27 October 2010. Genomewide association studies (GWAS) have turned up a handful of variants that explain a fraction of heritability for schizophrenia. Whether this shows that GWAS is just gaining traction in the particularly complicated field of psychiatric disease, or is a dead end that should be abandoned, remains hotly debated (see SRF related news story and SRF related story). Some argue that the genetic risk for schizophrenia lies in rare variants not tagged by GWAS approaches. Others, including many at the meeting, counter that GWAS have been statistically underpowered, and that increasing sample sizes will kick interesting variants up and over the very high bar set to achieve genomewide significance (see SRF related meeting report from WCPG 2007). [Ed. note: For an overview of the whole field, see the SRF's genetics series.]
It was plain at the meeting that researchers haven't given up on GWAS yet. Many explored ways to bridge the sizeable gap between the rather high heritability for schizophrenia and the fraction of this genetic risk—about 3 percent—explained by GWAS-derived common variants—the so-called "missing heritability."
Lumping and splitting SNPs
Naomi Wray of Queensland Institute of Medical Research in Australia argued that this heritability wasn't missing, but rather hidden in many common variants with effect sizes too small to make it to genomewide significance with the sample sizes used so far. Rare variants of small effects could also contribute, but may be impossible to detect. In a talk on Wednesday, 6 October, Wray showed how quantifying the combined contribution of all SNPs—regardless of their significance level in GWAS—can account for a larger portion of heritability for schizophrenia. Using a method developed by her colleagues to examine height, another polygenic trait with a "missing heritability" problem (Yang et al., 2010), Wray found that considering nearly 300,000 SNPs together could explain 30 percent of the variance in the International Schizophrenia Consortium dataset, similar to simulations done last year (see SRF related news story). Though this analysis doesn't pinpoint genetic loci, it does suggest that common variants tagged by SNPs so far can explain a substantial chunk—in this case 50 percent—of heritability in schizophrenia.
In more talks that day, researchers tried other—sometimes exotic, usually complicated—ways of gleaning insights from the GWAS approach. Recognizing that genes of similar function are often grouped together in the genome, Eske Derks of UMC Utrecht in the Netherlands did a segment-wise analysis of SNPs in the genome to find out whether there were regions containing a larger-than-expected-by-chance number of weakly associated SNPs. She and her colleagues tested 12,500 segments of the genome between 2 and 32 Mb wide, and found several segments associated with schizophrenia in three different samples, including a portion of chromosome 4q containing 9 genes never before implicated in schizophrenia. In another talk, Danielle Posthuma of VU University Amsterdam in the Netherlands took a hint from the idea that schizophrenia stems from synapse dysfunction in the brain, and asked whether the synaptome, the set of roughly 1,000 genes that encode synaptic proteins, contained common variants associated with schizophrenia. This limited search does not have as large a multiple test burden as genomewide tests do, so many synaptome SNPs reached significance—more so than a set of 1,000 randomly drawn genes.
In another session, Alex Richards of Cardiff University, U.K., combined gene expression data with SNPs nominally associated with schizophrenia and found that SNPs with a greater influence on gene expression predict schizophrenia status better than SNPs without such an effect. Xiangning Chen of Virginia Commonwealth University reported on newly published efforts (Chen et al., 2010) to re-examine GWAS datasets to find variants that have true effects on schizophrenia risk though they don't achieve genomewide significance. This data-mining turned up two non-synonymous SNPs in the cardiomyopathy-associated 5 gene (CMYA5), which were then verified in 23 other samples.
Is bigger GWAS better GWAS?
But maybe the brute force method of boosting sample sizes prevailed that day. On behalf of the Schizophrenia Psychiatric GWAS Consortium (PGC), Stephan Ripke of the Broad Institute, Cambridge, Massachusetts, presented results from the largest GWAS of schizophrenia to date. In discovery and replication cohorts, several signals went well above the threshold of genomewide significance, and a combined analysis of over 40,000 individuals offered up seven genomewide significant regions. One was the major histocompatibility complex (MHC) locus, a large region on chromosome 6 pinpointed by previous studies. A new and intriguing locus on chromosome 1 contains miR-137, a microRNA involved in adult neural stem cell proliferation and maturation (Smrt et al., 2010). Ripke noted that miR-137 targets genes already linked to schizophrenia, such as c10orf26, TCF4, and CACNA1C. He concluded that even greater sample sizes should garner more loci significantly associated with schizophrenia, adding: "We are only just on our way."
Maybe auspiciously, the talk ended with a crack of thunder—Zeus clapping from the nearby Acropolis?
A case for bigger GWAS was also made in the summing-up session on Thursday morning, which highlighted results from two other studies with large sample sizes. Jordan Smoller of Harvard Medical School briefly outlined findings from the Cross-Disorder PGC, which combines schizophrenia, bipolar disorder, and major depressive disorder cases, and currently has over 45,000 cases. This has produced several genomewide significant hits, including ITIH3, the MHC region, NT5C2, CACNA1C, and a signal near TCF4, but these results await replication. Smoller also noted that the polygenic score method of asking how much variance in one disorder predicts variance in another showed greater overlap between bipolar disorder and schizophrenia, consistent with the idea of a shared genetic vulnerability between them.
Pamela Sklar of the Broad Institute briefly presented results from the Bipolar Disorder PGC. A combined analysis of over 63,000 samples provided more support for CACNA1C, a gene which encodes a calcium channel subunit (see SRF related news story) and is also a schizophrenia suspect. Despite the hints of genetic overlap among common variants for schizophrenia and bipolar disorder, Sklar did note that the large CNVs described in schizophrenia and other brain disorders are not turning up in bipolar disorder so far (see SRF related news story).
The GWAS debate continues
Optimism about GWAS was echoed by other researchers in this session, with many arguing that it would be a mistake to abandon the GWAS approach just as it was starting to work. Pat Sullivan of University of North Carolina in Chapel Hill said that GWAS work when the sample size is large enough, and that even undersized studies could reveal useful insights into disease. Nick Martin of Queensland Institute of Medical Research was more emphatic, noting "spectacular success in schizophrenia," and saying that researchers should be "triumphant" about GWAS working for psychiatric disease. Michael O'Donovan of Cardiff University summed up the current state of schizophrenia research, counting 16 independent GWAS signals and nine CNV loci. "We shouldn't give up the GWAS," he said. As a demonstration of how powerful GWAS can be, several researchers referred to the newly published GWAS of human height in which a sample size of 180,000 turned up 180 loci (Lango Allen et al., 2010).
But David Curtis of Barts and the Royal London School of Medicine argued that height was not the same thing as schizophrenia, in that boosting sample sizes might be less useful for notoriously heterogeneous psychiatric diseases. Lumping together hundreds of thousands of cases to create a sufficiently powered GWAS runs the risk of diluting the specificity of a phenotype and swamping any signal. He also cautioned researchers to be aware of the limits of their SNP arrays: if they miss 25 percent of the genome, then increasing sample sizes will obviously not help them find signals in these missed regions. Others worried that the number of loci gained per increase in sample size may fall short of what is needed to explain a disorder like schizophrenia, which has been estimated to involve thousands of variants.
Though the funding situation presses people to take sides on whether the hunt for common or rare variants will be more fruitful, many during the meeting seemed to recognize that both approaches will yield important insights into disease. Saying that neither approach alone would uncover everything, Nick Martin likened it to doing a puzzle, in which you put together the easiest pieces first, then gradually fill in with the more difficult. "This is a road to discover all bits contributing to a disorder."—Michele Solis.