A meta-analytical review of 20 studies (n = 3907) of the association between DRD4 polymorphism and novelty seeking suggests the following conclusions: (a) on average, there is no association between DRD4 polymorphism and novelty seeking (average d= 0.06 with 95% CI of ± 0.09), where 13 reports suggest that the presence of longer alleles is associated with higher novelty seeking scores and seven reports suggest the opposite; (b) there is a true heterogeneity among the studies (ie, unknown moderators do exist) but the strength of the association between DRD4 polymorphism and novelty seeking in the presence of any (unknown) moderator is likely to be weak; (c) search for moderators has not yielded any reliable explanation for the variability among studies. We propose that to find such moderators, theory-driven research for potential interaction, coupled with larger sample sizes should be employed. The growing availability of powerful statistical techniques, high-throughput genotyping and large numbers of polymorphic markers such as single nucleotide polymorphisms makes such proposed studies increasingly feasible.
Evidence based on twin and adoption studies suggests that human personality traits are partially heritable.1 However, the specific genes responsible for trait variability are not known. Human personality is usually assessed using self-report questionnaires and one such questionnaire is the tridimensional personality questionnaire or TPQ designed by Cloninger2 which has been widely used in genetic studies. The TPQ is based on a biosocial theory of personality that draws on human and animal work to suggest that behavior is mediated by certain neurotransmitters which underlie three basic and largely heritable dimensions, called ‘Novelty Seeking (NS),’ ‘Harm Avoidance (HA)’ and ‘Reward Dependence (RD).’ Cloninger suggested that the trait of NS is mediated by genetic variability in dopamine transmission whereas HA is mediated by serotonergic transmission and RD by noradrenergic transmission. Research undertaken to test this hypothesis identified the dopamine D4 receptor gene (DRD4) exon III polymorphism3,4 as a candidate gene that may be associated with the NS trait.5,6 This polymorphism is an unusually variable repeat region in the third cytoplasmic loop of the receptor coding for 16 amino acids. The two most common repeats in most populations are the 4 and 7. However, reports concerning the association between the DRD4 exon III polymorphism and the personality trait of NS have yielded an inconsistent pattern. Whereas initial reports suggested that long alleles (mostly representing the 7 repeat since the 6 and 8 repeats are relatively rare) are associated with higher scores on the personality scale of NS, others have suggested that either the two are not associated or even the short alleles (usually 4 repeats) are associated with higher NS scores. This inconsistent pattern has been noted in several review articles which offered a narrative review of the pertinent reports.7,8 Yet, these narrative reviews were inconclusive as well.
An alternative to the narrative review is a meta-analysis that integrates all the available data numerically. A meta-analysis has three advantages over the narrative review. First, it can assess the effect (the strength of the association) in the population more efficiently than any of the original reports by utilizing the increase in sample size. Second, it can assess the variability of the effect in the population to determine whether inconsistent reports reflect mere statistical fluctuation (ie, no two reports are likely to yield identical results even if in the population the effect is constant, merely due to random sampling errors). Third, if it is determined that some of the between-studies inconsistency is not random, then one can investigate the potential moderators (or modulating factors) of the association under investigation. However, meta-analytic techniques have their own methodological limitations—including a potential publication bias against negative or non-significant results,9 mixing different types of variables as if they were measuring the same construct,10 and statistical biases in the meta-analysis estimates.11 Thus, our goal in this paper is to review the literature regarding the link between the DRD4 exon III polymorphism and NS utilizing the advantages of the meta-analysis, while safeguarding as much as possible against the potential limitations of the meta-analysis techniques.
To reap the maximum benefit from a meta-analysis, as many reports with similar methodology as possible should be reviewed. In the pertinent literature regarding the DRD4 polymorphism and NS, several types of research designs were used both with regards to the grouping of the genetic polymorphism and with regards to the specific instrument used for personality assessment. Comparison between genetic polymorphisms is sometimes assessed by comparing most common specific genotypes (eg, 4,4 vs 4,7) and at times by using a ‘short’ vs ‘long’ classification (2–5 vs 6–8 repeats). Since most studies used the short vs long scheme we used this classification scheme as well. Most of the literature has used Cloninger's TPQ instrument and only a few used the NEO-PI12,13,14 and the Karolinska Scale of Personality.15 Therefore, to ensure homogeneity (to avoid the problem of ‘mixing apple and oranges’10) we restricted our analysis only to those investigations employing the TPQ. The TPQ also measures other traits: HA and RD. If the DRD4 polymorphism has a unique association with NS, one would expect no association with the other traits. Hence, wherever reported we also analyzed the association between the presence of long vs short alleles with HA and RD.
We conducted a computerized search (Medline) with the following search criteria: ‘(Personality or Novelty Seeking or Extroversion) and (DRD4 or Poly- morphism)’. In addition, we used back references from identified papers to find as many pertinent papers as possible. We found a total of 19 relevant papers5,6,13,14,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30 (Table 1) yielding 20 different studies.
For each report we calculated the strength of the association. Given that the association is calculated between a dichotomous variable (long vs short allele) and a continuous variable (NS), the appropriate index of effect size is the statistics d that measures the standardized difference in mean NS scores between people who have and people who do not have the long allele. That is, in a given study a d of 1 means that the mean NS of those who have the long allele was one standard deviation higher on NS than the mean of those who do not have the long allele. A d of zero means that in a given study there was absolutely no mean difference in NS scores between those who carry the long allele and those who do not. Finally, a d of −0.5, for example, would mean that the mean NS of the carrier of the long allele was lower by half a standard deviation than the mean NS score of those who do not carry a long allele. Note that d can be converted to a correlation coefficient (Pearson's r) but the choice of meta-analysts is to work with the unit of measurement that is characteristic of the literature.
There are two slightly different meta-analytical approaches to combine research data. Whereas typically the conclusions produced by the various methods are similar, in cases where the effect sizes are small (the association exists but is weak), the methods can diverge in their conclusions.11 Of course, one's trust in the meta-analytic results will be enhanced if it is shown that the conclusion is independent of the methods. Thus, we have applied two methods that can potentially differ in their conclusions. Specifically, we used both the methods of Hedges and Olkin31 and those of Hunter and Schmidt.10 The differences in estimates are explained in the Results section.
For each study we computed d (Hunter & Schmidt's method) and d+ (Hedges & Olkin's method) from the reported means and standard deviations. Whenever a report did not provide means and standard deviations, but provided a t-test score, we computed d directly from t. Whenever a report provided means for subgroups of either people or sub-scales, we first averaged the means and variances of NS across the sub groups, or sub scales, to derive mean and standard deviation NS for those with and those without the long allele. We performed this aggregation with one exception because it provided two large samples; one for alcoholics and one for non-alcoholics (see coding below). When a study dichotomized NS to high/low score, we treated the proportion differences as mean differences by calculating standard deviation for proportions.
Finally, we coded for each study the following demographic variables that according to the literature could potentially moderate the association: mean age of the sample, proportion of males in the sample, whether the sample was based on alcoholics (1) or non-alcoholic population (0), and whether the sample was from continental Europe (1) or not (0). The later variable was chosen because most of the negative reports came from continental Europe.
Prior to performing the meta-analysis we explored the distribution of the literature with a stem-and-leaf display (Table 2). The results in the literature appear to be normally distributed without noticeable outliers that can bias the meta-analysis estimates. This table helps to obtain a bird's eye view that suggests that the entire literature is distributed between moderate-negative effects to moderate-positive effects (d < |0.2| is considered weak, |0.2| < d < |0.5| is moderate and d > |0.8| is strong32).
The distribution of the results in this literature (Table 2) suggests that this literature does not suffer from a serious publication bias because many reports were published despite reporting either negative effects or non-significant effects. This observation was further explored with a funnel plot9 with d values on one axis and standard errors on the other axis. As can be expected from the almost normal distribution shown in Table 2, the funnel plot was largely symmetrical ruling out a serious publication bias. This increases the confidence one can place in the meta-analytical results presented next.
Both meta-analytical methods suggested similar conclusions (Table 3). The mean difference in NS between those who carry the DRD4 long allele and those who do not carry it is negligible (about 1/20 of a standard deviation). The confidence interval for the mean includes zero (eg, the mean effect in the literature is not statistically significantly different from zero). However, both methods indicate that whereas the mean effect is close to zero, the differences between studies are due in part to true heterogeneity, ie, different studies estimating different ‘true’ underlying association parameters. Specifically, some studies find positive association and some studies find negative association and this divergence in results is not likely to be random (the Q-statistic is highly significant for the Hedges & Olkin method and according to the Hunter & Schmidt's method only 24.8% of the variability in results across studies could be attributed to random sampling error). Hence both methods strongly indicate that there are unknown moderators of the association between DRD4 polymorphism and NS.
Given the statistical evidence for the presence of moderators, we calculated the weighted correlation between the coded variables and d (Table 4). The weights for the correlations are based on the sample size of each effect.31 using the WEIGHT option of the readily available PROC CORR procedure in SAS.
All the demographic variables showed non-significant trends (weighted r’s) of attenuating the results (Table 4). The strongest trend is for samples from Continental Europe, which are most likely to yield negative association between the long allele and NS. Similarly, the older the mean age of the sample the weaker, or even negative, were the associations between the long allele and NS. The other potential moderators yielded non-significant and extremely small correlations.
Our results are consistent with a recent trend found in 36 meta-analyses that led to the conclusion that ‘The first study often suggests a stronger genetic effect than is found by subsequent studies’ (p 1).33 Indeed, the mean association between the DRD4 long allele, on the basis of data aggregated over almost 4000 individuals, is negligible and statistically not significant. This conclusion is strengthened by noting that the association between DRD4 polymorphism and NS is similar in magnitude (Table 3) to the association between DRD4 polymorphism and other traits (HA, RD; −0.14, 0.10, respectively). These were not the primary targets of investigation and yet yielded similar and numerically stronger effects. These could be considered as control variables that should show weaker effects than the one found for NS. Thus the pattern of results suggests, contrary to initial reports, and consistent with recent reviews8 that there is no simple association between DRD4 polymorphism and NS.
However, the variability in the direction and magnitude of the associations between DRD4 polymorphism and NS found in the literature indicates that there are as yet unknown causes for finding weak to moderate positive effects in some studies, null findings in some other studies, and weak negative effects in yet other studies. Some of this variability was estimated to be due to true heterogeneity between studies (ie, the true association between DRD4 polymorphism and NS varies in the population due to unknown causes) but the range in which the association is likely to vary is mostly in the range of small effects (d < |0.2|). Thus, the reasons for this variability may be very difficult to detect.
An important question is not whether an effect exists, because both weak positive and weak negative associations are likely to exist in the population, but rather whether the range of likely effects estimated in the current meta-analysis is meaningful. Specifically, the statistically true variability falls in a range that suggests that most true effects are very weak. For example, even if we assume, as the data suggest, that in some conditions the link between presence of long alleles and NS is about d = 0.16 (Hunter & Schmidt method, Table 3), or that the correlation coefficient between presence of long alleles and NS is 0.08, it is questionable whether practically these true effects justify a search for moderators. Detecting moderators of such a small effect size statistically requires very large data sets. However, note that weak statistical effects could be important but that the judgement is based not on statistics but on substantive knowledge of meaning of the effect.34 For example, if one translates findings regarding the association between smoking (yes/no) and occurrence of a heart attack (yes/no), the d statistics will be roughly 0.14. However, it corresponds to an odds ratio of 6:1, where 6% of smokers are expected to suffer from heart attack as opposed to 1% of non-smokers.34 In that case, the very small statistical effect size (most people whether smokers or not will not suffer heart attack) is very meaningful. The meaningfulness is independent of effect size. Thus, researchers in the field have to assess whether or not variable small true association between DRD4 and NS is theoretically meaningful. This may become clearer, perhaps, with the identification of other genetic correlates of NS.
We, nevertheless, attempted to identify potential moderators in the data set analyzed in the current meta-analysis. Four variables that might be contributing to differences in reports and that were identifiable in the published studies included the mean age of sample, ethnic groups (European vs others), gender and clinical characteristics of the sample (alcoholic vs non-alcoholic). None of the four moderators we analyzed proved significant although there was a trend for age and ethnicity to affect the purported association between the DRD4 polymorphism and NS. Given the small number of studies in this meta-analysis (k = 20), the moderator analysed results must be interpreted with great caution due to second order sampling error.10,35 That is, unlike the estimate of the association between DRD4 and NS that is based on approximately 4000 observations (high statistical power), the estimates of the moderators are based on 20 samples (extremely low statistical power).
One potential concern with this meta-analysis is the comparison of the long vs short allele. First, there is evidence of sequence variation within particular repeat lengths that may also play a role in receptor functioning. At least 35 distinct repeat variant motifs were observed for the DRD4 gene.36 The two most common haplotypes for the 4 and 7 repeat alleles respectively account for 95% and 89% of the observed sequence variants.36 Due to the infrequence of the additional haplotypes, it seems to us unlikely that they influenced our estimates of the association between NS and allelic type. However, this possibility cannot be excluded and one particular sequence variant of the 7 or 4 repeat allele may be responsible for the effect of one or both of these alleles on NS. Although no information is currently available concerning possible functional differences among variant repeats of the same length, it might be worthwhile to address this issue in future genetic and physiological studies.
Second, on the basis of a worldwide sample it is argued that the 7 repeat allele is evolutionarily younger than the common 4 repeat allele. Furthermore, on the basis of this report, it is suggested that the 7 repeat allele is not simply related to other common alleles and that it has increased to high frequency in humans by positive selection.36 Thus, the reports that collapse several alleles into one category of ‘long’ may be missing important information. However, in our meta-analysis 15 out of the 20 samples compared the specific effect of 7 repeat presence/absence on NS and yielded practically the same results (eg, mean d of these 15 studies is 0.01 and the heterogeneity estimates remain largely unchanged also). Nevertheless, future research regarding the DRD4 gene should consider these recent findings regarding the 7 repeat allele.
The failure to find a moderator is not surprising given the combination of small number of studies (k) and small range of effects (all d < |1|). To discover unknown moderators, future studies should search for an interaction between the presence of the long allele and other factors in affecting NS. These factors may be either genetic (eg, allelic or locus heterogeneity), environmental (eg, meaning of NS behavior in different cultures), or measurement related (eg, NS response may not be equivalent across cultures and languages).
Perhaps a more fruitful approach towards identifying factors that truly contribute to the variability in association studies between DRD4 and NS would be based on the highly polymorphic nature of this gene. At least 10 polymorphisms (outside of the repeat region) of the DRD4 receptor have been described to date and nine are discussed in a recent review by Paterson et al.8 The most recently reported polymorphisms are a group of SNPs (single nucleotide polymorphisms) in the promoter region, one of which (−521 C/T) is observed to regulate transcription of the receptor.37 This plethora of DRD4 receptor region polymorphisms raises the possibility that the reported association between the exon III polymorphism and NS is due to linkage disequilibrium between the long repeat and another one or more of the reported polymorphisms at this locus. Allelic heterogeneity can create situations in which multiple different alleles are associated with NS, rather than a single specific allele. Studying haplotypes can to some degree overcome hypothetical weak linkage equilibrium between the 7 allele and a true functional polymorphism.38 For example, a haplotype approach has been used to further analyze the role of the DRD4 receptor in attention deficit disorder. Indeed, such an analysis strengthens the connection between the gene and the disorder although the relationship between DRD4 and attention deficit is not simple.39
Additionally, in the case of personality traits locus heterogeneity is also likely to be present and many genes dispersed in the genome may be partially contributing to NS. For example, Cloninger et al40 in a linkage study of anxiety-related traits has identified a number of epistatic loci on different chromosomes. We have provisionally identified two additional functional polymorphisms that may be contributing to NS.41 It should also be noted that in Chinese and Japanese populations the long repeat alleles, including the 7 repeat, are extremely rare and cannot be contributing to NS. Indeed, in the Japanese population an association has been reported between NS and one of the promoter region polymorphisms.37
Human personality is a complex phenotype in which both genetic and environmental factors play a role. Moreover, similar to other complex traits it is likely that a number of genes, each yielding a small effect size, contribute to the phenotype and any one polymorphism may neither be necessary nor sufficient to determine the trait. Since different investigators draw on diverse populations in studies of personality, the genetic background of these populations likely varies for common polymorphisms, thus potentially confounding the identification of any specific genes of small effect size that contributes to personality traits. Validation of a role of polymorphisms in contributing to a complex trait may well depend on knowledge of additional genes contributing to this trait, and when studies are carried out across ethnic groups, on the frequency of these other polymorphisms across populations. The growing availability of powerful statistical techniques, high-throughput genotyping and large numbers of polymorphic markers such as single nucleotide polymorphisms makes such proposed studies increasingly feasible.
This research was supported by the Recanati Fund of the School of Business Administration at the Hebrew University (ANK) and a grant from the Israel Science Foundation founded by the Israel Academy of Sciences and Humanities (RPE). The authors wish to thank EM Berry and two anonymous reviewers for their useful comments on an earlier draft of this paper.
About this article
Self-Reported Sexual Behavioral Interests and Polymorphisms in the Dopamine Receptor D4 (DRD4) Exon III VNTR in Heterosexual Young Adults
Archives of Sexual Behavior (2016)