Combining Homologous Recombination-Deficient Testing and Functional RAD51 Analysis Enhances the Prediction of Poly(ADP-Ribose) Polymerase Inhibitor Sensitivity

Abstract

PURPOSE

To meet the urgent need for accessible homologous recombination-deficient (HRD) test options, we validated a laboratory-developed test (LDT) and a functional RAD51 assay to assess patients with ovarian cancer and predict the clinical benefit of poly(ADP-ribose) polymerase inhibitor therapy.

METHODS

Optimization of the LDT cutoff and validation on the basis of samples from 91 patients enrolled in the ENGOT-ov24/NSGO-AVANOVA1&2 trial (ClinicalTrials.gov identifier: NCT02354131), previously subjected to commercial CDx HRD testing (CDx). RAD51 foci analysis was performed and tumors with ≥five foci/nucleus were classified as RAD51-positive (homologous recombination-proficient).

RESULTS

The optimal LDT cutoff is 54. Comparing CDx genome instability score and LDT HRD scores show a Spearman's correlation of rho = 0.764 (P < .0001). Cross-tabulation analysis shows that the sensitivity of the LDT HRD score is 86% and of the LDT HRD status is 91.8% (Fisher's exact test P < .001). Survival analysis on progression-free survival (PFS) of LDT-assessed patients show a Cox regression P < .05. RAD51 assays show a correlation between low RAD51 foci detection (<20% RAD51⁺ cells) and significantly prolonged PFS (P < .001).

CONCLUSION

The robust concordance between the open standard LDT and the CDx, especially the correlation with PFS, warrants future validation and implementation of the open standard LDT for HRD testing in diagnostic settings.

Combining open standard HRD testing and RAD51 foci assay enhances prediction of PARPi sensitivity in ovarian cancer.

INTRODUCTION

Ovarian cancer accounts for more than 300,000 new cases worldwide per year, with approximately 90% classified as epithelial ovarian cancer.¹ Currently, the standard-of-care treatment for epithelial ovarian cancer is cytoreductive surgery and platinum-based first-line chemotherapy. Nevertheless, resistance is common, with the disease reoccurring in ∼70% of patients.² The discovery of a synthetic lethal relationship between breast cancer gene 1 and 2 (BRCA1/2) and poly(ADP-ribose) polymerase (PARP) enzymes led to the development of PARP inhibitors (PARPis)^3,4 and revolutionized the management of patients with ovarian cancer. Mechanistically, PARPi is thought to stabilize PARP enzymes onto the damaged DNA (termed as PARP trapping), thereby preventing DNA repair, replication, and transcription, leading to cell death.^5,6 PARPis were initially proven beneficial for patients who suffered from a homologous recombination-deficient (HRD) tumor⁷ because of germline mutations in the BRCA1/2 genes. To explore the hypothesis that PARPis could also be efficacious for patients without detected pathogenic mutations in BRCA1/2, several clinical trials were conducted on non–BRCA-mutated patients showing significantly increased progression-free survival (PFS) upon PARPi treatment among patients with tumors classified as HRD by the US Food and Drug Administration–approved HRD test MyChoice CDx (Myriad Genetics, hereafter referenced as CDx).^8,9 Nevertheless, introducing commercial HRD tests raises several concerns and an increased arsenal of HRD testing platforms has arisen.

CONTEXT

• Key Objective

• Does functional RAD51 staining in combination with homologous recombination-deficient (HRD) testing increase the prediction of poly(ADP-ribose) polymerase inhibitor (PARPi) sensitivity in patients with ovarian cancer?

• Knowledge Generated

• Our laboratory-developed test can assess HRD status to the same standard as the US Food and Drug Administration–approved CDx test. With the addition of functional RAD51 staining assay, we can further distinguish current functional HRD status from the original HRD status, which is merely based on genomic instability parameters. Thus, functional RAD51 staining can explain PARPi resistance among patients with ovarian cancer.

• Relevance

• This study reveals the importance of taking advantage of multiple testing platforms when accurately predicting PARPi sensitivity in ovarian cancer patients. Moreover, these results may pave the way for a more accurate diagnostic work-up before initiating targeted PARPi treatment.

In addition to displaying a heavily scarred genome, tumors lacking the homologous recombination (HR) repair pathway also exhibit lack of RAD51 nucleoprotein filament formation. The formation of RAD51 nucleoprotein filament on single-stranded DNA is a crucial step in the repair of DNA double-strand breaks by the HR pathway. In fact, RAD51 nucleoprotein filament formation commits the repair of damaged DNA only through the HR pathway. Thus, the absence or presence of RAD51 nucleoprotein filament is a clear indicator of HRD and HR-proficient (HRP) tumors, respectively. The RAD51 status in the ovarian tumor formalin-fixed paraffin-embedded (FFPE) tissues can be evaluated by either estimating the number of RAD51 foci or overall RAD51 expression and has been shown to be a reliable biomarker for identifying patients who would benefit from targeted treatment with PARPi.^10-16 Moreover, RAD51-based functional HRD tests can assess the current HR status, that is, functional HRD or HRP (fHRD or fHRP), of the tumor and effectively identify the reason behind PARPi resistance when compared with next generation sequencing-based approaches such as the CDx test.

We performed a validation of our laboratory-developed test (LDT) for HRD assessment on the basis of the ENGOT-ov24/NSGO-AVANOVA1&2 study^17,18 (from now on, referred to as AVANOVA1&2). The trial investigated the effect of niraparib plus bevacizumab versus niraparib alone for treatment of high-grade serous/endometrioid platinum-sensitive relapsed ovarian cancer tumors. All tumors were originally analyzed with the CDx test. Moreover, we examined the prognostic probability of the LDT in terms of PFS as well as the alternative RAD51-based functional HRD test for assessing PARPi sensitivity (Fig 1).

FIG 1.

Analysis overview. Cutoff optimization: an in-house cohort of 14 BRCA-mutated ovarian cancer samples and 57 BRCA-mutated ovarian cancer samples from the TCGA cohort were analyzed with the laboratory-developed HRD test script (LDT). A HRD score cutoff was set on the basis of the 20th percentile of HRD score values. LDT validation: samples from 91 patients included in the AVANOVA1 (n = 11) and AVANOVA2 (n = 80) trials were tested by both CDx and our in-house LDT. Of the 91 AVANOVA patients, CDx GIS was derived from 79 samples, and LDT HRD score was derived from 85 samples. CDx was unable to analyze 12 samples and LDT unable to analyze six samples, with one of the samples overlapping between the two tests. The performance of our LDT HRD scores was assessed by comparing the CDx and LDT HRD scores of 74 patients with data available from both tests. A positive or a negative score was determined for each test on the basis of the appropriate cutoffs. Combining the positive/negative HRD score with an evaluation of the BRCA status reveals an overall HRD status for each sample. Samples that did not yield a conclusive HRD score were categorized as negative (HRP). Further evaluation of LDT-assed HRD status were carried out comparing LDT and CDx HRD status for all patients (N = 91). Finally, the prognostic utility of CDx-assessed and LDT-assessed HRD status was compared using PFS recorded for each patient across treatment arms. RAD51 foci investigation: 19 FFPE tissue samples from the AVANOVA1/2 trials were investigated to assess the potential of functional RAD51 assay in predicting PARPi treatment outcomes. FFPE, formalin-fixed paraffin-embedded; GIS, genomic instability score; HRD, homologous recombination-deficient; HRP, HR-proficient; LDT, laboratory-developed test; PARPi, poly(ADP-ribose) polymerase inhibitor; PFS, progression-free survival; TCGA, The Cancer Genome Atlas.

METHODS

Study Cohorts and Platforms

Explorative cohorts: (1) in-house ovarian cancer samples (N = 14) already diagnosed with pathogenic BRCA1/2 mutation status and analyzed by copy number variation (CNV)-array platform (see the Data Supplement, Table S1, for details). For these samples, the HRD was computed as described in Allele-Specific Copy Number (ASCN) analysis and LDT HRD score sections. (2) Ovarian cancer samples (N = 57) from the The Cancer Genome Atlas (TCGA) pan-cancer cohort (pancan12),¹⁹ with confirmed pathogenic BRCA1/2 mutation status. TCGA samples were profiled with Affymetrix SNP 6.0 platform. The mutational status for BRCA1 and BRCA2 was obtained from the TCGA article (S8²⁰; see the Data Supplement, Table S2, for details). Segmented copy-number data, as computed by the ABSOLUTE software, were downloaded from Synapse web portal.²¹ For these samples, the HRD score was computed as described in the LDT HRD score section. Validation cohort: tumor DNA from primary cytoreductive surgery and survival data obtained from 91 patients enrolled in the AVANOVA (ClinicalTrials.gov identifier: NCT02354131) study cohort; 11 patients from AVANOVA 1 and 80 patients from AVANOVA 2. All samples were analyzed both by CNV-array as well as BRCA mutational screening platforms as described in the BRCA mutation screening analysis section. Also, for these samples, the HRD score was computed as described in ASCN analysis and LDT HRD score sections (see the Data Supplement, Table S3, for details). A subset of the AVANOVA validation cohort (n = 19) was used to investigate the performance of the RAD51 functional assay with FFPE tissue samples analyzed as described in the Immunofluorescence staining and RAD51 foci scoring section (see the Data Supplement, Methods, for details).

Tumor Samples and DNA Extraction

DNA from FFPE tissues was extracted using the QIAsymphony GeneRead DNA FFPE Treatment Kit (Qiagen, Hilden, Germany) and purified using the Maxwell RSC instrument (Promega, Madison, WI). Extracted DNA was quantified using the Qubit Fluorometer (Thermo Fisher Scientific, Waltham, MA).

BRCA Mutation Screening Analysis

To detect pathogenic and likely pathogenic BRCA1 and BRCA2 mutations in the 91 samples from the AVANOVA cohort, we performed next-generation sequencing analysis. DNA libraries were prepared from a minimum of 200 ng FFPE-DNA, hybridized using the TruSight Oncology (TSO) 500 HT gene panel (Illumina, San Diego, CA) and sequenced on the NovaSeq6000 platform to a minimum median coverage of 300×. Sequencing reads were processed using the TSO500 Illumina pipeline (version: ruo-2.2.0.12). Mutations present in ≥5% of the sequencing reads were identified and manually classified using the Qiagen Clinical Insight Interpret Translational software. All pathogenic or likely pathogenic identified mutations were manually confirmed in the sequencing reads using the Integrative Genomics Viewer.

Copy-Number Variation Array

OncoScan and CytoScan CNV assay (Thermo Fisher Scientific, Waltham, MA) platforms were used with a minimum input of 80 ng and 250 ng DNA, respectively, and scanning of the arrays was performed with the GeneChip Scanner 3000 7G, resulting in .CEL files for downstream processing.

ASCN Analysis

ASCN analysis was performed via the EaCoN R package version 0.3.6, available on GitHub.²² The EaCoN suite consists of a series of R packages that perform ASCN analysis from raw .CEL files of Thermo Fisher Scientific microarrays (for a detailed ASCN analysis protocol, see the Data Supplement, Methods). For this study, data from CytoScan HD and OncoScan SNP microarray platforms were used.

LDT HRD Score

The LDT HRD score was computed using the ovaHRDscar package, available on GitHub.²³ The computation was modeled after the Telli et al²⁴ publication, where the HRD score is defined as the unweighted sum of loss of heterozygosity (LOH), telomeric allelic imbalance (TAI), and large-scale state transitions (LSTs). The fitness of these chromosomal aberrations as HRD biomarkers has been validated in previous studies.^25-27 They were computed as described in the study by Perez-Villatoro et al, from the genomewide ASCN profile generated via the EaCoN package.³⁸ In their work, Perez-Villatoro et al revisited the definition of the three chromosomal aberrations to maximize their fitness as HRD biomarkers in high-grade serous ovarian cancer. LOH count was defined as the number of homozygous regions between 1 and 15 Mb, TAI count as the number of regions with allelic imbalance that extends to one of the subtelomeres but do not cross the centromere, and LST count as the number of chromosomal breaks between contiguous regions larger than 10 Mb, with a distance between them not exceeding 3 Mb. In accordance with previous studies, samples that did not yield a numeric HRD score were considered negative.¹⁷

Immunofluorescence Staining and RAD51 Foci Scoring

Immunofluorescence staining of FFPE samples was carried out according to previous protocols.^12,13,28 For the detailed protocol, see the Data Supplement (Methods).

Qualitative image analyses for fluorescence studies were carried out using the point scanning confocal microscope LSM780 (Zeiss, Oberkochen, Germany), using an ×63 oil objective and ZEN Software Black edition (Zeiss). Microscope images were adjusted and combined using Photoshop CS6 (version 24.4.1; Adobe, San Jose, CA).

For RAD51 foci scoring, geminin-positive (GMN⁺) cells were identified within vital tumor areas and 40 GMN⁺ cells were selected at random. Scoring of RAD51 foci was performed blindly from images, and cells were considered RAD51-positive (RAD51⁺) if they had ≥five RAD51 foci/nucleus.²⁹ At least two biological replicates were analyzed and the percentage of RAD51⁺/GMN⁺ cells was determined manually by two independent investigators. The HRD proficiency threshold was set to more than 20% GMN⁺ cells with ≥five RAD51 foci/nucleus in accordance with previous studies.²⁹

Statistical Analyses

Statistical analyses were performed with SPSS Statistics 28.0.0.0 (IBM, Armonk, NY) and Prism version 9.4.1 (GraphPad, Boston, MA). A P value of <.05 was considered significant. The chosen statistical test for each analysis is indicated in figure legends. All figures were combined and visually adjusted using Illustrator (version 27.11, Adobe, San Jose, CA).

Ethics Approval and Consent to Participate

The AVANOVA trial protocol and all amendments were approved by the research ethics committee/institutional review boards (IRBs) and the competent authorities of the participating hospitals and countries (ClinicalTrials.gov identifier: NCT02354131). The trial was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. All patients provided written informed consent and consented to upfront HRD testing by Myriad. The reuse of the same archival tumor tissue samples for genomic subanalysis, including the results presented in this paper, has been approved by the research ethics committee/IRBs of the participating hospitals and countries. Approval was given to analyze tissue from patients already deceased or terminally ill without renewal of consent. Patients still alive were asked for written consent.

RESULTS

Optimization of the LDT Cutoff

Samples from 14 in-house BRCA-mutated ovarian cancer patients were used to determine the LDT HRD score threshold. The LDT-derived HRD scores for the 14 samples (Data Supplement, Table S1) was obtained as described in the ASCN analysis and LDT HRD score methods sections. The LDT HRD score threshold was defined as the 20th percentile of the 14 samples' HRD score distribution (Data Supplement, Fig S1A), resulting in a LDT HRD score threshold equal to 54. For validation in an external larger cohort, segmented copy-number data from 57 BRCA-mutated ovarian cancer TCGA patients (Data Supplement, Table S2) were analyzed. The LDT-derived HRD score for the 57 samples was obtained as described in the LDT HRD score methods section. The LDT HRD score threshold was likewise defined as the 20th percentile of the 57 samples' HRD score distribution, confirming a LDT HRD score threshold equal to 54 (Data Supplement, Fig S1B).

Samples from 91 patients included in AVANOVA1 (n = 11) and AVANOVA2 (n = 80) were then tested by our LDT HRD assay (Data Supplement, Table S3), as described in the ASCN analysis and LDT HRD score methods sections. Six samples could not be analyzed with the LDT script, because allele-specific copy number analysis of tumors could not converge, resulting in HRD scores obtained for 85 patients (Fig 1). The LDT HRD score distribution stratified for BRCA mutation status is visualized in the Data Supplement (Figs S2A and S2B) and shows that the majority of BRCA-mutated samples have a LDT HRD score above the established cutoff of 54, represented by the dotted line. BRCA wild-type (wt) samples are shown to have wider distribution of LDT HRD scores, both above and below the threshold.

Comparison of LDT HRD Scores and CDx Genome Instability Score

The obtained LDT HRD score from the AVANOVA1/2 samples were compared with available CDx genome instability score (GIS) from the same FFPE samples, which were originally analyzed and used for stratification in the AVANOVA trial (Data Supplement, Table S3). Twelve samples failed the CDx HRD test (two samples because of reported insufficient tumor cell content; Fig 1). One failed both CDx and LDT, resulting in test score results from 74 patients eligible for direct comparison (Fig 1). As depicted in Figure 2, a strong correlation between score results from the two tests was observed with Spearman's correlation coefficient, rho = 0.764, and a one-tailed P < .0001.

FIG 2.

Correlation analysis with score values from patient samples with available LDT HRD scores and CDx GIS (n = 74). Spearman's correlation plot showing a high degree of association, with Spearman's rho = 0.7635 and P < .0001 (one-tailed), between CDx GIS and LDT HRD score obtained from the same 74 patients. The correlation is visualized using best fit model (line). GIS, genomic instability score; HRD, homologous recombination-deficient; LDT, laboratory-developed test.

To further explore the concordance between the HRD test results, each score was interpreted as a HRD group. A CDx GIS ≥42 is considered HRD-positive²⁴ and on the basis of the previously established cutoff of 54, a LDT HRD score ≥54 is considered HRD-positive (Data Supplement, Table S3). The contingency table of positive and/or negative scores derived from each of the two tests is shown in the Data Supplement (Table S4) and has a Fisher's exact test P < .001, with sensitivity = 86.0%, specificity = 80.6%, PPV = 86.0%, and NPV = 80.6%, as summarized in Table 1.

TABLE 1.

Fisher's Exact Test P, Sensitivity, Specificity, PPV, and NPV Were Derived From Comparisons Between HRD Score: CDx GIS Versus LDT HRD Scores (n = 74), and Test Assessed HRD Status: CDx Versus LDT (n = 91)

Test	HRD Score (n = 74)	HRD Status (n = 91)
Fisher's exact test (P)	<.001	<.001
Sensitivity, %	86.0	91.8
Specificity, %	80.6	73.8
PPV, %	86.0	80.4
NPV, %	80.6	88.6

Abbreviations: GIS, genomic instability score; HRD, homologous recombination-deficient; LDT, laboratory-developed test; NPV, negative predictive value; PPV, positive predictive value.

Concordance Between LDT-Assessed and CDx-Assessed HRD Status

Combining the HRD score with BRCA mutation identification defines the HRD status for each sample (Data Supplement, Table S3). Samples with a HRD score above the established threshold and/or a pathogenic/likely pathogenic BRCA mutation are classified as having a HRD status. Samples with a negative or inconclusive HRD score and BRCA wt were classified as having a HRP status. The number of samples classified as HRD and/or HRP on the basis of each of the two tests are visualized in a contingency table (Data Supplement, Table S5), and the Fisher's exact P < .001, sensitivity = 91.8%, specificity = 73.8%, PPV = 80.4%, and NPV = 88.6% are summarized in Table 1.

The Prognostic Probability of the HRD Status

After establishing that our LDT HRD score and status align with the CDx GIS and HRD status, we next sought to evaluate the ability of our LDT HRD status to predict PFS for patients enrolled in the two AVANOVA trials. The survival data analyzed in Figure 3 include 11 patients enrolled in AVANOVA1 receiving niraparib-bevacizumab combination therapy, 37 patients enrolled in AVANOVA2 receiving single-agent niraparib (arm 1) and 41 patients enrolled in AVANOVA2 receiving niraparib-bevacizumab combination therapy (arm 2) as depicted in Figure 1 and listed in the Data Supplement (Table S3). In the interest of evaluating the prognostic utility of the HRD status upon PARPi treatment, the two patients enrolled in AVANOVA2 only receiving single-agent bevacizumab (arm 3) were excluded from the analysis. Overall, the Kaplan-Meier survival curve depicted in Figure 3 shows that a positive HRD status, either derived by CDx or LDT, is associated with longer PFS compared with HRP patients. When adjusting for treatment arm as a categorical variable, the test on a Cox regression showed significant differences in PFS distribution between both LDT-assessed HRD patients and HRP patients (P = .043), and between CDx-assessed HRD patients and HRP patients (P = .006).

FIG 3.

A positive HRD status is associated with longer PFS. Survival plot showing cumulative PFS stratified by HRD status and type of HRD assay (n = 89). The test on a Cox regression showed significant differences in the survival distribution between both CDx-assessed HRD and HRP patients (P < .01) and between LDT-assessed HRD and HRP patients (P < .05), when adjusting for treatment arm as a categorical variable. HRD, homologous recombination-deficient; HRP, HR-proficient; LDT, laboratory-developed test; PFS, progression-free survival.

Investigation of the Prognostic Probability of the RAD51 Foci Assay

Nineteen samples from the AVANOVA1 and 2 study underwent further analysis with a functional RAD51 assay (representative images are shown in Fig 4). One sample was excluded because of low γH2AX score implying insufficient endogenous DNA damage. Staining for GMN was conducted to ensure that selected cells were in S-phase and the percentage of GMN⁺ cells with ≥five RAD51 foci was noted for each sample (Data Supplement, Table S6). The percentage of GMN⁺RAD51⁺ cells and PFS of each patient is plotted in Figure 5A and significant correlation between low RAD51 foci detection and longer PFS visualized with a fitted line (Spearman's rho = –0.523, P < .013, one-tailed). In accordance with previous studies, the functional HR proficiency threshold was set to 20% GMN⁺RAD51⁺ cells.²⁹ As depicted in Figure 5B, survival curves stratified on the basis of functional HR showed a significantly increased PFS among fHRD-assessed (<20% GMN⁺RAD51⁺ cells) patients compared with fHRP-assessed (≥20% GMN⁺RAD51⁺ cells) patients (P < .001, log-rank Mantel-Cox). In comparison, survival analysis over the same cohort stratified on the basis of the CDx test (Fig 5C) showed no significant increase in survival among HRD-assessed patients (P = .384, log-rank Mantel-Cox, one-tailed). A similar survival analysis was conducted on the basis of LDT assessment (Fig 5D), showing a significant stratification of PFS between the HRD and HRP patient groups (P = .024, log-rank Mantel-Cox, one-tailed), yet not as convincing as the functional RAD51 assay. In line with a recent study where the functional HR proficiency threshold was set to 10% GMN⁺RAD51⁺ cells,³⁰ we also confirmed a significant difference in PFS between fHRD and fHRP in the same cohort when applying a threshold of 10% (Data Supplement, Fig S3).

FIG 4.

Functional RAD51 assay performed on FFPE tissue slides. Representative images of AVANOVA FFPE samples (n = 18). All FFPE samples were stained for DAPI, GMN, and RAD51. The nucleus is marked by the dotted line. Scale bar = 10 μm. FFPE, formalin-fixed paraffin-embedded; fHRD, functional HRD; fHRP, functional HR-proficient; HRD, homologous recombination-deficient.

FIG 5.

A low fHRD percentage is associated with longer PFS. (A) Correlation plot showing the percentage of GMN+RAD51+ cells and PFS of each patient (n = 18). Red dot indicates fHRD status (<20% GMN+RAD51+ cells) and green dot indicates fHRP status (≥20% GMN+RAD51+ cells). The correlation between PFS and functional HR status is visualized with a best fitted line and a Spearman's correlation test applied to determine the degree of association between the two variables: Spearman's rho = –0.523 and P = .013 (one-tailed). (B-D) Survival plots (n = 18) showing cumulative PFS stratified by (B) functional HRD status with P > .001, (C) CDx assessment with P = .384, and (D) LDT assessment with P = .024. Statistical analyses were performed using log-rank Mantel-Cox tests. fHRD, functional HRD; fHRP, functional HR-proficient; HRD, homologous recombination-deficient; LDT, laboratory-developed test; PFS, progression-free survival.

DISCUSSION

Tumors that are devoid of the HR repair pathway characteristically display heavily scarred genome and lack RAD51 nucleoprotein filament formation. Indeed, the presence of a heavily scarred genome is a hallmark of HRD tumors that is currently being exploited to identify patients to benefit from personalized treatments such as PARPi.^31-33

As reflected in the AVANOVA cohort, many BRCA wt tumors were shown to be HRD-positive, emphasizing the need for HRD testing beyond classic BRCA testing. However, a subgroup of the BRCA-mutant tumors also scored below the HRD threshold of 54. Although deleterious BRCA1/2 mutations biologically should lead to genome instability and thereby a positive HRD score, there could be several reasons why genotype and phenotype do not always match. One reason could be that the HRD score could be assessed from a tumor subclone that is not dominated by the BRCA1/2 mutations. Also, the BRCA1/2 mutation may be in a domain where it might impair protein function without completely losing the function, resulting in somehow maintained genomic integrity and the HRD platform may not capture the genome instability. Taken together, the combined HRD assessment is the most thorough approach.

Recent data suggest that previous PARPi treatment likely reduces the efficacy of subsequent platinum-based chemotherapy in patients with ovarian cancer, compared with patients who did not receive previous PARPi treatment.^34,35 A rise in platinum resistance among PARPi-treated patients could be linked to newly disclosed data reporting an increase in BRCA reversion mutations in ∼20% of patients with ovarian cancer after treatment with PARPi.³⁶ BRCA reversion mutations are not reflected in tests based on detection of pathogenic BRCA mutation or genomic instability, which primarily capture previous HRD exposure. As an alternative to these methods measuring a historical HRD status, the current HRD status of the patient can instead be measured with functional RAD51 assays.³⁷

We here explored the RAD51 assay as a diagnostic add-on tool to evaluate HR repair capability, beyond HR gene panels and genomic instability assessment. In accordance with recent work,¹⁰ our results link RAD51 foci detection below the HRD threshold (<20% RAD51⁺ cells) to prolonged survival among patients with ovarian cancer after PARPi treatment, indicating that the lack of RAD51 nuclear staining is a significant predictor for PARPi sensitivity. In two cases (patients 2 and 13), we detected RAD51 foci formation above the set HRD threshold, classifying them as fHRP, and saw a corresponding short PFS interval, although both patients were classified as HRD on the basis of both BRCA status and assessment with CDx and LDT. These cases likely reflect the consequences of BRCA reversion mutations and emphasize the need for functional add-on tests to comprehend the full complexity of assessing HRD status. We suggest that these functional add-on tests could be offered to patients classified with a HRD-positive status by genomic testing, to ensure that a HRD status based on previous genomic instability still corresponds to the present PARPi sensitivity and thereby avoid exposing patients to unnecessary treatment.

In conclusion, we have developed a LDT to assess HRD in patients with ovarian cancer, optimizing the numeric threshold for HRD classification. Our LDT shows a high concordance with the CDx, both when comparing test scores, when evaluating overall HRD status assessment, and when correlating HRD status with clinical outcome measured by PFS. In line with current benchmarking studies, this encourages future validation and implementation of laboratory-developed HRD testing in a diagnostic setting, as an alternative to the available commercial tests. We show that our approach is valid across several platforms for detecting HRD. Besides HRD tests based on detection of previous HRD exposure, functional RAD51 assays present themselves as a valuable add-on when assessing the current HR capabilities of the cancer. In our preliminary study we showed a significant correlation between the lack of RAD51 nuclear staining and PARPi sensitivity, when implementing a fHRD cutoff at 20% RAD51⁺ cells with ≥5 RAD51 foci/nucleus. HRD assessment based on the functional RAD51 assay was even able to explain bad prognosis of two BRCA-mutated patients, implying that although genomic HRD testing has a generally high sensitivity, functional HRD testing improves specificity. For future clinical assessment among patients with ovarian cancer, we therefore suggest that functional RAD51 assays should be considered as an add-on biomarker for already HRD-classified patients.

DATA SHARING STATEMENT

All data generated or analyzed during this study are included in the manuscript and its supplementary information files. Array and sequencing files will be uploaded in public repositories upon acceptance of the manuscript.

AUTHOR CONTRIBUTIONS

Conception and design: Maj Kjeldsen, Mansoor Raza Mirza, Maria Rossing

Provision of study materials or patients: Maj Kjeldsen

Collection and assembly of data: Maj Kjeldsen, Lorenzo Perino, Luca Mariani, Gitte-Bettina Nyvang, Elisabeth Kristensen, Mansoor Raza Mirza, Maria Rossing

Data analysis and interpretation: Maj Kjeldsen, Lorenzo Perino, Frederik O. Bagger, Mansoor Raza Mirza, Maria Rossing

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).