Shallow Whole-Genome Sequencing to Assess Genomic Instability and Predict Treatment Response in Metastatic Castration-Resistant Prostate Cancer

Abstract

PURPOSE

Patients with metastatic castration-resistant prostate cancer (mCRPC) and homologous recombination deficiency (HRD) benefit from poly (ADP-ribose) polymerase (PARP) inhibitors (PARPis) or platinum-based chemotherapy (PlCh). Pathogenic variants in homologous recombination repair genes are used as a proxy for the phenotype. Genomic Instability Score (GIS) is a metric for the effect of HRD. This study used shallow whole-genome sequencing (sWGS), a cost-effective alternative to full-depth WGS, to calculate GIS and evaluate its predictive value for PARPi or PlCh response.

PATIENTS AND METHODS

We analyzed 288 tumor samples from 266 patients with mCRPC, consisting of 120 whole-genome samples from formalin-fixed paraffin-embedded material newly profiled at a sequencing depth of 2× and 168 from fresh frozen material downsampled in silico from 106× to 2×. The GIS was defined as the sum of large-scale transitions, telomeric-allelic imbalances, and genomic loss of heterozygosity.

RESULTS

The median GIS was 26 (IQR, 19-37). Samples with pathogenic variants in BRCA1, BRCA2, or PALB2 (BRCA complex) had a significantly higher median GIS (38 v 26, P < .0005). The median GIS increased significantly with tumor ploidy. Patients showing ≥50% prostate-specific antigen (PSA) declines had higher GIS on both PARPi (median 40 v 28, P = .0003) and PlCh (41 v 26, P = .0113). Similarly, patients with a partial response by RECIST1.1 had higher GIS than those with progressive disease (PARPi: 39 v 26, P = .0356; PlCh: 44 v 25, P = .0046). Adding GIS to BRCA-complex status improved PSA response prediction in logistic models (AUC PARPi: 0.849 v 0.781, P = .0308; PlCh: 0.731 v 0.658, P = .1888).

CONCLUSION

GIS assessment by sWGS is feasible and associated with response to PARPis and PlCh. The combination of BRCA-complex status and GIS is superior to BRCA-complex status alone in predicting PSA response to PARPis but not to PlCh.

INTRODUCTION

The heterogenous efficacy of poly (ADP-ribose) polymerase inhibitors (PARPis) in patients with metastatic castration-resistant prostate cancer (mCRPC) harboring pathogenic variants (PVs) in homologous recombination repair (HRR) genes has highlighted the need for a more precise and refined approach to patient selection. While PVs in key HRR genes, such as BRCA1 and BRCA2, have been associated with increased sensitivity to PARPis and platinum-based chemotherapy, not all patients with these aberrations respond equally.^1-4 This variability in clinical outcomes suggests that the current patient classification, relying on specific gene mutations, may not fully capture the complexity of homologous repair deficiency (HRD) and its implications for therapeutic response. Moreover, the functional impact of PVs in HRR genes depends on whether they result in monoallelic or biallelic inactivation, but detecting biallelic loss can be challenging.⁵

CONTEXT

• Key Objective

• The aim of this study was to assess the relationship between the Genomic Instability Score (GIS) determined by shallow whole-genome sequencing and homologous recombination repair (HRR) gene alterations in metastatic castration-resistant prostate cancer, while also evaluating its predictive value for poly (ADP-ribose) polymerase (PARP) inhibitor and platinum-based chemotherapy outcomes.

• Knowledge Generated

• Patients with pathogenic variants (PVs) in BRCA-complex genes had significantly higher median GISs than patients without, and GIS increased significantly with tumor ploidy. Patients achieving ≥50% prostate-specific antigen (PSA) declines or partial responses by RECIST 1.1 on both PARP inhibitor and platinum chemotherapy had higher GISs, and adding GIS assessment to BRCA complex genotyping improved PSA response prediction on PARP inhibition.

• Relevance

• As a stand-alone diagnostic for HRR deficiency in prostate cancer, GIS has limited utility because of the absence of a validated score cutoff. However, it may have future value as a companion diagnostic alongside BRCA genotyping to guide treatment selection.

The Genomic Instability Score (GIS), composed of large-scale transitions (LSTs), telomeric allelic imbalances (tAIs), and genomic loss of heterozygosity (gLOH), represents a promising biomarker for refining the stratification of HRD tumors.^6-8 Shallow whole-genome sequencing (sWGS) has emerged as a cost-effective approach for detecting copy number variations and, consequently, evaluating GIS across tumor genomes.⁹ Despite the growing interest in leveraging GIS as a biomarker, the association between GIS derived from sWGS and PVs in specific genes involved in HRR in mCRPC remains to be explored. Additionally, it remains unknown whether sWGS-based GIS can improve the prediction of response to PARPis or platinum-based chemotherapy compared with PV analysis. Addressing these knowledge gaps could improve cost-effective precision oncology strategies by maximizing the clinical benefit of PARPi and platinum-based chemotherapy in mCRPC. The aim of this study was to assess the relationship between GIS determined by sWGS and HRR gene alterations in mCRPC while also evaluating its predictive value for PARPi and platinum-based chemotherapy outcomes.

PATIENTS AND METHODS

Study Cohort

This study included 288 tumor samples from 266 patients with mCRPC, preserved in two different methods: 120 formalin-fixed paraffin-embedded (FFPE) samples and 168 fresh frozen samples. All patients were at one time under the care of the medical oncology department at Radboudumc.

The 120 FFPE samples, collected through routine care, were profiled using sWGS with a median depth of 2.05×. Sample selection was intentionally enriched for HRR-associated PVs on the basis of prior targeted sequencing data, either by a commercial service provider (Foundation Medicine; FoundationOne CDx) or in house using a targeted sequencing panel covering 523 cancer-related genes (Illumina; TruSight Oncology 500).¹⁰ DNA isolation and sWGS followed the methodology described by Elze et al.¹¹ The 168 fresh frozen samples were profiled using WGS by the nonprofit Hartwig Medical Foundation in the CPCT-02 study. These samples were also used in the study by Pamidimarri Naga et al¹² and were in silico downsampled from a median sequencing depth of 106× to 2×, as described therein. Genome-wide copy number variations were determined using IchorCNA (v0.3.2), and gLOH was assessed using bcftools (v1.15) for both fresh frozen and FFPE samples, following the methodology described by Elze et al.¹¹

Genetic Variant Analysis

Genetic variants identified by the Hartwig Medical Foundation were reported according to their established bioinformatic pipeline¹³ and subsequently reassessed on the basis of guidelines from the American College of Medical Genetics and Genomics and the Association for Molecular Pathology.^14,15 Variants found by commercial providers or in-house panels were re-evaluated using the same criteria. Genes of interest with known direct or indirect involvement in HRR included BRCA1, BRCA2, PALB2, CDK12, ATM, BARD1, BRIP1, RAD51B, RAD51C, RAD51D, RAD54L, CHEK1, CHEK2, FANCA, FANCL, FANCM, PPP2R2A, NBN, and MRE11A. TP53 was also included because of its association with higher GIS.¹⁶ Missense mutations in TP53 were classified as pathogenic if they had a nonfunctional transcriptional activity according to The TP53 Database (R21, Jan 2025 version).^17,18 Variants with partially functional transcriptional activity were considered nonpathogenic.

HRD Assessment

Genome-wide copy number plots from sWGS were independently reviewed by two researchers to ensure correct interpretation of ploidy and purity. In cases of discordance, a third researcher provided an independent assessment, and the median score was taken. For the FFPE subcohort, authors RvD and PS served as first reviewers, with NM as the second reviewer. For the fresh frozen subcohort, authors YQ and PS were first reviewers, with NM as the second. LSTs and tAIs were quantified per sample and combined with gLOH to calculate the GIS in line with the study method used by Elze et al.¹¹

Response Data

Response data for patients receiving PARPis or platinum-based chemotherapy were retrospectively collected from medical records in the TROPIC study, of which the requirement for approval was waived by the Medical Review Ethics Committee Oost-Nederland (CMO 2023-16778, October 18, 2023).

Progression-free survival (PFS) was defined as the time from the first administration of the therapy under investigation until radiologic or clinical progression, including death and censoring at the end of follow-up if the treatment was ongoing or if no progression occurred at next systemic treatment. Prostate-specific antigen (PSA) responses were assessed according to the Prostate Cancer Clinical Trials Working Group 3 (PCWG3) criteria and dichotomized by ≥50% PSA decline (PSA50).¹⁹ Radiologic responses were reported according to RECIST 1.1 criteria.²⁰

Statistical Analysis

BRCA1, BRCA2, and PALB2 are essential for the formation of the BRCA complex and are therefore crucial for HRR. Differences in GIS resulting from PVs in BRCA-complex genes and non–BRCA-complex genes were assessed using Wilcoxon test. Differences in GIS for PSA50 status and best radiologic response were assessed using Mann-Whitney U tests and Kruskal-Wallis tests, followed by post hoc Dunn tests, respectively. A multiple linear regression model was used for analyzing the impact of TP53 mutations on GIS, corrected for BRCA-complex mutations. To assess the predictive performance of GIS and PV assessment for PSA50 response, receiver-operating characteristic (ROC) curve analysis was conducted. Logistic regression models were built using GIS alone, BRCA-complex status alone, and the combination thereof. The AUC was calculated for each model and compared using DeLong test. Time-to-event outcomes were analyzed using Cox proportional hazards models and visualized with Kaplan-Meier survival curves. All statistical tests were two-sided, and P values <.05 were considered statistically significant. All analyses were performed using R (version 4.4.1) with RStudio (version 2024.04.2).

RESULTS

In total, 288 samples from 266 patients with mCRPC were analyzed for GIS by sWGS. Of these, 120 FFPE samples were resequenced with a mean depth of 2.05× and 168 fresh frozen samples were in silico downsampled from 106× to 2×. The median time between CRPC diagnosis and biopsy was 11 months (IQR, 1-28 months), with 83% of the samples being collected in the CRPC setting. The most common biopsied sites were lymph node, followed by bone and prostate (Data Supplement, Fig S1). Baseline characteristics are presented in Table 1.

TABLE 1.

Baseline Characteristics of the Full Patient Cohort

Characteristic	Patients in Analysis	Median (IQR) or n (valid percentage)
Age at CRPC, years	259	67 (61-73)
PSA at initial diagnosis	260	49.8 (17.3-150.0)
Gleason grade group	251
1		16 (6.4)
2		28 (11.2)
3		27 (10.8)
4		62 (24.7)
5		118 (47.0)
De novo metastatic	248	149 (60.1)

Abbreviations: CRPC, castration-resistant prostate cancer; n, number of patients; PSA, prostate-specific antigen.

The median GIS of all samples in this study was 26 (IQR, 19-37, Fig 1A). Samples with a monoallelic or biallelic PV in BRCA1, BRCA2, or PALB2 (BRCA complex, n = 56) had a median GIS of 38 (IQR, 24-47.3), significantly higher than the samples without BRCA-complex PVs (median, 26, IQR, 18-34, P = 5.3 × 10⁻⁶, Data Supplement, Fig S2). We identified 122 samples with a monoallelic or biallelic PV linked directly or indirectly to HRR, their median GIS was 31 (IQR, 22-42, Data Supplement, Fig S2), and the GISs per altered HRR gene are presented in Figure 1B and numerically presented in Data Supplement (Table S1). Samples without any of these PVs had a median GIS of 24 (n = 166, IQR, 18-33).

FIG 1.

GIS per gene and ploidy. (A) Overview of GIS per sample and the type of mutations in the HRR-associated gene or TP53. The GIS is determined by the sum of LST, gLOH, and tAI. Multihit is defined as more than 1 type of alteration in the same gene. (B) GIS per altered gene, split by monoallelic and biallelic alterations. (C) GIS per ploidy. All boxplots represent the IQR with the median indicated by a solid line within the box; lines attached to the boxes extend to 1.5×IQR. Asterisks indicate significance levels (not significant = ns, *P < .05, **P < .01, ***P < .001, ****P < .0001). CNV, copy number variation; FFPE, formalin-fixed paraffin-embedded; GIS, Genomic Instability Score; gLOH, genomic loss of heterozygosity; HRR, homologous recombination repair; LST, large-scale transition; tAI, telomeric allelic imbalances.

GISs were significantly higher in the triploid (n = 73) and tetraploid (n = 71) samples compared with diploid samples (n = 144), with adjusted P values of 1.1 × 10⁻¹⁰ and 1.1 × 10⁻¹², respectively (Fig 1C). There was no statistical difference between the tissue obtained from hormone-sensitive prostate cancer or CRPC (Data Supplement, Fig S3) or between the different sites of biopsy (Data Supplement, Fig S4).

Pathogenic TP53 alterations are linked to higher GIS in the literature¹⁶; in our cohort, TP53-mutated samples had a median GIS of 27 (IQR, 22-37) when compared with 26 (IQR, 18-37) for TP53-wildtype samples. The difference did not reach statistical significance after correction for the presence of BRCA-complex PVs (P = .0512).

Response Data

Of the 266 patients analyzed for GIS by sWGS, 89 received a treatment directed at HRD: 45 were treated with a PARPi, 18 with platinum-based chemotherapy (PlCh), and 26 with both (sequentially). Table 2 presents baseline characteristics for the patients treated with PARPis or PlCh. The median GIS of patients treated with a PARPi or PlCh was 31 (IQR, 22-42). Patients receiving PARPis had a median GIS of 32 (IQR, 24-46). Patients receiving PlCh had a median GIS of 33 (IQR, 23-45).

TABLE 2.

Baseline Characteristics of the HRD-Directed Therapy‑Treated Patient Cohort

Characteristic	Patients in Analysis	Median (IQR) or n (valid percentage)
Age at CRPC, years	88	63.5 (57.0-70.3)
Months from ADT to CRPC	88	18 (11-33)
PSA at initial diagnosis	72	56.5 (21.0-147.0)
Gleason grade group	82
1		5 (6)
2		6 (7)
3		9 (11)
4		18 (22)
5		44 (54)
Metastatic at initial diagnosis	89	48 (54)
Disease volume conform CHAARTED criteria	74
High		24 (32)
Low		50 (68)
Disease risk conform LATTIDUE criteria	74
High		27 (36)
Low		47 (64)
Systemic therapies for CSPC	89	35 (39)
Docetaxel		32 (36)
Abiraterone		2 (2)
Enzalutamide		1 (1)
Prior systemic treatment lines before first HRD-targeting therapy	89	2 (1-3)
0		5 (6)
1		22 (25)
2		28 (31)
3		18 (20)
4		10 (11)
≥5		6 (7)

Abbreviations: ADT, androgen-deprivation therapy; CRPC, castration-resistant prostate cancer; CSPC, castration-sensitive prostate cancer; HRD, homologous recombination deficiency; PSA, prostate-specific antigen.

Response to PARPis

Of the 71 patients treated with a PARPi, 52 received olaparib, 16 talazoparib, and three rucaparib. A ≥50% PSA decline (PSA50) was observed in 33 patients (47%) treated with a PARPi while 37 (53%) did not witness a PSA50 (Fig 2A). The median GIS for patients with and without PSA50 was 40 (IQR, 31-55) versus 28 (IQR, 20-39; P = .0003), respectively. In a logistic regression for PSA50 response, the odds ratio per GIS point was 1.08 (95% CI, 1.04 to 1.13).

FIG 2.

GIS and response to PARP inhibitor treatment. (A) GIS for patients with and without ≥50% PSA response (PSA50). (B) GIS per radiologic response to PARP inhibitor treatment. (C) Progression-free survival on PARP inhibitor treatment for the patients with the lowest GIS quartile, middle half, and highest quartile. All boxplots represent the IQR with the median indicated by a horizontal line across the box; lines attached to the boxes extend to 1.5×IQR. Asterisks indicate significance levels (not significant = ns, *P < .05, **P < .01, ***P < .001, ****P < .0001). GIS, Genomic Instability Score; PARP, poly (ADP-ribose) polymerase; PSA, prostate-specific antigen.

To evaluate the predictive value of GIS relative to BRCA-complex mutational status for PSA50, we constructed additional logistic models on the basis of BRCA-complex status alone and a combined model incorporating BRCA complex and GIS. All models demonstrated reasonably strong discriminative performance, with AUCs of the ROCs of 0.751 (GIS), 0.781 (BRCA complex), and 0.849 (combined, Data Supplement, Fig S5). No significant difference was observed between GIS and BRCA-complex status (P = .6692). The combined model significantly outperformed both GIS alone (P = .0212) and BRCA-complex status alone (P = .0308).

For 60 of the 71 PARPi-treated patients, radiologic evaluation was available: 22 (31%) had a partial response, 27 (38%) stable disease, and 11 (15%) progressive disease as the best response (Fig 2B). The median GIS was 39 for partial response, 30 for stable disease, and 26 for progressive disease. There was a significant difference between the groups, driven by the difference between partial response and progressive disease (P = .0156).

For the PFS analysis, patients treated with PARPis were stratified on the basis of their GISs: The lowest 25% had a GIS less than 23.5, the middle 50% had a GIS from 23.5 to 45.5, and the highest 25% had a GIS more than 45.5. The median PFS for each group was 3.5 months (95% CI, 3.2 to 14.2), 6.6 months (95% CI, 4.6 to 9.1), and 10.7 months (95% CI, 7.1 to not reached), respectively (Fig 2C). The Cox proportional hazards model did not show statistical significance (P = .0823) on the basis of the comparison between all three groups.

Response to Platinum-Based Chemotherapy

Of the 44 patients treated with platinum-based chemotherapy, 43 received carboplatin and 1 received cisplatin in combination with etoposide. Carboplatin was combined with cabazitaxel (n = 21), docetaxel (n = 14), or etoposide (n = 2) or given as monotherapy (n = 6).

PSA response could be calculated for 41 of the 44 patients; 21 (51%) witnessed a PSA50 and 20 (49%) did not (Fig 3A). The median GIS for patients with a PSA50 was 41 (IQR, 31-58) compared with 26 (IQR, 22-38) for patients without a PSA50 (P = .0113). In a logistic regression for PSA50 response, the odds ratio per GIS point was 1.06 (95% CI, 1.01 to 1.12).

FIG 3.

GIS and response to platinum-based chemotherapy. (A) GIS for patients with and without ≥50% PSA response (PSA50). (B) GIS per radiologic response to platinum-based chemotherapy. (C) Progression-free survival on platinum-based chemotherapy for the patients with the lowest GIS quartile, middle half, and highest quartile. All boxplots represent the IQR with the median indicated by a horizontal line across the box; lines attached to the boxes extend to 1.5×IQR. Asterisks indicate significance levels (not significant = ns, *P < .05, **P < .01, ***P < .001, ****P < .0001). GIS, Genomic Instability Score; PSA, prostate-specific antigen.

Again, we evaluated the predictive power of GIS relative to BRCA-complex status for PSA50 prediction by comparing the AUC of the ROC of the logistic models with GIS, BRCA-complex status, and both (Data Supplement, Fig S6). All three models demonstrated reasonably strong discriminative performance, with AUCs of 0.732 (GIS), 0.658 (BRCA-complex status), and 0.731 (combined), without statistically significant differences between the models (GIS v BRCA complex P = .3064; combined v GIS P = .9566; combined v BRCA complex P = .1888).

Radiologic evaluation was available for 33 of the 44 patients treated with platinum-based chemotherapy: 12 (36%) had a partial response, 8 (24%) stable disease, and 13 (39%) progressive disease as the best response (Fig 3B). The median GIS were 44, 32, and 25 for the three groups, respectively, with a significant difference among them, driven by the difference between partial response and progressive disease (P = .0021).

For the PFS analysis, patients were grouped by the lowest quartile of GIS (<23), middle half (≥23 to <45), and highest quartile (≥45). The median PFS per group was 2.4 months (95% CI, 1.2 to not reached), 3.5 months (95% CI, 2.4 to 7.2), and 9.2 months (95% CI, 5.0 to not reached), respectively, without a statistical significance between the three groups (P = .0673; Fig 3C).

DISCUSSION

Our findings highlight the complementary value of sWGS-derived GIS, alongside BRCA-complex genotyping, in predicting treatment response to PARPis and platinum-based chemotherapy in mCRPC. The GIS strongly correlates with response to both therapies, even after selection on the basis of HRR genotyping. However, its predictive performance in mCRPC seems more modest than that reported for HRD-targeting therapies in ovarian and breast cancers.^21-28 Future studies on PARPis and PlCh in prostate cancer should consider integrating both GIS and HRR gene genotyping to improve patient stratification and therapeutic decision making.

BRCA1 and BRCA2 PVs leading to higher GIS can be seen across cancer types, as shown in a pancancer analysis by Elze et al.¹¹ However, the median GIS and the corresponding cutoff for classifying a sample as HRD vary per cancer type. Previous studies have established and validated a GIS threshold of ≥42 (sometimes ≥33), which corresponds with BRCA mutational status and is prognostic of treatment response in ovarian and breast cancers.^21-28 However, ovarian and breast cancers generally have a higher GIS than prostate cancer, underscoring the need for cancer-specific thresholds.^11,29

Zhao et al proposed a prostate cancer‑specific GIS threshold of >43 to be considered HRD, which is not in line with the generally lower GIS for prostate cancer compared with breast and ovarian cancers.^7,30 The study by Zhao et al included only 16 patients treated with PARPis, while our cohort comprises 71 PARPi-treated and 44 platinum-based chemotherapy‑treated patients. Our GIS data obtained by sWGS seems consistent with this threshold of 43, as most samples with a GIS >43 were from patients with biochemical or radiologic response, although we did not formally test its predictive value as this study is underpowered to define a clinically implementable threshold.

HRD scores by genetic subgroups in the study by Zhao et al⁷ differ substantially from those in our cohort. Although overall median GIS cannot be directly compared because of intentional enrichment for HRR-mutated samples in our study, notable differences exist in subgroup distributions. For samples without HRR PVs, the median GIS was 2 (IQR, 1-4) in the study by Zhao et al⁷ versus 24 (IQR, 18-33) in our cohort. For samples with HRR PVs, Zhao et al⁷ reported a median GIS of 15 (IQR, 10-25), whereas we observed a median of 31 (IQR, 22-42). This discrepancy can only partly be explained by the broader gene set that considered HRR genes used by Zhao et al,⁷ which included 36 genes. Both studies calculated GIS as the sum of gLOH, tAI, and LST scores, yet differences may arise from variations in the algorithms used. Additionally, Zhao et al⁷ performed sequencing at a depth of 150×, whereas our sWGS was at 2× coverage. The lower sequencing depth can affect the detection of smaller allelic imbalances, particularly gLOH. Our previous methodologic study demonstrated that sequencing depths as low as 1×, at tumor purity ≥20%, are sufficient for reliable detection of LST, and increasing the depth to 2-5× does not improve LST calling.¹² It remains to be determined whether higher depth enhances the accuracy of tAI and gLOH detection.

Overall, it should be noted that the GIS in prostate cancer is less strongly correlated with BRCA or BRCA-complex pathogenic variants than has been reported in breast or ovarian cancer.^21-28 This is illustrated by the largely overlapping GIS distributions observed between patients with and without BRCA-complex variants in our study. Consequently, identifying a robust and clinically meaningful discriminative cutoff for prostate cancer remains challenging, as the currently available data lack sufficient power to reliably dichotomize PARP inhibitor responders and nonresponders. Although this limits the stand-alone clinical implementation of the GIS in mCRPC, our data found value in the use of the GIS as a complementary tool alongside (BRCA) genotyping.

Within a cancer type, ploidy is another factor—alongside HRD—that significantly influences GIS in our data set. Higher ploidy was associated with elevated GIS, consistent with findings by Popova et al,⁸ who proposed ploidy-adjusted thresholds for HRD detection using LSTs. However, when LSTs are combined with tAI and gLOH into a composite GIS, correction for ploidy is generally not performed.^{7,21,22,28,31}

Responses to platinum-based chemotherapy and PARPis are also strongly influenced by BRCA mutational status, which is in line with our data.^1-4 Patients with biallelic BRCA1 or BRCA2 PVs show the most favorable PARPi responses.³² In our cohort, GIS values for BRCA2 overlapped between monoallelic and biallelic inactivation, although biallelic inactivation generally yielded higher scores. It is important to note that not all biallelic events will be recognized as such and thus the monoallelic group may contain samples with biallelic inactivating PVs.

Patients with a PSA50 and GIS >43 but without BRCA-complex PVs were uncommon, with only 1 patient case observed per treatment group. On PARPis, this patient carried an ATM heterozygous splice-site variant (NM_000051.4: c.6095G>A; p.?), while in the platinum-based chemotherapy group, no HRR PV was identified. The small number of such patient cases does not allow definitive conclusions. The prospective, tumor-agnostic Drug Rediscovery Protocol trial (ClinicalTrials.gov identifier: NCT02925234) currently includes a cohort evaluating talazoparib treatment based on GISs in tumors without apparent HRR PVs.

This study has several limitations. First, the use of different sequencing platforms and tissue storage conditions (newly profile versus in silico downsampled; FFPE versus fresh frozen) may affect GIS assessment. We cannot determine the effect of these differences in our data set as FFPE samples were intentionally enriched for HRR PVs, contributing to a higher GIS. Second, the retrospective nature of the clinical data collection introduces common limitations. Because 17% of samples were obtained from hormone-sensitive prostate cancer, differences in biopsy timing may result in variations in GIS, as TP53 PVs are known to emerge during disease progression. However, in matched samples, there is no evidence of BRCA-complex PV enrichment in the CRPC phase.³³ The interpretation of response data is complicated by treatment selection: PARPi was administered only within clinical trials with genetic selection or as the standard of care for BRCA-mutated patients. PlCh is not approved for mCRPC in the Netherlands and is typically reserved for patients with suspected neuroendocrine differentiation or dedifferentiation or as a last-resort option for patients with HRD. As a result, PARPi-treated patients were highly selected for HRR aberrations, while platinum-treated patients are expected to have an above-average aggressive disease.

In conclusion, this study supports the integration of GIS assessment by sWGS as a complementary tool to (BRCA) genotyping in mCRPC, while also highlights its limited utility as a stand-alone marker. The added predictive value of combining GIS with genotyping was observed for PARP inhibitor treatment but not for platinum-based chemotherapy. These findings suggest that integrating GIS with BRCA-complex genotyping may help improve prediction of response to PARPis in mCRPC.

DATA SHARING STATEMENT

A data sharing statement provided by the authors is available with this article at DOI https://doi.org/10.1200/PO-26-00065.

Data is available upon reasonable request from bona fide researchers from the online DANS Data Station Life Sciences repository (https://doi.org/10.17026/LS/WGS2LL).

AUTHOR CONTRIBUTIONS

Conception and design: Peter H.J. Slootbeek, Niven Mehra

Financial support: Niven Mehra

Administrative support: Niven Mehra

Provision of study materials or patients: Niven Mehra

Collection and assembly of data: Peter H.J. Slootbeek, Julian J.R. Kokke, Niven Mehra

Data analysis and interpretation: Peter H.J. Slootbeek, Yarah M. Quint, Samhita Pamidimarri Naga, Maria Victoria Luna-Velez, Marjolijn J.L. Ligtenberg, Richarda M. de Voer, Niven Mehra

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).