Functional Assays for Classification of BRCA2 Variants of Uncertain Significance

The assessment of the influence of many rare BRCA2 missense mutations on cancer risk has proved difficult. A multifactorial likelihood model that predicts the odds of cancer causality for missense variants is effective, but is limited by the availability of family data. As an alternative, we developed functional assays that measure the influence of missense mutations on the ability of BRCA2 to repair DNA damage by homologous recombination and to control centriole amplification. We evaluated 22 missense mutations from the BRCA2 DNA binding domain (DBD) that were identified in multiple breast cancer families using these assays and compared the results with those from the likelihood model. Thirteen variants inactivated BRCA2 function in at least one assay; two others truncated BRCA2 by aberrant splicing; and seven had no effect on BRCA2 function. Of 10 variants with odds in favor of causality in the likelihood model of 50:1 or more and a posterior probability of pathogenicity of 0.99, eight inactivated BRCA2 function and the other two caused splicing defects. Four variants and four controls displaying odds in favor of neutrality of 50:1 and posterior probabilities of pathogenicity of at least 1 × 10⁻³ had no effect on function in either assay. The strong correlation between the functional assays and likelihood model data suggests that these functional assays are an excellent method for identifying inactivating missense mutations in the BRCA2 DBD and that the assays may be a useful addition to models that predict the likelihood of cancer in carriers of missense mutations. [Cancer Res 2008;68(9):3523–31]

The BRCA2 tumor suppressor gene displays substantial allelic diversity. Several thousand unique mutations have been identified through mutation testing and are listed in the Breast Cancer Information Core Database,⁸

which serves as a public record of BRCA1 and BRCA2 mutations. Protein truncating mutations in BRCA2 predispose women to early-onset breast and ovarian cancer (1, 2), and account for 15% to 30% of familial breast cancer. These large and small insertions and deletions, nonsense mutations, and splicing variants in large part exclude the nuclear localization signals of BRCA2, mislocalize BRCA2 to the cytoplasm (3), and disrupt the homologous recombination (HR)–dependent DNA repair activity of BRCA2. In addition, these mutations exclude a COOH-terminal Rad51 binding domain that is involved in regulation of the DNA repair function of the protein.

In contrast, the influence of many missense mutations, intronic variants, and in-frame deletions and insertions in the BRCA2 gene, also called variants of uncertain significance (VUS) or unclassified variants, has not been determined. Because many of these mutations are very rare, there is limited available genetic information from families carrying the mutations for assessment of cancer risk. In addition, the influence of the majority of these VUS on BRCA2 function is not known. The absence of knowledge about the risk of cancer associated with these variants is an important obstacle to identifying and providing optimal care for individuals found to carry these genetic alterations.

Several approaches to classification of VUS have been proposed. These include analysis of segregation of mutations with disease in families (4), cross-species sequence variation (5), and evaluation of the frequency of VUS in unaffected controls (6). An integrated approach to classification of VUS into deleterious and neutral categories has recently been proposed (7). This likelihood model combines data on cosegregation of VUS with cancer in families and co-occurrence of VUS in trans phase with known deleterious mutations (8) and is cross-validated with results from cross-species sequence analysis (9). It has recently been extended to include pathologic characteristics of breast tumors associated with deleterious BRCA1 mutations (10) and to incorporate an analysis of personal and family history of cancer associated with VUS (11). Although effective for classification of the more commonly observed VUS, this combined likelihood model is limited by the availability of family data and has not been effective in classifying VUS that are identified in small numbers of families. In addition, some efforts have been made to characterize BRCA2 VUS using functional assays that assess the effect of amino acid changes on BRCA2 protein function. These studies have focused primarily on the influence of VUS on the DNA repair activity of BRCA2 (12–14) and have been limited to small numbers of VUS.

Here, we report on the evaluation of the influence of 22 BRCA2 VUS on BRCA2 activity using two functional assays. These functional assays measure the HR repair activity of BRCA2 and the ability of BRCA2 to regulate centrosome duplication in the context of a full-length BRCA2 protein. We focused our study on 22 BRCA2 missense mutations from the DNA binding domain (DBD) of BRCA2 that have been identified in multiple breast cancer families by Myriad Genetic Laboratories, Inc. Based on these data, we showed that there is a strong correlation between the assay results and the odds of cancer causality/neutrality and the posterior probability of cancer defined by the likelihood model for these VUS.

Site-directed mutagenesis. Missense mutations were incorporated into full-length BRCA2 cDNA expression constructs as previously reported (13). In brief, nucleotide changes generated using the QuickChange site-directed mutagenesis kit (Stratagene) were incorporated into a 3× FLAG-tagged full-length BRCA2 cDNA expression plasmid. Primer sequences and conditions are available from the authors. The presence of the mutations was confirmed by sequencing.

Reverse transcription-PCR studies. RNA was harvested from lymphocytes of patients that harbor the R2659K, R2659T, and E2663V missense mutations with Trizol according to the manufacturer's instructions. mRNA from each sample was used for reverse transcription-PCR (RT-PCR) analysis with forward primer ATGGAAAGGCTGGAAAAGAA at the end of exon 16 and reverse primer AAGAATCCAAGTTTGGTATA at the start of exon 19. PCR products were size selected on agarose gels and all differentially sized products were extracted from the agarose gel and sequenced in both directions.

Cell culture and transfection. 293T cells were cultured in RPMI (Life Technologies) and BRCA2-deficient V-C8 cells (15) were maintained in DMEM-F12 (BioWhittaker). Both were supplemented with 10% bovine calf serum (HyCLone), 2 mmol/L l-glutamine, 100 units/mL penicillin, and 100 μg/mL streptomycin. Transfections were performed using FuGENE-6 according to the manufacturer's protocol (Roche).

Immunoprecipitation and immunoblotting. BRCA2 protein was immunoprecipitated from cell lysates with Protein-G agarose beads conjugated with M2 Flag monoclonal antibody (Sigma). Immunoprecipitates were subjected to SDS-PAGE on a 4% to 15% Tris-HCl gradient gel and transferred onto a polyvinylidene difluoride membrane. Immunoblotting was performed with a polyclonal anti-BRCA2 antibody against the COOH terminus of BRCA2 and an anti-Rad51 polyclonal antibody.

Centrosome amplification assay. 293T and V-C8 cells transfected with various FLAG-tagged BRCA2 missense mutant constructs and a wild-type control were grown for 72 h, fixed in cold methanol, and labeled with anti-FLAG and anti–centrin-2 antibodies (20H5, kindly provided by Jeffrey Salisbury, Department of Biochemistry and Molecular Biology, Mayo Clinic College of Medicine, Rochester, MN) followed by Alexa 568 goat anti-mouse and Alexa 488 goat anti-rabbit secondary antibodies. Centrioles from 100 cells were enumerated by confocal microscopy. Each experiment was performed in duplicate.

HR assay. V-C8 cells containing a single copy of the DR-GFP reporter plasmid were cotransfected with FLAG-tagged BRCA2 wild-type and VUS containing constructs and the I-Sce1–expressing pcBASce plasmid (13). After 72 h, cells were harvested and the number of green fluorescent protein (GFP)–expressing cells was assessed by flow cytometry. In parallel, we determined transfection/expression efficiency for BRCA2 by fluorescently labeling cells from these transfection experiments with an anti-FLAG antibody and counting the number of FLAG-BRCA2–expressing cells per 1,000 cells using a fluorescence microscope. The ratio of GFP-expressing cells induced by wild-type or mutant BRCA2 compared with vector control in the HR assay was then plotted after adjustment for transfection/expression efficiency.

Ascertainment of pedigrees with variants. The data analyzed in this report come predominantly from the large database of full sequence testing for mutations in the BRCA1 and BRCA2 genes performed at Myriad Genetics Laboratories, Inc., as of December 2005. This sequence testing involves complete sequence analysis of PCR products containing all coding regions within the BRCA1 and BRCA2 genes. These products are derived from genomic DNA extracted from blood samples and are amplified using primers in flanking introns. Individuals found through the full sequence testing to carry a VUS are offered free testing of additional family members in an effort to determine if the VUS segregates with cancer. In addition, a small number of families were recruited through an institutional review board–approved Mayo Clinic study of families with VUS. All family members enrolled in the Mayo Clinic study provided data on personal and family history of cancer and a blood sample for DNA extraction. All DNA samples were genotyped for the relevant variants. Families commonly recruited by Myriad Genetic Laboratories and Mayo Clinic were identified by comparing family history and age of diagnosis for cases.

Cosegregation with cancer in families. The segregation of VUS with cancer was evaluated in families using information provided by Myriad Genetics Laboratories or collected as part of an ongoing study of VUS containing families at the Mayo Clinic. Specifically, the likelihood ratio of cancer causality for each VUS was calculated under the hypothesis that a VUS has the same penetrance as the “average” established deleterious mutation, compared with the hypothesis that the VUS segregates independently of disease in the pedigree(s) under study (8, 11).

Co-occurrence with known deleterious mutations. Because co-inheritance in trans phase of two deleterious mutations of BRCA2 either induces embryonic lethality or causes Fanconi anemia (16), variants that co-occur in trans phase with known deleterious mutants are unlikely to be deleterious. Evidence of in trans phase co-occurrence of a variant with known deleterious mutations can be used to generate a likelihood ratio of cancer causality (7, 11). The odds of cancer causality based on co-occurrence for the VUS in this study were calculated using data from Myriad Genetic Laboratories (Table 1), in which the phase of the mutations with known deleterious mutations was established through examination of haplotypes of common polymorphisms in the BRCA2 gene (7, 8).

Family history. Individuals carrying known deleterious BRCA2 mutations exhibit personal and family history profiles of cancer that differ from those with wild-type BRCA2 in the Myriad Genetics Laboratories data set. Comparisons of known deleterious and known neutral variants based on disease status, age at diagnosis, and number and age of relatives with breast or ovarian cancer allows calculation of the probability of disease causality for each category and the derivation of a likelihood ratio of cancer causality for each mutation (11). This likelihood ratio was calculated for each BRCA2 VUS in the study.

Prior probability of pathogenicity based on evolutionary sequence analysis. VUS were classified as (a) enriched deleterious 1, (b) enriched deleterious 2, (c) unclassified, (d) enriched neutral, and (e) splicing alteration/in-frame deletion (9) based on the extent of conservation of the corresponding wild-type residue in a sequence alignment of BRCA2 from the pufferfish Tetraodon to human⁹

and on the chemical difference scores between residues based on the Grantham matrix (5, 8, 9). In addition, heterogeneity analysis of 1,433 variants in the Myriad Genetics Laboratories database (11) was used to estimate the proportion of variants in each of these categories that were deleterious (0.60 for enriched deleterious 1; 0.18 for enriched deleterious 2; 0.06 for unclassified; 0.01 for enriched neutral; and 0.96 for variants that were predicted to disrupt splicing; ref. 11). These values were then used to calculate prior probabilities of a variant being deleterious for the purposes of classification of BRCA2 VUS (Table 2).

Definition of thresholds of activity in the functional assays. A two-component mixture modeling approach was used to define the distributions of the results from both assays. The NOCOM program was used to estimate the variables of the mixture model for each assay using the average result of two replicates for each VUS (17, 18). Because the assays contained measurement error (the replicate to replicate R² were 78.5% for centrosome amplification and 92.8% for HR), a correction was made to the variance estimates from the mixture model. The final estimates included the mean and SD of the normal densities that approximated the distributions of assay results for VUS that belonged either to the “inactivating” or “no effect” groups. From these means and SDs, the probability that a VUS belonged either to the inactivating or to the no effect assay groups was calculated assuming that the component distributions were Gaussian. Final cutpoints/thresholds for both assays were estimated based on the final mixture distributions and were defined as the result at which a VUS had equal probability of being in the inactivating and no effect groups.

Estimation of sensitivity and specificity of functional assays. To estimate sensitivity and specificity of the assays, we compared the results with the posterior probability of pathogenicity for each VUS (defined using genetic and sequence data). Each VUS was first determined to be either negative or positive for each assay based on selected thresholds of activity and each VUS was also sampled as being pathogenic or neutral based on the posterior probability shown in Table 2. From this data set, the sensitivity (proportion of the VUS previously shown to have a high probability of pathogenicity that displayed loss of function) and specificity (proportion of the VUS shown to have a low probability of pathogenicity that displayed wild-type activity) were calculated for each assay. Having obtained estimates of sensitivity and specificity, we used properties of the binomial distributions to obtain exact 95% confidence intervals (95% CI) of these estimates.

Selection of VUS. We chose to study mutations from the DBD of BRCA2 because this domain is highly conserved at the amino acid level across evolution and seems to be critical for BRCA2 DNA repair activity. Fifteen VUS were chosen based on the availability of genotyping data from several members of three or more families (Table 1; Fig. 1). In addition, N319T from the NH₂ terminus of BRCA2 was included as a non-DBD variant to determine whether the assays can be applied to mutations in other domains of BRCA2 as has been suggested by other studies (14). We also selected the R2520Q, L2647P, R2659T, D2723A, R2784W, and L2865V mutations that occur in evolutionarily conserved residues and the A2643G variant from a nonconserved residue in the DBD (Fig. 1). Limited genetic data for these seven variants were available. The E462G, Y42C, N372H, and K3326X mutations were added as known neutral controls (Table 1; ref. 13). To estimate the frequency of the VUS in controls, we used 4,000 known carriers of BRCA1 deleterious truncating mutations from the Myriad Genetics Laboratories data set. The BRCA1 pathogenic mutation carriers were used as controls because their presence in the Myriad data set is independent of BRCA2 mutation status. All of the VUS were absent from these controls except N319T and V2908G, which occurred once each, suggesting that the VUS are rare in the population.

Combined odds of causality/neutrality. Genotyping of DNA from multiple members of participating families for the presence of the relevant VUS was conducted. An example of one large family is shown in Fig. 2A. Genotyping and cancer history data from these families were combined with information from Myriad Genetics Laboratories and individual likelihood ratios for co-occurrence, cosegregation, and family history for each VUS were calculated. These likelihood ratios were added to provide the combined odds of cancer causality. Overall, D2723H, R3052W, R2659K, I2627F, and G2748D displayed odds in favor of disease exceeding 1,000:1. Two other variants (E2663V, D2723G) reached odds in favor of disease causality of at least 100:1 and T2722R reached odds of 93:1 (Table 1). Conversely, R2973C, K2729N, V2908G, N319T, and the neutral controls Y42C, E462G, N372H, and K3326X all displayed combined odds in favor of neutrality of 50:1 or greater. The prior probabilities based on evolutionary sequence analysis were then combined with the combined odds of causality/neutrality (Table 1) to compute the posterior probability that each variant was pathogenic (Table 2). We noted that the nine VUS with odds >50:1 in favor of causality from the genetic data and the R2659T VUS were associated with posterior probabilities of cancer causality of >0.99. Conversely, the four variants and four controls displaying combined odds against disease causality of >50:1 all occurred in poorly conserved amino acid residues and had low probability of causality (high probability of neutrality).

Exon-skipping BRCA2 variants. We noted that four mutants in the study (R2336H, R2659K, R2659T, and E2663V) were located in consensus splice sites. The R2336H VUS, with odds of 50:1 in favor of cancer causality (Table 1) and a posterior probability of pathogenicity of 0.99 (Table 2), has been shown to induce aberrant splicing, deleting the 70-bp exon 13 from the mRNA and causing a frameshift and premature stop codon in exon 14 (19). For this reason, R2336H was not further evaluated by functional assay. Similarly, R2659K alters a splice consensus site in exon 17 resulting in an in-frame deletion of the 171-bp exon 17 through an exon-skipping event (20). We have previously shown that this in-frame deletion of exon 17 inactivates BRCA2 function (13). We further characterized the influence of the R2659T and E2663V variants on splicing using RNA from EBV-immortalized cell lines derived from individuals with the R2659T and E2663V variants. RT-PCR analysis of R2659T EBV cell line RNA using primers in flanking exons yielded a single product (Fig. 2B). Sequence analysis failed to detect the R2659T variant but revealed the absence of the 171-bp exon 17 (Fig. 2C). This shows that R2659T induces exon 17 skipping similarly to R2659K. The absence of the wild-type allele may reflect nonsense-mediated RNA degradation. Two RT-PCR products were obtained from the E2663V cell line (Fig. 2B). Sequence analysis of the smaller PCR product detected an out-of-frame deletion of the 355-bp exon 18, resulting in a premature stop codon within exon 19 (Fig. 2D). In contrast, the larger product exhibited wild-type sequence (Fig. 2D). RT-PCR analysis of 20 EBV cell lines from individuals with no BRCA2 variants or with the R496C or D2723A VUS (Fig. 2B) also detected alternative splicing of exon 18. However, the intensity of the exon-skipping product associated with the E2663V variant was significantly higher than in these controls, suggesting that the VUS enhances the efficiency of the exon-skipping event. Thus, R2659T and E2663V induce aberrant splicing of BRCA2 and likely inactivate BRCA2 function.

BRCA2 protein expression and Rad51 binding. Next, we studied the influence of these variants on BRCA2 function using both functional assays. We used full-length BRCA2 cDNA expression constructs for all assays to evaluate the influence of each VUS in the context of the entire protein. A BRCA2 construct with a 171-bp in-frame deletion of exon 17 was generated to represent the effects of the R2659K and R2659T exon-skipping variants (13, 20). E2663V- and R2336H-encoding variants were not generated because these mutations cause aberrant splicing and result in frameshift mutations. Ectopic expression of FLAG-tagged BRCA2 proteins in 293T cells after transient transfection was verified by immunoprecipitation and immunoblotting (Fig. 1B). All constructs expressed equivalently. In addition, no difference in the amount of Rad51 that associated with the wild-type and various mutant forms of BRCA2 was observed (Fig. 1C). This finding suggests that these VUS do not alter the conformation of BRCA2 sufficiently to disrupt the association with Rad51 and that any disruption of BRCA2 activity by VUS is likely domain specific.

Homology-directed recombination repair. BRCA2 has been implicated in HR repair of DNA double strand breaks through its ability to bind the Rad51 recombination protein and ssDNA at sites of damage (21–23). To investigate the influence of missense variants on BRCA2-dependent HR repair, we used a GFP-dependent homology-directed repair reporter assay (24) that was previously established in V-C8 BRCA2-deficient cells (13). FLAG-tagged wild-type and mutant BRCA2 constructs were introduced into DR-GFP containing V-C8 cells along with an I-Sce1 expression plasmid and the number of GFP-expressing cells was determined (Fig. 3A). After adjusting for transfection efficiency, the fold change in GFP-positive cells for wild-type and each VUS relative to vector control was established. Wild-type BRCA2 induced a 6-fold increase in GFP-expressing cells relative to vector (Fig. 3A and B). Similarly, N319T, R2520Q, A2643G, K2729N, L2865V, V2908G, and R2973C VUS displayed substantial increases in homology-directed recombination repair (HDR) activity (Fig. 3B). These effects are in keeping with those previously determined for the E462G, Y42C, and N372H known neutral controls (13). In contrast, I2627F, L2647P, L2653P, T2722R, D2723H, D2723G, D2723A, G2748D, R2784W, R3052W, and D3095E displayed reduced HDR activity (Fig. 3B). Similarly, the 171-bp in-frame deletion mutant representing R2659K and R2659T was deficient in HDR activity. We detected a substantial separation between the VUS with apparent wild-type activity and those that inactivated BRCA2 function in this HDR assay. The limits of these groups were defined by R2520Q (4.0-fold) and T2722R (2.5-fold). The optimal cutpoint, or threshold of activity for the assay, was estimated at 3.4-fold.

Induction of centrosome amplification. Because disruption or depletion of BRCA2 leads to centrosome amplification (25, 26), it has been postulated that regulation of centriole duplication is an important function of BRCA2 in tumor suppression. Here, we evaluated the ability of BRCA2 VUS to induce both centriole and centrosome amplification in 293T cells (13). BRCA2 variants and wild-type BRCA2 were ectopically expressed in these cells, and following immunofluorescence against FLAG-tag and the centrin-2 centriole marker, the proportion of cells displaying an abnormal centriole complement was enumerated. The N319T, R2520Q, A2643G, K2729N, R2784W, L2865V, V2908G, R2973C, and R3052W variants maintained the background level of centriole and centrosome amplification found in wild-type cells (Fig. 4A and B). In contrast, I2627F, L2647P, L2653P, R2659K (and R2659T), T2722R, D2723A, D2723G, D2723H, G2748D, and D3095E induced at least a 2-fold increase in the proportion of cells undergoing aberrant centriole amplification (Fig. 4A and B), suggesting that these variants disrupt BRCA2 function.

We noted that the R3052W variant displayed the highest level of centrosome amplification (14%) from the group of variants with wild-type BRCA2 activity, whereas D2723A displayed the lowest level of amplification (16.2%) from the group with aberrant centrosome amplification. As there is not much separation between these variants, we confirmed these results in a second centrosome amplification assay. V-C8 cells deficient in BRCA2 were reconstituted with wild-type and mutant forms of BRCA2 and the ability of the variants to reduce the levels of centrosome amplification similarly to wild-type was measured. In this assay, A2643G, R3052W, and K2729N reduced the frequency of centriole amplification to the same extent as wild-type protein, whereas D2723A showed no ability to rescue centriole amplification (Fig. 4C). These results are consistent with the centriole amplification induction assay and identify D2723A as a VUS that disrupts this function of BRCA2 while verifying that the R3052W VUS has no effect on this function of BRCA2. The optimal cutpoint or threshold of activity that discriminates between wild-type and inactive BRCA2 was estimated at 16.2%.

Sensitivity and specificity. As noted above, we conducted HR and centrosome amplification assays for eight VUS that displayed odds in favor of causality of >0.99 and four VUS along with four known neutral controls that displayed odds in favor of neutrality of 1 × 10⁻³ (Table 2). Two other VUS with prior probabilities in favor of causality of >0.99 were not evaluated because they truncated the BRCA2 protein through induction of aberrant splicing. We observed a perfect correlation between the outcome of the HR assay and the posterior probability of pathogenicity for these 16 VUS. Similarly, we observed a strong correlation with the probability estimate for the centrosome amplification assay results, with only the R3052W VUS displaying a difference (Table 2). The estimated sensitivity for HR was 100% (95% CI, 63-100), whereas for the centrosome amplification assay, the estimate for HR was 85% (95% CI, 47-100). In terms of specificity, the estimate for both the HR and centrosome amplification assay was 100% (95% CI, 63-100). Thus, based on these 16 variants, the HR assay has higher sensitivity and the same specificity as the centrosome amplification assay, although the differences are slight.

In this study, we have used genetic data to identify 10 BRCA2 VUS with posterior probabilities of pathogenicity >0.99 and four with posterior probabilities of pathogenicity <1 × 10⁻³ using a combined likelihood model and evolutionary sequence comparisons (Table 1). Although these estimates may be confounded by possible differential penetrance of VUS that is not accounted for in the combined likelihood model, our findings still suggest that this approach is a powerful method for classifying VUS when substantial family data are available. However, the need for multiple families carrying the same mutation for classification is also a limitation of the approach. Even variants such as D3095E that has been observed in 13 families did not reach odds in favor of causality of 1,000:1 because the families carrying the mutations were small and provided limited family history or because insufficient family members were recruited into the study. The efficacy of this method is strongly related to the ability of researchers and clinical groups to obtain detailed family histories and mutation data on multiple members of families known to carry BRCA2 or BRCA1 variants.

Because of the limitations of the genetic approach for classifying BRCA2 variants, we developed two assays of BRCA2 function with the intent of establishing that these assays can discriminate between VUS that inactivate or have no effect on BRCA2 and may contribute to multivariate models for classification of the cancer causality of BRCA2 VUS. To evaluate the ability of the assays to discriminate between VUS that inactivate or have no effect on BRCA2, we compared the results from the assay with the results from the multifactorial likelihood model for 16 VUS with posterior probabilities of pathogenicity of >0.99 or <1 × 10⁻³. After selecting thresholds of activity that discriminated between active and inactive BRCA2, we found that the outcomes from the in vitro assays correlated well with the genetic and sequence-derived posterior probabilities of pathogenicity. We formally calculated the sensitivity and specificity of the assays and found that the HR assay exhibited 100% sensitivity and specificity whereas the centrosome amplification assay showed 85% sensitivity and 100% specificity. Although these findings are based on only 16 VUS, which results in wide CIs, it seems that these assays can effectively distinguish between VUS in the DBD that inactivate or have no effect on BRCA2 function. Further sensitivity and specificity studies relative to the posterior probability of pathogenicity involving additional VUS must now be undertaken.

We noted that the centrosome amplification assay displayed lower sensitivity than the HDR assay. Specifically, the R3052W variant displaying a posterior probability of pathogenicity of 1.0 did not induce centrosome amplification but disrupted HR. Similarly, the R2784W DBD VUS with a posterior probability of pathogenicity of 0.643 did not induce centrosome amplification but inactivated BRCA2 HR activity. These findings suggest that the BRCA2 protein may have completely independent functions that can be measured as effects on HR repair of DNA and centrosome amplification. As additional VUS are analyzed in this way, we also expect to identify variants that cause centrosome amplification but have no effect on HDR activity. In terms of defining the influence of VUS on BRCA2 function, these findings indicate that all VUS should be analyzed by both methods. If only one assay is used, it is possible that a VUS that inactivates a different function of BRCA2 might be identified as having no functional effect. Furthermore, the multifunctionality of BRCA2 raises the possibility that other functions of BRCA2 that can be disrupted by VUS are not measured by the two assays described here. Although this is formally possible, the finding that the two assays exhibited 100% sensitivity for the eight VUS with high posterior probabilities of pathogenicity suggests that few inactivating VUS in the DBD will not be accounted for by the HR and centrosome amplification assays.

Although the high sensitivity and specificity of the two assays relates to VUS in the BRCA2 DBD, it is possible that the assays could be applied to the characterization of VUS from other domains of BRCA2. Separate studies from our group have shown a strong correlation between variants in the NH₂-terminal PALB2 interaction domain of BRCA2 that disrupt PALB2 binding and both the HR and centrosome amplification assays (26). However, the influence of VUS in other domains of BRCA2 on protein function has not been determined using this methodology. More extensive studies that evaluate mutations in other domains of BRCA2 are needed to determine the relevance of the two assays to overall BRCA2 integrity. In addition, it is now possible to perform detailed structure-function studies of the BRCA2 COOH-terminal DBD using the assays described here in conjunction with the predicted three-dimensional structure of the domain. These studies should lead to an improved understanding of the DNA repair and centrosome regulatory functions of BRCA2 and may lead to improved therapeutic targeting of wild-type and mutant forms of BRCA2.

In the course of our studies, we made the observation that 4 of the 10 variants with posterior probabilities of pathogenicity of >0.99 resulted in aberrant splicing or exon skipping that inactivated the BRCA2 protein. These findings suggest that all functional testing of BRCA2 VUS should begin with an evaluation of splicing. This can be accomplished by RT-PCR of RNA extracted from blood or tissue specimens of patients as shown here or perhaps by using in vitro approaches. These studies can be complicated by tissue-specific expression of splice forms and by the absence of the complete genomic context when using in vitro models. Furthermore, as shown for the R2659K VUS, the absence of the mutant VUS due to nonsense-mediated mRNA degradation can be problematic. However, in terms of BRCA2, the most difficult aspect of splicing analysis is associated with the many rare alternative splice products that can be detected in normal tissue. As shown in the splicing analysis of E2663V, the VUS enhances the expression of a rare isoform of BRCA2 containing an in-frame deletion of exon 18. Thus, to avoid overinterpretation of results, it is imperative that control mRNAs are used and that the efficiency of the aberrant splicing is carefully evaluated.

Overall, the strong concordance between the genetic data and the functional assays strongly suggests that the functional assays can be used to differentiate between VUS that inactivate or have no influence on BRCA2 function. However, these results do not necessarily reflect the influence of the VUS on cancer risk. Further studies will be needed to establish the relevance of the functional assay result to cancer risk. However, the strong concordance between the assays and the probability of pathogenicity suggests that these assays may in time prove useful as a component of a multivariate model for cancer risk assessment of VUS carriers.

Grant support: National Cancer Institute Breast Cancer Specialized Program in Research Excellence grants P50 CA116201 and R01 CA116167, American Cancer Society award RSG-04-220-01-CCE, and the Breast Cancer Research Foundation (F.J. Couch); U.S. Army Medical Research and Materiel Command grant W81XWH-06-1-0480 (D.J. Farrugia); grants from the National Breast Cancer Foundation, the National Health and Medical Research Council (NHMRC), and the Queensland Cancer Fund; the Cancer Councils of New South Wales, Victoria, Tasmania, and South Australia; and the Cancer Foundation of Western Australia [The Kathleen Cunningham Foundation Consortium for Research into Familial Breast Cancer (kConFab)].

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.

We thank Jennifer Scott for assistance with the preparation of the manuscript; Heather Thorne, Eveline Niedermayr, all the kConFab research nurses and staff, the heads and staff of the Family Cancer Clinics, and the Clinical Follow-up Study (funded by NHMRC grants 145684 and 288704) for their contributions to this resource; and the many families who contribute to kConFab.