21 February 2012. Common variants capture at least 23 percent of liability for schizophrenia, according to a new analysis published February 19 in Nature Genetics. Led by Naomi Wray and Peter Visscher at the University of Queensland in Brisbane, Australia, the study estimates the collective effect of all single nucleotide polymorphisms (SNPs) in the large dataset of the Psychiatric GWAS Consortium on Schizophrenia (PGC-SCZ). The results argue that many common variants with small effect sizes substantially contribute to risk for the disorder.
“It fits with a mutational load kind of model, where many things have to go wrong at the same time to increase risk,” Wray told SRF. “To me, the biology of it makes sense in that we have a robust system with redundancy, so that when any one of these things goes wrong, individually it doesn't have much impact.”
While ever-larger schizophrenia GWAS have detected a handful of common variants meeting the very high bar for genomewide significance (see SRF related news story), together these variants account for only about 3 percent of the risk, falling short of explaining the rather high heritability of schizophrenia. This “missing heritability” problem is seized upon by some as an argument against common variants as making any meaningful contribution to schizophrenia risk, leaving rare variants as more important culprits (see SRF genetics overview). Amid this debate, the new analysis offers a tangible, quantitative measure of the relative importance of common variants, and argues that some heritability is not missing, but hidden among many common variants with very small effect sizes lurking below the genomewide significance threshold, but which would emerge with larger GWAS.
Further analyses showed that this signal was due to common causal variants rather than rare ones, and was enriched for SNPs within central nervous system genes. The findings don’t rule out a contribution of rare variants to the as-yet unaccounted for heritability, however. “I 100 percent expect there to be rare variants,” Wray says. “But I think following up the common ones may be more informative for the population as a whole than studying the rare ones.”
Using methods first applied to human height GWAS (Yang et al., 2010), first author Sang Hong Lee and colleagues estimated the total contribution of 915,354 SNPs to schizophrenia liability in 9,087 individuals with schizophrenia and 12,171 controls—the first time these methods have been applied to disease.
The researchers estimated how genetically different cases were from controls with a measure of genomic variance based on comparisons between the patterns of the 915,354 SNPs in each individual. This computationally intensive endeavor found that the genomic variance accounted for 23 percent of the phenotypic variance, summarized as liability for schizophrenia, equivalent to 30 percent of its heritability. This estimate was consistent with one Wray previously made on a subset of the sample (see SRF related conference story), and another in a related analysis in the GWAS conducted by the International Schizophrenia Consortium (see SRF related news story).
The researchers then set about exploring how much of this reflected a true signal versus artifacts of case-control studies. Genotyping bias, which can lead to seemingly disease-related results when case samples are processed differently from control ones, was ruled out based on an analysis showing that subsets of the data collected by different research groups provided similar results. Population stratification, another potential artifact in GWAS in which different allele frequencies occur between cases and controls because of differences in ancestry rather than disease status, was also deemed to be minimal in the new analysis. When population stratification is driving a GWAS signal, a causal variant on one chromosome could be tagged by a SNP on a different chromosome with the same ancestry. To address this possibility, the researchers partitioned the SNP data by chromosome, and asked what proportion of the variance each contributed. Considering the chromosomes separately, then adding up their individual contributions to the variance, picks up on cross-chromosomal signals, and showed 26 percent—an estimate that was not dramatically higher than the 23 percent found when considering all chromosomes simultaneously. This similarity suggests that the GWAS signal was not an artifact.
The amount of variation attributed to SNPs from each chromosome also correlated with the length of the chromosome (r = 0.89, p = 2.6x10-8)—something that fits with a polygenic model of schizophrenia. Furthermore, despite clinical differences between men and women with schizophrenia, subdividing the data by gender did not reveal a difference in the variance in liability captured by SNPs on the autosomes, and on the X chromosome. This suggests that males and females share the same genetic basis for schizophrenia.
The common-to-rare spectrum
The researchers also partitioned the variance in liability captured by SNPs by function in order to examine the amount of the variance—the 23 percent figure from above—explained by SNPs within genes highly expressed in the central nervous system (CNS), by SNPs in other genes, and by SNPs not within genes. This revealed a similar proportion of variance in each of these three broad categories; however, the 2,725 CNS genes accounted for more variance (31 percent) than expected, given their length and the number of their SNPs (they represent only 20 percent of the genome). This argues that the genomic variance captured by SNPs includes signals pertaining to the brain.
Finally, the researchers had a look at how the variance distributed itself across common and less common SNPs to grapple with the type of variant responsible for their signal. Dividing the SNPs by their minor allele frequency (MAF), the researchers found that the least common ones (0.01 <-MAF <-0.1, meaning they made up 1-10 percent of all gene copies in the population) contributed 2 percent to the 23 percent estimate. The others, ranging from 0.1 to 0.5 MAF and hence fitting the definition of a common variant, contributed the rest. The researchers also simulated a rare, variant-only model of disease and turned up a different distribution of how variance was allocated, with little resemblance to what had been observed. These analyses finger common variants as responsible for the genetic susceptibility to schizophrenia captured by SNPs.
What’s left? The authors suggest that the remaining missing heritability can be found in causal variants that are not yet tagged consistently by SNPs with current microarray technology. This includes both common and rare variants, and finding the rare ones is exacerbated by the fact that it is hard to correlate common SNPs with something that is rare. While convinced that many, many common variants of small effect form a substantial part of the genetic architecture of schizophrenia, the authors recognize the potential contributions of rare variants, concluding:
“Hence, causal risk variants for schizophrenia range across the entire allelic frequency spectrum.”—Michele Solis.
Lee SH, DeCandia TR, Ripke S, Yang J, The Schizophrenia Psychiatric Genome-Wide Association Study Consortium (PGC-SCZ), The International Schizophrenia Consortium (ISC), The Molecular Genetics of Schizophrenia Collaboration (MGS), Sullivan PF, Goddard ME, Keller MC, Visscher PM, Wray NR. Estimating the proportion of variation in susceptibility to schizophrenia captured by common SNPs. Nat Genet 2012. Abstract