Traditionally, BRCA1/2 genetic testing has been performed in individuals with a strong personal or family history suggestive of HBOC. These tests are effective for detecting known pathogenic variants, while the pathogenicity of VUS identified through such testing is typically re-evaluated by co-segregation or functional assays [8, 27]. Building on this, the implementation of CGP in cancer genomic medicine has significantly expanded the scope of BRCA testing. CGP enables simultaneous analysis of hundreds of cancer-related genes in a tumor-agnostic manner, leading to the incidental identification of BRCA1/2 variants even in patients without classical HBOC phenotypes. Supporting this broader clinical context, a large-scale cohort study by Momozawa et al. reported that carriers of germline BRCA1/2 pathogenic variants are at increased risk not only for breast, ovarian, prostate, and pancreatic cancers, but also for several other malignancies compared to the general population [28].
In this study, we characterized the landscape of BRCA1/2 variants identified through 2172 real-world CGP tests and applied an in silico filtering framework to prioritize VUS with suspected pathogenicity. In our CGP series, some BRCA1/2 pathogenic variants were identified in non-classical HBOC-associated cancers, and BRCA1/2 VUS were predominantly found in such contexts. This expansion of BRCA testing beyond traditional HBOC-associated tumors has broadened its clinical utility but has also markedly increased the detection of VUS, posing new challenges for clinical interpretation and therapeutic decision-making. Indeed, BRCA1/2 VUS accounted for 75.3% of all variants in our dataset, highlighting the overwhelming proportion of variants lacking definitive classification. In addition to these testing-related factors, population genetics may also contribute to the high proportion of VUS observed in our cohort. BRCA1/2 exhibit substantial allelic heterogeneity, and many rare variants remain unclassified in underrepresented populations such as East Asians. For example, a Japanese-specific BRCA2 variant was recently reclassified only after population-level evidence became available [29], illustrating how ancestry-related variants are often absent from global databases and thus more likely to remain VUS. This genetic background likely contributed to the elevated VUS frequency observed in our CGP cohort.
To address this issue, we developed a filtering strategy that integrates multiple in silico predictors. Importantly, this framework was designed for variant prioritization rather than ACMG/AMP classification. In this context, computational predictors were used solely as screening tools to nominate VUS that merit deeper investigation, and not as evidence codes contributing to PP3 or BP4 assignments. Accordingly, the numerical thresholds derived from Franklin were applied only as reproducible filtering criteria, without implying any ACMG/AMP evidence strength. Furthermore, concordance among multiple predictors was not interpreted as stronger pathogenic evidence and was not used to modify PP3 weight; instead, it was considered only as descriptive information within the screening process. This conceptual distinction is essential because ENIGMA BRCA1/2 VCEP recommendations appropriately define PP3 evidence strength for clinical classification, whereas our objective was upstream triage of VUS identified through real-world CGP. Although prior ENIGMA-based studies have evaluated the clinical performance and calibration of in silico predictors [19, 30], our analysis did not aim to reassess the clinical performance of these predictors but rather to characterize their behavior within CGP-derived VUS for the purpose of prioritization.
This approach prioritized 10 candidate VUS with suspected pathogenicity, among which BRCA2:c.67 G > C was selected for further investigation. While its pathogenicity had long remained undetermined, Bose and colleagues recently demonstrated deleterious effects of BRCA2:c.67 G > C in vitro CRISPR-based functional assays, supported by a minigene splicing assay showing aberrant exon 2 splicing [26]. Consistent with these findings, our clinical analyses confirmed that BRCA2:c.67 G > C is of germline origin and induces exon 2 skipping in patient-derived specimens. Moreover, the patient with duodenal cancer harboring this variant exhibited marked sensitivity to platinum-based chemotherapy, providing in vivo evidence consistent with HR deficiency biology (Supplementary Fig. S2). In addition, although the observation period was limited to four months, the patient with castration-resistant prostate cancer harboring this variant also showed sensitivity to talazoparib combined with enzalutamide (Supplementary Fig. S3). Because this patient also carried a pathogenic BRIP1 frameshift variant, it is possible that HR deficiency resulting from BRIP1 contributed to the PARP inhibitor sensitivity. Nevertheless, the concordant evidence of platinum sensitivity in the duodenal cancer case, together with functional demonstration of exon 2 skipping, strongly supports BRCA2:c.67 G > C as a pathogenic allele. These findings are consistent with recently published European recommendations for RNA-based diagnostics, which emphasize that transcript-level analysis using patient-derived RNA is appropriate when evaluating splice-altering variants, whereas minigene assays serve as an alternative when native RNA cannot be obtained [31]. Importantly, in line with these recommendations, we encountered well-recognized technical limitations of FFPE-derived RNA for splicing analysis, likely due to RNA fragmentation and restricted amplicon size. In our case, RT-PCR from FFPE tumor tissue was unsuccessful for these reasons, whereas peripheral blood RNA allowed reliable assessment of exon 2 skipping for the germline BRCA2:c.67 G > C variant.
Compared with Bose’s study, which suggested PS3_Moderate evidence, our work further contributed PS1 (Splicing)_Moderate based on similarity to other splice donor variants such as BRCA2:c.67+1 A > G and BRCA2:c.67+2 T > C (See also Supplementary Table S4) [32, 33]. Collectively, these results reinforce the pathogenicity of BRCA2:c.67 G > C and highlight the value of integrating real-world clinical data with experimental assays. Large-scale functional studies, including SGE and HDR assays, have generated comprehensive catalogs of variant effects [22,23,24,25,26,27, 34]. Although they are highly effective for evaluating missense variants, they are less suited for splicing alterations or in-frame indels, for which transcript-level effects or the combinatorial complexity of indel designs may limit comprehensive assessment. Our study complements these efforts by confirming variant occurrence in real-world clinical practice and by directly validating splicing alterations in patient-derived samples, as demonstrated with BRCA2:c.67 G > C. This highlights the complementary value of CGP-based validation as a clinically applicable framework.
In this study, we compared the performance of multiple in silico predictors in a large real-world CGP cohort. As a result, we successfully narrowed 153 VUS to a set of prioritized candidates for functional interpretation. Previous work has also examined the role of in silico predictors in variant interpretation. For example, Wilcox et al. demonstrated how the choice of predictor criteria can influence clinical variant classification in hereditary testing cohorts [35]. Another study compared predictor performance for BRCA1/2 variants [36], underscoring variability in concordance between tools. While these investigations provided valuable methodological insights, they were not conducted in the context of large-scale, real-world CGP data. Our study complements this literature by assessing predictor performance in a CGP cohort and linking these predictions to variants directly encountered in clinical oncology practice.
Importantly, our filtering framework not only prioritized BRCA2:c.67 G > C but also illustrates a generalizable approach for identifying clinically relevant variants concealed among VUS. This conceptual framework can be extended to other hereditary cancer predisposition syndromes where VUS remain a major challenge. For instance, VUS are frequently identified in mismatch repair genes (MLH1, MSH2, MSH6, PMS2) through CGP, which are responsible for Lynch syndrome (LS). In this context, in silico predictors may be complemented by orthogonal molecular indicators such as microsatellite instability high or deficient mismatch repair status to refine variant interpretation. Extending our approach to LS and other hereditary cancer syndromes could help establish a more universal strategy for variant prioritization and interpretation, thereby enhancing the clinical utility of CGP in diverse hereditary contexts. Thus, although our analysis centered on BRCA1/2 as a clinically illustrative example, the same filtering and prioritization framework can be readily adapted for VUS interpretation in other hereditary cancer genes identified through CGP.
There are several limitations in this study. First, our analysis was based on 2172 CGP tests from a single regional cohort in Japan; therefore, the frequency and spectrum of BRCA1/2 VUS, as well as the performance of in silico predictors, may not be fully generalizable to other populations. Second, although our approach integrated multiple in silico predictors, functional validation was restricted to BRCA2:c.67 G > C. This reflects our intention to provide a proof-of-concept demonstration rather than a comprehensive functional survey. Nevertheless, consistency between our prioritized variants and previously published HDR and SGE results supports the robustness of the filtering strategy. This alignment between independent large-scale functional assays and our CGP-based prioritization increases confidence that the framework can scale to other variants. Third, the choice of in silico predictors and thresholds inevitably involves arbitrary selections. To enhance immediate clinical applicability, we focused on predictors commonly used in clinical variant interpretation frameworks such as ACMG/AMP and ClinGen. While not exhaustive, this proof-of-concept demonstrates feasibility and lays the groundwork for systematic evaluation in future studies. Fourth, reliance on clinically approved CGP panels with heterogeneous designs and coverage may have limited the detection of certain variant classes, such as structural variants, copy number changes, or deep intronic alterations. Finally, the somatic versus germline origin of variants was not systematically determined, as tumor-based CGP is not optimized for this purpose, and VUS do not routinely trigger germline confirmatory testing in real-world oncology practice. Germline analysis is generally reserved for clearly pathogenic or likely pathogenic variants, whereas most BRCA1/2 VUS are managed without additional testing unless hereditary cancer is clinically suspected or the patient requests further evaluation. Consistent with this standard practice, germline testing was selectively performed only for BRCA2:c.67 G > C because pathogenicity was strongly suspected, while the origin of other VUS remained undetermined. Although this represents an inherent limitation of CGP-based analyses, it also reflects its clinical utility in identifying potentially actionable variants regardless of germline or somatic origin. Despite these limitations, our framework demonstrates practical utility in prioritizing BRCA1/2 VUS from real-world CGP data and provides a foundation for refinement and broader clinical application. Crucially, integration with international variant databases and multi-institutional collaborations will be essential to accelerate variant interpretation efforts and enhance the global clinical utility of CGP.
In conclusion, our study illustrates a proof-of-concept filtering framework that integrates multiple in silico predictors with functional evidence to prioritize and interpret BRCA1/2 VUS identified through real-world CGP testing. While our dataset highlights the broadened landscape of BRCA1/2 variants beyond conventional hereditary testing, this framework offers a practical approach for prioritizing VUS in a clinically applicable manner. By complementing large-scale functional assays, it provides a useful bridge between genomic profiling and variant interpretation and may be extended to other cancer susceptibility genes and hereditary cancer syndromes.
