William Check, PhD
If it’s true that a good trial lawyer never asks a question without first knowing the answer, then molecular geneticists would make bad trial lawyers. Several hundred times each day across North America a molecular genetics laboratory sets up a clinical sample for whole-gene sequencing without knowing beforehand what the outcome will be. Of course, no laboratory knows beforehand whether any sample will be positive or negative. However, sequencing entails a whole new dimension of uncertainty. Like a traditional clinical laboratory test, a sequence can be negative—no mutation or a benign polymorphism; or positive—a mutation known to be associated with the hereditary condition in question. But, like a lawyer questioning a witness without benefit of a deposition, sequencing sometimes pops up a wild card—a mutation whose possible association to the hereditary condition in question is a complete unknown.
Vicky Pratt, PhD, FACMG, director of molecular genetics at Quest Diagnostics-Nichols Institute in Chantilly, Va., calls such variations “mutations of unknown clinical significance.” Dr. Pratt co-organized a workshop on this topic during last year’s meeting of the Association for Molecular Pathology. The session was called “Missense Mutations Detected During Sequencing” because most ambiguous or uncertain variants are missense mutations; frameshifts or stop codons (truncating mutations) would most likely be deleterious. “The job of the laboratorian is to try to figure out the clinical impact of that change,” Dr. Pratt told CAP TODAY. “It is hard for a laboratory to make a good interpretation in these situations and hard for clinicians to understand these reports and to know what to do next.” The workshop was designed to present the thoughts of laboratorians who have experience sequencing a large number of genes or many samples of a few genes. “We asked them to tell us how they resolve these unknown mutations,” she says.
Dr. Pratt predicts that, as more knowledge is gathered about uncertain variants, the problem will resolve to some degree. But with large complex genes like BRCA1 and BRCA2, mutations of unknown clinical significance will continue to be discovered. More important, Dr. Pratt and other experts agree that an increasing number of genes will be sequenced to aid clinical diagnosis. So the conundrum posed by mutations of unknown clinical significance won’t go away, and it’s likely to grow.
“We are increasingly moving into the realm of whole-gene analysis rather than detection of a limited panel of mutations,” Elaine Lyon, PhD, told CAP TODAY. “We can detect any mutation and any variant, and many times we find a variant that has not been described before and for which no clinical studies have been done,” says Dr. Lyon, who is assistant professor of clinical pathology at the University of Utah and medical director of molecular genetics at ARUP Laboratories, Salt Lake City. As a result, “The clinician gets a report saying that we found something that we are calling an uncertain variant, which leaves the patient maybe with more uncertainty than they started with.”
“I think this problem will only grow,” says Wayne Grody, MD, PhD, professor in the Divisions of Molecular Pathology and Medical Genetics in the Departments of Pathology and Laboratory Medicine, Pediatrics, and Human Genetics, UCLA School of Medicine. “As sequencing technology gets cheaper and easier, we will consider sequencing more genes and even sequencing the whole genome on some patients,” Dr. Grody says. Myriad Genetics’ work with the BRCA1 and BRCA2 genes, the longest and most extensive experience with whole-gene sequencing for clinical purposes, foreshadows what can be expected, he says. In the BRCA genes, uncertain variants can arise at virtually any position. “I don’t think the BRCA genes are more mutable than any other genes,” Dr. Grody says. “The fact that Myriad has found thousands of variants is a foreboding that we will find the same thing in any and every gene that we subject to sequencing.”
Sequencing for clinical investigation of hereditary diseases, most often a hereditary predisposition to cancer, is a relatively new practice. Myriad Genetics has been sequencing BRCA1 and BRCA2 genes for about eight years. Dr. Lyon notes that sequencing of the beta globin gene for structural anomalies that produce hemoglobinopathies (sickle cell disease) and altered amounts of beta-globin transcripts (thalassemias) has been done clinically for about five years. However, she says, “The beta globin gene is so well studied that we have yet to find a mutation that hasn’t already been characterized.” More recent additions to the sequencing menu are the CFTR gene in cystic fibrosis and several genes that give rise to hereditary colon cancers, which are as problematic as the BRCA genes.
Interpreting a mutation of unknown clinical significance is difficult. To make the situation even more stressful, medical geneticists know that major clinical actions can ride on their decisions. In an asymptomatic woman with a family history of breast or ovarian cancer, for instance, finding a mutation and calling it deleterious can have “serious life impact,” Dr. Pratt says. A deleterious mutation in one of the BRCA genes can lead to prophylactic bilateral mastectomy or oophorectomy.
Brian Ward, PhD, is senior vice president at Myriad Genetic Laboratories, which sequences both BRCA genes and several predisposition genes for colon cancer. He lists many possible implications of calling a mutation deleterious. First is a drastic rise in the individual’s lifetime risk of cancer. “For almost all of these cancer predisposition genes,” he says, “if you are asymptomatic and get tested because of a family history, finding a deleterious mutation increases your risk from the population risk of 10 percent to a lifetime risk of 80 percent [breast cancer] to almost 100 percent [colorectal cancer].” For ovarian cancer, the risk goes from about one percent to as high as 30 percent–40 percent. (All of this information is at Myriad Genetics’ site: www.myriadtests.com/inherited.htm.)
Calling a mutation deleterious also has implications for the person’s family. “If an affected individual carries a deleterious mutation, 50 percent of that person’s first-degree relatives—children, brothers, and sisters—are at risk of the same mutation,” Dr. Ward says. Conversely, if an asymptomatic person with a family history is found not to have the responsible mutation, that person’s risk goes down to almost the general population risk.
In the presence of a deleterious mutation, lifestyle changes may be recommended, especially increased surveillance. A chemopreventive drug may be prescribed: Oral contraceptive pills reduce risk for women at risk of breast or ovarian cancer. For those who already have a diagnosis of cancer, more aggressive surgery may be done. At the most extreme end are prophylactic surgeries.
Because of these extensive consequences, Dr. Ward says, “Any family in which cancer appears at an early age, especially under age 50 for breast and colon cancer, should be considered a family that should be tested [for a predisposition gene].” Because of the low prevalence of ovarian cancer, gene testing is indicated if anyone in the family has it at any age. “One of the greatest disservices we geneticists have done is to create a complicated algorithm for clinical testing,” Dr. Ward says. “The challenge we are now facing in human genetics is to simplify our message to increase access to these tests.”
When a patient is tested and an uncertain variant appears, laboratorians shoulder the responsibility for interpreting the variant—and for the consequences of their interpretation. “If we don’t do it, who will?” Dr. Grody says. “We can’t just report out the presence of a variant and leave the clinician to make sense of it. We have to make some attempt at least to couch the report in the context of the best information that we have for the clinician who has to make the decisions. We need to explain what we know and what we don’t know.” It’s not only professional ethics but also existing guidelines that mandate this effort. “This matter of interpreting missense variants previously unreported has been dealt with in both CAP and ACMG guidelines,” Dr. Grody says, citing CAP checklist item MOL. 34928 as an example.
At the AMP workshop, Dr. Grody provided a general outline for how to classify novel variants. A first step would be to check published genetic databases, such as Genbank, to examine the impact of that particular amino acid substitution. Another molecular clue would be the position of the affected residue in the 3-D protein structure. Most informative would be to study additional family members to see if the mutation tracks with the phenotype in multiple affected and unaffected individuals. (Any report concerning an uncertain variant should include a recommendation to study additional family members.) A more extensive step would be a general population study, checking 300 people to see if the mutation is randomly distributed, in which case it is most likely a benign polymorphism. In vitro functional studies of the protein may also be helpful.
Specific tools for interpreting uncertain variants were described by workshop
co-organizer and moderator William E. Highsmith, Jr., PhD, co-director of the
Molecular Genetics Laboratory in the Department of Laboratory Medicine and
Pathology at Mayo Clinic. One tool is the Grantham score, which evalutes differences
in amino acid side chain composition between variant proteins. Dr. Highsmith
cited a published study of a predisposition gene for colon cancer that used
this method (Eliason K, et al. J
Med Genet. 2005;42:95–96).
He also referred to two online databases that evaluate variant proteins based
on evolutionary and structural principles, SIFT (Sort Intolerant from Tolerant;
and PolyPhen (Polymorphism Phenotype;
SIFT is a product of the Fred Hutchinson Cancer Research Center (Ng PC, Henikoff S.
Nucleic Acids Res. 2003;31:3812– 3814). Each of these programs employs
an algorithm based on phylogenetic alignment to calculate the probability that
a substitution is deleterious.
One group has evaluated the accuracy of SIFT and PolyPhen using two extensively
studied genes, beta globin and glucose-6-phosphate dehydrogenase (Tchernitchko
D, et al. Clin Chem. 2004;50:1974–1978). Sensitivity of the two
programs for the two genes ranged from 74 percent to 98 percent and specificity
from 47 percent to 89 percent. For uncertain variants, Dr. Highsmith said, “This
is about as good as you can get online.”
Unfortunately, this level of accuracy is not good enough for clinical work. “We don’t do reports based on [SIFT and PolyPhen],” Dr. Lyon says. “We use them for internal purposes, but for my reports to clinicians I need to be more convinced.” Arlene Buller, PhD, associate scientific director of molecular genetics at Quest Diagnostics, San Juan Capistrano, Calif., expresses a similar view. “My personal experience with PolyPhen is that it is much easier to tell when something is benign,” she says. “It is more difficult for the model to predict that a mutation is deleterious.” Dr. Buller finds this true across all models because they are constructed to be conservative. Both of these laboratories are developing their own algorithms for evaluating uncertain variants.
Several panelists described in some detail how they handle uncertain variants. At Myriad Genetic Laboratories, whole-gene sequencing is done for a number of genes that predispose to cancer: BRCA1 and BRCA2 (Myriad holds a patent on this test so it is the only laboratory to sequence these genes); three genes that predispose to colon cancer (MLH1, MSH2, and MSH6); a gene for adenomatous polyposis (APC and MutY, also called MYH); and one for melanoma (p16). (To see who performs whole-gene sequencing, as well as other tests, for any hereditary condition, consult the laboratory directory at the Web site GeneTests [www.genetests.org], which is managed by Roberta A. Pagon, MD, professor of pediatrics at the University of Washington.)
Dr. Ward says that sequencing work at Myriad turns up about 40 new mutations per week, which are worked up through a standard algorithm. If a mutation is truncating, which can be determined from basic molecular biology, it is classified as deleterious. The next step is to review the published literature. If insufficient literature information is available, they ask whether the mutation deleteriously affects a splice site, which can also be determined from basic principles. If it does, it is classified as suspected deleterious (functionally equivalent to deleterious). If it does not affect a splice site and is a missense mutation, it is either suspected deleterious, if the mutation is in the conserved area of the protein’s functional domain, or uncertain variant.
Decisions about uncertain variants are made through a process similar to that outlined by Dr. Grody. Information is sought about the occurrence of the mutation in control groups and whether it tracks with the disease in multiple families. Myriad’s database is consulted to see if the uncertain variant co-occurs with deleterious mutations and in what phase. If the phase is trans—the uncertain variant and deleterious mutation are on opposite chromosomes—it is called benign. “For the most part, you cannot have two deleterious mutations in the same gene on two different chromosomes in the same individual,” Dr. Ward says. Myriad also uses statistical analysis, family segregation data, information on species conservation of amino acids, and knowledge about the functional domains of the gene to assist in classifying mutations.
Myriad’s goal is to decrease the number of uncertain variants. According to Dr. Ward, over eight years, they have reclassified more than 2,000 patients from “uncertain variant” to “polymorphism.” Over the same period, the rate of reporting “uncertain variant” has dropped by about two-thirds, from more than 20 percent to about seven percent among Caucasians and from almost 30 percent to the low teens in African-Americans.
At ARUP Laboratories, Dr. Lyon oversees sequencing of the beta globin gene and the CFTR gene in cases in which a patient has symptoms of cystic fibrosis and a positive sweat chloride test but two mutations are not found on the standard 32-mutation panel. In addition, they sequence two genes that are responsible for hereditary hemorrhagic telangiectasia, or HHT, an autosomal dominant condition for which the University of Utah has a clinic and which Dr. Lyon discussed as a model of their approach.
While not common, HHT is “not extremely rare,” Dr. Lyon says, with a frequency of one in 8,000 to 10,000 people. “We are starting to think it may be underdiagnosed,” she says, “since some people only get mild symptoms such as nosebleeds.” HHT can be more serious if the patient has arteriovenous malformation in the lung, liver, or brain.
In one family, sequencing detected an R479G mutation. Pedigree analysis gave a probability of 444:1 in favor of causality. In another family, sequencing of the genes in four family members gave a likelihood of 19:1 in favor of the uncertain variant being causal. The discrepancy between these two estimates shows how probability can be affected by such factors as the number of family members available for sequencing and the distance between the proband and affected family members (more-distant relatives provide more useful information). Additional family members are being collected for testing in each kinship.
Further evidence arises from comparison to sequence variations in the same gene in other species (called orthologs) and in different genes with similar functions in humans (called paralogs). Such comparisons can tell whether a site is evolutionarily conserved. Analysis of control populations is also revealing. The mutation found in the second family was absent in 200 control human chromosomes.
Dr. Lyon and colleagues are working with a quantitative model that allows
combining all these measures and that was developed initially for breast cancer
by genetic epidemiologist David Goldgar, PhD, of the International Agency for
Research in Cancer (Goldgar DE, et al. Am
J Hum Genet. 2004;75:535–544).
Dr. Goldgar’s model yields a statistical estimate that a given mutation is
associated with a hereditary condition. “A lot of papers say, ‘We tested other
family members and the mutation tracked with the disease,’” Dr. Lyon says. “But
that doesn’t tell us how many family members were tested and therefore how
strong the evidence of association is.” Dr. Goldgar’s model remedies those
problems. In research studies, Dr. Lyon notes, the data must reach a statistical
probability of 1,000:1 to show linkage for a marker. “Can we use this research
model to provide clinical probabilities?” she asks. “And, if we can, should
we lower the required statistical likelihood from 1,000:1 to come up with a
number that is convincing for clinical purposes?” She and her colleagues are
addressing these questions now.
Dr. Lyon is a strong believer in providing quantitative estimates on her sequencing reports. She spoke with some of the clinicians she works with and found that only a minority want a statistical likelihood. Nevertheless, she says, “We should put a number on our reports, even if most clinicians say they don’t want numbers.” Physicians who are comfortable with quantitative probabilities will use them. And, Dr. Lyon adds, the numbers serve a further purpose: Family members may be tested in another laboratory, which will be able to interpret the statistics and benefit from them.
At Quest Diagnostics in San Juan Capistrano, Dr. Buller’s laboratory performs sequencing for the CFTR gene, beta globin, and predisposition genes for colon cancer (MLH1, MSH2, and MSH6) and Rett syndrome (MECP2, MEN2), RET protooncogene, and a few other rarer disorders. A group at Quest Diagnostics, led by Steve Potts, PhD, and Matthew McGinness, PhD, is working on a qualitative and quantitative model, which the group calls MIST, for Mutation Inference Scoring Tool.
For qualitative evaluation, MIST uses hand-curated sequence alignments and structural models to evaluate novel missense mutations. For the CFTR gene, for instance, alignment is based on 31 orthologs and 60 paralogs, along with more than 100 literature references. Quantitatively, MIST gives a deleterious/benign score for missense mutations based on the evolutionary conservation of amino acid properties and protein structure.
Dr. Buller and her coworkers are now comparing MIST to SIFT and PolyPhen for their ability to score missense mutations as deleterious or benign. “We often find something that MIST calls deleterious and PolyPhen or SIFT calls benign and vice versa,” Dr. Buller says. Adjudicating these disagreements should prove informative. “We are now doing our own validation,” she says, “after which we would like to make MIST accessible to physicians on our Web site.”
In his main UCLA laboratory, Dr. Grody does only targeted mutation testing. At another UCLA laboratory he runs, the Orphan Disease Testing Center, he does sequencing for rare disorders such as congenital adrenal hypoplasia and some mitochondrial diseases. He emphasizes the need for collaboration in gathering data that make adjudication of uncertain variants possible. “When you get a previously undescribed variant,” he says, “it is very difficult or impossible for any one laboratory to determine whether it is a pathological change or a polymorphism. It takes many patient studies, both affected and unaffected, to make that determination. It is hard to do even with mutations that we are pretty sure are deleterious,” Dr. Grody says. He notes that making such a determination is analogous to clinical validation of an entire test. To facilitate interpretation of novel variants in the BRCA genes, an international consortium called the Breast Cancer Information Core acts as a repository for information from laboratories. Myriad is a major contributor to this database.
Dr. Grody subscribes to Dr. Lyon’s view that sequencing reports should contain numbers, even though most clinicians and patients don’t like them. “Unfortunately, in genetics it is hard to get away from numbers,” he says. “Many of our results confer a risk. That’s certainly true with uncertain variants.”
What’s the take-home message of these experiences of laboratories that perform sequencing? Says Dr. Pratt, “My takeaway message would be that if I was planning to get into sequencing on a routine basis I would conclude it is still a very complex type of testing. There are online tools to help you get answers, but those tools are imperfect.” Expertise, knowledge, and judgment are still required, as well as a penchant for a lot of hard work.
“Even if the people who attended the workshop are not doing whole gene sequencing right now,” Dr. Lyon says, “they may be considering doing it in the future. Whole-gene sequencing is one area of growth in molecular genetics. It allows us to test for genetic diseases that we would otherwise not be able to do, where there is not a set of common mutations on a panel but every family has its own private mutation.”
Because of this advantage, Dr. Lyon believes that whole-gene sequencing will expand and the problem of uncertain variants will expand along with it. “We need to get this information to physicians,” she says, “so they can understand how prevalent this problem is and they can properly explain it to families.”
William Check is a medical writer in Wilmette, Ill.