Current genetic testing methods often fall short in diagnosing pediatric patients with developmental delay, intellectual disability, or congenital abnormalities. Sometimes, this is because certain genetic variants are technically very difficult to detect. Other times, tests reveal genetic mutations that clinicians simply don’t know how to interpret: it’s not clear whether they are pathogenic, so it’s impossible to say if they’re related to a disease or not.
Over the past several years, researchers have found that several rare conditions associated with these symptoms are caused by mutations in epigenetic genes, such as those encoding histone modifying enzymes or components of DNA methylation machinery. These result in wide-scale disruptions in methylation patterns across the genome, creating a distinct “episignature.”
Now, a team of Canadian and American scientists have developed a computational tool that can diagnose 14 rare, hereditary disorders based on a patient’s episignature. They report their findings today (March 28) in the American Journal of Human Genetics. The researchers hope the tool will prove useful in getting more clinical diagnoses for patients with such conditions.
“There are a lot of times in genetics where we get through the whole work-up, and we still have no answer for the family,” says Jill Fahrner, a physician and assistant professor of pediatrics and genetics at Johns Hopkins Medicine, who has discussed collaborations with the group but was not involved in the new research. “There’s probably many conditions that lead to disrupted epigenetics, and I think this tool might be one of the things that help us identify these.”
It’s powerful, if they have one of those diseases.—Andrew Sharp, Icahn School of Medicine at Mount Sinai
To develop the tool, the team collected peripheral blood DNA samples from 383 patients with a conclusive diagnosis of one of 14 neurodevelopmental or congenital disorders. These included, for instance, Kabuki, Williams, Sotos, and Nicolaides-Baraitser syndromes, all characterized by varying degrees of intellectual disability and certain physical features. The team focused on these conditions because they are known to be caused by defects in chromatin-regulating genes, explains Bekim Sadikovic, the senior author and a clinical molecular geneticist at Western University in London, Ontario.
He and his colleagues used a commercial methylation array to characterize the methylation pattern across the patients’ genomes. For each syndrome, they identified the top 1,000 loci with the most significant methylation differences compared to a healthy control group. Using data from 75 percent of the patients, they trained a machine learning algorithm—a type of artificial intelligence—to distinguish between the 14 conditions. Testing the algorithm on the remaining 25 percent of patients showed it was correct around 99 percent of the time.
To try out the tool in a clinical scenario, the team applied it to 67 patients suspected, but not proven, to have one of the 14 conditions. For 21 of these, the epigenetic signature found a match. Some classifications corresponded to clinicians’ suspected diagnoses, but others didn’t. In the majority of the 21 cases, the team was able to verify that the tool’s conclusion was correct with further disease-specific genetic tests. As for the rest of the patients who didn’t get an episignature match with one of the 14 diseases, “it is possible that they may have another condition for which epi-signature is yet to be defined,” Sadikovic writes to The Scientist in an email.
The group then applied the model to a cohort of nearly 1,000 patients who had undergone extensive genetic testing such as copy number array or exome sequencing. The tool diagnosed 15 patients with one of the conditions—all but one of which were confirmed with additional, disease-specific tests.
To investigate if the remaining patients had any other methylation-related disorders, the investigators specifically looked for genomic segments with methylation patterns different from a control group of nearly 3,000 healthy genomes. For instance, they found an additional 7 individuals to have an imprinting defect, conditions that result in either a gain or loss of function in an imprinted gene. The team also found 106 patients that had rare epi-variants, individual loci with abnormal DNA methylation patterns, which in some cases appeared to be related to the patients’ phenotype.
Overall, Sadikovic estimates the approach could increase the rate of clinical diagnoses for rare hereditary disorders by 3–5 percent. A major advantage is the ability to diagnose patients for whom previous genetic testing has turned up a “variant of unknown clinical significance,” he explains. In these cases, it’s not clear if the mutations are causing the disease, or if they’re just benign polymorphisms. Finding out typically requires functional investigations of the gene, or family studies to establish whether the variant is inherited along with phenotypes. When these options aren’t possible or fail, and clinicians hit a dead end. The tool can determine whether a patient actually has the episignature for a given disease, and whether a particular variant is causing the disease or not.
Sadikovic and his colleagues have already licensed the tool to two diagnostic laboratories in the EU and the US, and plan on launching it commercially next week at the annual meeting of the American College of Medical Genetics.
Fahrner, who works in a specialized epigenetics and chromatin clinic, says she thinks the tool could prove particularly useful for patients who present with very unspecific symptoms. For instance, it could be used as a first-pass diagnosis in parallel with routine genetic analyses. “If someone comes in with unspecific features like developmental delay, I could see us sending a SNP array to look for copy number changes, and also this DNA methylation array which would look for things like Fragile X” and other disorders, she says. “One of the benefits of this type of test is that you wouldn’t have to send three different tests” to investigate each suspected diagnosis individually, she adds.
Sadikovic says he hopes the success rate of his tool will improve over time, as more conditions are identified that have such episignatures. He suspects many more exist, but more data are needed to characterize them. For instance, no episignature is known for Rett disorder, even though this often caused by mutations in a gene encoding a methylation-binding protein. “The more data, the better these types of approaches are. . . . I think we’re at the tip of the iceberg here.”
Andrew Sharp, a professor of genetics and genomics sciences at Icahn School of Medicine at Mount Sinai who was also not involved in the study, agrees. “My guess is that there’s going to be a number or other diseases where there’s a distinct epigenetic signature attached to mutations in certain genes, so probably the subset of individuals that can be discovered by this technique will grow over time.”
For now though, the approach is only useful for a very small proportion of patients where previous genetic tests have been inconclusive, and only if they have a particular set of conditions. “It’s powerful, if they have one of those diseases,” Sharp says.
E. Aref-Eshghi et al., “Diagnostic utility of genome-wide DNA methylation testing in genetically unsolved individuals with suspected hereditary conditions,” Am J Human Genet, doi:10.1016/j.ajhg.2019.03.008, 2019.