22 October 2007. There has been great anticipation in the world of psychiatric research over the past year, with the community awaiting the results of a number of genome-wide association studies (GWASs). The preliminary results of some of the schizophrenia and bipolar studies were presented in oral sessions at the recent World Congress on Psychiatric Genetics, 7-11 October, in New York City. Similar pictures emerged for both disorders—no strong (p <10-6) replications across studies, no candidates with strong effects on disease risk (i.e., no odds ratios approaching 2.0), and no clear replications of genes implicated by candidate gene studies. However, the field can now investigate the hypothesis that this large crop of candidates contains small-effect susceptibility genes for these disorders.
Speakers did agree on two points: that it is necessary to practice “aggressive data-sharing” to increase sample sizes, and to apply ruthless quality-control techniques on initial results to weed out false positives.
Schizophrenia GWA studies
The schizophrenia GWA session was held on 11 October and chaired by Ridha Joober of Douglas Hospital in Montreal. Among the six talks were new data from the recently published schizophrenia GWAS led by Todd Lencz of Zucker Hillside Hospital in Glen Oaks, New York (see SRF related news story), as well as unpublished GWA studies presented by Jennifer Stone of the Broad Institute in Cambridge, Massachusetts, Mick O'Donovan of Cardiff University in the United Kingdom, Lin He of Shanghai Jiao Tong University in China, Patrick Sullivan of the University of North Carolina, Chapel Hill, and Sven Cichon of Bonn University in Germany. While the researchers did not find markers associated with schizophrenia at experiment-wide significance values, their "hits"—markers with p values in the 10-7 to 10-5 range—do offer new candidate genes for further examination.
The Cichon study used the Illumina HumanHap550 platform and the Stone, O'Donovan, He, Lencz, and Sullivan studies used the Affymetrix 500K platform (as well as the Affymetrix 1 million chip in the case of Stone). In addition to a number of markers outside coding regions, three groups mentioned genes by name that they deemed to have evidence for association with schizophrenia: HLA-DRB1, HLA-DRB5, SETBP1, CRLS1, ZNF323, and C20orfg26 (Stone); ZNF804A, RPGRIP1L, and NOS1 (O'Donovan); CUTL1 (Cichon).
While Sullivan noted that none of the markers in the CATIE sample showed genome-wide significant association with schizophrenia, his group's subsequent analysis showed preliminary support for NRG1 and DISC1, and indeed, for DISC1 the findings cluster at the breakpoint identified in the original reports by the Porteous group.
There was one departure from the many genes of small-effect paradigm: Lencz described a new approach by the Malhotra group wherein they probed their dataset for recessive genes that might contribute larger genetic effects in schizophrenia. Specifically they looked for regions with improbably long stretches ("runs") of homozygosity—statistically improbable sequences in which the same nucleotide sequence was inherited from both parents. While this is apparently a common feature of the genome even in healthy subjects, this phenomenon was observed to occur with significantly greater frequency in patients with schizophrenia. They identified a greater number of runs of homozygosity among patients than among controls in nine distinct genomic regions, four of which contain genes previously suggested to be susceptibility genes for schizophrenia (CAPON [aka NOS1AP], NSF. ATF2, and PIK3C3).
Conference organizer Lynn Delisi of New York University raised the question of why there was no replication of popular candidates such as NRG1 or DTNPB1. It was noted by several speakers that the current GWA technologies, despite covering 500,000 SNPs, are still sampling just a small fraction of the genome, and were not designed for specific coverage of these gene candidates, which consequently are not well covered. Sullivan suggested it would take in the range of two million SNP chips to densely cover the top current candidates.
One audience member cautioned against using a "sliding scale" for picking out interesting candidates, noting that gene markers with significance figures in the range of 10-4 may contain a small percentage of true positives. Daniel Weinberger of NIMH commented that, similar to the candidate gene approach, it would make sense to look more closely at "hits" with lower than genome-wide significance but prior evidence from other lines of research.
Bipolar GWA studies
The bipolar GWASs presented on 8 October were a mixture of published and unpublished work. The Wellcome Trust study presented by Nick Craddock of Cardiff University, and the NIMH study, presented by Amber Baum of NIMH in Bethesda, Maryland, were published in the past half year (see SRF related news story). These authors presented new data from their groups' efforts to verify their results in each other's publicly available datasets. Unpublished bipolar GWASs were presented by Pamela Sklar of the Broad Institute, Laura Scott of the University of Michigan in Ann Arbor, and Sven Cichon of Bonn University. The Baum, Scott, and Cichon studies used the Illumina HumanHap550 platform and the Craddock and Sklar studies used the Affymetrix 500K platform.
A number of genes, all of small effect at best, were highlighted by the different authors as ranking at the top of their respective lists of significant associations: PALB2, GABRB1 (and other GABA receptor subunit genes), GRM7, and SYN3 (Craddock); CACNAIC, MYO5B, and EGRF (Sklar), MAN2A, PRKCE, and DAGLA (Scott); DGKH, ZIP3, JAM3 (Baum).
Session chair Steve Faraone complimented the speakers both on their willingness to show SNP numbers and on the level of current and promised data-sharing. Craddock said that the experience with GWA studies in type 2 diabetes show you need big sample sizes, but warned that the p values and odds ratios will not be impressive, meaning that some true risk genes may not show up even in larger samples.
Several groups highlighted the importance of quality control. For example, Craddock noted that all of their initial top "hits" were thrown out during quality control steps. This seems to be a particular problem with the Affymetrix chips, for which researchers said that up to 25 percent of results have to be discarded.—Hakon Heimer.