Molecular Psychiatry (2004) 9, 197–202 & 2004 Nature Publishing Group All rights reserved 1359-4184/04 $25.00 www.nature.com/mp ORIGINAL RESEARCH ARTICLE A meta-analysis of the association between the serotonin transporter gene polymorphism (5-HTTLPR) and trait anxiety JA Schinka1,2, RM Busch1,3 and N Robichaux-Keene1,4 1Veteran’s Administration Medical Center, Tampa, FL, USA; 2Department of Psychiatry, University of South Florida, Tampa, FL, USA; 3Department of Psychology, University of Cincinnati, Cincinnati, OH, USA; 4Department of Psychology, Baylor University, Waco, TX, USA Studies of the association between polymorphisms of the serotonin transporter gene (5-HTT) and trait anxiety have produced inconsistent results, raising questions about the strength of the relationship and the methodological conditions under which the relationship holds. We conducted a meta-analysis of existing studies to provide formal statistical measures of the strength of the linked polymorphic region of the serotonin transporter gene (5-HTTLPR)– anxiety relationship. For the entire collection of 26 studies, results provided no support for a relationship between anxiety and the presence of the short form of the 5-HTTLPR polymorphism. There was strong evidence of the presence of moderating variables, however, and subsequent analysis revealed that choice of the measure of trait anxiety was significant. Studies using the Neuroticism scale of Costa and McCrae were found to produce a small positive effect (d¼0.23). Other potential moderators (country of study origin, type of subject) did not have a meaningful impact on d statistics. These findings indicate that 5-HTTLPR may in fact have a small but reliable influence on personality, particularly in the manifestation of trait anxiety when measured with a neuroticism scale based on the five-factor model of personality. Our results suggest that the success of future personality genetics research will be maximized by the use of personality measures from both the psychobiological and five-factor models. Molecular Psychiatry (2004) 9, 197–202. doi:10.1038/sj.mp.4001405 Keywords: personality; genetics; questionnaire; association study; serotonin; meta-analysis Introduction numerous 5-HTT-anxiety studies have been reported, with most examining the LPR polymorphism. These Twin studies have consistently demonstrated that studies have produced a large set of both positive and individual differences in reliably measured dimen- negative results and have generated extensive com- sions of personality are substantially influenced by ment about the strength of the relationship and genetic factors.1,2 Only in the past few years, however, the methodological conditions under which the have specific genetic polymorphisms been identified relationship hold.5–8 To date, however, no formal as factors influencing specific personality traits. In statistical analysis has addressed the collection of 1996, Lesch et al3 examined the relationship between studies examining the relationship between 5- the linked polymorphic region of the human seroto- HTTLPR and trait anxiety. In this study, we con- nin transporter gene (5-HTTLPR) and anxiety-related ducted meta-analyses of existing studies to provide traits (which we will refer to throughout as trait formal statistical measures of the strength of this anxiety) in a sample of healthy American volunteers. relationship. They found a small (d¼0.28) but significant relation- ship between Neuroticism, a measure of trait anxiety 4 contained in the revised NEO Personality Inventory, Methods and 5-HTTLPR. Specifically, individuals with the short form of 5-HTTLPR (s/s or s/l) had higher scores Studies on Neuroticism than individuals with the homozy- We conducted a search for all available studies of the gous long form (l/l). Since this original finding, association between 5-HTTLPR and measures of anxiety in adults by examining journal abstracts available through March 1, 2003 in the following Correspondence: JA Schinka, PhD, VA Medical Center/116B, databases: PubMed at the National Library of Medi- 13000 Downs Blvd, Tampa, FL 33612, USA. E-mail: [email protected] cine, PsycInfo, Science Direct, and MedLine. The Received 07 April 2003; revised 23 May 2003; accepted 12 June search identified 33 studies, seven of which were 2003 excluded from the analysis. We excluded one study9 5-HTTLPR and anxiety JA Schinka et al 198 because the same data are analyzed in two earlier ‘anxiety factor’ score by averaging scores across the reports5,10 that are included in the meta-analysis. We scales. excluded three studies11–13 because they examined allele and/or genotype frequencies among individuals Results scoring at high and low levels on an anxiety measure, rather than testing differences in trait anxiety scores Table 1 presents descriptive information and effect between genotype groups. One study14 was not sizes for 28 samples described in 26 separate studies, included because sufficient descriptive data were comprising 7657 subjects, which examined the not provided to allow independent calculation of relationship between anxiety and the presence/ effect sizes. Finally, we excluded two studies15,16 absence of the short genotype. The range of d values because they did not provide clearcut or well-studied across samples and studies is quite substantial (À0.31 measures of anxiety-related traits. The meta-analysis to 0.58). Eight (29%) of the d values are negative in was therefore conducted on results provided in 26 direction and the majority (19 of 28, 68%) are less published studies.3,5,6,10,17–38 As one study presented than 0.2. data for multiple samples, 28 samples were available Table 2 presents the results of the meta-analysis for for analysis. the entire collection of studies. The mean difference Each article was coded according to the following: in measured anxiety between short and long groups is (1) reference, (2) scale or test used as a measure of an negligible (d¼0.10, 1/10 of a standard deviation). The anxiety-related trait, (3) type of subject (normal vs. confidence interval for the mean includes zero, patient), (4) country of sample origin, (5) sample size indicating that the mean effect is not significantly by genotype, (6) mean scores for the trait anxiety different from zero. However, the set of d values is measure by genotype, and (7) standard deviation of characterized by a substantially large standard devia- scores by genotype. tion (0.18) and a relatively small estimate for Although providing data on only a single sample, sampling error (Vare¼0.001). According to the Hunter several studies used more than one measure of trait and Schmidt method, the ratio of sampling error anxiety. In all, 13 studies used the harm avoidance variance to the variance of d (Vare/Vard) provides an (HA) scale of either the Tridimensional Personality estimate of the variability in results across studies Questionaire40 (TPQ) or the Temperament and Char- that can be attributed to sampling error. For the entire acter Inventory41 (TCI) as a measure of trait anxiety, collection of studies, only approximately 2% of the while 11 studies used the Neuroticism (N) scale of the variability in results across studies can be attributed revised NEO Personality Inventory (NEO-PI-R). Three to sampling error. In addition, a formal test of the studies used the Neuroticism scale of either the assumption of homogeneity produced a significant Eysenck Personality Questionnaire42 or the Eysenck result (Q¼171.92, df¼27, Po0.001), indicating het- Personality Inventory.43 Two studies21,32 provided erogeneity in the distribution of d values. These data for all of the anxiety scales of the Karolinska results argue strongly for the presence of unknown Personality Inventory.44 moderators of the association between 5-HTTLPR and The meta-analysis was conducted on effect sizes trait anxiety. and the method of analysis was guided by the work of Given the statistical evidence for the presence of Hunter and Schmidt.45 Effect size (d) was calculated moderators, we calculated additional d values for four as the difference in anxiety measure means between possible factors: studies with extreme d values, the groups (ie, s/s and s/l vs l/l genotypes) divided by the country of origin of the samples, type of subject pooled standard deviation of the two groups. All ds (normal vs patient), and the measure of trait anxiety. were calculated so that a higher mean for the Examination of summary statistics in Table 2 suggests combined s/s and s/l (short) group was reflected in a that the results are only minimally influenced positive value for d.Ad of 1 thus means that the short materially by studies with the most extreme d values group had a mean that was one standard deviation (À0.31 and 0.58), as the d statistic changes little by higher than for the l/l (long) group. A d of zero would exclusion of those studies. The d statistic for the 13 indicate that there was no difference in means studies conducted with samples from the USA does between the short and long groups. By convention, increase over that for all studies, but the confidence ds of |0.2|, |0.5|, and |0.8| are considered to be interval continues to include zero, indicating that the indications of small, medium, and large effect sizes, mean effect is not significantly different from zero. respectively.46 Exclusion of studies using patient subjects, leaving Several studies did not provide means and standard only studies of normal adults, also has little impact on deviations for the short and long groups, but did the d statistics, as might be expected from the fact that provide correlation coefficients or the results of t or F over 80% of the subjects were from normal samples. tests. In these cases, d was calculated from r, t,orF, The moderator analyses do indicate a substantial using standard formulas.45 For studies reporting the and significant effect, however, for the choice of results of more than one measure of trait anxiety, we anxiety measure.