Meta-analysis of the interaction between serotonin transporter promoter variant, stress, and posttraumatic stress disorder

Exposure to stress predicts the occurrence of posttraumatic stress disorder (PTSD) in individuals harboring the serotonin transporter promoter variant 5-HTTLPR. We carried out a meta-analysis of studies investigating the interaction between 5-HTTLPR, stress, and PTSD to clarify the interrelatedness of these factors. We reviewed all relevant studies published in English before May 2016. The Lipták–Stouffer z-score method for meta-analysis was applied to combined data. The z score was separately calculated for the stressful life events, childhood adversity, bi- and triallelic loci, and cross-sectional and longitudinal studies subgroups. A total of 14 studies with 15,883 subjects met our inclusion criteria. We found strong evidence that the presence of 5-HTTLPR influenced the relationship between stress and PTSD (P = 0.00003), with the strongest effects observed in the cross-sectional and longitudinal groups (P = 0.01 and 2.0 × 10−6, respectively). Stressful life events and childhood adversity separately interacted with 5-HTTLPR in PTSD (P = 2.0 × 10−8 and 0.003, respectively). When the studies were stratified by locus classification, the evidence was stronger for the triallelic (P = 4.0 × 10−8) than for the biallelic (P = 0.054) locus subgroup. There was strong evidence that 5-HTTLPR influences the relationship between stress and PTSD.

identified 18,19,22 , including a polymorphism in the serotonin (5-hydroxytrptamine, 5-HT) transporter (5-HTT) gene (SLC6A4) promoter region 18,23 . SLC6A4 has been reported to modulate sensitivity to stress and susceptibility to psychopathology 24 . SLC6A4 promoter is modified by elements within the proximal 5 regulatory region known as the 5-HTT gene-linked polymorphic region (5-HTTLPR) 25 , which contains a polymorphism with a rare short (S) and more common long (L) allele; the former is associated with reduced transcriptional efficiency of the promoter 26 , which has been linked to suicidal behavior 27 , depression 28 , and PTSD 29 . 5-HTTLPR variants have a third functional allele: Lg contains an A > G polymorphism at position 6 of the first two 22-bp imperfect repeats that define the 16-repeat L allele (the single nucleotide polymorphism rs25531) 30 . Thus, 5-HTTLPR is a triallelic locus from which Lg and S alleles (both reclassified as S') are expressed at similar levels 13 and La (reclassified as L') has higher expression 31,32 .
Two recent meta-analysis assessed the set of studies exploring the association between 5-HTTLPR and PTSD and concluded that evidence did not support the presence of the association 47,48 . However, none of two meta-analyses has assessed the interaction between 5-HTTLRP and stress type in PTSD. In order to clarify the role of SLC6A4 in the etiology of PTSD, we sought to carried out a meta-analysis of published studies on the relationship between 5-HTTLPR, stress, and PTSD. Specifically, three subgroup analysis stratified by type of stressor, study design, and locus classification was first used to detect the interaction effect and sensitivity analysis was conducted to detect publication bias in overall and subgroup analysis. The various studies employed different study designs, making it difficult to merge the results into a single traditional meta-analysis. The Lipták-Stouffer z-score, which is useful in situations where equivalent raw data are not available across relevant studies 49 , has been used to combine P values from many studies of gene and environment interaction [50][51][52] . Here we used the Lipták-Stouffer z-score to combine information at the level of significance tests to assess whether variations in 5-HTTLPR influences the relationship between stress and PTSD.

Materials and Methods
Studies. In accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines 53 , we identified candidate studies by examining previous meta-analyses and review articles retrieved from PubMed, Wolters Kluwer, Web of Science, EBSCO, and Elsevier Science Direct from study inception up to May 2016 using the following search terms without restrictions: "posttraumatic stress disorder" or "PTSD", "serotonin transporter gene" or "5-HTTLPR", "stress", "trauma", and " gene-environment interaction" 47,48 . In addition, we carried out related searches online using Baidu Scholar. We also examined references in prior meta-analyses and review articles to identify eligible publications.
Two investigators independently reviewed the remaining articles to establish eligibility based on predefined inclusion criteria. We included only human studies published in English before May 2016 that investigated the effect of 5-HTTLPR on the relationship between PTSD and stress, or the relationship between 5-HTTLPR polymorphisms and stressful events and/or childhood adversity. The genotype distributions in the studies were in Hardy-Weinberg equilibrium. One study was excluded because it used the same data as another study included in the analysis 54 . A total of 14 independent investigations with 15,883 study subjects met the inclusion criteria.
To ascertain whether results were affected by study design characteristics, we analyzed subgroups based on three variables, some of which have been defined in recent review articles 51,52 . We stratified studies by type of stressor (stressful life events and childhood adversity), study design (cross-sectional and longitudinal), and by locus classification (bi-and triallelic loci).
Quality assessment. The methodological quality of eligible studies was evaluated according to a quality checklist derived from Strengthening the Reporting of Observational Studies in Epidemiology checklists 55,56 , which have been used in some gene-environment interaction meta-analyses [50][51][52] . In accordance with current guidelines and prior studies [50][51][52] , included studies were not weighted by quality scores or excluded based on low scores. We nonetheless describe data quality in Supplementary Materials of Table S1 for readers to evaluate. P value extraction. Two authors independently extracted P values from each included study without divergence. If a study did not report an exact statistical outcome (e.g., the article stated only P > 0.05), the authors were contacted to obtain more precise values. If that was unsuccessful, a P value of 1 (indicating a lack of outcome) was assigned. In some instances, several P values were reported due to differences in PTSD scales or sample subsets among studies. Therefore, weighted mean P values were used in our analyses. If reported analyses corresponded to different groups, the mean of P values for each group were incorporated into the overall analysis.

Statistical analysis.
After combining eligible studies, we applied the Lipták-Stouffer z-score to obtain an aggregate value based on the significance level of tests weighed by sample size. We first converted extracted P values to one-tailed metrics where P values Z<0.50 indicated greater sensitivity to S/Lg allele stress and those >0.50 corresponded to greater sensitivity to L/La stress in PTSD. We then converted these P values to z scores, with positive and negative z scores corresponding to P values less and greater than 0.05, respectively. The z scores were incorporated into the following formula: where the weighting factor W i corresponds to the study sample size; Z i is the study z score; and k is the total number of studies. Z w conformed to a normal distribution and the corresponding probability was obtained from a standard normal distribution table. This statistical procedure was applied to all studies and to the stratified analyses.
To determine whether any single study had a disproportionate influence on our results, we carried out sensitivity analyses by computing Z w after removing each study in turn. To gauge publication bias, we calculated the fail-safe N for the overall analysis and for each stratified analysis, since unreported studies could potentially influence the conclusions of our meta-analysis. Consistent with previous reports 50, 51 , we counted the number of studies with an assigned P value of 0.50 and the average sample size of studies incorporated into the weighted Lipták-Stouffer analysis that yielded a non-significant result. The ratio of fail-safe N to the number of published studies provided an estimate of publication bias in our results.
Date availability statement. The authors declare that the data in this research is available.

Results
Literature search results. The study selection procedure is shown in Fig. 1. We identified 1644 potentially relevant records through literature searches; 403 were duplicated articles and were excluded. After screening titles and abstracts, 1113 additional articles were excluded, leaving 128 full-text articles for eligibility assessment. We ultimately included 14 studies in our meta-analysis that fulfilled all inclusion criteria. The characteristics of these studies are shown in Table 1 and Supplementary Materials of Table S2.
Overall meta-analysis. The 14 studies (with a total of 15,883 study subjects) were pooled to assess the interaction between 5-HTTLPR polymorphism, stress, and PTSD. We found strong evidence that 5-HTTLPR influences the relationship between stress and PTSD, with the S allele associated with an increased risk of developing  (Fig. 2). The results remained robust when each study was removed in turn from the overall analysis-that is, the overall P values remained significant (3.0 × 10 −5 < P < 0.0001) ( Table 1).
With respect to publication bias, a non-significant result (P = 0.05) in the overall analysis would require more than 226 unpublished or undiscovered studies with an average sample size of n = 1135 and non-significant results (P = 0.50), corresponding to a fail-safe ratio of 16 studies excluded from the meta-analysis for every included study.
Previous studies provided evidence for distinct interaction effects of child adversity and stressful life events with 5-HTTLPR 50 and brain-derived neurotrophic factor 51 in depression; with respect to studies included in the present meta-analysis, results pertaining to the impact of these stressors and the 5-HTTLPR gene in PTSD were examined.
Childhood adversity. Four studies were pooled for a total of 11,409 study subjects; the combined results revealed a significant interaction between 5-HTTLPR and childhood adversity (P = 0.003) (Fig. 2). The sensitivity analysis showed that the results remained significant after each study was removed in turn (1 × 10 −4 < P < 0.03) ( Table 2). Obtaining a significant result (P = 0.05) would require more than 21 unpublished or undiscovered studies with an average sample size of n = 2852 and non-significant results (P = 0.50). This yields a fail-safe ratio of five excluded studies for every study included in the present meta-analysis.

Stressful life events.
Twelve studies were pooled for a total of 8771 study subjects; the results revealed a significant interaction between stressful life events and 5-HTTLPR polymorphism (P = 2.0 × 10 −8 ) (Fig. 2). The sensitivity analysis showed that the P value remained significant after removing each study in turn and calculating the outcome (3.0 × 10 −9 < P < 4.0 × 10 −7 ) ( Table 3). More than 166 unpublished or undiscovered studies with a null effect (P = 0.50) and average sample size of n = 731 would be required for this to be a non-significant outcome in the stratified analysis (P = 0.05), corresponding to a fail-safe ratio of 14 excluded studies for every study included in this meta-analysis.
The time at which a stressor is measured can affect the gene and environment interaction effect 57 , while recalling adversity over long periods of time may increase the risk of forgetting or discounting events 24 . This type of   bias can largely be avoided in longitudinal studies. We carried out a subgroup analysis based on study design to determine whether outcomes differed between cross-sectional and longitudinal studies.
Cross-sectional studies. We identified nine studies with a cross-sectional study design (9864 subjects) for which results were available in separate cross-sectional studies. The meta-analysis revealed a significant interaction between 5-HTTLPR and stress (P = 0.01) (Fig. 2), and the sensitivity analysis revealed that this effect persisted after each study was removed in turn (3.0 × 10 −5 < P < 0.04) ( Table 4). More than 49 unpublished or undiscovered studies with a sample size of n = 1096 and a non-significant result (P = 0.50) would be required for the outcomes in the stratified analysis to be non-significant (P = 0.05). This yields a fail-safe ratio of five studies excluded for every study included in the meta-analysis.
Longitudinal studies. Five studies were pooled with a total of 6019 study subjects, revealing a significant interaction between 5-HTTLPR and life stress (P = 2.0 × 10 −6 ) (Fig. 2). The sensitivity analysis suggested that the results were still significant after removing each study in turn (8.0 × 10 −7 < P < 9.0 × 10 −5 ) ( Table 5). For these results to be non-significant (P = 0.05), more than 56 unpublished analyses or undiscovered studies with an   average sample size of n = 1203 and non-significant results (P = 0.50) would be required, yielding a fail-safe ratio of 11 excluded studies for every study included in the meta-analysis.

Biallelic locus.
We conducted a subgroup analysis stratified by locus to explore the interaction between 5-HTTLPR polymorphism, life stress, and PTSD. Seven studies were pooled with a total of 8023 study subjects; the results revealed a trend towards a significant interaction between a biallelic 5-HTTLPR locus, life stress, and PTSD (P = 0.054) (Fig. 2). The sensitivity analysis indicated that the results remained non-significant after removing each study in turn (0.054 < P < 0.086), with the exception of one study 35 (Table 6).

Triallelic locus.
Six studies were pooled with a total of 4815 subjects to assess the interaction between a triallelic 5-HTTLPR locus, life stress, and PTSD; a significant interaction was observed (P = 4.0 × 10 8 ) (Fig. 2). The results remained significant in the sensitivity analysis when each study was removed in turn from the analysis (2.0 × 10 −7 < P < 2.0 × 10 −6 ) ( Table 7). To render the outcomes non-significant (P = 0.05), more than 67 unpublished or undiscovered studies with a sample size of n = 802 and non-significant results (P = 0.50) would be required. This yielded a fail-safe ratio of 11 excluded studies for every study included in the meta-analysis.

Discussion
This is the first meta-analysis investigating the interaction between 5-HTTLPR polymorphism, stress, and PTSD. We found that 5-HTTLPR polymorphism influenced the relationship between stress and PTSD, with the less frequent S allele associated with increased stress sensitivity. When the meta-analysis was stratified by type of stressor, we found that stressful life events and childhood adversity independently interacted with 5-HTTLPR in PTSD; when the analysis was stratified by study design, we found interactions between 5-HTTLPR and stress in both cross-sectional and longitudinal groups; and when the analysis was stratified by allele classification, the results were more robust for the triallelic models group while the interaction effect for biallelic models failed to reach the predetermined level of significance. The overall results are consistent with a recent qualitative review 58 , and are in accordance with studies that have reported increased stress reactivity among 5-HTTLRP S allele carriers 59,60 . Animal studies have also demonstrated that functional variations in the 5-HTT gene affect behavioral response to stress. Specifically, 5-HTT knockout mice showed increased hypothalamic-pituitary-adrenal axis activation in response to both physical and psychological stressors 61,62 , as well as defects in cortical development and altered expression of 5-HT receptor subtypes [63][64][65] . Moreover, evidence from studies in mice and non-human primates have shown that 5-HTT gene variants are associated with changes in central nervous system biochemistry and behaviors linked to stress sensitivity 66,67 .
Of the 14 studies that investigated the influence of stressful life events and childhood adversity on PTSD, 12 examined the interaction effects between stressful life events and 5-HTTLPR; eight of these reported a significant interaction for PTSD 29,34,35,37,38,40,44,45 , whereas four did not find any evidence of interaction 36,41,42,46 . Four of the 14 studies investigated the interaction between childhood adversity and 5-HTTLPR, with three reporting a significant interaction for PTSD 35,42,43 and one finding no supporting evidence 39 . The meta-analysis revealed significant  Table 4. Studies included in the cross-sectional group meta-analysis.
interaction effects for stressful life events and childhood adversity separately interacting with 5-HTTLPR in PTSD. Our results are consistent with studies that have reported an association between 5-HTTLPR and neural responses to traumatic reminders and cognitive control of emotions in PTSD patients and the persistent effects of stressful life events and childhood adversity on hippocampal volume [68][69][70][71] .
Nine of the 14 studies used a cross-sectional design 29,35,36,39,[41][42][43]45,46 , with four reaching a conventional significance level 29,35,43,45 ; meanwhile, significance was attained by all five studies that used a longitudinal design 34,37,38,40,44 . The meta-analysis of both cross-sectional and longitudinal studies showed that 5-HTTLPR interacted with stress and PTSD. At least one study has failed to detect a gene-environment interaction between 5-HTTLPR polymorphism and life events in the months immediately preceding PTSD onset 72 . However, most studies have measured life events in the 5 years prior to PTSD. Retrospective recall of adversity is associated with increased risk of forgetting or discounting events 24 ; thus, when only a lifetime diagnosis of PTSD is available, information about the relative timing of stressors and PTSD is lost. This bias can be avoided to a greater extent in longitudinal than in cross-sectional studies.
We evaluated the association between loci using bi-and triallelic models to determine whether the latter better reflects the interaction between the 5-HTTLPR polymorphism, stress, and PTSD. We found evidence supporting the interaction in the triallelic but not in the biallelic group. Our approach was based on the reclassification of alleles according to lower and higher levels of expression, which is more precise for exploring the interaction between 5-HTTLPR polymorphism, stress, and PTSD 30,32 .
There were some limitations to the present meta-analysis. Firstly, several of the included studies may have limited power due to their small sample size 41,42,44,46 . Secondly, since we combined studies at the level of P values, the quality of primary studies may have affected our results. Some primary studies conducted separate tests on different sample subgroups or multiple PTSD measures; we guarded against false-positive results resulting from     Table 7. Studies included in the S allele group meta-analysis. this potential bias by using an average of reported P values. Finally, we could not estimate the magnitude of genetic effect and how it compares to the interaction effect size 73 .
In conclusion, we found that 5-HTTLPR influences the relationship between stress and PTSD. Further studies which focus on Gene × Environment interaction are needed to better understand the role of this polymorphism in PTSD risk. Our analysis identified study characteristics that could potentially affect study results such as type of stressor, study design, and allele classification. Childhood adversity and stressful life events could be two good candidate environmental risk factors in G × E research. The triallelic models approach (S/La/Lg) altered the results of meta-analysis comparing with the biallelic models approach(S/L). Special attention should be paid to the triallelic polymorphism in the relationship between 5-HTTLPR, stress and PTSD. These findings provide a basis for designing more rigorous studies on gene-environment interactions in PTSD in the future.