Structural Variants at the BRCA1/2 Loci are a Common Source of Homologous Repair Deficiency in High-grade Serous Ovarian Carcinoma

Abstract

Purpose: The abundance and effects of structural variation at BRCA1/2 in tumors are not well understood. In particular, the impact of these events on homologous recombination repair deficiency (HRD) has yet to be demonstrated.

Experimental design: Exploiting a large collection of whole-genome sequencing data from high-grade serous ovarian carcinoma (N = 205) together with matched RNA sequencing for the majority of tumors (N = 150), we have comprehensively characterized mutation and expression at BRCA1/2.

Results: In addition to the known spectrum of short somatic mutations (SSM), we discovered that multi-megabase structural variants (SV) were a frequent, unappreciated source of BRCA1/2 disruption in these tumors, and we found a genome-wide enrichment for large deletions at the BRCA1/2 loci across the cohort. These SVs independently affected a substantial proportion of patients (16%) in addition to those affected by SSMs (24%), conferring HRD and impacting patient survival. We also detail compound deficiencies involving SSMs and SVs at both loci, demonstrating that the strongest risk of HRD emerges from combined SVs at both BRCA1 and BRCA2 in the absence of SSMs. Furthermore, these SVs are abundant and disruptive in other cancer types.

Conclusions: These results extend our understanding of the mutational landscape underlying HRD, increase the number of patients predicted to benefit from therapies exploiting HRD, and suggest there is currently untapped potential in SV detection for patient stratification.

Figure 1. Abundance, location and size of structural variants overlapping BRCA1/2 in three HGSOC cohorts.

a) Alignment of structural variants overlapping BRCA1 across the AOCS, TCGA and SHGSOC cohorts with breakpoints marked in grey according to their position on chromosome 17. Location of BRCA1 marked by a blue line with deletions (blue), duplications (orange) and inversions (purple). b) The distribution of sizes of structural variants (Mb), overlapping BRCA1 across all cohorts. c) Alignment of structural variants overlapping BRCA2 across the three cohorts with breakpoints marked in grey according to their position on chromosome 13. Location of BRCA2 marked by a blue line. d) The distribution of sizes of structural variants (Mb) overlapping BRCA2 across all cohorts with deletions (blue), duplications (orange) and inversions (purple).

Figure 2. Structural variation and expression of BRCA1/2 in the combined HGSOC cohort.

a)c) Expression of BRCA1/2 (variance stabilising transformed RNA-seq counts) across samples ordered from lowest to highest expression. Median BRCA1/2 expression is indicated by a black dashed line. Sample bars are coloured by BRCA1/2 mutational category. b)d) Boxplot of BRCA1/2 expression for each category of BRCA1/2 mutation. The BRCA1 deletion category is split into those samples with SNVs and deletions and those with only deletions as their expression is significantly different (Supplementary Figure 5a). This is not the case for BRCA2 so all samples with deletions are considered together to maximise the available sample size (Supplementary Figure 5b).

Figure 3. BRCA1/2 mutation classes and repair deficiency in three HGSOC cohorts.

a) Predictions of HRD for three large cohorts of HGSOC coloured by BRCA1/2 structural variant status and outlined by BRCA1/2 mutation status. Categories of mutation include GSMs and SSMs. Categories of structural variation include deletions, duplications, inversions and complex overlapping combinations thereof as formally described in the methods and absence of structural variants at BRCA1/2. The HRDetect scores range from 0, least likely to be HR deficient to 1, most likely to be HR deficient. The red dashed line represents the threshold of 0.7 representing HRD³³³³³³. b) The number of HRD tumours with different categories of BRCA1/2 short variants, deletions or non-deleting SVs. c),d) The increase in log odds ratio of HRD (HRDetect score > 0.7) associated with different categories of mutation and structural variation at BRCA1/2 in comparison to the frequency of the reference category where samples lack evidence of BRCA1/2 inactivation (GSV, SSV, SV). All categories apart from the BRCA1 promoter methylation category itself also exclude tumours with BRCA1 promoter methylation where this is known (TCGA and AOCS). ORs are defined using Fisher’s Exact tests for enrichment. Error bars represent 95% confidence intervals. Mutually exclusive categories of mutation examined include GSV only, SSV only, the presence of a deletion at one or both genes without a GSV or SSV, the presence of a short variant together with deletion of one or both genes, non-deleting SVs in samples without short variants or deletions, samples with BRCA1 promoter methylation and no mutational BRCA1/2 deficiencies.

Figure 4. Integrative modelling of repair deficiency in HGSOC.

a) Median effect sizes of genomic features selected to predict HRD, using elastic net regularised regression on 100 training/test set splits. Model performance was measured for each split and average AUC = 0.75. Binary mutational status variables (e.g. presence/absence of BRCA1 somatic SNV) were included as factors and continuous variables were standardised to allow comparisons between variables. b) Distributions of effect size for each variable on HRD (log odds) in each training/test set split. Variables in red are selected for inclusion by the model in more than half of the training sets.

Figure 5. Predicted HRD is associated with patient survival in the absence of short variants at BRCA1/2.

a) The effect of HRD on overall survival time after diagnosis (in days) in HGSOC. (N (events) =190 (144)). b) The effect of HRD on overall survival time after diagnosis (in days) in HGSOC patients without BRCA1/2 GSV/SSV. (N (events) =145 (113)). c) Forest plot showing the effects of HRD on overall and progression-free survival, unadjusted and adjusted for age and stage at diagnosis in a multivariable model. Estimates are also shown with tumours with SNVs/indels at BRCA1/2 excluded. Kaplan-Meier plots compare survival times between HR deficient and HR proficient patients as defined by HRDetect score above and below 0.7. Hazard ratio estimates (on a log scale) are taken from Cox proportional hazards models and correspond to a 1 standard deviation increase in HRDetect score, stratified by cohort and adjusted for age and stage at diagnosis.

Figure 6. Deletions at BRCA1/2 in other cancer types and their impact on expression.

a) The proportion of samples with deletions at BRCA1, BRCA2 or both, by primary site in PCAWG. b) BRCA1 expression in tumours with and without deletions at BRCA1 coloured by primary site. c) BRCA2 expression in tumours with and without deletions at BRCA2 coloured by primary site. Gene expression visualised in variance stabilising transformed (VST) counts. P-values are from differential expression analyses conducted using DESeq2 for each primary site separately and are adjusted for multiple testing using the Benjamini-Hochberg approach. Only sites with at least 15 samples in total and at least 3 samples with a deletion at the gene in question were included.