The serotonin transporter gene promoter polymorphism (5-HTTLPR) has been repeatedly associated with antidepressant response in mood disorder patients, but findings are not consistent across studies. A meta-analysis was performed on 15 studies including data of 1435 subjects. We tested three phenotypes: remission rate, response rate and response rate within 4 weeks using the cochrane review manager. We observed a significant association of the s/s variant of 5-HTTLPR with remission rate (P<0.0001) and both s/s and s/l variants with response rate (P=0.0002). Response rate within 4 weeks was associated in both models (P=0.003–P<0.00001). This effect is quite robust to ethnic differences although a significant heterogeneity is present in Asian samples.
Mood disorders have a large impact on social health, with considerable both direct and indirect costs.1, 2, 3, 4 Selective serotonin reuptake inhibitors (SSRIs) treatment reduced their morbidity with a favorable side effect profile. Unfortunately, not all individuals benefit from treatment, and 30–40% of patients do not show a complete response to treatment.5, 6
The first step of SSRIs action is to inhibit the serotonin transporter (5-HTT) and thus modulate the serotonergic activity. The human gene encoding 5-HTT is located on chromosome 17q–11.1-q12, it spans 31 kb and consists of 14 exons. Heils et al.7 reported a functional polymorphism in the transcriptional control region upstream of the 5-HTT coding sequence (5-HTTLPR). Since then, this polymorphism has been the most widely studied in psychiatric genetics for its high number of effects.8 Of particular interest is the association with SSRI efficacy. Our group performed the first studies, and tested the hypothesis that 5-HTTLPR was related to the response to fluvoxamine and/or augmentation with pindolol in 102 inpatients with major depression.9 The carriers of the long allele showed a better response to fluvoxamine compared to those who were homozygous for the short allelic variant. Following this pilot study, the association between 5-HTTLPR and response in patients with mood disorder has been investigated by a large number of studies, but findings are not always consistent. Significant associations between the long variant and a good response have been reported, but other studies did not confirm these findings.10, 11
However, no article has pooled and analyzed all the data from studies testing associations of 5-HTTLPR with treatment response. The purpose of this study is the reevaluation of the association between 5-HTTLPR and clinical response to SSRIs treatment in patients with mood disorders pooling existing studies through a meta-analysis.
Materials and methods
To identify studies eligible for this meta-analysis, we searched Medline for all publications available up to July 2006 studying the association between SSRIs treatments and 5-HTTLPR in depressive patients with the key words affective, depression, mood, treatment response, serotonin transporter promoter region (SERTPR) and 5-HTTLPR. We also used reference lists from identified articles and reviews to find additional articles not indexed by MEDLINE. Studies were included in the current meta-analysis if they evaluated the association between clinical response to SSRIs treatments and 5-HTTLPR in patients diagnosed with major depressive disorder or bipolar disorder according to DSM criteria. Studies were excluded from the analysis if outcome was not evaluated as response or remission rate on a depression scale, and studies with overlapping patient samples were excluded to only include the study with the large number of patients. Authors were contacted when data were not reported in the article. Hardy–Weinberg equilibrium was examined in studies where genotype frequencies were included.9, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25 Given the lack of unequivocal data for 5-HTTLPR genotype pooling, we tested both dominant and recessive hypotheses: l/l versus l/s-s/s and l/l-l/s versus s/s. Outcome was defined with three phenotypes: remission rate, response rate, and response rate within 4 weeks. Subjects with pindolol augmentation were excluded given the observed influence of pindolol on the 5-HTTLPR effect.9
Remission was defined as a final Hamilton Rating Scale for Depression (HAM-D) total score of 7 or less. Response was defined as at least 50% decrease in HAM-D or Montgomery and Asberg Depression Rating Scale (MADRS) total score. In one study,17 response was defined as a 60% or greater decrease in MADRS. Remission is a quite robust outcome definition but response is more sensitive although quite non-specific. Moreover, the length of evaluation of response varied largely between studies. We therefore investigated separately response within 4 weeks of treatment and response at any length of observation (2–12 weeks). We focused on the response rate within 4 weeks because it is a sensitive measure to evaluate speed of response. Many studies presented the outcome of 4 weeks, while only three studies presented outcome within 2 weeks, and 6 weeks was too a long period to investigate speed of response.26
Remission was instead calculated at 6 weeks when possible, and in defect of data we used different observation lengths. When outcome data were not available in the article, we requested data from the author or estimated the number of response and remission from the presented figures. Data were entered into the Cochrane Collaboration review manager software (RevMan version 4.2) and analyzed by RevMan analysis 1.01. Heterogeneity between the studies was assessed with χ2 test. Individual and pooled odds ratio (OR) and associated 95% CIs were calculated. A fixed-effect model was used in all analyses. We used the fixed-effect despite a moderate heterogeneity across studies given that we had no a priori reason to hypothesize data coming from different populations and because of the main aim of the present analysis being the identification of the best estimate of a single effect size more than the range of effect sizes across populations.12
The literature search and selection produced 19 studies,9, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 33 but only 15 studies were included in current meta-analysis, as listed in Table 1. Studies by Kraft et al.28 and Minov et al.30 were excluded because they included patients treated by a mixture of antidepressants. The paper by Lee et al.29 was not included because it did not use HAM-D or MADRS but Clinical Global Impressions of Improvement (CGI) score. The paper by Murphy et al.31 was not included because data were not available. Two studies17, 19 used the DSM-III-R diagnostic criteria and all the other studies applied the DSM-IV diagnostic criteria.
Analysis in the whole sample
Figure 1 presents ORs for the individual studies and the pooled analyses in the l/l and l/s genotype versus the s/s genotype subjects for the remission rate, response rate and response rate within 4 weeks. All studies except two showed an OR higher than 1. The pooled OR of seven studies of remission rate including 745 subjects was highly significant (2.21, CI=1.53–3.21, P<0.0001) and there was no evidence for heterogeneity across studies. Pooled OR of nine studies of response rate including 944 subjects was not significant (1.20, CI=0.90–1.60, P=0.21) but with a significant heterogeneity across studies (P=0.001). When we considered only the five studies reporting response rate within 4 weeks including 633 subjects the effect of the 5-HTTLPR s/s genotype was again significant (1.72, CI=1.20–2.47, P=0.003) with no heterogeneity across studies.
We then pooled the l/l genotype versus the l/s and s/s genotype and results were presented in Figure 2. Pooled OR of seven studies of remission rate including 745 subjects was 1.42 (CI=0.98–2.04, P=0.06) with no heterogeneity across studies. On the other hand, that of 10 studies of response rate including 1031 subjects was 2.01 (CI=1.39–2.89, P=0.0002) with heterogeneity across studies, whereas the effect in seven studies of response rate within 4 weeks including 771 subjects was significant (OR=2.57, CI=1.70–3.88, P<0.00001) without heterogeneity across studies.
To investigate whether ethnicity played a role in confounding the association between 5-HTTLPR and SSRIs response, we examined studies with Caucasian and Asian subjects separately.
Analysis in Caucasian subjects
The pooled analyses and OR in the l/l and l/s genotype versus the s/s genotype within Caucasian subjects are presented in Figure 3. There was no evidence for heterogeneity and all OR were higher than 1. Pooled OR of five studies of remission rate including 544 subjects was highly significant (2.37, CI=1.56–3.58, P<0.0001), while the pooled OR of four studies of response rate including 345 subjects (1.53, CI=0.90–2.59, P=0.11) and the one of two studies of response rate within 4 weeks including 208 subjects (1.37, CI=0.65–2.88, P=0.40) were not significant, although greater than 1. Figure 4 presented the pooled analyses in the l/l genotype versus the l/s and s/s genotype. Pooled OR of five studies of remission rate including 544 subjects was not significant (1.37, CI=0.93–2.00, P=0.11), whereas the pooled OR of five studies of response rate including 432 subjects (1.74, CI=1.10–2.76, P=0.02) and those of four studies of response rate within 4 weeks including 346 subjects (1.75, CI=1.07–2.88, P=0.03) was significant. Overall in Caucasians, we observed an association of the s/s genotype with nonremission and of the s containing genotype with nonresponse but with some reduction in the significance possibly due to the smaller number of studies included.
Analysis in Asian subjects
Figure 5 presented results for Asian subjects. Pooled OR of two studies of remission rate including 201 subjects demonstrated a nonsignificant trend of favorable response to SSRIs in the l/l or l/s genotype carrier (1.70, CI=0.74–3.93, P=0.21) with no evidence for heterogeneity. Pooled OR of the five studies of response rate including 599 subjects was not significant (1.09, CI=0.78–1.53, P=0.62) but with evidence for large heterogeneity (P<0.0001). In fact the neutral OR of 1.09 is due to the sum of three studies showing a worse response effect of the s/s genotype and two studies reporting a protective effect of the s/s genotype. However, analyzing response rate within 4 weeks there was no evidence for heterogeneity across three studies including 425 subjects and pooled OR was significant in the same direction of Caucasians with the s/s genotype associated with poor response (1.85, CI=1.22–2.79, P=0.004). The alternative pooling of the l/l genotype versus the l/s and s/s genotype are presented in Figure 6. Pooled OR of the two studies of remission rate including 201 subjects was not significant (2.28, CI=0.59–8.86, P=0.23). However, pooled OR of the five studies of response rate including 599 was significant (2.52, CI=1.37–4.62, P=0.003) with large heterogeneity, whereas pooled OR of the three studies of response rate within 4 weeks including 425 subjects was highly significant (5.96, CI=2.70–13.17, P<0.0001) with no heterogeneity.
In this study, we retrieved 15 studies that included data from 1435 subjects to evaluate the association between 5-HTTLPR and clinical response to SSRIs treatment in patients with mood disorders. The results of our study indicate a significant association between 5-HTTLPR and clinical response in both remission rate and response rate. These findings suggest that 5-HTTLPR could be a predictor of response to SSRIs treatment.
The results also showed a more robust effect of 5-HTTLPR for remission rate when comparing the l/l and l/s versus the s/s and for response rate when comparing the l/l versus the l/s and s/s pooled genotypes. Nevertheless both comparisons showed a similar direction of effect. It is interesting to note that the effect of 5-HTTLPR for response within 4 weeks was the most robust and consistent in all comparisons except one analysis in Caucasian with some reduction in the significance due to the smaller number of studies included. This could suggest a major effect on speed of response of both s/s and s/l genotypes, while only the homozygote s/s effect could influence the overall remission rate. The dominance of 5-HTTLPR has not been addressed unequivocally, even though a dominant s allele effect has been repeatedly reported34, 35, 36 other studies did not confirm this model37 and SNPs are usually supposed to have a codominant effect.38 Our results favour the view of a dominant s effect, as indicated on response and early response. However, the remission effect was more clear with both s allele copies (s/s). We repeated the analysis excluding the first published study, which may inflate the overall estimate of effect to test for publication bias, but the outcome was similar to those with all together. OR in the l/l and l/s genotype versus the s/s genotype were 2.06 (P=0.0003) in the whole population and 2.18 (P=0.0005) within Caucasian whereas that in the l/l versus l/s and s/s genotype were 1.26 (P=0.23) in the whole population and 1.20 (P=0.38) within Caucasian. Consequently, the publication bias should have not influenced our results.
There are several differences among the studies included in this meta-analysis to discuss that could explain the significant heterogeneity we observed in some comparisons: different kind and dose of SSRIs, different subtype of affective disorder, different ethnic populations, and different lengths of assessment. Moreover other SNPs within the gene could be the causal variant and this might lead to the observed differences.39, 40 A wider list of polymorphism should be investigated to achieve a better resolution given that other control regions may be present for the 5-HTT expression, not to mention possible enhancers or silencers located in other regions far from the gene locus. In fact, Hamilton et al.28, 41 reported a significant association of a functional SNP (rs25531)42, 43 located just upstream of the 5-HTTLPR with antidepressant response to fluoxetine treatment and in linkage disequilibrium (LD) with 5-HTTLPR. In the presence of the g allele of this SNP, the l allele of 5-HTTLPR seems to be associated with nonresponse, while this is the case for the s allele in presence of the a allele of the SNP. In partial agreement, we found the same polymorphism influencing response, although in a different direction.44 These effects may be one of the sources of heterogeneity between studies.
Ethnicity and length of assessment might have also played a role in determining the heterogeneity among studies. Different allele frequencies between Caucasians and Asians, the s allele being present in 42% of Caucasians, but in 79% of Asians,45 are a strong cause of heterogeneity. In fact, when Caucasians were analyzed separately, the heterogeneity disappeared but it still remained when Asians were analyzed alone. The reason of this large amount of heterogeneity in Asian population could not be understood well, however, the length of assessment could be one of the sources. Observation length ranged from 2 to 12 weeks, this wide range mostly influences the outcome in term of response given its lower robustness compared to remission. In fact, there was no evidence for heterogeneity among studies that assessed response rate within 4 weeks both in Caucasians and in Asians. Indeed some evidence suggest that the association between 5-HTTLPR and treatment response is due to acceleration, more than overall response rate.24, 32, 46 The results of our study suggest the period of assessment could be an important factor to evaluate the response rate, while on the other hand the remission rate seemed less influenced by the period of assessment because our results showed no heterogeneity in all remission comparison. This could be due to the fact that remission was evaluated in a longer length of observation: only two studies within 4 weeks.23, 24 Finally, as previously discussed, a causative SNP located nearby the 5-HTTLPR could be the cause for association in the opposite direction in Asians.
We used a fixed-effect model for the meta-analysis. The two populations gene frequency differences may not fully justify the a priori population independence, as well as other minor methodological differences in the studies. Moreover the discrepant Asian results argue against a true population independence from Caucasians. Therefore, we presented the fixed-effect model to address the best estimate of a single effect size. However, the differences between the fixed-effects and the random-effects models were minimal in this analysis (data not shown).
To the best of our knowledge, our study is the first article investigating the association of 5-HTTLPR with response to SSRI treatment with a meta-analytic technique. Smits et al.11 collected and analyzed studies about association of 5-HTTLPR with treatment in depression. They result in a somewhat less favorable effect of SSRIs among Caucasian population with the s/s variants of the 5-HTTLPR as opposed to those with the l/l and l/s variant and no available evidence in Asian population. Our results were consistent with their ones, the conceivable difference between our study and theirs is that they analyzed a smaller number of studies focusing on the mean decrease in HAM-D and MADRS score. Another partial meta-analysis has been reported but in the context of a wider review and including only a small part of the studies here included and with a large heterogeneity of treatments.47
In conclusion, our meta-analysis confirmed the significant association of the l variant of 5-HTTLPR with a better response to SSRIs and this effect seemed independent from ethnic differences. The subjects with s/s genotype have difficulties to reach remission, and take a long time, over 4 weeks, to respond as well as the subjects with s allele take a long time to respond. The pooled OR in this meta-analysis resulted in up to 2.57 in all populations that seems a moderate effect and it is in line with minor effect genes.48 Other genes are expected to influence this complex trait and a comprehensive knowledge could enable clinical use of a gene profile as predictor to allow identification of the best therapeutic tools and avoid lengthy treatment trials.
About this article
The Pharmacogenomics Journal (2017)