presents the characteristics of the 14 studies that were included in the meta-analyses. There were a total of 14 250 participants of whom 1769 were classified as having depression, and 12 481 did not have depression. Published data were used from 4 studies and original data were used from 10 studies. There was wide variation in the demographic characteristics (eg, proportion of women, age distribution) and sample sizes, which ranged from 118 to 4175 participants. Most of the participants were white except for a multiethnic sample in one study26
and an Asian sample in another study.30
Genotype distributions were similar among white participants; however, Asian individuals had higher S
allele frequency than white individuals. Nine of the studies used a structured interview to assess either DSM-IV
whereas the other 5 assessed depressive symptoms via self-rated symptoms scales.16,18,20,23,26
Stressful life events measures were assessed consistently across most of the studies using the Brugha List of Threatening Experiences,42,43
whereas a few studies included measures of somatic illness, unemployment, and social stressors. The final column of describes whether the findings replicated those of Caspi et al.10
Characteristics of Studies Included in Meta Analyses of 5-HTTLPR Genotype, Stressful Life Events, and Depression
presents the results of the meta-analysis of the relationship between depression and number of stressful life events for each study for both sexes combined (n=12 520). Although the estimates for the individual studies were highly variable, the results of the meta-analysis across all studies reveals a direct association between the number of life events and depression (1 vs 0 stressful life event: OR, 1.31; 95% CI, 0.98-1.75; 2 vs 0 stressful life events: OR, 1.95; 95% CI, 1.33-2.87; and ≥3 vs 0, stressful life events: OR, 3.21; 95% CI, 2.07-4.99). Sex-specific estimates show a similar pattern (eTable 1
available at http://www.jama.com
Odds Ratios for Major Depression by Number of Stressful Life Events for Individual Studies and Meta-analysisa
presents the results of the first logistic regression analyses for sexes combined for 14 studies for the effect of 5-HTTLPR genotype, the effect of the number of stressful life events, and the interaction between genotype and number of life events on risk of depression.
Logistic Regression Analyses of Risk of Depression for 14 Studies
The analysis shows no significant allele frequency difference between those with and without depression in any of the single studies or in the meta-analysis of all studies combined (OR, 1.05; 95% CI, 0.98-1.13; ). Thus, the genotype alone did not predict depression.
Similar to the results of for the specific number of life events, the meta-analysis shows that the number of stressful life events was significantly associated with depression (OR, 1.40; 95% CI, 1.25-1.57; ). Finally, in the analysis of interaction (), the ORs were significant for the studies of Caspi et al10
and Wilhelm et al27
but were not significant for any of the other replication studies. The aggregate estimate did not yield a significant interaction (OR, 1.01; CI, 0.94- 1.10) indicating that 5-HTTLPR
genotype did not interact with stressful life events on the risk of depression.
In the analysis by sex, no significant allele frequency difference existed for either sex between those with and without depression in any of the single studies or in the meta-analysis of all studies within either sex or for sexes combined (female sex: OR, 1.07; 95% CI, 0.96-1.18; male sex: OR, 1.00; 95% CI, 0.87-1.14, all participants: OR, 1.04; 95% CI, 0.96-1.13; ). Thus, the genotype alone did not predict depression.
Logistic Regression Analyses of Depression for 10 Studies Stratified by Sex
However, the number of stressful life events was significantly associated with depression in both sexes in 4 studies,15,16,20,25
and in 2 studies of females ().17,19
The meta-analysis showed a significant association for females alone (OR, 1.39; 95% CI, 1.23-1.58), males alone (OR, 1.43; 95% CI, 1.16-1.75), and for all participants (OR, 1.41; 95% CI, 1.27-1.57). Assuming a normal distribution for the β coefficient (log of the OR), this estimate is statistically significant, with a Z
score of 6.35 and a P
value of <.001.
Finally, in the analysis of the interaction between genotype and number of life events on risk of depression, the regression coefficient was not significant for females in any of the studies and was nominally significant for males in 2 of the 10 studies.16,27
The meta-analysis showed no interaction effect for either females (OR, 0.95; 95% CI, 0.86-1.04), males (OR, 1.02; 95% CI, 0.86-1.20), or both sexes combined (OR, 0.98; 95% CI, 0.90-1.07). Thus, there was no evidence that the serotonin transporter genotype alone or in interaction with stressful life events is associated with an elevated risk of depression in males alone, females alone, or both sexes combined. The only significant finding across studies was the potent association of stressful life events with the risk of depression.
When examining the studies for the possibility of heterogeneity of effect—consistent with , eTable 2, and eTable 3
available at http://www.jama.com
)—we found no evidence of heterogeneity for the genotype alone (P
=.68); however, we did find a significant degree of heterogeneity for life events alone (P
<.001). Heterogeneity of the interaction was also not significant (P
>.05), and none of the individual interaction coefficients was significantly deviant from the overall average.
eTable 2 and eTable 3
show the results of another series of meta-analyses that test gene-environment correlation, gene-environment interaction, and whether stratification of those with and without depression by the environmental risk factor (stressful life events) would enhance the power to detect an association between the 5HTTLPR
genotype and depression. We divided the samples of those with and without depression into strata defined by the number of stressful life events and examined the difference in the frequency of the S
alleles between those with and without depression within each stressful life events stratum. eTable 3
presents these frequencies further stratified by sex.
We first examined the association between allele frequency and number of stressful life events within those without depression, our proxy for testing for gene-environment correlation in the population. No significant relationship was found between genotype and number of life events among those without depression for both sexes combined (eTable 2
) or for males or females alone (eTable 3
) in most of the individual studies, or in the metaanalysis (β, −0.003; SE, 0.014). Likewise, there was no association between stressful life events and allele frequency within those with depression either in the individual studies, except for the studies of Caspi et al10
(positive regression) and Cervilla et al15
(negative regression), or in the meta-analysis (β, −0.004 ; SE, 0.034). However, the meta-analysis yielded no significant differences in the β coefficients between those with and without depression across studies (β, −0.001; SE, 0.047) and sexes, indicating a lack of gene environment interaction.
In the test of whether stratification by exposure to life events would enhance power to detect an association between depression status and genotype, we found no relationship between number of stressful life events (the exposure level) and allele frequency difference between those with and without depression as indicated by the δ coefficient. Similar findings occurred for females alone and males alone (eTable 3
In fact, the largest allele frequency difference occurred at 1 life event (δ, 0.045; SE, 0.020), which was of nominal significance, but none of the differences in any of the strata was statistically significant after adjusting for multiple testing. Furthermore, the findings of an increase in the S frequency at 1 life event is clearly at odds with the original findings of Caspi et al10
), for which the allele frequency difference is in the opposite direction (δ, −0.091; SE, 0.069). Therefore, in this case, stratifying the allele frequency test based on the environmental exposure did not provide additional support for a genetic association.
All of the above analyses involving a dosage model or allele frequencies were repeated for genotypes, for which either the SS genotype was compared with the SL plus LL genotypes (recessive model) or the SS plus SL genotypes were compared with the LL genotype (dominant model). None of these analyses provided evidence different from that presented above based on the original models.