6 Apr 2015
April 7, 2015. The acclaimed genomewide association study (GWAS) of schizophrenia brought to light 108 regions in which single nucleotide polymorphisms (SNPs) infinitesimally increase risk for the disorder (see SRF related news report). Adding these risk SNPs together amounts to more substantial increases in risk, but strains to account for schizophrenia's heritability. GWAS proponents expect that uncovering more risk SNPs through increased sample sizes will account for even more heritability, but a symposium Monday morning, March 30, pointed to another possible source of disease risk: gene-gene interactions, also called epistasis, which might explain a substantial proportion of the genetic contributions to schizophrenia.
Despite epistatic interactions found throughout biology, it was the sense in the room that it was a neglected possibility. "Why has the schizophrenia field been so reluctant to embrace epistasis?" asked John Waddington of the Royal College of Surgeons in Ireland, Dublin, who co-chaired the session.
A gene-gene interaction means that the effect of one variant depends on the genetic context in which it finds itself, and the combined effect on risk would be more than the sum of its parts. Such an effect might explain the lack of replication that dogged earlier genetic studies of schizophrenia—for example, one population simply did not have the needed genetic interactor that another had. GWAS, however, seeks the most common denominators of risk across a heterogeneous population, which one speaker called "low-hanging fruits," and another cast as "antithetical to precision medicine."
Scott Williams of Dartmouth University in Hanover, New Hampshire, gave a primer on epistasis, pointing out that the brain is an especially good venue for gene-gene interactions because 84 percent of genes are expressed there—far more than in other organs. But detecting epistasis can be tricky, because contributing variants by themselves may offer no hint of an effect on phenotype. This, he said, might mean that the current list of genomewide-significant hits for schizophrenia is not a good starting place for finding epistasis—and, indeed, the PGC GWAS did not find much evidence for it in a two-SNPs-at-a-time analysis of their hits. Williams said that analytical methods such as multifactor dimensionality reduction can help detect the combinations of genetic factors associated with risk.
Examples of gene-gene interactions in schizophrenia have come to light in work from Daniel Weinberger of the Lieber Institute of Brain Development in Baltimore, Maryland, who reviewed several published studies. One found interactions between DISC1 and NKCC1, which encodes a chloride transporter that regulates neuronal excitability; the interaction has effects both on neurite structures in vitro as well as on risk for schizophrenia (see SRF related news report). This suggests that GWAS are not finding variants in DISC1—a gene discovered in a Scottish family—because its effects are diluted when considered in isolation in a large heterogeneous population. Another interaction between CYFIP1 and ACTR2, genes with roles in cytoskeletal dynamics, has also been found for cellular phenotypes and schizophrenia risk alike (see SRF related news report); in fact, one allele for CYFIP either increased or decreased risk, depending on the ACTR2 allele it was paired with. Weinberger isn't bothered that these genes haven't turned up in GWAS, saying, "In large heterogeneous populations, epistasis almost by definition [is] exclusive of GWAS loci."
John Waddington then presented work in transgenic mice in which two genes—DISC1 and neuregulin (NRG1)—had been manipulated. Systematically working through the different combinations of genotypes, he showed that very different phenotypes can result. For example, measures of paired pulse inhibition showed no evidence for gene-gene interactions, but measures of mouse sociability were impaired in mice heterogenous for a NRG1 variant and homozygous for a DISC1 variant. But other NRG1 and DISC1 genotypes showed no effect on sociability, either alone or in combination, which illustrates specificity in these gene-gene interactions.
But humans have about 24,000 genes, and the head-spinning number of possible combinations is enough to send anyone running. Unruffled, Kristen Nicodemus of the University of Edinburgh, Scotland, oriented the audience to new methods for detecting the full range of epistatic interactions. One involved an analytical process called a random forest approach that works through the SNPs to ultimately classify them into groups that predict a phenotype. Another involved a linear regression approach she used in a study published last year, which found that epistasis increased the variation explained in working memory above that found for an additive contribution (Nicodemus et al., 2014). She is heading the PGC's schizophrenia epistasis working group and invited interested people to contact her.—Michele Solis.