Summary of BARD1 Mutations and Precise Estimation of Breast and Ovarian Cancer Risks Associated with the Mutations

Over the last two decades, numerous BARD1 mutations/pathogenic variants (PVs) have been found in patients with breast cancer (BC) and ovarian cancer (OC). However, their role in BC and OC susceptibility remains controversial, and strong evidence-based guidelines for carriers are not yet available. Herein, we present a comprehensive catalog of BARD1 PVs identified in large cumulative cohorts of ~48,700 BC and ~20,800 OC cases (retrieved from 123 studies examining the whole coding sequence of BARD1). Using these resources, we compared the frequency of BARD1 PVs in the cases and ~134,100 controls from the gnomAD database and estimated the effect of the BARD1 PVs on BC and OC risks. The analysis revealed that BARD1 is a BC moderate-risk gene (odds ratio (OR) = 2.90, 95% CIs:2.25–3.75, p < 0.0001) but not an OC risk gene (OR = 1.36, 95% CIs:0.87–2.11, p = 0.1733). In addition, the BARD1 mutational spectrum outlined in this study allowed us to determine recurrent PVs and evaluate the variant-specific risk for the most frequent PVs. In conclusion, these precise estimates improve the understanding of the role of BARD1 PVs in BC and OC predisposition and support the need for BARD1 diagnostic testing in BC patients.


Introduction
Inherited genetic factors are responsible for a substantial portion of breast cancer (BC) and ovarian cancer (OC) cases. Multigene panel testing has revealed that germline loss-of-function mutations/pathogenic variants (PVs) in various cancer-associated genes are identified in more than 10% and 20% of patients with BC and OC, respectively [1][2][3]. Attempts have been made to define the genetic/familial risk of BC/OC associated with these genes, and subsequently, management recommendations for carriers of PVs in some of the high-penetrance genes have been established [4,5]. Despite this, several genes in which PVs confer low-or moderate-penetrance effects still require more evidence and more convincing assessments of BC/OC risk to utilize them in recommendations for carriers. A group of genes with insufficient and/or conflicting data includes the BRCA1-associated RING domain 1 (BARD1) gene.
The BARD1 gene encodes a protein of 777 amino acids with several functional domains, including an N-terminal RING-finger domain, three ankyrin repeats (ANK), and two C-terminal BRCT domains. The protein shows structural homology with BRCA1 within the BRCT and RING-finger domains, and through the latter domains, the two proteins form a stable heterodimer. The BARD1-BRCA1 heterodimer has E3 ubiquitin ligase activity and acts in multiple cellular processes essential for maintaining genomic stability, including DNA double-strand break repair through homologous PVs are shown alongside the BARD1 coding sequence with the indicated exon structure and the protein functional domains. The size of a PV symbol (circle) is proportional to the number of PVs, and color indicates the type of PV. A) PVs detected in BC (above) and OC (below) cases. B) PVs reported in noncancer gnomAD controls. The total number of detected PVs and the total number of cases and controls tested for the variants are indicated in parentheses; note that the total number of tested subjects differs substantially between groups.
Although BARD1 PVs in BC and OC patients are distributed over the entire coding sequence, there are two regions of increased density of PVs, i.e., from exons 2 to ~230AA in exon 4, overlapping the RING-finger domain and from exon 5 to exon 10, overlapping the ANK repeat and BRCT I domains.
In the BC cohort, 10 of the PVs were identified in three or more cases (defined here as recurrent), with the highest occurrence of c.1690C>T (Q564*), c.1935_1954dup20 (E652Vfs*69), c.1652C>G (S551*), c.334C>T (R112*), and c.1921C>T (R641*) reported in 16,11,7,6, and 6 cases, respectively. R112* and R641* recurrent PVs were identified in both Caucasian and Asian populations, while the remaining PVs, including the most frequent, Q564* and E652Vfs*69, were identified only in the Caucasian population. None of the recurrent PVs were specific to the Asian population. Interestingly, 10 out of 16 Q564* PVs were reported by studies analyzing mainly patients with a family history of BC/OC. Only one recurrent PV, i.e., Q564*, was reported in the OC cohort (five patients). In population controls, an additional recurrent PV was observed (not identified in BC/OC cohorts), i.e., Figure 1. Maps of deleterious PVs in BARD1. PVs are shown alongside the BARD1 coding sequence with the indicated exon structure and the protein functional domains. The size of a PV symbol (circle) is proportional to the number of PVs, and color indicates the type of PV. (A) PVs detected in BC (above) and OC (below) cases. (B) PVs reported in noncancer gnomAD controls. The total number of detected PVs and the total number of cases and controls tested for the variants are indicated in parentheses; note that the total number of tested subjects differs substantially between groups.
Although BARD1 PVs in BC and OC patients are distributed over the entire coding sequence, there are two regions of increased density of PVs, i.e., from exons 2 to~230AA in exon 4, overlapping the RING-finger domain and from exon 5 to exon 10, overlapping the ANK repeat and BRCT I domains.
In the BC cohort, 10 of the PVs were identified in three or more cases (defined here as recurrent), with the highest occurrence of c.1690C>T (Q564*), c.1935_1954dup20 (E652Vfs*69), c.1652C>G (S551*), c.334C>T (R112*), and c.1921C>T (R641*) reported in 16, 11, 7, 6, and 6 cases, respectively. R112* and R641* recurrent PVs were identified in both Caucasian and Asian populations, while the remaining PVs, including the most frequent, Q564* and E652Vfs*69, were identified only in the Caucasian population. None of the recurrent PVs were specific to the Asian population. Interestingly, 10 out of 16 Q564* PVs were reported by studies analyzing mainly patients with a family history of BC/OC. Only one recurrent PV, i.e., Q564*, was reported in the OC cohort (five patients). In population controls, an additional recurrent PV was observed (not identified in BC/OC cohorts), i.e., c.55G>T (E19*). However, the PV was characteristic only for the Latino population, which represents a small proportion of the entire population of BC/OC cohorts.

Association of BARD1 Pathogenic Variants with Breast Cancer
The prevalence of BARD1 PVs was higher in BC patients (0.25%) than in population controls (0.09%), with a cumulative OR = 2.90 (95% CIs:2.25-3.75; p < 0.0001), classifying BARD1 as a moderate BC risk gene. The risk was slightly higher for familial BC patients (OR = 3.67; 95% CI:2.52-5.34; p < 0.0001), evaluated based on data from studies that mainly analyzed cases with a BC/OC family history. To check whether the mixed population affected the results, similar association analyses were performed separately for Caucasian and Asian populations. As shown in Figure 2, BC risk estimates for Caucasians and Asians do not differ substantially, and both are only slightly lower than the risk estimates for mixed populations (OR = 2.73; 95% CIs:1.94-3.86; p < 0.0001 for Caucasians and OR = 2.50; 95% CIs:1.43-4.35; p = 0.0012 for Asians). We repeated the calculations excluding the three large studies from the analysis, which reported results of multigene testing of samples tested in diagnostic companies [125,139,141], as some small fractions of these samples may represent unrecognized duplicates of patients analyzed in other studies. Excluding these studies did not change the risk estimates substantially (OR = 3.14; 95% CIs:2.41-4.09; p < 0.0001). Moreover, to estimate the potential bias associated with the use of older technologies, which may be less sensitive when detecting some variants, we repeated the analysis including only studies reporting variants detected using NGS; again, the risk estimates showed no substantial difference (OR = 2.83; 95% CIs:2.18-3.67; p < 0.0001).
For five recurrent PVs, identified in at least five cases that have also been reported in controls, we calculated variant-specific ORs. Surprisingly, for 3 of these PVs, i.e., R112*, S551*, and Q564*, the variant-specific BC risk turned out to be substantially higher (OR > 4) than the overall gene-specific risk. The variant-specific risks of the other two recurrent PVs were in the range of the gene-specific risk, although it must be noted that due to the low number of particular PVs, the statistical power of the variant-specific risk estimations is generally low, and for one of the variants did not reach the nominal significance threshold (p = 0.05).

Association of BARD1 Pathogenic Variants with Ovarian Cancer
The prevalence of BARD1 PVs was only slightly higher in OC patients (0.12%) than in population controls (0.09%), with a slight non-significantly increased cumulative OR = 1.36 (95% CIs:0.87-2.11; p = 0.1733). Therefore, the association was insufficient to classify BARD1 as a low-risk gene for OC. The lack of association of BARD1 PVs with OC was also supported by the analysis performed for extracted Caucasian and Asian populations, for which even lower ORs were obtained (OR = 1.22; 95% CIs:0.72-2.07; p = 0.4501 for Caucasians and OR = 0.77; 95% CIs:0.10-5.75; p = 0.8007 for Asians). The exclusion of the two largest studies reporting results from diagnostic companies and accounting for~50% of cases [18,34] and limiting the analysis only to the studies using NGS did not affect the risk estimates (OR = 1.40; 95% CIs:0.77-2.54; p = 0.2648 and OR = 1.36; 95% CIs:0.87-2.11; p = 0.1727, respectively). Surprisingly, the variant-specific risk estimated for the most frequent OC PV, Q564*, suggests that the PV may be an OC high-risk allele (OR = 6.45; 95% CIs:1.87-22.27; p = 0.0032). This result, however, must be considered carefully, as it may result from the unequal proportion of Caucasian and Asiatic patients in OC and control populations. Indeed, as this PV is characteristic of the European population [14], the recalculation of the risk estimate presented a lower, yet still quite high, OR (OR = 3.32; 95% CIs:0.89-12.36; p = 0.0738).
studies from the analysis, which reported results of multigene testing of samples tested in diagnostic companies [125,139,141], as some small fractions of these samples may represent unrecognized duplicates of patients analyzed in other studies. Excluding these studies did not change the risk estimates substantially (OR = 3.14; 95%CIs:2.41-4.09; p < 0.0001). Moreover, to estimate the potential bias associated with the use of older technologies, which may be less sensitive when detecting some variants, we repeated the analysis including only studies reporting variants detected using NGS; again, the risk estimates showed no substantial difference (OR = 2.83; 95%CIs:2.18-3.67; p < 0.0001).

Large Mutations in the BARD1 Gene
As the identification of large mutations (CNVs; deletions/duplications of single or several exons) in NGS data is a challenging task, CNV mutations were not reported in most of the selected studies. Therefore, we did not take them into account in the formal risk estimates in our study. Only five CNVs were identified in the BARD1 gene in the selected studies. Deletion of exon 1, deletion of exon 2, and deletion of the entire gene (exons 1-11) were identified in BC patients [26,81,125], while deletion of exons 8-11 and deletion of the entire gene were identified in OC patients [34]. Of note, no large deletions were reported in the BARD1 gene in gnomAD population controls (~10,800 cases tested), and only 1 deletion (exon 4-11) was identified among noncancer FLOSSIES controls (9884 cases tested). Comparing these numbers and considering the lack of CNV analysis in most included studies suggests that CNVs in BARD1 are more frequent in BC/OC patients than in the general population and, therefore, most likely also contribute to BC/OC risk. However, as CNVs documented in BARD1 account for a small fraction of all PVs in the gene, this observation is still insufficient to determine the real contribution of the BARD1 CNV mutations to the risk of BC/OC.

Discussion
We performed a pooled analysis of the BARD1 screening studies to catalog PVs identified in BC/OC patients over a 20-years period and to estimate a reliable BARD1-associated BC and OC risk. The association of BARD1 with BC/OC has not been convincingly and unequivocally determined, and existing data, including our own [14,143], are still insufficient to ensure the solid interpretation of identified BARD1 variants and make recommendations for carriers [4,5,144]. Defining the correct risk ratio for infrequently mutated genes, such as BARD1, whose PVs confer a low or moderate effect on phenotype, might be challenging; therefore, only large-scale studies can provide strong evidence.
In this study, we described a wide mutational spectrum of BARD1, encompassing 69 distinct PVs identified in 144 BC and OC patients. Generally, PVs were distributed over the whole BARD1 coding sequence, although two regions of increased mutation density were also observed, the first overlapping the RING-finger domain and the second overlapping the ANK repeat and BRCT I domains. When comparing the PVs distribution between BC/OC patients and population controls, no evident hotspot characteristic of BC/OC could be defined. However, several recurrent PVs (n ≥ 5), which have been noted in BC patients, had a lower prevalence rate in population controls. Considering the generally dispersed distribution of PVs in BARD1, sequencing of the entire protein-coding region of the gene is recommended for diagnostic purposes; however, the genotyping of the most frequent PV, i.e., Q564*, may also have practical utility in central European populations [14,15].
The collection of~48,700 BC and~20,800 OC cases in the pooled analysis allowed us to achieve higher statistical power and more precisely estimate the overall BC and OC risk attributed to BARD1 PVs than in previous individual studies. The prevalence of deleterious PVs in BARD1 in the BC group (0.25%) was significantly higher than in population controls (0.09%); however, the frequency of PVs in the OC group (0.12%) was similar to that observed in controls. The pooled analysis of all studies revealed that PVs in BARD1 were associated with a moderate BC risk (OR = 2.9), and only slightly lower ORs were obtained for European and Asian cases than population-matched controls.  [20], while our previous large-scale case-control study, based on genotyping of one recurrent PV, indicated an OR of 2.3 (p = 0.04; n = 13,935) [14]. The other studies did not confirm the association of BARD1 PVs with BC or reported that increases in mutation frequencies were not significant [17,122,145], most likely due to an insufficient number of analyzed cases. Interestingly, Weber-Lassalle et al. reported a much stronger association (OR = 5.4; p < 0.00001; n = 4469) in BC patients with a family history of BC and an even higher risk (OR = 12.0; p < 0.00001; n = 782) for BC patients diagnosed under 40 years of age, suggesting that BARD1 might be a risk gene for early-onset familial BC [15]. In our recent study, in which we performed a large-scale association analysis of central European founder PV, i.e., Q564*, we did not observe a higher risk for BC patients diagnosed under 40 years of age, however, the risk was elevated for triple-negative BC (OR = 3.6; p = 0.02; n = 1120) and bilateral BC (OR = 5.1; p = 0.02; n = 498), which are also indicators of hereditary BC [14]. Similar observations were reported in other studies [145][146][147], indicating that BARD1 is a risk gene for familial BC. Consistently with the above, we observed a slightly increased risk for BC cases with a BC/OC family history.
In this study, we identified several recurrent PVs (potentially founder PVs in particular populations) that have been difficult to recognize in individual studies, and subsequently, we aimed to determine the variant-specific BC risk for these PVs. Surprisingly, three out of the five most frequent recurrent PVs were associated with substantially higher BC risk (OR > 5) than the overall risk estimated for the entire gene and two other recurrent PVs, which resembles an elevated risk associated with individual variants in other BC risk genes, e.g., the c.7271T>G (V2424G) variant in ATM, c.1036C>T (R346C) in CHEK2, and c.3113G>A (W1038*) in PALB2 [148]. These results confirm the possibility of varied cancer risks associated with specific variants within the gene. Although the variant-specific risks estimated in our study may be affected by the relatively low statistical power of the tests (reflected in wide 95% CIs) and should be interpreted with caution, this is the first attempt to estimate the risk for distinct BARD1 PVs. Future studies are required to define the risk more precisely and for a larger set of distinct PVs.
Our results indicate that BARD1 PVs do not substantially affect OC risk, and a small increase in the OR was not statistically significant (OR ; however, this result should be interpreted with caution, as two out of four carriers were also positive for BRCA1 PVs [19]. Despite the obtained results, we cannot exclude some very small effect of the PVs on OC risk or the effect of some individual PVs, such as Q564*, which was shown to be associated with a substantial risk of OC. The findings reported in the study should be interpreted within the context of certain limitations. First, the heterogeneity of the studies (i.e., differences in the inclusion criteria for patients and methodological aspects) and the use of population controls from public databases (not perfectly matched for the geographic area) may affect the results. Second, the association analysis for distinct PVs was performed only for recurrent variants and was of relatively low statistical power. Third, we cannot exclude the presence, in the tested cohorts' rare cases, of duplicated patients or not reported family members that may increase the recurrence of some variants and have some impact on calculations. Finally, the overall BC and OC risks for BARD1 were estimated based on loss-of-function variants. The consequences of other variants (including missense variants, silent variants, in-frame deletions/insertions, and multiple non-protein-coding variants) are difficult and/or impossible to define at this time. It is worth mentioning that several of the BARD1 missense variants have been shown to cosegregate in families with cancer [47,58,78]. Several missense variants in the ANK domain and the BRCT domain were also found to be nonfunctional in the homology-directed repair (HDR) assay; however, most of the tested missense variants did not perturb the HDR function of BARD1 [149][150][151]. Additionally, an apparent synonymous variant, i.e., R659R, was shown to affect splicing and was suggested to affect cancer risk [143,152]. Due to difficulties in detecting CNV mutations, they are rarely reported in mutation screening studies and therefore were not included in the risk. However, although large CNV mutations are not common in BARD1 [143], it is unlikely that they do not contribute to BC risk and therefore should be considered in mutation testing. Additionally, it was shown that common variants in BARD1 that are not the subject of this study may also modify cancer risk, including lung cancer [153] and neuroblastoma [154].

Conclusions
In conclusion, in this study, we provide a comprehensive summary of BARD1 PVs, and we provide reliable evidence that BARD1 is a BC moderate-risk gene but not an OC risk gene. The practical implication of the results is the support that they provide for a substantively justified interpretation of diagnostic results, which could help to establish BC management for BARD1 PV carriers.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4425/11/7/798/s1, Figure S1. The flow diagram indicating the strategy and criteria used for the selection of articles for the analysis, Table S1. Characteristics of the studies included in the study, Table S2. The list of BARD1 mutations identified in patients with BC/OC, with their prevalence in BC/OC patients and population controls.
Funding: This work was supported by grants from the Polish National Science Centre (NCN 2015/17/B/NZ2/01182, and 2016/22/A/NZ2/00184). The funders reviewed the design of the grants but otherwise had no role in study design, data collection, data analysis, data interpretation, or the writing of this report.