NEW YORK (GenomeWeb) – Investigators have developed a diabetes risk and diagnosis algorithm, which they said now does a better job of discriminating type 1 from type 2 diabetes than prior tests.
The improved assay could also be a much more efficient tool for recruiting children for interventional trials and could serve as a future clinical screening test, they said.
In a study published last month in Diabetes Care, investigators described the new algorithm, called T1DGRS2, which is an updated version of their previously developed polygenic score method. The team calculated that the updated test is about twice as efficient in identifying babies at a high risk of type 1 diabetes as prior genetic score methods, which were themselves a significant improvement over the simpler HLA-typing initially used in this area.
The group, led by a team at the University of Exeter, published on their first polygenic assay in 2016. Since then, other groups have also developed scores, in some cases finding that combinations of different scores offered improved prediction. A team of German researchers, for example, recently created a 41-SNP score, which they found improved prediction when combined with the Exeter team's assay.
Richard Oram, the study's senior author and a University of Exeter research fellow, said in an interview this week that he and his team approached the creation of both their initial diabetes genetic risk score and the new updated version "with a view for [them] to be used for diagnosis, and research, and prediction."
In diagnosis, an assessment of an individual's genetic risk of type 1 diabetes helps distinguish difficult cases — where it's unclear whether someone has type 1 or type 2 disease. Separate research by the Exeter team, for example, has calculated that up to about half of all cases of type 1 diabetes develop in adulthood, and can often be misdiagnosed as type 2.
Oram said that his team's previous score is already being used in this manner — offered through the genetics lab at the University of Exeter "for resolving difficult diagnosis and to help decide when to do genome sequencing for monogenic forms of diabetes."
To expand availability, the academic team has also partnered with a diagnostic company, Randox, to make a chip version of the test, which could be used in other labs. "We are in the process of CE marking that," Oram said.
While diagnosis is one application for polygenic assays in diabetes, another exciting aspect of the group's new improvement of the T1DGRS algorithm is that it could make it more useful for ongoing and novel interventional clinical trials: to select children for studies of specific drugs, vaccines, or other preventative treatments.
"This [has already been started] using our score plus the Winkler score in a large worldwide study in New York," under the Global Platform for the Prevention of Autoimmune Diabetes, or GPPAD.
As pharmaceutical companies and public health organizations pursue various drug and diet interventions to prevent T1D, there is also already some evidence that risk-testing itself — and providing the results to families — can improve outcomes.
"As part of an effort to develop screening [tools] … the question we have to challenge ourselves with is, why would we screen. Why identify people at high risk and give that information back to them?" Oram said. "One reason is if these interventions [that are being trialed] work, that’s a very strong argument for screening."
But knowledge of genetic risk itself also might impact outcomes. For example, Oram said, in a US study called TEDDY (The Environmental Determinants of Diabetes in the Young) researchers performed polygenic risk analysis and followed thousands of infants through their first 10 years of life. In the end, there were significantly lower rates of severe and life-threatening presentations in the nearly 400 babies who developed T1D than are seen in the general population.
In developing their newly published second-generation GRS, Oram and his colleagues analyzed genetic variation across the genomes of 6,581 people with type 1 diabetes from the Type 1 Diabetes Genetics Consortium and compared the results to 9,247 control participants without the disease.
By modeling interactions between variants to focus on strongly associated HLA haplotypes and performing various statistical analyses, the team was able to generate the new improved T1D GRS2, which includes 67 SNPs.
According to Oram, the group looked at the performance of the new GRS relative to its previous version and other genetic risk methods that have been published using two methods. The team first calculated areas under the receiver operating curve, reporting that the GRS2 had an AUC of 0.92 overall, and an even higher 0.96 AUC for early-onset T1D
They also examined the test's predictive ability specific to a screening application. "Given the purpose was identifying people at high risk, [we calculated that] if you do HLA typing alone, you can identify children at 5 percent risk, but no higher," he said.
Simulating the context of newborn screening, "the T1D GRS2 was nearly twice as efficient as HLA genotyping alone and 50 percent better than current genetic scores in general population T1D prediction," the authors wrote.
"The reason this matters so much is [for] intervention trials," Oram said. By picking up kids at higher risk, trials can be powered to pick up a smaller effect, or costs can be reduced by enrolling fewer individuals — because a doubling of risk would allow for the number of children who need to be randomized to be sequentially halved.
Although there could be room for even further refinement or improvement of the score, Oram said that although the current assay profiles a relatively small area of the genome, it already accounts for "nearly all the heritable risk that has been identified," for type 1 diabetes.
"We have gotten quite close," to explaining full heritability he said, and much closer than efforts in other disease areas. In type 2 diabetes, for example, only about 20 percent of heritability is accounted for by the latest risk scores, which can include hundreds of variants.
In other words, Oram said, while there are certainly areas where there is more genetic real estate to find, "for autoimmune diseases that are strongly HLA-linked, a smaller number of SNPS really can capture the majority of risk."
He and his team did compare some established genome-wide score methods as part of their latest effort, he added, and found them to be no better than the GRS2 signature, though this was part of the study review process and wasn't included in the final paper.
Whether for more accurate diagnosis or for potential future screening efforts, Oram stressed that the "the low hanging fruit" is combining genetic risk with other orthogonal factors. In the study published using their first-generation GRS, combining the genetic score with measuring autoantibodies and clinical factors like body mass yielded "near perfect discriminatory power."
The team is now working on the same type of combined analysis using the second-generation score, focusing efforts on analyzing their algorithm in a US population, as part of a US Centers for Disease Control and Prevention and National Institutes of Health study called Search for Diabetes in Youth. This would allow the researchers to answer key questions about whether results hold up in non-white populations, namely African-Americans and Hispanics.
Although analyses from this work has yet to be published, Oram said that it's looking positive. While the team's first-generation score was not as predictive in these populations as in white/European children, the new algorithm looks like it might perform more uniformly across the different groups.
The group is also pursuing a similar effort in India.