Abstract
Background
Molecular alterations leading to homologous recombination deficiency (HRD) are heterogeneous. We aimed to identify a transcriptional profile shared by endometrial (UCEC), breast (BRCA) and ovarian (OV) cancers with HRD.
Methods
Genes differentially expressed with HRD genomic score (continuous gHRD score) in UCEC/BRCA/OV were identified using edgeR, and used to train a RNAseq score (ridge-regression model) predictive of the gHRD score (PanCanAtlas, N = 1684 samples). The RNAseq score was applied in independent gynaecological datasets (CARPEM/CPTAC/SCAN/TCGA, N = 4038 samples). Validations used ROC curves, linear regressions and Pearson correlations. Overall survival (OS) analyses used Kaplan–Meier curves and Cox models.
Results
In total, 656 genes were commonly up/downregulated with gHRD score in UCEC/BRCA/OV. Upregulated genes were enriched for nuclear/chromatin/DNA-repair processes, while downregulated genes for cytoskeleton (gene ontologies). The RNAseq score correlated with gHRD score in independent gynaecological cancers (R² = 0.4–0.7, Pearson correlation = 0.64–0.86, all P < 10−11), and was predictive of gHRD score >42 (RNAseq HRD profile; AUC = 0.95/0.92/0.78 in UCEC/BRCA/OV). RNAseq HRD profile was associated (i) with better OS in platinum-treated advanced TP53-mutated-UCEC (P < 0.001) and OV (P = 0.013), and (ii) with poorer OS (P < 0.001) and higher benefit of adjuvant chemotherapy in Stage I–III BRCA (interaction test, P < 0.001).
Conclusions
UCEC/BRCA/OV with HRD-associated genomic scars share a common transcriptional profile. RNAseq signatures might be relevant for identifying HRD-gynaecological cancers, for prognostication and for therapeutic decision.
Subject terms: Predictive markers, Prognostic markers, Gynaecological cancer, Translational research, Predictive medicine
Introduction
Homologous recombination (HR) is the main mechanism involved in DNA double-strand break repair [1]. Its deficiency (HRD) leads to chromosomal instability in breast cancer (BRCA), high-grade serous ovarian carcinoma (OV) and uterine corpus endometrial carcinoma (UCEC) [2–4] (acronymes as per TCGA). HRD was initially related to pathogenic, germline or somatic BRCA1/2 mutations in OV/BRCA, and more recently to multiple genomic and epigenomic alterations affecting DNA-repair pathways [1]. Most TP53-mutated UCEC (serous-like) are characterised by high copy-number alterations (CNH) [5] and previous reports suggested that half of these tumours exhibits HRD-like genomic instability profiles [4]. However, less than 10% of serous UCEC have been shown to be BRCA1/2-mutated [6, 7] or silenced [1, 4]. HRD tumours have been shown to be particularly sensitive to platinum-based chemotherapy and PARP inhibitors (PARPi) in BRCA and OV patients [8–12], strengthening the need to identify properly these tumours.
Multiple biomarkers were reported to identify HRD tumours [13]. Somatic or germline mutations in HR genes are not found in all HRD tumours and their interpretation can be difficult because of variants of uncertain significance [14] and reversion mutations [15–17]. HRD is associated with the accumulation of genomic scars in the form of large DNA copy-number alterations, specific profiles of chromosomal rearrangement, and mutational signatures [3, 13, 18]. A genomic instability score, referred in literature as a genomic (g)HRD score, is computed by the unweighted sum of three SNP array-based scores that quantify specific types of genomic scar associated with HRD [3]. A gHRD score above 42 has been associated with BRCA1-2 mutations [3], platinum sensitivity [9] and PARPi efficacy in BRCA and OV [12].
However, a gHRD score below 42 did not exclude the benefit of PARPi in several randomised trials in OV [19, 20]. This could be due to the threshold used to discriminate HR-“proficient” tumours from HRD tumours. False-positive results (i.e. high gHRD score but resistance to platinum and PARPi) may be related to the imprinted dimension of these scars in a dynamic biological system affected by HRD reversion [21].
Gene expression profiles associated with HRD were reported in OV [22, 23] and BRCA [24, 25]. However, these studies are limited by their restriction to one cancer type and the lack of external validation. Whether HRD-gynecologic cancers share a transcriptional profile is unknown. A common transcriptional signature associated with a functionally deficient HR could help to better understand the biology of HRD tumours and to identify these tumours, beyond primary origin and heterogenous genomic alterations. Such a signature would be of particular interest in UCEC, where no recurrent molecular alterations associated with HRD have been identified.
We aimed to identify whether gynaecological cancers share a common transcriptional profile associated with a high gHRD score and to identify a RNAseq signature predictive of the gHRD score for validation on independent gynaecological cancer cohorts.
Materials and methods
Patients and cohorts
Discovery gynaecologist cohorts
Data from PanCanAtlas UCEC, BRCA, and OV were used to identify genes and RNAseq signatures associated with gHRD score (https://gdc.cancer.gov/about-data/publications/PanCan-DDR-2018) [1].
Independent gynaecologist cohorts
Validation gynaecologist cohorts were selected from literature and open databases with the two following objectives: (A) accounting for the biological diversity in each cancer, and (B) exploring the association between HRD-associated transcriptional profile and (i) prognostic in patients with advanced UCEC/OV treated with platinum-based chemotherapy, and (ii) prognostic with or without adjuvant chemotherapy in patients with localised BRCA. Endometrial carcinoma: Clinical Proteomic Tumour Analysis Consortium (CPTAC) cohort patient data were used as a validation cohort for endometrioid UCEC [26]. In addition, 210 patients treated for Stage I–IV UCEC in two University Hospitals in Paris [2010–2017] were post hoc included and referred as to the CAncer Research for Personalised Medicine Institute upon written consent or non-opposition (CARPEM) cohort (ethical approvals from National Ethical Committee CPP Ile-de-France I II & IV, DC-2020/144; DC-2019-3677; DC-2009-950, Data Protection Committee approval ID APHP2020-1109121347). Among these patients, 193 for whom (i) primary tumour sample had >50% of tumour cell (pathology review), and (ii) RNAseq was successfully performed (library quality), were included (see Supplemental Method and [27] for full technical details). Breast cancer: Data from 125 TCGA patients not belonging to the PanCanAtlas were used for validation of the RNAseq signatures in BRCA. Data from 3409 BRCA included in the Sweden Cancerome Analysis Network (SCAN) cohort [2010–2015] [28] were used to validate its translational relevance (GSE96058). Ovarian cancer: Data from 210 TCGA patients not belonging to the PanCanAtlas were used for validation of the RNAseq signatures and translational validation (N = 197 with RNAseq and survival data available).
Non-gynaecological cohorts
Data from PanCanAtlas prostate, pancreatic, lung and colon cancers were used to assess whether RNAseq signatures discovered in gynaecological cancers might be associated with gHRD score in non-gynaecological cancers.
Clinical and molecular data
For in silico cohorts, data were obtained from open repositories (PanCanAtlas repository [1]; Genomic Data Common (GDC) repository for TCGA [29] and CPTAC [26, 29], Gene Expression Omnibus (GEO) for SCAN [30]).
For the CARPEM cohort (UCEC), clinical, pathology, and follow-up data were post hoc extracted from standardised electronic medical records. TP53-mutated tumours (surrogate of CNH/serous-like tumours) [31, 32] were identified as follows: (i) tumours with a deficient mismatch repair machinery (dMMR, loss of PMS2 or MSH6 expression in immunohistochemistry) were classified as dMMR, (ii) tumours with pathogenic variants in the DNA polymerase ɛ catalytic domain (targeted sequencing) were classified as POLE-mutated, (iii) among tumours without POLE mutation and dMMR, tumours with a TP53 mutation by targeted sequencing were classified as TP53-mutated. By note, one serous tumour was observed with a TP53 abnormal immunostaining and no mutation in targeted sequencing, and was finally classified as TP53-mutated as surrogate of CNH UCEC because of a high gHRD score. The remaining tumours were classified as nonspecific molecular profile (NSMP) (Supplemental Method).
RNAseq data
For TCGA/PanCanAtlas/CPTAC samples, RNAseq HTseq count files were obtained from GDC. For SCAN samples, gene-level FPKM expression data were obtained from GEO (GSE96058).
For CARPEM tumour samples, tumour RNAs were extracted from FFPE tissue block sections using the Maxwell® 16 LEV RNA FFPE Kit. PolyA-RNAseq libraries were prepared using the QuantSeq 3’mRNA-Seq Kit FWD for Illumina (LexogenTM). Libraries were sequenced on NovaSeq6000. Fastq RNAseq files were trimmed by cutadapt [33] and mapped by STAR [34]. Reads-per-gene count was performed with HTseq [35] All RNAseq data were normalised using edgeR [36] (see Supplemental Method and ref. [27]).
Homologous recombination deficiency genomic score
The score used as a measure of HRD-associated genomic instability/scars was the gHRD score, that is the unweighted sum of three measures of HRD-associated genomic scars: loss of heterozygosity, telomeric allelic imbalance, and large-scale state transitions [9].
For PanCanAtlas, gHRD scores were available in repository. For TCGA/CPTAC samples, allele-specific copy-number segment files (available at GDC) generated from Affymetrix SNP 6.0 or whole-genome sequencing (WGS) were used for gHRD score calculation.
For CARPEM cohort patients, in order to generate data complementary to the CPTAC-UCEC cohort (that accounts only for endometrioid tumours), and because of sample availability constraints, only TP53-mutated and/or non-endometrioid and/or endometrioid high-grade and/or Stage IV tumours were considered for gHRD score estimation (N = 69 tumours/patients). Tumour DNA was extracted from macrodissected FFPE tissue block sections using the Maxwell® 16 FFPE Plus LEV DNA Purification Kit (PromegaTM, France). Copy-number alterations were estimated using the OncoScan FFPE Assay Kit (Thermo Fisher Scientific) on 80 ng of FFPE-extracted DNA. CEL files were normalised using the EACON R package [37]. LogR/BAF data generated were used for allele-specific segmentation using ASCAT [38]. All segmented data were finally used for gHRD score calculation using the scarHRD R package [39].
Statistical analyses (Fig. 1)
Fig. 1. Analytical workflow.
HRD homologous recombination deficiency. PanCancAtlas: https://gdc.cancer.gov/about-data/publications/pancanatlas. GO gene ontology, CPTAC Clinical Proteomic Tumour Analysis Consortium, Spl. samples, CARPEM CAncer Research for Personalised Medicine, SCAN Population-Based Multicenter Sweden Cancerome Analysis Network—Breast Initiative (GEO: GSE96058), TCGA The Cancer Genome Atlas. *Breast and ovarian cancer samples considered are TCGA samples not overlapping PanCancerAtlas. Numbers indicate the number of samples with RNAseq data available (among which 90, and 207 samples had segment copy-number variation data available for TCGA breast cancer and TCGA ovarian cancer, respectively).
Differential gene expression analyses (DGEA) were performed to identify genes differentially expressed with gHRD score as a continuous variable. DGEA were performed independently for each gynaecological tumour type. Genes commonly up/downregulated with higher gHRD score were identified by overlapping lists of differentially expressed genes. A conservative threshold of significance (false-discovery rate <0.05) was used to define a gene as significantly differentially expressed, considering the finding of a gene in three independent analyses as a stronger validation than individual significance.
Exploration of the biological meaning of differentially expressed genes sets was performed using gene ontology enrichment analyses (GOEA) using edgeR [36], ViSEAGO [40], and the PANTHER Overrepresentation Test tool (http://geneontology.org/). As for DGEA, a conservative threshold of significance (false-discovery rate <0.05) was used to define a GO as significantly enriched.
Gynaecological PanCanAtlas RNAseq training data were used to train a 50-fold cross-validated penalised ridge-regression models (R function: cv.glmnet; hyperparameter: alpha = 0; penalisation parameter Lambda = λmin) predictive of gHRD scores (as continuous variable) [41]. Genes found commonly up or downregulated across all PanCanAtlas datasets (UCEC, OV, BRCA) were used as input of variable lists for training two ridge-regression models (one model from upregulated genes, and one model from downregulated genes). A third linear regression model was trained from the two previous models for regression on the gHRD score. A continuous score (RNAseq profile) was computed for each sample using trained models. Performances of the RNAseq profile to fit gHRD score were estimated by linear regression and Pearson correlation. Performances of the RNAseq profile to identify tumours with gHRD score >42 [9] were estimated by the area under ROC curve analyses (AUC). Threshold identification was performed in training data based on ROC curves.
Associations between qualitative/continuous variables were analysed using Student/Wilcoxon/ANOVA test. Associations between continuous variables were analysed using linear regression/Pearson correlations. Associations between categorical/binary variables were assessed using chi-square/Fisher’s exact test. For in silico cohorts, the primary outcome of interest was the overall survival (OS). For the CARPEM cohort, cause of death being known, disease-specific survival (DSS) (survival until death related to UCEC) was used. Associations between patient/cancer characteristics and DSS/OS were assessed by Cox regression to estimate hazard ratio (HR) with [95% confidence interval (95%CI)]. Survivals were analysed by Kaplan–Meier curves (with median (interquartile range)), and compared using the log-rank test. The potential of RNAseq profiles to predict treatment benefit was estimated using an interaction test [42] (stratified on propensity to receive the treatment in the context of non-randomised cohorts [43]). All test were two-sided. Significance was defined by P < 0.05. Analyses were performed using R v4.
Results
Homologous recombination-deficient gynaecological cancers share a common transcriptional programme
In the PanCanAtlas, a continuous gradient of gHRD scores was found in OV and basal BRCA rather than a clear bimodal distribution (Fig. 2a). In UCEC, only CNH tumours exhibited a significant proportion of tumours with gHRD score above 42 (N = 38/161, 24%) (Fig. 2a). Deleterious BRCA1/2 mutations (OncoKB annotation (cbioportal.org)) were observed with high gHRD scores in BRCA and OV (Fig. 2b). In UCEC, 43/547 tumours exhibited deleterious BRCA1/2 mutations, mostly in POLE-mutated tumours (N = 32) and microsatellite-instable tumours (N = 10) and were not associated with high gHRD scores (Fig. 2b). Loss of PTEN or ARID1A (truncating mutations/homologous deletion), potentially associated with response to PARPi [44, 45], were not associated with higher gHRD scores (Supplementary Fig. S1).
Fig. 2. Homologous recombination deficiency-associated transcriptional programme.
a Distribution of genomic homologous recombination deficiency (gHRD) scores in gynaecological cancers (PanCanAtlas data, density distribution plots). UCEC/BRCA/OV uterine corpus endometrial carcinoma, breast cancer, and ovarian (high-grade serous) cancer. POLE/MSI/CN_LOW/CN_HIGH: endometrial carcinoma molecular subgroups 2013). LumB/LumA/Her2/Basal: breast cancer intrinsic subtypes. b Distribution of gHRD score according to BRCA1/2 mutation in gynaecological cancers (deleterious mutations—OncoKB annotation on cbioportal.org data). c Genes commonly up/downregulated with genomic HRD scores in PanCanAtlas gynaecological cancer datasets. d Gene ontology terms (GO) commonly enriched in genes up/downregulated with genomic HRD scores. See Supplemental Data 1. e, f Linear association between mean expression values of HRD-associated genes (upregulated genes: nuclear genes (e); downregulated genes: cytoplasmic genes (f)) and genomic HRD scores. R²/slope/p computed using linear regression models.
When clustering tumours based on transcriptome data, a continuous gradient of similarity between tumours with low- and high gHRD scores was observed rather than clearly distinct clusters (Supplementary Fig. S2A, B). DGEA identified 271 and 385 genes commonly up- and downregulated in BRCA, OV and UCEC, according to gHRD (Fig. 2c and Supplementary Fig. S2C, D). Shared upregulated genes were enriched for nuclear architecture, chromatin, DNA repair and endoplasmic reticulum GO. Downregulated genes were enriched for axoneme, cytoskeleton, and cell mobility/projections GO (Supplementary Fig. S2E and Supplementary Data 1). Among GO terms enriched in differentially expressed gene sets, 84 and 67 GO terms were found commonly enriched in up- and downregulated gene sets and supported the over-representation of GO terms related to nuclear processes/structures versus cytoplasmic processes/structures, respectively (Supplementary Data 1). This observation led to refer to these gene lists as “nuclear genes” and “cytoplasmic genes”.
A transcriptional burden, as the mean in gene expression levels of all nuclear versus all cytoplasmic genes, was estimated by averaging Z score-transformed gene expression values of each gene list: two continuous variables were observed, with linear associations between higher burden of nuclear gene expression and lower burden of cytoplasmic gene expression along with higher gHRD score (Fig. 2e, f). These associations were not found when randomly selecting genes (10,000 random permutations) (Supplementary Fig. S3).
A transcriptional signature predictive of homologous recombination deficiency (training)
Two ridge-penalised regression models were trained from PanCanAtlas data to predict gHRD score from gene expression levels of nuclear (nuclear model) and cytoplasmic (cytoplasmic model) genes (Supplementary Fig. S4). The computed predictions were gathered as a two-dimensional regression model to estimate a unique RNAseq score predictive of gHRD score (R² = 0.82, P < 2.10−16, Fig. 3a). The RNAseq score accurately predicted the HRD phenotype (gHRD score >42) in gynaecological cancers as a whole (OV, BRCA and UCEC) (AUC = 0.96) and separately in each localisation (Fig. 3b, c). A common threshold was selected on whole gynaecological training data to identify tumours with gHRD score >42 based on RNAseq score (defining a RNAseq HRD profile). Because of a distinct distribution of gHRD score in OV with HRD values, a specific threshold was selected in this subset for defining OV with RNAseq HRD profile with improved specificity (Fig. 3c—blue threshold).
Fig. 3. RNAseq signature associated with HRD: RNAseq HRD profile.
Training data. a Association between RNAseq-derived nuclear score, cytoplasmic score, and genomic HRD score. A 2-variable linear regression model based on RNAseq nuclear score and cytoplasmic score has been fit to predict the genomic HRD score. Plan: graphical projection of the 2-variable planar regression model (R² = 0.82, P < 2.10−16). b, c Receiver operating characteristic (ROC) curves: predictive performance of the RNAseq score to predict a genomic HRD score >42 in gynaecological cancers. Dashed line: threshold selected in the RNAseq score distribution based on sensitivity/specificity in all gynaecological cancers (training data). Blue dashed line: additional threshold for ovarian cancers to improve specificity in detecting ovarian cancers with genomic HRD score >42. AUC area under the curve. For colour figure, please refer HTML version.
In prostate, pancreatic, lung squamous, lung adenocarcinoma and colon carcinoma cohorts, the RNAseq score showed also significant correlations with gHRD score in these cohorts (Supplementary Fig. S5). BRCA1/2 deleterious mutations were not associated with a higher gHRD score, despite the low rate of mutations limited definitive conclusion (Supplementary Fig. S6).
Validation of the RNAseq score in gynaecological cancers (independent samples)
Gathering the CPTAC cohort (N = 116 UCEC) and the CARPEM cohort (N = 69 UCEC) (Supplementary Table S1), RNAseq score was highly correlated with gHRD score and highly predictive of HRD score >42 in UCEC (Fig. 4a, b and Supplementary Fig. S7). In the 193 UCEC patients of the CARPEM cohort with RNAseq available, half of the TP53-mutated UCEC is shown with RNAseq HRD profiles (Fig. 4c), and none but two non-TP53-mutated UCEC showed similarly high RNAseq score (two clear-cell carcinomas).
Fig. 4. Validation of the RNAseq HRD profile in gynaecological cancers (independent samples).
Validation for endometrial (a–c, CPTAC and CARPEM cohorts), breast (d, e, TCGA and f, SCAN cohort) and ovarian (g–i, TCGA cohort) cancers. a, d, g Association between the RNAseq score and genomic HRD score. R²/slope/p computed using linear regression models. r: Pearson correlation coefficient. Red dashed line at genomic HRD score = 42. Black dashed line at the threshold calibrated in the TCGA dataset at indicative of a genomic HRD score >42. Blue dashed line at the ovarian cancer-specific threshold. b, e, h ROC curves: predictive performance of the RNAseq score to predict a genomic HRD score >42. AUC area under the curve. c RNAseq score across endometrial molecular subgroups (CARPEM cohort). POLE POLE-mutated tumours, dMMR mismatch repair-deficient tumours. TP53-mut tumours with TP53 mutation and without POLE or dMMR, NSMP tumours with no specific molecular profile. f Association between RNAseq score and breast cancer subgroups in the SCAN cohort (FPKM data). HER2 HER2-positive cancers, ER+ oestrogen-receptor-positive cancers. TN triple-negative cancers. Black dots: mean. ANOVA analysis of variance. (For the computation of the RNAseq HRD score in the SCAN cohort, please refer to Supplementary Fig. S7). Because of the distinct RNAseq score scale due to FPKM data, no threshold is depicted on the panel. i RNAseq score in BRCA1/2-mutated or wild-type ovarian cancer in the TCGA cohort (pathogenic/likely pathogenic mutation as per OncoKB annotation (cbioportal.org annotation)). For colour figure, please refer HTML version.
In TCGA BRCA samples, RNAseq score also correlated with gHRD scores and accurately predicted gHRD score >42 (Fig. 4d, e). Of note, only 575 genes were common between the RNAseq signature and gene expression data of the SCAN dataset, leading to train an additional model with equivalent predictions for external application in this cohort (Supplementary Fig. S8). RNAseq scores were significantly higher in basal-like (triple-negative) and HER2-positive than in oestrogen-receptor-positive tumours (including luminal A and B tumours) (Fig. 4f; gHRD scores unavailable).
In TCGA OV samples not belonging to the PanCanAtlas, RNAseq score also correlated with gHRD scores but to a lesser extent (R² = 0.4) and also predicted gHRD score >42 but with lower accuracy (AUC = 0.78) (Fig. 4g, h). All but one tumour with pathogenic BRCA1/2 mutation were shown with RNAseq HRD profile. This last tumour showed a gHRD score of 56 and a BRCA1 loss-of-function alteration (p.Ser1217Argfs*21, with shadow deletion). Reversely, one BRCA1-mutated tumour (p.Cys47Trp, predicted loss-of-function [46], with shadow deletion), but with a gHRD score of 29, was predicted with a RNAseq HRD profile.
Translational potential of the RNAseq score in gynaecological cancers
Within the UCEC CARPEM dataset (N = 193), 27 patients were found with RNAseq HRD tumours (Supplementary Table S2A and Supplementary Fig. S7). In advanced (FIGO Stage III–IV) TP53-mutated tumours treated with platinum-based chemotherapy (N = 25), both RNAseq HRD profile and gHRD score >42 were associated with better survival, potentially reflecting higher platinum sensitivity (Fig. 5a and Supplementary Fig. S7E), despite the higher proportion of Stage III tumours among tumours with RNAseq HRD profile may contribute to this better prognosis (Supplementary Table S2B).
Fig. 5. Translational applicability of the RNAseq HRD profile in gynaecological cancers.
Kaplan–Meier curves. p: log-rank P value. a OS in patients treated with chemotherapy for Stage III–IV TP53-mutated endometrial carcinoma (CARPEM cohort, N = 25 patients). b OS in patients with Stage III–IV ovarian cancer (TCGA cohort). FIGO International Federation of Gynaecologist and Obstetrics 2010 staging system (N = 197 patients with RNAseq and survival data available). c OS in patients with Stage I–III breast cancer according to RNAseq HRD profile and adjuvant chemotherapy administration (SCAN cohort) (for computation of the RNAseq HRD score in the SCAN cohort, please refer to Supplementary Fig. S7).
Consistently with observation in advanced TP53-mutated UCEC, in 197 TCGA patients with OV treated with platinum-based chemotherapy (Supplementary Table S3), the RNAseq HRD profile was significantly associated with better OS (Fig. 5b) (adjusted on Stage III versus Stage IV: HR = 0.62 [0.44, 0.88]). Exploratory analyses showed that tumours identified with a HRD phenotype according to both gHRD score and RNAseq (true positive) were found with the best OS (median 57 months [44; 81]) compared to true negative tumours (both gHRD and RNAseq score negative) but also to tumours with discordant results between genomic and RNAseq scores (Supplementary Fig. S9).
In patients with Stage I–III BRCA (SCAN), the RNAseq HRD profile did not overlap entirely the triple-negative or PAM50-Basal phenotypes, despite a significant association (28% and 67% of tumours with RNAseq HRD profile being categorised as triple-negative or PAM50-Basal, respectively) (Supplementary Table S4). The RNAseq HRD profile was independently associated with poor OS (unadjusted-HR = 2.10 [1.60, 2.76]; adjusted-HR = 1.94 [1.34, 2.83], Table 1). Adjuvant chemotherapy was associated with a benefit of higher magnitude in patients with RNAseq HRD tumours (HR adjusted on breast cancer subgroup: HR = 0.27 [0.15, 0.50]) than in patients without (Fig. 5c, d and Supplementary Fig. S10) (adjusted-HR = 0.40, [0.29, 0.56]) ([RNAseq HRD profile] × [chemotherapy] interaction test: stratified on propensity score for chemotherapy administration: P = 0.024) (Table 1). The PAM50-Basal phenotype was equivalently associated with poor survival but did not reach significance for the prediction of chemotherapy benefit (interaction test, P = 0.225, Supplementary Table S5).
Table 1.
Prognostic and predictive potential of the RNAseq HRD profile in breast cancer (SCAN cohort).
| Variables | Hazard ratio [95% CI] | P |
|---|---|---|
| Multivariable model | ||
| RNAseq HRD profile (Ref: no) | 1.94 [1.34, 2.83] | <0.001 |
| Age (for each year older) | 1.08 [1.06, 1.09] | <0.001 |
| Tumour size (for each cm larger) | 1.02 [1.01, 1.03] | <0.001 |
| Positive lymph node status (Ref: N0) | 1.13 [0.89, 1.44] | 0.315 |
| Breast cancer group (Ref: ER+) | ||
| HER2+ | 1.35 [0.96, 1.88] | 0.083 |
| TN | 1.46 [0.95; 2.24] | 0.088 |
| Tumour grade (Ref: grade 1) | ||
| Grade 2 | 1.15 [0.75, 1.79] | 0.520 |
| Grade 3 | 1.81 [1.15, 2.86] | 0.010 |
| Interaction test model | ||
| RNAseq HRD profile (Ref: no) | 4.96 [3.36, 7.30] | <0.001 |
| Adjuvant chemotherapy (Ref: no) | 0.58 [0.43, 0.76] | <0.001 |
| Interaction [RNAseq HRD profile]*[chemotherapy] | 0.38 [0.22, 0.67] | <0.001 |
| Interaction test model stratified on propensity score | ||
| RNAseq HRD profile (Ref: no) | 4.90 [3.05, 7.84] | <0.001 |
| Adjuvant chemotherapy (Ref: no) | 0.52 [0.35, 0.80] | 0.002 |
| Interaction [RNAseq HRD profile]*[chemotherapy] | 0.45 [0.23, 0.90] | 0.024 |
Hazard ratio, 95% confidence intervals (95%CI), and P value computed using Cox regression models. Multivariable model: association between overall survival and RNAseq HRD profile, adjusted on a priori selected prognostic variables available (N = 2979 patients, N = 289 events). Interaction test model: association between overall survival and the interaction of [RNAseq HRD profile]×[chemotherapy] to identify RNAseq HRD profile as predictive of higher chemotherapy benefit (N = 3388 patients, N = 346 events). Interaction test model with propensity score: same model integrating a propensity score associated with the likelihood to receive adjuvant chemotherapy in the SCAN cohort (generalised linear model including age, tumour size, lymph node status, tumour group and grade) (N = 2965 patients, N = 287 events). Model stratified on quartiles of the propensity score.
Discussion
Our results demonstrate that BRCA, UCEC and OV tumours share a transcriptional profile associated with higher gHRD scores, suggested as (i) an increase in the expression of genes associated with nuclear structure, chromatin remodelling, and DNA repair, versus (ii) a decrease in the expression of genes associated with cytoskeleton and cell motility/projections. On the basis of the identified gene expression signatures, a RNAseq-based model was demonstrated to be externally applicable on multiple datasets, including FFPE samples collected in routine clinical practice, and accurately identified BRCA, UCEC, and OV tumours with high gHRD scores with or without BRCA1/2 mutations. The RNAseq HRD profile was associated with a better prognostic in advanced TP53-mutated UCEC and OV treated with platinum-based chemotherapy, in the setting of a biomarker potentially associated with higher platinum sensitivity. Conversely, in Stage I–III BRCA, the RNAseq HRD profile was associated with a poorer prognostic, as a feature partially associated with the basal phenotype, and was demonstrated as a potential predictive biomarker for a higher benefit of adjuvant chemotherapy, while the basal phenotype was not (interaction test). Therefore, in the three cancer types, the RNAseq HRD profile appeared as a potential feature associated with tumour sensitivity to chemotherapy.
Our results support that around half of TP53-mutated UCEC exhibits a HRD phenotype [4], which could be associated with a better prognosis of platinum-treated advanced disease. We also observed rare HRD phenotype in non-TP53-mutated tumours. TCGA data confirmed the low frequency of BRCA1/2 pathogenic mutations in UCEC, particularly in TP53-mutated tumours. The observation that BRCA1/2 mutations occurred mainly in POLE-mutated or dMMR UCEC, which have the highest mutational load, and without high gHRD scores, supports the passenger feature of these mutations in this setting. This suggests that PARPi efficacy would be limited to a subset of TP53-mutated tumours, and may not extent to PTEN-/ARID1A-mutated endometrioid tumours, as previously hypothesised [5, 47], or through mechanisms independent of HRD [48].
When splitting tumour types into molecular subtypes, distribution of HRD scores across gynaecological tumours in the PanCanAtlas showed a less clear-cut bimodal distribution than previously reported [9], but was consistent with other studies [49]. Both distribution of HRD score (Fig. 1A) and transcriptomic profiles (Supplementary Fig. S2) are consistent with the observation of two groups of tumours: (i) one with extreme levels of HRD-associated chromosomal damages among basal/triple-negative BRCA and OV, that is, BRCA1/2-mutated and BRCA-like tumours, and (ii) one with no any HRD-associated chromosomal damages, that is, most non-CNH UCEC, most luminal A BRCA, and a fraction of luminal B BRCA. Between these two extreme groups, we further observed a gradient of HRD-associated chromosomal damages and a consistent gradient of transcriptional phenotype, in CNH UCEC, luminal B and HER2 + BRCA, and a subset of OV and basal BRCA. This finding is consistent with previous studies reporting that only half of CNH UCEC would be affected by BRCA-deficient-like levels of HRD [4], and with the non-bimodal distribution observed for RAD51 foci (functional immune-assay for HRD) in OV [50]. These considerations suggest the hypothesis, beyond BRCA1/2-altered tumours, of a gradient of intermediate levels of deficiency, related either with incomplete genetic or epigenetic alterations in HR genes, or with compensatory/feedback loop mechanisms of nuclear/chromatin sustainment, suggested by the increase in nuclear/chromatin/DNA-repair gene expression along with HRD scores found in our study. Together, these results may explain why PARP-inhibition benefit has been found even in OV tumours without gHRD score >42 or BRCA1/2 alteration, by suggesting the existence of intermediate state of HRD in tumours with gHRD score <42 [12, 20].
Our results also suggest that some OV with gHRD score >42 but without RNAseq HRD profile may have a poorer prognosis as compared to cases consistently classified as HRD. Reversely, a subset of OV with a RNAseq HRD profile but with gHRD score <42 was shown particularly aggressive. Together, these results suggest that combining genomic and transcriptomic approaches may be complementary to better assess HRD phenotype in OV.
In BRCA, our results support that evaluation of transcriptional profiles beyond BRCA subtypes may be of interest to identify patients most eligible for adjuvant chemotherapy. In the present study, the RNAseq HRD profile was found both associated with poor outcomes, probably because of significant overlap with the basal/basal-like phenotype, and with a higher benefit of chemotherapy. This last observation is consistent with the better survival observed in UCEC or OV-advanced cancer treated with chemotherapy, and support this RNAseq HRD profile as indicator of tumour sensitivity to chemotherapy. Overall, in BRCA, our results suggest that, beyond prognostication, testing tumours for transcriptomic HRD profile may be of interest to better select patients eligible for adjuvant chemotherapy, particularly in the context of multiple available adjuvant treatments for HER2+ and ER+ tumours.
Some limitations of the present study should be pointed out. None of the cohorts did account for data from randomised clinical trial, and no large RNAseq dataset was available from patients treated with PARPi. The number of patients with advanced TP53-mutated UCEC precluded multivariable analysis to explore whether HRD phenotype would be independently associated with prognostic. Whether OV patients with low/intermediate gHRD score but RNAseq HRD profile would exhibit distinct prognostic and/or platinum- or PARP-inhibition sensitivity remains to be confirmed. Application of the RNAseq profile to samples derived from OV patients in platinum-sensitive or platinum-resistant relapse setting will be critical to better understand the dynamic of transcriptomic profiles, lacking in the present article. Indeed, HRD-associated genomic scars have been reported as imprinted, and therefore less relevant, in this setting [51].
Despite these limitations, our results demonstrate that HRD breast, endometrial, and ovarian cancers share a common transcriptional programme, and that a HRD phenotype can be estimated through RNAseq in these tumour subtypes. RNAseq-based detection of HRD tumours may be of interest for a better prognostication, and to better tailor therapeutic strategies in multiple settings in patients with gynaecological cancers.
Supplementary information
Supplemental data 2 - HRD RNAseq cytoplasmic model
Supplemental data 2 - HRD RNAseq nuclear model
Acknowledgements
The work was conducted in a research team supported by the Ligue Nationale Contre le Cancer (LNCC, Program “Equipe labelisée LIGUE”; no. EL2016.LNCC). FFPE sections were provided by the Biological Resources Centers and Tumour Bank Platforms of Cochin Hospital (BB-0033-00023 certification) and HEGP (BB-0033-00063 certification). RNA sequencing was performed at the sequencing platform of the Institut du Cerveau et de la Moelle Institute (Mr. Yannick Marie, Mme Mundwiller, CNRS UMR 7225—Inserm U 1127—Sorbonne Université UM75, Paris, France). Targeted sequencing and OncoScan microarray analyses have been performed at the GENOM’IC sequencing platform (Institut Cochin, U1016, Paris, France) and at the Department of Biochemistry and Molecular Oncology, Hopital Européen Georges Pompidou (APHP.Centre, Paris, France). Bioinformatical analyses used the French Institute of Bioinformatic clusters and local R studio on R v4. The authors would like to thank Mme De Jesus (Department of Gynaecological Surgery, Cochin Hospital), Mme Lannoy (Department of Medical Oncology, Cochin Hospital), Mme Philibert (Department of Gynecological Surgery, HEGP), Mme Hermary, Mme Le Lay (Tumour Bank Platform, Cochin Hospital), Mme Geromin, Mme Largeau, Dr. Védie, Mme Carron, Mme Le Dannois, Mme Valognes, Mme Moussy, Mme Bruneau, Mr. Maisonneuve, Mme Chabert (Tumour bank platform & pathology department, HEGP), Mme Leger, Mme Urban, Mme Goyer, Mme Auribault (Department of biochemistry/molecular oncology, HEGP), Mme Mulot, Mme Didelot, Mme Chaba, Mme Agueff, Mme Bourreau, Mme Mazoyer (Centre de Recherche des Cordeliers, Paris, France), Mr. Ladeiro and Mme Lusson (CARPEM) for their technical and administrative support. Data used in this publication were generated by the (i) National Cancer Institute Clinical Proteomic Tumour Analysis Consortium (CPTAC), (ii) the Cancer Genome Atlas, (iii) the Sweden Cancerome Analysis Network and (iv) the CARPEM institute.
Author contributions
Conceptualisation: GB, BB and JA. Data curation: GB, P-AJ, M-ALFB, MK, SG and KL. Formal analysis: GB. Funding acquisition: GB and JA. Investigation: GB, PAJ, MALFB, MK, SG, PL-P, A-SB, BB, JA. Patient accrual and data collection: GB, MK, ND, CG, CD, CC, FG, A-SB, BB, JA. Methodology: GB, P-AJ, MALFB, PL-P, A-SB, BB, JA. Project administration: GB and JA. Resources: GB, PAJ, MALFB, MK, HB, SJ, BT, CB, PL-P, A-SB, BB and JA. Supervision: PL-P, BB and JA. Writing—original draft: GB and JA. Writing—review & editing: all authors.
Funding
This work was supported by ITMO Cancer AVIESAN (Alliance Nationale pour les Sciences de la Vie et de la Santé/ National Alliance for Life Sciences & Health) within the framework of the Cancer Plan and by GHU-Assistance Publique-Hopitaux de Paris Centre (translational research program). The work was conducted within the SIRIC CARPEM translational research platform.
Data availability
Materials, data, and protocols described in the manuscript will be made available upon reasonable request at the corresponding author. Full details on data generation and data quality have been reported elsewhere [27]. The R objects to be used for the prediction on external data are provided as Supplementary Data 2.
Competing interests
GB: institutional funding from ITMO Cancer AVIESAN (French National Cancer Institute); JA: research funding from MSD; advisory board: GSK, MSD, AstraZeneca, Clovis, Eisaï. No external entities had any role in the design and conduction of the study, the collection, management, analysis, and interpretation of the data, the preparation, review and approval of the manuscript or the decision to submit the manuscript for publication. The remaining authors declare no competing interests.
Ethics approval and consent to participate
Patients treated for Stage I–IV UCEC in two University Hospitals in Paris [2010–2017] were post hoc included and referred as to the CAncer Research for Personalised Medicine Institute upon written consent or non-opposition (CARPEM) cohort (ethical approvals from National Ethical Committee CPP Ile-de-France I II & IV, DC-2020/144; DC-2019-3677; DC-2009-950, Data Protection Committee approval ID APHP2020-1109121347).
Consent to publish
Not applicable.
Footnotes
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
These authors contributed equally: Pierre-Alexandre Just, Marie-Aude Le Frere Belda.
These authors contributed equally: Anne-Sophie Bats, Bruno Borghese, Jérôme Alexandre.
Supplementary information
The online version contains supplementary material available at 10.1038/s41416-022-01900-9.
References
- 1.Knijnenburg TA, Wang L, Zimmermann MT, Chambwe N, Gao GF, Cherniack AD, et al. Genomic and molecular landscape of DNA damage repair deficiency across The Cancer Genome Atlas. Cell Rep. 2018;23:239–54.e6. doi: 10.1016/j.celrep.2018.03.076. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Timms KM, Abkevich V, Hughes E, Neff C, Reid J, Morris B, et al. Association of BRCA1/2 defects with genomic scores predictive of DNA damage repair deficiency among breast cancer subtypes. Breast Cancer Res. 2014;16:475. doi: 10.1186/s13058-014-0475-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Watkins JA, Irshad S, Grigoriadis A, Tutt AN. Genomic scars as biomarkers of homologous recombination deficiency and drug response in breast and ovarian cancers. Breast Cancer Res. 2014;16:211. doi: 10.1186/bcr3670. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.de Jonge MM, Auguste A, van Wijk LM, Schouten PC, Meijers M, ter Haar NT, et al. Frequent homologous recombination deficiency in high-grade endometrial carcinomas. Clin Cancer Res. 2019;25:1087–97. doi: 10.1158/1078-0432.CCR-18-1443. [DOI] [PubMed] [Google Scholar]
- 5.Siedel JH, Ring KL, Hu W, Dood RL, Wang Y, Baggerly K, et al. Clinical significance of homologous recombination deficiency score testing in endometrial Cancer. Gynecol Oncol. 2021;160:777–85. doi: 10.1016/j.ygyno.2020.12.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Sokol ES, Pavlick D, Khiabanian H, Frampton GM, Ross JS, Gregg JP, et al. Pan-cancer analysis of BRCA1 and BRCA2 genomic alterations and their association with genomic instability as measured by genome-wide loss of heterozygosity. JCO Precision Oncol. 2020;4:442–65. [DOI] [PMC free article] [PubMed]
- 7.Jones NL, Xiu J, Reddy SK, Burke WM, Tergas AI, Wright JD, et al. Identification of potential therapeutic targets by molecular profiling of 628 cases of uterine serous carcinoma. Gynecol Oncol. 2015;138:620–6. doi: 10.1016/j.ygyno.2015.06.034. [DOI] [PubMed] [Google Scholar]
- 8.Tutt A, Tovey H, Cheang MCU, Kernaghan S, Kilburn L, Gazinska P, et al. Carboplatin in BRCA1/2-mutated and triple-negative breast cancer BRCAness subgroups: the TNT Trial. Nat Med. 2018;24:628–37. doi: 10.1038/s41591-018-0009-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Telli ML, Timms KM, Reid J, Hennessy B, Mills GB, Jensen KC, et al. Homologous recombination deficiency (HRD) score predicts response to platinum-containing neoadjuvant chemotherapy in patients with triple-negative breast cancer. Clin Cancer Res. 2016;22:3764–73. doi: 10.1158/1078-0432.CCR-15-2477. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Robson M, Im SA, Senkus E, Xu B, Domchek SM, Masuda N, et al. Olaparib for metastatic breast cancer in patients with a germline BRCA mutation. N. Engl J Med. 2017;377:523–33. doi: 10.1056/NEJMoa1706450. [DOI] [PubMed] [Google Scholar]
- 11.Ray-Coquard I, Pautier P, Pignata S, Pérol D, González-Martín A, Berger R, et al. Olaparib plus bevacizumab as first-line maintenance in ovarian cancer. N. Engl J Med. 2019;381:2416–28. doi: 10.1056/NEJMoa1911361. [DOI] [PubMed] [Google Scholar]
- 12.González-Martín A, Pothuri B, Vergote I, DePont Christensen R, Graybill W, Mirza MR, et al. Niraparib in patients with newly diagnosed advanced ovarian cancer. N. Engl J Med. 2019;381:2391–402. doi: 10.1056/NEJMoa1910962. [DOI] [PubMed] [Google Scholar]
- 13.Hoppe MM, Sundar R, Tan DSP, Jeyasekharan AD. Biomarkers for homologous recombination deficiency in cancer. JNCI: J Natl Cancer Inst. 2018;110:704–13. doi: 10.1093/jnci/djy085. [DOI] [PubMed] [Google Scholar]
- 14.Eccles DM, Mitchell G, Monteiro ANA, Schmutzler R, Couch FJ, Spurdle AB, et al. BRCA1 and BRCA2 genetic testing—pitfalls and recommendations for managing variants of uncertain clinical significance. Ann Oncol. 2015;26:2057–65. doi: 10.1093/annonc/mdv278. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Sakai W, Swisher EM, Karlan BY, Agarwal MK, Higgins J, Friedman C, et al. Secondary mutations as a mechanism of cisplatin resistance in BRCA2-mutated cancers. Nature. 2008;451:1116–20. doi: 10.1038/nature06633. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Drost R, Bouwman P, Rottenberg S, Boon U, Schut E, Klarenbeek S, et al. BRCA1 RING function is essential for tumor suppression but dispensable for therapy resistance. Cancer Cell. 2011;20:797–809. doi: 10.1016/j.ccr.2011.11.014. [DOI] [PubMed] [Google Scholar]
- 17.Labidi-Galy SI, Olivier T, Rodrigues M, Ferraioli D, Derbel O, Bodmer A, et al. Location of mutation in BRCA2 gene and survival in patients with ovarian cancer. Clin Cancer Res. 2018;24:326–33. doi: 10.1158/1078-0432.CCR-17-2136. [DOI] [PubMed] [Google Scholar]
- 18.Sztupinszki Z, Diossy M, Börcsök J, Prosz A, Cornelius N, Kjeldsen MK, et al. Comparative assessment of diagnostic homologous recombination deficiency associated mutational signatures in ovarian cancer. Clin Cancer Res. 2021;27:5681–87. [DOI] [PubMed]
- 19.Coleman RL, Oza AM, Lorusso D, Aghajanian C, Oaknin A, Dean A, et al. Rucaparib maintenance treatment for recurrent ovarian carcinoma after response to platinum therapy (ARIEL3): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet. 2017;390:1949–61. doi: 10.1016/S0140-6736(17)32440-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Miller RE, Leary A, Scott CL, Serra V, Lord CJ, Bowtell D, et al. ESMO recommendations on predictive biomarker testing for homologous recombination deficiency and PARP inhibitor benefit in ovarian cancer. Ann Oncol. 2020;31:1606–22. doi: 10.1016/j.annonc.2020.08.2102. [DOI] [PubMed] [Google Scholar]
- 21.Pettitt SJ, Frankum JR, Punta M, Lise S, Alexander J, Chen Y, et al. Clinical BRCA1/2 reversion analysis identifies hotspot mutations and predicted neoantigens associated with therapy resistance. Cancer Discov. 2020;10:1475–88. doi: 10.1158/2159-8290.CD-19-1485. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Jazaeri AA, Yee CJ, Sotiriou C, Brantley KR, Boyd J, Liu ET. Gene expression profiles of BRCA1-linked, BRCA2-linked, and sporadic ovarian cancers. J Natl Cancer Inst. 2002;94:990–1000. doi: 10.1093/jnci/94.13.990. [DOI] [PubMed] [Google Scholar]
- 23.Konstantinopoulos PA, Spentzos D, Karlan BY, Taniguchi T, Fountzilas E, Francoeur N, et al. Gene expression profile of BRCAness that correlates with responsiveness to chemotherapy and with outcome in patients with epithelial ovarian cancer. JCO. 2010;28:3555–61. doi: 10.1200/JCO.2009.27.5719. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Severson TM, Wolf DM, Yau C, Peeters J, Wehkam D, Schouten PC, et al. The BRCA1ness signature is associated significantly with response to PARP inhibitor treatment versus control in the I-SPY 2 randomized neoadjuvant setting. Breast Cancer Res. 2017;19:99. doi: 10.1186/s13058-017-0861-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Peng G, Chun-Jen Lin C, Mo W, Dai H, Park YY, Kim SM, et al. Genome-wide transcriptome profiling of homologous recombination DNA repair. Nat Commun. 2014;5:3361. doi: 10.1038/ncomms4361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Clinical Proteomic Tumor Analysis Consortium (CPTAC) | NCI Genomic Data Commons [Internet]. [cité 24 août 2021]. Disponible sur: https://gdc.cancer.gov/about-gdc/contributed-genomic-data-cancer-research/clinical-proteomic-tumor-analysis-consortium-cptac
- 27.Beinse G, Belda MALF, Just PA, Bekmezian N, Koual M, Garinet S, et al. Development and validation of a RNAseq signature for prognostic stratification in endometrial cancer. Gynecologic Oncol. 2022. Disponible sur: https://www.gynecologiconcology-online.net/article/S0090-8258(22)00006-3/fulltext [DOI] [PubMed]
- 28.Saal LH, Vallon-Christersson J, Häkkinen J, Hegardt C, Grabau D, Winter C, et al. The Sweden Cancerome Analysis Network - Breast (SCAN-B) Initiative: a large-scale multicenter infrastructure towards implementation of breast cancer genomic analyses in the clinical routine. Genome Med. 2015;7:20. doi: 10.1186/s13073-015-0131-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Repository [Internet]. [cité 24 août 2021]. Disponible sur: https://portal.gdc.cancer.gov/repository
- 30.GEO Accession viewer [Internet]. [cité 24 août 2021]. Disponible sur: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi
- 31.Kommoss S, McConechy MK, Kommoss F, Leung S, Bunz A, Magrill J, et al. Final validation of the ProMisE molecular classifier for endometrial carcinoma in a large population-based case series. Ann Oncol. 2018;29:1180–8. doi: 10.1093/annonc/mdy058. [DOI] [PubMed] [Google Scholar]
- 32.Publication of the WHO Classification of Tumours, 5th Edition, Vol. 4: Female Genital Tumours – IARC [Internet]. [cité 4 août 2021]. Disponible sur: https://www.iarc.who.int/news-events/publication-of-the-who-classification-of-tumours-5th-edition-volume-4-female-genital-tumours/
- 33.Martin M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet J. 2011;17:10–2. doi: 10.14806/ej.17.1.200. [DOI] [Google Scholar]
- 34.Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29:15–21. doi: 10.1093/bioinformatics/bts635. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Anders S, Pyl PT, Huber W. HTSeq—a Python framework to work with high-throughput sequencing data. Bioinformatics. 2015;31:166–9. doi: 10.1093/bioinformatics/btu638. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010;26:139–40. doi: 10.1093/bioinformatics/btp616. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.EaCoN [Internet]. Gustave Roussy; 2021 [cité 24 août 2021]. Disponible sur: https://github.com/gustaveroussy/EaCoN
- 38.Loo PV, Nordgard SH, Lingjærde OC, Russnes HG, Rye IH, Sun W, et al. Allele-specific copy number analysis of tumors. Proc Natl Acad Sci USA. 2010;107:16910–5. doi: 10.1073/pnas.1009843107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Sztupinszki Z, Diossy M, Krzystanek M, Reiniger L, Csabai I, Favero F, et al. Migrating the SNP array-based homologous recombination deficiency measures to next generation sequencing data of breast cancer. npj Breast Cancer. 2018;4:1–4. doi: 10.1038/s41523-018-0066-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Brionne A, Juanchich A, Hennequet-Antier C. ViSEAGO: a Bioconductor package for clustering biological functions using Gene Ontology and semantic similarity. BioData Min. 2019;12:16. doi: 10.1186/s13040-019-0204-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Soft. 2010;33. Disponible sur: http://www.jstatsoft.org/v33/i01/ [PMC free article] [PubMed]
- 42.Ballman KV. Biomarker: predictive or prognostic? JCO. 2015;33:3968–71. doi: 10.1200/JCO.2015.63.3651. [DOI] [PubMed] [Google Scholar]
- 43.Martens EP, Boer A, de, Pestman WR, Belitser SV, Stricker BHC, Klungel OH. Comparing treatment effects after adjustment with multivariable Cox proportional hazards regression and propensity score methods. Pharmacoepidemiol Drug Saf. 2008;17:1–8. doi: 10.1002/pds.1520. [DOI] [PubMed] [Google Scholar]
- 44.Park Y, Chui MH, Suryo Rahmanto Y, Yu ZC, Shamanna RA, Bellani MA, et al. Loss of ARID1A in tumor cells renders selective vulnerability to combined ionizing radiation and PARP inhibitor therapy. Clin Cancer Res. 2019;25:5584–94. doi: 10.1158/1078-0432.CCR-18-4222. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Dedes KJ, Wetterskog D, Mendes-Pereira AM, Natrajan R, Lambros MB, Geyer FC, et al. PTEN deficiency in endometrioid endometrial adenocarcinomas predicts sensitivity to PARP inhibitors. Sci Transl Med. 2010;2:53ra75. doi: 10.1126/scitranslmed.3001538. [DOI] [PubMed] [Google Scholar]
- 46.Findlay GM, Daza RM, Martin B, Zhang MD, Leith AP, Gasperini M, et al. Accurate classification of BRCA1 variants with saturation genome editing. Nature. 2018;562:217–22. doi: 10.1038/s41586-018-0461-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Philip CA, Laskov I, Beauchamp MC, Marques M, Amin O, Bitharas J, et al. Inhibition of PI3K-AKT-mTOR pathway sensitizes endometrial cancer cell lines to PARP inhibitors. BMC Cancer. 2017;17:638. doi: 10.1186/s12885-017-3639-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Romero I, Rubio MJ, Medina M, Matias-Guiu X, Santacana M, Schoenenberger JA, et al. An olaparib window-of-opportunity trial in patients with early-stage endometrial carcinoma: POLEN study. Gynecol Oncol. 2020;159:721–31. doi: 10.1016/j.ygyno.2020.09.013. [DOI] [PubMed] [Google Scholar]
- 49.Takaya H, Nakai H, Takamatsu S, Mandai M, Matsumura N. Homologous recombination deficiency status-based classification of high-grade serous ovarian carcinoma. Sci Rep. 2020;10:2757. doi: 10.1038/s41598-020-59671-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Blanc-Durand F, Yaniz E, Genestie C, Rouleau E, Berton D, Lortholary A, et al. Evaluation of a RAD51 functional assay in advanced ovarian cancer, a GINECO/GINEGEPS study. JCO. 2021;39:5513–5513. doi: 10.1200/JCO.2021.39.15_suppl.5513. [DOI] [Google Scholar]
- 51.Patel JN, Braicu I, Timms KM, Solimeno C, Tshiaba P, Reid J, et al. Characterisation of homologous recombination deficiency in paired primary and recurrent high-grade serous ovarian cancer. Br J Cancer. 2018;119:1060–6. doi: 10.1038/s41416-018-0268-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental data 2 - HRD RNAseq cytoplasmic model
Supplemental data 2 - HRD RNAseq nuclear model
Data Availability Statement
Materials, data, and protocols described in the manuscript will be made available upon reasonable request at the corresponding author. Full details on data generation and data quality have been reported elsewhere [27]. The R objects to be used for the prediction on external data are provided as Supplementary Data 2.