-dependent pattern. Other common interactions are listed in Supplementary Material, Table S3 for each cell line. Clearly, some gene regions were common among these cell lines such as intra-chromosomal interaction genes (MYC, PVT1, FAM84B and GSDMC) and inter-chromosomal interaction genes (CD96 and CXorf36). Other gene regions were cell line-specific such as RRP12 in LNCaP, USP14 in DU-145 and SMIN3 in LCL.

To evaluate the reproducibility of the probe capture-based 3C assay, we made an independent 3C-MTS library using LNCaP cell line (LNCaP-R) and obtained 201 304 long-range interactions from a total of 7.89 million on-target reads. Due to relatively low long-range read counts in the LNCaP-R library, we performed Pearson correlation coefficient analysis in the dataset with at least one read in the repeat library and observed significant correlation (r = 0.89) between original and the repeat library (Supplementary Material, Table S4). This correlation was further enhanced if considering the dataset with at least three read counts in the repeat library (r = 0.91). To examine if intra-chromosomal ligations caused the significant correlation, we performed the same analysis after excluding all interactions mapped to Chr 8. The correlations were still significant with r = 0.75 in ≥1 read dataset and 0.82 in ≥3 read dataset. In all four scenarios, the LNCaP-R always showed higher correlation with LNCaP than any other cell lines, indicating robust reproducibility of the 3C-MTS. Meanwhile, we also observed significant correlation among the six different cell lines (Supplementary Material, Table S4).

Frequent physical contacts between 8q24 and 3q13

To fine map the most common 8q24-3q13 interaction, we examined the normalized interaction frequency from each pair of fragments between the two loci. Among the fragments defined by 77 EcoRI sites across three PC risk regions at 8q24, we found a cluster of fragments between sites E59 and E62 in PC risk region 1 contributing to the observed interaction signal (Fig. 4A). For 3q13, the interaction signals originated from three fragments defined by two EcoRI sites only, 3L at chr3:111274268 and 3R at chr3:111274323 (55 bp apart from each other). Surprisingly, the two 3q13 cut sites showed significant position-dependent interactions with different 8q24 cut sites of risk region 1. For example, in the LNCaP cell line, the 3R fragment (right side) at 3q13 was frequently interacted with 8q24 fragments between E59 and E61. The 3L fragment (left side) at 3q13, however, was highly interacted with 8q24 fragments between E61 and E62 (Fig. 5). This position-dependent interaction was observed among all cell lines tested (Supplementary Material, Figs. S3 and S4).

Figure 4.

Interaction of 8q24 risk locus with other genomic loci. (A) Heatmap between risk regions at 8q24 and both chr 3 and 8 in PC-3 cell line. The Y-axis is the risk regions defined by 77 EcoRI sites (96 probes). The X-axis is all EcoRI-defined fragments on chr 3 and 8. (B) Detail heatmap between risk regions at 8q24 and CD96 region at 3q13. The X-axis is CD96 region defined by 35 EcoRI sites. The dotted circle indicates an interaction hot spot between 3q13 fragments (3L and 3R) and 8q24 fragments (E59–62). (C) Detail heatmap between risk regions and MYC region. The X-axis is MYC and part of PVT1 region defined by 73 EcoRI sites. Multiple interaction hot spots are shown in dotted circles. Under the heatmap are positions of MYC's predicted promoter (red bar), insulator (blue bar), repressor (gray bar) and enhancer (yellow bar). (D) Heatmap of ICP index in six cell lines. Fragments with low ICP values (in blue) coincide with the fragments with active interactions in B and C. (E) Predicted functional domains in PC risk regions. Black, yellow and blue bars indicate low ICP, predicted enhancer and insulator regions, respectively.

Figure 5.

Long-range interaction between 8q24 E59–E62 and 3q13 CD96 intron 2 in the LNCaP cell line. (A) Physical map of EcoRI sites, capture probes and 3C-qPCR primers. The anchor primers 3L and 3R are close to two EcoRI sites on chr.3. The test primers on chr.8 are close to EcoRI sites from E56 to E65 (position 128511051-128549431). Each EcoRI site has black/grey bars and arrows on either side. The bars represent the location of capture probes, and the arrows show positions of the qPCR primers. Black and gray represent upstream and downstream from each EcoRI site, respectively. (B) Sequencing-based interaction read counts between 3q13 and 8q24. The Y-axis shows the normalized unique read counts. The X-axis shows the positions of the EcoRI cutting sites. (C) 3C qPCR-based interaction signal between anchor primers and test primers. Error bars represent standard deviation from the means of at least two independent experiments. In both (B) and (C), the graph representing interaction with 3R is in black; with 3L is in gray. U and D represent upstream and downstream of each corresponding EcoRI site.

To validate the most common interaction, we performed a 3C-qPCR assay using two anchor primers close to the 3L/3R sites at 3q13 and 19 test primers at 8q24 covering E56–E65 for a total of 2 × 19 possible ligation products. The results showed that anchor primer 3R generated higher interaction signals between E59 and E61. The anchor primer 3L, however, showed higher interactions between E61 and E62 in the LNCaP cell line (Fig. 5). The other five cell lines also showed similar results (Supplementary Material, Fig. S4). Consistent with 3C-MTS data, the interaction signals were several folds higher in cancer cell lines than in non-cancer cell lines. To examine whether 8q24 or 3q13 amplification caused the increased interactions in cancer cell lines, we performed the qPCR-based copy number tests in the six cell lines (see Supplementary Material online) and observed amplification at 8q24 and 3q13 in PC-3 cell line. We found slight copy number decrease at 3q13 in DU-145 cell line and no significant change in other normal cell lines (LNCaP, BPH-1, RWPE-1 and LCL) (Supplementary Material, Fig. S5). We also browsed Cancer Cell Line Encyclopedia (http://www.broadinstitute.org/) for copy number variation and observed 8q24 and 3q13 amplification in PC-3, and slight copy number change in DU-145 but not LNCaP cancer cell line. Clearly, the copy number variation is not a significant contributor of the frequent 8q24–3q13 interactions.

To further confirm the most common interaction, we performed fluorescence in situ hybridization (FISH) analysis in LNCaP cells using a probe covering CD96 intron 2 (green) and another probe covering 8q24 risk region 1 (red). We also used two additional probes at 4p15 (red) and 9q21 (green) as negative controls. Co-localization of the probe signals was scored as a fusion if the red and green signals were touching each other within the same focal plane (Fig. 6A). Also recorded were red and green signals that were less than or equal to one probe distance apart from one another. We counted 622 nuclei for test probes and 480 nuclei for control probes in two separate counting. We observed 45 fusion signals for the test probes (7.23%) and 10 fusion signals for the control probes (2.08%) (Fisher's exact test, P < 0.0001). The same analysis did not show any significant difference for the one probe distance signals (P > 0.05) (Fig. 6B).

Figure 6.

FISH analysis confirms trans interaction between 8q24 E59–E62 and CD96 intron 2 on 3q13. Two color FISH in LNCaP cells is used to determine the co-localization of 8q24 risk region 1 and CD96 intron 2 on 3q13. (A) Representative FISH signals for test probes covering 3q13 intron 2 (green) and 8q24 region 1 (red) show co-localization and one probe distance. (B) Bar plot showing FISH analysis result. Co-localizations (fusion signals) between target probes are more frequent than control probes (P < 0.0001). The same analysis for one probe distance signal does not show statistical significance (P = 0.28). Yellow color arrows refer to co-localizations, white color arrows refer to one probe distance.

Multiple interaction hot spots between 8q24 risk locus and MYC locus

To further investigate the interaction between the 8q24 risk locus and the MYC locus, we performed a fine mapping analysis by examining the contact frequency at each pair of EcoRI-defined fragments between the 8q24 risk locus (77 EcoRI sites) and the extended MYC gene region (73 EcoRI sites at chr8:128652000-128920000). For the PC-3 cell line, we found multiple clusters of EcoRI fragments at PC risk regions 1 and 3 showing frequent interaction with the MYC region (Fig. 4B). Specifically, we observed at least 11 interaction hot spots between five major fragment clusters at the risk locus and two major fragment clusters at MYC upstream sequences. For risk region 1, the fragment cluster E59–E60 was more likely to interact with the region (chr8:128741351-128754066) containing the MYC promoter and predicted enhancer region (chr8:128675845-128687963). For risk region 3, four clusters (E1–6, E10–11, E22–24 and E26–27) actively interacted with the MYC promoter and enhancer regions. Of those clusters, fragments E22-24 showed an additional hot spot at a predicted repressor between the promoter and enhancer regions. However, the interaction patterns with MYC were cell type- and cell line-dependent. For example, the most active fragments were E1–E3 in LCL, but the same fragments were less active in DU-145, LNCaP, BPH-1 and RWPE-1. The most active fragments were E22–23 in DU-145, E22–23 and E59–60 in LNCaP. For the two PC cell lines (DU-145 and LNCaP), the interactions were limited to the MYC promoter or predicted repressor region. For the two normal prostate cell lines, BPH-1 and RWPE-1, the interactions occurred more frequently at enhancer region (Supplementary Material, Fig. S3). These results were supported by previous reports showing the interaction of region 3 (within the E23–24 fragment containing functional SNP rs6983267) and region 1 (within the E60–61 fragment) with MYC locus in LNCaP (15,16). The results were also supported by another study showing the role of the functional SNP rs378854 in the E3–4 fragment, which interacts with MYC and PVT1 in the RWPE-1 cell line (14). A recent study using 5C assay further strengthened the observation (28).

Functional prediction of active interaction regions

It has been reported that there are active and inactive regions within each chromosome and that the inter-chromosomal contact probability (ICP) index can reflect the active status of a chromatin region (27). To examine the chromatin status of different EcoRI-defined fragments at the 8q24 risk locus, we first summarized the total number of chr8 and non-chr8 contact frequencies for each fragment. We then calculated the ICP index in each cell line using the following formula (27): the sum of a fragment's non-chr8 contact frequencies divided by the sum of chr8 and non-chr8 contact frequencies. The result showed significantly low ICP in several fragment clusters including E72–73, E1–6, E10–11, E22–23, E26–27 and E58–62 (Fig. 4C and D). Based on the ENCODE epigenomic database, these low ICP clusters were predicted to have multiple regulatory elements such as enhancers and insulators. Some clusters also showed enrichment in the H3K27 acetylation site, PolII ChIP-PET chromatin interaction signal, transcription factor binding sites and DNaseI hypersensitivity sites. Interestingly, most of the low ICP clusters demonstrated high contact frequencies with the MYC locus (Fig. 4B). For example, all four interaction clusters in risk region 3 and one interaction cluster in risk region 1 were located in the same fragment clusters as the low ICP. These data indicate that functional elements at 8q24 risk regions act more likely in a form of intra- than inter-chromosome interaction.

To predict the functional significance of these long-range interactions, we extracted all captured genomic fragments with Z score ≥5 in at least one cell line for the six cell lines. We also extracted the fragments with Z score ≥3 for each individual cell line. When Z score = 3, the corresponding normalized read counts were 5.25 in PC-3, 6.57 in LNCaP, 10.95 in DU-145, 7.22 in BPH-1, 9.16 in RWPW-1 and 5.06 in LCL, respectively. We applied Genomic Regions Enrichment of Annotations Tool (GREAT; http://bejerano.stanford.edu/great/public/html/) in the suggestive interaction regions to predict potential functional consequences of these frequent contacts. This online program detected the significant enrichment of the candidate target regions containing genes under a variety of gene ontology terms such as regulation of mesenchymal cell proliferation, organ morphogenesis and ureteric bud development [false discovery rate (FDR) ≤ 9.88E−67, fold enrichment ≥ 2.87] (Fig. 7). We observed significant enrichment in various pathways such as regulation of nuclear β-catenin signaling, and Wnt signaling pathways (FDR ≤ 1.66E−57, fold enrichment ≥ 2.61). Under disease ontology, 18 of top 20 significantly enriched diseases were tumor-related. Interestingly, male genital cancer was among the most significant enriched disease ontology term (FDR = 7.29E−109, fold enrichment = 3.21). Although the enriched gene terms were often shared among individual cell lines tested, we did observe cell line-dependent gene term enrichment. For example, LNCaP was significantly enriched for regulation of chromosome organization (FDR = 4.62E−4, fold enrichment = 2.25). DU-145 was significantly enriched for SH3 domain (FDR = 1.78E−42, fold enrichment = 3.06). RWPE-1 was enriched for protein kinase inhibitor activity (FDR = 1.33E−8, fold enrichment = 2.87) and CCCTC-Binding factor multivalent nuclear factor (FDR = 4.15E−11, fold enrichment = 3.56). Both LNCaP and BPH-1 were significantly enriched for genes relevant to prostatic neoplasm (FDR ≤ 2.56E−5, fold enrichment ≥ 2.14). The interaction regions were also enriched for several promoter motifs that matched transcription factors. For example, the motif NNGGGNCGCAGCTGCGNCCCNN that matched NHLH1 was the most significantly enriched (FDR = 6.13E−67, fold enrichment = 2.95). The significantly enriched motifs also included TTGGCGCGRAANNGNM for E2F1 (FDR = 3.30E−52, fold enrichment = 2.37) and NNNMRCAGGTGTTMNN for TCF3 (FDR = 5.63E−81, fold enrichment = 2.08). The complete list of these significant enrichments in each cell line is listed in Supplementary Material, Table S5.

Figure 7.

Genomic region enrichment analysis of frequent contact fragments in combined dataset consisting of all six cell lines. Only captured genomic regions with Z score ≥5 in at least one cell line were used for analysis. Multiple gene ontology terms, disease ontologies, pathways and promoter motifs are significantly enriched. The black bars represent –log 10 (binomial P value).

DISCUSSION

There is currently a substantial knowledge gap between disease associations derived from genome-wide association studies and an understanding of how these genetic variants contribute to human diseases (29–32). Recent studies have shown that the association signals are strongly associated with enhancer elements and other regulatory regions (33,34), suggesting an important role of many associated variants in the regulation of critical genes. One regulatory mechanism is believed to regulate target genes through long-range chromatin interactions (13,14,35). To fully examine the potential long-range interactions, we applied a newly developed 3C-MTS technology to investigate the most common PC risk locus at 8q24 and its target genomic regions in six different cell lines. We observed frequent interactions of this risk locus with numerous other gene regions, in particular, inter-chromosomal region at the CD96 locus and intra-chromosomal region at the MYC locus. We provided complete fragment-by-fragment interaction map between the 8q24 risk locus and the CD96/MYC loci and showed multiple active interaction regions at these loci. Our results demonstrated significant enrichment of the interaction regions in important ontology terms and pathways. Successful capture of this interaction signal further validates 3C-MTS as a powerful tool for high-resolution survey of multiple risk-loci (such as multiple long LD blocks) for their potential regulatory targets in other genomic regions.

3C-MTS was developed based on two well-established technologies, multiple target capture sequencing and the 3C assay. Sequence capture technology has been successfully used to enrich regions of interest for target sequencing analysis (36). 3C has been used to investigate the long-range interactions between candidate genes and regulatory elements (17,20). However, current 3C-based assays are not suitable for disease risk loci because long LD blocks and unknown target genes make the assay difficult to design. Compared with the traditional 3C method, 3C-MTS does not rely on prior knowledge of the potential interacting partners and is able to detect associations with any genomic region at the whole genome level. It allows systematic examination of multiple regions of interest for their regulatory targets simultaneously, overcoming the limitation of enhanced 4C (37) and its variants (38), which can only be used to investigate one or limited fragments of choice. Furthermore, like other 3C variants such as 4C (22,23), 5C (24), ChIA-PET (25) and Hi-C (26), 3C-MTS allows genome-scale mapping of long-range genomic interactions (20,39). Recently, a similar capture-based 3C sequencing technology (capture-C) has just been published and also demonstrated great potential to analyze hundreds of regulatory landscapes at high resolution in a single, high-throughput experiment (40). Therefore, the powerful 3C-MTS technology will fill the gap between Hi-C (high coverage but low resolution) and 4C (low coverage but high resolution), providing an optimal approach for candidate gene discovery at risk loci with unknown functions.

The shared enrichment of certain genomic regions in both tumor and normal cell lines suggests a common regulatory mechanism. CD96, the most common and strongest interaction locus, may play a role in the adhesive interactions of activated T and NK cells during the late phase of the immune response. Based on Biograph's knowledge base (41), in the context of CD96, PC ranked number 4 among 6021 disease concepts, strongly suggesting a potential role of this gene in PC development. Furthermore, significant genomic region enrichment in various gene ontology terms and pathways (Supplementary Material, Table S5) indicates the involvement of the 8q24 risk locus in regulation of these important functions. For example, our data show that β-catenin and Wnt signaling pathways are among the most significantly enriched in the frequent interaction regions. Recently, Wnt signaling is reported to regulate prostate bud growth and luminal epithelial differentiation (42). Wnt pathway has been widely implicated in mesenchymal differentiation. Coincidentally, our data show significant enrichment in regulation of mesenchymal cell proliferation and ureteric bud development. More interestingly, two transcription factors (E2F1 and TCF3), their promoter motifs most significantly enriched in the frequent interaction regions, participate in the regulation of Wnt/β-catenin activity (43,44). TCF3 is also reported to promote cell proliferation and acts as a tumor promoter in PC (45). In addition, previous studies have shown that the risk allele of a regulatory variant at the 8q24 risk locus confers potential to an enhanced Wnt signaling (13,46) and genomic regions containing PC risk variants show significant enrichment of Wnt signaling genes (47). Taken together, our results suggest that the 8q24 risk regions may regulate multiple genes through physical contact, in particular, the genes involved in β-catenin/Wnt signaling.

It is believed that each chromosome is composed of discrete topologically associating domains (48–51). Chromatin looping is a main mechanism by which regulatory elements communicate with their cognate target genes. This looping mechanism ensures genomic elements that are widely spaced in the linear genome into close spatial proximity (52,53). The 8q24 risk locus is part of a single large chromatin topological domain containing the MYC (54), which may explain the high frequency of interactions between the risk locus and the MYC. Consistent with this finding, a recent 5C-based assay showed that multiple functional domains in the risk regions frequently interact with the MYC gene (28). Meanwhile, the territories of different chromosomes can also form extensive interactions where chromatin loops of different chromosomes may intermingle (55,56), which supports our observation that active elements of the 8q24 risk locus interact with multiple other chromosomal loci across the genome via trans-regulatory mechanisms. The position-dependent interaction between 8q24 and 3q13 indicates that the inter-chromosome interactions occur at a fixed spatial structure in at least a fraction of cells. These results strongly suggest that chromatin at the 8q24 risk locus may constitute a regulatory hub where protein complexes mediate long-range interactions with multiple other genomic loci. The fact that multiple regulatory regions at this risk locus show frequent contacts with multiple other gene regions further supports the chromatin regulatory hub at the 8q24 risk locus.

Like any other 3C-based methods, however, 3C-MTS does not distinguish functional from non-functional association nor does it reveal the mechanisms that led to co-localization. This technology does not directly measure the dynamics and cell-to-cell variation in chromosome folding. Further study using other technologies such as eQTL analysis (57,58), genome editing (59,60) and GENECAP (61) will facilitate functional characterization. Additional tests by examining effect of androgen treatment or interfering chromatin structural proteins will also be important to elucidate the functional role of these interactions. Nevertheless, to our knowledge, this is the first report to systematically examine the whole genome for potential target genes of a cancer risk locus with multiple LD blocks, a process that is currently very challenging. This study not only demonstrates the power of 3C-MTS for high-resolution survey of long-range chromatin interactions on a genome-wide scale, but also provides convincing evidence showing the 8q24 risk locus as a potential regulatory hub. Further understanding genetic effect and biological mechanism of these chromatin interactions will shed light on the newly discovered regulatory role of the risk locus in PC etiology and progression.

MATERIALS AND METHODS

Cell lines and cell culture

PC cell lines DU-145, LNCaP and PC-3, as well as normal-like prostate cell lines BPH-1 and RWPE-1 were purchased from American Type Culture Collection (Manassas, VA). One Epstein–Barr virus transfected LCL was obtained from Mayo Clinic. DU-145, LNCaP and PC-3 were maintained in RPMI 1640, 10% fetal bovine serum (FBS) and antibiotics; BPH-1 was cultured in RPMI 1640, 5% FCS and antibiotics. RWPE-1 cells were cultured in keratinocyte serum-free medium (Gibco, Grand Island, NY) and supplemented with 5 ng/ml human recombinant EGF and 0.05 ng/ml bovine pituitary extract (Sigma, St. Louis, MO).

3C library preparation

3C libraries were prepared as previously described (62). Briefly, 10 million exponentially growing cells were cross-linked with 1% formaldehyde on a rocking platform for 10 min, quenched with a final concentration of 0.125 mM glycine at room temperature for 5 min while shaking, and followed by 15 min on ice to stop cross-linking completely. Cells were counted and placed into aliquots of 5 × 10⁶ cells and stored at −80°C until use. Two aliquots of cells were lysed with 500 µl of 1× cold lysis buffer (10 mm Tris–HCl, pH 8.0, 10 mm NaCl and 0.2% IGEPAL CA-630) including 1× protease inhibitor (Roche, Indianapolis, IN). After incubation on ice for at least 15 min, cells were further lysed with a Dounce homogenizer by moving pestle B up and down 10 times, incubated on ice for 1 min, followed by 10 more strokes with the pestle. The resulting cell nuclei were pelleted and washed twice with 500 µl of ice cold 1× EcoRI buffer. Then the pellet was re-suspended in 500 µl of 1× EcoRI buffer, and sodium dodecyl sulfate (SDS) was added to each tube to a final concentration of 0.3%. After incubating for 1 h at 37°C while shaking at 100 rpm, Triton X-100 was added to a final concentration of 1% along with 600 U of EcoRI enzyme. The samples were then incubated at 37°C overnight. The restriction endonuclease was inactivated by the addition of SDS to a final concentration of 1.6% and incubated at 65°C for 20 min. Ligation mixes, prepared in 15-mL tubes containing 745 μl of 10× T4 Ligase buffer, 10% Triton-X 100, 80 μl of 10 mg/mL BSA, 6 ml of water, 575 μl of cell lysate and 10 μl of 1 U/μl T4 ligase (Invitrogen, Grand Island, NY), were incubated at 16°C for 4 h and then at room temperature for 30 min. The cross-links were reversed by incubating at 65°C overnight. DNA was extracted with phenol–chloroform followed by an ethanol precipitation and quantified by using Qubit dsDNA BR Assay Kits (Life Technologies, Cat# Q32850).

EcoRI digestion and ligation efficiency test

The digestion efficiency was estimated by agarose gel electrophoresis and real-time qPCR by comparing digested DNA to un-digested DNA. For the qPCR assays, three primer sets were designed across three different EcoRI sites at 11q13 and a primer pair that was not across the EcoRI site served as the internal control. The primer sequences are listed in Supplementary Material, Table S2. Digestion efficiency was calculated according to the following formula (62): digestion efficiency (%) = 100 − 100/[2^{(ΔCTdigested−ΔCTundigested)}], where ΔCT is the product of the C_T value from the primer pairs across the EcoRI site minus the C_T value of the primer pairs not across the EcoRI site. The ligation efficiency was tested by agarose gel electrophoresis.

3C-sequencing library

3C libraries that passed quality control were fragmented to a target peak of 500 bp (Covaris E210, Woburn, MA) using a 5% duty cycle at intensity 3 for 80 s with 200 cycles per burst. The fragmented DNA was subject to agarose gel electrophoresis for size selection. The size-selected 400–600 bp DNA fragments were used to construct a 3C sequencing library using a NEBNext DNA Library Prep kit (New England Biolabs, Ipswich, MA). After end repair and dA tailing, the adaptors were ligated, followed by pre-amplification for six cycles using NEB universal primer and an indexed primer. The PCR product was purified by Agencourt AMPure beads (Beckman Coulter, Inc., Cat# A63881) and quantified by Qubit dsDNA BR Assay kits (Life Technologies, Cat# Q32850).

8q24 Target regions and probes design

The target regions were selected based on previous publications showing independent contributions of PC risk in three different regions at the 8q24 locus (13,14,16) (Fig. 1). A total of 77 EcoRI sites that covered the three regions (region 2: 128077370-128138344, region 3–region 1: 128310265-128551675) were used to enrich the DNA fragments containing possible interactions from 3C-sequencing libraries. Each EcoRI site named from E1 to E77 in the order of region 3, region 1 and region 2. xGen Lockdown probes (IDT, Coralville, IA) were designed from both upstream and downstream of each EcoRI site in these regions (Supplementary Material, Table S6). After filtering the repeat sequences, a total of 96 probes met the requirements for probe design. The probes were named according to their EcoRI sites and positions. For example, E1D and E1U represented downstream and upstream of EcoRI site 1, respectively.

Targets enrichment

The targets enrichment procedure followed NimbleGen SeqCap EZ Library SR User's Guide (version 3.0) with some modifications. In brief, 1 μg of 3C-sequencing library DNA, 5 μl of 1 mg/ml COT DNA, 1 μl of 1000 μM universal blocker and 1 μl of index blocker (1000 μM) were mixed and dried in a 1.5 ml tube. Then, 7.5 μl of 2× hybridization buffer and 3 μl of hybridization component A were added to suspend the DNA mixture. After denatured at 95°C for 10 min, the DNA was mixed with 4.5 μl of biotin-labeled probe pool (total 13 pmol) in a 0.2 ml PCR tube. The hybridization was incubated in a Master-cycler pro (Eppendorf, Hamburg, Germany) at 47°C for 72 h. After hybridization, the captured targets were selected by streptavidin-coated magnetic beads (Dynabeads M-270 Streptavidin; Life Technologies, Oslo Cat #2014-08). The captured DNA fragments were amplified for additional 11 amplification cycles and purified by Agencourt AMPure beads (Beckman Coulter, Inc., Cat # A63881). Insert size of the final library was checked using the Agilent High Sensitivity DNA Assay before sequencing.

Target enrichment efficiency was tested by qPCR assays. Relative fold enrichment was calculated by the relative abundance of unrelated control regions (2q31, 7p22, 12q13 and 16p11) and 8q24 target regions in the pre- and post-captured 3C-sequencing libraries. For the target regions, the primer pairs (Supplementary Material, Table S2) were carefully designed so that one primer was located outside the capture probe, hence qPCR could not amplify the capture probe. The relative enrichment efficiency was calculated by ΔCt value according to the following formula: enrichment efficiency = 2^{−(CTpost-captured library−CTpre-captured library)}.

Next-generation sequencing and data analysis

All sequencing was performed on an Illumina Genome Analyzer HiSeq2000 with 100 bp PE reads. Two indexed libraries were pooled and sequenced in one lane. Paired-end sequences were aligned to a reference genome using Bowtie (v.2.0). The seed for alignment is 22 bases and no mismatches are allowed. For a certain sequence pair (R1 and R2), various level categories were classified, including unmapped sequence pairs, mapped sequence pairs, sequence pairs mapped outside/inside target regions, non-specific mapped sequence pairs, same fragment ligations and nearby ligations and long-range ligation sequence pairs. Sequences that were unrelated to the 8q24 risk regions and sequences with both sequence ends within 8q24 target regions (±25 kb) were filtered out. To correct for potential bias of probe capture efficiency, raw interaction numbers were normalized based on the relative enrichment efficiency of each probe. The final number of interactions was reported as normalized read counts per 10 kb window and per EcoRI fragment, respectively.

Quantitative polymerase chain reaction

EcoRI digestion and target enrichment efficiency were both tested by real-time PCR using SYBR green PCR master mix (Applied Biosystems, Foster City, CA). Each 20 μl reaction consisted of 1× SYBR master mix, 2.5 μm constant primer, 2.5 μm testing primer and 10 ng of template DNA. PCR cycles were as follows: an initial denaturing step for 10 min at 95°C; 45 cycles of 15 s at 94°C, then 60 s at 60°C. A StepOne plus Real-Time machine (Applied Biosystems, Foster City, CA) was used for the quantitation. Each PCR reaction was performed in duplicate, and the data presented in this paper were the average of results for all PCR reactions.

Taqman qPCR technology was used to quantify the ligation frequency of CD96 on 3q13 region and 8q24 region 1. All PCR reactions were performed using Taqman Universal Master Mix II (Applied Biosystems, Foster City, CA, Cat# 4440038). Each of 10 μl reaction consisted of 1× Taqman Universal MasterMix II, 1 μl 5 uM anchor primer, 1 μl test primer, 1 μl Taqman probe (2.5 μm) and 40 ng 3C DNA. PCR cycles were as follows: an initial step 2 min at 50°C, 10 min at 95°C, 50 cycles of 15 s at 95°C and 60 s at 60°C. Each PCR reaction was performed in duplicate, and the data presented were the average of at least two independent experiment results for all PCR reactions.

3C primers design and normalization controls

Two sets of primers were designed for the detection of interactions between CD96 on 3q13 and 8q24 region 1. Two anchor primers next to two EcoRI sites at 3q13 were designed with one near the 3L position (chr3:111274262) and the other near the 3R position (chr3:111274323). Nineteen test primers were designed on 8q24 region 1 around 10 EcoRI cutting sites from position chr8:128511051 to position chr8:128551221, which corresponded to capture probe sites from E56 to E65. Each EcoRI site had two primers, one upstream and the other downstream (Fig. 5). Each test primer was paired with each of the anchor primers. The sequences of the primers are listed in Supplementary Material, Table S7. The contact frequency of each interaction pair was normalized using BAC clones as control templates that covered all ligation products in equal amounts. The BAC clones used to generate the control templates were RP11-12P11 (3q13) and RP11-1150B6 (8q24) (Empire Genomics, Buffalo, NY). Adjacent fragment ligation frequency was used to normalize the different loading, fixation and ligation efficiencies between different cell lines (63,64).

Fluorescence in situ hybridization

FISH analysis was performed using labeled custom FISH probes (Empire Genomics, Buffalo, NY). Custom test FISH probes on 3q13 (RP11-12P11) and 8q24 (RP11-1150B6) were labeled with Green 5-Fluorescein dUTP and Red 5-ROX dUTP, respectively. Custom control probes were selected on 4p15 (RP11-420H4 labeled with Red 5-ROX dUTP) and 9q21 (RP11-698K16 labeled with Green 5-Fluorescein dUTP). Cells were fixed with 1% formaldehyde and washed three times in Carnoy's (methanol-glacial acetic acid) fixative, dried to glass slides, and aged overnight at room temperature. The FISH procedure was carried out according to the manufacturer's instructions. FISH signals were examined using an Olympus BX61 microscope (Olympus America, Center Valley, PA). Images were taken separately using appropriate filters and assembled using the Olympus cellSens standard software.

Gene enrichment analysis

An online bioinformatics tool (GREAT v2.0.2) (65) was applied to examine the functional significance of long-range interactions. The tool was designed to test gene and DNA motif enrichments in candidate regions of interest. The association rule in this GREAT analysis was Basal + extension (5 kb upstream, 1 kb downstream and 1000 kb max extension). We used term annotation count range of (20, Inf) and accepted all default parameters defined by the program. The whole genome was set as background. We selected genomic regions with Z score ≥3 for the GREAT analysis. For each captured genomic region, Z score was calculated using the following formula: (normalized read counts – mean of all captured read counts)/standard deviation. A Z score of 3 indicates that the captured read counts are 3 SD above the mean.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

Conflict of Interest statement. None declared.

FUNDING

Medical College of Wisconsin Cancer Center Seed Fund (3305738 to L.W.), National Institute of Health (R01CA157881 to L.W.) and (CA151254 to S.N.T.).