The molecular genetics of cognition: dopamine, COMT and BDNF


* J. Savitz, MRC/UCT Human Genetics Research Unit, University of Cape Town Medical School, Observatory, 7925, South Africa. E-mail:


The important contribution of genetic factors to the development of cognition and intelligence is widely acknowledged, but identification of these genes has proven to be difficult. Given a variety of evidence implicating the prefrontal cortex and its dopaminergic circuits in cognition, most of the research conducted to date has focused on genes regulating dopaminergic function. Here we review the genetic association studies carried out on catechol-O-methyltransferase (COMT) and the dopamine receptor genes, D1, D2 and D4. In addition, the evidence implicating another promising candidate gene, brain-derived neurotrophic factor (BDNF) in neuropsychological function, is assessed. Both the COMT val158met polymorphism and the BDNF val66met variant appear to influence cognitive function, but the specific neurocognitive processes involved continue to be a matter of debate. Part of the difficulty is distinguishing between false positives, pleiotropy and the influence of a general intelligence factor, g. Also at issue is the complexity of the relevant neuromolecular pathways, which make the inference of simple causal relationships difficult. The implications of molecular genetic cognitive research for psychiatry are discussed in light of these data.

The genetics of cognition and intelligence have long fascinated both the scientific community and society at large. Ever since Francis Galton's famous study of eminent families, the pièce de résistance of the 1865 ‘Hereditary Talent and Character’, controversy has raged over the role hat genes play in the development of the intellect. While Galton's conceptualization of intelligence as an instinct (see Plotkin 1997) is no longer tenable, most modern studies acknowledge the importance of genetic factors, with heritability estimates of standardized intelligence quotient (IQ) tests converging on the 50–80% range (Bouchard 1998). Discrete (as opposed to global) cognitive processes such as processing speed, short-term and working memory have also been shown to be significantly influenced by genes with heritability values ranging from 30 to 60% (Luciano et al. 2001).

Cognition is a complex trait and is therefore likely to be underpinned by many genes, each with a relatively small effect. To reduce the complexity involved, most investigators have attempted to fractionate global cognitive function into discrete cognitive processes via neuropsychological investigations. Performance in these specific domains can then be statistically linked to the functional activity of particular proteins, and by extension to the genetic variants accounting for these functional differences. This is known as an association study and accounts for a major component of genetic research into complex traits like cognition. Genes are usually considered good contenders for involvement in the genesis of a particular phenotype, if they are functionally polymorphic and have some prima facie physiological relevance to the cognitive trait under consideration (Glatt & Freimer 2002). These criteria will inform our evaluation of two of the most promising genes implicated in neurocognitive function, catechol-O-methyltransferase (COMT) and brain-derived neurotrophic factor (BDNF). Given the pivotal role of dopamine (DA) in the functioning of the prefrontal cortex (PFC) (vide infra), the burgeoning database of genetic association studies involving DA receptor genes will also be covered here.

Dopamine and the prefrontal cortex

Our linguistic and cognitive abilities are the characteristics that set us apart from other primates, and these distinctive functional differences presumably have neurological correlates. Rilling and Insel (1999), in a magnetic resonance imaging (MRI) investigation, compared neuroanatomical data across 11 different species of primates. Humans were found to have significantly larger neocortices and had significantly more gyrified (a measure of cortical folding) prefrontal cortices than expected for a non-human primate of the same size (Rilling & Insel 1999). Secondly, a plethora of data from neuropsychological studies of brain-injured patients and more recently neuroimaging work has indicated that the PFC is critical for ‘executive functioning’, a miscellaneous category of cognitive processes that includes working memory, response inhibition, planning, concentration, attention, perceptual organization, judgement, decision making and self-monitoring (Benton 1994).

Genes and gene families that are expressed in the PFC are therefore obvious candidates for involvement in cognition. Unfortunately, with approximately half of the 32 000 Homo sapiens genes expressed in the brain and significant portion of these genes expressed in the PFC, narrowing down the list of candidates becomes a speculative exercise. Previc (1999) has hypothesized that DA is the key regulator of six predominantly left-hemisphere cognitive skills – the ‘executive functions’ that are critical to human language and thought. In Previc's schema, physiological adaptations to ecological changes in sub-Saharan Africa catalysed the expansion of dopaminergic pathways and with it our ascending intellectual abilities. While this hypothesis may be speculative, the evidence linking DA to prefrontal function and cognition is well established (Cools & Robbins 2004). Given the pivotal role of DA in the mediation of PFC function (vide infra) and the empirical evidence garnered from genetic association studies, we have in the first part of the paper focussed our attention on three candidates, COMT and the DA receptors, D2 and D4.

One of the first clues that DA is involved in cognition came from studies of patients with Parkinson's disease (PD). Working memory and executive type deficits are often present in the early stages of PD (Cools & Robbins 2004). Dubois and Pillon (1997) presented data indicating that about 93% of PD patients present with cognitive – mostly executive – deficits. Administration of L-DOPA to PD patients has been reported to improve working memory (Mattay et al. 2002), cognitive flexibility (Cools et al. 2001) and planning as evinced by the Tower of London (TOL) (Cools et al. 2002). In addition, the neurotoxin 1-methyl-4-phenyl-1,2,3,6-tetrohydropyridine, which results in a degeneration of the nigrostriatal pathway, produces similar executive dysfunction to PD (Stern & Langston 1985).

The second strand of evidence implicating DA in cognition comes from animal studies. Non-human primate studies of working memory make widespread use of the delayed response task. The animal is shown the location of a piece of food, the food is then hidden from view by an opaque screen, and the primate must then choose the correct location of the treat from two or more options (Goldman-Rakic 1995). Working memory deficits on this task were shown to result from experimental depletion of PFC DA in rhesus monkeys (Brozoski et al. 1979), and more contemporary studies have shown that deficits on tests of working memory can be reversed with DA agonists (Goldman-Rakic 1995). In addition, spatial working memory tasks have been shown to increase dopaminergic activity in the dorsolateral PFC of primates (Kodoma et al. 1997).

In human studies, Luciana et al. (1998) improved the working memory performance of individuals by using the DA agonist bromocriptine, while set-shifting ability was reduced by DA antagonists (Lange et al. 1992). Ageing populations with decreased PFC DA levels and cognitive inflexibility have also been shown to benefit from the administration of DA agonists (Ollat 1992). More contemporary neuroimaging studies have demonstrated a correlation between the density of dopaminergic fibres in the caudate nucleus and performance on a range of executive and memory tests (Remy & Samson 2003). Recent data suggest, however, that there is not a simple linear relationship between DA and cognition.

Dopamine does not always enhance cognitive performance and may in fact retard it (Cools & Robbins 2004). Bromocriptine, a DA D2 receptor agonist, was demonstrated to improve cognition in individuals with a low-baseline memory capacity but impair performance in high performers (Kimberg et al. 1997). Similarly, Mattay et al. (2000) showed that dextroamphetamine, a drug that potentiates dopaminergic activity, improved cognitive performance in subjects with low-baseline levels of DA but that once DA receptor stimulation reached a certain threshold, working memory function deteriorated. Thus, an inverted ‘U’ shaped curve characterizes the relationship between DA levels and cognitive performance, with both suboptimal and supra-optimal DA activity impairing prefrontal function (Cools & Robbins 2004).

These hypothesized effects may be mediated by the differential effects of D1 and D2 receptor binding. Theoretically, D2 receptor binding signals the presence of important (often reward-based) information and allows the PFC network to respond to this new information by updating its working memory system (Weinberger et al. 2001). D1 receptor stimulation, on the other hand, plays a gating role by controlling the threshold of significance above which information must pass before it can be admitted to working memory and processed by the PFC. D1 receptor activation thus stabilizes the pattern of activity within the PFC neural network and protects the system from distracters until the appropriate behavioural response is generated (Weinberger et al. 2001).

When the D1/D2-binding ratio is disturbed, cognitive performance is theoretically impaired in one of two directions: (a) A lack of D1-mediated signalling in the PFC (a hypodopaminergic state) theoretically causes neural representations to be vulnerable to decay and competing inputs, resulting in poor differentiation of target from background (Winterer & Weinberger 2004). This may be manifested in impulsive and distractible behaviours and poor working memory and planning, abilities which are mediated by the PFC. Attention is drawn to contextually weak stimuli, leading perhaps in extreme cases to the delusions, paranoia and neurocognitive dysfunction characteristic of schizophrenia (Winterer & Weinberger 2004). (b) Hyperdopaminergic states associated with excessive D1 stimulation, on the other hand, may allow representations in the PFC to be maintained but not effectively updated, leading in extreme cases to perseverative or stereotypic behaviour (Weinberger et al. 2001) and reduced performance on tasks requiring cognitive flexibility (Cools & Robbins 2004).

Given the delicate balancing act that the regulators of PFC DA activity need to achieve for optimal cognitive function, these enzymes make excellent candidate genes for influencing inter-individual variation in cognition. Dopamine catabolism in the PFC is regulated by monoamine oxidase which is found on the mitochondrial membranes of catecholamine neurons and COMT which is concentrated in the extrasynaptic spaces (Deutch & Roth 1999). The dopamine transporter (DAT) is found primarily in dopaminergic neurons of the substantia nigra and ventral tegmental area (VTA) and allows for the rapid reuptake of DA from the synaptic cleft (Deutch & Roth 1999).

While all three of these enzymes are important regulators of subcortical and cortical DA transmission, COMT appears to play the pivotal role in the modulation of fronto-striatal networks. An evolutionary recent functional single nucleotide polymorphism (SNP) of the COMT gene (val158met) located on chromosome 22q11 that results in the substitution of valine (val) with methionine (met) at codon 158 of the protein sequence has been described (GenBank accession no. Z26491). The met allele produces an enzyme that is unstable at body temperature and has only a quarter the activity of the val-containing polypeptide (Egan et al. 2001).

In the PFC, burst firing of VTA neurons causes synaptic DA release in pyramidal cells, but these cortical neurons contain very little DAT (Dougherty et al. 1999), allowing high levels of DA to diffuse out of the synaptic cleft and bind to extrasynaptic D1 receptors where the neurotransmitter is inactivated principally by COMT (Bilder et al. 2004). The high-activity val allele decreases extrasynaptic DA concentrations and therefore D1 activation, shifting the balance in favour of intrasynaptic D2 receptor activation (Bilder et al. 2004; Winterer & Weinberger 2004).

Given the hypothetical differential effects of D1 and D2 receptor stimulation discussed above, the effect of COMT genotype on PFC-mediated cognitive functions may be as follows: The val (high-activity) allele theoretically results in lowered stability of PFC neural networks which may manifest in decreased ability to maintain information in working memory (Bilder et al. 2004; Winterer & Weinberger 2004) but enhanced ability to update the contents of working memory with new information or switch cognitive sets (Cools & Robbins 2004). The met allele, which is associated with an elevated PFC DA level and increased D1 receptor binding, is predicted to enhance the stability of PFC networks and thus the performance of cognitive tasks involving maintenance of information, but lead to decreased cognitive flexibility (Bilder et al. 2004; Winterer & Weinberger 2004).

This hypothesis offers an elegant model of PFC modulation and cognition, but are these theoretical speculations congruent with empirical data linking genotype with cognitive performance? We are aware of 26 studies that have examined the association between the COMT val158met polymorphism and cognitive function, and these data are summarized in Table 1. The results are impressive even if one acknowledges the possible effects of publication bias. Twenty out of 26 studies report an association between the COMT val158met polymorphism and cognitive function, and all of them bar two suggest that the low-activity met allele allows for better performance on cognitive tasks that have a working memory component.

Table 1.  Association between the Catechol-O-methyltransferase (COMT) val158met polymorphism and cognitive performance
StudySample (mean age)EthnicityCognitive tasksOutcome
  1. ADHD, attention deficit hyperactivity disorder; C, controls; CANTAB, Cambridge neuropsychological test automated battery; CPT, continuous performance test; DSST, digit symbol substitution test; ERP, event related potential; MARS, maudsley attention and response suppression battery; MFFT, matching familiar figures test; MMSE, mini mental state examination; RCF, rey complex figure; SOPT, self-ordered pointing task; TEA-Ch, tests of everyday attention for children; TMT, trail-making test; TOL, tower of London; WAIS, Wechsler adult intelligence test; WCST, Wisconsin card sorting test; WISC, Wechsler intelligence scale for children; WMS, Wechsler memory scale.

Eisenberg et al. (1999)48 ADHD triads (9.4)Ashkenazi and
non-Ashkenazi jews
CPTMet/met subjects made less errors of commission (false alarms) on the CPT, indicating lower levels of impulsivity
Egan et al. (2001)175 schizophrenics (36), 219 unaffected siblings and 55 C (34)Caucasian (USA)WCSTValine allele is associated with a greater number of perseverative errors on the WCST and reduced efficacy of prefrontal function during a working memory task
Bilder et al. (2002)58 individuals with chronic schizophrenia (40.5)Mixed (USA)Executive (digit-span, WCST, letter fluency, block design, visual reproductions); declarative memory (paragraph recall, word-list learning); processing speed and attention (TMT A + B, DSST); simple motor (tapping)Met allele associated with better performance in the processing speed and attention domain
Fossella et al. (2002)220 healthy adults (18–50)Not stated (USA)Attention network test (ANT)There was a non-significant trend towards higher executive attention scores in the met/met homozygotes
Joober et al. (2002)94 schizophrenics (not given) + 31 C (not given)Not stated (Canadian)WCSTMet/met homozygotes performed better on the WCST compared with heterozygotes or val/val subjects
Malhotra et al. (2002)73 healthy subjects (31.3)Mixed (USA)WCSTMet/met homozygotes performed better on the WCST compared with heterozygotes or val/val subjects
Bates et al. (2003)79 healthy subjectsNot stated (USA)Not stated (long-term visual memory)Met/met homozygotes performed better than val/val homozygotes on the visual memory task
Gallinat et al. (2003)49 schizophrenics (32.8) + 170 C (43.9)Caucasian (German)Auditory oddball ERPMet/met subjects had lower frontal P300 amplitudes than val/val homozygotes. Frontal P300 amplitude correlates negatively with prefrontal cognitive task performance
Goldberg et al. (2003)74 schizophrenics (37), 108 siblings (37) and 68 C (35)94% Caucasian (USA)N-back working memory task + CPTCPT and 0-back performance (attention) not related to genotype. Val/val genotype associated with worst performance in 1 + 2-back conditions (working memory)
Mattay et al. (2003)25 healthy subjects (33) imaged on amphetamineNot stated (USA)WCST and N-back working memory taskVal/val subjects on placebo made more errors on the WCST than met/met subjects. Val/val subjects improved on the amphetamine, while met/met subjects performed worse on the WCST and N-back task
Tsai et al. 2003)120 healthy females (19–21)Han ChineseWAIS + WCSTMet/met homozygotes showed reduced P300 response latencies. There was no difference between the groups on the cognitive tasks
Bearden et al. (2004)44 children (11.1) with 22q11 deletion syndromePredominantly Caucasian (USA)Animal naming, TMT-B, arithmetic (WISC), digit spanMet hemizygotes performed better than val hemizygotes on digit span and TMT-B
Bertolino et al. (2004)30 schizophrenics (28.6) treated with olanzapine over 8 weeksNot stated (Italian)N-back working memory taskNo difference between groups at 4 weeks treatment. At 8 weeks, met carriers performed better on 2-back task. Met/met homozygotes showed least dorsolateral prefrontal activation on fMRI
de Frias et al. (2004)286 men (57.8)Not stated (Swedish)Episodic and semantic memory (battery)Met/met homozygotes performed better than heterozygotes or val/val homozygotes on the episodic and semantic memory tasks
Diamond et al. (2004)39 healthy children (9)Predominantly Caucasian (USA)Dots-mixed (working memory and inhibition); self-ordered pointing task (working memory only); recall memory; mental rotation (posterior parietal lobe)Met/met homozygotes performed better than val/val subjects on the dots-mixed task only. There was no genotype effect on the other tasks
Foltynie et al. (2004)288 patients with Parkinson's disease (65.5)Not stated (British)TOL, CANTAB (pattern and spatial recognition), verbal fluencyMet allele associated with poorer performance on the TOL only
Mills et al. (2004)114 children with ADHD (9.2)Caucasian (British)Digit span, MFFT, CPT, Go No Go (MARS)No significant differences
Nolan et al. (2004)26 schizophrenics (41.4)Predominantly Caucasian (USA)Competing programs task (tests imitation and reversal of rule)Met allele associated with better imitation (cognitive stability) but val allele associated with better performance in the reversal condition (flexibility)
Rosa et al. (2004)89 sibling pairs discordant for schizophrenia (not given)Not stated (Spanish)WCSTNo association between genotype and performance in schizophrenics.Val/val genotype associated with greater number of perseverative errors in healthy siblings
Stefanis et al. (2004)543 males from the general population (21)Caucasian (Greek)N-back working memory task; CPT; antisaccade taskNo significant association between genotype and task performance
Strauss et al. (2004)63 young adults with childhood-onset mood disorder (18.4)65% Caucasian (USA)Logical memory tests from the WMS; paired associates (delayed recall); RCF; WAIS; WISCNo significant association between genotype and task performance
Taerk et al. (2004)118 ADHD children (9)Not stated (Caucasian)SOPT, TOL, WCSTNo significant association between genotype and task performance
Tsai et al. (2004)154 schizophrenics (43.5)ChineseMMSENo significant differences reported
Weickert et al. (2004)20 schizophrenics (approximately 32) given antipsychotics or placeboNot statedN-back working memory task; WCST; verbal fluency; WAIS; WMSMet/met subjects on medication improved on the N-back task. Val/val subjects showed no improvement with treatment and were more impaired than met/met homozygotes
Bellgrove et al. (2005)61 children and adolescents with ADHD (12)Caucasian (Irish)WISC; TEA-ChVal/val homozygotes performed better than heterozygotes or met/met homozygotes on the ‘Walk Don’t Walk' subtest which measures sustained attention and response inhibition
Ho et al. (2005)159 schizophrenics (26.5) and 84 C (27)Caucasian (USA).WCST, digit span (R), TMT (A + B); N-back working memory taskNo significant differences reported

The latter point is important, because if all the positive findings reported to date were spurious, then it would be expected that the relevant studies would implicate all alleles of COMT with equal frequency. In fact, 90% of statistically significant results reported in the literature suggest that the low-activity met allele is associated with enhanced working memory. Assuming that no publication bias exists, these data suggest that the COMT val158met polymorphism may exert a genuine effect on cognition.

Egan et al. (2001), in the original study, reported that the high-activity val allele was associated with poorer performance on the Wisconsin Card Sorting Test (WCST), a putative measure of ‘executive’ function, reduced efficiency of physiological response in the dorsolateral PFC during a working memory task and was transmitted more often than expected by chance to probands with schizophrenia. Egan et al. (2001) suggested that the COMT genotype accounted for 4% of the variance in performance on the WCST, and this was echoed by Joober et al. (2002), Bilder et al. (2002) and Goldberg et al. (2003). Despite this seemingly small effect size, however, positive results have been generated by studies with very modest sample sizes. Fourteen of the 20 studies reporting significant genotype – cognition associations – made use of sample sizes below 100 (Table 1). Whether this is a tribute to the sensitivity of the neuropsychological and neuroimaging techniques or whether the COMT val158met polymorphism exerts a more significant effect on cognition is unclear. There is some evidence for the latter possibility. Bilder et al. (2002) reported that the COMT genotype accounted for 11% of the variance in processing speed and attention. Diamond et al. (2004) found that 26% of the variance on their dots-mixed task was due to the effects of genotype, while Nolan et al. (2004) suggested that 28–41% of the statistical variance on their competing programs task was due to COMT variants.

Despite the impressive congruency in genotype-cognition correlations, the specific cognitive functions or modules that are sensitive to dopaminergic modulation remain a source of debate. Most of the initial studies used the WCST as a measure of prefrontal cognitive function. The original test developed by Berg in 1948 was designed to assess cognitive flexibility, hypothesis testing and problem solving. Frontal cortex dysfunction may interfere with the planning of tasks and the ability to generate and select alternative hypotheses after negative feedback (shifting cognitive set), traits which the WCST putatively measures (Lezak 1995). The inability to shift cognitive set is usually recorded as ‘percentage of perseverative errors’ made by the individual.

A consistent association between the val/val genotype and greater percentage of perseverative errors on the WCST characterizes the literature (Egan et al. 2001; Joober et al. 2002; Malhotra et al. 2002; Mattay et al. 2003; Rosa et al. 2004), a trend that appears to run counter to the hypothesized sequelae of hypodopaminergic states, discussed above. The val allele, reducing D1 receptor binding, should allow for more cognitive flexibility, the negative aspect of which may be greater distractibility as evinced by a WCST measure such as ‘failure to maintain cognitive set’. On the other hand, individuals with the met/met genotype should theoretically be more cognitively stable, less distractible and in certain cases therefore more likely to make perseverative errors.

The difficulty with the WCST is that it is a complex problem-solving task with multiple cognitive components and thus reasons for failure (Bilder et al. 2004; Goldberg et al. 2003). Nevertheless, the complexity of the task does not explain the consistent association (five studies) between the presence of the val allele and the increased number of perseverative errors on the WCST. Perhaps val/val homozygotes are less able to inhibit prepotent responses (see below) – in this case, the prepotent response may be the tendency to continue with the same strategy – than met allele carriers, and therefore make more perseverative errors.

The lack of specificity of the WCST has led to the use of simpler neuropsychological tasks with more focussed effects. Goldberg et al. (2003) made use of the N-back-working memory test in which the individual must continuously recall information that is 0 or 1, or more, back in a sequence. Catechol-O-methyltransferase genotype was not related to performance in the 0-back condition, which is more a measure of attention than working memory (Goldberg et al. 2003). In the 1-back and 2-back conditions, however, individuals with the met allele outperformed their counterparts, suggesting that the process of information updating and maintenance in the face of competing stimuli is sensitive to prefrontal DA levels (Goldberg et al. 2003). In a sample of schizophrenics treated with olanzapine, met carriers also performed better than val/val individuals on the 2-back task (Bertolino et al. 2004). Bilder et al. (2004), however, argue that the N-back task requires both stability (the maintenance of information) and flexibility (updating the currently relevant stimulus with new information) and thus suffers from some of the same drawbacks as the WCST.

Diamond et al. (2004) tested a group of children on a task that putatively requires both working memory and inhibition (the dots-mixed task) and a task that requires working memory only (self-ordered pointing). The dots-mixed task presents subjects with two types of dots, shaded or striped. For the shaded dots, the person is required to press a button with the left or right hand according to the side of the screen that the dot appears on, while for the striped dots, the subject is required to use the opposite hand (Diamond et al. 2004). In the self-ordered pointing task, a series of line drawings or abstract designs are presented. The subject's task is to touch each stimulus only once, but after each response, the computer screen refreshes, and the order of the stimuli changes (Diamond et al. 2004). Although both tasks depend on the integrity of the dorsolateral PFC, only the dots-mixed test was associated with genotype (met homozygotes performed better), and this led Diamond et al. (2004) to propose that working memory per se is not sensitive to PFC DA levels and COMT genotype. It is the ability to inhibit prepotent responses and shut out distracting stimuli, along with working memory that is sensitive to DA regulation (Diamond et al. 2004).

Nolan et al. (2004) examined the effect of the COMT genotype on a computerized task that requires shifting between two rules of responding, imitation and reversal. In the imitation condition, the subject presses once on a key in response to a single stimulus and twice in the case of two stimuli. The rules are then reversed, and the individual is required to press once in response to two stimuli and twice in the case of a single stimulus. Acquisition and maintenance of the imitation rule reportedly requires cognitive stability, whereas the reversal condition requires flexibility and the ability to inhibit the previously learned, more intuitive response (Nolan et al. 2004). In line with the hypothesized effects of D1 and D2 receptor binding, Nolan et al. (2004) found that the met allele was associated with better cognitive stability but poorer flexibility.

While it may be premature on the basis of a small number of studies to engage in theoretical speculation about the evolutionary trade-off between these two alleles, the high frequency of the low performance val allele in most populations may be explained by hypothesizing that greater cognitive stability and enhanced working memory makes the PFC system less sensitive to internal or external stimuli, a trade-off that presumably incurred costs in a hostile evolutionary environment.

A recent report has suggested that the influence of COMT on cognition may extend beyond working memory and inhibition. de Frias et al. (2004) found an association between the COMT genotype and episodic and semantic memory. The effect was specific to recall rather than recognition, with met/met individuals outperforming heterozygotes, but not val/val homozygotes (de Frias et al. 2004). In an earlier paper, Bates et al. (2003) found that met/met homozygotes performed better than val/val homozygotes on a delayed visual memory-recall task. Whether these are false-positive results or are indicative of a role for DA in memory function remains to be seen. Diamond et al. (2004) and Strauss et al. (2004) failed to implicate COMT in memory performance, although these studies may have been under-powered with sample sizes of 39 and 63, respectively.

Ho et al. (2005) have hypothesized that age differences between samples may be responsible for the failure to detect a significant gene – cognition association in some studies. The rationale behind this argument is a putative decline in frontal lobe DA function of 11–13% per decade, which may render older groups more sensitive to a genetically mediated (val allele) increase in DA catabolism (Ho et al. 2005). Because the mean age of the sample of Ho et al. (2005) was approximately 26, we divided the data in Table 1 into two groups: those studies that sampled individuals below 30 and those studies with a mean subject age of greater than 30. We excluded the data derived from children with attention-deficit hyperactivity disorder (ADHD) and studies where age was not reported.

Preliminary support for the hypothesis of Ho et al. was detected. Three out of five studies including individuals with a mean age of less than 30 were unable to detect a significant gene – neurocognitive function relationship, while 10 out of 11 studies with a mean sample age of more than 30 reported positive results. However, the effect of COMT variants on cognitive function in children is not entirely consistent with this hypothesis. Out of six studies that sampled children, three reported the met allele to be associated with better performance on tests of ‘executive’ function, one found the met allele to be associated with worse cognitive function, and two studies were non-significant.

The difficulty with the interpretation of childhood studies of cognition is that the PFC is yet to reach full maturity. Segalowitz & Davies (2004) hypothesized that the PFC develops until late adolescence, while Welsh et al. (1991) extended this neurodevelopmental process until early adulthood. In the absence, therefore, of a lucid understanding of the developmental status of these inchoate dopaminergic pathways, extrapolating genetic association data from children to adults may be problematic.

In summary then, on the basis of the genetics literature, neurocognitive tasks that require the individual to hold information ‘on-line’ (i.e. working memory) and perhaps inhibit responses to prepotent or distracting stimuli appear to be most sensitive to differential COMT activity. Some preliminary evidence suggests that the effect of COMT variants may extend to other cognitive domains such as memory (Bates et al. 2003; de Frias et al. 2004). The problem with simply reporting scores obtained on neuropsychological tests of memory in isolation, however, is that low scores may be indicative of poor executive functioning rather than a bona fide memory deficit (that is a deficit mediated by hippocampal dysfunction). In our opinion, therefore, the balance of evidence indicates that it may be premature at this stage to invoke a role for COMT in mediating medial temporal function.

While debate will doubtless continue over the nature of the neurocircuitry modulated by COMT, important implications for the treatment of a variety of psychiatric conditions loom on the horizon. Mattay et al. (2003) demonstrated that amphetamine, which blocks the action of dopamine, serotonin and noradrenaline transporters, improves working memory in individuals with suboptimal D1 receptor activation (i.e. val/val individuals) but has a detrimental effect on PFC cognition in met/met individuals. In contrast, antipsychotics were found to improve cognitive function as evinced by the N-back task in met homozygotes, but val/val individuals did not improve cognitively on treatment (Weickert et al. 2004).

Inada et al. (2003) examined the response of 100 schizophrenics to neuroleptics and found that met/met homozygotes were on a significantly higher dose of maintenance therapy than the other patients. This effect was partially replicated by Illi et al. (2003) who detected a significantly higher rate of conventional neuroleptic treatment resistance in met/met patients. Thus, the effect of the met allele which shifts the D1/D2 activation ratio in favour of D1 is presumably accentuated by D2 receptor antagonists.

DRD1 and DRD2

The important influence of COMT variants on the D1/D2 receptor-activation ratio raises the possibility that functional polymorphisms of these two receptor genes may also contribute to population variation in cognitive performance. The D1 receptor is highly expressed in the glutamatergic pyramidal cells of the PFC (Lidow et al. 1991). Arnsten et al. (1994) demonstrated that a D1 receptor antagonist impaired the working memory of monkeys, while a D1 agonist improved memory performance. Similarly, D1 receptor knockout mice display spatial learning deficits (El-Ghundi et al. 1999), and antipsychotic induced downregulation of PFC D1 receptors was shown to produce severe working memory deficits which could be reversed by D1 receptor stimulation (Castner et al. 2000). Okubo et al. (1997) reported a positive correlation between the density of prefrontal D1 receptors and performance on the WCST in a schizophrenic population, and Muller et al. (1998) showed that visuospatial working memory improved in individuals administered pergolide, a D1 and D2 receptor agonist.

D2 receptors are most commonly found on the GABAergic interneurons of the PFC and appear to play a role in depressing N-Methyl-D-Aspartate (NMDA) receptor-mediated excitatory neurotransmission (Kotecha et al. 2002). Mehta et al. (1999) administered sulpiride, a D2 receptor antagonist, to healthy volunteers and found that the drug impaired spatial working memory, planning and attentional set-shifting. In a follow-up study, the same group showed that shifting of cognitive set is particularly sensitive to D2 receptor antagonism (Mehta et al. 2004). Kimberg et al. (1997) demonstrated that a D2 receptor agonist improved cognition in individuals with a low-baseline memory capacity. In addition, an age-related decrease in D2 receptors has been hypothesized to contribute to cognitive decline in the elderly. As a result, the D2 agonist piribedil has been used in the treatment of patients with mild cognitive impairment, apparently with some success (Peretti et al. 2004).

On the basis of the pharmacological data described above, it is entirely plausible that executive functions such as working memory are influenced by functional variation in both the D1 and D2 receptors. We were unable to find any studies that made reference to a relationship between cognitive performance and polymorphisms of the D1 receptor. A handful of publications have, however, addressed the role of D2 receptor polymorphisms on cognition (Table 2). All of these studies concentrated on one particular SNP within the DRD2 gene, the TaqIA polymorphism.

Table 2.  Relationship between dopamine D2 variants and cognitive performance
StudySampleEthnicityCognitive tasksOutcome
  1. CPT, continuous performance test; ERP, event related potential; JLO, judgement of line orientation; MMSE, mini mental status examination; RAVLT, Rey auditory verbal learning test; TMT, trail making test; TOH, tower of Hanoi; WAIS-R, Weschler adult intelligence scale (revised); WISC-R, Wechsler intelligence scale for children (revised).

Noble et al. (1994)98 10–14-year-old boysCaucasian (USA)CPTERP-derived P300 latency was prolonged in Taq A1 allele 1 carriers
Berman and Noble (1995)182 10–14-year-old boysCaucasian (USA)JLOReduced visuospatial performance associated with the Taq A1 allele
Petrill et al. (1997)51 high IQ, 51 average IQ, and 35 low IQ childrenCaucasian (USA)WISC-RNo significant association between IQ and genotype
Bartres-Faz et al. (2002)49 memory impaired subjects (mean age 65 years)Not stated (Spanish)MMSE, RAVLT, logical memory, visual reproduction, visual paired associates, TMT-A + B; FAS; TOH; JLOSubjects homozygous for the Taq A2 allele exhibited reduced left caudate volumes and performed worse on the RAVLT and MMSE
Tsai et al. (2002)112 female volunteers, 19–21 years oldHan ChineseWAIS-RThe A1/A1 group scored significantly higher on the performance scale than the A2/A2 group

The DRD2 gene is about 270 kb long and contains eight exons, with the TaqIA locus located in the 3′ untranslated region of the gene (Noble 2000). The functional status of the TaqIA variant is not entirely clear, but some preliminary evidence indicated that A1 allele was associated with reduced density of DRD2 receptors in the striatum (Pohjalainen et al. 1998). More recently, Duan et al. (2003) found that a synonymous SNP (C957T) affects mRNA stability and therefore receptor expression, and this was confirmed in vivo by Hirvonen et al. (2004). The C957T variant was found to be in linkage disequilibrium with the TaqI A variant in a Caucasian but not an African-American sample (Duan et al. 2003).

Five studies examining the effect of the DRD2 TaqIA polymorphism and cognition are listed in Table 2. Tsai et al. 2002) and Petrill et al. (1997) made use of standardized IQ tests. The former reported that A1 homozygotes outperformed their counterparts on the performance subscale of the Wechsler Adult Intelligence Scale (WAIS)-R, while the latter found no significant association between genotype and IQ. Using the Rey Auditory Verbal Learning Test, a measure of verbal memory and learning, Bartres-Faz et al. (2002) reported a result in the same direction as Tsai et al. (2002): reduced performance in A2 homozygotes. However, in a sample of 10–14-year-old boys, Noble et al. (1994) and Berman and Noble (1995) reported prolonged P300 latency and reduced visuospatial performance in A1 allele carriers. These contradictory findings together with the modest sample sizes of the relevant studies suggest the possibility of false positives. Nevertheless, given the wide distribution of the DRD2 receptor in the caudate, putamen, nucleus accumbens, amygdala, hippocampus and cerebral cortex and the effect of COMT on the D1/D2 receptor-activation ratio, further investigation of this gene may prove worthwhile. We suggest that because of its putative functional effect, use of the C957T polymorphism may provide more decisive results than have been achieved up to now.


Wang et al. (2002) proposed that DRD4 impacts prefrontal cognition by virtue of its effects on GABAergic signalling. The authors demonstrated that activation of DRD4 receptors, located in the vicinity of GABAA receptors on dendrites, reduces GABAA receptor-mediated currents in the PFC (Wang et al. 2002). GABAergic activity in the PFC has been shown to modulate the processing of information and the planning of future actions (Constantinidis et al. 2002). Thus, both DRD4 and genes coding for components of the GABAergic system are also candidates for influencing cognition. Egan et al. (2004) showed that a potentially functional variant in the metabotropic glutamate receptor, GRM3, impacts verbal fluency and memory and is associated with abnormal PFC and hippocampal activation as evinced by functional magnetic resonance imaging (fMRI). To our knowledge, this is the only genetic association study that has correlated GRM3 variants with cognition. More work has been carried out on DRD4, most probably because of its hypothesized role in ADHD.

Attention-deficit hyperactivity disorder is characterized by deficits in impulse control, attention, working memory, planning and self-regulation (Barkley 1997), traits that are usually associated with damage to or dysfunction of PFC circuits. The treatment of choice, methylphenidate (Ritalin), which promotes the release of DA and blocks its reuptake, is suggestive of an intimate link between this neurotransmitter and ADHD. Thus, any genes involved in the aetiology of ADHD may be good candidates for modulating cognitive function. One such candidate is the DA D4 receptor gene. The association between an exonic DRD4 variable number tandem repeat (VNTR) variant (see below) and ADHD is one of the most robust findings in psychiatric genetics (DiMaio et al. 2003). Both case-control and family-based association studies have suggested that the longer or 7R allele is a risk factor for ADHD (DiMaio et al. 2003).

The DRD4 belongs to the same class of receptor as DRD2, exerting an inhibitory effect on the adenylyl cyclase-mediated secondary messenger system and is found primarily in the limbic regions and on PFC pyramidal neurons (Wang et al. 2002). The DRD4 gene is located on the short arm of chromosome 11 and contains a 48 bp VNTR polymorphism in the third exon of the gene which has been widely typed in genetic association studies (Ding et al. 2002). Between 2 and 11 repeated elements have been reported in the literature, although the two predominant alleles in Caucasians consist of four (4R) and seven repeats (7R), respectively (Ding et al. 2002). The polymorphic repeated segment codes for amino acids in the third intracellular loop of the receptor, a region that couples to G proteins and therefore mediates intracellular signalling (Asghari et al. 1995). The strategic position of the polymorphism suggests that the DRD4 receptor variants may exert a functional effect. Asghari et al. (1995) found differences in cyclic adenosine monophosphate inhibition between the 4R and 7R alleles, as well as between a 2R and 4R combined group, and the 7R allele. It is unclear whether the various alleles in question are differentially sensitive to endogenous DA, although the 4R allele has been reported to be more responsive to pharmacological DA agonists (Fossella et al. 2002).

We are aware of four papers that have described an association between alleles of the DRD4 48 bp VNTR polymorphism and cognitive functioning (see Table 3). Interestingly, three out of the four studies were carried out on individuals with ADHD, possibly because the DRD4 gene is a candidate gene for this disorder. While Swanson et al. (2000), Fossella et al. (2002) and Manor et al. (2002) found that children with a 4R allele or other shorter (2–5R) alleles performed more poorly on tests of attention and executive functioning, Langley et al. (2004) report increased impulsivity and therefore worse performance on a test of behavioural inhibition in carriers of the 7R allele.

Table 3.  Associations between the dopamine D4 receptor gene and cognition
StudySampleEthnicityCognitive tasksOutcome
  1. ADHD, attention deficit hyperactivity disorder; ANT, attention network test; CPT, continuous performance test; MFFT, matching familiar figures test; TOVA, test of variables of attention

Swanson et al. (2000)32 ADHD subjects68% Caucasian (USA)Stroop, cued detection task, go-change taskIndividuals with a 7R allele performed at the same level as the control group. ADHD children without a 7R allele performed poorly on the battery
Fossella et al. (2002)200 adultsNot stated (USA)ANTIndividuals without a 4R allele performed better on the executive attention component than carriers of the allele
Manor et al. (2002)178 ADHD triadsNot stated (Israeli)CPT (TOVA)Children with the short (2–5) exon III repeats performed worse on the CPT
Langley et al. (2004)133 children with ADHDCaucasian (British)CPT, MFFT, Go/No Go Task, Stop TaskChildren with the 7-R allele were more impulsive on the MFFT and stop task

If the 4R variant does indeed increase sensitivity to DA, then one would expect more efficient signal transduction in 4R carriers and a subsensitive response to DA in 7R carriers. Theoretically, the 7R allele could therefore be associated with under-activity of the mesocorticolimbic DA pathway (Swanson et al. 2000), although, to our knowledge, this remains to be conclusively demonstrated. However, this hypothesis will have to reconcile better neurocognitive performance in ADHD individuals with the DRD4 7R allele (presumably with an under-active mesocortical DA pathway) and the association between improved cognition and the COMT met allele (which elevates the PFC DA level).

The interpretation of these results is further complicated by the hypothesis of Swanson et al. (2000) that there may be two basic aetiological subtypes of ADHD. One sector of the ADHD population displays a profile suggestive of significant cognitive (and neuropsychological task) impairment. The balance of the population is hypothesized to carry the 7R polymorphism implicated in novelty-seeking behaviour (Savitz & Ramesar 2004). This genetic subtype may represent an extreme of temperament, manifesting the behavioural but not the neurocognitive problems associated with ADHD (Swanson et al. 2000). If this postulate is correct, then those individuals with a ‘genetic subtype’ of ADHD (i.e. carriers of the 7R allele) may outperform their counterparts who have sustained some exogenous form of damage to the brain during development. This latter ADHD group will probably not display an excess of any particular DRD4 VNTR allele. Thus, in ADHD samples, although the 7R allele may be associated with better cognitive performance, it may play no role in cognition and may simply be a proxy for ADHD aetiology.

In addition, a C to T transition 521 bp upstream from the transcription-initiation site has attracted attention because of its possible functional effect. According to Okuyama et al. (1999), the T variant is transcribed with 40% less efficiency than the C allele. More recently, D'Souza et al. (2004) reported that the longer allele of a 120-bp tandem repeat in the 5′-flanking region of the DRD4 gene, which is known to contain predicted consensus sequence transcription factor-binding sites, has lower levels of transcriptional activity than the shorter allele. These variants have also been implicated in ADHD (Kustanovich et al. 2004) and novelty-seeking behaviour (Rogers et al. 2004) and may be worthwhile genotyping in future investigations of the genetics of cognition. Theoretically, the 48 bp VNTR could be in linkage disequilibrium with either or both of these polymorphisms (or some other unknown variant), although no linkage disequilibrium between the −521 C/T polymorphism and the 48 bp VNTR was detected by Fossella et al. (2002) in their sample.

Brain-derived neurotrophic factor

Neurotrophins are regulatory factors that mediate the differentiation, proliferation and survival of cholinergic, dopaminergic and serotonergic neurons (Poo 2001). Brain-derived neurotrophic factor, one of these neurotrophins, is expressed throughout the brain, particularly in the PFC and hippocampus (Pezawas et al. 2004) where it exerts long-term effects on neuronal survival, migration, dendritic and axonal growth (Pang & Lu 2004). Brain-derived neurotrophic factor has been shown to prevent the spontaneous death of dopaminergic rat neurons (Hyman et al. 1994), exert a protective effect in the presence of neurotoxins (Hung & Lee 1996) and elevate forebrain 5-HT neuronal fiber density (Mamounas et al. 1995).

Brain-derived neurotrophic factor also appears to act on a much shorter temporal scale. In vitro experiments demonstrate that BDNF application causes rapid changes in synaptic transmission at both excitatory and inhibitory synapses (Baldelli et al. 2002). Brain-derived neurotrophic factor is also secreted in response to neuronal activity (Farhadi et al. 2000; Goodman et al. 1996, cited in Egan et al. 2003) where it acts as a retrograde or paracrine messenger allowing neuronal activity to modify synaptic connections (Egan et al. 2003; Poo 2001). This modulation of synaptic plasticity and neuronal transmission is particularly salient in the hippocampus where BDNF-mediated enhancement of long-term potentiation (LTP) has been shown to facilitate memory formation (Poo 2001).

Brain-derived neurotrophic factor is widely expressed in the rat hippocampus, and studies of BDNF knockout mice have demonstrated impairment of hippocampal LTP (Pozzo-Miller et al. 1999), and treatment with antisense BDNF oligonucleotide or anti-BDNF antibody impaired spatial memory formation in rats (Mu et al. 1999). Knockout mice, with a deletion of the TrkB gene which codes for the BDNF receptor, show LTP and maze-learning deficits (Minichiello et al. 1999). The causal mechanism behind BDNF-mediated modulation of LTP is not clear, but one suggestion is that the protein enhances synaptic vesicle docking by phosphorylating synaptic proteins (Pang & Lu 2004).

The BDNF gene is located on chromosome 11p13 and is composed of five or more exons, each with its own promoter region allowing for differential mRNA splicing (Jiang et al. 2005). A frequent, non-conservative SNP in the gene (dbSNP number rs6265), producing a val to met amino acid substitution at codon 66 (val66met) of the pro-BDNF sequence, was shown by Egan et al. (2003) to affect the activity-dependent secretion of BDNF. Depolarization-dependent secretion of BDNF is impaired by the met allele. The met-BDNF may not be correctly transferred from the Golgi apparatus to its appropriate secretory granules, despite the fact that mature protein function is unaltered by the polymorphism (Egan et al. 2003).

On the basis of these data, Egan et al. (2003) and Hariri et al. (2003) reasoned that the val66met variant may impact human hippocampal function and memory. In a sample of schizophrenic and healthy subjects, the met allele was associated with poorer episodic memory as evinced by immediate and delayed recall of Wechsler Memory Scale stories, a disruption of the normal hippocampal fMRI disengagement pattern during a working memory task, and lower hippocampal n-acetyl aspartate, an intracellular marker of neuronal function (Egan et al. 2003). Similarly, Hariri et al. (2003) demonstrated that met-BDNF carriers displayed reduced hippocampal engagement during both the encoding and retrieval of a spatial task and made significantly more recognition errors on this task than val/val homozygotes. These results were partially replicated by Dempster et al. (2005). Another study of declarative memory in affectively ill individuals, however, failed to confirm these original findings (Strauss et al. 2004). In MRI investigations, Pezawas et al. (2004) and Szeszko et al. (2005) extended the findings of Egan et al. (2003) and Hariri et al. (2003) by demonstrating that val/met heterozygotes displayed lower hippocampal volumes than their val/val counterparts.

The putative effects of the val66met variant on working memory and retrieval of encoded information suggest that the influence of BDNF may extend beyond its role in LTP (see Table 4). Rybakowski et al. (2003) found that carriers of the met allele performed significantly worse than val/val homozygotes on the WCST, a measure of PFC function. This was confirmed by the recent Pezawas et al. (2004) report, demonstrating that met-BDNF carriers display age- and gender-independent dorsolateral PFC gray matter volume reductions. Pezawas et al. (2004) propose that this variation in cortical morphology is at least partly the result of long-term BDNF-mediated effects on the developmental process and is not simply a reflection of transient val66met-moderated neurotransmitter fluctuations during memory formation.

Table 4.  Association between the BDNF val66met polymorphism and cognition
StudySampleEthnicityCognitive tasksOutcome
  1. C, controls; CANTAB, Cambridge neuropsychological test automated battery; CTT, corsi tapping task; CVLT, California verbal learning test; IAPS, international affective picture system; IMCT, international memory and concentration test; NS, not significant; Schiz, schizophrenics; Sibs, siblings; TOL, tower of London: WAIS, Wechsler adult intelligence test; WCST, Wisconsin card sorting test; WMS, Wechsler memory scale.

Egan et al. (2003)203 schizophrenics and 305 siblings and 133 C90% Caucasian (USA)WMS (episodic memory) and CVLTMet/met homozygotes performed worse thanheterozygotes or val/val homozygotes on WMS. No difference on CVLT
Hariri et al. (2003)28 healthy subjectsPredominantly Caucasian (USA)Visual declarative memory (IAPS)Memory-related hippocampal activity greaterduring both encoding + retrieval in val/val homozygotes. Met carriers made more errors on recognition memory task
Rybakowski et al. (2003)54 bipolar patientsNot stated (Polish)WCSTVal/val subjects performed significantly better than met allele carriers
Nacmias et al. (2004)83 Alzheimer patients and 97 CCaucasian (Italian)IMCT, digit span, CTT, babcock story, paired wordsNS
Strauss et al. (2004)63 adults with childhood-onset mood disorder(Canadian)Declarative memoryNS
Tsai et al. (2004b)114 healthy femalesHan ChineseWAISVal/met heterozygotes but not met/met homozygotes performed worse than the val/val group on the performance scale of the WAIS
Dempster et al. (2005)92 schizophrenics and 114 unaffected relativesNot stated (British)WMSThe met allele was associated with worse performance on the delayed score of the logical memory subscale
Foltynie et al. (2005)291 individuals with Parkinson's diseaseNot stated (British)TOL, verbal fluency, spatial and pattern recognition (CANTAB)Patients with a met allele performed better on the TOL than val/val homozygotes

Tsai et al. (2004) presented data indicating that val/val homozygotes performed significantly better than carriers of the met allele on the performance scale of the WAIS, although the statistical significance was marginal, and the result stands in contradistinction to that of Egan et al. (2003). On the other hand, Foltynie et al. (2005) reported that in patients with PD, the met allele was associated with better performance on the TOL, a measure of planning ability. The apparently differential effect of the BDNF val66met polymorphism in PD may possibly be explained by the complex interactions between BDNF secretion and degeneration of the nigro-striatal dopaminergic neurons which may result in a prefrontal-striatal DA imbalance (Foltynie et al. 2005).

Thus, out of six studies that have implicated BDNF variants in cognition, four have reported an association between the met allele and worse performance on sundry neurocognitive tasks. One study is equivocal (Tsai et al. 2004), and one study (Foltynie et al. 2005) finds evidence for improved performance in met carriers (see Table 4).

A potentially confounding variable is the discovery by Jiang et al. (2005) of a functional polymorphism 281 bp upstream from the putative transcription-initiation site of exon 1 of BDNF. The ‘A’ variant which may be protective against anxiety and psychiatric illness appears to exert an independent, but opposite effect to the met 66 allele which has been implicated in psychopathology (Jiang et al. 2005). The two genetic variants may, however, be in linkage disequilibrium with each other, although, given the differential anxiety-related effects of the −281C>A variant on the val 66 haplotype background (Jiang et al. 2005), future research may benefit from controlling for this SNP.

There appear to be at least three different mechanisms through which BDNF impacts cognition. One possibility is that stress, especially chronic stress, may result in excessive release of glucocorticoids from the adrenal gland, directly and indirectly causing atrophy and death of vulnerable neurons through the actions of cortisol and the inhibition of BDNF synthesis, respectively (Duman et al. 1997). Because the hippocampus is particularly vulnerable to the action of glucocorticoids (Nestler et al. 2002), this may explain the memory deficits described above. A potentially useful future study would be to examine the effects of the val66met polymorphism on cortisol release and hypothalamic pituitary-adrenal axis dysfunction.

A second possibility is that the val66met variant influences the efficacy of both early- and late-phase LTP in the hippocampus (Egan et al. 2003; Hariri et al. 2003), consistent with reported BDNF-driven variation in verbal and visual declarative memory. Interestingly, three out of four studies hypothesizing a significant role for BDNF in neurocognition have reported that heterozygotes perform more poorly on cognitive tasks than val/val homozygotes. Whether this is an artefact of the low frequency of met homozygotes in the population, limiting potential group comparisons in genetic association studies, or indicative of the sensitivity of LTP to BDNF level is debatable.

Thirdly, BDNF has diffuse effects on monoaminergic neurotransmitter systems. It promotes the sprouting of mature serotonergic neurons (Mamounas et al. 1995), augments central serotonergic activity following a midbrain infusion (Siuciak et al. 1996), moderates the firing of neurons in the dorsal raphe nucleus (Celada et al. 1996) and modulates serotonin transporter function (Mossner et al. 2000). Brain-derived neurotrophic factor has also been shown to influence dopaminergic activity (Narita et al. 2003). Administration of BDNF has been shown to potentiate nigrostriatal dopaminergic function and locomotor activity (Horger et al. 1999; Pierce et al. 1999, cited in Narita et al. 2003), and chronic BDNF treatment enhances the reward response to cocaine (Horger et al. 1999).

In summary, (a) Animal studies suggest that BDNF protects neurons from the effects of damage, plays a role in modifying synaptic connections and modulates hippocampal LTP. (b) Human studies have demonstrated that the met allele is associated with lower hippocampal volume or functional activity. (c) Genetic studies with the functional val66met variant have indicated that the met allele is associated with weaker performance on tests of memory and ‘executive’ function. Given these parallel sources of evidence, we suggest that it is likely that val66met BDNF variant exerts an effect on memory performance and perhaps executive function, although more evidence is required for the latter.


A number of theoretical obstacles may impede the resolution of gene-cognition relationships.

(a) Gene-cognition correlations reported in the literature usually rest on the assumption that the effect of the genetic variant in question is specific to a particular cognitive process and is not reflective of variation in a general intelligence factor. It has long been recognized that scores on different IQ tests emphasizing distinctive types of cognitive tasks are positively correlated with each other, suggesting the existence of a general intelligence factor eponymously named Spearman's g (Spearman 1904). Persuasive evidence that IQ scores predict concurrent neuropsychological performance across the entire spectrum of intelligence in neurologically normal individuals has also been mustered (Jung et al. 2000). As the number of discrete cognition-genotype correlations reported in the literature grows, so does the difficulty of distinguishing between (a) false positives, (b) the potentially pleiotropic influence of the polymorphism on multiple but discrete cognitive processes and (c) the possible effect of genotype on g– that is a higher-order over-arching cognitive process. This can be seen in the data reviewed above. While the COMT val158met variant has been reported to impact general executive function (Egan et al. 2001), working memory (Goldberg et al. 2003), working memory and inhibition (Diamond et al. 2004) and episodic and semantic memory (de Frias et al. 2004), the BDNF val66met polymorphism has been postulated to influence hippocampal function and memory (Egan et al. 2003; Hariri et al. 2003) and different components of executive function (Foltynie et al. 2005; Rybakowski et al. 2003).

A related problem is that the use of neuropsychological tests to pinpoint cognitive modules that are sensitive to genetic variation is limited by the absence of a simple linear relationship between task performance and activation of a particular neural circuit. This is because there are many reasons for failure on a neuropsychological task. Goldberg and Weinberger (2004) suggest that the use of an alternative phenotype, such as fMRI activity, which may be more closely related than psychological testing to the underlying function of the neural system, may ameliorate this difficulty.

(b) The complexity of neuromolecular pathways makes it dangerous to draw simple causal relationships between a genetic variant and a cognitive process. The putative relationship between, for example, the COMT val158met SNP and working memory probably indicates that COMT is one cog in a complex circuit involved in PFC function. Does the COMT val158met variant exert a direct causal effect on PFC function, or is it simply correlated with a true causal factor such as the efficacy of DA receptor binding or the developmental trajectory taken by the embryonic brain?

(c) If a true but weak relationship between COMT and BDNF variants and cognitive function exists, then it would be expected that a greater percentage of positive findings would have been derived from studies employing large numbers of participants. In fact, in the case of COMT, the pattern is diametrical. Four of the largest studies with sample sizes of 159 schizophrenics and 84 controls (Ho et al. 2005), 154 schizophrenics (Tsai et al. 2004), 543 healthy males (Stefanis et al. 2004) and 220 healthy adults (Fossella et al. 2002) failed to show statistically significant gene-cognition associations. On the other hand, 15 studies making use of sample sizes of 100 or less report positive results (see Table 1). The relatively low rate of positive findings in studies with large sample sizes and statistical power is a cause for concern and raises the possibility that some of these results may be spurious. Concerning BDNF, the two studies with the largest sample sizes, 203 schizophrenics, 305 siblings and 133 controls (Egan et al. 2003), and 291 patients with PD (Foltynie et al. 2005) both reported significant results, while the two non-significant findings were obtained with sample sizes of less than 100 (see Table 4). This picture is more consistent with a small but genuine effect of the val66met BDNF variant on cognition.

These crude estimates of publication bias are best assessed with a formal meta analysis or funnel plot. The difficulty in this field is the diversity of tasks used to measure cognition. Different neuropsychological tasks vary in their properties and therefore exert a differential influence on effect size and study outcome. Secondly, results are differentially reported across studies: for example, the homozygote and heterozygote groups are sometimes grouped together and sometimes considered separately in the statistical analyses. Given the additional frustration that the data required to calculate effect size and mean standard error are not always available, more formal estimates of publication bias are not presented here.

(d) False-positive results due to lack of correction for multiple testing is a potential problem when many different neuropsychological measures are being used. Theoretically, the cut-off value for significance (a value) should be divided by the number of comparisons being made in order to avoid Type I errors. Some researchers, however, contend that conservative corrections such as the Bonferonni method are unnecessary when the research is driven by an a priori hypothesis and hampers the detection of weak effects. The best approach is probably some kind of compromise between the two extremes. Numerous multiple comparisons will be problematic for the interpretation of results, but unnecessarily strict methodological requirements will hamper the detection of weak effects.

The genetics of cognition: implications for psychiatry

The détente between the once disparate worlds of cognition and emotion research (see Damasio 1994; Panksepp 1998) has coincided with the growing recognition that cognitive impairment is a central facet of many psychiatric disorders. Deficits of executive function, working memory and declarative memory are widely reported in the schizophrenia literature (Hoff & Kremen 2003). Similarly, two recent reviews of the neuropsychology of bipolar disorder concluded that executive dysfunction and verbal memory deficits are pathognomonic of a bipolar diathesis (Glahn et al. 2004; Savitz et al. 2005). Cognitive dysfunction has also been reported in a multitude of other psychiatric conditions including ADHD (Barkley 1997), unipolar depression (Davidson et al. 2002), eating disorders (Lena et al. 2004), obsessive compulsive disorder (Shin et al. 2004) and borderline personality disorder (Monarch et al. 2004).

Given the extensive iterative connections between the PFC and the limbic system, these data suggest that an occult biological anomaly in the aforementioned neural networks underpins both a disruption of cognition and affect. Another possibility is that a primary affective or cognitive disturbance leads to dysfunction of the other modality. Regardless of which possibility turns out to be correct, ‘cognitive genetics’, far from being an esoteric backwater of modern psychiatry, is most probably indispensable to a comprehensive account of both the aetiology and pathophysiology of psychiatric illness.


Traditional behavioural genetics methodologies have long suggested a role for genes in cognition, but it is only in the past few years with the rapid evolution of molecular and computational technology for genetic analysis, coupled with the extraordinary flow of data from the human genome project, that identification of these genes has become a viable proposition. One of the defining features of our species is a significantly enlarged neocortex and gyrified PFC, and this together with a plethora of evidence implicating in particular, the PFC in cognition, suggests that the genes expressed in this part of the brain are good candidates for involvement in cognitive function.

One such of these candidates is COMT, an enzyme which influences the activity of DA in the PFC by virtue of its catabolic effects on the neurotransmitter. Given its pivotal role in the regulation of dopaminergic activity and the large number of studies that have implicated the val158met SNP in cognition, we suggest that functional variation in this gene contributes to inter-individual differences in cognitive function. The DA receptor genes, in particular D1, D2 and D4, are all theoretically good candidates for a role in cognition, but the number of genetic association studies carried out thus far is too small to draw any firm conclusions. Another candidate gene is BDNF which appears to exert a more diffuse effect via its role in LTP and neuronal survival and repair. Physiological, genetic and neuroimaging data all favour the hypothesis that BDNF influences neurocognitive function. As is the case with COMT though, a resolution of the precise cognitive processes regulated by BDNF remains elusive.


The support of the Medical Research Council of South Africa is acknowledged.