2 Nov 2015
November 3, 2015. It was standing room only for a symposium at the 2015 World Congress on Psychiatric Genetics in Toronto, Canada, devoted to polygenic risk scores (PRS) on Saturday, October 17. As a measure of the total burden of common risk variants—the PRS sums a person's risk alleles, weighted by their effect size—the PRS was first marshaled in genomewide association studies (GWAS) of schizophrenia to detect a genetic signal from common variants. Since then, the PRS's uses have expanded, including a way to measure genetic overlap between disorders, a biomarker-like correlate to many phenotypes, and it may be used to stratify people into subgroups according to their risk burden.
Yet the biostatisticians in the house saw room for improvement in getting the PRS to capture risk. In fact, according to Frank Dudbridge of the London School of Hygiene and Tropical Medicine, the PRS may produce clinically useful scores before it fully explains genetic risk (heritability). He noted that in the case of type 2 diabetes, a PRS helps classify cases from controls when added to clinical information. He also said that calculating the PRS solely from single nucleotide polymorphisms (SNPs) that made it over the high bar of genomewide significance was "self-defeating" and did not improve PRS accuracy, which underlines the importance of incorporating as much information as possible. He presented a recently published method for calculating a PRS from GWAS summary statistics, which obviates the need for individual genotypes (Palla et al., 2015). This lends itself well to the giant datasets amassed by collaborations such as the Psychiatric Genomics Consortium (PGC).
Jack Euesden of King's College London talked about PRSice, software he developed to standardize PRS calculation (Euesden et al., 2015). One feature gave a high-resolution picture of how the PRS changes depending on the threshold for including an SNP in the score. Typically, researchers designate a particular p value threshold, based on looking at several ranges; by contrast, Euesden's high-resolution method looks at 10,000 different p value ranges, which can result in a clear peak PRS.
Euesden also made the case that the way PRS is calculated may depend on the scientific question, with the example of using SNPs identified by the PGC's schizophrenia GWAS to measure genetic overlap with major depressive disorder (MDD). The schizophrenia-discovered SNPs may include schizophrenia-specific factors as well as more general contributors to psychiatric disorders, and these may differ by their p values. Euesden explored how the p value threshold could be optimized for different parts of the genome, divided into 5 Mb chunks. This produced a more significant PRS for MDD than that obtained by applying the same threshold across the genome.
To make the PRS better at capturing risk, Robert Maier of the University of Queensland in Brisbane, Australia, looked toward multiple psychiatric disorders. He showed how using genetic correlations between psychiatric disorders—specifically, schizophrenia, bipolar disorder, and MDD—improved PRS accuracy in distinguishing cases from controls (Maier et al., 2015). These refinements may point the way to a clinically useful PRS that could identify subgroups of people based on their genetic risk.
Hilary Finucane of the Massachusetts Institute of Technology in Cambridge presented her recently published method called "stratified LD regression," which extracts biological information from SNPs identified by GWAS (Finucane et al., 2015). Rather than focusing on the top most statistically significant SNPs, the method uses information about SNP function—for example, whether the SNP is located within a coding region, a promoter, or some other regulatory part of the genome—to allow an understanding of how much each functional category contributes to heritability. Applying the method to 17 complex diseases or traits, including schizophrenia, Finucane found that SNPs in regions conserved among mammals had the largest effects. When looking at cell-specific categories (based on histone marks), the researchers found that schizophrenia showed enrichment for fetal brain.
The symposium wrapped up with remarks from Gerome Breen of King's College London. Breen proposed that it was time for psychiatric genetics to re-engage with clinicians, with the PRS in a position to revitalize the discovery of biomarkers and to inform clinical trials, given that results might vary according to a person's burden of risk. Dave Curtis of University College London wasn't convinced, however, saying that it wasn't yet clear whether the PRS added anything above asking about a person's family history of a disorder. Others countered that family history doesn't give extra information for some phenotypes, such as drug response, and Naomi Wray of the University of Queensland said that PRS could be informative for the 60 percent of people with schizophrenia who report no family history.—Michele Solis.