Characterising the contribution of rare protein-coding germline variants to prostate cancer risk and severity in 37,184 cases

Abstract

The etiology of prostate cancer, the second most common cancer in men globally, has a strong heritable component. While rare coding germline variants in several genes have been identified as risk factors from candidate gene and linkage studies, the exome-wide spectrum of causal rare variants remains to be fully explored. To more comprehensively address their contribution, we analysed data from 37,184 prostate cancer cases and 331,329 male controls from five cohorts with germline exome/genome sequencing and one cohort with imputed array data from a population enriched in low-frequency deleterious variants. Our gene-level collapsing analysis revealed that rare damaging variants in SAMHD1 as well as genes in the DNA damage response pathway (BRCA2, ATM and CHEK2) are associated with the risk of overall prostate cancer. We also found that rare damaging variants in AOX1 and BRCA2 were associated with increased severity of prostate cancer in a case-only analysis of aggressive versus non-aggressive prostate cancer. At the single-variant level, we found rare non-synonymous variants in three genes (HOXB13, CHEK2, BIK) significantly associated with increased risk of overall prostate cancer and in four genes (ANO7, SPDL1, AR, TERT) with decreased risk. Altogether, this study provides deeper insights into the genetic architecture and biological basis of prostate cancer risk and severity.

Figure 1:. Manhattan plot of all meta-analysis gene-level association tests with risk of developing overall prostate cancer.

The x-axis is the genomic position of the gene, and the y-axis is the −log₁₀ transformed unadjusted P-values for all qualifying variant models as indicated in the legend. The light grey dashed line represents the suggestive significance threshold (P = 2.6×10⁻⁶) and the dark grey dashed line the study-wide significance threshold (P = 1×10⁻⁸). Genes which reach the suggestive significance threshold are labelled, and only the most significant qualifying variant model for each gene is labelled. Qualifying variant models defined in Supplementary Table 1.

Figure 2:. Forest plot showing the association of genes with prostate cancer risk (All PCa Vs Ctrls) and severity (Agg. PCa Vs Ctrls) which reached the suggestive significance threshold (P = 2.6×10⁻⁶) for non-synonymous qualifying variant (QV) models.

Gene and QV model are as indicated in legend. For genes where more than one QV model (defined in Supplementary Table 1) passed the suggestive significance threshold the most significant is plotted. ptv = rare protein truncating variant QV model; flexdmg = rare damaging non-synonymous QV model. PCa = prostate cancer; Agg = aggressive. Non-agg = non-aggressive.

Figure 3:. Summary of ExWAS variants which reached study-wide significance (P<1×10⁻⁸) in the meta-analysis for the risk of developing prostate cancer.

The x-axis is the variant minor allele frequency in non-Finnish Europeans, and the y-axis is the variant effect estimate. Gene labelled variants are those which are rare in non-Finnish Europeans (Allele Frequency < 1%) and had a posterior inclusion probability of being a causal variant greater than 0.05 in the FinnGen study. The P-value used to determine significance is from the Stouffer’s meta-analysis and as this does not generate an effect-size we report here the effect estimate from the FinnGen cohort.