- Top of page
- Materials and methods
The CHRM2 gene is thought to be involved in neuronal excitability, synaptic plasticity and feedback regulation of acetylcholine release and has previously been implicated in higher cognitive processing. In a sample of 667 individuals from 304 families, we genotyped three single-nucleotide polymorphisms (SNPs) in the CHRM2 gene on 7q31–35. From all individuals, standardized intelligence measures were available. Using a test of within-family association, which controls for the possible effects of population stratification, a highly significant association was found between the CHRM2 gene and intelligence. The strongest association was between rs324650 and performance IQ (PIQ), where the T allele was associated with an increase of 4.6 PIQ points. In parallel with a large family-based association, we observed an attenuated – although still significant – population-based association, illustrating that population stratification may decrease our chances of detecting allele–trait associations. Such a mechanism has been predicted earlier, and this article is one of the first to empirically show that family-based association methods are not only needed to guard against false positives, but are also invaluable in guarding against false negatives.
Individual performance across a single aspect of cognitive ability is highly predictive of performance on other aspects of cognitive ability. Indeed, about 40% of the population variance of measures of these individual cognitive processes can be accounted for by a single general intelligence factor (Plomin et al. 2004). Multivariate genetic analyses indicate that this general intelligence factor is highly heritable (Boomsma & van Baal 1998; Cherny & Cardon 1994; Plomin et al. 1994; Posthuma et al. 2001) and that there is a substantial overlap in the genes influencing different aspects of cognitive ability. This implies that genes associated with one aspect of cognitive ability are likely to be associated with other aspects as well. As noted by Plomin et al. (2004), these quantitative genetic findings make general intelligence an excellent target for molecular genetic research.
In spite of its high heritability, reports on the actual genes influencing intelligence are scarce. Recently, Comings et al. (2003) reported an association between a variant of the cholinergic muscarinic receptor 2 (CHRM2 ) gene explaining 1% of the variance in scores on full-scale IQ (FSIQ) and years of education. We recently conducted an autosomal genome scan for intelligence using two independent, unselected samples consisting of 329 Australian families and 100 Dutch families, totalling 625 sib pairs (Posthuma et al. 2005). Although the most promising regions were 2q and 6p, we also found modest evidence for linkage with performance IQ (PIQ) at 7q31–36 right above the CHRM2 gene.
The cholinergic muscarinic receptor family (M1-M5) belongs to the superfamily of G-protein-coupled receptors. These receptors activate a multitude of signaling pathways important for modulating neuronal excitability, synaptic plasticity and feedback regulation of acetylcholine (ACh) release and cognitive processes, including learning and memory (Hulme 1990; Volpicelli & Levey 2004; Wess 1996). On the basis of its putative role in cognitive ability, we genotyped three tagging single-nucleotide polymorphisms (SNPs) in the CHRM2 gene in a sample of 667 Dutch individuals from 304 twin families. The current sample overlaps only marginally (17%) with the sample used in the linkage analysis. A family-based genetic association test was used, which allows evaluating evidence for association that is free from spurious effects of population stratification (Abecasis et al. 2000; Fulker et al. 1999; Posthuma et al. 2004).
- Top of page
- Materials and methods
Comparison of MZ and DZ twin similarities for IQ measures showed that the observed variation in IQ could be attributed to additive genetic variance and unique environmental variance and not to shared environmental variance. Heritabilities of PIQ, VIQ and FSIQ were 0.73 (95% CI 0.63–0.80), 0.70 (95% CI 0.59–0.78) and 0.80 (95% CI 0.72–0.85), respectively, in the young cohort. Using the complete adult cohort, the respective heritabilities for PIQ, VIQ and FSIQ were 0.71 (95% CI 0.62–0.77), 0.78 (95% CI 0.72–0.83) and 0.78 (95% CI 0.72–0.83). These heritability estimates are comparable with those reported previously for these age cohorts in the Dutch population (Bartels et al. 2002; Posthuma et al. 2001).
In total, 667 subjects were available for SNP genotyping. On the basis of blind controls and MZ checks, no genotyping errors were found. For SNP rs2061174, 2.8% of the genotypes could not be called (648 genotypes succeeded), for SNP rs324640, 4.0% of the genotypes could not be called (640 genotypes succeeded) and for SNP rs324650, 5.7% of the genotypes could not be called (629 genotypes succeeded). SNP rs2061174 (A/G) in intron 4 had an MAF of 0.34. Two SNPs in intron 5, rs324640 (A/G) and rs324650 (A/T), had similar MAFs between 0.48 and 0.49. Observed haplotype frequencies were estimated using haploview 3.11 that implements the EM-algorithm (http://www.broad.mit.edu/mpg/haploview/). Only one twin from each MZ pair was included. LD was calculated from the estimated haplotype frequencies. The two SNPs in intron 5, lying 4 kb apart, are in very strong LD (r2 > 0.90). The SNP in intron 4 lies about 28 kb apart from the SNPs in the downstream region and shows lower LD with the two SNPs in intron 5 (r2 < 0.35).
Genotypic means per cohort are summarized inTables 2a and b. The three SNPs were in Hardy–Weinberg equilibrium in both cohorts as well as in the combined cohort. As the heritabilities were comparable across cohorts, as well as the allele frequencies and the directions of the genotypic effects, we pooled the two cohorts for the association tests.
Table 2a. Means (SD) per genotype for performance IQ (PIQ), verbal IQ (VIQ) and full-scale IQ (FSIQ) in the young cohort
| ||Genotype|| ||Total N|
| Frequency||0.43||0.45||0.12|| |
| Mean PIQ (SD)||100.16 (13.75)||102.20 (12.55)||105.26 (11.57)||381|
| Mean VIQ (SD)||99.81 (18.71)||97.56 (19.75)||101.00 (18.32)||382|
| Mean FSIQ (SD)||99.64 (15.52)||99.93 (15.40)||103.62 (14.16)||381|
| Frequency||0.28||0.50||0.22|| |
| Mean PIQ (SD)||100.60 (14.02)||101.81 (12.82)||102.67 (12.27)||381|
| Mean VIQ (SD)||98.78 (18.82)||98.41 (18.82)||100.38 (18.96)||382|
| Mean FSIQ (SD)||99.18 (16.35)||100.10 (14.67)||101.88 (15.42)||381|
| Frequency||0.29||0.50||0.21|| |
| Mean PIQ (SD)||100.50 (14.15)||101.76 (12.74)||102.96 (12.051)||379|
| Mean VIQ (SD)||97.67 (19.47)||98.94 (18.93)||100.05 (18.81)||380|
| Mean FSIQ (SD)||98.56 (16.11)||100.35 (14.83)||101.89 (15.05)||379|
Table 2b. Means (SD) per genotype for performance IQ (PIQ), verbal IQ (VIQ) and full-scale IQ (FSIQ) in the adult cohort
| ||Genotype|| ||Total N|
| Frequency||0.45||0.44||0.11|| |
| Mean PIQ (SD)||98.59 (12.55)||100.95 (12.79)||102.76 (9.21)||259|
| Mean VIQ (SD)||91.94 (14.08)||94.15 (15.71)||93.62 (9.69)||260|
| Mean FSIQ (SD)||94.06 (12.08)||97.27 (12.67)||96.03 (8.12)||256|
| Frequency||0.26||0.44||0.30|| |
| Mean PIQ (SD)||98.17 (13.39)||100.68 (12.64)||101.09 (11.5)||252|
| Mean VIQ (SD)||91.26 (16.770)||93.28 (13.76)||93.559 (13.33)||251|
| Mean FSIQ (SD)||94.60 (12.23)||95.71 (12.92)||96.17 (10.78)||248|
| Frequency||0.27||0.45||0.29|| |
| Mean PIQ (SD)||98.84 13.07||100.41 (12.78)||100.07 (12.51)||238|
| Mean VIQ (SD)||92.33 16.13||93.22 (13.23)||93.39 (13.70)||239|
| Mean FSIQ (SD)||95.33 11.72||95.48 (12.48)||95.94 (11.09)||235|
The models used in QTDT included the effects of age and sex on the means and modeled additive allelic between- and within-family effects. When testing for the presence of population stratification (i.e. equivalence of the between- and within-family effects), we found significant evidence for the presence of population stratification in the association between rs324650 and both PIQ and FSIQ (P < 0.05), indicating the association effects across the total population are not equal to the association effects as found within families. As it seems obvious that population stratification is caused by pooling the two age cohorts, we also tested for population stratification in each cohort separately and found evidence for population stratification for the same SNPs, in the same direction (i.e. between-family effects smaller than within-family effects), in the young cohort (rs324650 with PIQ: rs324650 with FSIQ: P < 0.05), but not in the adult cohort. The within-family effects were comparable across both cohorts.
Using the within-family association, a significant association of PIQ with all three SNPs was found. The strongest effect was seen with the T allele of rs324650 (P < 0.001), which was associated with an increase of 4.6 IQ points among family members (Table 3). Put otherwise, the difference between AT and TT genotypes or AT and AA is 4.6, whereas the difference between AA and TT genotypes is estimated at 9.2.
Table 3. Tests for genetic association between the CHRM2 gene and intelligence
| ||Within-family association||Population-based association|
| ||N||χ2 (nominal P-value)||Genotypic effect (increaser allele)||N||χ2 (nominal P-value)||Genotypic effect (increaser allele)|
| PIQ||175||7.3 (<0.01)||3.7 (G)||648||9.0 (<0.01)||2.4 (G)|
| VIQ||175||0.4||1.2 (G)||648||0.0||0.2 (G)|
| FSIQ||174||2.4||2.3 (G)||644||2.3||1.4 (G)|
| PIQ||209||7.7 (<0.01)*||3.7 (G)||640||5.2 (<0.05)*||1.8 (G)|
| VIQ||209||1.9||2.5 (G)||640||1.2||1.1 (G)|
| FSIQ||207||4.6 (<0.01)||3.1 (G)||636||2.9||1.4 (G)|
| PIQ||193||12.1 (<0.001)*||4.6 (T)||629||6.0 (<0.05)*||1.9 (T)|
| VIQ||193||3.9 (<0.05)||3.6 (T)||629||1.8||1.4 (T)|
| FSIQ||191||8.0 (<0.01)*||4.1 (T)||625||4.0 (<0.05)*||1.7 (T)|
Notably, the effect sizes of the increaser alleles were all reduced, although still significant, in the total association test as compared with the effect sizes based solely on the within-family association. As within-family associations are not sensitive to spurious associations due to population stratification, whereas between-family associations are, this means that stratification acted to hide a true association.Figure 2 shows the observed mean difference in IQ points between different genotypes for individuals within families, for associations significant at the 0.01 level.
Figure 2. Observed mean difference in IQ scores between siblings (i.e. within family pairs) with different genotypes for those single-nucleotide polymorphisms (SNPs) in the CHRM2 gene that show a significant association. Sibling pairs include dizygotic pairs and non-twin sibling pairs. The number of pairs on which the difference scores are based is: rs2061174: 97 (AG-AA), 64 (GG-AG) and 15 (GG-AA); rs324640: 87 (AG-AA), 97 (GG-AG) and 9 (GG-AA); rs324650: 75 (AG-AA), 98 (GG-AG) and 11 (GG-AA). FSIQ, full-scale IQ; PIQ, performance IQ.
Download figure to PowerPoint
- Top of page
- Materials and methods
To investigate the possible role of the CHRM2 gene in intelligence, we employed a family-based genetic association test. Significant evidence was found for an association between the CHRM2 gene and IQ, showing an effect size of 4.6 IQ points for the increaser allele of SNP rs324650 (P < 0.001).
We also found that the effect sizes based on the within-family effects were 1.5-2.5 times as large as the population-based effect sizes, suggesting that population stratification resulted in an underestimation of the genuine allelic effect.
The attenuation of allelic effects due to population stratification occurs when across subpopulations higher trait values tend to go together with a lower frequency of the increaser allele, or vice versa. This is consistent with findings from mouse model systems in which it has been shown that the same allele at the same locus may cause a major disease in one mouse strain, but no disease phenotype in a strain with a different genetic background (e.g.Linder 2001; Liu et al. 2001; Montagutelli 2000). The same has been reported for effects on gene expression in different environmental backgrounds (Cabib et al. 2000; Crabbe et al. 1999). In humans, the presence of different genetic (or environmental) backgrounds that derive from mixed strata may differentially affect the expression of gene variants (G × E interaction). As non-Mendelian traits are likely to be influenced by multiple (risk-) factors which in turn are likely to interact with each other, neglecting the presence of population stratification may realistically hide genuine allele-trait associations, and may be responsible in part for the difficulties in replicating reported associations. We previously predicted this phenomenon based on theory and simulations (Posthuma et al. 2004) and now show it to occur in practice as well.
The most significant association was seen with rs324650 where the presence of the T allele was associated with an increase of 4.6 points for PIQ. This extends earlier findings of Comings et al. (2003) who found a weaker association between the CHRM2 gene and both total IQ score and years of education, which was only significant after stratifying on the parental origin of transmission. The current study therefore looked at different SNPs than the SNP that was used by Comings et al. (2003). Their SNP was in the 3′ UTR region and is not classified as a tagging SNP. The SNPs used in the current sample are tagging SNPs, two of which (rs324640 and rs324650) are in LD with the SNP used in the study by Comings et al. (2003). These two SNPs are also the ones that show a significant association with IQ in the current study, albeit stronger than in the study by Comings et al. (2003). This study thus provides further evidence of a role of the cholinergic muscarinic receptor in cognition.
The M2 subtype cholinergic muscarinic receptor is, like the M1 and M4 subtypes, predominantly expressed in the CNS (Volpicelli & Levey 2004). The M2 receptors are predominantly located at the presynaptic level (Levey et al. 1991; Mrzljak et al. 1993), spread throughout the brain but with the highest levels in the cerebral cortical, forebrain cholinergic nuclei, cervical spinal cord region, cerebellum and thalamus (Flynn & Mash 1993; Piggott et al. 2002; Spencer et al. 1986; Wei et al. 1994). M2 receptors are selectively coupled to G-proteins of the Gi/Go family, which mediate the inhibition of voltage-sensitive Ca2+ channels. Furthermore, the M2 receptor subtype is likely to have an additional role in cholinergic modulation of excitatory and inhibitory hippocampal circuits acting as autoreceptor (Akam et al. 2001; Kitaichi et al. 1999a, 1999b; Rouse et al. 2000; Shapiro et al. 1999; Zhang et al. 2002), inhibiting ACh release from cholinergic terminals.
The association found in the present study as well as by Comings and coworkers was all with SNPs in non-coding regions of the gene. We found association with SNPs located in intron 4 (rs2061174) and intron 5 (rs324640 and rs324650) of the CHRM2 gene. Comings et al. (2003) found an association between the CHRM2 gene and IQ with an SNP in the 3′ UTR of the gene. The transcription of the CHRM2 gene is complex. Krejci et al. (2004) determined that the 5′ UTR of CHRM2 consists of four non-coding regions whose different combinations give rise to eight splice variants. In addition, expression is regulated by two promoters. One promoter regulates expression at the cardiac cell level, whereas the second promoter could be considered neuron specific. Experiments using reporter genes demonstrated that additional regulatory sequences are present further upstream of the proximal promoter(s) and, even more interesting, within the intronic regions (Krejci et al. 2004).
Identifying genes for variation in the range of normal intelligence could provide important clues to the genetic etiology of disturbed cognition in, for example, autism, reading disorder and attention deficit and hyperactivity disorder (ADHD). It is worth to mention the first genome-wide linkage screen in autism, performed by the International Molecular Genetic Study of Autism Consortium, involving sib pairs from the United Kingdom. Interestingly, the strongest linkage signal for autism occurred at 7q near the CHRM2 gene (for a review on linkage scans for autism, see Wassink et al. 2004). With regard to attention problems, it is of note that a number of recent findings clearly implicate deviant ACh neurotransmission in attentional processing (Beane & Marrocco 2004; Yu & Dayan 2005). Thus far, candidate gene approaches for attention disorder have focused only on genetic variation in nicotinergic receptors (Sacco et al. 2004; Todd et al. 2003). The results in the current study tentatively suggest that muscarinergic signaling may be involved as well.