Classification of BRCA2 Variants of Uncertain Significance (VUS) Using an ACMG/AMP Model Incorporating a Homology-Directed Repair (HDR) Functional Assay

Genetic testing has been integrated into clinical management due to advances in molecular genetics and sequencing technology. This testing benefits patients and society through enhanced cancer surveillance, prevention measures, and therapeutic options. However, the clinical relevance of many inherited variants (variants of uncertain significance, VUS) in disease-related genes identified through genetic testing has not been determined. The American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP) has developed variant classification guidelines based on a general framework that includes information on variant frequency, computational prediction, functional analysis, and segregation studies (1). However, these general ACMG/AMP framework guidelines lack a uniform classification framework for specific genes, which can lead to discrepancies in variant classification among groups conducting testing.

BRCA2 (MIM: 600185) is a frequently mutated gene in the general population (1 in 1,000 in unaffected individuals; ref. 2), for which inherited pathogenic variants have been associated with high risks of breast, ovarian, prostate, and pancreatic cancers (2, 3). Current clinical management of individuals with BRCA2 pathogenic variants can involve specific cancer-screening and surveillance measures, surgical prevention, and therapeutic options for patients and family members. However, individuals found to carry VUS cannot benefit from these clinical management strategies due to uncertainty about the clinical relevance of the VUS.

As of January 5, 2022 there were 6,176 BRCA2 VUS listed in the ClinVar database, including 5,534 missense VUS. Furthermore, of all 6,506 BRCA2 missense variants in ClinVar, 5,534 (85%) were VUS, 667 (10.4%) had conflicting interpretations, and 153 (2.4%) were benign. Another 85 were considered pathogenic/likely pathogenic, but most appeared to influence splicing. Only 21 missense variants that do not influence splicing have been established as pathogenic/likely pathogenic and 35 as benign of likely benign by the ENIGMA expert panel for BRCA1/2 variant classification (before development of ACMG/AMP rules-based analysis; refs. 4–6). All of these are located in the BRCA2 DNA-binding domain (DBD; amino acids 2,479–3,192). BRCA2 VUS characterization is an area of active research with recent efforts focused on functional evaluation (4, 6–10), computational algorithm development (11, 12), and expanded case–control analysis of variants (5). Among efforts focused on functional evaluation of BRCA2 VUS, a homology-directed repair (HDR) cell-based DNA repair assay has demonstrated high sensitivity and specificity for established pathogenic and benign missense variants from the BRCA2 DBD (4, 6). This assay has been applied only to missense variants in the BRCA2 DBD because this hotspot region contains all established non-splice P/LP missense variants in BRCA2, and because it is not known whether the HDR assay can measure the functional effects of variants in other domains of BRCA2. Results from this functional assay have recently been integrated into an ACMG/AMP framework VUS classification model developed by Ambry Genetics (6). In the current study, we report on re-standardization of the HDR assay and systematic evaluation and classification of 133 BRCA2 DBD missense variants using an alternative ACMG/AMP model, as applied by the GeneDx clinical genetic testing laboratory, that incorporates the HDR assay results.

A total of 133 missense variants in the BRCA2 DBD domain (amino acids position 2479–3192) that were evaluated for effects on BRCA2 activity using an HDR functional assay and were clinically observed by GeneDx (up to 12/22/2020) were included in the current study. These 133 were among 450 BRCA2 DBD missense variants selected for evaluation by the HDR assay (data not shown) because of high prediction scores for deleterious/non-functional variants from the BayesDel (13) and BRCA-ML (12) in silico prediction models and presence in the NIH supported ClinVar database (133 of 1,413; 9.4%). Variants with observed splicing effects or predicted to have potential splicing effects by in silico models were excluded. Variants were annotated according to HGVS recommendations and RefSeq transcript NM_000059.3 (Supplementary Table S1).

HDR functional analysis of BRCA2 DBD missense variants was carried out as described previously (4). Briefly, variants were introduced into a full-length FLAG-tagged BRCA2 mammalian expression plasmid by site-directed mutagenesis. The presence of variants was verified by Sanger sequencing and related protein expression was verified by Western blot using anti-FLAG antibodies. Individual plasmids expressing BRCA2 and the iSceI restriction endonuclease were co-transfected into brca2-deficient V-C8 cells containing a stably integrated DR-GFP reporter with an iSceI recognition site. Repair of the iSceI induced double-strand break by BRCA2-dependent homologous recombination (HR) resulted in GFP expression. The proportion of GFP-expressing cells for each transfection was quantified by flow cytometry. The established pathogenic missense variant p.Asp2723His and wild-type BRCA2 were used as internal controls for normalizing the number of DNA repair-dependent GFP-positive cells for each BRCA2 construct to a 1 to 5 scale. All variants were evaluated using at least two independent clones in duplicate experiments.

A Bayesian regression model was used to estimate the distribution of HDR scores and 95% confidence intervals (2.5 and 97.5 percentiles of the posterior distributions) for all variants. All calculations were performed in R using the rstanarm package. On the basis of the log-normal distribution of HDR scores for 21 pathogenic/likely pathogenic and 35 benign/likely benign missense variant standards from the BRCA2 DBD (previously classified by the ENIGMA; evidence-based Network for the Interpretation of Germline Mutant Alleles) expert panel or reported in previous publications (Supplementary Table S1; refs. 4–6), 99.9% probability thresholds for both pathogenic and benign variants were calculated.

The ACMG/AMP framework is a combination of 27 sources of evidence from population, computational and predictive, functional, segregation, and other relevant data, in which each contributing variable is weighted as very strong (PVS1), strong (PS1, PS2, PS3, PS4), moderate (PM1, PM2, PM3, PM4, PM5, PM6), and supporting (PP1, PP2, PP3, PP4, PP5) for pathogenicity, or stand-alone (BA1), strong (BS1, BS2, BS3, BS4), and supporting (BP1, BP2, BP3, BP4, BP5, BP6, BP7) for benign effects (14). The five variant classifications based on the ACMG/AMP framework are pathogenic, likely pathogenic, benign, likely benign, and VUS. In this study ACMG/AMP scoring rules (14) modified by GeneDx were used for clinical classification of BRCA2 variants. The study was conducted in accordance with guidelines set forth by the Western Institutional Review Board, which waived authorization for use of de-identified data. The strength of all pathogenic and benign ACMG/AMP criteria was considered. Thus, variant-specific functional data in support of neutrality (BS3) outweighed less specific evidence such as rarity in populations, which might otherwise support pathogenicity (PM2). For BRCA2 missense variants with no reported effects on RNA splicing, the very strong evidence of pathogenicity (PVS1) rule was not applied. ACMG/AMP criteria used for evaluation of BRCA2 missense variants were as follows:

This is the only stand-alone evidence applied within the ACMG/AMP framework and is used to assign benign impact based on variant frequency in populations. GeneDx evaluated population frequency relative to the disease incidence as stand-alone data for classifying a variant as benign. For rare disorders, proportionally lower allele frequencies are accepted as stand-alone criteria relative to the disease incidence (14).

These two strong forms of evidence for either pathogenicity (PS3) or benign impact (BS3) are based on well-established in vitro or in vivo functional studies supportive of a deleterious (PS3) or neutral (BS3) effect on protein function. We applied results from the HDR assay, with demonstrated high sensitivity and specificity for 21 pathogenic and 35 benign variants, as strong functional evidence.

This evidence is applied when allele frequency is greater than expected for disorders (1). In addition, GeneDx uses a gene-specific lower threshold with supporting evidence weight (BS1_Supporting; ≥0.1%).

Characterized by location of a variant in a mutational hot spot and/or critical and well-established functional domain without benign variation. Because all evaluated variants were in the BRCA2 DBD domain, PM1 was applied to all evaluated variants (15).

Absent or very rare in controls (gnomAD, ExAC).

Variant observed in trans with a BRCA2 pathogenic variant in individual(s) with phenotypes consistent with Fanconi anemia (FA). As BRCA2 bi-allelic pathogenic mutations are associated with FA, a variant that occurs in trans with a PVS1 variant in an FA case is attributed PM3 evidence. PM3 is applied as strong (PM3_Strong) evidence with multiple independent occurrences (≥2), or supporting (PM3_Supporting) evidence when with incomplete segregation (phase is not confirmed for one or more observations).

Novel missense change at an amino acid residue where a different missense change determined to be pathogenic has been seen before.

Co-segregation with disease in multiple affected family members. PP1 is also applied as moderate evidence (PP1_Moderate) with increasing segregation data.

Multiple lines of computational evidence support a deleterious effect (PP3) or no impact (BP4) on the gene or gene product. The in silico algorithm PROVEAN was used for protein prediction.

Observed in trans or phase unknown with a BRCA2 pathogenic variant, in the absence of reported phenotypes consistent with FA.

Variant observed in a case with an alternate molecular basis for disease. Variants co-occurring with established pathogenic variants are considered BP5. Strong evidence is applied (BP5_Strong) when multiple co-occurrences (>10) were observed.

Multifactorial prediction was applied as supporting (PP_Multifactorial/BP_Multifactorial) or strong (PP_Multifactorial_Strong) evidence based on multifactorial probability models that incorporate evidence from segregation, co-occurrence, and pathology (5, 16, 17). This multifactorial prediction was not identified as a source of evidence in the original ACMG/AMP classification model (1).

Results from other functional studies, including mouse embryonic-stem cell (m-ESC)–based functional analysis (7, 10, 18–20), a BRCA-deficient cell line–based drug assay (MANO-B; ref. 8), and a prime editing–based saturation genome–editing (SGE) analysis (9) were compared with results for the HDR assay.

Variant classification from ClinVar submitted by clinical laboratories meeting ClinGen minimum requirements for data sharing to support quality assurance and the ENIGMA (Evidence-based Network for the Interpretation of Germline Mutant Alleles) consortium expert panel was used for comparison with the GeneDx ACMG/AMP model (including functional data).

All data are presented in the article and/or are available directly from the authors.

The HDR functional assay for BRCA2 DBD missense variants was previously shown to have 100% sensitivity (95% CI, 79%–100%) and 100% specificity (95% CI, 93%–100%) for pathogenic variants based on 18 known benign/likely benign and 12 known pathogenic/likely pathogenic variants (4). However, a subset of these “standards” is now known to influence RNA splicing. Here, the HDR assay was recalibrated using a larger set of 21 known pathogenic/likely pathogenic and 35 known benign/likely benign variants, none of which were predicted to cause aberrant splicing by Splice AI prediction analysis or shown to cause aberrant splicing by RT-PCR (Supplementary Table S1). Thresholds for 99.9% probability for pathogenic and benign effects were estimated on the basis of these standards. An HDR score <1.49 was considered non-functional with probability of pathogenicity >0.99, whereas variants with HDR score >2.50 were considered functional, with probabilities of neutrality >0.99. BRCA2 missense variants with HDR scores between 1.49 and 2.50 were considered to have potential hypomorphic/partial effects on function with uncertain clinical significance and were excluded from the current study.

The HDR scores for 133 clinically observed BRCA2 DBD missense variants evaluated in this study ranged from 0.83 (95% CI, 0.76–0.91) for p.His2623Arg to 5.67 (95% CI, 5.05–6.36) for p.Asp3112Asn (Tables 1 and 2, Fig. 1). These included 67 VUS evaluated for the first time in this study and 66 missense variants previously evaluated by the HDR assay and other functional assays (Tables 1 and 2; refs. 4, 11, 12, 21). Of the 133 variants, 44 were designated as non-functional with HDR scores <1.49 and 89 were designated as functional with HDR scores >2.50. Both non-functional and functional variants were evenly distributed within the DBD domain, except for a reduced number of observed variants in the OB2 domain (Fig. 1). Multiple residues contained more than one amino acid alteration. Of these, substitutions in 14 of 15 residues had the same functional impact (all functional or all non-functional), such as the p.Arg2488Gly/Ser functional and the p.Asp2723His/Ala/Val non-functional variants. However, p.Arg2625Lys was functional, whereas p.Arg2625Ile was non-functional.

We compared results from the HDR assay for the 133 variants with available results from other functional studies (Supplementary Table S2). The other functional studies included mESC-based survival and HR assays (7, 10, 18–20), drug–response assays for DLD1 BRCA2-deficient cells reconstituted with BRCA2 (MANO-B; ref. 8), and prime editing–based SGE (9). Among 39 variants evaluated with the mESC-based methods, 19 variants that were functional in the HDR assay showed full complementation in the mESC-based survival analysis with 50% to 116% BRCA2 protein activity in mESC-based HR assays. All 20 variants that were non-functional in the HDR assay showed complete loss or reduced mESC survival and low or no HR activity in the mESC-based HR assays. Among 41 variants also evaluated for effects on drug response in the MANO-B study, which rated variants as class 1–5 for the sensitivity to four drugs (Olaparib, niraparib, rucaparib, and cisplatin), 12 of 18 variants considered functional in the HDR assay were consistently resistant (class 1 or 2) in the four drug–response assays (Supplementary Table S2). However, of the remaining 6 functional HDR variants, p.Val2908Gly (HDR, 2.53; 95% CI, 2.36–2.72) was class 4/5 (non-functional) for all 4 drug assays; p.Gly2508Ser (HDR, 3.22; 95% CI, 2.95–3.52) was class 3/4 (uncertain/non-functional) for the 4 drug assays; p.Leu2972Trp (HDR, 2.78; 95% CI, 2.47–3.12) and p.Pro2771Leu (HDR, 2.81; 95% CI, 2.44–3.24) were class 3 (uncertain) for the 4 drug assays; p.Arg2520Pro (HDR, 3.54; 95% CI, 3.08–4.08) and p.Leu3011Pro (HDR, 3.12; 95% CI, 2.89–3.36) were mixed class 3 (uncertain) and class 2 (non-functional) in the drug assays. In contrast, 22 of 23 variants designated non-functional by HDR assay were class 4 or 5 (non-functional) in the MANO-B study. Only p.Trp2626Arg was class 3 (uncertain) for Olaparib and Rucaparib response (Supplementary Table S2).

Comparisons with the prime editing–based SGE analysis (9) were less informative because only variants in part of exon 15 (c.7545_7552, residues 2515–2518) and portions of exon 17 (c.7826_7842, residues 2609–2614); (c.7916_7924, residues 2639–2642), and (c.7930_7960, residues 2644–2654) of the BRCA2 DBD were evaluated by prime editing. Of 5 variants evaluated by HDR and by prime editing, 4 were non-functional by both methods [p.Gly2609Val (HDR, 1.08; 95% CI, 0.97–1.22), p.Leu2647Pro (HDR, 1.00; 95% CI, 0.89–1.12), p.Val2652Gly (HDR, 1.14; 95% CI, 0.99–1.32), and p.Leu2653Pro (HDR, 0.92; 95% CI, 0.80–1.06)], and one was functional by both methods (p.Cys2646Trp; HDR, 3.39; 95% CI, 2.95–3.91; Supplementary Table S2). This suggests a potentially strong correlation between the two assays, but more variants must be analyzed by both methods for a detailed comparison.

The ClinGen SVI recommendations for applying weight to functional studies within the ACMG/AMP ClinGen framework suggest that the PS3 and BS3 rules can be applied as strong lines of evidence when an assay has sensitivity and specificity for known pathogenic and benign variants that yields an odds of pathogenicity (oddspath) >18.7 (6, 22). The HDR assay in this study was assigned strong evidence of pathogenicity under the PS3/BS3 functional assay rule because the assay yielded an oddspath of 35.0 on the basis of widely accepted standard variants (21 pathogenic and 35 benign), that were correctly identified by the assay as deleterious or neutral, respectively (Supplementary Table S1). This was consistent with a previous study of the HDR assay that yielded an oddspath of 46.0 (6). Thus, among the 133 variants under evaluation, those found to be non-functional in the HDR assay were assigned PS3 (strong evidence of pathogenicity) in the ACMG/AMP model, whereas variants considered functional were assigned BS3 (strong evidence of benign).

Of the 133 variants, 66 (49.3%) had previously been classified as pathogenic/likely pathogenic (n = 22) or benign/likely benign (n = 44) by GeneDx based on an internal ACMG/AMP framework model. Results from previously reported HDR assays and other functional studies contributed to classification of 43 of these 66 variants (16/22 pathogenic/likely pathogenic and 27/44 benign/likely benign). No changes in classification occurred after applying new HDR functional data for 61 of these 66 variants previously classified as benign, likely benign, pathogenic, or likely pathogenic. However, 5 variants had stronger evidence of pathogenicity and changed from likely pathogenic to pathogenic (p.Leu2510Pro, p.Leu2686Pro, p.Asp2723Val, p.Gly2724Trp, and p.Leu3101Arg; Table 1; Supplementary Table S2, Fig. 2). In addition, 67 variants had a classification of VUS before the current study. After evaluation by the HDR assay (Figs. 1 and2), 92.5% (62/67) of the VUS were reclassified as likely benign (n = 40) or likely pathogenic (n = 22) by the ACMG/AMP classification model that included the BS3 or PS3 rules (Fig. 1, Tables 1 and 2; Supplementary Table S2). Although 5 VUS (p.Cys2646Trp, p.Pro2735Arg, p.Cys2765Gly, p.Pro2771Leu, and p.Pro2785Leu) had strong evidence of benign effects in the HDR assay (BS3-strong), these variants were not reclassified as benign or likely benign due to lack of other supporting benign criteria (Supplementary Table S2). Other than the PS3/BS3 functional data, moderate evidence from variant location in a functional domain (PM1; 100.0%; 133/133) and population frequency (PM2/BA1/BS1_Supporting; 86.5%; 115/133), and supporting evidence from computational prediction (PP3/BP4; 100%; 133/133) were the most used forms of evidence for classification (Table 2; Supplementary Table S2).

The classifications for 133 variants based on incorporation of the HDR functional assay results into the GeneDx ACMG/AMP classification model were compared with the classifications submitted to ClinVar by other clinical-testing laboratories (Supplementary Table S2). Among the 129 variants reported to ClinVar, 14 consistently classified as likely benign or benign by ClinGen-designated high-quality testing groups were reclassified as benign/likely benign by the GeneDx ACMG/AMP model. A further 31 consistently classified as VUS in ClinVar were reclassified as benign/likely benign (n = 10), pathogenic/likely pathogenic (n = 16) and VUS (n = 5). All 15 variants consistently classified as likely pathogenic or pathogenic in ClinVar were reclassified as likely pathogenic or pathogenic using the GeneDx model. However, 69 variants displayed conflicting classifications in ClinVar. After incorporating the HDR functional assay results, 95.0% (95 out of 100) variants with at least one entry as VUS in ClinVar were reclassified as either pathogenic/likely pathogenic or benign/likely benign by the GeneDx ACMG/AMP classification model, and 83.9% (26 out of 31) variants consistently classified as VUS in ClinVar were reclassified as either pathogenic/likely pathogenic or benign/likely benign by GeneDx (Fig. 2; Supplementary Table S2).

We evaluated 133 clinically observed BRCA2 DBD missense variants with an HDR functional assay and then incorporated the findings into an ACMG/AMP framework variant classification model. Of these, 128 variants were classified as pathogenic/likely pathogenic or benign/likely benign, including 62 VUS that were re-classified as likely pathogenic (n = 22) or likely benign (n = 40). These results suggest that over 30% of VUS in the BRCA2 DBD hotspot may be reclassified as pathogenic/likely pathogenic 32.8% (22 of 67) and that approximately 60% may be downgraded to benign/likely benign (59.7%; 40 of 67). This is consistent with results for BRCA2 DBD VUS evaluated using the separate Ambry Genetics ACMG/AMP classification model (25.3% pathogenic/likely pathogenic, 60.4% benign/likely benign; ref. 6). However, these findings contrast with studies showing that <10% of unique VUS from the entire BRCA2 gene are upgraded to pathogenic/likely pathogenic (23), suggesting that very few VUS outside of the DBD affect BRCA2 function and predispose to cancer.

Carriers of the BRCA2 variants classified as pathogenic/likely pathogenic will benefit from these studies by now qualifying for the same clinical management strategies as carriers of known pathogenic protein truncating variants in BRCA2, such as enhanced screening with MRI and/or consideration for risk reducing prophylactic mastectomy/oophorectomy. For those already diagnosed with cancer the reclassifications provide eligibility for PARP inhibitor therapy. Furthermore, the information can be used to provide risk assessment and management for relatives. Carriers of BRCA2 variants classified as benign/likely benign also benefit by eliminating the uncertainty associated with a genetic testing VUS diagnosis.

Publicly accessible data and computational predictions (PM1, PM2, and PP3/BP4) have been applied most frequently to variants within the ACMG/AMP framework rules-based classification model. However, these sources provide only mainly moderate or supporting evidence and do not resolve the classification of numerous VUS, which pose challenges to clinical management. In contrast, the HDR functional assay provides strong evidence of pathogenic or benign effects and leads to the classification of the great majority (128 of 133) of BRCA2 DBD missense variants observed clinically by GeneDx. These findings suggest that the majority of missense VUS observed in this domain of BRCA2 can be reclassified when adding this or other validated functional assays to ACMG/AMP framework rules-based classification models.

The HDR assay results were consistent with those from an mESC-based functional assay and CRISPR prime editing–based SGE, but discrepancies were observed between the HDR assays and the MANO-B drug–response assays. As the HDR assay has been validated using a large number of known pathogenic (n = 21) and benign missense variants (n = 35; Supplementary Table S1), the suggestion is that the drug–response analysis may need to be fine-tuned for variant classification. In addition, the implications of the inconsistencies for PARP inhibitor therapy remain to be determined.

The ACMG/AMP framework BP1 rule providing evidence for benign effects of missense variants in a gene that primarily has disease-causing truncating variants was not used in the current study because established pathogenic missense variants in BRCA2 have exhibited similar risks for breast cancer as truncating variants. Similarly, the PP2 rule providing evidence of pathogenicity that applies to a gene with a low rate of benign missense variation, in which missense variants are common mechanisms of disease, was not applicable to BRCA2. Among the very strong and strong sources of evidence for pathogenicity within the ACMG/AMP framework, PVS1 was not applicable for the majority of missense variants unless there was a demonstrated splicing effect by RNA analyses. The PS1 rule for a variant that causes the same amino acid change as a known pathogenic variant was rarely applicable. The PS2 rule for de novo variant was limited by availability of family data, and the PS4 rule for enrichment in cases over controls in a population was not applicable for the majority of these rare VUS. In addition, the PM5 rule for pathogenic missense hotspots, should be applied with caution. For example, observed amino acid alterations in the same residue can have different clinical impact (p.Arg2625Lys as likely benign vs. p.Arg2625Ile as likely pathogenic). PS3/BS3 functional evidence was the major strong source of evidence applied in the current study. This was sufficient, when combined with other moderate and supporting evidence, to classify 128 of 133 variants.

The five variants that were not reclassified and remained as VUS (p.Cys2646Trp, p.Pro2735Arg, p.Cys2765Gly, p.Pro2771Leu, and p.Pro2785Leu) were considered functional in the HDR assay, but no other evidence to support classification as benign variants was available. PROVEAN in silico prediction for all five variants yielded PP3 evidence, which conflicted with the BS3 functional results. Similarly, PROVEAN provided BP4 evidence in favor of a benign classification for 36 variants that were finally classified as pathogenic/likely pathogenic. Application of in silico prediction models optimized for the gene of interest such as BRCA-ML (ref. 11; which has been developed specifically for BRCA1 and BRCA2 variants), correctly predicted the five VUS as functional with BP4 evidence (Supplementary Table S2). However, generalized in silico models as opposed to gene-specific in silico models are often preferred by clinical labs testing a wide variety of genes. Overall, the PROVEAN predictor used by GeneDx contributed effectively to classification of 83 benign variants with BP4 evidence and 8 pathogenic/Likely Pathogenic variants with PP3 evidence. The limitations of all in silico models further validate the importance of functional assays for classifying BRCA2 VUS.

To date (January 5, 2022), over 5,500 missense BRCA2 VUS have been reported to the ClinVar database. Although the current study demonstrates the important contribution of a functional assay with defined high sensitivity and specificity for pathogenic variants to variant classification, more efforts in development, validation, and application of functional assays are needed to keep pace with VUS identification. Classification of variants in BRCA2 will likely benefit from development of high-throughput methodologies and subsequent validation of findings in the near future, with results having direct effects on patient care.

L.R. Susswein reports other support from GeneDx during the conduct of the study. M.E. Roberts reports other support from GeneDx during the conduct of the study. M.L. Marshall reports other support from GeneDx outside the submitted work. S. Hiraki reports other support from GeneDx outside the submitted work. W. Berkofsky-Fessler reports employment with GeneDx. S.M. Domchek reports personal fees from AstraZeneca and from GSK outside the submitted work. K.S. Hruska reports employment with GeneDx and have stock options. No disclosures were reported by the other authors.

C. Hu: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. L.R. Susswein: Data curation, formal analysis, writing–original draft, writing–review and editing. M.E. Roberts: Data curation, formal analysis, methodology, writing–original draft, writing–review and editing. H. Yang: Formal analysis, investigation, methodology. M.L. Marshall: Data curation, formal analysis, investigation. S. Hiraki: Data curation, formal analysis, investigation, writing–review and editing. W. Berkofsky-Fessler: Data curation, writing–review and editing. S. Gupta: Formal analysis, investigation, methodology, writing–review and editing. W. Shen: Formal analysis, writing–review and editing. C.A. Dunn: Data curation, formal analysis, investigation. H. Huang: Conceptualization, data curation, formal analysis, investigation, methodology, writing–review and editing. J. Na: Formal analysis, validation. S.M. Domchek: Formal analysis, methodology, writing–review and editing. S. Yadav: Conceptualization, investigation, writing–review and editing. A.N.A. Monteiro: Conceptualization, funding acquisition, investigation, writing–original draft, writing–review and editing. E.C. Polley: Formal analysis, validation, investigation, writing–review and editing. S.N. Hart: Data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. K.S. Hruska: Data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. F.J. Couch: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

The study was funded by NIH grants R35CA253187, R01CA225662, a specialized program of research excellence (SPORE) in breast cancer to Mayo Clinic (P01 CA106201), and the Breast Cancer Research Foundation (to F.J. Couch).

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.