Assessing the Clinical Relevance of BRCA1 BRCT Domain Variants of Uncertain Significance

Breast and ovarian cancer are among the most common cancers in Canadian women. Approximately 5-10% of breast and 20-25% of ovarian cancers are inherited, with pathogenic germline BRCA1 and BRCA2 variants causing the majority of hereditary cases. While genetic testing is used to identify pathogenic BRCA variant carriers who would subsequently benefit from personalized screening, prophylactic and treatment steps, its widespread use has resulted in the discovery of thousands of variants of uncertain significance (VUS). VUSs pose a critical clinical challenge as they are unable to be effectively interpreted, limiting clinicians’ ability to accurately assess cancer risk and recommend appropriate management steps. We sought to build a computational classifier specific to BRCA1’s BRCT domain to accurately predict missense VUS pathogenicity and stratify VUSs to prioritize for functional analyses like phosphopeptide binding assays. All BRCA1 BRCT missense variants were collected from the ClinVar database and were analyzed using 50 different in silico tools. Molecular Feature Selection Tool (MFeaST) ranked tools based on their ability to discriminate pathogenic and benign variants. Supervised classifiers were then trained using combinations of the most discriminative in silico tools, with the most accurate classifier being used to score all VUSs. Phosphopeptide binding assays were conducted on select suspected pathogenic and benign VUSs to assess their impact on binding activity and folding. Our results show that an Ensemble Subspace kNN classifier trained with 9 in silico tools (CADD hg19, MetaRNN, ClinPred, VEST4, BayesDel AF, EVE, Eigen PC, gMVP and PolyPhen2) demonstrated the best performance out of all trained supervised models, with 91.1% and 87.9% accuracy on the training and validation sets, respectively. Compared to individual in silico and AI protein language models, our model demonstrated the highest accuracy on the training set and comparable accuracy on the validation set of BRCA1 BRCT variants. Results from the phosphopeptide binding assays show that suspected pathogenic VUSs demonstrated either reduced BRCT binding ability, reduced BRCT protein levels, or a combination of both. Suspected benign VUSs demonstrated retained BRCT binding ability and BRCT protein levels. The computational and functional evidence obtained through this study will contribute to the reclassification of BRCT VUSs to pathogenic or benign, strengthening and broadening variant classification databases essential for clinicians to make decisive management recommendations for BRCA1 variant carriers. Additionally, this study highlights the potential of domain-specific computational approaches for characterizing missense variants in other multi-domain cancer susceptibility genes.

Department dc:contributor.department

Pathology and Molecular Medicine

Year dc:date.issued

2024

Author dc:creator

• Torretto, Gabriella

Advisors dc:contributor.supervisor

• Davey, Scott
• Feilotter, Harriet

• BRCA1
• Breast Cancer
• Machine Learning
• Functional Assays
• Hereditary Breast and Ovarian Cancer
• Ovarian Cancer
• BRCA1 BRCT Domain
• In Silico
• Genetic Testing

Statement dc:rights

• Attribution-NonCommercial-NoDerivatives 4.0 International

Licence dc:rights.uri

• http://creativecommons.org/licenses/by-nc-nd/4.0/

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/1974/33446

OAI identifier oai:identifier

oai:queensu.scholaris.ca:1974/33446