Patients seeking certainty in genetic tests, such as tests for inherited susceptibility to cancer, often receive a perplexing result. Many people learn they carry a “variant of unknown significance” of a disease-linked gene. Such variants might—or equally might not—increase disease risk.
In the latest issue of GENETICS, Starita et al. characterized nearly 2000 variants of the breast cancer-associated gene BRCA1, demonstrating the potential of a new approach for sorting out which variants are harmful and which are harmless.
Because genetic tests increasingly use more comprehensive multi-gene and whole-genome sequencing methods, it’s becoming more common for patients to learn they carry a variant of unknown significance. For example, a 2014 study showed 42% of breast cancer patients who received results from a 25-gene hereditary cancer genetic test carried a variant of unknown significance in one of the scanned genes.
“There’s not much you can do with this information, except worry,” says lead author Lea Starita of the University of Washington. “We hope to reduce some of the uncertainty by driving forward technologies for efficient functional testing of gene variants.”
To understand more about variants of unknown significance, the authors are exploring methods to assay protein function in massively parallel formats. The approach—called deep mutational scanning—relies on relatively simple assays that can be scaled up to cover all possible missense substitutions in a given gene. The results could help clinicians to prioritize variants for more detailed studies, or to provide preliminary classifications of newly discovered variants.
The team used the BRCA1 gene as a test case for their approach because more is known about the impact of sequence variation in BRCA1 than many other genes associated with disease.
Among other jobs, the BRCA1 protein regulates homology-directed DNA repair. People who carry a known pathogenic BRCA1 variant have a higher risk of breast and ovarian cancers because the faulty protein fails to properly regulate DNA repair and allows cancer-causing mutations to accumulate.
But not all BRCA1 variants are pathogenic. It’s difficult to tell whether a gene variant confers a higher disease risk until enough people with the variant have been identified to allow rigorous statistical analysis of their disease rates. For extremely rare or unique variants, such studies might never be possible.
In the new research, the team combined data from two different functional tests of a key part of the BRCA1 protein called the RING domain. This domain binds to another RING domain in the protein BARD1, and together the two domains form a ubiquitin ligase. BRCA1 variants that cannot heterodimerize in this way lose their tumor suppressing ability, and around 58% of known pathogenic BRCA1 variants affect this domain.
One of the tests measured BRCA1 RING domain ubiquitin ligase activity using a type of phage display assay. The second test measured heterodimerization of the BRCA1 RING domain with the BARD1 RING domain using a yeast two-hybrid assay. Overall, the data from both assays were consistent with previous studies of RING domain function.
Next, the team evaluated how well this high-throughput data matched results from the “gold standard” of BRCA1 functional assays, homology-directed repair rescue. This more labor-intensive assay quantifies the ability of full-length BRCA1 variants to repair double-strand DNA breaks in a cell culture system. The homology-directed repair assay is also the measure that best correlates with disease risk in patients.
The researchers assayed homology-directed repair for 28 BRCA1 variants and added an additional 17 results curated from the literature. Using models built from the deep mutational scan data, they were able to predict results in these homology-directed repair assays. These predictions were most reliable when based on data from both the ubiquitin ligase and BARD1 binding assays. Crucially, these predictions were substantially more reliable than those made by widely-used computational methods that are currently used in genomic studies to predict the severity of mutations.
Starita cautions that although the results show remarkable promise, the data are not yet ready for use in the clinic. Clinicians can’t use the data in isolation to make conclusions about a variant, says Starita. Their model is better than computational methods, but it is still not perfect.
The team is working on related large-scale approaches for other genes. For example, many genetic variants linked to autism are located in genes related to chromatin remodeling. The researchers are developing massively-parallel chromatin remodeling assays that could be used to predict the severity of many autism-associated variants.
They are also using new genome editing technologies like CRISPR to create thousands of genetic variants directly in the genomes of cultured cells. These technologies would allow many other biochemical assays—such as protein-protein interaction, enzyme catalysis, and protein stability assays—to be performed at enormous scales, even for relatively poorly characterized genes.
As genetic testing becomes both cheaper and more comprehensive, Starita says we will need a variety of approaches to translate the deluge of genetic data into practical information on individual health risks. Deep mutational scans are one tool to help meet this urgent need.
Lea M. Starita, David L. Young, Muhtadi Islam, Jacob O. Kitzman, Justin Gullingsrud, Ronald J. Hause, Douglas M. Fowler, Jeffrey D. Parvin, Jay Shendure, and Stanley Fields (2015). Massively Parallel Functional Analysis of BRCA1 RING Domain Variants. Genetics 200(2):413-422 doi: 10.1534/genetics.115.175802