Among patients with ovarian cancer who had received PARPi between September 2017 and June 2022 at the Yonsei Cancer Center (Seoul, Korea), those with archival FFPE tissue samples prior to PARPi exposure (“pre-PARPi”) and/or after progression on PARPi (“post-PARPi”) were selected. The study was approved by the Institutional Review Board of Yonsei University (approval number 4-2020-0386) and adhered to the principles of the Declaration of Helsinki. Informed consent was obtained from all participants.

For histologic analysis, hematoxylin and eosin-stained tumor-containing slides were independently reviewed by gynecologic pathologists and a representative FFPE tumor tissue block was selected.

For IHC analysis, we used three markers: RAD51 for defining the HR status, geminin for identifying cells in the G2/S cell-cycle phase, and γH2AX as a DNA double-strand break marker (26). The following antibodies were used for the staining: RAD51 (1:1,000, clone 14B4; GeneTex, Irvine, CA, USA), geminin (1:1,000, clone 10802-1-AP; ProteinTech, Chicago, IL, USA), and γH2AX (1:10,000, clone JBW301; Sigma-Aldrich, Dorset, UK). Approximately 4-μm-thick sections of FFPE blocks were immunostained using a Ventana BenchMark XT automated stainer (Ventana Medical Systems, Tucson, AZ, USA) and using the UltraView Universal DAB Detection Kit (Ventana Medical Systems). Positive and negative controls were stained concurrently to validate the staining method. Germ cells from normal testicular testis tissue was used as a positive control for RAD51 and geminin, while normal palatine tonsil tissue was the positive control for γH2AX. Tests were also conducted without the primary antibody to act as negative controls for all antibodies.

Cases with low cellularity (<300 tumor cells/high-power field) were excluded. To evaluate cells in the G2/S phase and with DNA damage, tumors with <25% γH2AX-positive cells or <3% geminin-positive cells were excluded (20). For RAD51 IHC analysis, nuclear expression was evaluated using the H-score method, which is a semi-quantitative system with a 0–300 score range. The percentage of positive cells (0–100%) was multiplied by the dominant staining intensity score. Scores were defined as follows: 0, no staining; 1, barely detectable staining; 2, distinct brown staining; and 3, strong dark brown staining. Applying the optimal cutoff for RAD51 expression determined based on maximally selected rank statistics formulated using the Contal and O’Quigley method, RAD51 H-score <20 was defined as RAD51-low and RAD51 H-score ≥20 was defined as RAD51-high (27). This binary definition was used in all analyses, unless otherwise specified. All slides were evaluated by two experienced pathologists (E.P. and K.K.) in a blinded manner. If discrepancies occurred, consensus was reached.

BRCA1/2 genetic testing was performed using genomic DNA from peripheral blood samples. Sanger sequencing analysis was performed using a 3730 DNA Analyzer with a BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, CA, USA) and the Sequencher 5.3 software (Gene Codes, Ann Arbor, MI). Next-generation sequencing was performed on a proportion of patients using a custom panel, including BRCA1 and BRCA2, on a MiSeq sequencer (Illumina, San Diego, CA, USA) using the MiSeq Reagent Kit v2 (300 cycles). Bioinformatic analysis was performed using the Burrows–Wheeler Aligner, Genome Analysis Toolkit, Ensembl Variant Effect Predictor, and a custom pipeline. Experienced geneticists made the final interpretations.

Whole-exome data were obtained from fresh frozen or FFPE samples. Genomic scarring was estimated by determining the copy number alterations in the sequencing data using Sequenza-utils (v.3.0.0) (28) based on LOH, TAI, and LST estimated using the R package scarHRD (v.0.1.1) (29). The sum of these values was defined as the genomic scar score (GSS). Details of the sequencing methodology have been described previously (12).

Cell-free DNA extracted from serially collected whole blood samples was used for next-generation sequencing using an Illumina NovaSeq 6000 System. The sequencing library was prepared using the Library Prep Reagent for Illumina (Dxome, Seoul, Korea). Target enrichment was performed using the TMB 500 panel, whereby 531 cancer-related genes were targeted (Dxome) and analyzed using the PiSeq algorithm (Dxome). All variants were classified into a four-tiered system based on the standards and guidelines of the Association of Molecular Pathology/American Society of Clinical Oncology/College of American Pathologists. Details of the ctDNA preprocessing steps have been described previously (30).

Based on the ctDNA mutation profiles from pre- and post-treatment matched samples, mutations specifically present in post-treatment samples (i.e., not present in pre-treatment samples) were identified in each patient (31) and used to identify resistance mechanisms. As performed in our previous approach (30), individual genes in the TMB500 panel were individually reviewed to determine their implication in PARPi resistance. Among patients who were tested for both RAD51 IHC and ctDNA, resistance mechanisms were classified into HR restoration, replication fork stabilization, survival upregulation, target loss, drug efflux, and BRCA reversion (30).

Clinical variables were evaluated to assess RAD51 IHC efficacy. Basic clinical variables, including histology, International Federation of Gynecology and Obstetrics stage, and BRCA status, were collected. Clinical variables associated with PARPi, such as drug type, treatment setting, and line of treatment, were collected. The treatment setting was defined as maintenance if PARPi was administered immediately following a complete or partial response to chemotherapy, or as salvage if PARPi was administered as a treatment without any preceding chemotherapy. Medical records were reviewed for PARPi start date, disease recurrence date, or PARPi end date for causes other than recurrence or death. Disease recurrence was determined based on a radiological assessment, which was performed every three months for the first two years and every six months thereafter. PFS was defined as the period from the PARPi start date to the date of disease progression or PARPi end date, whichever occurred first. Patients who did not experience a second disease progression or died were considered censored.

A standard box-and-whisker plot was used to summarize the continuous data by indicating the median, the first and third quantiles within the box, whiskers extending to 1.5 times the interquartile range, and any outlier points. Statistical significance was calculated using Fisher’s exact test or the chi-squared test for categorical variables and Student’s t-test for continuous variables. The Shapiro-Wilk test was used to check the normalization assumption; based on these results, data are presented as either the median with range or the mean with standard deviation, as appropriate. The Kaplan–Meier method was used for PFS analysis. A cox proportional hazards regression model was used to evaluate the impact of RAD51 H-score as well as other relevant prognostic variables on disease progression on PARPi. Statistical significance was set at P < 0.05. All statistical analyses were performed using R v.4.0.3 (R Foundation for Statistical Computing, Vienna, Austria).

The ability to functionally assess HR status, besides germline BRCA mutations or genomic scars, is clinically important in de novo resistance, which affects the initial response to PARPi therapy, and in acquired resistance, which has implications for subsequent therapy after disease progression on PARPi. We investigated both aspects in tissue samples using RAD51 IHC, which is highly convenient and cost-effective and can, therefore, be readily incorporated into clinical settings. Our data showed that RAD51 IHC status, unlike the GSS, effectively predicted the response of patients with BRCA mutations to PARPi therapy. Furthermore, in matched pre- and post-PARPi samples, the HR capacity indicated by the RAD51 H-score increased post-progression, suggesting the potential for stratification of post-progression therapy in the PARPi era.

Several studies have functionally assessed HR based on RAD51 expression, mostly in breast cancer (15, 20–24). In ovarian cancer, one of the earliest studies examined RAD51 IF expression after irradiating patient-derived primary cells to predict primary chemotherapy response and survival (32). However, the research in RAD51 foci assessment for ovarian cancer has been rapidly evolving, partly triggered by the need to make the assay more accurate and practical for clinical use. For instance, Compadre et al. showed the feasibility and consistency of RAD51 IF across different sample types, such as cell lines, organoids, and FFPE samples (33). Pikkussari et al. assessed RAD51 IF across different time points in newly diagnosed ovarian cancer patients receiving neoadjuvant chemotherapy (34). Both studies attempted automate the counting process. An alternative to the IF-based approach is IHC, which has also been explored in ovarian cancer. For instance, Hoppe et al. conducted one of the largest studies with quantitative IHC in ovarian cancer, using multispectral imaging and an automated approach to study RAD51 foci in FFPE tissues from a platinum monotherapy trial (35). These studies suggest the need to study RAD51 as biomarker, specifically in the context of PARPi treatment. A recent study by Guffanti et al. showed that RAD51 foci with IF correlates with olaparib response in PDX models (36). However, to the best of our knowledge, no study has examined RAD51 IHC expression and its time-lagged changes in FFPE samples from patients with ovarian cancer treated with PARPi.

Besides assessing RAD51 IHC, we analyzed its correlation with the GSS in patients who underwent both tests. RAD51 IHC status outperformed the GSS in accurately predicting the response to PARPi. The limitations of genomic markers, failure to reflect longitudinal and dynamic changes in HR capacity and need for real-time functional markers have been recognized (37). Exposure to three cycles of neoadjuvant therapy can modulate the HR status (11). The degree of HR capacity in each patient may differ after first-line platinum-based chemotherapy immediately before PARPi therapy initiation. Nevertheless, a comparison between the GSS and RAD51 H-score indicated that a RAD51-low status was associated with a high GSS, both being associated with a favorable response to PARPi. Furthermore, the RAD51 H-score was negatively correlated with individual GSS components, particularly LST. Further functional and biological studies may help elucidate the reason for this observation.

Notably, our data suggested that the RAD51 IHC status can identify poor responders to PARPi among patients with deleterious BRCA mutations who are expected to respond well to PARPi monotherapy. Currently approved biomarker testing, including the Myriad HRD test, considers all BRCA-mutated patients as HR-deficient; that is, currently, the BRCA status overrides any other test aiming to identify favorable candidates for PARPi therapy. However, our findings suggest that identifying poor responders to PARPi may be equally important, especially considering that patients with high RAD51 expression account for approximately one-third of patients with BRCA mutations. Patients with high RAD51 expression before treatment may benefit from PARPi-based combination therapies, such as PARPi combined with anti-angiogenic agents (38), immune checkpoint inhibitors (39), or HR inhibitors (40, 41).

Using matched pre- and post-PARPi- samples, we observed an increase in the HR capacity, albeit with varying magnitudes. This trend was particularly pronounced in patients with low pre-treatment values. The overall increasing trend is expected from the implication of acquired HR proficiency as part of the PARPi resistance mechanism (4). In case of a steep increase in the RAD51 H-score upon PARPi progression, as demonstrated in the exemplary cases, agents other than PARPi or platinum-based chemotherapy, such as targeted therapy, may help improve the outcome of subsequent therapy. Conversely, approximately one-third of the cohort exhibited high pre-PARPi RAD51 H-scores. This suggested that HR restoration could have occurred before PARPi administration due to factors such as prior chemotherapy or other potential mechanisms of BRCA reversion (16, 17). In these patients with high pre-PARPi RAD51 H-scores, RAD51 H-score did not necessarily increase post-treatment, with several instances showing a decreasing trend. We postulate that in a heavily pre-treated setting or in instances where the pre-PARPi RAD51 H-score is already high, the post-PARPi RAD51 H-score will be less interpretable because of confounders. To better understand the dynamics, we think that further prospective studies with more patients with matched samples, involving serial collection of pre- and post-PARPi samples starting from the initial diagnosis and continuing through various treatment stages, are needed.

The limitations of this study include the small sample size and its retrospective nature. Given that RAD51 expression was evaluated only in proliferating cells that had DNA double-strand breaks, 20% of the samples were excluded from analysis. These failed samples indicated inherent biological limitations that can be expected from RAD51 IHC. Therefore, the sample size was smaller than expected, and only a subgroup of patients underwent both RAD51 IHC and GSS tests or had matched samples. Moreover, the GSS was based on WES of FFPE tissue, which is not a clinically established method. We based the cutoff of 42 on our previous experience with this method (12, 42), but this cutoff is still arbitrary. Therefore, any conclusions which relate to GSS can only be considered as hypothesis-generating. In the future, a more rigorous approach to assess HRD status should be taken. For example, a more established methods such as Myriad HRD test or HRDetect, or simultaneous testing of different HRD testing approaches could be implemented to get a robust assessment of GSS/HRD status. We assessed bio-banked tissues of all patients with ovarian cancer who received PARPi at our institution over a specific timeframe. In cases where several biopsy specimens were obtained at a single timepoint, we utilized one representative tissue with the largest tumor component. Given the retrospective nature of the study design, there may have been a selection bias.

Given the expanding therapeutic potential of PARPi beyond ovarian cancer, it will be interesting to validate the RAD51 IHC test in other cancer types, such as breast, prostate, and pancreatic cancer. Furthermore, a dedicated functional study, which simultaneously assesses RAD51 via IHC and conducting RNAseq on newly collected, prospectively obtained samples, would yield more definitive insights. Once validated, the RAD51 IHC can be applied to a prospective trial to help stratify patients identify patients who may not respond well to PARPi despite the presence of BRCA mutation and provide clues for the choice of subsequent therapy after progression on PARPi.