Integrative multi-omics analyses to identify the genetic and functional mechanisms underlying ovarian cancer risk regions

Abstract

To identify credible causal risk variants (CCVs) associated with different histotypes of epithelial ovarian cancer (EOC), we performed genome-wide association analysis for 470,825 genotyped and 10,163,797 imputed SNPs in 25,981 EOC cases and 105,724 controls of European origin. We identified five histotype-specific EOC risk regions (p value <5 × 10^-8) and confirmed previously reported associations for 27 risk regions. Conditional analyses identified an additional 11 signals independent of the primary signal at six risk regions (p value <10^-5). Fine mapping identified 4,008 CCVs in these regions, of which 1,452 CCVs were located in ovarian cancer-related chromatin marks with significant enrichment in active enhancers, active promoters, and active regions for CCVs from each EOC histotype. Transcriptome-wide association and colocalization analyses across histotypes using tissue-specific and cross-tissue datasets identified 86 candidate susceptibility genes in known EOC risk regions and 32 genes in 23 additional genomic regions that may represent novel EOC risk loci (false discovery rate <0.05). Finally, by integrating genome-wide HiChIP interactome analysis with transcriptome-wide association study (TWAS), variant effect predictor, transcription factor ChIP-seq, and motifbreakR data, we identified candidate gene-CCV interactions at each locus. This included risk loci where TWAS identified one or more candidate susceptibility genes (e.g., HOXD-AS2, HOXD8, and HOXD3 at 2q31) and other loci where no candidate gene was identified (e.g., MYC and PVT1 at 8q24) by TWAS. In summary, this study describes a functional framework and provides a greater understanding of the biological significance of risk alleles and candidate gene targets at EOC susceptibility loci identified by a genome-wide association study.

Keywords: GWAS; epithelial ovarian cancer risk; fine mapping; functional mechanisms.

Figure 1

Study design for the identification and functional analysis of common low penetrance risk variants for epithelial ovarian cancer (EOC) (A) Study design: HGSOC, high-grade serous ovarian cancer; ENOC, endometrioid ovarian cancer; LGSOC, low-grade serous ovarian cancer; CCOC, clear-cell ovarian cancer; MOC, mucinous ovarian cancer; NMOC, all non-mucinous ovarian cancer; fine mapping of risk regions and epigenomic annotation to identify credible causal risk variants (CCVs) at each risk locus; cell-type-specific epigenomic enrichment and partitioning heritability analysis; expression QTL-based approaches to identify candidate susceptibility genes in EOC risk regions; and 3D looping analysis to identify gene-CCV interactions at EOC risk loci. (B) Genome-wide association analyses identified five novel EOC risk loci at p < 5 × 10⁻⁸ for different EOC histotypes; bubble plot illustrates both the effect size and the statistical significance value for each region by histotype. (C) Risk associations by histotype for previously reported EOC risk regions including 22 regions for which the association signal replicates in one or more histotypes at p < 5 × 10⁻⁸ and 11 risk regions that do not replicate at genome-wide significance.

Figure 2

Regional plots show primary and secondary risk association signals for different histotypes at the 8q24.1 and 3q31.1 risk loci SNPs are colored by histotype with light points representing p values from primary (unadjusted) analyses and dark points representing p values from CCVs. (A) At the 8q24.21 risk locus we identified two independent risk signals for HGSOC (dark red points), two independent risk signals for NMOC (dark blue points), and one signal for MOC (dark orange points at 128,087,904). p values from unadjusted association analyses are plotted beneath CCVs in light red, blue, and orange for HGSOC, NMOC, and MOC, respectively. (B) At the 2q31.1 risk locus, primary signals for the HGSOC, MOC, and NMOC histotypes were co-localized to the same region with evidence for a secondary signal for MOC located approximately 500 kb distal to the primary signal for the same histotype.

Figure 3

Estimates of SNP-heritability (h_g²) and enrichment in regulatory features (A) Overall SNP-heritability estimates for each EOC histotype. The error bars represent 95% confidence intervals. (B) Enrichment of 24 functional annotations for NMOC and HGSOC risk loci calculated as the proportion of estimated SNP-heritability explained by the proportion of SNPs in each of several functional categories. Statistically significant annotations (p < 0.05) are shown in orange. (C) Enrichment analyses for CCVs by EOC histotype (HGSOC, high-grade serous ovarian cancer; LGSOC, low-grade serous ovarian cancer; CCOC, clear-cell ovarian cancer; MOC, mucinous ovarian cancer; FT, fallopian tube epithelial cells; OSE, ovarian surface epithelial cells; and EEC, endometriosis epithelial cells). Enriched histotype-specific chromatin states are shown in red; depleted chromatin states are shown in blue; density of the color indicates strength of enrichment/depletion; size of the circle indicates the probability of enrichment, circles outlined met significance (probability > 0.98).

Figure 4

Looping interactions between candidate genes, CCVs, and REs at the 2q31.1 locus (A) Locus plot of 2q31.1 showing the integration of genetic fine mapping data (NMOC CCVs in purple, MOC CCVs in pink) with chromatin state calls by cell/tissue type. The profiles of cis-interactions in the fallopian tube cell line FT33 and HGSOC cell line Kuramochi were very similar. A heatmap shows the CCV/gene scores associated with each gene in the region. The greater the intensity the stronger the evidence of the interaction. This is summarized across CCVs separately for each histotype to the right of the heatmap. (B) The gray box zooms into a region of strongest interaction between HOXD-AS2 and two CCVs associated with NMOC risk about 500 kb distal to the gene. A cluster of NMOC CCVs are located in an active enhancer/promoter region of HAGLROS in both FT33 and Kuramochi cells. Kuramochi cells exhibit looping between the CCV rs6755766 (chr2:177,043,205) within a Remap ChIP-seq RUNX2 peak that breaks a RUNX2 motif. (C) Highlights this motif disruption.

Figure 5

Looping interactions between candidate genes, CCVs, and REs at the 8q24.1 risk locus (A) Locus plot of 8q24.1 showing the integration of genetic fine mapping data (HGSOC CCVs in red, NMOC CCVs in green) with chromatin state calls by cell/tissue type, and cis-interaction analysis in the fallopian tube cell line FT33 and HGSOC cell line Kuramochi. A heatmap shows the CCV/gene scores associated with each gene in the region. The greater the intensity, the stronger the evidence of and interaction. This is summarized across CCVs separately for each histotype to the right of the heatmap. (B and C) Zoom in on the two regions highlighted: (B) highlights chromatin state calls in around the MYC gene, which shows greatest interaction with HGSOC SNPs located in region C. Epigenomic profiling indicate activation of the MYC promoter in both FT33 and Kuramochi cell lines. HiChIP identifies a strong interaction with a single HGSOC CCV that disrupts a Tead3/4 TF motif in a remap PAX8-TEAD3/4 ChIPseq co-binding site (see black bars in top tracks) in (C), implicating a biological connection between MYC and PAX8/TEAD3/4. Consistent with this, knock down of PAX8 (siPAX8) in HGSOC cells resulted in a reduction in MYC expression compared to two controls (siNT1 and siNT2) (error bars show standard error) (B).