Introduction

By the year 2030, depressive disorders are projected to rank first among disorders contributing to the global disease burden [1]. Major depression is a complex and heterogeneous disorder with a highly variable course. Lifetime prevalence rates vary significantly by age and sex, with young adults and females having higher rates of depression than males and older adults [2–4].

Stressful life events (SLE) are one of the most consistently documented exposures among those with depression [5–8]. Stressful events can include, but are not limited to, unemployment, acute or significant health problems, unexpected or persistent financial hardship, and relationship problems. While these are common experiences, not everyone who experiences them becomes depressed. Although the nature, kind, and timing of stressful life events have been implicated as sources of variability in the relationship between depression and SLEs, so too have biological factors such as serotonergic functioning [9, 10].

The neurotransmitter serotonin (5-HT) is synthesized in the brain by a small number of neuronal fibers with profuse projections throughout the brain, spinal cord and nervous system [10]. Because of this wide distribution, it has been implicated in a variety of mental and physical disorders [11–13]. Through removal of the 5-HT molecule at the synapse, the serotonin transporter (5-HTT) plays an important role in determining the extent and duration of serotonergic signaling. An insertion/deletion polymorphism (5HTTLPR) in the promoter region of the 5-HTT gene has been implicated as a source of variation in 5-HTT gene transcription rates such that the short (S) allele is less transcriptionally efficient than the long (L) allele [14, 15]. In 2003, Caspi et al [16] examined the role that 5HTTLPR polymorphism has in the relationship between SLEs and depression finding that carriers of one or more S-alleles were more likely to develop depression in the context of more SLE’s than those with two L-alleles.

Since that landmark study [16], there have been numerous replication attempts and three meta-analyses [17–24]. Differences in the conceptualization of stressful life events, depression, and their quantification have likely contributed to a literature that has not provided any definitive evidence for or against the original hypothesis. Small sample sizes, low statistical power, and an inconsistent preferred mode of examining the hypothesized interaction have also contributed to mixed findings [25], leaving some to contend that there have been few direct replication attempts of the original hypothesis [20, 21]. Here, we report findings from a study of the 5HTTLPR, stressful life events, and depression using an approach as close to an a priori direct replication as we could achieve in a large prospective nationally representative sample. Our first test examined the hypothesis that carriers of one or more S-alleles would be more sensitive to stress than individuals carrying no S-alleles. In a post-hoc manner and similar to Caspi et al [16], we examined the role of the 5HTTLPR in moderating the impact of stressful life events on suicide ideation and moderating the impact of childhood maltreatment on self-reported diagnosed depression. Based on prior published findings since Caspi et al [16] showing that the stress-moderating effects of the 5HTTLPR may differ between males and females [21, 26–31], we also examined sex as a moderator of stressful life events x 5HTTLPR effects on depression and suicidal ideation in post-hoc analyses.

Methods

Results

We examined the gene-environment hypothesis in a sample of white young adult males and females, ages 22–26, who participated in Wave III (2001–2002) of Add Health (n = 4724). In our analysis sample, 23.6% (n = 1117) had two copies of the short allele (s/s genotype), 48.8% (n = 2303) had one copy of the short allele (s/l genotype), and 27.6% (n = 1304) had no copies of the short allele (l/l genotype). The three-genotype groups were in Hardy-Weinberg equilibrium (χ2 = 2.597, df = 1, p = 0.11) and there were no significant differences in genotype frequencies between males and females (χ2 = 0.32, p = 0.85).

Stressful life events were assessed using 21 separate events including health, housing, employment, financial, and relationship experiences. Seventy-nine percent of our sample reported experiencing at least 1 stressful life event and of those, 23.3% reported 2 events, 12.7% reported 3 events, and 7.2% experienced 4 or more. There were no significant differences in SLE scores (F(2,4723) = 0.41, p = 0.66) as a function of genotype status and indicated that the 5HTTLPR polymorphism was not associated with exposure to stressful life events.

Thirteen percent (n = 626) of our analysis sample self-reported being diagnosed with depression between 22 and 26 years of age. Females were more likely than males to report having a diagnosis of depression (odd ratio = 2.17; 95% confidence interval, (CI): 1.85 to 2.55) and report having taken medication for depression (odds ratio = 3.4, 95% CI: 2.36 to 3.91). Males and females were equally likely to report having spent time in a hospital for mental illness in the previous five years (odds ratio = 1.10, 95% CI: 0.74 to 1.63) and to have attempted or seriously thought about committing suicide in the past year (odds ratio = 0.70, 95% CI: 0.48 to 1.02).

We began our logistic regression analyses by examining whether the risk for depression increased as a function of 5HTTLPR genotype and the number of stressful life events experienced. Sample sizes for our primary and post-hoc analyses are provided in S2 Table. For both males and females, as the number of life events increased the risk for depression also increased. Prevalence estimates for depression were similar across 5HTTLPR genotype status, though depression risk appeared to increase more steeply for males who have one or more of the short alleles (Fig 1a and 1b, S3 Table). We interpreted this pattern of results as providing preliminary support for the originally hypothesized gene-environment interaction [16]. Formal tests of this possibility, accounting for the sampling design of Add Health, revealed a main effect for SLEs [b = 0.67, t = 8.75, p < 0.01 95% CI: 0.52 to 0.83], a main effect of the 5HTTLPR genotype [b = 0.40, t = -2.66, p < 0.01, 95% CI: 0.10 to 0.69], and for a statistically significant interaction [b = -0.16, t = 0.07, p = 0.02, 95% CI: -0.29 to -0.02] (Table 1). Post-hoc analyses of this replication finding revealed that while stressful life events were a risk factor for depression for both sexes, the role of the 5HTTLPR and its interaction with stressful life events approached significance (p = 0.007, 95% CI: -0.54 to -0.09) only for males [Fig 1a and 1b, Table 1]. A formal test of whether the interaction terms differed between the sexes revealed a non-significant sex-5HTTLPR-stressful life events interaction [b = 0.25, t = -1.82, p = 0.07, 95% CI: -0.02 to 0.53].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Prevalence Rates for Depression as a function of 5-HTTLPR’ genotype status, two stressful environmental experiences and suicide ideation among young adult males and females.

Figs A and B (Top Row) graphically illustrate the prevalence of depression as a function of stressful life events; Figs C and D (Middle Row) illustrate the prevalence depression as a function of suicide ideation; and Figs E and F (Bottom Row) illustrate the prevalence of depression as a function of childhood maltreatment experiences.

https://doi.org/10.1371/journal.pone.0148373.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Unstandardized Logistic Regression Estimates and Standard Errors Predicting Depression and Suicide Ideation.

https://doi.org/10.1371/journal.pone.0148373.t001

Discussion

In the current report we examined the hypothesis that 5HTTLPR genotype moderated the relationship between experiences of stressful life events and depression during young adulthood. We sought to test this hypothesis by adopting a direct replication strategy that included an analytic approach and a stressful life events measure similar to those employed previously [16]. Importantly, as a group we specified the exact tests that would be conducted prior to undertaking our analyses to limit the number of tests conducted and preserve the integrity of interpretation for those that were conducted. Results supported the hypothesis that the 5HTTLPR plays a role in moderating the impact of SLEs on depression status, though at a statistically significant level only in males and using a self-reported ‘ever’ diagnosis of depression. We did not find support for a similar role of the 5HTTLPR in the relationship between stressful life events and suicide ideation; however, post-hoc analyses were suggestive of a stress-moderating effect of the 5HTTLPR on suicide ideation that differed between males and females. Further, while having experienced maltreatment as a child was a significant risk for depression, we did not detect a significant maltreatment-5HTTLPR interaction in these data as had been reported previously [16]. Post-hoc considerations revealed, that in the context of increasing numbers of stressful life events, common variation in 5HTTLPR genotype could contribute to differences in mental health problems between the sexes.

In the absence of stressful life events, we found rates of depression that were similar across genotype status. This indicated that at low levels of stress, 5HTTLPR genotype was not related to the risk for depression. As the number of stressful life events increased, the prevalence of depression also increased among those carrying one or more copies of the short allele. This relationship between stressful life events and depression was most evident at the extremes of the distribution where sample sizes were smaller, though still larger than the original study [16]. Moreover, the higher prevalence of depression among those with many stressful life events was significant only for males. Finally, as new genetic variation in the 5HTTLPR had been identified since the original study [16] we included SNP rs25531 in our analyses. Doing so negligibly changed the observed relationships as compared with results including just the 5HTTLPR. This is consistent for a SNP with a low minor allele frequency (MAF) among Caucasians [> 5%; 35] such that the number of participants reclassified on the basis of functional binning of the 5HTTLPR was limited. This, however, would not be the expectation in genetically diverse samples with greater MAFs.

Gene-environment interaction studies offer an opportunity to understand biological mechanisms important for predicting risk and evaluating changes in modifiable environments [37–39]. Further, these kinds of studies can help to identify environmental effects that act primarily in genetically susceptible individuals. There have been many attempts at replicating the observation that the 5HTTLPR S-allele moderates the relationship between depression and stressful life events [16–20, 25], often with conflicting results. Similarly, findings from meta-analyses [20–22] have also been inconclusive, with some reporting no effect and others full support for the original hypothesis. Against this literature, our results agree with the notion that the interaction effect of 5HTTLPR genotype and stressful life events on depression is marginal and that the effect size is smaller than that originally reported. Further, similar to one prior study [19], our results suggest that it is important to capture the specific forms of life stress originally set forth in Caspi and colleagues [16] original report [40, 41].

Suicide ideation is often reported among patients with depression and those experiencing acute stress [42]. Because of the magnitude of impact suicide has on society there is a great interest in understanding its biological and social causes. To this end, our results suggest the possibility that in the presence of an increasing number of stressful life events different 5HTTLPR alleles may confer a susceptibility to suicide ideation that differs between males and females. If true, this finding would extend Caspi et al [16] and support previous studies showing sex moderation of the 5HTTLPR effects on depression [28, 30, 31] and negative affect [43], and may offer one explanation for the heterogeneity in association tests of the 5HTTLPR and suicide behaviors that implicate both the L-allele and S-allele [44, 45]. If replicated, these results would aid efforts to better identify those at risk for suicide and implicate environments and social experiences that may be modified to reduce such risk.

Similar to other replication attempts, our study is not without limitations. First, we were unable to establish a temporal relationship between the experience of a stressful life event and the onset or occurrence of depression. As such, some of the measured stressful life events may have occurred after the onset of depression, which may have the effect of biasing our interaction parameter [46]. Similarly, stressful life events examined here were retrospectively reported as having occurred sometime within the preceding five-years while depression diagnosis could have occurred at any point before the Wave III interview. Stressors most salient to depression are those which are acute, major, occurred within the previous six-months, and are participant focused [40, 41]. Although our stressful life events measure was very similar to that specified by Caspi and colleagues (16), we were not able to include each of its aspects. Second, our coding of SLEs, though exactly the same as Caspi et al [16], limits the potential importance of experiencing multiple types of SLE’s within one domain or multiple occurrences of the same event within a domain. This has, for example, the consequence of scoring someone who has experienced multiple forms of violence over years in an intimate relationship but no other SLE the same as a person who moved states for a job and no other SLE’s. Said differently, scoring SLE domains as we have limits the impact of the severity of an experience. A coding scheme that accounts for this by possibly weighting subsequent experiences may provide a better allocation of subjects to SLE groups and reduce sample heterogeneity among those experiencing fewer SLE’s. Third, though we were able to detect a significant interaction between the 5HTTLPR and stressful life events in a much larger sample than examined by Caspi and colleagues [16], the effect size was less robust. One potential reason for this could be differences between interview and self-reported adversity. Previous meta-analyses [45] and systematic reviews [39] have suggested that the use of interviews and/or multiple informants may afford better sensitivity for the detection of gene-environment (GxE) interactions. Fourth, reports of childhood maltreatment were retrospective and therefore may have been influenced by distorted memories or recall bias [47–50]. Lastly, for replication purposes [16] we restricted our sample to include only Caucasians and as such are unable to establish the extent to which findings would be similar in other race/ethnic populations. Given race differences in the prevalence of depression [4, 50] and allele frequencies of the functional SNP rs25531 [51, 52], examining this hypothesis in diverse ethnic samples should be a priority for future research.

In conclusion, by committee, we set out to examine the hypothesis that SLEs moderated the impact of the 5HTTLPR S-allele on depression among young adults. To do so, we attempted to approximate as closely as possible the measures originally examined by Caspi et al [16] in a large well-characterized sample followed longitudinally since adolescence. Towards this end, a notable advantage of the Add Health sample is the extensive environmental measures combined with the unique availability of simple sequence data [34]; in particular rs25531 and the 5HTTLPR. Although these types of samples provide strong statistical power for hypothesis testing, they may not always assess each phenotype of interest in as an extensive of a manner as smaller samples designed to test circumscribed hypotheses. As noted above, we could not approximate every phenotype as originally [16] put forward. Despite this, our results utilizing a highly similar measure of SLEs, a clinically orientated depression measure, a more transcriptionally accurate genotype classification, and the same analytic approach, we obtained evidence, albeit not unequivocal, that supported the original finding reported by Caspi et al [16]. Specially, young adults with lower 5HTT transcription and greater numbers of SLEs reported higher rates of depression as compared with those carrying the L-allele and experiencing fewer SLEs. Although larger sample sizes are desirable to further explore the questions asked here, it is important to do so in samples with extensive environmental characterization using alternative methodologies and strategies that elucidate the connections between sequence variation, gene regulation, and depression related neural pathophysiology. In this way we can realize the identification of environmental effects that act in susceptible individuals.

Supporting Information

Acknowledgments

This research uses data from Add Health, a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill, and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Special acknowledgement is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Information on how to obtain Add Health data files is available at the Add Health website (http://www.cpc.unc.edu/addhealth). This research was also supported by grant P01-HL36587 from the National Heart, Lung, Blood Institute to Redford Williams and by grant AG046938 from the National Institute on Aging to Chandra A. Reynolds and Sally Wadsworth (Co-PI’s)