GFP-Fragment Reassembly Screens for the Functional Characterization of Variants of Uncertain Significance in Protein Interaction Domains of the BRCA1 and BRCA2 Genes

Genetic testing for BRCA1 and BRCA2 genes has led to the identification of many unique variants of uncertain significance (VUS). Multifactorial likelihood models that predict the odds ratio for VUS in favor or against cancer causality, have been developed, but their use is conditioned by the amount of necessary data, which are difficult to obtain if a variant is rare. As an alternative, variants mapping to the coding regions can be examined using in vitro functional assays. BRCA1 and BRCA2 proteins promote genome protection by interacting with different proteins. In this study, we assessed the functional effect of two sets of variants in BRCA genes by exploiting the green fluorescent protein (GFP)-reassembly in vitro assay, which was set-up to test the BRCA1/BARD1, BRCA1/UbcH5a, and BRCA2/DSS1 interactions. Based on the findings observed for the validation panels of previously classified variants, BRCA1/UbcH5a and BRCA2/DSS1 binding assays showed 100% sensitivity and specificity in identifying pathogenic and non-pathogenic variants. While the actual efficiency of these assays in assessing the clinical significance of BRCA VUS has to be verified using larger validation panels, our results suggest that the GFP-reassembly assay is a robust method to identify variants affecting normal protein functioning and contributes to the classification of VUS.


Introduction
Genetic testing for the breast cancer susceptibility genes, BRCA1 (MIM# 113705) and BRCA2 (MIM# 600185), has become part of routine clinical care for patients with a personal or family history indicative of Hereditary Breast/Ovarian Cancer Syndrome (HBOC). A recent prospective study estimates the average cumulative risks by age 70 years of BRCA1 carriers to be 60% and 59% for breast and ovarian cancer, respectively. For BRCA2 carriers, the corresponding values are 55% and 16.5%, respectively [1]. Most pathogenic variants identified in the BRCA1 and BRCA2 genes are nonsense or frameshift that, by introducing a premature termination codon, lead to non-functional proteins. Splice site variants with a similar effect on the protein have also been identified [2]. In addition to these clearly loss-of-function alterations, a growing number of variants of uncertain significance (VUS) are currently being identified. These include missense substitutions, small in-frame deletions or insertions, alterations of both exonic and intronic sites potentially affecting splicing, and variants in regulatory sequences. The inability to classify the VUS either as disease causing or as rare non-pathogenic (benign) variation represents a challenge for genetic counselling, risk evaluation, and adoption of preventive and risk-reduction measures.
A five-tier classification scheme has been developed for the assessment of the clinical relevance of VUS by the International Agency for Research on Cancer (IARC) Working Group [3]. According to this scheme, Class 1 (posterior probability of pathogenicity < 0.001) and Class 2 (0.001 < posterior probability < 0.049) are considered non-pathogenic and likely non-pathogenic variants, Class 3 (0.05 < posterior probability < 0.949) VUS remain unclassified because of a lack of sufficient information for classification, Class 4 (0.95 < posterior probability < 0.99) and Class 5 (posterior probability > 0.99) are likely pathogenic and pathogenic variants, respectively. Posterior probability may be estimated using a multi-factorial model. For missense variants, this model combines a prior probability of causality, based on the amino acid evolutionary conservation model (A-GVGD, URL: http://agvgd.iarc.fr/) [4], with a likelihood ratio for causality derived from genetic and family-based data and histopathological tumor features [5][6][7][8]. However, since most BRCA1 and BRCA2 VUS are detected in a single or a few families only, this model is not able to classify the majority of them. Therefore, functional assays have been developed to independently classify VUS based on their effect on established functions of the BRCA1 and BRCA2 proteins [9,10].
Mounting evidence implicates both BRCA1 and BRCA2 proteins in all phases of the cell cycle and in the regulation of its orderly events [11] through their association with different proteins [12,13]. As a consequence, a defective protein-protein interaction may lead to uncontrolled cell replication and genomic instability, which are both hallmarks of tumor formation and progression. Thus, biochemical in vitro assays analyzing the formation of protein complexes provide a context in which to assess the functional effect of the BRCA1 [14,15] and BRCA2 VUS [16][17][18].
One of the most important BRCA1-interacting proteins is BARD1 (MIM# 601593). Under physiological conditions, BRCA1 is prevalently found to be associated with BARD1. This interaction is mediated by the N-terminal RING-finger domains of both proteins [19], which suggests that the heterodimer is the physiological relevant form of BRCA1 and the two proteins require each other for most of their cellular functions. Several lines of evidence support this view: BRCA1-BARD1 binding appears to stabilize BRCA1 protein expression [20] and to increase its nuclear accumulation by masking the nuclear export sequence (NES) flanking the RING-finger domain [21]. The two proteins co-localize in nuclear dots during the S-phase of the cell cycle and in nuclear foci at sites of DNA damage [22][23][24][25]. Most importantly, the binding with the E2 ubiquitin-conjugating enzyme UbcH5a (MIM# 602961) enables the BRCA1-BARD1 complex to acquire an ubiquitin ligase activity, linked to BRCA1 tumor suppression function by regulating transcription and double stand break (DSB) repair [26][27][28]. This is consistent with the reported hypersensitivity to Ionizing Radiation (IR) and to agent crosslinking the DNA, such as Cisplatin or Mitomycin C (MMC), of cells carrying BRCA1 RING pathogenic variants [29][30][31].
Concerning BRCA2, the strong physical interaction with the small (70-residues) DSS1 peptide (MIM# 601285), through its highly conserved C-terminal DSS1/ssDNA Binding Domain (DBD), appears to be crucial for the maintenance of its stability and correct conformation [32,33], for the repair of DNA DSBs by homologous recombination (HR) and the formation of DNA damage induced RAD51 foci [34]. Recently, it was shown that the nuclear retention of BRCA2 is regulated by DSS1 interaction through the masking of a NES, such that variation impairing BRCA2-DSS1 binding mis-localizes BRCA2 to the cytoplasm [35].
In this study, we analyzed the effect of a group of VUS mapped to the BRCA1 and BRCA2 gene regions encoding the RING-finger domain of BRCA1 and the DBD of BRCA2 on the binding with the interacting proteins BARD1, UbcH5a, and DSS1, using the previously established GFP-fragment reassembly screening [15,36]. In this assay, the GFP is dissected into two fragments (NfrGFP and CfrGFP) that, when expressed together in E. coli bacterial cells, do not spontaneously reassemble into a fluorescence protein. However, if each of the two fragments of the GFP are individually fused to two interacting protein domains, this interaction can mediate the reassembly of the GFP in co-transformed bacteria with consequent cellular fluorescence. We evaluated the performance of these approaches in correctly discriminating between pathogenic and non-pathogenic analysis of a panel of variants classified according to the IARC 5-class model (Leiden Open Variation Database, URL: http://hci-exlovd.hci.utah.edu/variants.php). In addition, we verified the consistency of the VUS classification with the results of other published functional analyses investigating the same variants [14,30,[37][38][39][40][41][42][43][44][45][46][47][48][49][50].

Selection of the BRCA1 and BRCA2 Variants
We selected a total of 58 BRCA1 (n = 36) and BRCA2 (n = 22) variants mapping to the gene regions coding for the conserved motifs encompassing the RING-finger domain of BRCA1 and the DBD of BRCA2 [13] (Figure 1). Of these, 28 BRCA1 (n = 13) and BRCA2 (n = 15) variants, classified by the IARC-5 class system as pathogenic or non-pathogenic, were used for assay validation (Table 1). Thirty BRCA1 (n = 23) and BRCA2 (n = 7) variants were VUS (Table 2) of which 21 were identified among families referred for BRCA gene testing. Of the remaining nine variants, seven were extracted from the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar/), while two, p.Cys27Ala and p.His41Ala, were synthetic variants not yet observed in HBOC patients and were included so that at least one variant was examined for each of the critical cystein and histidine residues of the RING domain.    The N-terminal RING domain (aa 1-109) mediates the E3-ubiquitin ligase activity of BRCA1. It consists of a RING finger motif (aa 23-76) and two flanking α helices. It is characterized by a conserved pattern of seven cysteine and one histidine residues, which form two Zn 2+ -binding sites (termed Site I and Site II) stabilizing the RING structure. The RING finger motif assembles in a globular structure with three strand β-sheet and a central helix.
All 36 mutant constructs were co-transformed in bacterial cells with the pET11a-NfrGFP-BARD1 and pET11a-NfrGFP-UbcH5a constructs expressing the wild-type BARD1 (BD N ) and UbcH5a (Ub N ) proteins, which is in frame with the N-terminal fragment of the GFP protein, respectively. In addition, pET11a-NfrGFP-BARD1 and pET11a-NfrGFP-UbcH5a were co-transformed with pMRBAD-BRCA1-CfrGFP expressing wild-type BRCA1 (BR1 C ) and pMRBAD-Z-CfrGFP expressing the leucine zipper (Z) peptide (Z C ), both in a frame with the C-terminal fragment of the GFP protein, as positive and non-cognate negative controls. The results are listed in Table 3.
Under inducing conditions, a bright fluorescence was observed in bacterial cells co-expressing the strong interacting anti-parallel leucine zipper NfrGFP-Z/Z-CfrGFP (Z N /Z C ) fusion peptides and the positive controls (BD N /BR1 C , Ub N /BR1 C ). No fluorescence was observed in bacterial cells co-expressing the non-cognate negative controls (BD N /Z C , Ub N /Z C ) (Figures 2 and 3). The following variants of the validation panel showed a fluorescence resembling that of the positive controls: p.Lys45Gln and p.Asp67Tyr (Class 1), p.Met18Thr (Class 4), p.Thr37Lys and p.His41Arg (Class 5), while no fluorescent signal was observed for the validation panel variants p.Leu22Ser, p.Cys39Arg, p.Cys39Tyr, p.Cys44Phe, p.Cys44Ser, p.Cys44Tyr, p.Cys61Gly and p.Cys64Tyr (all class 5). Among the analyzed VUS, p.Ile15Thr, p.Leu28Pro, p.Ile31Met, p.Ser36Tyr, p.Ile42Val, p.Met48Lys, p.Cys64Gly, p.Cys64Arg, p.Thr77Met, and p.Gly98Val showed bright fluorescence. Conversely, no fluorescence was observed in bacteria expressing the variants p.7-10delinsGln, p.Ile15Val, p.Met18Lys, p.Cys24Arg, p.Cys27Ala, p.Pro34Ser, p.Thr37Arg, p.Cys39Gly, p.His41Ala, p.Cys47Gly, p.Met48Arg, p.Cys64Tyr, p.Thr69Met., and p.Val83Gly (Figure 2). Under inducing conditions, a bright fluorescence was observed in bacterial cells co-expressing the strong interacting anti-parallel leucine zipper NfrGFP-Z/Z-CfrGFP (Z N /Z C ) fusion peptides and the positive controls (BD N /BR1 C , Ub N /BR1 C ). No fluorescence was observed in bacterial cells coexpressing the non-cognate negative controls (BD N /Z C , Ub N /Z C ) (Figures 2 and 3). The following variants of the validation panel showed a fluorescence resembling that of the positive controls: p.Lys45Gln and p.Asp67Tyr (Class 1), p.Met18Thr (Class 4), p.Thr37Lys and p.His41Arg (Class 5), while no fluorescent signal was observed for the validation panel variants p.Leu22Ser, p.Cys39Arg, p.Cys39Tyr, p.Cys44Phe, p.Cys44Ser, p.Cys44Tyr, p.Cys61Gly and p.Cys64Tyr (all class 5). Among the analyzed VUS, p.Ile15Thr, p.Leu28Pro, p.Ile31Met, p.Ser36Tyr, p.Ile42Val, p.Met48Lys, p.Cys64Gly, p.Cys64Arg, p.Thr77Met, and p.Gly98Val showed bright fluorescence. Conversely, no fluorescence was observed in bacteria expressing the variants p.7-10delinsGln, p.Ile15Val, p.Met18Lys, p.Cys24Arg, p.Cys27Ala, p.Pro34Ser, p.Thr37Arg, p.Cys39Gly, p.His41Ala, p.Cys47Gly, p.Met48Arg, p.Cys64Tyr, p.Thr69Met., and p.Val83Gly (Figure 2). The screening assessing the effect of BRCA1 variants on the binding to UbcH5a revealed bright fluorescence for bacterial cells co-expressing NfrGFP-UbcH5a together with BRCA1-CfrGFP bearing The screening assessing the effect of BRCA1 variants on the binding to UbcH5a revealed bright fluorescence for bacterial cells co-expressing NfrGFP-UbcH5a together with BRCA1-CfrGFP bearing the class1 variants p.Lys45Gln and p.Asp67Tyr and the VUS p.Ile31Met, p.Ile42Val, p.Thr77Met, p.Ile15Val, and Val83Gly in the helical bundle interface ( Figure 3). Conversely, all tested class 4 and class 5 variants were not able to bind to UbcH5a. These results indicate 100% sensitivity and specificity of this assay. All remaining tested VUS, including some of those that maintained the interaction with BARD1, namely p.Ile15Thr, p.Leu28Pro, p.Ser36Tyr, and p.Met48Lys, abrogated the binding with UbcH5a.  Figure 3). Conversely, all tested class 4 and class 5 variants were not able to bind to UbcH5a. These results indicate 100% sensitivity and specificity of this assay. All remaining tested VUS, including some of those that maintained the interaction with BARD1, namely p.Ile15Thr, p.Leu28Pro, p.Ser36Tyr, and p.Met48Lys, abrogated the binding with UbcH5a. To confirm that the loss of fluorescence was attributable to the lack of binding to BARD1 and/or UbcH5a and not to poor expression of the BRCA1 mutants, the cell lysates were visualized by Western blotting against the GFP tag. The results showed that all fusion peptides were expressed to a similar extent ( Figure 4A). To confirm that the loss of fluorescence was attributable to the lack of binding to BARD1 and/or UbcH5a and not to poor expression of the BRCA1 mutants, the cell lysates were visualized by Western blotting against the GFP tag. The results showed that all fusion peptides were expressed to a similar extent ( Figure 4A).

Evaluation of the Effect of the BRCA2 Variants on the Interaction with the DSS1 Protein by the GFP-Reassembly Assay
The C-terminal region of BRCA2 contains a DSS1/ssDNA Binding Domain (DBD) consisting of five components. The first is described as the helical domain (HD) due to its 190 amino acids consisting mainly of α-helices. This is followed by three oligonucleotide/oligosaccharide binding folds (OB1, OB2 and OB3). In addition, within OB2, a 130 amino acid insertion exists, which was named the tower domain due to its tower-like structure protruding from the OB fold. The crystallographic study of BRCA2-DBD revealed a stoichiometric complex with DSS1, which is almost entirely mediated by highly conserved residues spanning the HD and OB1 motifs. Solely, one DSS1 residue interacts with the first portion of the OB2 motif [32].
We replaced the leucine zipper peptides in the original constructs (pET11a-NfrGFP-Z and pMRBAD-Z-CfrGFP) with HD, the three OB motifs, each of which is in a distinct construct and the protein DSS1 is in all the possible orientations (i.e., pET11a-NfrGFP-HD/OB1/OB2/OB3 pET11a-NfrGFP-DSS1, pMRBAD-HD/OB1/OB2/OB3-CfrGFP, and pMRBAD-DSS1-CfrGFP). Upon cotransformation in bacterial cells, bright fluorescence was observed only when the HD and OB1 were fused to the NfrGFP and DSS1 was fused to the CfrGFP while a less pronounced fluorescence was observed in bacterial cells co-expressing BRCA2-OB2 and DSS1 inserted in the same orientation ( Figure 5). No fluorescence was detected when the above inserts were fused in the opposite

Evaluation of the Effect of the BRCA2 Variants on the Interaction with the DSS1 Protein by the GFP-Reassembly Assay
The C-terminal region of BRCA2 contains a DSS1/ssDNA Binding Domain (DBD) consisting of five components. The first is described as the helical domain (HD) due to its 190 amino acids consisting mainly of α-helices. This is followed by three oligonucleotide/oligosaccharide binding folds (OB1, OB2 and OB3). In addition, within OB2, a 130 amino acid insertion exists, which was named the tower domain due to its tower-like structure protruding from the OB fold. The crystallographic study of BRCA2-DBD revealed a stoichiometric complex with DSS1, which is almost entirely mediated by highly conserved residues spanning the HD and OB1 motifs. Solely, one DSS1 residue interacts with the first portion of the OB2 motif [32].
We replaced the leucine zipper peptides in the original constructs (pET11a-NfrGFP-Z and pMRBAD-Z-CfrGFP) with HD, the three OB motifs, each of which is in a distinct construct and the protein DSS1 is in all the possible orientations (i.e., pET11a-NfrGFP-HD/OB1/OB2/OB3 pET11a-NfrGFP-DSS1, pMRBAD-HD/OB1/OB2/OB3-CfrGFP, and pMRBAD-DSS1-CfrGFP). Upon co-transformation in bacterial cells, bright fluorescence was observed only when the HD and OB1 were fused to the NfrGFP and DSS1 was fused to the CfrGFP while a less pronounced fluorescence was observed in bacterial cells co-expressing BRCA2-OB2 and DSS1 inserted in the same orientation ( Figure 5). No fluorescence was detected when the above inserts were fused in the opposite orientation and co-transfected likely due to a conformational restraint of GFP-reassembly under these conditions. Consistently with the BRCA2 DBD-DSS1 crystal structure, no fluorescence was observed in bacteria co-expressing BRCA2-OB3 and DSS1 in either orientation ( Figure 5 and Supplemental Figure S1). orientation and co-transfected likely due to a conformational restraint of GFP-reassembly under these conditions. Consistently with the BRCA2 DBD-DSS1 crystal structure, no fluorescence was observed in bacteria co-expressing BRCA2-OB3 and DSS1 in either orientation ( Figure 5 and Supplemental Figure S1). Based on the above results, we decided to evaluate the effect of the BRCA2 variants mapped to the gene regions encoding for the HD and OB1 motifs only. All 22 mutants and the corresponding wild-type forms were co-transformed with the pMRBAD-DSS1-CfrGFP expressing the wild-type DSS1. In addition, the pET11a-NfrGFP-HD was co-transformed with the pMRBAD-Z-CfrGFP expressing the leucine zipper peptide as a non-cognate negative control. Based on the above results, we decided to evaluate the effect of the BRCA2 variants mapped to the gene regions encoding for the HD and OB1 motifs only. All 22 mutants and the corresponding wild-type forms were co-transformed with the pMRBAD-DSS1-CfrGFP expressing the wild-type DSS1.
In addition, the pET11a-NfrGFP-HD was co-transformed with the pMRBAD-Z-CfrGFP expressing the leucine zipper peptide as a non-cognate negative control.
As reported in Figure 5, no fluorescence was observed in bacterial cells co-expressing non-cognate negative control NfrGFP-HD/Z-CfrGFP (HD N /Z C ) while bright fluorescence was detected in bacteria co-expressing the positive controls Z N /Z C , NfrGFP-HD/DSS1-CfrGFP (HD N /DSS1 C ), and NfrGFP-OB1/DSS1-CfrGFP (OB1 N /DSS1 C ). Comparable fluorescence was observed in bacterial cells co-expressing DSS1-CfrGFP with NfrGFP fused to the BRCA2 peptides bearing all the variants of Class 1/2 and the following VUS: p.Arg2502His, p.Tyr2543Cys, p.Ser2546Pro, p.Ile2550Thr, p.Asp2566Glu, and p.Asn2644Asp (HD). No fluorescence was observed in all the bacteria expressing BRCA2 peptides carrying the Class 4/5 mutations and the VUS p.Phe2562Leu (HD). The expression level of all the NfrGFP-HD/OB1 peptides was verified by Western blotting analysis using an anti-GFP antibody and no significant differences were observed ( Figure 4B).

Validation of the GFP-Reassembly Screenings
The re-assembled complexes were purified by Immobilized Metal Affinity Chromatography (IMAC) from the soluble fraction of co-transformed bacterial cell lysates exploiting the hexahistidine tag (H 6 ) fused to the N-terminus of the NGFP fragments [14,15]. Equal aliquots of the cleared lysates of co-transformed E. coli BL21 (DE3) cells were incubated with Ni-NTA agarose resin and the affinity-captured material was analyzed by Western blotting using a polyclonal anti-GFP antibody (Supplemental Figures S2-S4). The detection of two bands corresponding to the complex components confirmed interactions with the binding proteins for all BRCA1 and BRCA2 constructs that tested positive in the fluorescence reassembly assay. No bands, or bands of very low intensity, were observed in case BRCA1 and BRCA2 variants for whom no fluorescence was detected, which confirms the lack of interaction. Assay specificity was supported by the absence of signals detected in cell lysates co-expressing the NfrGFP-BARD1, NfrGFP-UbcH5a, or the NfrGFP-HD with the CfrGFP-Z peptide (non-cognate negative controls).

Evaluation of the Effect of VUS on mRNA Splicing
Exonic variants may affect protein functions not only by altering the amino acid sequence, but also abolishing naturally-occurring splice sites or introducing/activating new or cryptic splice sites. To account for this possibility, we performed an in silico analysis on all 30 VUS included in the study, from which only three were predicted to be spliceogenic by ≥three algorithms of the Alamut software. These included BRCA1 c.110C>G and c.115T>G, predicted to introduce novel acceptor splice sites, and BRCA1 c.190T>G, predicted to reinforce the strength of an alternatively used donor splice site in exon 5. Existing mRNA assay data reports no effect on mRNA splicing for BRCA1 c.110C>G [57], while BRCA1 c.190T>G has been shown to lead to increased levels of a non-productive naturally-occurring isoform (∆5q) [58], which carries the out of frame deletion of the distal portion of exon 5 [59]. To the best of our knowledge, no experimental data has been reported as for the c.115T>G. No spliceogenic effect was predicted for all remaining VUS.

Discussion
The reporting of VUS can cause uncertainty and anxiety regarding cancer risk and clinical management. In addition, appropriate medical care may be delayed for individuals carrying VUS that are later determined to be pathogenic. In the near future, the relevance of these questions is likely to rise, as the relative frequency of VUS on the total number of variants identified in HBOC individuals is expected to increase dramatically with the development of next generation sequencing (NGS) analyses and the possibility to screen gene panels containing several verified or putative breast/ovarian cancer predisposition genes at minimal additional costs [60]. Furthermore, a study evaluating the coding regions and exon-intron boundaries of a 42-cancer gene sequencing panel (including BRCA1 and BRCA2) identified one or more VUS in approximately 90% of the patients [61]. This underlines the need to develop accurate methods to discriminate VUS that are actually related to cancer risk (pathogenic) from those without clinical relevance (non-pathogenic or benign).
In this study, we evaluated the use of an in vitro protein-fragment complementation assay and the GFP-fragment reassembly screening for the functional characterization of BRCA1 and BRCA2 VUS. We had previously applied this method to functionally characterize an Italian founder variant (p.Cys64Arg) mapped to the RING-finger domain of BRCA1 [15]. This variant was shown to abolish the interaction of BRCA1 with the partner protein BARD1, in a manner similar to the known cancer predisposing variant p.Cys61Gly [62]. In this case, we applied the same approach to analyze the effect of 25 BRCA1 VUS mapped to the RING-finger domain on the binding with BARD1 and UbcH5a proteins and seven BRCA2 VUS mapped to the DBD domain on the binding with the DSS1 protein.
To evaluate the accuracy of the assays in discriminating pathogenic from non-pathogenic variants, we tested the effect of a panel of 13 BRCA1 and 15 BRCA2 variants (validation panels) previously classified according to the IARC 5-class system. The assays analyzing the BRCA1/UbcH5a and BRCA2/DSS1 interactions show 100% sensitivity (percentage of correctly classified pathogenic variants) and 100% specificity (percentage of correctly classified non-pathogenic variants). In fact, all constructs carrying the Class 1 and 2 variants included in the validation panels displayed complex formation with the selected binding partners, while all constructs carrying the Class 4 and 5 variants did not. Based on these findings, our investigation shows that 19 (18 BRCA1 and 1 BRCA2) VUS behaved as pathogenic and 11 (5 BRCA1 and 6 BRCA2) as neutral variants (Figures 3 and 5, Tables 3 and 4).
Two issues need to be considered when evaluating the performance of the GFP-fragment reassembly assay. First, since the assays described are based on an artificial experimental model, where functional domains of human proteins are expressed in prokaryotic cells, the behavior of mutant constructs that we observed might not reflect the properties of full-length BRCA proteins expressed in their 'natural' context. However, it must be considered that, although our observations might actually provide clues for the molecular mechanisms mediating the tumor suppressor activity of BRCA proteins (see below), the aim of this study was essentially to collect evidence for the clinical relevance of BRCA VUS by the use of functional assays that efficiently dichotomize variants into pathogenic or non-pathogenic. Second, some exonic non-synonymous changes can affect gene functionality by altering mRNA splicing. In a few cases, i.e., when the spliceogenic allele retains the ability to produce a full-length protein, this occurrence may co-exist with a pathogenic effect of the missense substitution such as in the case of the BRCA2 c.8168A>G (p.Asp2723Gly) [47,56]. Previous data and in silico analyses revealed for two of the BRCA1 VUS we tested, an actual (c.190T>G) or potential (c.115T>G) spliceogenic effect. Based on the outcomes of the BRCA/UbcH5a interaction assay, all the above variants were predicted to be pathogenic. However, no data are presently available to ascertain if full-length proteins carrying such variants are actually expressed, and, therefore, no data are present to hypothesize dual mRNA/protein defects.
Similar results were observed using alternative approaches in independent studies (Table 3). Morris and co-workers reported a comprehensive analysis of missense mutations mapped to the RING-finger domain of BRCA1 and, using yeast 2-hybrid (Y2H) and in vitro ubiquitin ligase assays, they showed that, in the presence of the p.Met18Thr variant, BRCA1: (a) retains the interaction with BARD1, (b) is no longer able to bind UbcH5a, and (c) loses the E3 ubiquitin ligase activity [37]. Ransburgh and co-workers exploiting the co-immunoprecipitation approach showed that the variant p.Met18Thr only modestly affects the interaction between BRCA1 and BARD1 [38]. Moreover, a recent analysis measuring the effect of approximately 2000 BRCA1 missense substitutions, both on BARD1 binding and E3 ubiquitin ligase activity, reported that the BRCA1 proteins carrying the variants p.Met18Thr, p.Thr37Lys, and p.His41Arg are able to interact with BARD1 and lose the E3 ligase activity [43]. All these observations, together with our results, suggest that, in the absence of the BRCA1-UbcH5a complex, the maintenance of the binding with BARD1 is dispensable for the BRCA1 tumor suppressor activity. Thus, we speculate that the reduced or abrogated E3 ubiquitin ligase activity, through loss of binding with UbcH5a, is correlated with disease susceptibility and the prediction of pathogenicity. The observations of Drost and co-workers support this hypothesis, who reported that mice carrying the pathogenic p.Cys61Gly variant, inactivating BRCA1 E3 ligase activity [37], are embryonically lethal because of a developmental delay and that, in addition, the latency and nature of mammary tumors with one Brca1 allele carrying the p.Cys61Gly variant and one Brca1 conditional null allele were similar to those carrying two Brca1 null alleles [63]. Consistently with these findings, two of the BRCA1 VUS that, in our analysis, did not inhibit the BARD1 interaction, but lose the UbcH5a, and for which data on enzymatic properties are available (p.Ile15Thr, p.Leu28Pro) and lack ubiquitin ligase activity [37] (Table 3). However, it has to be mentioned that, in another study, the missense substitution p.Ile26Ala, abrogating E2 binding and thus ubiquitin-ligase activity, was found to prevent tumor formation to the same degree as wild-type Brca1 in three conditional genetic models. Furthermore, the DNA damage response remained intact with no changes in chromosome stability or sensitivity to genotoxic stress in mouse embryonic fibroblasts [64]. Given the discrepancy of the results from the two studies, the implication of BRCA1 E3 ligase activity in tumor suppression remains to be elucidated.
In the presence of p.Thr77Met, BRCA1 is able to bind both BARD1 and UbcH5a proteins. However, as reported by Morris and co-workers, the proteins lose the ubiquitin ligase activity [37]. A possible explanation is that a weak interaction between BRCA1 and UbcH5a is not sufficient to promote the ubiquitin ligase activity. Actually, the GFP-fragment reassembly approach can detect very weak and transient interactions and the fluorescence intensity observed does not necessarily correlate quantitatively to the interaction affinity of the proteins, which represents a limitation for this assay. Therefore, hypomorphic variants (i.e., variants weakening, but not completely disrupting, the interaction of BRCA1 and BRCA2 with partner binding proteins, which retains partial protein activity) are likely to produce a positive fluorescent signal and can be categorized as non-pathogenic with this approach. In effect, using the Y2H assay, Morris and co-workers showed a reduction in binding to the UbcH5a of BRCA1 carrying the p.Thr77Met variant. These observations suggest that the p.Thr77Met may be a hypomorphic variant associated with moderate-to-low risk of breast and ovarian cancers. Furthermore, this variant was observed in trans with a pathogenic variant in an individual without clinical signs of Fanconi anemia (FA) or FA-like disease, which is consistent with the suggestion that the p.Thr77Met is hypomorphic and is not associated with a high risk of cancer. Further analyses, using additional approaches such as the Y2H assay and the split biosensor assays measuring protein interaction kinetics (the tripartite split-GFP association assay [65]), are required to define the thresholds of binding activity that discriminate between pathogenic and neutral variants and assess potential intermediate effects.
A series of important considerations have to be applied to functional assays evaluating the BRCA1 and BRCA2 VUS [10,11]. In particular, a panel of different validated assays, representative of different protein functions, should be used, in order to avoid the misinterpretation of the clinical relevance of BRCA gene variants. We observed good correlation of results obtained from the GFP-reassembly screenings with those of other reported functional assays, including homology-directed repair (HDR), mouse embryo stem cells (mESCs) viability, and drug resistance assays (Tables 3 and 4). In fact, BRCA1 and BRCA2 VUS that were proficient in these assays, retain binding activity with UbCH5a and DSS1, respectively. All variants that were not proficient in other reported assays showed a loss of the binding activities. These included the BRCA2 p.Phe2562Leu, which is the only VUS that, in our assay, abolishes the ability of BRCA2 DBD to bind DSS1, which is reported to be deficient in HDR activity [48].
To date, the most accurate medium throughput functional approaches in evaluating BRCA1 and BRCA2 VUS are the yeast-cell based colony size phenotype (SCP) assay [40,66] and the HDR assay in mammalian cells [47,48], respectively (Tables 3 and 4). Compared to the GFP-reassembly screening, the SCP assay shows lower sensitivity (96%) and specificity (93%). Importantly, contrary to the SCP assay, the GFP-reassembly screening analyzing the effect of BRCA1 variants on UbcH5a binding correctly classifies the p.Cys61Gly variant as pathogenic (Table 3). While the GFP-reassembly assay and the HDR functional assay show the same level of accuracy, the latter is more technically and experimentally demanding and time consuming.
While this manuscript was in preparation, Findlay and co-workers published the results of a massive functional characterization of BRCA1 single nucleotide variants, (SNVs) based on a highly sensitive and specific saturation genome editing (SGE) approach [44]. Findlay et al. measured the effect on the survival of the near-haploid HAP1 cell line of almost 4,000 SNVs corresponding to 96.5% of all possible base-pair changes in 13 exons that encode the RING and BRCT domains. The assay was based on the observation that the maintenance of BRCA1-mediated HDR is necessary for the viability of HAP1 cells. Notably, we observed a 100% correlation between the functional classification based on SGE scores and the outcomes of our GFP-reassembly assay evaluating the BRCA1/UbCH5a interaction (Table 3). This strict consistency suggests that BRCA1/UbCH5a interaction, which mediates BRCA1 ligase activity, is essential for proficient HDR. In addition, since HDR deficiency is targeted by drugs exploiting synthetic lethality, such as Poly (ADP-ribose) polymerase 1 (PARP1) inhibitors [67], our assay could be used to predict a response to therapeutic treatments. SGE has demonstrated to be a viable and highly productive strategy for functionally classifying thousands of variants in clinically actionable genes such as BRCA1, including a large number of variants that, to date, have not been observed in humans. In fact, a major strength of this approach is represented by the possibility of examining variants located throughout the entire gene sequence, while, on the contrary, the GFP-reassembly assay may be applied only to gene regions coding for protein-interaction domains. However, it must be noted that the vast majority of pathogenic BRCA gene variants lie in these domains [6,51]. On the other hand, SGE analyses are limited by the extreme difficulty of generating all possible gene variants that may occur in patients. In particular, substitutions and insertions/deletions involve more than one nucleotide. For example, in our study, we could analyze three variants identified in families referred for BRCA gene testing, namely the BRCA1 c.20_29del9 (p.7-10delinsGln) and c.206_207delCCinsTG (p.Thr69Met) in frame deletions, and the c.107C>A (p.Ser36Tyr) SNV, that were not evaluated in the high-throughput study of Findlay and co-workers.
This represents an important and attractive advantage of the split-GFP reassembly in vitro method, which, together with its easiness of use and robustness, makes it suitable for routine analyses of naturally occurring genetic variants. Moreover, this assay could be applied to large-scale studies that are required to achieve robust estimates of the accuracy of functional approaches for VUS classification and clinical applicability.

Variant Selection
All examined variants were observed in high risk HBOC families fulfilling the ascertainment criteria for BRCA1/2 testing in use at the collaborating institutions, including the Fondazione IRCCS Istituto Nazionale dei Tumori (Milan, Italy) [68], the Catalan Institute of Oncologia (Hospitalet de Llobregat, Barcelona, Spain) [69], and the Rigshospitalet (Copenhagen, Denmark) [70].

Expression of Recombinant Peptides
E. coli BL21 (DE3) cells were transformed with pET11a-NfrGFP vector series or pMRBAD-CfrGFP vector series. Single colonies were picked and used to inoculate 20 mL of LB broth medium containing 100 µg/mL of ampicillin (pET11a-NfrGFP vector series, Addgene, Watertown, MA, USA) or 35 µg/mL of kanamycin (pMRBAD-CfrGFP vector series, Addgene, Watertown, MA, USA). The colonies were then incubated at 37 • C until the Optical Density measured at a wavelength of 600 nm (OD 600 ) reached 0.6. Then, induction of recombinant peptides was carried out with the addition of IPTG to a final concentration of 20 µM (pET11a-NfrGFP vector series) and 0.2% of L-arabinose (pMRBAD-CfrGFP vector series). After growing for 24 h at 30 • C followed by two days at room temperature (RT), the cells were harvested by centrifugation. Each pellet was resuspended in 1 mL of lysis buffer [50 mM Tris-HCl, 300 mM NaCl, 0.1% v/v Triton X-100, 100 µM EDTA pH8.0, 0.5 mg/mL lysozime, 20 mM Imidazole, protease inhibitors (0.5 mM PMSF, 0.4 µg/mL leupeptin, 0.5 µg/mL aprotinin), and 5 µg/mL DNase and RNase]. The resuspended cells were lysed by sonication. The cell debris were removed by centrifugation at 4 • C for 20 min at 13,000 rpm. The protein concentration of the whole cell extracts was determined by the Bradford method [71], using the Bio-Rad protein assay kit (Bio-Rad Laboratories, Hercules, CA, USA). Equal amounts of protein (20 µg) were subjected to 13% SDS-PAGE and visualized by Western blotting using an anti-GFP antibody (# 600-101-215, Rockland, Limerik, PA, USA).

In Silico Analyses
The Alamut Visual version 2.10 (Interactive Biosoftware, Rouen, France) was used to evaluate the possible effect of examined VUS on mRNA transcript splicing.

Conclusions
The concordance between genetic data and the functional assays analyzing the BRCA1/UbcH5a and the BRCA2/DSS1 interactions, observed for the validation panels of previously classified variants, strongly suggests that the approaches described in the present study are able to differentiate between VUS that either inactivate or have no effect on BRCA1 and BRCA2 functions and support their utility in the assessment of the clinical relevance of VUS. Caution is warranted when the interpretation of VUS is based only on results from an assay focusing on a single specific biochemical activity. Therefore, these assays should be used in combination with other evidence when assessing variant pathogenicity, and the efficacy will be further verified using larger validation panels.