Homologous Recombination Deficiency Landscape of Breast Cancers and Real-World Effectiveness of Poly ADP-Ribose Polymerase Inhibitors in Patients With Somatic BRCA1/2, Germline PALB2, or Homologous Recombination Deficiency Signature

Abstract

PURPOSE

Poly ADP-ribose polymerase inhibitors (PARPi) are approved for patients with human epidermal growth factor receptor 2–negative metastatic breast cancer (mBC) and germline pathogenic/likely pathogenic variant (hereafter mutation) in the BRCA1/2 genes (gBRCA); however, clinical benefit has also been demonstrated in mBC with somatic BRCA1/2 mutations (sBRCA) or germline PALB2 mutations (gPALB2). This study aims to describe the genomic landscape of homologous recombination repair (HRR) gene alterations in mBC and assess PARPi treatment outcomes for patients with gBRCA compared with other HRR genes and by status of a novel homologous recombination deficiency signature (HRDsig).

METHODS

A real-world (RW) clinico-genomic database (CGDB) of comprehensive genomic profiling (CGP) linked to deidentified, electronic health record–derived clinical data was used. CGP was analyzed for HRR genes and HRDsig. The CGDB enabled cohort characterization and outcomes analyses of 177 patients exposed to PARPi. RW progression-free survival (rwPFS) and RW overall survival (rwOS) were compared.

RESULTS

Of 28,920 patients with mBC, gBRCA was detected in 3.4%, whereas the population with any BRCA alteration or gPALB2 increased to 9.5%. HRDsig+ represented 21% of patients with mBC. BRCA and gPALB2 had higher levels of biallelic loss and HRDsig+ than other HRR alterations. Outcomes on PARPi were assessed for 177 patients, and gBRCA and sBRCA/gPALB2 cohorts were similar: gBRCA versus sBRCA/gPALB2 rwPFS was 6.3 versus 5.4 months (hazard ratio [HR], 1.37 [0.77-2.43]); rwOS was 16.2 versus 21.2 months (HR, 1.45 [0.74-2.86]). Additionally, patients with HRDsig+ versus HRDsig– had longer rwPFS (6.3 v 2.8 months; HR, 0.62 [0.42-0.92]) and numerically longer rwOS (17.8 v 13.0 months; HR, 0.72 [0.46-1.14]).

CONCLUSION

Patients with sBRCA and gPALB2 derive similar benefit from PARPi as those with gBRCA alterations. In combination, HRDsig+, sBRCA, and gPALB2 represent an additional 19% of mBC that can potentially benefit from PARPi. Randomized trials exploring a more inclusive biomarker such as HRDsig are warranted.

This is real-world evidence supporting benefit of PARPi in sBRCA, gPALB2, and HRDsig.

INTRODUCTION

Poly ADP-ribose polymerase inhibitors (PARPi) are small molecules that cause synthetic lethality in tumors with homologous recombination deficiency (HRD).¹ Although PARPi were not effective in the treatment of unselected and heavily pretreated patients with metastatic triple-negative breast cancer (TNBC),² olaparib and talazoparib are beneficial to those with a germline BRCA1/2 (gBRCA) pathogenic or likely pathogenic variant (hereafter mutation; aka gBRCA carriers).^3-5

CONTEXT

• Key Objective

• Poly ADP-ribose polymerase inhibitors (PARPi) are approved for the treatment of breast cancers in patients with a germline pathogenic variant in the BRCA1/2 genes (gBRCA), which represents only a small fraction of all patients with breast cancer. We sought to quantify the frequency of alterations across homologous recombination repair-genes, the prevalence of homologous recombination deficiency signature (HRDsig), and compare outcomes with PARPi on the basis of these biomarkers.

• Knowledge Generated

• Among patients with breast cancer, the frequency of each individual gene alteration is low but the population with HRDsig is large. Compared with gBRCA, patients with sBRCA and gPALB2 have similar benefit from PARPi and HRDsig+ tumors have longer progression-free survival than HRDsig– tumors.

• Relevance

• We present real-world evidence supporting the findings of single-arm phase II studies revealing benefit of PARPi in sBRCA and gPALB2. Further, we demonstrate that HRDsig could be used to identify a much larger population of breast cancer patients who may benefit from PARPi.

In some tumor types (ie, breast and pancreatic cancers), PARPi are approved only for gBRCA carriers, and others (ie, prostate and ovarian cancers) have more inclusive criteria beyond gBRCA. For prostate cancer, olaparib and rucaparib are approved for patients with alterations (germline or somatic) in any of 14 homologous recombination repair (HRR) genes. For ovarian cancer, benefit with PARPi has been observed in patients with gBRCA and sBRCA mutations,⁶ as well as those with elevated tumor genomic instability score (GIS), genome-wide loss-of-heterozygosity, and in some patients with advanced disease and no HRD biomarkers, although the degree of benefit in HRD-negative tumors is modest.^6-10 To identify patients with metastatic breast cancer (mBC) beyond those with gBRCA mutations who may benefit from PARPi, TBCRC-048 (the Olaparib Expanded trial) evaluated olaparib in gBRCA-negative patients with a germline or somatic mutation in an HRR pathway gene, and reported an 80% response rate in patients with gPALB2 mutations and a 50% in those with somatic BRCA (sBRCA) mutations. Evaluation in larger cohorts to confirm these results are ongoing.¹¹

TBCRC-048 highlighted the difficulty in studying the predictive effect of rare gene alterations. Thus, a unifying genomic biomarker of HRD could be useful to further identify patients who might benefit from PARPi. HRD signature (HRDsig) is a novel genomic scar–based algorithm that has been shown to predict PARPi response for HRD tumors.¹²

We analyzed 28,920 mBC cases and evaluated variant characteristics, germline status, allelic status, and HRDsig, with a focus on HRR and cancer predisposition genes. Using a real-world (RW) clinico-genomic database (CGDB), we explored characteristics and outcomes for a subset of patients to identify those likely to benefit from PARPi.

The primary aims of this study were to compare outcomes in patients with gPALB2 or sBRCA mutations compared with patients with gBRCA mutations treated with PARPi. In addition, we compared outcomes between patients whose tumors had an HRDsig (HRDsig+) with those whose did not (HRDsig–).

METHODS

Comprehensive Genomic Profiling

Twenty-eight thousand nine hundred twenty consecutive centrally reviewed breast carcinomas underwent comprehensive genomic profiling (CGP) in a Clinical Laboratory Improvements Act-College of American Pathologists, New York State–regulated reference laboratory (Foundation Medicine, Inc, Cambridge, MA), as previously described.¹³ Approval for this study, including a waiver of informed consent and a Health Insurance Portability and Accountability Act waiver of authorization, was obtained from WCG institutional review board (protocol No. 20152817). Further details are provided in the Data Supplement (Methods S1).

Somatic and Germline Variant Determination

Recurrent predicted benign germline variants were excluded from all analyses; allelic status and somatic/germline status for mutations were computationally predicted without matched normal tissue with the accuracy of 95% for somatic and 99% for germline.¹⁴ HRR gene alteration allelic status was categorized as biallelic, monoallelic, or unknown as described previously.¹⁵ Further details are provided in the Data Supplement (Methods S1).

Homologous Recombination Deficiency Biomarkers

HRR genes (Data Supplement, Table S1) were defined by the inclusion criteria for the TBCRC-048 trial.¹¹ HRDsig was called using a machine learning–based algorithm.¹² Briefly, copy-number features and select indel features—as defined by Alexandrov et al¹⁶—were extracted from segmented copy-number profiles and used as inputs into an extreme gradient boosting machine learning model. Training data labels were defined as biallelic BRCA1/2 for true positives and HRR wild-type (WT for BRCA1, BRCA2, ATM, BARD1, BRIP1, CDK12, CHEK1, CHEK2, FANCL, PALB2, RAD51B, RAD51C, RAD51D, and RAD54L alterations) for true negatives. A simulated GIS was calculated from segmented copy-number profiles and represented a sum of qualifying loss of heterozygosity segments,¹⁷ telomeric allelic imbalance (TAI) events,¹⁸ and large-scale transition (LST) adjusted for ploidy¹⁹ by subtracting 15.5 × ploidy as previously described,²⁰ and the established cutoff of 42 was used to classify patients as GIS+.²⁰ Herein, we refer to GIS rather than HRD score because we calculated only the GIS component, whereas the Myriad MyChoice CDx HRD score—not used in this manuscript—is considered positive if either GIS is high (≥42) or a BRCA1/2 mutation is present. Moreover, we advocate that HRD score is a more general term, and GIS is more specific to the score that incorporates LOH, LST, and TAI.

Patient Selection and Outcomes Analysis

Clinical analyses for a subset of patients included in the genomic landscape analysis used the nationwide (US-based) deidentified Flatiron Health-Foundation Medicine breast cancer CGDB (FH-FMI CGDB). The deidentified data originated from approximately 280 US cancer clinics (approximately 800 sites of care). Retrospective longitudinal clinical data were derived from electronic health records, comprising patient-level structured and unstructured data, curated via technology-enabled abstraction of clinical notes and radiology/pathology reports, which were linked to genomic data derived from FMI testing by deidentified, deterministic matching.²¹

The FH-FMI CGDB included 11,773 patients diagnosed with breast cancer from January 2011 to March 2022 who underwent FMI CGP. We selected unique patients who met the following inclusion criteria: (1) a confirmed visit in the FH network within 90 days of metastatic diagnosis, (2) an FMI tissue CGP result, (3) 60 days or less between last FH network visit and CGP result, (4) PARPi treatment after CGP specimen collection, (5) documented assessment of RW progression (rwP), and (6) entrance to at-risk population before rwP or relevant censor date. Patients were grouped according to the presence of germline or somatic mutations in BRCA1/2 and gPALB2. Additionally, HRDsig was used to classify tumors as HRDsig+ or HRDsig– (Fig 1). Since patients with other HRR gene mutations (eg, ATM and CHEK2) have been found to not respond to PARPi, and the responsiveness of tumors with mutations in other HRR genes (eg, RAD51C/D) has not been determined in breast cancer,¹¹ patients without BRCA1/2 or gPALB2 mutations were compared based on the HRDsig status.

FIG 1.

Patient data were analyzed from two databases. FMI's genomic data set was leveraged to explore the landscape of HRR alterations and HRDsig in breast carcinoma and the FH-FMI CGDB was used to assess outcomes to PARPi. CGP, comprehensive genomic profiling; FH-FMI CGDB, Flatiron Health-Foundation Medicine breast cancer clinico-genomic database; gLOH, genome-wide loss-of-heterozygosity; HRDsig, homologous recombination deficiency signature; HRR, homologous recombination repair; PARPi, poly ADP-ribose polymerase inhibitors; SGZ, somatic germline zygosity; WT, wild-type.

rwP, RW progression-free survival (rwPFS), and RW overall survival (rwOS) definition events are described in detail in the Data Supplement (Methods S1).

RESULTS

HRR Gene Alteration Landscape

BRCA1/2-predicted inactivating alterations were detected in 8.9% (n = 2,571) of cases, with BRCA1 and BRCA2 alterations accounting for 3.9% (n = 1,126) and 4.9% (n = 1,412), respectively, whereas 0.1% (n = 33) harbored alterations in both genes. PALB2 alterations were observed in 1.4% (n = 412) of breast cancer cases. BRCA2 short coding mutations and indels (muts/indels) were the most frequently detected HRR alteration (4.3%, n = 1,217), and computational assessment classified 49% as germline (n = 594), 30% as somatic (n = 365), and 21% as unknown (n = 256). BRCA1 muts/indels were detected in 3.2% of cases (n = 908) and 43% were classified as germline (n = 386), 32% as somatic (n = 289), and 25% as unknown (n = 225). PALB2 muts/indels (n = 371) consisted of 51% germline (n = 188), 26% somatic (n = 97), and 22% unknown (n = 81) alterations. Homozygous deletions (BRCA2, n = 136; BRCA1, n = 101; PALB2, n = 9) and genomic rearrangements (BRCA2, n = 118; BRCA1, n = 178; PALB2, n = 37) were relatively uncommon compared with muts/indels, and somatic and germline predictions were not available for these alteration classes (Fig 2A). Other commonly altered HRR genes included ATM (2.8%, n = 801) and CHEK2 (2.4%, n = 683; Fig 2A). Biallelic inactivation was observed in more than 70% of tumors with gBRCA, sBRCA, and gPALB2 mutations and in fewer than 40% of tumors with sPALB2 and other HRR mutations (Fig 2B).

FIG 2.

(A) In addition to BRCA1/2 germline mutations, a diverse landscape of somatic and germline mutations, along with copy-number loss and genomic rearrangements, are detected in genes associated with homologous recombination. For cases with multiple alterations in the same gene, alteration type is prioritized in the following order: (1) germline mut/indel, somatic mut/indel, unknown mut/indel, homozygous deletion, and rearrangement. (B) Biallelic inactivation is necessary to induce HRD. Biallelic inactivation is commonly observed in all BRCA1/2 short coding variants (mut/indel) and cases with PALB2 germline mut/indel. Cases with multiple alterations in a single gene were considered biallelic, and if at least one mutation was germline, the case was classified as germline. (C) HRDsig within BRCA1, BRCA2, PALB2, and other HRR pathogenic and likely pathogenic variants compared by germline/somatic status. (D) HRDsig within BRCA1, BRCA2, PALB2, and other HRR pathogenic and likely pathogenic variants compared by biallelic/monoallelic status. (E) Frequency of HRDsig within HRR WT tumors, tumors harboring other HRR alterations (including sPALB2), or BRCA/gPALB2 alterations. For outer ring, percentages in parenthesis represent percent of full cohort. HRDsig, homologous recombination deficiency signature; HRR, homologous recombination repair; WT, wild-type.

Clinical and genomic features were compared according to gBRCA1/2, sBRCA1/2, and gPALB2 status (Table 1; Data Supplement, Table S2). Overall, no notable differences in breast cancer subtypes or sex were seen, but patients with gBRCA or gPALB2 tumors had a lower median age than their somatic counterparts, as expected.

TABLE 1.

Clinical and Genomic Characteristics for Those Patients Whose Tumors Were Classified as gBRCA, sBRCA, or gPALB2

Similar to biallelic inactivation, high rates of HRDsig positivity (HRDsig+) were seen in gBRCA-mutated (88%), sBRCA-mutated (76%), and gPALB2-mutated (77%) tumors, suggesting that all three mutation classes commonly result in functional HRD (Fig 2C). In addition, 70% of tumors with BRCA mutations of unknown origin (germline or somatic) also had high frequency of HRDsig+, whereas only 52% of sPALB2-mutated were HRDsig+, mirroring the lower rate of biallelic inactivation in these cases. Other germline and somatic HRR-mutated had substantially lower rates of HRDsig+ (both 17%; Fig 2C). Given the association of germline alterations with both allelic status and HRDsig, we investigated the association between allelic status and HRDsig for BRCA1, BRCA2, PALB2, and other HRR alterations (somatic, germline, or unknown). We found a strong association between HRDsig+ and biallelic inactivation for BRCA1, BRCA2, and PALB2 alterations (93%, 91%, and 93%, respectively) compared with specimens with monoallelic inactivation (47%, 23%, and 28%, respectively). Similar to allelic status and germline status, the association between biallelic inactivation in other HRR genes and HRDsig was marginal (21% v 15%; Fig 2D), and even when both alleles of these genes were inactivated, they were typically HRDsig–. Details for specific gene alterations are provided in the Data Supplement (Table S3).

Overall, of 28,920 mBCs, predicted gBRCA alterations were detected in 3% (n = 980/28,920). The group of sBRCA/unkBRCA/gPALB2 patients comprised an additional 6% (n = 1,774/28,920). HRDsig+ was observed in 21% of mBC cases (6,041/28,920), including 78% (2,156/2,754) of specimens with BRCA or gPALB2 alterations, 21% (721/3,493) with other HRR alterations, and 14% (3,164/22,673) with HRR WT (Fig 2E).

Outcomes on PARPi Stratified by Gene and Mutation Origin

A total of 177 patients treated with PARPi were included in the CGDB and met inclusion criteria (Fig 1). Patients were grouped according to genomic alterations and included patients with gBRCA (for which PARPi are already approved) and patients with sBRCA or gPALB2 (sBRCA/gPALB2). Within the 177 patients, there were gBRCA (n = 71), sBRCA (n = 23), and gPALB2 (n = 8) on the basis of FMI tissue CGP results. The remaining patients (n = 75) were included in the assessment of outcomes using HRDsig. Prognostic factors such as hormone receptor status and histologic subtype were similar between the groups, whereas previous exposure to platinum chemotherapy was more common in patients with gBRCA mutations (gBRCA 30% v sBRCA/gPALB2 7%; P = .02). Additionally, patients with gBRCA tended to receive PARPi earlier (first/second line) during the treatment course of metastatic disease (30% v 23%; P = .05; Table 2). Similar to the full genomics data set, both cohorts had high levels of biallelic inactivation of the relevant gene and HRDsig+ (Table 2; Data Supplement, Fig S1). In the gBRCA cohort, only 1% (1/71) had predicted monoallelic inactivation and 77% (55/71) had confirmed BRCA biallelic inactivation (83% of BRCA1 and 74% of BRCA2), whereas status was unknown in 20% (14/71). The sBRCA/gPALB2 cohort had biallelic loss confirmed in 65% (16/23) for sBRCA and 75% (6/8) for gPALB2.

TABLE 2.

Clinical and Genomic Characteristics at the Start of PARPi for Those Patients Whose Tumors Were Classified as gBRCA or sBRCA/gPALB2

PARPi treatment outcomes, as assessed by rwPFS and rwOS, were similar for the gBRCA and sBRCA/gPALB2 cohorts. The median rwPFS for gBRCA patients was 6.0 months (95% CI, 4.6 to 8.2) versus 5.4 months (3.6 to 8.6) for those with sBRCA/gPALB2 tumors (hazard ratio [HR] gBRCA [ref], 1.56 [0.87-2.82]; Figs 3A and 3B). The median rwOS from the start of PARPi was also similar in the sBRCA/gPALB2 compared with the gBRCA cohort (21.2 [14.2-NR] v 20.9 [13.7-33.8] months), and hazard of death (HR gBRCA [ref], 1.63 [0.82-3.24]) was not significantly different between the two groups when accounting for other clinical factors (Figs 3C and 3D). There were clinical factors that were associated with outcomes. Previous exposure to platinum chemotherapy was associated with increased hazard of progression (HR, 2.35 [1.22-4.51]), whereas ECOG ≥2 was associated with increased risk of death (3.09 [1.35-7.09]).

FIG 3.

Kaplan-Meier curves for (A) RW progression-free survival and (C) RW overall survival from the start of PARPi for patients with tumors classified as gBRCA or sBRCA/gPALB2. (B and D) Hazard ratios from multivariate Cox proportional hazards models. ECOG, Eastern Cooperative Oncology Group; ER+/PR+, estrogen receptor-positive or progesterone receptor-positive; HER2, human epidermal growth factor receptor 2; PARPi, poly ADP-ribose polymerase inhibitors; RW, real-world; TNBC, triple-negative breast cancer.

Although the sample size was not well powered statistically, numerical comparison of rwPFS and rwOS for sBRCA patients (n = 24) and gPALB2 patients (n = 8) did not reveal major differences: median rwPFS (sBRCA, 4.6 months [3.0-8.1]; gPALB2, 8.6 months [6.4-NR]) and median rwOS (sBRCA, 21.2 months [13.8-NR]; gPALB2, 19.0 months [14.2-NR]).

Additionally, we compared outcomes for patients with BRCA (gBRCA, sBRCA, or unknown BRCA) or gPALB2 (n = 141) to patients with other HRR mutations (n = 16) or HRR WT tumors (n = 20). We saw significantly worse rwPFS and rwOS in the other HRR group (rwPFS, 1.9 [1.5-2.9] months; rwOS, 8.7 [3.7-NR] months) and for HRR WT (rwPFS, 2.5 [1.7-NR] months; rwOS, 13.9 [9.4-NR] months) compared with patients with BRCA or gPALB2 (rwPFS, 6.3 [5.0-7.6] months; rwOS, 20.9 [14.1-23.8] months; Data Supplement, Fig S2).

Outcomes of PARPi on the Basis of HRDsig

We further stratified patients using HRDsig, rather than individual genes, and compared demographics and outcomes of PARPi therapy. Patients with HRDsig+ were enriched for poorer-prognosis TNBC (Table 3). Nevertheless, patients with HRDsig+ had a significantly longer median rwPFS on PARPi compared with HRDsig– (5.9 v 3.0 months; HR HRDsig– [ref], 0.65 [0.43-0.96]; Figs 4A and 4B). The point estimate of rwOS of HRDsig+ patients was superior to HRDsig– (median, 18.4 v 13.0 months) but hazard of death (HR, 0.69 [0.43-1.12]) was not significantly different (Figs 4C and 4D). As in our analysis of patients with gBRCA versus sBRCA/gPALB2, previous platinum was associated with increased hazard of progression (HR, 2.47 [1.60-3.83]), but patients with TNBC had similar outcomes to those with other subtypes despite having higher rates of platinum exposure at the time of PARPi treatment (33% v 21%).

TABLE 3.

Clinical and Genomic Characteristics at the Start of PARPi Stratified by HRDsig Status

FIG 4.

Kaplan-Meier curves for (A) RW progression-free survival and (C) RW overall survival from the start of PARPi for stratified on HRDsig status. (B and D) Hazard ratios from multivariate Cox proportional hazards models. ECOG, Eastern Cooperative Oncology Group; ER+/PR+, estrogen receptor-positive or progesterone receptor-positive; HER2, human epidermal growth factor receptor 2; HRDsig, homologous recombination deficiency signature; PARPi, poly ADP-ribose polymerase inhibitors; RW, real-world; TNBC, triple-negative breast cancer.

Given the utility of GIS in ovarian cancer, patient outcomes based off GIS were also explored. For both rwPFS (5.2 v 5.3 months; HR GIS– [ref], 1.00 [0.72-1.47]) and rwOS (19.5 v 14.1 months; HR GIS– [ref], 0.86 [0.57-1.30]), no significant differences were seen between the GIS+ and GIS– cohorts (Data Supplement, Fig S3). Given the difference in predictive value for patients with breast cancer treated with PARPi observed for HRDsig and GIS, we assessed the overlap between our simulated GIS and HRDsig in our large genomic database of breast cancer samples and found that although 14% of samples were positive for both, 7.0% were HRDsig+ but GIS– and 7.8% were HRDsig– but GIS+; the remaining 71% were negative for both. Among the HRDsig– cohort, the majority of samples had a GIS of <42, but for the HRDsig+ cohort, approximately one third of samples had GIS <42 (Data Supplement, Fig S4).

The duration of individual responses to PARPi for patients without BRCA1/2 or gPALB2 mutations are shown in the Data Supplement (Fig S5), which includes patients with other HRR mutations/HRDsig+ (n = 8), other HRR alterations/HRDsig– (n = 11), HRR WT/HRDsig+ (n = 4), and HRR WT/HRDsig– (n = 13). Notably, there are patients, including several with HRR WT disease or biallelic mutations in ATM or BARD1, with long duration on therapy (>6 months). However, several patients remained on therapy despite evidence of progressive disease, suggesting these patients may have had more indolent disease or exhausted other more cytotoxic regimens. Some of these responsive HRR WT cases are captured by HRDsig+ and some are not.

DISCUSSION

These results suggest that the benefit of PARPi may extend beyond gBRCA carriers. Our results provide real-world evidence supporting outcomes from the TBCRC-048 trial and suggest that patients with mBC with sBRCA or gPALB2 mutation may also benefit from PARPi. Another recent trial, Talazoparib Beyond BRCA, showed similar results for gPALB2 and tumors positive for HRD as assessed by GIS.²² In our study, high rates of biallelic inactivation and HRDsig+ were seen across patients with all three mutation types. Even in patients with BRCA alterations with unknown germline/somatic status, biallelic inactivation and HRDsig+ were common, indicating that any BRCA alteration can guide the selection of patients for PARPi irrespective of known germline or somatic origin. The similar rwPFS for patients with sBRCA or gPALB2 alterations is encouraging and clinically meaningful. Most patients with mBC eventually progress on first-line chemotherapy. PARPi present an attractive alternative or subsequent treatment option, as they rarely cause alopecia and are associated with better quality of life.^4,23

In our analysis, the inclusion of tumors with sBRCA or gPALB2 mutations and those with HRDsig+ broadens the population of patients with mBC who could benefit from PARPi. Conversely, data supporting the role of alterations in other HRR genes, such as ATM and CHEK2, as predictors of response to PARPi for patients with mBC are still lacking. TBCRC-048 included 18 patients with germline or somatic mutations in ATM or CHEK2, none of whom achieved an objective response or stable disease for at least 6 months. It is still unclear whether the PARPi response will be restricted to pathogenic variants in specific genes, germline or somatic, and the search for biomarkers of response to PARPi is an area of active investigation.^24,25 Furthermore, a small number of patients without HRD biomarkers had >6-month duration of response on PARPi, and in two cases, no documented progression after 6 months. It is possible that PARPi may influence immune-dependent response that is independent of HRD.²⁶

Scar-based functional readouts such as HRDsig are a promising avenue to address these clinical challenges. We examined the association of HRR alterations with a novel HRDsig. It is provocative that the cohorts that most benefited from PARPi in TBCRC-048 had the most tumors classified as HRDsig+; these results are consistent with previous studies on the basis of the detection of COSMIC single-base substitution mutational signature 3.^27,28 Importantly, we find that 21% of mBCs are classified as HRDsig+, and HRDsig+ was associated with improved rwPFS compared with HRDsig–. Lower rates of HRDsig+ seen in sPALB2 and tumors with other HRR alterations suggest HRDsig+ can help avoid futile use of PARPi for patients with other HRR alterations not resulting in functional HRD. Consistent with these findings, van der Wijngaart et al²⁹ observed a high response rate in patients with biallelic BRCA inactivation across tumor types. We also note that in some cases, clinical benefit was observed in patients with tumors that were HRR WT but HRDsig+, underscoring that the currently approved biomarker, gBRCA, and even extension to sBRCA and gPALB2, may not fully capture all patients whose tumors have HRD.

A major strength of this study is that the large real-world genomic and clinicogenomic data sets enable the study of rare mutations, which are not with clinical trials. Limitations include that germline/somatic BRCA status and GIS were determined using computational predictions from tumor sequencing. The GIS reported was calculated based off published methods and not directly assessed using the CDx assay approved in ovarian cancer. There are also limitations inherent to the use of real-world evidence, including incomplete clinical variables (eg, tumor shrinkage for RECIST) and lack of a randomized standard of care/physician's choice arm.

In conclusion, overall, these findings suggest that the benefit of PARPi in patients with sBRCA and gPALB2 mutations are comparable with those with gBRCA. Clinical trials investigating PARPi for patients beyond gBRCA are ongoing, and a scar-based marker such as HRDsig may play a complementary role in identifying patients who will have durable response to PARPi.

DATA SHARING STATEMENT

The authors declare that all relevant aggregate data supporting the findings of this study are available within the article and its supplementary information files. In accordance with the Health Insurance Portability and Accountability Act, we do not have institutional review board approval or patient consent to share individualized patient genomic data, which contain potentially identifying or sensitive patient information and cannot be reported in a public data repository. Foundation Medicine is committed to collaborative data analysis and has well-established and widely used mechanisms by which qualified researchers can query our core genomic database of >500,000 deidentified sequenced cancers. More information and mechanisms for data access can be obtained by contacting the corresponding author or the Foundation Medicine Data Governance Council at data.governance.council@foundationmedicine.com.

AUTHOR CONTRIBUTIONS

Conception and design: Felipe Batalini, Russell W. Madison, Ethan S. Sokol, Kuei-Ting Chen, Garrett M. Frampton, Alexa B. Schrock, Nadine M. Tung

Collection and assembly of data: Dexter X. Jin, Dean C. Pavlick, Garrett M. Frampton

Data analysis and interpretation: Felipe Batalini, Russell W. Madison, Ethan S. Sokol, Dexter X. Jin, Kuei-Ting Chen, Brennan Decker, Dean C. Pavlick, Garrett M. Frampton, Gerburg M. Wulf, Judy E. Garber, Geoffrey Oxnard, Alexa B. Schrock

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).