Introduction

Working memory (WM) is a fundamental component of human intelligence.1, 2 It refers to the processes that support the short-term maintenance or manipulation of relevant information in the presence of distracting irrelevant information. Nonhuman primates physiology and human functional imaging studies support a critical involvement of the prefrontal cortex (PFC) in WM.3, 4 The PFC is the primary target of extensive dopamine (DA) projections from the midbrain, and several lines of evidence suggest that the frontal DA level is a critical modulator of WM performance.1, 5 Hence, genes involved in dopaminergic pathway metabolism have been of particular interest to explain individual differences in WM performance.6, 7, 8, 9, 10, 11 Amongst these, the catechol-O-methyltransferase (COMT) gene has been studied the most extensively. The COMT gene is located on chromosome 22q11, and contains six exons.12 It is involved in enzymatic activity that degrades DA, norepinephrine, and epinephrine.13 Two promoters encoding different isoforms, a membrane-bound COMT (MB-COMT) and a soluble COMT (S-COMT), are known. Both transcripts start at exon 3.14 Differential expression and activity profiles have been well characterized, MB-COMT is predominantly expressed in brain neurons,15 whereas S-COMT is predominantly expressed in other tissues, such as liver, blood, and kidney.14 The human MB-COMT plays a crucial role in regulation of DA signaling at the PFC level. It contains a common functional single-nucleotide polymorphism (SNP) (rs4680) that substitutes a Val for a Met residue at codon 158 (see Figure 1).

Figure 1
figure 1

COMT gene and functional single-nucleotide polymorphism (SNP) rs4680 on 22q11.21.

The Met allele encodes an enzyme with relatively lower activity,16 and is thought to be specific to humans; as no equivalent polymorphism has been found in any other species.17 Decrease in enzyme activity present in Met/Met individuals, compared to individuals homozygous for the Val allele, leads to a relatively higher DA availability, whereas Met/Val heterozygous display an intermediate enzyme activity.18 Because frontal DA level is a critical modulator of WM processes, the decreased COMT activity of Met carriers might be beneficial to their cognitive performance. In line with this, Savitz et al10 found that 20 of the 26 studies on the association between the COMT Val108/158Met polymorphism and cognitive function reported a significant association. All but two of these studies suggested that the low-activity Met allele yields better performance on cognitive tasks that have a WM component. However, these studies were often based on small and/or clinical samples (ADHD in children; schizophrenia in adults) and no significant association to WM was found in a much larger sample of healthy adult males.19 In addition, Mattay et al20 have shown that the role of the COMT Val108/158Met polymorphism in PFC function, particularly in WM performance, is not straightforward. Although homozygous individuals for the Met allele perform significantly better than individuals homozygous for the Val allele, when the Met homozygotes are given DA agonists their response actually deteriorates. In contrast, the response of the ‘dopamine-poor’ Val homozygotes improves with DA agonists. This suggests that the relation between DA availability at the PFC level as indexed by COMT activity and WM performance is not linear, but instead follows an inverted-U shape.

DA signaling, furthermore, is not only dependent on the availability of DA, but also on the efficiency of the DA receptor and its downstream signaling cascade. Because of their importance in reward processing, dopaminergic receptors, in particular the DA D2 receptor gene (DRD2), have been studied extensively in addiction research.21 The DRD2 gene is located on chromosome 11 at q22-q23 (see Figure 2). A DRD2 Taq IA variant, a restriction fragment length polymorphism (RFLP), located on the 3′ untranslated region (3′UTR) of the DRD2 gene, is associated with altered receptor density.22 Individuals with the A1 allele show a 30–40% reduction in D2 DA receptor density compared with those homozygous for the A2 allele.22, 23

Figure 2
figure 2

DRD2 gene and tagging SNP rs2075654 near the restriction site Taq A1 on 11q23. * indicates splicing site at exon 6.

Recently, Reuter et al24 conducted an association analysis using an adult cohort enriched for COMT and DRD2 homozygotes. They found a significant interaction between DRD2 Taq IA and COMT Val108/158Met polymorphisms, and performance on response interference on the Stroop color–word conflict task. Met homozygotes performed better than Val allele carriers, but only if they had the DRD2 genotype associated with low receptor density. In fact, Met homozygotes also bearing two DRD2 A2 alleles showed a significantly worse performance compared to all other genotypes. Although response interference and WM are not unitary constructs, we showed a significant correlation (r=−0.26, P<0.05) between these two measures of PFC function.25 Hence, we hypothesize that a COMT by DRD2 interaction may also be found for WM performance.

In the present study, which included 762 genotyped subjects, from two independent family-based Dutch samples of 371 (mean age 12.4 years) and 391 (mean age 36.2 years) subjects, respectively, our principal goal was to test for association of the COMT Val108/158Met polymorphism with WM performance. The use of a family-based sample made it possible to test for association in a combined within- and between-family design to estimate genetic effects, which are free from spurious effects of population stratification.26 As a secondary analysis, we tested for an interactive effect of the COMT Val108/158Met polymorphism and genetic variation in the DRD2 gene on WM performance comparable to the effect reported by Reuter et al24 for Stroop interference. Because DA receptor sensitivity has been shown to decline with aging in both animal27 and human studies,28 our analyses will allow this interaction to be different in children and adults.

Materials and methods

Subjects

All twins and their siblings were part of two larger cognitive studies and were recruited from the Netherlands Twin Registry.29 Informed consent was obtained from the participants (adult cohort) or from their parents if they were under 18 years of age (young cohort). The current study was approved by the institutional review board of the VU University Medical Center. None of the individuals tested suffered from severe physical or mental handicaps, as assessed through standard questionnaire.

Young cohort

The young cohort consisted of 177 twin pairs born between 1990 and 1992, and 55 siblings,30, 31 of which 371 were available for genotyping. The genotyped twins were 12.4 (SD=0.9) years of age and the siblings were between 8 and 15 years of age at the time of testing. There were 35 monozygotic male twin pairs (MZM), 28 dizygotic male twin pairs (DZM), 48 monozygotic female twin pairs (MZF), 23 dizygotic female twin pairs (DZF), 26 dizygotic opposite-sex twin pairs (DOS), 24 male siblings and 24 female siblings, and 3 subjects form incomplete twin pairs (1 male, 2 females). Participation in this study included a voluntary agreement to provide buccal swabs for DNA extraction.

Adult cohort

A total of 793 family members from 317 extended twin families participated in the adult cognition study.32 Participation in this study did not automatically include DNA collection, however, part of the sample, 276 subjects returned to the lab to provide blood samples, 115 provided buccal swabs through the biobanking project33 for DNA extraction. Mean age of the genotyped sample was 36.2 years (SD=12.6). There were 25 monozygotic male twin pairs (MZM), 15 dizygotic male twin pairs (DZM), 1 DZM triplet, 20 monozygotic female twin pairs (MZF), 28 dizygotic female twin pairs (DZF), 23 dizygotic opposite-sex twin pairs (DOS), 29 female siblings, 28 male siblings, and 109 subjects from incomplete twin pairs (41 males, 68 females).

Cognitive testing

WM tasks were assessed in the young cohort, using the Dutch adaptation of the Wechsler Intelligence Scale for Children-Revised (WISC-R)34 consisting of two subtests: arithmetic and digit span. WM performance was indexed as the sum score of the two subtests and corrected for age and sex. The Dutch adaptation of the Wechsler Adult Intelligence Scale III-Revised (WAIS-IIIR)35 was used to assess WM performance in the adult cohort and consisted of two subtests taxing WM (arithmetic and letter–number sequencing). WM was indexed as the sum score of arithmetic and letter-number sequencing and corrected for age and sex.

DNA collection and genotyping

DNA isolation from buccal swabs was performed using a chloroform/isopropanol extraction36 DNA was extracted from blood samples using the salting out protocol.37 Zygosity was assessed using 11 polymorphic microsatellite markers (Het>0.80). Genotyping was performed blind to familial status and phenotypic data. Both MZ twins of a pair were included, serving as additional quality control on genotyping. COMT genotyping was performed using fluorogenic probes in the high-throughput 5′-nuclease assay and following manufacturer's recommendations (TaqMan, PE Applied Biosystems, Foster city, CA, USA). For DRD2, instead of the A1 allele of the Taq IA polymorphism, a tag-SNP (rs2075654) lying 18 kb downstream of the Taq IA variant was genotyped. LD between rs2075654 and Taq IA was calculated using the CEPH population, which is presumably of similar genetic ancestry to the Dutch population. LD between these two polymorphisms was high (r2=0.65, LOD score 14.14). In view of the LD between the tag-SNP and Taq IA, we will refer to the T allele as ‘A1’. DRD2-SNP genotyping was performed as part of a SNPlex assay, which included multiple other genes, following a tagging approach.38 The SNPlex assay was conducted following the manufacturer's recommendations (Applied Biosystems). Here we focus on the DRD2 gene only. Results on cognitive effects of two other genes are described elsewhere.39, 40

Statistical analyses

Allele frequencies of the COMT Val108/158Met and DRD2 A1/A2 polymorphisms were estimated in both cohorts using Haploview (http://www.broad.mit.edu/mpg/haploview), in which a Hardy–Weinberg test is implemented, based on an exact calculation of the probability of observing a certain number of heterozygotes conditional on the number of copies of the minor SNP allele. Family-based genetic association tests were conducted using the program QTDT (http://www.sph.umich.edu/csg/abecasis/QTDT/), which implements the orthogonal association model proposed by Abecasis et al41 (see also Fulker et al42 extended by Posthuma et al43) This model allows one the decomposition of the genotypic association effect into orthogonal between- (βb) and within- (βw) family components and can incorporate fixed effects of covariates and can also model the residual sib-correlation as a function of polygenic or environmental factors. MZ twins can be included and are modeled as such, by adding zygosity status to the data file. They are not informative to the within-family association component (unless they are paired with non-twin siblings) but are informative for the between family component. The between-family association component is sensitive to population admixture, whereas the within-family component is significant only in the presence of true association. Testing for the equality of the βb and βw effects serves as a test of population stratification. If population stratification acts to create a false association, the test for association using the within-family component (βb) is still valid and provides a conservative test of association. If this test is not significant, the between- and within-family effects are equal and the more powerful association test that uses the whole population at once can be applied. The residual sib correlation was modeled as a function of residual genetic variance and non-shared environmental variance. The DRD2 genotype was recoded into carriers (A1+) versus non-carriers (A1−). We used one-sided hypothesis testing for the interaction effects as our hypotheses specify the direction of genetic effects.

Results

In total, 762 subjects were available for SNP genotyping. Based on blind controls and intrapair MZ comparisons, a low genotyping error rate was found (0.015%). For the total sample, the success rate was 98.5 and 93.4% for the COMT Val108/158Met polymorphism in the young and adult cohort, respectively. For DRD2 rs2075654, success rates were 97 and 100%, 365 adults and 360 children had genotype data for both COMT and DRD2. The distribution of genotype and allele frequencies of the COMT and DRD2 polymorphisms as well as means, standard deviations, and standard errors for WM performance are provided in Table 1. Phenotypic means are provided for the complete phenotypic sample as well as for the genotyped subsamples.

Table 1 Means and standard deviations WM scores for COMT and DRD2 genotypes

Stratification

Tests for the presence of population stratification were not significant at the 0.05 level (COMT: χ2(1360)=0.937, P=0.33; χ2(1360)=0.23, P=0.64; χ2(1724)=1.516, P=0.22; for children, adults and the combined sample, respectively) (DRD2: χ2(1355)=0.220, P=0.64; χ2(1388)=0.00, P=0.9976; χ2(1748)=0.084, P=0.77; for children, adults and the combined sample, respectively), indicating that genotypic effects within families were not significantly different from those observed between families, suggesting that the more powerful population-based association test can be meaningfully interpreted for both COMT and DRD2.

COMT polymorphism

WM performance was compared across the three possible genotype groups (Met/Met, Met/Val, and Val/Val) suggesting a positive heterosis pattern in both the young and the adult cohorts (see Table 1). Heterosis refers to a situation in which a given trait is significantly greater (or lesser) in individuals heterozygous at a specific gene marker than those homozygous for either allele. We tested for heterosis by adding a non-additive (dominance) genetic component to the population-based analysis in QTDT and testing whether the heterozygous genotypes were associated with better WM performance. A significant heterosis effect was found for the association between COMT Val108/158Met polymorphism and WM in the adult cohort 2(1360)=4.80, P=0.014) even after correction for multiple testing. In the young cohort, the association did not reach significance 2(1360)=1.54, P=0.107), although the effects were in the same direction as in the adult cohort. The strongest effect was found in the combined sample (χ2(1724)=5.70, P=0.008).

DRD2 polymorphism

No significant main effect was found on the DRD2 rs2075654 tag-SNP and WM performance (χ2(1,356)=0.42 P=0.52; χ2(1,391)=0.04 P=0.84; and χ2(1749)=0.10 P=0.75, for children, adults, and the combined sample, respectively).

COMT and DRD2 interaction

Figure 3 plots WM performance against six possible combined genotype groups (Met/Met,A−; Val/Met,A−; Val/Val,A−; Met/Met;A+; Val/Met,A+ and Val/Val,A+). The figure suggests that the heterosis found for the COMT gene is entirely limited to subjects with the DRD2 A+ genotype, the genotype previously linked to the reduced receptor density. In support of this, a significant interaction effect was detected between the DRD2 and COMT polymorphisms in the combined sample (χ2(1699)=2.72, one-sided P=0.050), which seemed confined to the adult cohort (see Table 2).

Figure 3
figure 3

Working memory (WM) means plotted for the six combined genotype groups of the COMT and DRD2 genes for young (left panel) and adult (right panel) cohort. Bars denote standard error. Note: A+ denotes carriers of the A1 allele, whereas A- denotes non-carriers.

Table 2 Results of population-based genegene interaction analysis – for COMT Val108/158Met, and DRD2 rs2075654 (tagging the TaqA1 polymorphism) for the young, adult, and combined cohorts

Discussion

In the present study, we tested the association of the COMT Val108/158Met polymorphism with WM performance. A significant association was found in the combined sample with stronger contribution from the adult than the young cohort. The association reflected positive heterosis such that the Met/Val heterozygotes performed better than both Met/Met and Val/Val homozygotes on the WM tasks. An age-dependent positive heterosis pattern has previously been reported in a longitudinal study by Harriset al.44 The COMT genotype was not associated with childhood intelligence measured at the age of 11 years in the Scottish Mental Survey of 1932. At the age of 79 years, COMT genotype was significantly related to differences in verbal declarative memory and to scores on the personality traits of intellect/imagination. For both traits, the elderly Val/Met heterozygotes had higher scores than both homozygous groups, which echo the pattern of heterosis on WM found in adult cohort. Because the COMT polymorphism has been hypothesized to have a nonlinear effect on DA availability in the prefrontal cortex,20 the finding of heterosis is in keeping with the idea that the relationship between DA signaling and cognitive performance follows an inverted U-shaped curve, with both suboptimal and supraoptimal DA activity, impairing prefrontal function.45 Burst firing of VTA neurons causes synaptic DA release in pyramidal cells in the PFC. Because these cells contain little DA transporter (DAT), most DA diffuses out of the synaptic cleft to bind to extrasynpatic D1 receptors, where it is inactivated by COMT.46 The higher activity Val allele decreases extrasynaptic DA levels and, therefore, D1 activation, shifting the balance in favor of intrasynaptic D2 receptor activation.46, 47 Cognitive performance may be critically dependent on the D1/D2 binding ratio, with a relative lack of D1 signaling causing impulsivity, distractibility, and poor WM performance with schizophrenia at the extreme end.47 A relative lack of D2 signaling, on the other hand, may fail to signal the presence of reward information, a signal that is required to engage the PFC in updating its WM system.48

The above suggests that the optimum level of DA signaling depends not simply on frontal DA availability, but on its combination with D2 receptor sensitivity. Therefore, individual differences in DA availability as well as D2 receptor sensitivity may come into play during the performance of WM tasks. We tested this expectation in a secondary analysis in which a DRD2 tagging SNP (rs2075654) was tested for an interactive effect with the COMT polymorphism. No significant main effect on WM was found for the rs2075654 tag-SNP in DRD2. However, in the combined cohort, the DRD2 and COMT polymorphisms had a significant interactive effect on WM performance. The interaction suggested that the Met/Val heterozygotes perform better than both Met/Met and Val/Val homozygotes only when they carry one or two A1 alleles. The A1 alleles have been associated with lower receptor density, suggesting that the U-curve-shaped effect of DA availability on WM performance disappears when receptor density is high. Such a pattern has been previously reported by Reuter et al,24 who reported a significant interactive effect between the DRD2 Taq IA and the COMT polymorphisms on the amount of response interference in the Stroop color–word conflict task. Inspection of Figure 3 suggests that the interaction is stronger in the adult than in the young cohort. Although p-values in neither cohort reach formal significance levels, this age difference may be real. Evidence for age-related changes regarding DA metabolism within the PFC has been postulated in both animal27 and human studies,28 with increased DA metabolism (eg, MAO, COMT) thought to be present at a more mature age.49 Furthermore, several lines of evidence showed a decrease of DA receptors with age.50, 51, 52 When age-related changes in overall levels of DA availability and DA receptor sensitivity are superimposed on the influence of genetic polymorphisms on these levels, a different change in overall DA signaling may occur with age in the various COMT–DRD2 haplotypes. When we add to this that there may be an inverted U-curved relation between DA signaling and WM performance, interactions between age, DRD2, and COMT genes should be the rule rather than the exception.

Clearly, full genetic contribution to dopaminergic variation in frontal executive function will rely on far more complex interactions between multiple receptor (eg, DRD1, DRD2, and DRD4), transporters and enzymatic polymorphisms (eg, DAT, COMT, and MAO).9, 53, 54 Further studies systematically involving such interactions are needed to obtain a clearer overview of the dopaminergic pathway. At the same time, denser SNP coverage of the area under study is needed in genes like DRD2 to reveal the true functional variants, which while tagged, are still undiscovered. In summary, our results are in keeping with previous findings suggesting that WM performance needs an optimal level of DA signaling within the PFC. This optimum level depends on enzymatic activity controlling DA level as well as on DA receptor sensitivity, both of which may differ as a function of age and genotype. We conclude that the effects of a single polymorphism in a dopaminergic gene on a well-defined cognitive trait may easily remain hidden if the interaction with age and other genes in the pathway are not taken into account.