NEW YORK (GenomeWeb) – The genomic architecture of schizophrenia is highly polygenic, according to an analysis conducted by a team of Boston-based researchers.
The Harvard School of Public Health's Alkes Price and his colleagues developed a new algorithm to gauge the genetic architecture of complex diseases. As they reported today in Nature Genetics, they applied their approach to the Psychiatric Genomics Consortium (PGC2) and Genetic Epidemiology Research on Aging (GERA) datasets to estimate components of heritability, polygenicity, and genetic correlates of schizophrenia and other complex diseases.
"We have introduced a new fast algorithm, BOLT-REML, for variance-components analysis involving multiple variance components and multiple traits, and we demonstrated that it enables previously intractable large-sample heritability analyses," Price and his colleagues wrote in their paper.
Variance-component analysis has helped aggregate signals across SNPs as a complement to genome-wide association studies, but the researchers noted that computational limitations have forced scientists to split larger datasets up and then perform a meta-analysis. To get around that, the investigators developed a new method called BOLT-REML that relies on a conjugate gradient-based iterative framework and a Monte Carlo average information restricted maximum-likelihood algorithm.
Using the GERA cohort, the researchers compared their approach's capabilities to those of the GCTA software for restricted maximum-likelihood analyses. From this, they reported that while the GCTA software could handle about half the cohort, their BOLT-REML approach could perform a number of heritability analyses on the full set of 50,000 samples.
From the PGC2 cohort, Price and his colleagues analyzed 22,177 schizophrenia cases and 27,629 controls with imputed genotypes at nearly half a million markers, and from the GERA cohort, they analyzed nine complex diseases in 54,734 randomly ascertained samples genotyped at 597,736 SNPs.
Through their analyses — which included 10 principal components covariates to account for population stratification — they estimated the liability scale for SNP heritability for schizophrenia and nine GERA diseases, including allergic rhinitis, asthma, dyslipidemia, and hypertension.For schizophrenia, they reported a liability scale SNP heritability of 0.274, and one of 0.074 for allergic rhinitis and 0.152 for asthma.
The researchers noted a significant downward bias of schizophrenia liability-scale SNP heritability estimates as the sample size increased, a trend that the randomly ascertained GERA cohort didn't exhibit.
"These analyses help explain a previously mysterious observation of decreasing [liability-scale SNP heritability] estimates for schizophrenia with increasing aggregation of cohorts," Price and his colleagues added. "This phenomenon was attributed to phenotypic heterogeneity. … Our analyses implicate ascertainment-induced downward bias of estimated [liability-scale SNP heritability] as an additional explanation of this effect."
The researchers also examined the polygenicity of schizophrenia and the GERA diseases dyslipidemia and hypertension by estimating the SNP heritability that's explained by 1-megabase chunks of the genome. Though the results for the various regions were individually noisy, the researchers reported that some regions harbored substantial SNP heritability, especially for dyslipidemia.
Through additional simulation-based analyses, the researchers estimated that there are more than 20,000 causal SNPs for schizophrenia. They further noted that schizophrenia is extremely polygenic as most 1-megabase regions seemed to contribute to disease heritability.
"Our inference that most 1-Mb regions of the genome (71 percent to 100 percent) contain schizophrenia-associated loci evokes the concern that increasingly powered complex trait GWAS will ultimately implicate the entire genome, becoming uninformative," Price and his colleagues added.
The researchers found other correlates linked to these conditions. For instance, they reported that there was a correlation between each disease and GC content. Previous work, they noted, has found that GC architecture influences recombination and replication timing as well as DNA methylation, though they added that further study is needed to tease out the underlying mechanisms.
They further found that adjusting for body-mass index led to a relative reduction in genetic correlates among the GERA diseases — six genetic correlates became non-significant after accounting for BMI. This, they noted, underscores the necessity of considering heritability covariates when performing and analyzing genetic analyses.
Still, they reported a significant correlation between asthma and allergic rhinitis, a link that they noted has long been known.