Summary of Findings and Previous Literature
A multi-step analysis strategy was utilized to understand better the associations between NRG1 and cognition in the context of schizophrenia, other psychopathology, and intelligence. At all three analytic steps, several variants in NRG1 were nominally associated with various cognitive domains, although none reached significance after correcting for multiple tests.
Given this context of many nominally significant but no experiment-wise significant associations, we will focus here on those associations that were nominally significant at all of the three analytic steps as potentially most promising for future study. Although statistically the most powerful test, associations in the total sample could be spurious due to secondary effects of schizophrenia or other psychopathology causing cognitive deficits, whereas analyses in the no-diagnosis sample eliminate this possibility and the analyses with intelligence as a covariate further narrow the field to those SNPs whose associations with cognition are not completely due to effects on general intelligence. Although some promising associations might not meet these criteria, those that do seem most likely to be of interest.
Associations that were nominally significant at all steps include four cognitive domains (abstraction and mental flexibility; attention; verbal memory; and sensorimotor dexterity) and four SNPs (rs10954855, rs3924999, rs2466060, and rs10503929). The LD within this group of SNPs was less than or equal to an r2 of 0.2, except for the following: rs3924999 and rs10954855 (r2=0.3), and rs3924999 and rs2466060 (r2=0.5). These associations are discussed below in order of SNP location within NRG1 (5′ to 3′). Effect sizes are shown in . In addition, previous findings from the literature for each of these SNPs are summarized.
Marker rs10954855 was nominally associated with verbal memory performance in each of the analyses, with the minor allele (A) always conferring a benefit. This SNP lies less than 1000bp from microsatellite 317J8-2123, which was associated with schizophrenia in a Chinese sample (Li et al., 2004
). To our knowledge, no studies of rs10954855 and cognition exist, although one previous study (Duan et al., 2005
) failed to find any significant association with schizophrenia.
SNP rs3924999 was nominally associated with attention in all analyses with the minor allele (T) improving performance. This marker is a non-synonymous exonic SNP that results in a change from arginine (R) to glutamine (Q) if a C to T (minor) allele substitution occurs. This SNP’s minor allele has been associated with schizophrenia (Yang et al., 2003
), as well as perceptual aberrations in adolescents (Lin et al., 2005
), although, another study failed to find association with schizophrenia (Hong et al., 2004
). To our knowledge, no studies of this SNP and cognition have been performed. In contrast to these previous studies, the minor allele of rs3924999 was associated with an improvement in cognitive performance in the current study.
SNP rs2466060 was also nominally associated with attention in all analyses with the minor allele (A) improving performance. Although not significant after correction for multiple testing, rs2466060 and attention produced the strongest consistent associations across analyses (uncorrected ps <.007) in this study. This marker is located near microsatellite 317J8-4858, which was associated with schizophrenia in a Chinese sample (Li et al., 2004
). To our knowledge, no studies of rs2466060 and cognition exist, although one previous study (Petryshen et al., 2005
) failed to find any significant association with schizophrenia.
Finally, rs10503929 was nominally associated with both abstraction and mental flexibility performance and sensorimotor dexterity in all analyses, where the minor allele (C) always reduced performance. This marker is a missense non-synonymous SNP that leads to the substitution from a methionine (M) to threonine (T) amino acid if a T to C (minor) allele change occurs. A recent meta-analysis (Allen et al., 2008
) of case-control studies found that rs10503929 was the only NRG1
SNP of 13 assessed that was significantly related to schizophrenia (odds ratio: 0.88), suggesting a protective effect of the minor allele. In contrast, the minor allele of rs10503929 was associated with a decrease in cognitive performance in the current study.
Other NRG1 Markers Associated with Cognition in Previous Literature
Despite some significant associations between rs6994992 and cognition in previous studies, this marker consistently failed the SNPlex Assay Design algorithm that was used to create the primer pool and could not be analyzed in the current study. Marker rs35753505 had been associated with sustained attention in a general population sample (Stefanis et al., 2007
), but in the current study it was associated with abstraction and mental flexibility in the “no diagnosis” sample (p=0.0178) but not with attention in any sample.
Association between NRG1 SNPs and Schizophrenia
There was one nominally significant association between NRG1 SNPs and the diagnosis of schizophrenia or schizoaffective disorder-depressed in the current study, although it was non-significant after correction. This relative lack of findings is likely due to the small sample of affected participants leading to low power to detect associations between the disorder and individual SNPs.
Strengths & Limitations
Relative to other studies on this topic, this study utilized a large, multiplex, multigenerational family sample of schizophrenia to assess the association between the largest number of NRG1 SNPs and most cognitive domains to date. In addition, a strategy of sequential analyses was employed in order to control for factors such as schizophrenia, general psychopathology, and intelligence that can significantly affect analyses.
In addition to these strengths, it also has some limitations. First, the number of participants with schizophrenia or schizoaffective disorder-depressed in the sample was relatively small, which prevented us from testing for associations with NRG1 and cognition in the patient-only sample. The general lack of associations between NRG1 and affected status also likely reflects this low power. Furthermore, although the largest such study to date, the number of relatives was not as large as would be ideal. At the nominal significance level of .05, statistical power was good in the total sample, being able to detect SNPs accounting for possibly only 1% of trait variance (N=419, power = .80, assuming complete LD between SNP and causal variant), and in the No Diagnosis sample, being able to detect SNP effects accounting for 4.3% of trait variance (N=178, power = .80). As expected, power was reduced when correcting for multiple comparisons (p<.0003), being able to detect effects accounting for 3.4% of trait variance in the total sample but only 7.7% in the No Diagnosis sample. Finally, although a literature based SNP set was utilized in this study, a more comprehensive tag SNP (tSNP) design may further elucidate the individual SNPs that are important in cognitive performance (estimated 348 tag SNPs needed due to the size of NRG1).
Many of the SNPs with significant associations with cognition exhibited significant stratification. Although stratification is generally expected to lead to greater associations between- than within-families, in the current study, the between-family component was typically less than the within-family. Such results may indicate that some between-family factor, such as socioeconomic status, is associated, perhaps due to the recruitment process, with the allele, leading to significant stratification in which the direction of the association is different between compared to within families. Regardless of the nature of such effects, the QTDT analyses and within-family beta weights reported here are robust to any stratification effects (Abecasis et al., 2000
It is also possible that these nominal associations with cognition are irrelevant to schizophrenia, but the literature on NRG1
and schizophrenia suggests otherwise. In addition, although several of the alleles that were previously associated with schizophrenia in the literature were associated with reduced cognitive function in the current study, some alleles previously associated with affected status were associated with improved cognition here. The causes of such inconsistencies or “allele-flipping” are frequently unknown, but recent studies (Clarke and Cardon, 2010
, Goldberg and Green, 2002
) suggest that these inconsistencies are not always merely alpha error but may be due to allelic and locus heterogeneity, population-related and sample variation-related differences in LD structure between markers, and environmental exposures.
None of the associations met the criteria for significance after correcting for multiple testing, although each sample had between 19–21 nominally significant (p≤.05) associations, encompassing multiple SNPs and multiple cognitive domains. The large effect sizes of these associations with cognition (absolute value βwithin across all analyses: 0.08–0.73) are noteworthy given the usually reported modest size of the association between NRG1 and schizophrenia (odds ratio: 1–2.2). The numerous nominally significant findings in progressively healthier sub-samples also may suggest that any possible NRG1 effects are not solely secondary to an effect of schizophrenia or psychopathology. If replicated and found significant experiment-wise, these associations may suggest a role for NRG1 in cognition that is a mediating risk factor for schizophrenia or psychopathology more generally, although it is possible that NRG1 could affect schizophrenia and cognitive function independently. Although the current lack of significant experiment-wise results does not provide evidence for convergent validity for NRG1 as a risk gene for schizophrenia, the consistent “nominally significant” findings may suggest hypotheses for future studies that focus on a few specific promising SNPs and cognitive functions and thus are more statistically powerful given a reduced penalty for multiple tests.
In particular, when considering the cognitive domains separately from individual SNPs, attention consistently had the most nominally significant associations over all of the samples assessed, with abstraction and mental flexibility, verbal memory, and sensorimotor dexterity also having interesting results, suggesting that future studies could focus on these cognitive processes specifically. Similarly, SNPs rs10954855, rs3924999, rs2466060, and rs10503929 appear to be the most promising for future study. Overall, although not significant experiment-wise, these findings may suggest more specific hypotheses that warrant future study concerning the relationships among NRG1, schizophrenia and cognition. In addition, studies of the NRG1-ErbB signaling pathway, with particular attention paid to ErbB4, may further elucidate the pathophysiology of schizophrenia-related cognitive dysfunction.