Application of Embryonic Lethal or Other Obvious Phenotypes to Characterize the Clinical Significance of Genetic Variants Found in Trans with Known Deleterious Mutations

Clinical molecular genetic testing will continue to grow in importance for patient management. Genetic tests provide valuable information to confirm diagnoses, assess patients' risks of developing diseases, and select appropriate therapies (1, 2). A persistent problem has been characterizing the clinical significance of genetic variants that do not pose an obvious effect on gene function. In particular, missense changes are proving the most difficult to understand.

Hereditary breast/ovarian cancer (HBOC) provides an interesting case in point where in the United States over 100,000 patients from cancer families have received clinical genetic testing. HBOC risk is largely explained by mutations in two large tumor suppressor genes, BRCA1 and BRCA2. Because mutations in both of these genes are distributed throughout the loci, accepted clinical protocols involve screening their entire coding regions (3). A variety of approaches including theoretical translation to identify termination codons, biochemical assays to assess protein function and RNA splicing, and population and family studies to associate variants with disease, are all employed to define the clinical risk associated with genetic variants (4, 5). Although the specific criteria to classify genetic variants differ between laboratories, they basically can be classified as deleterious mutations, variants of uncertain clinical significance, and benign polymorphisms. Some laboratories further subdivide the uncertain variants to include two additional groups: favor polymorphism/unlikely to convey cancer risk and suspected deleterious/likely to convey cancer risk (6). To date, ∼43.5% of over 3547 genetic variants that have been described in BRCA1 and BRCA2 are reported with an uncertain clinical significance.¹

Strong evidence exists that biallelic mutations in BRCA1 result in an embryonic lethal phenotype. Initially, this result was shown in multiple transgenic mouse models (7, 8). Reports of patients with biallelic BRCA1 mutations have been subsequently disproved or placed in doubt and none have been confirmed despite extensive clinical testing (9, 10). Considering the absence of biallelic mutations in the large numbers of patients that have been screened, there is an overwhelming statistical probability (P = 0.002) that embryonic lethality of compound heterozygous mutation carriers extends to humans (11). Based upon this conclusion, genetic variants in BRCA1 that can be shown to reside in trans with known deleterious mutations are probably polymorphisms with no clinical significance or mutations with greatly reduced severity. Two factors make BRCA1 an ideal locus to apply this approach for classifying variants. Foremost is that whole gene screening has been done on >60,000 patients worldwide. The vast majority of these patients have been assayed by direct DNA sequencing which provides comprehensive genotype data. Second, BRCA1 alleles are described almost completely by 10 canonical haplotypes derived from 14 prevalent single nucleotide polymorphisms (SNPs) that are detected during most genetic tests (12, 13). The low level of complexity for BRCA1 haplotypes results from a genome region or “block” where recombination is suppressed (14, 15). Two of these haplotypes predominate and account for ∼78% of all alleles in the North American population undergoing genetic screening. Therefore, BRCA1 is particularly well suited to statistical approaches that can assign the phase of genetic variants. Obviously, the ability to phase variants will correlate with the identification of uncertain variants in trans with known deleterious mutations.

Nomenclature. The descriptions of genetic variants throughout the article comply with those listed in the Breast Cancer Information Core.¹

Data set. Results for genetic variants detected during clinical BRCA1 mutation screening of 55,630 patients by direct DNA sequencing were made fully anonymous. The vast majority of these patients represent unrelated individuals, because accepted clinical practice focuses on administering the full gene sequencing test only to the available family member most likely to carry a mutation. Other family members are subsequently tested only to confirm results seen in the proband and are therefore not included in the data set used in these analyses.

Assignment of haplotype pairs to specimens. A computer program was written to assign haplotype pairs based upon BRCA1 genotype data for 14 SNPs obtained during clinical testing. This was accomplished by the assignment of binary numbers to represent the genotype data and canonical haplotypes. Each of the 10 canonical haplotypes were assigned a 14-bit binary number, where bits represent the allele at the SNP positions (“0” for the consensus base and “1” for the nonconsensus base) listed across the gene 5′ to 3′ (Table 1). The 55 possible haplotype pairs that can be produced by the 10 canonical haplotypes were described as a pair of two 14-bit binary numbers that define the presence of the polymorphic bases, a logical “OR” operation, and if they were homozygous, a logical “AND” operation. In this way, all 55 haplotype pairs were described by a unique pair of 14-bit binary numbers. A corresponding pair of 14-bit binary numbers were derived from patient genotypes and matched to the reference list of 55 to make haplotype assignments.

Assignment of genetic variants to haplotypes. A table was created that linked haplotypes derived for specimens with other genetic variants that were detected during the clinical tests. The most obvious association was made when a variant was seen in a specimen that was homozygous for a haplotype. The most parsimonious assignment to the common haplotype was made for variants identified in multiple heterozygous contexts. If there was any ambiguity about which haplotype a variant is associated with no attempt was made to make a partial assignment. Obviously, as more data is available, the majority of recurrent variants should be able to be assigned to haplotypes.

The alleles of BRCA1 can be almost completely described by haplotype assignment employing 14 SNPs that reside within or near exons and are therefore detected during clinical genetic testing (12). A computer program was developed to automate the process of haplotype pair assignment using genotype data obtained during clinical testing (see Materials and Methods). By this method, 55,086 of 55,630 (99%) patient genotypes could be assigned haplotypes. The prevalence of haplotypes in this specimen set shows strong agreement with previous reports that describe North American and European patients (Fig. 1; refs. 12, 16). The remaining 1% of genotypes could not be explained as a pair of the 10 canonical haplotypes. Some explanations for these specimens include rare haplotypes, point mutations, or intragenic recombination. Notably, similar genotypes were shown to result from hemizygous loci due to intragenic deletions and likewise these specimens are probably highly enriched for these mutations (17, 18). The low complexity of common haplotypes found in BRCA1 facilitated this analysis, because all possible haplotype pairs could be uniquely defined. More complex situations could arise in other genes or if greater numbers of SNPs are used to define the haplotypes. In these conditions, augmenting this approach with statistical methods such as expectation maximization or with biochemical analysis like allele-specific PCR would resolve ambiguities.

Haplotype assignments were correlated with the additional genetic variants that were detected during clinical testing. Variants could easily be assigned to haplotypes when observed in a homozygous context. Variants observed in multiple heterozygous contexts were assigned to the common haplotype. By these methods, 820 from a total of 1,477 variants in the data set, excluding haplotype-tagging SNPs, SNPs used to define a SNP-based haplotype, were assigned to a single haplotype (Table 2).

Another 57 variants were associated to two or three haplotypes (Table 3). The majority of these variants (19) reside in sequence contexts suspected to be hypermutable, which supports the conclusion that they arose by multiple independent mutation events. Twenty-one of these variants are located within repetitive sequences that are prone to mutation by polymerase “slippage” during DNA replication (20). It has previously been reported that this effect is responsible for the finding that the most prevalent BRCA1 mutation, 187delAG, is found on two haplotypes (12, 21). Nine instances of base changes that could represent mutations due to 5-methylcytosine were observed. Cytosine methylation, a common DNA modification, occurs at CG dinucleotides. Methylcytosine induces mutation by spontaneous or enzyme-induced deamination to thymine (22). The genotype calls for several variants assigned to multiple haplotypes based on the single SNP genotype with less than five observations were confirmed by reviewing the original clinical data. No discrepancies with the initial results were observed.

Erroneous assignments of variants to multiple haplotypes could occur through technical failures in genotyping, where data derived from one allele was interpreted as representing both alleles. Two obvious scenarios where this could occur have a genetic basis. First, rare polymorphisms at primer annealing sites could induce preferential amplification, whereby only the base at the unaffected allele is observed during sequencing. Second, an intragenic deletion mutation that affects a SNP used to define the haplotypes would also result in a homozygous call for the base present on the unaffected allele. In both of these cases, the genotype for the affected SNP from one allele would be mistakenly associated with both alleles and an inaccurate haplotype assignment could result. In fact, this second scenario has been exploited to identify patients that carry deletion mutations. Multiple laboratories have shown BRCA1 deletion mutations in patients whose genotypes cannot be explained by a pair of canonical haplotypes (17, 22). Finally, in keeping with a conservative approach, no attempts were made to assign haplotypes for variants observed once or multiple times in the same heterozygous context.

Forty-one genetic variants of uncertain clinical significance were observed in trans with known deleterious mutations (Table 4). Twenty-one of these uncertain variants were observed in patients that were heterozygous for two different haplotypes and where the mutation and uncertain variant were known to reside on alternative alleles (Table 4, heterozygous haplotypes). The second group of uncertain variants were observed in a homozygous haplotype context but where both the mutation and uncertain variant had also been seen in the absence of each other in other patients (Table 4, homozygous haplotypes). Seven of the variants, I379M, M1008V, E1250K, L1564P, L246V, M1628T, and Y105C, were seen in trans with multiple deleterious mutations. Two of the uncertain variants, V772A and M1008V, were observed in trans with mutations in both the heterozygous and the homozygous contexts. These 41 variants of uncertain clinical significance have been observed in 1,150 patients thus far, 956 of them as the only variant of uncertain clinical significance reported. These results provide for an improved clinical interpretation for all patients that harbor these particular variants.

Clinical genetic testing for hereditary disease syndromes is increasingly employed for patient management. Technical advances permit highly sensitive testing even for large genes with dispersed mutations through comprehensive whole gene direct DNA sequencing. This paradigm is particularly evident for dominantly transmitted hereditary cancer syndromes that result from mutations in tumor suppressor genes (6, 23, 24). Unfortunately, practical methods do not exist to determine the clinical significance of many genetic variants within the clinical laboratory setting, which reduces the value of these results for some patients.

This work describes an approach that can be applied to genes where biallelic mutations are embryonic lethal or cause detectable phenotypes. BRCA1 provides an advantageous model system to evaluate this technique for several reasons. The dispersed nature of mutations throughout the BRCA1 locus dictates that the entire coding region of the gene must be screened during clinical tests and thereby captures genotypes for common polymorphisms that can be used to assign haplotypes. In addition, BRCA1 is described by a small number of prevalent haplotypes believed to result from the location of BRCA1 in a region of the genome that undergoes reduced recombination (15). This suggests that this approach will be useful in other disease genes with similar characteristics. The especially high prevalence of the two most common haplotypes in this data set assured that many genetic variants, although rare, could be assigned to a haplotype. Whereas this approach is limited to data sets large enough where rare mutations and uncharacterized variants will coincide, this also means that it can be applied to some of the most prescribed genetic tests. This same paradigm means that within the data set, the most common variants will prove to be informative and thus serve a larger number of patients.

The application of this method provides information regarding the clinical significance of genetic variants detected during clinical testing. These results support the recommendation that BRCA1 genetic variants detected in trans with known deleterious mutations, which currently reside within the “uncertain” group, should be assigned to the “benign polymorphism” classification if they were seen with more than one known deleterious mutation, show lack of cosegregation with disease, or if other data supports the reclassification. Otherwise, the variants will be assigned under “favor polymorphism/unlikely to convey cancer risk” until their status can be confirmed.

The assessment of the use of this technique will benefit from continued validation within BRCA1 and hopefully in other genes and by other laboratories. Similar results that bear on this question have recently been reported with the discovery that patients from the Fanconi anemia complementation group D1 carry biallelic mutations in BRCA2 (25). Notably, no patients homozygous for the Ashkenazi BRCA2 mutation 6174delT, which has a carrier frequency of about 1%, have been observed during clinical testing for HBOC (26, 27). In addition, patients of Ashkenazi heritage are not represented in the Fanconi D1 patients with homozygous 6174delT mutations (28). Taken as a whole, it could be hypothesized that some mutations in BRCA2 convey a embryonic lethal phenotype in a biallelic context and that Fanconi D1 patients carry at least one hypomorphic mutation (28). Because many HBOC families carry the BRCA2 mutations that were detected in Fanconi patients, it remains to be investigated to determine if some of these mutations also convey a reduced cancer risk.

Several methods to assess the clinical significance of missense genetic variants in BRCA1 have been published. Recently, analysis of evolutionary conservation of amino acid residues was applied to this problem (11). These authors conclude that residues are conserved across species due to functional significance and that genetic variants that alter conserved residues are likely to be mutations. This method is advantageous because it can be applied to the classification of both mutations and polymorphisms. However, chance conservation of residues across the species employed for comparison can lead to the misclassification of mutations by this approach. For example, P871L is one of the most common benign polymorphisms known in BRCA1. Because leucine is conserved at this position throughout the phylogeny, the highly divergent yet consensus proline residue could appear as a mutation by this technique. Additionally, mutations that affect RNA splicing, specifically those affecting exonic splice enhancer sequences, would prove problematic for this technique.

Another method determines cosegregation of genetic variants with cancer in HBOC families (29). This technique benefits from data that directly relates genotype with phenotype and the analysis can be expanded with additional families or extended pedigrees. In practice, this approach proves inapplicable, because clinical testing frequently involves individual patients and comprehensive pedigrees with mutation status cannot be achieved. This method could be modified to include information about penetrance where variants carried by early-onset cancer patients should correlate with deleterious mutations and variants carried by older cancer-free patients should correlate with benign polymorphisms.

Several different in vitro functional assays have also been developed. However, these have also proven limited, because they are highly artificial and, for example, assess different BRCA1 functional domains in isolation or employ systems comprising both yeast and human components (19, 30).

The results from two recent publications that employed alternatives to co-occurrence in trans to classify genetic variants of uncertain clinical significance in BRCA1 were compared with the results presented here (11, 31). Concordant conclusions were drawn for all of the overlapping variants except one (Table 4). This approach, using the co-occurrence of mutations in trans with uncertain variants, benefits from a solid premise that fully penetrant biallelic mutations are undoubtedly lethal. However, this method is only informative for variants that convey reduced risk and does not assist the identification of deleterious mutations. Furthermore, it is unclear the extent of cancer risk that can be conveyed by genetic variants before they prove lethal in a biallelic setting. It is conceivable that deleterious mutations exist that, whereas able to increase lifetime cancer risk, are not penetrant enough to trigger embryonic lethality. Due to this, variants of unknown significance are not reclassified directly to benign polymorphism based on this data without additional sources of confirmatory data (5). It seems likely that comprehensive approaches that formulate probabilistic models or a Bayesian approach to incorporate results from a variety of different methods, as has recently been proposed, could yield the most definitive information towards the classification of missense variants (32).

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.