Authors: Lijun Xu, Brett Liddell, Ksenija Nesic, Franziska Geissler, Lauren M Ashwood, Matthew J Wakefield, Clare L Scott, Nicola Waddell, Olga Kondrashova
Categories: Cancer Gene Regulation, Chromatin, and Epigenetics, AcademicSubjects/SCI00030, AcademicSubjects/SCI00980, AcademicSubjects/SCI01060, AcademicSubjects/SCI01140, AcademicSubjects/SCI01180
Source: NAR Cancer
In ovarian and breast cancer, promoter methylation of BRCA1 or RAD51C is a promising biomarker for PARP inhibitor response, as high levels lead to homologous recombination deficiency (HRD). Yet the extent and role of such methylation in other cancers is not clear. This study comprehensively investigated promoter methylation of eight homologous recombination repair genes across 23 solid cancer types. Here, we showed that BRCA1 methylated cancers were associated with reduced gene expression, loss of heterozygosity (LOH), TP53 mutations and genomic features of HRD. We identified BRCA1 methylation in 3% of the copy-number high subtype of endometrial cancer, and as a rare event in six other cancer types, including lung squamous cell, pancreatic, bladder and stomach cancer. RAD51C promoter methylation was widespread across multiple cancer types, but HRD features were only observed for cases which contained high-level tumour methylation and LOH of RAD51C. While RAD51C methylation was frequent in stomach adenocarcinoma (6%) and low-grade glioma (2.5%), it was mostly detected at a low tumour level, suggestive of heterozygous methylation, and was associated with CpG island methylator phenotype. Our findings indicate that high-level tumour methylation of BRCA1 and RAD51C should be explored as a PARP inhibitor biomarker across multiple cancers.
The homologous recombination repair (HRR) DNA double-strand break repair pathway can be disrupted in cancer, resulting in homologous recombination deficiency (HRD). HRD cancers tend to accumulate additional DNA damage, leading to specific genomic scarring patterns (1). HRD can be exploited therapeutically using platinum agents or a class of DNA repair inhibitors, called poly (ADP-ribose) polymerase inhibitors (PARPi), which have had profound impact in the clinic. The efficacy of PARPi therapy was initially demonstrated in high-grade serous ovarian cancer (referred to as OV hereafter) and breast cancer (BC), in which HRD was driven by germline pathogenic variants in either the BRCA1 or BRCA2 genes (2). However, a substantial proportion of cancers without BRCA1/2 mutations can also manifest somatic mutational HRD molecular features, indicating the existence of other mechanisms of HRD. Biallelic inactivation of other core HRR genes including RAD51C, RAD51D (3) and PALB2 have been associated with HRD (4), while alterations in BRIP1 (5), RAD51B (6) and XRCC3 (7) have also been suggested to impact the competency of the HRR pathway.
Promoter-associated DNA CpG methylation has been reported in cancer for some HRR genes, specifically, BRCA1 and RAD51C (8–10). Homozygous promoter methylation of these genes (methylation of all gene copies) can result in silencing of the respective genes, leading to the HRD phenotype (11) and subsequently, susceptibility to PARPi therapy, as supported by pre-clinical and retrospective clinical cohort analyses (12). Promoter methylation of BRCA1 and RAD51C in OV and BC has been extensively characterised (9,13). In OV, the prevalence of BRCA1 methylation was reported in 15% of HRD cases and RAD51C methylation in 1% of HRD cases (13). A recent study used HRD-associated genomic patterns to provide evidence of HRD in 30 different cancer types, noting the presence of BRCA1 and RAD51C methylation (14). Furthermore, evidence has linked abnormal methylation of BRCA1 in normal tissue during early development stages (constitutional BRCA1 promoter methylation) to an increased risk of OV and BC (15). However, despite the clinical relevance, characterisation of HRR-methylated cases in other cancer types and association with an HRD phenotype has not yet been explored.
In this study, we explore publicly available solid cancer datasets, including The Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC) (16), encompassing 23 major cancer types to identify cancers with aberrant somatic promoter methylation in HRR genes. We assess the molecular profiles of cases with BRCA1 or RAD51C methylation by integrating methylation arrays, RNA sequencing and DNA sequencing to assess somatic copy number aberrations (CNAs), mutations and genomic ‘scarring’ predictive of HRD, as well as molecular and clinical subtypes. Finally, we estimate the methylation zygosity by adjusting for tumour purity and ploidy to predict gene silencing.
We used publicly available data from TCGA and ICGC. To ensure adequate sample size for prevalence assessment and downstream analysis, we chose 22 solid cancer types with at least 100 samples analysed using the Illumina Human Methylation 450 (HM450) array, including 7 625 primary cancer samples and 727 matched normal samples. We incorporated the ovarian dataset from ICGC (81 primary high-grade serous ovarian cancer samples and 81 matched normal samples) instead of TCGA OV, which was assayed with a platform of limited coverage across genome Illumina Human Methylation 27 (HM27). Additionally, 33 stomach adenocarcinoma (STAD) cell lines from the DepMap dataset were included in our analysis.
This project used publicly available datasets. The QIMR Berghofer Human Research Ethics Committee approved use of public data (P2095).
For TCGA, the masked HM450 raw intensity files were downloaded from Genomic Data Commons (GDC) between April – June 2022 using GDC Data Transfer Tool Client (v1.6.0) (https://portal.gdc.cancer.gov/). For ICGC, the HM450 raw intensity files were accessed from the Gene Expression Omnibus (GEO) accession GSE65821 in March 2022. The raw HM450 data was converted into beta values (Supplementary Methods). For DepMap, reduced representation bisulfite sequencing data for 33 STAD cell-lines was obtained from Sequence Read Archive (SRA) under the accession number PRJNA523380 (accessed in April 2023), was processed and converted to percentage methylation per CpG site (Supplementary Methods).
A set of probes was selected to assess promoter methylation status, including CpG probes proximate to promoter regions or overlapping CpG islands with low median methylation levels (≤0.2) and low variability (IQR ≤ 0.2) (Supplementary Table S1 and Supplementary Figure S1). To reduce background noise, we evaluated the median methylation levels for eight HRR genes in the normal tissue across each cancer type. As such, BRIP1 and RAD51B were excluded from breast cancer analysis due to high median methylation in normal tissue. Promoter methylation in primary tumours was assessed using at least 60% of selected probes, with methylation ≥0.25. Further details about probe selection and determination of the methylation status are provided in Supplementary Methods.
CpG island methylator phenotype (CIMP) classification was obtained from Yates et al., (2022) (17) for TCGA STAD, BC, low-grade glioma (LGG) and uterine corpus endometrial carcinoma (UCEC). For ICGC OV and TCGA testicular germ cell tumour (TGCT) dataset, we performed classification following the same methods (17).
Gene expression data for TCGA was obtained as pre-processed gene counts from GDC between April and June 2022 using the GDC Data Transfer Tool Client (v1.6.0) (https://portal.gdc.cancer.gov/). The ICGC and DepMap mRNA sequencing data were processed as described by Patch et al. (13). All gene counts data was filtered for lowly expressed genes and transformed to log2 counts-per-million (CPM) using edgeR (v3.40.2) (18).
Mutation data for TCGA was sourced from the Multi-Center Mutation Calling in Multiple Cancers project (MC3) (19). For ICGC, the method used to detect substitutions and indels is reported in Patch et al. (13). Somatic mutation for DepMap was obtained from the DepMap Portal (https://depmap.org/portal/; release 23Q2; accessed in Oct 2023).
CNAs for TCGA were obtained from GDC (June 2023). For ICGC, we used copy number segment information from whole genome sequencing (WGS) data using ascatNGS (v4.0.1) (20). For DepMap, the pre-processed copy number and LOH information was obtained from the CCLE ABSOLUTE (CCLE_ABSOLUTE_COMBINED_20 181 227, 2019 release).
HRD scores for TCGA were obtained from Knijnenburg et al. (6). For ICGC and DepMap, the HRD scores were calculated using scar HRD package (21) (v0.1.1).
Further details about determination of molecular characteristics are provided in Supplementary Methods.
Individual beta values for the selected probes as enlisted in the promoter methylation section were corrected using purity and gene-level copy number. We adopted the equation from VanLoo lab (22):
; ; ; ;
The normal copy number was assumed to be 2 for all specimens. The normal methylation rate was calculated as the mean beta value of the corresponding normal tissue within the same dataset. TCGA LGG did not have data available for matched normal tissue, therefore we used normal tissue from glioblastoma multiforme (GBM) instead.
Drug sensitivity information (drug sensitivity replicate-level dose (CTD^2)) was obtained from the DepMap Portal (https://depmap.org/portal/; release 23Q2; accessed in October 2023). Drug sensitivity curve was generated using R package drc (v3.0-1) (23).
Fisher's exact test was performed to test the independence between the CIMP classification/other molecular subtypes and BRCA1 or RAD51C methylation status. The comparison of HRD scores between groups was performed using the Mann–Whitney U test when the group number equals 2 or the Kruskal–Wallis test when the group number was >2. The posthoc Dunn's test corrected by Benjamini–Hochberg adjustment was performed following a significant Kruskal–Wallis test. P-value and q-value of <0.05 was considered statistically significant for statistical testing.
All other methods and sources for cancer subtypes, patient information and survival analysis are described in detail in the Supplementary Methods.
We evaluated the promoter methylation profile of eight HRR genes (BRCA1/2, RAD51B/C/D, BRIP1, PALB2 and XRCC3), using data from TCGA and ICGC. Our analysis was focused on solid cancer types with methylation array data available for at least 100 samples (except for OV, n = 81 samples), resulting in 7619 primary cancer samples from 23 solid cancer types assessed.
Our focus was on the promoter regions, which were extended to include overlapping CpG islands and directly adjacent lowly methylated CpGs (Supplementary Table S1). Of the eight HRR genes examined, the median methylation observed at the promoter regions of RAD51D and XRCC3 was ≤25% in all samples across all cancer types assessed (Supplementary Figures S2-S10), suggesting absence of promoter methylation for these two genes. Median methylation ≤25% was also found at the promoter regions of BRIP1 and BRCA2 for most cases, with only rare outliers. Notably, we observed one case of BRCA2 promoter methylation in BC (Supplementary Figure S2 and S4). BRIP1 promoter methylation (median value > 40%) was observed in one to two cases of cervical squamous cell carcinoma (CESC), liver hepatocellular carcinoma (LIHC), STAD and UCEC (Supplementary Figure S2 and S5).
Promoter methylation analysis of BRCA1 and RAD51C contrasted with that of the other genes described, revealing a more diverse methylation landscape, which formed the focus of exploration in this study. Analysing selected promoter probes of BRCA1 and RAD51C, we identified promoter methylation of BRCA1 (meBRCA1) or RAD51C (meRAD51C) (Supplementary Figure S2, S3 and S8) in 15 out of 23 cancer types (Figure 1A, B, Supplementary Table S2). MeBRCA1 was present in 10 cancer types (Figure 1a) and was prevalent in OV (12.4%) and BC (1.8%), consistent with previous reports (14). We also observed meBRCA1 in TGCT (6.7%), which has been previously reported by Shen et al. (24). MeBRCA1 was found in 0.93% of the UCEC cohort, which to our knowledge has not been previously reported in the literature. The distribution of meRAD51C was different from that of meBRCA1. MeRAD51C was detected in 13 cancer types (Figure 1B), with the highest prevalence observed in TGCT (8.7%), followed by stomach adenocarcinoma (STAD) (6.1%), OV (3.7%), low grade glioma (LGG) (2.5%) and BC (2.3%). In the eight cancer types where meBRCA1 and meRAD51C were observed (Figure 1a and b), these events were mutually exclusive in individual samples, except for TGCT (Supplementary Figure S11). Strikingly within TGCT, the methylation profiles among samples with meBRCA1 or meRAD51C or combined meBRCA1 and meRAD51C showed low methylation signal across all the BRCA1/RAD51C promoter probes examined (Supplementary Figure S11), suggesting a distinctive methylation pattern of these two genes in TGCT compared to the other cancer types.
Figure 1. RAD51C and BRCA1 promoter methylation across 23 solid cancer types. The prevalence of BRCA1 (A) and RAD51C (B) promoter methylation in 23 cancer types, combining data from TCGA (22 cancer types) and ICGC (high-grade serous ovarian cancer). Cancer types are arranged by the highest combined prevalence of methylation (meBRCA1 and meRAD51C). Promoter methylation status is determined based on specific CpG probes within 200 bp of the promoter region. The bar chart illustrates the percentage of cases with methylation. The percentage of cases with methylation, number of cases with methylation and total number of cases is indicated above each bar. MeBRCA1 and meRAD51C distributions within cancer types with more than three detected cases are detailed in terms of distinct histological subtypes (C) and molecular subtypes (D). Testicular germ cell tumour (TGCT) is exempted from molecular subtype classification due to the absence of well-established molecular subtypes. Both histological subtypes and molecular subtypes are presented as two an upper panel showing percentages of cases with meBRCA1 and meRAD51C and a lower panel displaying counts. Bars are colour-coded: black (co-exist meBRCA1 and meRAD51C), green (meBRCA1), blue (meRAD51C), grey (no promoter methylation detected in either gene). Abbreviations are high-grade serous ovarian cancer subtype (OV): CCNE1 amplification (CCNE1), homologous recombination deficiency (HRD) and homologous recombination wild-type (WT); stomach adenocarcinoma subtype (STAD): Epstein-Barr virus infected (EBV), chromosomal instability (CIN) and microsatellite instability (MSI); low grade adenocarcinoma subtype (LGG): IDH mutated with hemizygous codeletion of chromosome arms 1p and 19q (IDHmut codel), IDH mutated without codeletion (IDHmut non-codel) and IDH wild type (IDHwt); breast cancer (BC) basal, Claudin-low (CLOW), HER2 positive (Her2), HER2-enhanced (HER2E), luminal A (LumA), luminal B (LumB) and normal-like (Normal); uterine corpus endometrial carcinoma (UCEC) copy number high (CN HIGH), copy number low (CN LOW), microsatellite instability (MSI) and DNA polymerase epsilon mutation (POLE).
To explore potential drivers of meBRCA1 and meRAD51C, we assessed correlations with histological and molecular subtypes for cancer types with more than three cases of meBRCA1 or meRAD51C (Supplementary Table S3). With the exception of TGCT, where meRAD51C was enriched in mixed germ cell subtype (Fisher's exact; q-value = 0.003), we found no evidence of enrichment of histological subtype within meBRCA1 or meRAD51C samples (Figure 1C). Instead, we identified associations between meBRCA1 or meRAD51C and molecular subtypes of OV, STAD and BC (Figure 1D). In OV, we found both meBRCA1 and meRAD51C exclusively within the HRD molecular subtype, with meBRCA1 significantly enriched (Fisher's exact; q-value = 0.006). In STAD, meRAD51C was strongly associated with the Epstein-Barr Virus (EBV) infection molecular signature (Fisher's exact; q-value < 0.0001), while the sole STAD case with meBRCA1 was of the chromosomal instability (CIN) subtype. MeBRCA1 BC was enriched in the basal subtype (Fisher's exact; q-value < 0.0001) whereas meRAD51C BC did not exhibit specific subtype enrichment. Despite lacking statistical significance, all four meBRCA1 UCEC cases belonged to the p53-mutant copy number (CN) high subtype.
We next explored clinical features (age, sex, ancestry and tumour stage) and examined their association with meBRCA1 and meRAD51C. We observed that patients with meBRCA1 OV were marginally younger than those without meBRCA1 and meRAD51C (Figure 2A, Supplementary Table S4, Two-tail Wilcoxon; q-value = 0.007). This association was confirmed in TCGA OV cohort (Supplementary Figure S12a, Two-tail Wilcoxon; P-value < 0.0001). Among patients diagnosed with STAD, we found that primary tumours from male patients were more likely to carry meRAD51C in comparison to those from female patients (Figure 2B, Supplementary Table S5, Fisher's exact; P-value = 0.02). We observed differences in ancestry distribution for BC, where methylated cases were more frequently in patients with African ancestry (meBRCA1: 28.5%, meRAD51C: 44.4%) compared to those with no promoter methylation (no methylation of either 13.1%) (Figure 2C). As such, we observed that 10.9% of BC cases in patients with African ancestry have BRCA1 or RAD51C methylation compared to 4.2% of all BC cases (without accounting for ancestry). This is consistent with the high prevalence of triple-negative breast cancer (TNBC) subtype in individuals with African ancestry (25) (Supplementary Figure S12b) and is of potential clinical importance due to worse outcomes associated with TNBC (26). In contrast, we found no enrichment of methylated cases by cancer stage, across any cancer type assessed (Figure 2D).
Figure 2. Clinical characteristics of patients with primary tumours exhibiting BRCA1 and/or RAD51C promoter methylation. Only eight cancer types with more than two cases showing methylation in either BRCA1 or RAD51C are included. Four characteristics were (A) age, (B) sex, (C) ancestry and (D) stage. (A) Age boxplots show the distribution of age within a group. For groups with less than three samples, a dotplot is used. The boxplots are composed of the median (center bar) and IQR, with the whiskers extending to the maxima and minima, but no further than 1.5 × IQR. The outliers, defined as values beyond 1.5 × IQR, are represented by dots above or below box. A two-tail Wilcoxon test with Benjamini-Hochberg correction was used to compare between cancer types with at least three samples with BRCA1 or RAD51C methylation. (B) Sex the percentage of patients identified as female (in orange) or male (in blue) is delineated based on the methylation status of BRCA1 and RAD51C. TGCT, OV, UCEC, and CESC are excluded as they consist of only a single sex. (C) Ancestry the percentage of patients' ancestry, with colours indicating different ancestral origins as inferred from SNP arrays. The colour code is as African (afr): orange, African admix (afr_admix): light orange, American (amr): green, East Asian (eas): yellow, European (eur): blue, European admix (eur_admix): light blue, South Asian (sas): purple, South Asian admix (SAS_admix): light purple, black, missing grey. (D) Stage at diagnosis, LGG, UCEC and GBM are excluded as stage information is unavailable. Samples are grouped by stage and colours indicate the methylation status (meBRCA1: green, meRAD51C: blue, meBRCA1&meRAD51C: black, not methylated in either of the grey). Abbreviations are high-grade serous ovarian cancer (OV), testicular germ cell tumour (TGCT), stomach adenocarcinoma (STAD), low-grade glioma (LGG), breast cancer (BC), uterine corpus endometrial carcinoma (UCEC), colon adenocarcinoma (COAD), head-neck squamous cell carcinoma (HNSC).
To further explore the clinical associations of meBRCA1 and meRAD51C, we analysed the overall survival (OS) in the methylated and unmethylated cases (Supplementary Figure S13). Only subtypes enriched with meBRCA1 or meRAD51C were included in this analysis to reduce the potential confounding effect from molecular or histological subtypes. We did not observe differences in OS between methylated and unmethylated cases across the evaluated cancer types—OV (HRD subtype), testicular mixed germ cell tumour, STAD (EBV subtype), LGG, BC (TNBC subtype) and UCEC (CN high subtype). However, due to the low methylated case numbers and diverse treatment and pre-treatment histories, this analysis was likely underpowered.
To investigate the potential connection between meBRCA1 and meRAD51C and previously described global CpG island methylation patterns in cancer (17), we assessed methylation of CpG island-associated regions globally in each cancer case. CIMP is an epigenetic signature extensively explored in STAD (27) and LGG (28). The characterisation of this signature typically involves clustering of CpG-associated methylation probes (17,27,28). Using the clustering from Yates et al. (17), we found a strong enrichment of meRAD51C in CIMP+ samples in STAD, which were mostly EBV+ (Figure 3A, Supplementary Table S6, Fisher's exact; q-value < 0.0001). This association was absent in LGG, as the prevalence of CIMP+ samples in meRAD51C LGG (85%) closely mirrored that of the entire LGG cohort (82%).
Figure 3. Global methylation and BRCA1/RAD51C promoter methylation. The prevalence of the CpG island methylator phenotype (CIMP) (A) or distinct global methylation clusters (B) in meBRCA1 or meRAD51C samples in six cancer types. The cancer types with more than two cases methylated in either BRCA1 or RAD51C are included. The percent of patients are categorised into dark green (CIMP phenotype or higher global CpG island methylation), green (intermediate global CpG island methylation), and light green (absence of CIMP phenotype or lower global CpG island methylation) within their respective primary cancer cohorts. Abbreviations are stomach adenocarcinoma (STAD), low-grade glioma (LGG), high-grade serous ovarian cancer (OV), testicular germ cell tumour (TGCT), breast cancer (BC), uterine corpus endometrial carcinoma (UCEC).
While the presence of CIMP in OV, TGCT, BC and UCEC has not been described (17), subgroups with high global CpG island patterns of CpG methylation have been reported (17). In BC, both meBRCA1 (Fisher's exact test; q-value < 0.0001) and meRAD51C (Fisher's exact test; q-value = 0.003) were found to be associated with intermediate levels of global CpG island methylation (Figure 3B). In contrast, we observed no statistically significant enrichment of specific global CpG island methylation profiles in meBRCA1 or meRAD51C for OV, TGCT, and UCEC (Supplementary Table S6). Nonetheless, the majority of meBRCA1 OV (89%), meRAD51C OV (67%) and meBRCA1 TGCT (80%) exhibited relatively higher global CpG island methylation levels compared to the non-methylated samples within their respective cancer types (Figure 3B).
To determine if meBRCA1 and meRAD51C is directly associated with somatic mutation of cancer driving genes, we explored which genes were co-mutated with meBRCA1 and meRAD51C. We compared the frequency of non-synonymous coding mutations in 1 025 cancer-associated genes from OncoKB (29,30) between methylated and unmethylated cases (Supplementary Table S7). By examining aggregated mutation frequencies across all cancer types, we found that TP53 mutations were more prevalent across all cancer types in the meBRCA1 subgroup, with 66% of methylated cases harbouring a TP53 mutation compared to 43% of unmethylated cases (Supplementary Figure S14a). TP53 mutations were found to be less frequent in the meRAD51C subgroup (33%) compared with the unmethylated subgroup (43%) across all cancer types (Supplementary Figure S14b). However, across all cancer types no other mutated genes were found to be significantly enriched in either the meBRCA1 or meRAD51C subgroups.
To determine if mutations were associated with meBRCA1 or meRAD51C within specific cancer types, we narrowed our analysis to the six cancer types with at least four cases of meBRCA1 or meRAD51C. TP53 mutations were detected in all cases of OV and UCEC cancers with meBRCA1, as well as in most cases (83%) of meBRCA1 BC (Figure 4A), however enrichment was not significant, likely due to small sample number. In contrast, TP53 mutations were significantly less frequent in the meRAD51C STAD cases compared to the unmethylated cases (Figure 4B, 12.5% vs. 52.2%; Fisher's exact; q-value = 0.049). In LGG and BC, the prevalence of TP53 variants was similar between the meRAD51C and unmethylated groups (Figure 4B). TP53 variants were not detected in the TGCT cohort, regardless of BRCA1 and RAD51C status.
Figure 4. A comparison of somatic mutation frequencies in 1 025 cancer genes between samples with no methylation or BRCA1/RAD51C promoter methylation. (A) The frequency of somatic non-silent mutations in 1 025 cancer genes (OncoKB™) in samples with BRCA1 promoter methylation (x-axis) and samples lacking promoter methylation (y-axis). Cancer types with more than three cases of methylated BRCA1 are shown. (B) The frequency of somatic non-silent mutations in 1 025 cancer genes (OncoKB™) in samples with RAD51C promoter methylation (x-axis) and samples lacking promoter methylation (y-axis). Cancer types with more than three cases of methylated RAD51C are shown. Each data point is colour-coded based on its p-value from Fisher's exact test, adjusted using the Benjamini-Hochberg correction. Red data points indicate q-values < 0.05, signifying statistical significance, while grey data points indicate no significance. (C) The cancer subtype and somatic non-synonymous coding variants of TP53, PIK3CA, IDH1 and ARID1A in BRCA1/RAD51C methylated cases. Cases are grouped into six cancer types, each bar shows profile of one case. Cases are ordered by their status of promoter methylation of BRCA1 (green), RAD51C (blue) or co-occurring BRCA1 and RAD51C (black). The global CpG methylation clustering of each case within their cancer types are colour coded into high (dark green), intermediate (green) and low (light green). Cancer subtypes of testicular germ cell carcinoma (TGCT), stomach adenocarcinoma (STAD) and low-grade glioma (LGG), breast cancer (BC) and uterine corpus endometrial carcinoma (UCEC) are colour coded for each cases accordingly. The subsequent coloured boxes indicate the presence or absence of somatic non-synonymous coding variants of TP53, PIK3CA, IDH1 and ARID1A. Colour of the boxes corresponding to the types of non-silent somatic variants predicted, while black lines represent pathogenic (P)/likely pathogenic (LP) variants from ClinVar. STAD Epstein-Barr virus infected (EBV), chromosomal instability (CIN), and microsatellite instability (MSI). LGG IDH mutated with hemizygous codeletion of chromosome arms 1p and 19q (IDHmut codel), IDH mutated without codeletion (IDHmut non-codel), and IDH wild type (IDHwt). BC basal, HER2 positive (Her2), luminal A (LumA), luminal B, and normal-like (Normal). UCEC copy number high (CN HIGH) and microsatellite instability (MSI).
For the meRAD51C STAD cases, PIK3CA and ARID1A mutations were significantly enriched compared to unmethylated samples (Figure 4B, Fisher's exact; q-value = 0.001 and 0.049, respectively). Corroborating the previous observation that RAD51C is not differently expressed between IDH1 mutant LGG and IDH1 wildtype LGG (31), IDH1 variants were found at similar high frequency between meRAD51C group (85%) and the unmethylated group (77%). In addition, we found IDH1 mutants to be exclusive to LGG in the meRAD51C cases across all cancer types.
To summarise, within BC, we observed an association of meBRCA1/meRAD51C with intermediate global CpG island methylation patterns, basal subtype, and the presence of TP53 mutations (Figure 4C). In the case of STAD, our findings corroborate previous reports (27), with the co-occurrence of the CIMP subtype (high methylation), EBV infection, and PIK3CA mutations in meRAD51C samples (Figure 4C).
We assessed if HRR gene promoter methylation was associated with reduced expression of the corresponding HRR gene, using matched RNA-seq data from individual cases. Significantly reduced gene expression was observed for meBRCA1 (Figure 5A) and meRAD51C (Figure 5B) cases in the cancer types with at least three methylated cases. For the cancer types with lower numbers of methylated cases, the methylated cases showed a gene expression level that was below the median of the unmethylated cases. Similarly, gene expression was below the lowest quartile of the unmethylated cases for BRIP1 (Supplementary Figure S15a) and BRCA2 (Supplementary Figure S15b) methylated cases, with the exception of meBRIP1 STAD case.
Figure 5. BRCA1 and RAD51C promoter methylation associated gene expression. Gene expression of BRCA1 (A) and RAD51C (B) in samples grouped by the presence of promoter methylation. Each box plot is a cancer type with normalised mRNA gene expression (log
2CPM) on y-axis and samples on x-axis grouped by gene promoter methylation status. The boxplots are composed of the median (center bar) and IQR, with the whiskers extending to the maxima and minima, but no further than 1.5 × IQR. Outliers, defined as values beyond 1.5 × IQR, are not shown in the boxplot. All data points are depicted with dots overlaid on the boxplot. Significant gene expression differences between methylated and unmethylated samples are shown with a two-tailed Wilcoxon signed-rank test. Abbreviations: high-grade serous ovarian cancer (OV), testicular germ cell tumour (TGCT), stomach adenocarcinoma (STAD), low-grade glioma (LGG), breast cancer (BC), uterine corpus endometrial carcinoma (UCEC), lung squamous carcinoma (LUSC), thyroid cancer (THCA), pancreatic adenocarcinoma (PAAD), bladder urothelial carcinoma (BLCA), skin cutaneous melanoma (SKCM), colon adenocarcinoma (COAD), liver hepatocellular carcinoma (LIHC).
Having identified cases with promoter methylation and reduced corresponding gene expression in BRCA1, RAD51C, BRCA2 and BRIP1, we proceeded to examine the correlation between promoter methylation levels and genomic HRD biomarkers, specifically the genomic scarring (HRD score) (1,32) and COSMICv2 Signature 3 (33). HRD scores were established and validated for application in OV and BC (32), and are calculated as a sum of three loss of heterozygosity (LOH), telomeric allelic imbalance (TAI) and large-scale state transition (LST).
In samples with meBRCA1, we observed higher HRD scores (Figure 6A) and/or Signature 3 (Figure 6B) across OV, BC, UCEC, LUSC and bladder urothelial carcinoma (BLCA), when compared to cases without promoter methylation (Supplementary Table S8). Increased HRD scores and Signature 3 were also found in all meRAD51C OV and the single LUSC case, as well as a subset of BC and STAD cases. Within the HRR genes other than BRCA1 and RAD51C, we found elevated HRD scores and/or proportion of Signature 3 in the single BRCA2-methylated BC, PALB2-methylated LUSC and BRIP1-methylated LIHC cases, but not in the two BRIP1-methylated UCEC cases (Figure 6A, B). A similar trend was also observed when we reviewed the three individual components of HRD scores (Supplementary Figure S16).
Figure 6. Molecular features of homologous recombination deficiency (HRD) in cases with promoter methylation of homologous recombination repair (HRR) genes. (A) Distribution of HRD scores and (B) proportion of single nucleotide substitutions explained by COSMICv2 signature 3 grouped by the promoter methylation status in BRCA1, RAD51C, BRCA2, BRIP1, PALB2 (x-axis). Each box plot is a cancer type with HRD scores or Signature 3. A two-tailed Wilcoxon signed-rank test with Benjamini-Hochberg correction showed differences in HRD score or Signature 3 distribution in cancer types with at least three methylated samples. (C) The plot details various factors, including promoter methylation status of BRCA1 (green) and RAD51C (blue), HRD score, COSMICv2 signature 3 contribution, of the selected promoter CpG probes in cases with BRCA1 and RAD51C promoter methylation (raw: orange circle, corrected by purity and copy blue cross). The plot includes only cancer types with cases showing promoter methylation in either of the gene, however TGCT is excluded from this analysis due to the unavailability of methylation information for the matched normal tissue. The threshold for high methylation (median beta value of 0.7 post correction) is indicated with dashed line. Subsequent rows provide information on the estimated tumour purity, copy number status for the corresponding methylated gene, and the presence of loss of heterozygosity (LOH) for the corresponding methylated gene. The oncoplot shows the presence of somatic mutations in HRR genes with variant annotation of pathogenic (P) or likely pathogenic (LP) from ClinVar. Samples are ordered by HRD scores. Abbreviations: high-grade serous ovarian cancer (OV), testicular germ cell tumour (TGCT), stomach adenocarcinoma (STAD), low-grade glioma (LGG), breast cancer (BC), uterine corpus endometrial carcinoma (UCEC), glioblastoma multiforme (GBM), liver hepatocellular carcinoma (LIHC), colon adenocarcinoma (COAD), head-neck squamous cell carcinoma (HNSC), lung squamous carcinoma (LUSC), thyroid cancer (THCA), cervical squamous cell carcinoma (CESC), bladder urothelial carcinoma (BLCA).
To determine tumour-specific levels of BRCA1 promoter methylation within samples, we corrected the methylation level by tumour purity and BRCA1 gene-level copy number. We identified a subset of OV (e.g. AOCS-158, AOCS-057), BC (TCGA-D8-A1JL, TCGA-A2-A0YM, TCGA-D8-A27F and TCGA-EW-A1PB) and UCEC (TCGA-DF-A2KS, TCGA-D1-A1NW) cases that displayed relatively low tumour tissue levels of BRCA1 methylation (Figure 6C) (<70% median methylation post-correction), suggesting potential subclonal or heterozygous methylation in these cases. Previous literature has described treatment driven methylation loss of RAD51C and BRCA1 in patient-derived models (12,34,35). Therefore, we further investigated the treatment status of these meBRCA1 cases with low tumour tissue levels of methylation. These cases were all primary surgical cases. One of the two OV samples (AOCS-057) had no history of chemotherapy for other malignancies, while AOCS-158 was exposed to neoadjuvant treatment prior to the sample collection (Supplementary Table S9). For TCGA, none of the low level meBRCA1 samples reported history of other malignancy or exposure to neoadjuvant treatment. Of note, the primary samples for TCGA-DF-A2KS (UCEC) and TCGA-D8-A27F (BC) were collected on the same day as or prior to the diagnosis, which suggest a low possibility of other treatment exposure (Supplementary Table S9). We did not detect any variants likely to affect protein function in the HRR-related genes in these cases with low level meBRCA1, yet all these samples had high HRD scores suggestive of a HRD phenotype (≥42) (Figure 6C). These cases also had one or more copies of the BRCA1 gene with evidence of LOH. This supported that both high and low levels of BRCA1 methylation could be associated with HRD scarring in multiple cancer types, including OV, BC and UCEC, and could indicate a history of HRD. It also suggested that partial loss of methylation can occur during cancer development or early exposure to treatment.
For meRAD51C, we also identified a subset of cases with low tumour tissue levels of meRAD51C (defined as less than 70% median methylation post-correction). Most STAD (15 of 22) and all LGG cases showed low levels of methylation post-correction (Figure 6C), suggesting the methylation is heterozygous. These cases, as well as low-level meRAD51C colon adenocarcinoma (COAD) and HNSC, also mostly exhibited low HRD scores, absence of Signature 3 and absence of RAD51C LOH (Figure 6C). Additionally, unlike meBRCA1-low cases, BC and UCEC cases with low levels of RAD51C methylation, displayed HRD scores below the HRD threshold of 42 (TCGA-E2-A108, TCGA-D1-A1NW, TCGA-EY-A548) and lacked Signature 3. We should note, however, that a number of meRAD51C cases had low tumour purity, which may affect accurate determination of tumour methylation level. Additional complicating factors included the number of probes covering the RAD51C promoter being lower than for the BRCA1 promoter (7 probes versus 21 probes), and the observed (Supplementary Figure S3, S8 and S11) and reported (34) heterogeneity for meRAD51C being higher.
Due to the high prevalence of meRAD51C in STAD being associated with low HRD scores, we wanted to further explore this event without the issue of contaminating stroma. To achieve this, we analysed an independent dataset of 33 STAD cell-lines from DepMap (36). We detected meRAD51C in five of the cell-lines (Figure 7A), with one of the five methylated cell-lines, SNU601, exhibiting a consistently high level of methylation across all CpG sites examined (termed homozygous meRAD51C). In contrast, the other four cell-lines displayed lower and heterogeneous methylation patterns, which we defined as heterozygous meRAD51C (Figure 7A). Interestingly, the SNU601 cell-line with homozygous meRAD51C was established from a STAD sample (ascites), from a patient who had received prior chemotherapy treatment, whereas the other methylated cell-lines were established from untreated primary cancer or metastatic lymph nodes (37,38). None of the methylated cell-lines showed evidence of truncating somatic variants in HRR genes (Figure 7B).
Figure 7. RAD51C promoter methylation, global CpG island methylation and homologous recombination deficiency (HRD) in stomach adenocarcinoma (STAD) cell-lines. (A) Methylation levels of eight individual CpG site within the RAD51C promoter region are represented on a colour scale from 0 (blue) to 1 (red) in 33 STAD cell-lines from DepMap. (B) An integrated summary showcases the RAD51C promoter methylation, global methylation (CIMP) status, HRD scores with the widely used threshold (in BC and OV) of 42 highlighted with dashed line. The presence of loss of heterozygosity (LOH) and copy number of RAD51C gene region are provided. The oncoplot shows the presence of somatic variants in HRR genes (BRCA1/2, RAD51C/D, PALB2, BRIP1, PIK3CA, ARID1A and TP53). Coloured boxes indicate types of non-silent somatic variants predicted by Depmap mutation pipeline, while black lines represent pathogenic (P) or likely pathogenic (LP) variants from ClinVar. (C) Normalised mRNA expression (log2 CPM) of RAD51C (y-axis) in samples grouped as homozygous (N = 1), heterozygous (N = 4) RAD51C promoter methylation (meRAD51C) and no methylation (N = 28) (x-axis). P-values shown from a two-tail Wilcoxon signed-rank test. (D) HRD scores (y-axis) in samples grouped as homozygous, heterozygous meRAD51C and cases with no meRAD51C detected (x-axis). (E) Olaparib sensitivity in 27 cell-lines (data unavailable for TGBC11TKB, RERFGC1B, GCIY, SNU520, NUGC2 and OCUM1). The changes of cell viability (y-axis) with Olaparib concentration (x-axis) are described with samples grouped primarily by RAD51C promoter methylation status (homozygous RAD51C promoter dark blue; heterozygous RAD51C promoter blue) and then HRD scores (no RAD51C promoter methylation and HRD score greater than or equal to pink; no RAD51C promoter methylation and HRD score smaller than grey).
Consistent with findings in the TCGA STAD cohort, we observed that meRAD51C cell-lines exhibited a high global CpG island methylation pattern (CIMP subtype) and an enrichment of PIK3CA non-silent variants (Figure 7B & Supplementary Figure S17; CIMP Fisher's P-value < 0.0001, PIK3CA Fisher's P-value = 0.0017). However, none of the meRAD51C cell-lines carried somatic ARID1A variants, contrasting with the TCGA STAD cohort's results.
Heterozygous meRAD51C samples were associated with reduced RAD51C gene expression compared to those with no methylation (Figure 7C, two-tail Wilcoxon P-value = 0.0029). Homozygous meRAD51C had a greater reduction in RAD51C expression (Figure 7C) and a higher HRD score (39 versus 2–17; Figure 7B–D), when compared to the samples with heterozygous meRAD51C. Although the HRD score for the homozygous meRAD51C sample fell below the conventional threshold of 42 (Figure 7D) commonly used to predict HRD in BC and OV, the HRD score has not been validated for STAD (39). To determine the clinical consequence of meRAD51C in STAD cell lines, we assessed response to Olaparib (PARPi). SNU601 with homozygous meRAD51C displayed a greater sensitivity to Olaparib compared to the heterozygous meRAD51C cell-lines (Figure 7E). Furthermore, our results revealed that the conventional HRD score threshold did not align with Olaparib sensitivity in STAD; only one out of 14 unmethylated cell-lines with an HRD score greater than 42 displayed a higher sensitivity to Olaparib than SNU601 (Figure 7E), for which we did not find a genomic explanation.
While the impacts of meBRCA1 and meRAD51C have been extensively studied in OV and BC (12,34,35,40,41), their implications in other cancer types remain largely unexplored. In this study, we characterised meBRCA1 and meRAD51C and associated molecular and clinical features across multiple solid cancer types. We demonstrated variable prevalence of these two methylation events across cancer types, ranging from rare (e.g. BRCA1 methylation in lung cancer) to highly prevalent (e.g. RAD51C methylation in STAD and BRCA1 methylation in individuals of African ancestry with BC). Our findings suggest the two epigenetic alterations, meBRCA1 and meRAD51C, stand as distinctive molecular events that vary across cancer type, dependent upon molecular subtype, with distinct methylation profiles, molecular consequence, and associated driver events.
BRCA1 methylation was detected in several cancer types, albeit as a rare event. Only OV, BC and UCEC had a BRCA1 methylation frequency of 1% or more, which was further enriched in specific molecular subgroups. In BC, meBRCA1 was enriched in TNBC, as expected (9). This corresponded to an enrichment of both meBRCA1 and meRAD51C, observed in >10% of individuals of African ancestry with BC, where the triple-negative subtype is more prevalent (42), with potentially important clinical implications, as it may contribute to the observation of a poorer prognosis in this context (42).
We also identified BRCA1 methylation in UCEC, with all cases found in the p53-mutant copy number (CN) high subgroup, which is a subgroup correlated with unfavourable clinical outcomes (43). To our knowledge this is the first report of BRCA1 methylation in endometrial carcinoma, with one previous report of BRCA1 methylation in uterine leiomyosarcoma (44). We identified a consistent correlation between BRCA1 methylation and genomic biomarkers for HRD across all cancer types with BRCA1 methylation. The concurrent presence of TP53 mutations was also observed, known to be a key element for the survival of cells with inactivated BRCA1 function, as demonstrated by Xu et al. (45) and Na et al. (46). While occurrence of BRCA1 methylation was relatively rare across all cancer types, its clinical significance and potential therapeutic implications warrant further investigation. In particular, given the rise in precision-personalised medicine and the successful N-of-1 treatment approach demonstrated in the I-PREDICT study (47), BRCA1 methylation beyond OV and BC should be explored as a potential biomarker for PARPi sensitivity, particularly in the CN high endometrial subgroup, where we observed BRCA1 methylation at >3% frequency. Furthermore, existing HRD genomic scarring biomarkers may not be appropriate for cancers with frequent genomic instability not caused by HRD, due to reduced specificity. As such, tumour-corrected promoter methylation testing of HRR genes may offer a more suitable direct readout of the HRD status.
RAD51C methylation was observed more broadly across cancer types, especially in those characterised by the distinctive CpG island methylator phenotype, CIMP. However, unlike for BRCA1 methylation, the relationship between RAD51C methylation and the HRD biomarker, was less straightforward. Whilst the previously observed (34,35,48) association between RAD51C methylation and HRD was validated for OV and TNBC, it was absent in other cancer types. With the exception of OV and TNBC, RAD51C methylation was also generally detected at a low tumour level (<70%, suggestive of heterozygous or subclonal status) and without accompanying LOH or HRD scarring. Furthermore, in stomach cancer, RAD51C methylated cases were characterised by mutual exclusivity with TP53 mutations, the opposite trend to BRCA1 methylation across multiple cancer types. Altogether, these observations point to different roles for BRCA1 and RAD51C methylation in cancer development, progression and therapeutic potential.
Promoter methylation of both BRCA1 and RAD51C has previously been reported as a constitutional event, likely arising during early embryogenesis before germ layer formation (49). Additionally, constitutional methylation is more frequently detected in OV and BC cases compared with controls (50,51), and is enriched in early-onset OV (52) and BC (10,50). In our analysis, we observed younger age at diagnosis for patients with BRCA1-methylated OV in two independent cohorts, confirming previous reports (53,54). Collectively, these observations support the contribution of constitutional methylation of BRCA1 and RAD51C to OV and TNBC development.
In other non-CIMP-driven cancer types, including other subtypes of BC, it is not clear whether the observed tumour BRCA1 or RAD51C methylation arose from constitutional methylation or from an early cancer-acquired event. We hypothesise that it is the former, but for other cancer types, meRAD51C may not result in a second LOH hit, as has been documented for OV and TNBC (13,14,41). Hence, cancer cells with heterozygous RAD51C promoter methylation likely retain HRR function, without the acquisition of the HRD phenotype. Taken together, we propose a model where the variation in meBRCA1 and meRAD51C methylation levels and impact on HRD can be explained (Figure 8). On the other hand, in EBV-driven CIMP+ stomach cancer, RAD51C methylation is likely a consequence of the global CpG island methylation (55,56). Interestingly, RAD51C promoter was almost ubiquitously methylated in CIMP+ stomach cancer cases, but mostly at low tumour methylation level (likely heterozygous), suggesting that homozygous RAD51C methylation may offer a growth disadvantage in the stomach cancer development. Therapeutic relevance in this context is less clear, perhaps requiring combination, rather than single-agent PARPi therapy to induce tumour response.
Figure 8. Proposed mechanism for variation in meBRCA1 and meRAD51C methylation levels and impact on homologous recombination deficiency (HRD). The initial event of methylation may occur on one of the alleles. This could be a result of aberrant de novo methylation acquired as part of the epigenetic reprogramming during early embryogenesis (before germ layer formation), which has also been termed constitutional methylation. Alternatively, methylation could be established as part of the global promoter hypermethylation phenotype in cancer, for example in association with EBV infection. At this stage, the cell remains competent in homologous recombination repair. Subsequently, in some cases, a loss of heterozygosity ensues, leading to methylation in both alleles, resulting in an HRD phenotype and the accumulation of HRD genomic scarring. Finally, selection pressures, such as treatment exposure, can induce partial loss of methylation and consequently rescuing homologous recombination repair functionality, leading to PARP inhibitor resistance. Created with BioRender.com
Since HRD scarring was an infrequent event for RAD51C methylation beyond OV and TNBC, meRAD51C is unlikely to be relevant for most cancer types as a PARPi single-agent treatment biomarker. However, we did observe a subset of RAD51C methylated stomach cancer cases with LOH and evidence of HRD-associated scarring (albeit not at the HRD score threshold of 42). We also observed in vitro PARPi sensitivity for the homozygous meRAD51C cell-line compared with heterozygous meRAD51C cell-lines. Notably, the homozygous RAD51C methylated cell line was the only one established from a post-treatment sample, also containing LOH of RAD51C. Exploration of RAD51C methylation in the context of LOH and treatment-exposure would aid further characterisation with regard to its gene silencing effects, impact on HRD phenotype and therapeutic relevance across cancer types.
In addition to the initial establishment of BRCA1 and RAD51C methylation, several studies have reported that complete or partial methylation loss, contributes to the development of resistance to platinum chemotherapy and PARPi (12,34). In OV, frequent methylation has been reported in cases treated with three or more cycles of chemotherapy (41). In this study, we identified a subset of primary OV and TNBC cases exhibiting low tumour level methylation in BRCA1. While we were unable to determine if all these samples were chemo-naïve, a recent report has documented two of 11 chemo-naïve TNBC cases as having low level BRCA1 methylation (57). Considering the implication of BRCA1 methylation loss for PARPi resistance (12,58), this observation raises crucial questions relevant for further exploration. Follow-up studies examining the frequency of low tumour BRCA1 or RAD51C methylation in the chemo-naïve setting will be critical for understanding early treatment resistance in OV and BC.
The complexity of the tumour methylation level analysis, which is reliant on accurate tumour purity and copy number estimation, is a known limitation of BRCA1 and RAD51C methylation characterisation studies. Moreover, the limited coverage of the RAD51C promoter region on methylation arrays and the presence of the heterogeneous RAD51C methylation pattern, further complicates assessments of methylation level. Additionally, refinement and validation of methods to assess HRD genomic scarring in non-OV/BC types is needed to gain a comprehensive understanding of the role of BRCA1 and RAD51C methylation in tumourigenesis and drug resistance. Finally, future studies should consider population size and diversity, given the rare frequency of BRCA1 and RAD51C methylation in some cancers, and differences in molecular subtype-enrichment across populations and ancestries (e.g. TNBC and CN high endometrial cancer in individuals of African ancestry (42,59) or EBV-associated stomach cancer frequencies across populations (60,61)).
In conclusion, this pan-cancer study broadens our understanding of BRCA1 and RAD51C methylation and identifies important avenues for further exploration, especially due to significant therapeutic implications. As we navigate the complexity of cancer epigenetics, the interplay between DNA methylation, histone modifications and genomic events present new opportunities for translational discoveries to drive novel therapeutic interventions and enhance patient outcomes.
Authors would like to thank all the patients included in this study. We would also like to acknowledge members of the Statistics Unit, Medical Genomics and Genome Informatics teams at the QIMR Berghofer Medical Research for their technical support. We are grateful for the members of the Australian Ovarian Cancer Study (AOCS) Group for providing additional clinical information. The results shown here are in part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. Graphical Abstract and Figure 8 were created with BioRender.com.
Author contributions: Lijun Xu: Conceptualisation, Methodology, Software, Formal analysis, Investigation, Writing – Original Draft, Writing – Review & Editing, Visualisation. Brett Liddell: Data Curation, Visualisation, Writing – Review & Editing. Ksenija Nesic: Writing – Review & Editing. Matthew J. Wakefield: Writing – Review & Editing. Clare L. Scott: Writing – Review & Editing. Franziska Geissler: Writing – Review & Editing. Lauren M. Ashwood: Writing – Review & Editing. Nicola Waddell: Conceptualisation, Writing – Review & Editing, Supervision. Olga Kondrashova: Conceptualisation, Writing – Review & Editing, Visualisation, Supervision.
Lijun Xu, Cancer Research Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; The University of Queensland, Brisbane, QLD, Australia.
Brett Liddell, Cancer Research Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; The University of Queensland, Brisbane, QLD, Australia.
Ksenija Nesic, The Walter and Eliza Hall Institute of Medical Research, Melbourne, VIC, Australia.
Franziska Geissler, The Walter and Eliza Hall Institute of Medical Research, Melbourne, VIC, Australia.
Lauren M Ashwood, Cancer Research Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; The University of Queensland, Brisbane, QLD, Australia.
Matthew J Wakefield, The Walter and Eliza Hall Institute of Medical Research, Melbourne, VIC, Australia; Department of Obstetrics and Gynaecology, University of Melbourne, Melbourne, VIC, Australia.
Clare L Scott, The Walter and Eliza Hall Institute of Medical Research, Melbourne, VIC, Australia; Department of Obstetrics and Gynaecology, University of Melbourne, Melbourne, VIC, Australia.
Nicola Waddell, Cancer Research Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; The University of Queensland, Brisbane, QLD, Australia.
Olga Kondrashova, Cancer Research Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; The University of Queensland, Brisbane, QLD, Australia; The Walter and Eliza Hall Institute of Medical Research, Melbourne, VIC, Australia.
This project used publicly available dataset from TCGA GDC portal (https://portal.gdc.cancer.gov/) and cBioPortal (https://www.cbioportal.org/). ICGC methylation data is available from Gene Expression Omnibus (GEO) under the accession GSE6582. The whole genome and transcriptome sequencing data is available from European Genome-phenome Archive (EGA) repository under the accession code EGAD00001000877. DepMap mutation and drug sensitivity data is accessible from https://depmap.org. The raw genomic data is available on the Sequence Read Archive (SRA) under the accession number PRJNA523380. The source table containing beta values per probe per patient is available from https://doi.org/10.6084/m9.figshare.25249060.v1.
The R code (R version 4.0.2) used for analyses in this study is available from https://doi.org/10.6084/m9.figshare.25249060.v1.
Supplementary Data are available at NAR Cancer Online.
L.X. is supported by the University of Queensland Graduate School Scholarship and QIMR Berghofer PhD Top-up scholarship; F.G. has received research support by the Swiss Cancer Research foundation & Swiss Cancer League [KFS-5445-08-2021]; K.N. is supported by AACR-AstraZeneca Ovarian Cancer Research Fellowship; M.J.W. and C.L.S. received funding support from the Stafford Fox Medical Research Foundation; C.L.S. is supported by a National Health and Medical Research Council of Australia (NHMRC) Investigator Grant [GNT 2009783]; N.W. is supported by a NHMRC Investigator Grant [GNT 2018244]; O.K. is supported by a NHMRC Emerging Leader 1 Investigator Grant [GNT 2008631]; This work and research were performed on QIMR Berghofer computing infrastructure supported by the Ian Potter Foundation and the John Thomas Wilson Endowment.
Conflict of interest statement. C.L.S. has served on advisory boards for Clovis Oncology, AstraZeneca, Takeda, Roche Australia, MSD, Eisai Co., has received travel support from AstraZeneca, has received in kind research support from Clovis Oncology, Roche, Beigene and funded research support from AstraZeneca, Sierra oncology and Eisai Co. N.W. is co-founder and Board member of genomiQa. O.K. has received consulting fees from XING Genomic Services and has received travel honorarium from AstraZeneca.
This project used publicly available dataset from TCGA GDC portal (https://portal.gdc.cancer.gov/) and cBioPortal (https://www.cbioportal.org/). ICGC methylation data is available from Gene Expression Omnibus (GEO) under the accession GSE6582. The whole genome and transcriptome sequencing data is available from European Genome-phenome Archive (EGA) repository under the accession code EGAD00001000877. DepMap mutation and drug sensitivity data is accessible from https://depmap.org. The raw genomic data is available on the Sequence Read Archive (SRA) under the accession number PRJNA523380. The source table containing beta values per probe per patient is available from https://doi.org/10.6084/m9.figshare.25249060.v1.
The R code (R version 4.0.2) used for analyses in this study is available from https://doi.org/10.6084/m9.figshare.25249060.v1.