Purpose:

Homologous recombination (HR) deficiency (HRD) is one of the key determinants of PARP inhibitor response in ovarian cancer, and its accurate detection in tumor biopsies is expected to improve the efficacy of this therapy. Because HRD induces a wide array of genomic aberrations, mutational signatures may serve as a companion diagnostic to identify PARP inhibitor–responsive cases.

Experimental Design:

From the The Cancer Genome Atlas (TCGA) whole-exome sequencing (WES) data, we extracted different types of mutational signature–based HRD measures, such as the HRD score, genome-wide LOH, and HRDetect trained on ovarian and breast cancer–specific sequencing data. We compared their performance to identify BRCA1/2-deficient cases in the TCGA ovarian cancer cohort and predict survival benefit in platinum-treated, BRCA1/2 wild-type ovarian cancer.

Results:

We found that the HRD score, which is based on large chromosomal alterations alone, performed similarly well to an ovarian cancer–specific HRDetect, which incorporates mutations on a finer scale as well (AUC = 0.823 vs. AUC = 0.837). In an independent cohort these two methods were equally accurate predicting long-term survival after platinum treatment (AUC = 0.787 vs. AUC = 0.823). We also found that HRDetect trained on ovarian cancer was more accurate than HRDetect trained on breast cancer data (AUC = 0.837 vs. AUC = 0.795; P = 0.0072).

Conclusions:

When WES data are available, methods that quantify only large chromosomal alterations such as the HRD score and HRDetect that captures a wider array of HRD-induced genomic aberrations are equally efficient identifying HRD ovarian cancer cases.

Translational Relevance

Ovarian cancer cases that initially respond to platinum-based therapy are eligible for PARP inhibitor–based maintenance. The clinical benefit of PARP inhibitor treatment significantly varies among patients. Cases with homologous recombination (HR) deficiency (HRD), especially those with loss-of-function mutations of BRCA1 or BRCA2, benefit the most, while cases without HRD only moderately benefit. The clinical challenge is to identify the well-responding patients. Since HRD induces a wide array of genomic aberrations, mutational signatures may serve as a companion diagnostic to identify PARP inhibitor–responsive cases. Here we use whole-exome sequencing (WES) data to compare FDA-approved methods (Myriad myChoice CDx and FoundationFocus CDxBRCA LOH) that are based on large chromosomal alterations and HRDetect, that combines HRD-induced point mutations and short indels with large-scale chromosomal alterations. We found that HRD score performed similarly well to HRDetect. SNP array–based estimates of HRD score also indicated that anatomical location of biopsy and the time point during the course of disease may influence determining HRD status.

Seventy percent to 80% of ovarian cancer cases initially respond to platinum-based therapy. When they achieve response to platinum-based chemotherapy, patients receive PARP inhibitor–based maintenance therapy with significant progression-free survival benefit (1). The clinical benefit significantly varies among patients. Cases with homologous recombination (HR) deficiency (HRD), especially those with loss-of-function mutations of BRCA1 or BRCA2 benefit the most, while cases without HRD only moderately benefit (1–3). HRD in the clinical setting is usually determined by two sets of criteria (4). The first identifies loss-of-function mutations in key HR genes, such as BRCA1 and BRCA2, by direct sequencing and when such mutations cooccur with loss of heterozygosity the cases are classified as HR deficient. However, HRD can be present in ovarian cancer even in the absence of such mutations. In these cases, a second set of criteria is used to determine HR-deficient status in the form of a specific DNA-aberration profile (“DNA scarring”), which is induced by HRD. The importance of determining HRD was recently highlighted by several clinical trials. Patients in the PRIMA trial (5), with loss-of-function mutations of BRCA1/2 benefitted the most from niraparib maintenance therapy. Patients without BRCA1/2 mutations but with HRD-associated mutational signatures benefitted less and patients without HRD benefitted the least. In the PAOLA1 trial (3), patients without HRD did not have clinical benefit at all from olaparib-based first-line maintenance therapy.

The current clinically used HRD-associated mutational signatures were developed before the widespread introduction of next-generation sequencing (NGS) and quantify three types of mainly large-scale (larger than kB range) genomic aberrations (6–8). These large-scale genomic aberrations were later adjusted to NGS-based technology and they were combined in the first FDA-approved diagnostic test, the Myriad myChoice HRD score (9). This diagnostic test was used in clinical trials to determine which patients benefit from olaparib (3) and niraparib (1, 2, 5) treatment in ovarian cancer. For rucaparib-based trials (ARIAL2 trial; ref. 10, ARIAL3 trial; ref. 11) another FDA-approved genomic aberrations–based score was used, the FoundationFocus CDxBRCA LOH.

Because these “genomic scar”–based measures were developed based on hybridization-based microarray technology, they did not capture HRD–induced mutations on a finer scale.

NGS, especially whole-genome sequencing (WES), allowed the identification of mutations down to the level of single-nucleotide variations. WES analysis of BRCA1/2-mutant breast cancer showed that functional loss of these genes induces characteristic mutational signatures on at least three levels: (i) a single-nucleotide variation–based mutational signature [“Catalogue of Somatic Mutations in Cancer (COSMIC) signature 3”; ref. 12]; (ii) a short insertions/deletions–based mutational profile, a sign of alternative repair mechanisms joining double-strand breaks in the absence of HR (13, 14); and (iii) large-scale rearrangements such as nonclustered tandem duplications of a given size range (mainly associated with BRCA1 loss-of-function) or deletions in the range of 1–10 kb (mainly associated with BRCA2 loss-of-function; ref. 15). Several of these DNA-aberration profiles are directly induced by the loss-of-function of BRCA1 or BRCA2 or other key HR genes (14, 16). It was also suggested that combining the above listed mutational signatures into a composite signature may provide a more accurate measure of HRD (HRDetect) in breast cancer (17, 18).

Using HRDetect for ovarian cancer was proposed in the original publication in which this method was described for breast cancer (17), but a WES-based ovarian cancer–specific HRDetect was not derived. Here, we determined the weights of an ovarian cancer–specific lasso logistic regression-based classifier applicable to WES data and investigated its clinical relevance.

Patients and cohorts

The various classifiers in our analysis were trained and tested utilizing data from 425 patients from The Cancer Genome Atlas (TCGA) WES (19) cohort, and 109 patients from the Pan-Cancer Analysis of Whole Genome (PCAWG) Consortium whole-genome sequencing cohort (20). To evaluate the prognostic power of HRDetect (17), a dataset of 41 patients with high-grade serous ovarian cancer, of whom 20 had exceptional long-term survival (10+ years), while 21 patients developed primary resistance to platinum-based chemotherapy and died in less than 2 years (short term; Table 1; ref. 21).

Table 1.

Summary of cohorts analyzed.

DatasetNumber of patientsNumber of samples
TCGA-WES (19) 425 425 
PCAWG-WGS (20) 109 109 
Yang and colleagues (21) 41 41 
GSE40546 (28) 16 119 
Pongor and colleagues (30) 15 
DatasetNumber of patientsNumber of samples
TCGA-WES (19) 425 425 
PCAWG-WGS (20) 109 109 
Yang and colleagues (21) 41 41 
GSE40546 (28) 16 119 
Pongor and colleagues (30) 15 

Mutation and copy-number alteration calling

For the TCGA WES cohort germline mutations were called with HaplotypeCaller, while somatic point-mutations and indels were called using Mutect2 (GATK 3.8). Allele-specific copy-number profiles were estimated by using Sequenza (22).

Genotyping – BRCA status

The mutations were annotated using InterVar (23). Variants predicted as pathogenic or likely pathogenic were considered deleterious, while variants with unknown significance were marked differently. Copy-number status of BRCA1/2 were based on Sequenza results.

Mutational signatures

Ovarian cancer–specific somatic point-mutational signatures were extracted with the deconstructSigs R package (24), by using the COSMIC signatures as a mutational-process matrix. The extraction of rearrangement signatures was executed as described previously (15). The updated single-nucleotide mutational signatures, double-base mutational signatures, and indel signatures were also extracted (25).

Genomic scar scores

The calculation of the genomics scar scores [LOH (6); large-scale transitions (LST; ref. 7); and number of telomeric allelic imbalances (ntAI; ref. 8)] were determined using the scarHRD R package (26).

HRDetect-Breast cancer-WES

The WES-based HRDetect was trained on breast cancer samples as previously described (27).

HRDetect-Ovarian cancer-WES predictor of BRCA1/BRCA2 deficiency in cancer

Because the TCGA ovarian cancer dataset has sufficient numbers of bona fide HRD cases (BRCA1/2 mutation and LOH), a modified, ovarian cancer–specific HRDetect (17) model could be created. After the identification of n = 117 samples with quiescent genomic profiles, a lasso logistic regression model was applied to log-transformed and standardized genomic features: counts of single-nucleotide mutational signatures, indel profiles, and genomic scar scores. To achieve a robust model, a ten-fold nested cross-validation (where 90% of samples were used for model parameter selection and the weights for each parameter were tested on the remaining 10% of samples) was applied of 500 sub-sampling and training iterations on the randomly preselected 80% of the cohort.

HRDetect-Ovarian cancer-WGS

Davies and colleagues (17) determined the WGS-based HRDetect trained on breast cancer data, using COSMICv2 signatures. Here we introduce HRDetect-Ovarian cancer-WGS (HRDetect-OV-WGS), trained on 109 ovarian cancer WGS samples using the COSMICv3 signatures (Supplementary Methods).

Genome-wide LOH

The proportion of the genome with LOH (gLOH) of autosomes was estimated for the investigated dataset. (Supplementary Methods; refs. 10, 11)

Heterogeneity of genomic scar scores and HRDetect

We investigated the level of heterogeneity of HRD measurements in two cohorts with multiple biopsies from the same ovarian tumors, often obtained at different time points. In the first cohort, 119 samples from 16 patients were profiled for copy-number aberrations using Affymetrix Genome-Wide SNP 6.0 arrays (28). The SNP array–based genomic scar scores were determined as described previously (29). In this cohort, biopsies were obtained using different surgical intervention techniques (laparoscopic surgery, primary surgery, interval debulking surgery), from different anatomical locations (omentum, peritoneum) or at different time points (primary surgery vs. relapse).

In the second cohort (30), three DNA samples were extracted from the same primary surgery material in the following ways and whole-exome sequenced: (i) DNA extracted from a small diagnostic biopsy size tissue core, (ii) DNA extracted from a larger sub-region of the tumor (“local sample”), and (iii) DNA extracted from the entire tumor. Features were determined as in the case of the TCGA dataset.

Frequency of BRCA1/2 gene aberrations in ovarian cancer in the TCGA whole-exome sequenced cohort

In the 425 TCGA ovarian cancer samples, 101 showed biallelic loss-of-function of the BRCA1 (Fig. 1; BRCA1 germline mutation and LOH was detected in 42 cases, BRCA1 somatic mutation and LOH was detected in 19 cases, BRCA1 promoter hypermethylation and LOH was detected in 40 cases). Furthermore, we detected biallelic loss of the wild-type (WT) BRCA2 in 40 cases (BRCA2 germline mutation and LOH was detected in 30 cases, BRCA2 somatic mutation and LOH was detected in 10 cases). These cases served as positive control for validating the HRD-associated mutational signatures. We also selected 117 cases with quiescent genome as described (17) with likely HRD proficiency, which served as negative controls. There were several other samples with BRCA1/2 mutations with unknown clinical significance (BRCA1: 17, BRCA2: 30 cases; Supplementary Table S1). In the cohort with clinical response data (21), nine tumors had biallelic inactivation of either BRCA1 or BRCA2 deficiency.

Figure 1.

HRDetect-OV-WES and its performance in TCGA cohort. A, Box plots of the weights for the genomic features contributing to the HRDetect-OV-WES predictor. The range of values from 500 replicates of training in cross-validation is shown. Red crosses indicate the final weights used in HRDetect-OV-WES. In the box plots, the midline represents the median, the two edges of the box represent the lower and upper interquartile range (IQR), upper whisker = min[max(x), Q3 + 1.5 × IQR], and lower whisker = max[min(x), Q1 − 1.5 × IQR]. B, Box plots of the HRDetect-OV-WES scores of 425 ovarian cancer samples grouped by their detailed BRCA status. germ, germline; som, somatic; mut., mutation. C, The HRDetect-OV-WES scores of 425 ovarian cancer samples ordered from lowest to highest score across the x-axis from left to right. Colored bars represent both samples with monoallelic and biallelic mutations, as well as cases with BRCA mutation of unknown pathologic significance. The black asterisks above the bars indicate samples with BRCA1/2 mutation with unknown significance and LOH, and crosses represent single-allele mutations of BRCA1/2. signif., significance. D, The HRD scores of 425 ovarian cancer samples. E, ROC curves demonstrating the performance of HRDetect-OV-WES, HRDetect-BC-WES (retrained on TCGA breast cancer WES data), HRD score, and gLOH predicting biallelic BRCA loss (including all 425 samples). F, ROC curves demonstrating the performance of HRDetect-OV-WES, HRDetect-BC-WES, HRD score, and gLOH predicting biallelic BRCA loss (including only BRCA-deficient and quiescent samples).

Figure 1.

HRDetect-OV-WES and its performance in TCGA cohort. A, Box plots of the weights for the genomic features contributing to the HRDetect-OV-WES predictor. The range of values from 500 replicates of training in cross-validation is shown. Red crosses indicate the final weights used in HRDetect-OV-WES. In the box plots, the midline represents the median, the two edges of the box represent the lower and upper interquartile range (IQR), upper whisker = min[max(x), Q3 + 1.5 × IQR], and lower whisker = max[min(x), Q1 − 1.5 × IQR]. B, Box plots of the HRDetect-OV-WES scores of 425 ovarian cancer samples grouped by their detailed BRCA status. germ, germline; som, somatic; mut., mutation. C, The HRDetect-OV-WES scores of 425 ovarian cancer samples ordered from lowest to highest score across the x-axis from left to right. Colored bars represent both samples with monoallelic and biallelic mutations, as well as cases with BRCA mutation of unknown pathologic significance. The black asterisks above the bars indicate samples with BRCA1/2 mutation with unknown significance and LOH, and crosses represent single-allele mutations of BRCA1/2. signif., significance. D, The HRD scores of 425 ovarian cancer samples. E, ROC curves demonstrating the performance of HRDetect-OV-WES, HRDetect-BC-WES (retrained on TCGA breast cancer WES data), HRD score, and gLOH predicting biallelic BRCA loss (including all 425 samples). F, ROC curves demonstrating the performance of HRDetect-OV-WES, HRDetect-BC-WES, HRD score, and gLOH predicting biallelic BRCA loss (including only BRCA-deficient and quiescent samples).

Close modal

Features of BRCA deficiency in ovarian cancer WES data

The myChoice HRD score consists of three large-scale (>1 kb) genomic aberrations–based components: (i) the HRD-LOH score (6), (ii) the LST score (7), and (iii) the ntAI (8). As expected, all three components showed significant increase in the BRCA1- and BRCA2-deficient cases (Supplementary Fig. S1).

In addition to the above-listed large-scale genomics aberrations, HRD also induced mutations at a finer scale such as single-nucleotide variations or indels ranging from 1 to over 10 bp (14). The first COSMIC v2 contained only single-nucleotide substitution profiles. Consistent with previous publications (13, 31, 32), the number of mutations from single-nucleotide mutational signature 3 (12) was significantly increased in the BRCA1/2-deficient cases (Supplementary Fig. S2). In addition to the number of deletions, the number of deletions longer than 9-bp and the number of microhomology-mediated deletions were also significantly elevated in the BRCA1/2-deficient cases as described before (refs. 9, 31, 32; Supplementary Figs. S1 and S2).

Recently, the list of signatures of mutational processes in human cancer was updated and it now includes single- and double-base substitution and short insertion/deletion signatures as well (COSMIC mutational signatures 3.1; ref. 25). Using this updated list of signatures, we found that single-nucleotide substitution signatures SBS3 and SBS8, double-nucleotide substitution signatures DBS2 and DBS9, and the indel signature ID8 were significantly increased in the cases with loss-of-function of BRCA1/2 (Supplementary Figs. S3–S6).

HRDetect-OV-WES score

The original HRDetect classifier was trained on WGS data derived from fresh frozen material (17). Currently, however, in most clinical trial cohorts only formalin-fixed, paraffin-embedded (FFPE)-derived WES is available for further analysis. In principle, HRDetect can be derived from WES data, albeit with a more restricted set of features to start with. We trained an ovarian cancer WES-specific HRDetect model based on both sets [version 2 (12) and version 3 (25)] of COSMIC-mutational signatures. Since version 2 contains only single-nucleotide substitution signatures, we added several other mutational features for the analysis, such as microhomology-mediated deletions, etc. (for details see Supplementary Methods). The model was trained on cases with BRCA1/2 deficiency versus cases with quiet genome as outlined above. The resultant logistic regression-based model contained seven mutational features: the number of LSTs, the number of microhomology-mediated deletions (at least 2 bp long) to total number of deletions ratio, the number of HRD-LOH, the number of ntAI, the number of at least 10 bp–long deletions to total number of deletions ratio, and the single-nucleotide substitution signature 3 (Fig. 1A). We also derived the ovarian cancer WES HRDetect based on the new version of COSMIC signatures combined with the number of LSTs, HRD-LOH, and telomeric allelic imbalances (TAI). Details are described in the Supplementary Methods. Interestingly, this version of WES HRDetect contained only LST, HRD-LOH, TAI, and DBS9 and thus HRDetect trained on the new version of COSMIC signatures derived from WES data was essentially reduced to the HRD score. As expected, the majority of samples with a biallelic loss of WT BRCA1/2 had a higher than 0.7 HRDetect-OV-WES score. Cases with single-allele mutations and cases with BRCA1/2 mutations with unknown significance had a significantly lower average HRDetect-OV-WES score (Fig. 1B and C). We compared the performance of HRDetect trained on ovarian cancer–derived data (HRDetect-OV-WES) versus HRDetect trained on breast cancer–derived data (HRDetect-BC-WES). We found that the HRDetect-OV-WES score was a significantly better predictor of BRCA1/2 deficiency than HRDetect-BC-WES (ROC analyses: HRDetect-BC-WES: AUC = 0.795, HRDetect-OV-WES: AUC = 0.837; P = 0.0073; Fig. 1D). This suggests that transferring HRDetect from one tumor type to another may be less accurate than an HRDetect predictor trained on data derived from the same tumor type. We also compared the performance of HRDetect-OV-WES with another capture-based sequencing–derived measure of HRD gLOH (10, 11). We found that both HRDetect-OV-WES and HRD score identified BRCA1/2-deficient cases more accurately than gLOH (Fig. 1E and F)

HRD-associated mutational signatures in WGS data

WGS data contain significantly more HRD-associated mutations than capture-based sequencing including WES, such as tandem duplications or larger than 1-kb deletions (15). These are usually not detected by components of the HRD score. We investigated whether WGS-based HRDetect is less sensitive to the constraints seen for WES data in the previous section.

We compared the original HRDetect trained on breast cancer–derived WGS data, and HRDetect trained on WGS data from 109 ovarian cancer cases. The HRDetect-WGS-OV used different features and weights (Supplementary Figs. S7 and S8) compared with the previously published breast cancer–specific HRDetect but did not significantly outperform it (Supplementary Fig. S8).

Based on the analysis of 29 ovarian cancer cases with both WGS and WES data, the genomic scar scores showed an overall good correlation (between r = 0.61 and r = 0.69) but in some cases the WGS- and WES-based scores produced discrepant results (Supplementary Fig. S9). These discrepancies are likely due to differences in the output of the segmentation algorithms, and our analysis showed that the HRD-LOH (6) and the HRD score was not significantly higher when calculated based on WES data (Supplementary Fig. S9).

Finally, we compared HRDetect values derived from WES and WGS data on the 29 cases for which both types of sequencing data were available. This subcohort had 10 BRCA1/2-deficient cases. WES- and WGS-derived HRDetect values showed an overall correlation of r = 0.71. The two methods produced more than 0.7 HRDetect values on 6 of the 10 BRCA1/2 deficient cases with both methods giving a <0.7 value for 2 different cases each (Supplementary Fig. S10). While the small number of samples precludes a definitive conclusion, these preliminary results suggest that the WES-derived HRDetect for ovarian cancer may not have a significantly inferior accuracy of identifying HRD cases.

HRDetect-OV-WES score predicting survival after platinum-based treatment

The Yang and colleagues dataset contains WES profiles of 41 platinum-treated patients with ovarian cancer. The majority of these patients had no mutation in BRCA1/2 and the cohort displayed a high level of variation in survival benefit after platinum treatment. We calculated the HRD scores and HRDetect-WES-OV values (i.e., using HRDetect trained on the ovarian cancer WES cohort as described above) for each patient in this cohort (Fig. 2). As expected, the 9 BRCA1/2-mutant cases had a significant survival benefit from treatment (>2 years). However, another 10 cases without BRCA1/2 mutations also showed similar survival benefit. On the other hand, 21 patients, all without BRCA1/2 mutations, showed significantly shorter (<2 years) survival benefit. Eight of the 10 BRCA1/2 WT cases with longer survival had a higher than 0.7 HRDetect-WES-OV score. In case of the short survival [overall survival (OS) <2 years] group, 12 of the 21 patients had less than 0.7 HRDetect score. (Fig. 2A and B). The HRDetect-OV-WES model had an AUC of 83.9% (confidence interval, 71.1%–96.8%) predicting long survival benefit from platinum treatment. The HRDetect-OV-WES model significantly outperformed the breast cancer sequencing data–trained HRDetect (P = 0.039; Fig. 2C).

Figure 2.

Performance of HRDetect-OV-WES predicting survival after platinum-based treatment. A, The HRDetect-OV-WES scores of 41 ovarian cancer samples of the Yang and colleagues' cohort ordered from lowest to highest score across the x-axis from left to right. Colored bars represent the survival group and the BRCA status. B, Box plots of the HRDetect-OV-WES scores of 41 ovarian cancer samples of the Yang and colleagues' cohort grouped by their BRCA status and survival. C, ROC curves demonstrating the performance of HRDetect-OV-WES, HRDetect-BC-WES, HRD score, and gLOH predicting extremely long survival (OS > 10 years) following platinum-based treatment.

Figure 2.

Performance of HRDetect-OV-WES predicting survival after platinum-based treatment. A, The HRDetect-OV-WES scores of 41 ovarian cancer samples of the Yang and colleagues' cohort ordered from lowest to highest score across the x-axis from left to right. Colored bars represent the survival group and the BRCA status. B, Box plots of the HRDetect-OV-WES scores of 41 ovarian cancer samples of the Yang and colleagues' cohort grouped by their BRCA status and survival. C, ROC curves demonstrating the performance of HRDetect-OV-WES, HRDetect-BC-WES, HRD score, and gLOH predicting extremely long survival (OS > 10 years) following platinum-based treatment.

Close modal

The HRD score, the WES-trained ovarian cancer–specific HRDetect, and WES-trained ovarian cancer gLOH produced similarly accurate predictions for survival benefit in this data set.

Heterogeneity of genomic scar scores and HRDetect

Therapeutic decisions are often based on a single diagnostic biopsy. However, HRD may change during tumor progression and it could also be impacted by intratumor heterogeneity. Therefore, we investigated the level of heterogeneity of HRD measurements in two cohorts. In the SNP array–based cohort (21), individual patients had up to 27 biopsies taken from different locations or at different time points and profiled. In 8 of the 16 cases the HRD score values for an individual patient showed significant variation with values both above and below the currently accepted threshold of HRD of 42 (Supplementary Fig. S11; ref. 9).

In the second, WES-based cohort (30), where three DNA samples were extracted from the same primary surgery material, the HRD measurements showed a high level of consistency (Supplementary Fig. S12).

PARP inhibitors have a significant clinical benefit in ovarian cancer (1–3, 5, 33). It is also clear, however, that ovarian cancers with different HR competence show different sensitivity to this treatment (2, 3, 34). Therefore, developing accurate measures of HRD may greatly improve the efficacy of targeted therapy in ovarian cancer.

The currently FDA-approved companion diagnostics for HR deficiency, myChoice HRD score, and FoundationFocus CDxBRCA LOH, were originally derived from hybridization microarray-based measurements and quantified the number of HRD-associated, relatively large scale (>1kb) copy-number and/or LOH variations. These methods have several technical advantages, such as they are performed on FFPE material (as opposed to fresh frozen material) and the lack of need for germline DNA sequencing.

With the widespread introduction of NGS (whole-exome and whole-genome), it has become apparent that HRD, caused by the functional loss of key HR enzymes (BRCA1, BRCA2, PALB2, etc.) also induces genomics aberrations at a finer scale, single-nucleotide variations, deletions/insertions as small as 1 bp, etc. (13–15). Based on this observation it was proposed that HRD measures that also considered these finer-scale genomic aberrations may provide a more accurate detection of HRD in the clinical setting (17). Initial analysis (18, 35) validated the ability of this method to identify HR-deficient cases in the clinical setting. However, the more complex nature of HRDetect also posed additional technical complications. Sequencing germline DNA is necessary to extract, e.g., the SNV-based signatures and the weights of HRDetect may need to be adjusted to a given tumor type. We evaluated HRDetect in this context for ovarian cancer. We found that transferring the weights trained of breast cancer WES data produces inferior results identifying BRCA1/2 deficiency in ovarian cancer relative to retraining HRDetect on ovarian cancer–derived WES data. This may be due to a number of biological differences between breast and ovarian cancer. For example, there are several SNV-based mutational signatures that are present in either breast or ovarian cancer but not in both (12, 25), which will likely have an impact on the extraction of SNV mutational signatures. We also found that the technically less challenging HRD score performed as well as the WES-derived HRDetect both for identifying BRCA1/2-deficient ovarian cancer cases and predicting survival benefit after platinum treatment in BRCA1/2 WT cases. It remains to be seen, however, whether HRD score or the ovarian cancer WES-derived HRDetect predicts benefit from PARP inhibitor therapy with a higher accuracy. In principle there are several existing cohorts (3, 5) that could be used to investigate this question if the biopsies were profiled by NGS.

Intratumor heterogeneity has been widely investigated as a confounding factor to obtain accurate diagnostic measurements from a single biopsy (36). It was shown previously that in breast cancer (37) the HRD score showed limited intratumor heterogeneity. Here we observed a similar, low level of variation of the HRD score when analyzed within the same primary tumor. However, biopsies obtained at different time points from different anatomical locations from the same patient often produce conflicting HRD status, which warrants further studies and a careful evaluation of the diagnostic decision process of PARP inhibitor therapy in ovarian cancer.

Z. Sztupinszki reports grants from Velux during the conduct of the study. J. Borcsok reports grants from Velux Foundations during the conduct of the study. M.R. Mirza reports personal fees from AstraZeneca, GSK, and Zailab outside the submitted work. Z. Szallasi reports grants from Breast Cancer Research Foundation, Research and Technology Innovation Fund, Novo Nordisk Foundation Interdisciplinary Synergy Programme Grant, Department of Defense through the Prostate Cancer Research Program, Det Frie Forskningsråd, Sundhed og Sygdom, Kræftens Bekæmpelses Videnskabelige Udvalg, and Basser Foundation during the conduct of the study; in addition, Z. Szallasi is an inventor on a patent used in the myChoice HRD assay issued and with royalties paid. No disclosures were reported by the other authors.

Z. Sztupinszki: Conceptualization, investigation, visualization, methodology, writing–original draft. M. Diossy: Investigation. J. Borcsok: Investigation. A. Prosz: Investigation. N. Cornelius: Writing–original draft. M.K. Kjeldsen: Writing–original draft. M.R. Mirza: Conceptualization, writing–original draft. Z. Szallasi: Conceptualization, writing–original draft.

This work was supported by the Research and Technology Innovation Fund (KTIA_NAP_13–2014–0021 and 2017–1.2.1-NKP-2017–00002, to Z. Szallasi), Breast Cancer Research Foundation (BCRF-20–159, to Z. Szallasi), the Novo Nordisk Foundation Interdisciplinary Synergy Programme Grant (NNF15OC0016584, to Z. Szallasi), Kræftens Bekæmpelses Videnskabelige Udvalg (award number R281-A16566, to Z. Szallasi), Det Frie Forskningsråd, Sundhed og Sygdom (award number, 7016–00345B, to Z. Szallasi), Department of Defense through the Prostate Cancer Research Program (award number W81XWH-18–2-0056, to Z. Szallasi), and Basser Foundation (to Z. Szallasi). Z. Sztupinszki and J. Borcsok were supported by Velux Foundation 00018310 grant. The results shown here are based upon data generated by the TCGA Research Network: http://cancergenome.nih.gov/ and the International Cancer Genome Consortium (ICGC): https://icgc.org/.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Mirza
MR
,
Coleman
RL
,
Gonzalez-Martin
A
,
Moore
KN
,
Colombo
N
,
Ray-Coquard
I
, et al
The forefront of ovarian cancer therapy: update on PARP inhibitors
.
Ann Oncol
2020
;
31
:
1148
59
.
2.
Mirza
MR
,
Monk
BJ
,
Herrstedt
J
,
Oza
AM
,
Mahner
S
,
Redondo
A
, et al
Niraparib maintenance therapy in platinum-sensitive, recurrent ovarian cancer
.
N Engl J Med
2016
;
375
:
2154
64
.
3.
Ray-Coquard
I
,
Pautier
P
,
Pignata
S
,
Perol
D
,
Gonzalez-Martin
A
,
Berger
R
, et al
Olaparib plus bevacizumab as first-line maintenance in ovarian cancer
.
N Engl J Med
2019
;
381
:
2416
28
.
4.
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
.
5.
Gonzalez-Martin
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
.
6.
Abkevich
V
,
Timms
KM
,
Hennessy
BT
,
Potter
J
,
Carey
MS
,
Meyer
LA
, et al
Patterns of genomic loss of heterozygosity predict homologous recombination repair defects in epithelial ovarian cancer
.
Br J Cancer
2012
;
107
:
1776
82
.
7.
Popova
T
,
Manie
E
,
Rieunier
G
,
Caux-Moncoutier
V
,
Tirapo
C
,
Dubois
T
, et al
Ploidy and large-scale genomic instability consistently identify basal-like breast carcinomas with BRCA1/2 inactivation
.
Cancer Res
2012
;
72
:
5454
62
.
8.
Birkbak
NJ
,
Wang
ZC
,
Kim
JY
,
Eklund
AC
,
Li
Q
,
Tian
R
, et al
Telomeric allelic imbalance indicates defective DNA repair and sensitivity to DNA-damaging agents
.
Cancer Discov
2012
;
2
:
366
75
.
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
.
10.
Swisher
EM
,
Lin
KK
,
Oza
AM
,
Scott
CL
,
Giordano
H
,
Sun
J
, et al
Rucaparib in relapsed, platinum-sensitive high-grade ovarian carcinoma (ARIEL2 Part 1): an international, multicentre, open-label, phase 2 trial
.
Lancet Oncol
2017
;
18
:
75
87
.
11.
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
.
12.
Alexandrov
LB
,
Nik-Zainal
S
,
Wedge
DC
,
Aparicio
SA
,
Behjati
S
,
Biankin
AV
, et al
Signatures of mutational processes in human cancer
.
Nature
2013
;
500
:
415
21
.
13.
Nik-Zainal
S
,
Alexandrov
LB
,
Wedge
DC
,
Van Loo
P
,
Greenman
CD
,
Raine
K
, et al
Mutational processes molding the genomes of 21 breast cancers
.
Cell
2012
;
149
:
979
93
.
14.
Zamborszky
J
,
Szikriszt
B
,
Gervai
JZ
,
Pipek
O
,
Poti
A
,
Krzystanek
M
, et al
Loss of BRCA1 or BRCA2 markedly increases the rate of base substitution mutagenesis and has distinct effects on genomic deletions
.
Oncogene
2017
;
36
:
746
55
.
15.
Nik-Zainal
S
,
Davies
H
,
Staaf
J
,
Ramakrishna
M
,
Glodzik
D
,
Zou
X
, et al
Landscape of somatic mutations in 560 breast cancer whole-genome sequences
.
Nature
2016
;
534
:
47
54
.
16.
Poti
A
,
Gyergyak
H
,
Nemeth
E
,
Rusz
O
,
Toth
S
,
Kovacshazi
C
, et al
Correlation of homologous recombination deficiency induced mutational signatures with sensitivity to PARP inhibitors and cytotoxic agents
.
Genome Biol
2019
;
20
:
240
.
17.
Davies
H
,
Glodzik
D
,
Morganella
S
,
Yates
LR
,
Staaf
J
,
Zou
X
, et al
HRDetect is a predictor of BRCA1 and BRCA2 deficiency based on mutational signatures
.
Nat Med
2017
;
23
:
517
25
.
18.
Staaf
J
,
Glodzik
D
,
Bosch
A
,
Vallon-Christersson
J
,
Reutersward
C
,
Hakkinen
J
, et al
Whole-genome sequencing of triple-negative breast cancers in a population-based clinical study
.
Nat Med
2019
;
25
:
1526
33
.
19.
Cancer Genome Atlas Research Network
. 
Integrated genomic analyses of ovarian carcinoma
.
Nature
2011
;
474
:
609
15
.
20.
ICGC/TCGA Pan-Cancer Analysis of Whole Genomes Consortium
. 
Pan-cancer analysis of whole genomes
.
Nature
2020
;
578
:
82
93
.
21.
Yang
SYC
,
Lheureux
S
,
Karakasis
K
,
Burnier
JV
,
Bruce
JP
,
Clouthier
DL
, et al
Landscape of genomic alterations in high-grade serous ovarian cancer from exceptional long- and short-term survivors
.
Genome Med
2018
;
10
:
81
.
22.
Favero
F
,
Joshi
T
,
Marquard
AM
,
Birkbak
NJ
,
Krzystanek
M
,
Li
Q
, et al
Sequenza: allele-specific copy number and mutation profiles from tumor sequencing data
.
Ann Oncol
2015
;
26
:
64
70
.
23.
Li
Q
,
Wang
K
. 
InterVar: clinical interpretation of genetic variants by the 2015 ACMG-AMP guidelines
.
Am J Hum Genet
2017
;
100
:
267
80
.
24.
Rosenthal
R
,
McGranahan
N
,
Herrero
J
,
Taylor
BS
,
Swanton
C
. 
DeconstructSigs: delineating mutational processes in single tumors distinguishes DNA repair deficiencies and patterns of carcinoma evolution
.
Genome Biol
2016
;
17
:
31
.
25.
Alexandrov
LB
,
Kim
J
,
Haradhvala
NJ
,
Huang
MN
,
Tian Ng
AW
,
Wu
Y
, et al
The repertoire of mutational signatures in human cancer
.
Nature
2020
;
578
:
94
101
.
26.
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
:
16
.
27.
Diossy
M
,
Reiniger
L
,
Sztupinszki
Z
,
Krzystanek
M
,
Timms
KM
,
Neff
C
, et al
Breast cancer brain metastases show increased levels of genomic aberration-based homologous recombination deficiency scores relative to their corresponding primary tumors
.
Ann Oncol
2018
;
29
:
1948
54
.
28.
Schwarz
RF
,
Ng
CK
,
Cooke
SL
,
Newman
S
,
Temple
J
,
Piskorz
AM
, et al
Spatial and temporal heterogeneity in high-grade serous ovarian cancer: a phylogenetic analysis
.
PLoS Med
2015
;
12
:
e1001789
.
29.
Marquard
AM
,
Eklund
AC
,
Joshi
T
,
Krzystanek
M
,
Favero
F
,
Wang
ZC
, et al
Pan-cancer analysis of genomic scar signatures associated with homologous recombination deficiency suggests novel indications for existing cancer drugs
.
Biomark Res
2015
;
3
:
9
.
30.
Pongor
LS
,
Munkacsy
G
,
Vereczkey
I
,
Pete
I
,
Gyorffy
B
. 
Currently favored sampling practices for tumor sequencing can produce optimal results in the clinical setting
.
Sci Rep
2020
;
10
:
14403
.
31.
Decker
B
,
Karyadi
DM
,
Davis
BW
,
Karlins
E
,
Tillmans
LS
,
Stanford
JL
, et al
Biallelic BRCA2 mutations shape the somatic mutational landscape of aggressive prostate tumors
.
Am J Hum Genet
2016
;
98
:
818
29
.
32.
Sztupinszki
Z
,
Diossy
M
,
Krzystanek
M
,
Borcsok
J
,
Pomerantz
MM
,
Tisza
V
, et al
Detection of molecular signatures of homologous recombination deficiency in prostate cancer with or without BRCA1/2 mutations
.
Clin Cancer Res
2020
;
26
:
2673
80
.
33.
Mirza
MR
,
Pignata
S
,
Ledermann
JA
. 
Latest clinical evidence and further development of PARP inhibitors in ovarian cancer
.
Ann Oncol
2018
;
29
:
1366
76
.
34.
Moore
KN
,
Secord
AA
,
Geller
MA
,
Miller
DS
,
Cloven
N
,
Fleming
GF
, et al
Niraparib monotherapy for late-line treatment of ovarian cancer (QUADRA): a multicentre, open-label, single-arm, phase 2 trial
.
Lancet Oncol
2019
;
20
:
636
48
.
35.
Zhao
EY
,
Shen
Y
,
Pleasance
E
,
Kasaian
K
,
Leelakumari
S
,
Jones
M
, et al
Homologous recombination deficiency and platinum-based therapy outcomes in advanced breast cancer
.
Clin Cancer Res
2017
;
23
:
7521
30
.
36.
Biswas
D
,
Birkbak
NJ
,
Rosenthal
R
,
Hiley
CT
,
Lim
EL
,
Papp
K
, et al
A clonal expression biomarker associates with lung cancer mortality
.
Nat Med
2019
;
25
:
1540
8
.
37.
von Wahlde
MK
,
Timms
KM
,
Chagpar
A
,
Wali
VB
,
Jiang
T
,
Bossuyt
V
, et al
Intratumor heterogeneity of homologous recombination deficiency in primary breast cancer
.
Clin Cancer Res
2017
;
23
:
1193
9
.