Structure-Based Assessment of Missense Mutations in Human BRCA1: Implications for Breast and Ovarian Cancer Predisposition

Many germ-line mutations in the human BRCA1 gene are associated with inherited breast and ovarian cancers (1, 2). This information has allowed clinicians and genetic counselors to identify individuals at high risk for developing cancer. However, the disease association of over 350 missense mutations remains unclear, primarily because their relatively low frequency and ethnic specificity limit the usefulness of the population-based statistical approaches to identifying cancer-causing mutations. To address this problem, we use here the three-dimensional structure of the human BRCA1 BRCT domains to assess the transcriptional activation functions of BRCA1 mutants. Our study is made possible by the recently determined sequences (3, 4, 5, 6) and three-dimensional structures of the BRCA1 homologs (7, 8). In addition, we benefited from prior studies that attempted to rationalize and predict functional effects of mutations in various proteins (9, 10, 11, 12), including those of BRCA1 (13, 14).

BRCA1 is a nuclear protein that activates transcription and facilitates DNA damage repair (15, 16). The tandem BRCT domains at the COOH-terminus of BRCA1 are involved in several of its functions, including modulation of the activity of several transcription factors (15), binding to the RNA polymerase II holoenzyme (17), and activating transcription of a reporter gene when fused to a heterologous DNA-binding domain (18, 19). Importantly, cancer-associated mutations in the BRCT domains, but not benign polymorphisms, inactivate transcriptional activation and binding to RNA polymerase II (18, 19, 20, 21). These observations suggest that abolishing the transcriptional activation function of BRCA1 leads to tumor development and provides a genetic framework for characterization of BRCA1 BRCT variants.

The multiple sequence alignment (MSA) of orthologous BRCA1 BRCT domains from seven species, including Homo sapiens (GenBank accession number U14680), Pan troglodytes (AF207822), Mus musculus (U68174), Rattus norvegicus (AF036760), Gallus gallus (AF355273), Canis familiaris (U50709), and Xenopus laevis (AF416868), was obtained by using program ClustalW (22) and contains only one gapped position (Supplementary Fig. 1). According to PSI-BLAST (23), the latter six sequences are the only sequences in the nonredundant protein sequence database at National Center for Biotechnology Information that have between 30% and 90% sequence identity to the human BRCA1 BRCT domains (residues 1649–1859).

The multiple structure-based alignment of the native structures of the BRCT-like domains was obtained by the SALIGN command in MODELLER (Supplementary Fig. 2). It included the experimentally determined structures of the two human BRCA1 BRCT domains (Protein Data Bank code 1JNX; Refs. 8, 24), rat BRCA1 BRCT domains (1L0B; Ref. 7), human p53-binding protein (1KZY; Ref. 7), human DNA-ligase IIIα (1IMO; Ref. 25), and human XRCC1 protein (1CDZ; Ref. 13). Structure variability was defined by the root-mean-square deviation among the superposed Cα positions, as calculated by the COMPARE command of MODELLER. The purpose of these calculations was to gain insight into the variability of surface-exposed residues (left panel in Fig. 2). In conjunction with observed mutation clustering, these data may point to putative functional site(s) on the surface of BRCT repeats.

Comparative protein structure modeling by satisfaction of spatial restraints, implemented in the program MODELLER-6 (26), was used to produce a three-dimensional model for each of the 94 mutants. The crystallographic structure of the human wild-type BRCA1 BRCT domains was used as the template for modeling (8). The four residues missing in the crystallographic structure (1694 and 1817–1819) were modeled de novo (27). All of the models are available in the BRCA1 model set deposited in our ModBase database of comparative protein structure models (28).⁶

For the native structure of the human BRCT tandem repeat and each of the 94 mutant models, a number of sequence and structure features were calculated. These features were used in the classification tree in Fig. 3 (values for all 94 mutations are given in Supplementary Tables 1 and 2).

Accessible surface area of an amino acid residue was calculated by the program DSSP (29) and normalized by the maximum accessible surface area for the corresponding amino acid residue type. A residue was considered exposed if its accessible surface area was larger than 40Ų and if its relative accessible surface area was larger than 9% and buried otherwise. A mutation of a more exposed residue is less likely to change the structure and therefore its function.

The putative functional site was visually assigned to the surface groove that contains the deleterious mutants (Fig. 2); it consists of residue positions with accessible surface area >9%, as follows: 1654–1657, 1659, 1662, 1663, 1666, 1676, 1678, 1698–1702, 1773, 1774, 1811–1813, 1834, and 1836. A mutation in the functional site is more likely to abolish the function.

A residue isotropic temperature factor (B-factor) was the mean of its atomic B-factors as determined by X-ray crystallography of the native structure; it was also normalized by the mean and standard deviation (SD) of all residue B-factors in the protein chain. A mutation of a less rigid residue at a buried position is less likely to change the structure and therefore its function.

Neighborhood of a residue was defined by the residues that have at least one atom within 5Å of any atom of the central residue. Neighborhood B-factor of a residue was defined to be the mean value of the B-factors of the residues in the neighborhood; it was also normalized by the mean and SD of all neighborhood B-factors for the neighborhoods of the same size. A mutation of a buried residue in a less rigid neighborhood is less likely to change the structure and therefore its function.

Volume change was defined solely by the type of residue substitution (30). A large change in the volume of the amino acid residue type was considered destabilizing to the structure and therefore its function, especially when buried at a rigid position.

Charge change was defined solely by the type of residue substitution (30). Mutations corresponding to changes in charge at buried positions are more likely to change the structure and therefore its function.

Polarity change was defined solely by the type of residue substitution. Amino acid residue types were classified into three classes according to their hydrophobicity: 0 {LIFWCMVY}, 1 {PATGS}, 2 {HQRKNED} (30). Mutations corresponding to changes in polarity at buried positions are more likely to change the structure and therefore its function.

Mutation likelihood of a residue substitution at a given MSA position reflects the “background” substitution probabilities in the BLOSUM62 matrix as well as the actually observed substitutions at the given position in the MSA of the BRCA1 BRCT orthologs (Supplementary Fig. 1; Refs. 4, 5, 6, 9, 31). The scores are centered on zero, with negative values corresponding to less likely substitutions. Unlikely mutations are more likely to change the structure and function.

Sequence variability was defined for every column in the MSA of the BRCA1 BRCT orthologs (Supplementary Fig. 1) by Σ p ln p, where p is the observed relative frequency of residue type i (Fig. 2, right panel). Mutations at evolutionarily conserved positions are more likely to change the structure and function.

Secondary structure assignment was calculated by DSSP (29). A “helix breaker” mutation was defined as any mutation into the glycine or proline residues at a helical position; a “turn breaker” mutation was defined as any replacement of the glycine or proline residues at a turn position.

Visualization and analysis of the mutations were facilitated by the program DINO.⁷

Transcriptional assays for testing our predictions were performed as described previously (21). The wild-type BRCA1 sequence (residues 1396–1863) fused in frame to the LexA DNA-binding domain was used as the wild-type control and for introducing mutations M1775R, Y1853X, M1689R, and W1718C by site-directed mutagenesis. EGY48 yeast cells were cotransformed with the constructs and pRB1840, which contains a lacZ reporter gene under the control of a LexA operator, and tested in liquid β-galactosidase assays. For mammalian transcription assays, BRCA1 constructs were subcloned in frame to GAL4 DNA-binding domain in pCDNA3. GAL4 DNA-binding domain fusion constructs were cotransfected into human 293T cells with pG5Luc, which contains a firefly luciferase gene under the control of five GAL4 binding sites as a reporter for the assay. Transfections were normalized with an internal control phRG-TK (Promega), which contains a Renilla luciferase gene under a constitutive thymidine kinase-basal promoter.

We tabulated a total of 94 missense mutations in the human BRCA1 BRCT domains (Tables 1 and 2; Fig. 1). These mutations occurred naturally or were obtained via random mutagenesis. Transcriptional activation function of 37 of 94 mutants has been tested (Fig. 1, rows 1 and 2). Eight of the 37 mutants could be tentatively classified either as cancer-associated or benign following established genetic criteria (Ref. 32; Fig. 1, columns 1 and 2). The basis of each classification is outlined in Table 1.

For the eight mutations with intact transcriptional activation function, there is no evidence of cancer-association; the frequency in control population is equal or higher than in cancer cases, and there is absence of co-segregation in at least one family (Fig. 1, row 2; Table 1). Conversely, none of the 29 tested mutations without transcriptional activation function are thought to be benign, following the same criteria (Fig. 1, row 2; Table 1; Ref. 32). These observations indicate that tumor suppression by BRCA1 correlates with its ability to activate transcription.

We first rationalized the functional consequences of the 37 tested mutations in terms of their impact on folding and stability of the native structure and in terms of the integrity of binding sites of the BRCT domains. Next, we used this experience to develop generic rules for prediction of the functional impact of a missense mutation and applied these rules to the remaining 57 functionally uncharacterized mutations listed in Fig. 1 and Table 2. We then tested two of these predictions by functional assays and segregation analysis.

Missense mutations of BRCA1 are likely to affect transcriptional activation in either one of the following two ways: (a) when they are exposed to the solvent, they may substantially change the structure or chemical nature of functional sites that bind other molecules, or (b) when they are buried in the core, they may prevent folding of the BRCT domains into their native fold or, less likely, affect only the structure of functional sites.

The 37 mutations, including the 8 (Fig. 1, row 2) and the 29 mutations (Fig. 1, row 1) with and without transcriptional activation function, respectively, were mapped onto the crystallographic structure of the human BRCA1 BRCT domains (Ref. 8; Fig. 2). In addition to this structural mapping, we produced a three-dimensional protein structure model for each mutant and calculated a number of its sequence and structure features (“Materials and Methods”). We used the feature values and their changes between the wild type and the mutant to rationalize the observed effect on transcriptional activation (Table 1; Supplementary Table 1).

The markedly reduced transcriptional activation activity of the 29 mutations (Fig. 1, row 1) is readily explained by their likely disruption of the structure of the BRCT domain and/or its binding sites that are involved in transcriptional activation. Twenty-three of these 29 mutations occur at buried positions and involve amino acid residue types with dissimilar charge, hydrophobicity, size, or secondary structure propensity (Fig. 1, row 1, not underlined). Thus, they are expected to disrupt the structure and therefore its function. The remaining six mutations resulting in reduced transcriptional activation are exposed (Fig. 1, row 1, underlined). Among these six mutations, R1699W and R1699Q replace a conserved residue involved in a salt bridge between the BRCT repeats (8). Another one of the six mutations (G1743R) occurs at a position that is absolutely conserved among all available orthologs (Supplementary Fig. 1). For the remaining three exposed mutations that abolish transcription, we propose that they disrupt interactions with BRCA1 ligand(s). They are located in a groove formed by both BRCT repeats (L1657P and K1702E) and in the ridge that delimits the groove (E1660G). We suggest that this clustering of residues in a single patch on the surface of BRCA1 highlights a binding site with integrity that is necessary for transcription. Moreover, this putative binding site is relatively conserved in sequence and structure among the BRCA1 homologs (Fig. 2; “Materials and Methods”). These results are in agreement with in vitro binding studies that determined the binding site of BRCA1 to the RNA polymerase holoenzyme via RNA helicase A (17). Nonconservative substitutions of any amino acid residues that comprise the putative binding site are also likely to affect transcriptional activation. The existence of such a functional region indicates that the tandem BRCT repeats comprise one structural and functional unit, in accordance with previous conclusions (7).

The lack of impact of the eight mutations that do not disrupt transcriptional activation function (Fig. 1, row 2) can also be rationalized in structural terms. Three of them are exposed on the face away from the putative binding site proposed above, whereas the remaining five mutations are conservative changes in the core of the fold (Table 1; Fig. 2). Five mutations (G1706A, S1715N, F1761I, M1775V, and A1843T) in the total set of 37, with impact on the transcriptional activation function that is known, could not be readily explained. Possible reasons for the inability to explain these mutants are discussed in detail in Supplementary Table 3.

Because it was straightforward to use the three-dimensional structure of the BRCA1 domains to rationalize the known functional impact of 32 of the 37 mutations, we derived a set of rules for predicting whether a given mutation inactivates transcriptional activation and therefore correlates with predisposition to cancer. These rules rely on the sequence and structural features that proved informative in explaining the presence or absence of a functional impact of the tested mutations. For example, structural neighborhood around a mutated residue can be rigid, and a mutation can introduce a large change (over 30A³) in the side chain volume. Such a mutation is likely to destabilize or distort local conformation, thus having an effect on the function of the protein. In contrast, a flexible structural neighborhood is expected to accommodate such a substitution without major perturbations in the structure. Moreover, our rules allow for the possibility that very large changes in the side chain volume (over 90A³) can have a significant effect even in a flexible neighborhood. The rules were encoded hierarchically in a “classification tree” scheme (Fig. 3) and are implemented on the web.⁸

Using the classification tree, we predicted the functional consequences of the 57 missense mutations in the BRCT domains that presently lack definitive information about cancer association (Breast Cancer Information Core Database;⁹ Table 2; Supplementary Table 2). Our prediction algorithm assigned functional impact to 32 mutants, whereas 25 were predicted to be benign.

To validate our predictions, we tested prospectively two mutants (Fig. 4, A and B) for which segregation data were available. The W1718C mutation was found in a family with three generations affected where women were diagnosed with either early-onset (<40-years-old) breast cancer or ovarian cancer. It was present in four of four women with cancer and was absent in three of three at-risk women without cancer.¹⁰ The M1689R mutation also segregated with disease in a large family with several generations affected (Fig. 4,C). Both M1689R and W1718C displayed the loss of function phenotype and segregated with disease in agreement with our predictions (Fig. 4).

To illustrate prediction by the classification tree, we used the mutation M1689R that occurs at a buried position in a relatively rigid neighborhood (Table 2, Supplementary Table 2). Mutations at buried and rigid locations are tolerable only if changes in the residue volume, charge, and polarity are sufficiently small; a volume change of more than one methyl group (30ų) cannot be accepted. However, the arginine volume is only 18ų larger than that of methionine; therefore, the condition on the maximal allowed volume change is satisfied. At the next checkpoint, the side-chain charge is considered. Because M1689R introduces a positive charge at a buried position, the mutation is considered destabilizing, resulting into the prediction of a loss of function.

The tumor suppressor function of BRCA1 is likely to depend, at least partially, on its transcriptional activity, as suggested previously (18, 19, 20, 21) and supported by the mutation set in Table 1. Therefore, the ability to predict which mutations do and do not abolish the transcriptional activation function of BRCA1 can be useful in predicting their cancer association. Our analysis, based primarily on the considerations of protein three-dimensional structure, leads to a prediction of cancer association of the mutations that cannot be easily characterized by other approaches, such as population-based statistical genetics. Although all but two of the present predictions are unverified and thus cannot be used as a sole tool for risk assessment and genetic counseling, this structure-based approach may be helpful in an integrated effort, to identify mutations that predispose an individual to breast and ovarian cancers.

We are grateful to Drs. Andras Fiser and Anthony Brown for helpful discussions and to Dr. Ursula Pieper for help with the server setup.