The 5HTTLPR Genotype by Stress Interaction on Risk for Depression:
A Critical Review

Michael Bronstein

Capsi et al. (2003) found that there was no main effect of an individual's 5HTTLPR genotype on his or her risk of experiencing depression. I review the evidence for an against the 5HTTLPR genotype by stressful life events interaction on risk for depression [hereafter, a.k.a. "The Interaction"].

 

The heritability of major depressive disorder (MDD) has been estimated to be between 30 and 50 percent (Heim & Binder, 2012; Levinson, 2006). This significant amount of heritability suggests that genetic risk factors play a major role in determining who might develop MDD. Identifying these genetic risk factors is worthwhile for two main reasons. First, identifying these factors would constitute a major advance in our understanding of the mechanisms involved in the pathophysiology of MDD (Levinson, 2006). Second, identifying these factors could allow improved targeting of prophylactic treatment at individuals who might be especially prone to develop MDD. Given that MDD is one of the major worldwide causes of disability (Vos et al., 2012), having the capability to provide such treatment efficiently could alleviate the suffering and increase the wellbeing of a large number of individuals.

As a result of efforts to realize these benefits, the serotonin transporter (5HTT) gene has become the subject of many studies searching for the identity of genetic risk factors for depression. This gene is located on chromosome 17 (q11.1-12) and consists of 31 kilobases which code for 14 exons (Lasky-Su et al., 2005). The transcription of this gene is regulated by a promoter region (5HTTLPR) that is subject to a 44 base pair insertion/deletion mutation. As a result of this mutation, a polymorphism exists in the population: some individuals have a short allele consisting of 14 copies of a 20-24 base pair variable nucleotide tandem repeat element, whereas others have a long allele which has 16 copies of this element (Munafo et al., 2008a). This polymorphism in 5HTTLPR genotypes likely impacts neurotransmission by altering 5HTT density and the efficiency of individual 5HTT polypeptides (Munafo et al., 2008a; Levinson, 2006), although alternate mechanisms of impact have been proposed that center on changes in serotonin autoreceptor density as a function of 5HTTLPR genotype (Munafo et al., 2008a).

In the most famous study investigating this polymorphism, Caspi et al. (2003) found that there was no main effect of an individual's 5HTTLPR genotype on his or her risk of experiencing depression. However, they found an interaction between this genotype and the number of stressful life events an individual had experienced on that individual's risk for depression. Although this finding contributed to a paradigm shift in behavioral genetics research, it remains highly controversial because many studies attempting to replicate this gene by environment interaction with different populations and study designs have failed to do so (e.g., Heim & Binder, 2012; Karg et al., 2011; Gilliespie et al., 2005; Hariri et al., 2005). In order to move toward resolving this controversy, I will critically review the evidence for and against the 5HTTLPR genotype by stressful life events interaction on risk for depression posited by Caspi et al. (2003) [hereafter, a.k.a. "The Interaction"], beginning with the evidence that The Interaction does indeed exist.

One source of evidence supporting the existence of The Interaction is research demonstrating that in the presence of stress there is an association between the 5HTTLPR genotype and traits which are intermediates of depression. Several of these demonstrations come from research on rhesus macaques, who have variations in 5HTTLPR length that parallel those of humans (Caspi et al., 2003). In these animals, the intermediates of depression associated with 5HTTLPR in the presence of stress include impaired serotonergic function, which is associated with the short 5HTTLPR allele in macaques reared in stressful but not normal conditions (Caspi et al., 2003). These intermediates also include increased anxiety and hypothalamic-pituitary-adrenal (HPA) axis reactivity, which rhesus macaques display more readily in response to both early and later life stress when carrying the short allele (Caspi et al., 2010). The association of these intermediates with 5HTTLPR genotype is consistent with The Interaction because it provides putative biological mechanisms that could cause the short 5HTTLPR allele to increase risk for depression to a greater degree than the long allele in the presence of increasing stress.     

Studies of humans have also suggested such putative mechanisms and are therefore consistent with The Interaction. These studies include neuroimaging research demonstrating that in children the short 5HTTLPR allele is associated with increased brain activity in emotion processing regions in response to sad movies (Heim & Binder, 2012). This alteration suggests that 5HTTLPR genotype may alter the emotional processing that occurs during stressful situations encountered early in development and could therefore alter the depressogenic effects of these situations. These studies also include attentional assessments demonstrating that healthy adults with two short 5HTTLPR alleles have more difficulty disengaging attention from emotional stimuli than those with two long alleles (Beevers et al., 2009). Because rumination is possibly caused by the impaired disengagement of attention from information that has a negative emotional valence and is self-relevant (Koster et al., 2011), this difficulty might lead to increased rumination on the negative emotional consequences of stressful events in individuals with two short 5HTTLPR alleles vs. those with two long alleles. In light of the fact that rumination increases vulnerability to depression (Lyubomirsky et al., 2015), this attentional impairment could therefore account for the greater depressogenic effects of stressful events in those with a short 5HTTLPR allele and at least partially underlie The Interaction.

Stronger evidence that The Interaction exists comes from Caspi and colleagues' seminal study (2003). In this study, The Interaction was demonstrated in two ways. First, risk of depression at age 26 in response to childhood maltreatment in the first decade of life (a specific stressful life event) was shown to be moderated by 5HTTLPR genotype. Second, the association of risk for depression at age 26 with the number of stressful life events experienced in the previous five years was also found to be moderated by 5HTTLPR genotype. These two demonstrations of The Interaction with varied methodology are a major strength of Caspi and collegues' study because they suggest that The Interaction has external validity and did not arise from type I error. Another strength of Caspi et al.'s (2003) study is that they corroborated the results of a structured interview with informant reports in order to determine whether individuals were depressed. This procedure likely increased the accuracy of their sorting of individuals into diagnostic categories, increasing the internal validity of their conclusions. A final strength of this study was that it used a partially longitudinal design and therefore did not rely on retrospective reports to assess childhood maltreatment.

Despite these strengths, this study did have one major weakness. This was that stressful life events in the study (with the exception of childhood maltreatment) were assessed using a retrospective self-report measure. The use of this measure may explain the interaction between 5HTTLPR genotype and the number of stressful life events encountered in the last five years on risk for depression that was found in 26 year olds. For example, The Interaction could be spuriously produced in retrospective self report data by mood-congruent memory revision, which should lead to increased reporting of stressful life events by depressed individuals (Caspi et al., 2010). This possibility, however, is contingent on a main effect of 5HTTLPR genotype on risk for depression. This main effect was not detected in Caspi's (2003) study and thus this particular source of a spurious interaction is somewhat unlikely, although it is possible that not enough power was present to detect this effect. Even if a spurious interaction finding was generated in this way, mood-congruent memory revision cannot account for the replication of this interaction in Caspi and colleagues' examination of childhood maltreatment (which used a prospective design). In light of this and the strengths listed above, Caspi et al.'s (2003) study can be said to constitute strong evidence for the existence of The Interaction.

This strong evidence for the existence of The Interaction is complemented by more recent research replicating this interaction effect. One example of this research (Wilhem et al., 2006) replicated this effect using stressful life events over a five year period that varied by participant. For participants who had previously experienced MDD, this period consisted of the five years prior to the onset of an individual's first episode of MDD. For those who had never experienced MDD, this period consisted of the 5 years prior to the average age at which all participants' first episodes of MDD occurred. Replicating The Interaction with a sample that experienced depressive episodes at ages ranging from 15-50 supports the idea that this interaction exists and that it holds across a wide swath of the lifespan.

Despite this initial support for The Interaction, it should be noted that when reducing this five year time period over which stressful life events were considered to only one year, this same study (Wilhem et al., 2006) failed to replicate this interaction effect. This failure was probably the result of reduced power in the portion of this study considering stressful life events occurring over a one year period relative to the portion considering these events over a five year period. This power differential may have been caused by the fact that in the portion of this study considering stressful life events occurring over a one year period, these events were coded as a dichotomous variable (1+ or 0), whereas in the portion of this study considering these events over a five year period more categories were used (0,1,2,3+). This coding difference likely produced a significantly smaller variance in risk for depression between 5HTTLPR genotypes in the portion of this study considering stressful events that occurred over a one year period. This reduction in between-genotypes variance seems likely to result from this coding difference given that multiple studies have produced results suggesting that risk for depression is most similar among the 5HTTLPR genotypes when considering the data for individuals who experienced only one stressful life event in a given time period (see Figure 1 in Wilhem et al., 2006 and in Caspi et al., 2003). If this is the case, then including individuals who experienced one stressful life event in the same category as those who experienced more than one such event in a given time period would significantly reduce between-genotypes variance in risk for depression. This reduction, in turn, would reduce the statistical power of the portion of this study examining stressful events over a single year by reducing the effect size of the 5HTTLPR by stressful life event interaction. Because this failed replication is probably the result of insufficient power, the portion of this study replicating Caspi et al. (2003) should take precedence and this study should be considered good evidence supporting The Interaction.

Another study providing good evidence for the existence of this interaction replicated it with a sample of individuals ages 65 and older (Kim et al., 2007). This replication supports the notion that The Interaction exists as well as the aforementioned idea that this interaction holds over a large swath of the lifespan. Interestingly, this study is also consistent with the idea that the failed replication above resulted from insufficient power. This consistency results from the several steps taken by this study that may have improved its power over the failed replication above. These steps included focusing on a population of Korean elders (over age 65). Because a greater number of individuals in Asian populations than in populations of European descent have the short 5HTTLPR allele (Kim et al., 2007), this choice would have given this study a better chance of detecting an interaction involving 5HTTLPR genotype than the failed replication in Australian individuals above. These steps also included the fact that this study, unlike the failed replication above, coded only individuals with severe or moderate depression according to a structured interview as being depressed. This conservative choice likely resulted in the inclusion of only truly depressed individuals in the group of depressed individuals in this study, again increasing this study's power relative to the replication above.

Additional support for the interaction purported by Caspi et al. (2003) comes from a study of consecutive patients at several primary care centers in Spain (Cervilla et al., 2007). This study's focus on these patients yielded a sample that is more representative of the general population than the sample in many previous studies examining The Interaction (Cevilla et al., 2007). Thus, its replication of The Interaction strongly suggests that this interaction generalizes across a variety of population segments. This study also suggests that this interaction is stronger when considering more severe cases of depression as diagnosed according to the ICD-10, indicating that failure to replicate The Interaction is more likely in studies which include a greater proportion of individuals with more mild cases of depression. This indication corroborates the notion that the discrepancy between the support for The Interaction in this study and the lack of support for it in the other study above which examined stressful life events over a one year period (Wilhem et al., 2006) could be accounted for by greater statistical power in this study resulting from its more conservative criteria for what counts as a depressive episode.

Although these individual studies appear to strongly support the existence of The Interaction, it is always possible that unknown flaws in these individual examples exist that are driving these supporting results. Because this likelihood should be greatly reduced by examining the literature on this interaction as a whole, it is worth considering the support that meta-analyses give for this interaction as well. In the most extensive meta-analysis of this literature to date (Karg et al., 2011), studies were considered simultaneously by combining them at the level of significance tests in order to determine the overall level of support for this interaction in the literature. This procedure resulted in strong support for Caspi and colleagues' interaction. Notably, this support persisted when any one study was removed from the analysis, indicating that this support is robust against outliers (Karg et al., 2011). Unfortunately, the support of this meta-analysis for The Interaction must be qualified by the fact that critics of this meta-analysis have noted that its method could have resulted in an amplification of the biases present in any of the constituents of this meta-analysis and could thereby color its results (Risch et al., 2009). This explanation for the results of this meta-analysis cannot be definitively ruled out. However, some suggestion that these results are not due to amplifications of the bias from individual studies comes from the fact that downsampling the studies included in this meta-analysis to match the sample used in two other meta-analyses which failed to replicated The Interaction (Risch et al., 2009; Munafo et al., 2008b) produced results that concurred with these analyses findings. This concurrence is somewhat consistent with the lack of a strong bias in this meta-analysis toward supporting The Interaction.

Given the apparently strong evidence detailed above from both individual studies and meta-analyses supporting The Interaction, it is worth considering whether this support may be driven by a gene-environment correlation. Gene-environment correlations result when individuals with different alleles for a certain gene are at differential risk for some outcome of interest (e.g., depression) and also for selection into an environment with specific attributes (e.g., an environment conducive to stressful life events). In this case, it can spuriously appear that sequelae of the relevant environmental attribute (e.g., higher numbers of stressful life events) played a causal role in producing the outcome of interest. If a gene-environment correlation explained the interaction effects above, one would expect that the strength of the association between the genotype (5HTTLPR) and the sequelae of the environmental attribute (stressful life events) involved in the gene-environment correlation would differ between groups of individuals separated on the basis of the outcome of interest (depression). This is not the case in studies of the interaction between 5HTTLPR and stressful environments (Risch et al., 2009). Further, if a gene-environment correlation explained the findings of an interaction between 5HTTLPR genotype and stressful life events presented above, then there should be reports of a correlation between these factors in the literature. This is also not the case. In fact, research has explicitly demonstrated that this correlation is unlikely to exist (Wilhem et al., 2006). Taken together, these two lines of reasoning indicate that the interaction observed by Caspi et al. (2003) is not underlied by gene-environment correlation.

At odds with the above research is a large body of evidence that refutes the existence of The Interaction. Much of this evidence comes from individual studies that were unable to endorse the alternative hypothesis that such an interaction was occurring in their samples (i.e., that yielded null findings). Methodological considerations may account for the null findings of many of these studies. For example, in one study that failed to replicate this interaction (Gillespie et al., 2005), the vast majority of participants reported no exposure to stressful life events. Because optimal power to detect this interaction would occur if there were equal numbers of participants in each category of exposure to stressful life events used in this study (Caspi et al., 2010), this imbalance should have limited this study's power to detect the expected interaction effect. Thus, this study should have had an elevated risk of making a type II error. In another study that produced null findings that may be accounted for by methodological considerations (Chipman et al., 2007), the sample examined was not in Hardy-Weinberg equilibrium (HWE). A lack of HWE in a sample can be indicative of population stratification, non-random genotyping error, or non-random mating (Lunetta, 2008). All of these possible causes of violations of HWE can lead to increases in type II error rates and could therefore account for this study's lack of support for the interaction purported by Caspi and colleagues.

In a third such study with a method that increased the risk of type II errors being committed, no 5HTTLPR by stressful life events interaction was found in a sample of adults from the United Kingdom when considering stressful life events encountered in adulthood or in childhood (Surtees et al., 2006). In this study, this lack of an interaction held when using both a traditional diagnosis of depression (DSM-IV) and a diagnosis of possibly more severe depression (characterized by a DSM-IV diagnosis of depression and high neuroticism) as outcome variables. This indicates that the failure of this study to replicate Caspi et al. was probably not due to the outcome variable used. Differences in the proportion of participants experiencing their first episode of depression in this study (9%) and in Caspi et al.'s study (where this percentage was likely much higher) better account for the discrepancy between these studies' results (Surtees et al., 2006). This is because the effect of stressful life events on risk for depression decays with each subsequent depressive episode, an effect known as the kindling hypothesis (Gillespie et al., 2005). As a result of kindling, variation in the number of stressful life events recently experienced by a group of individuals should produce less variation in risk for depression if this group of individuals previously experienced a greater number of depressive episodes. Thus, power to detect a 5HTTLPR by stressful life events interaction should be more drastically reduced in studies whose sample of depressed individuals is comprised of a greater proportion of participants for whom the index episode is not their first episode of depression. Because this study of adults from the United Kingdom (Surtees et al., 2006) fits this profile to a greater degree than Caspi et al.'s study (2003) and has a sample size similar to that in Caspi et al.'s study, it is unsurprising that the former study produced null findings while the latter one produced significant ones. This discrepancy therefore cannot be considered strong evidence against The Interaction.

Stronger evidence against The Interaction comes from studies lacking such methodological predispositions to committing type II errors. These studies include one of the few analyses of African American individuals in the literature (Ressler et al., 2010), which found no evidence for The Interaction. These studies also include several meta-analyses. In one such meta-analysis (Munafo et al., 2008b), a simulation was run using data from multiple studies examining The Interaction. The parameters of this simulation were purposefully chosen to emulate Caspi et al.'s original study. This, in tandem with the detailed power analyses accompanying this meta-analysis which indicated that it had reasonably adequate power to detect interaction effects, suggests that the failure of this meta-analysis to replicate The Interaction directly conflicts with the results purported by Caspi et al. (2003). This strong evidence against the existence of this interaction is echoed by a second well done meta-analysis (Risch et al., 2009), which replicated these null findings by conducting a logistic regression on the data from fourteen studies.

More theoretical arguments ostensibly complement these stronger studies in refuting The Interaction. One such theoretical argument stems from the pharmacological link that selective serotonin reuptake inhibitors (SSRIs) provide between decreased serotonin reuptake and relief of depression. This link makes it difficult to believe that the short 5HTTLPR allele, which according to Beevers et al. (2009) also causes decreased serotonin reuptake (compared to the long 5HTTLPR allele), would confer increased risk for depression compared to the long 5HTTLPR allele in the presence of increasing stress. One possible resolution to this apparent paradox of decreased serotonin reuptake both increasing and decreasing risk for depression is that the short 5HTTLPR allele might decrease sensitivity to serotonin by down-regulating post-synaptic serotonin receptors (Hariri et al., 2005). If this were the case, this down-regulation might cause the short 5HTTLPR allele to actually decrease the effective amount of serotonin in the synapse, in contrast to the effects of SSRIs. Thus, the actions of SSRIs and 5HTTLPR genotypes would more neatly align with their effects on risk for experiencing depression.

Another such theoretical argument against the existence of The Interaction is based in the nature of the interaction purported by Caspi et al. (2003). In this interaction, risk for depression increased to a greater degree from one level of stress exposure to the next in individuals with a short 5HTTLPR allele (vs. those with a long allele). Further, risk for depression was quite similar in this interaction for individuals with all 5HTTLPR genotypes at baseline (i.e., who were exposed to 0 stressful life events). Given this potential interaction, on average the risk for depression in individuals with the short allele should be higher than in those with the long allele (Munafo et al., 2008b) assuming that the distribution of stressful life events among individuals with different 5HTTLPR genotypes is random. Thus far, research has demonstrated that this assumption is valid (i.e., no gene-environment correlation has been demonstrated in this literature). Therefore, there should be a main effect of 5HTTLPR genotype on risk for depression. However, this main effect has been found in almost none of the studies reviewed here (the one exception being Clarke et al., 2010).

One plausible explanation for this discrepancy between expected and actual results pertaining to this main effect of 5HTTLPR stems from the fact that in many studies focused on The Interaction the majority of participants report being exposed to zero stressful life events (e.g., Chipman et al., 2007; Surtees et al., 2006; Gillespie et al., 2005; Caspi et al., 2003). Given that an interaction will be increasingly strong relative to a main effect in studies with greater numbers of participants who were not exposed to one of the putatively interacting variables (Munafo et al., 2008b), this low rate of exposure to stressful experiences among participants could cause these studies to have more power to detect an interaction than to detect a main effect. If this is the case, then the lack of main effects of 5HTTLPR genotype in the literature are a poor basis for an argument against the interaction posited by Caspi et al. (2003).

A second plausible explanation for the lack of main effects of 5HTTLPR genotype on risk for depression in published literature similarly suggests that the lack of these effects does not constitute evidence against the interaction posited by Caspi et al. This second explanation is that if the short 5HTTLPR allele increased one's sensitivity to both positive and negative (e.g., stressful) life events, then even if The Interaction existed one would not expect there to be a main effect of 5HTTLPR genotype on depression as the effects of these increased sensitivities would cancel each other out (Culverhouse et al., 2013). This explanation seems probable in light of previous research demonstrating the increased sensitivity to both positive and negative aspects of one's environment conferred by the short 5HTTLPR allele (e.g., Drury et al., 2012).

Given that neither of these well-supported explanations preclude the existence of The Interaction in the absence of a main effect of 5HTTLPR genotype on risk for depression, the lack of this expected main effect makes a poor basis for an argument against the existence of The Interaction. Considering this information in tandem with the material reviewed above, it is clear that most theoretical and empirical evidence against the existence of The Interaction is rather weak. However, as mentioned earlier, there is a small subset of stronger empirical evidence against this interaction which cannot be as easily discounted. Thus, this review has found at least some sound empirical evidence both for and against this interaction. This discrepant evidence warrants consideration of what factors may have contributed to the different findings of studies reviewed above.

One factor to which these discrepant findings have been attributed is the wide methodological variations between research that affirms and refutes The Interaction (Risch et al., 2009). These variations frequently include researchers' choice of the genetic model (e.g., additive vs. multiplicative models) under which to consider this interaction (Uher et al., 2011) and the statistical procedures used to test for the presence of this interaction. Despite the fact that it is common to attribute the discrepancies in this literature to these factors, they are unlikely to explain the conflicting findings above given that in a recent meta-analysis (Sharpley et al., 2014) no methodological differences were found between the 26% of included studies that failed to replicate The Interaction and those that supported it.

Interestingly, this large failure rate suggests an alternative reason for the discrepancies between the above findings: The Interaction may have limited external validity. The fact that many studies which fail to replicate the interaction support it in post-hoc analyses on a specific age group is consistent with this idea (Uher et al., 2011). However, this consistency is limited by the fact that the multiple comparisons common in post-hoc tests cause them to have a high risk of type I error that is further multiplied by the nature of interaction testing (Munafo et al., 2008b). Because of this and the corpus of evidence discussed earlier which suggests that The Interaction has high external validity across age groups, the limited external validity of this interaction is unlikely to account for the discrepancies in the above findings.

After ruling out these explanations for the discrepancy between the results of studies that confirm and refute The Interaction, it seems most likely from this review that variations in the power of these studies accounts for variations in their findings. As discussed previously, these variations may result from the type of depression examined (first episode vs. not, more vs. less severe) and the race of the participants examined (Asian vs. not). They may also result from between-samples variation in factors, such as social support levels (Kaufman et al., 2004) or brain derived neuroprotective factor (BDNF) genotype (Kim et al., 2007), that moderate the strength of The Interaction. This putative variation in power makes it difficult to interpret the evidence for and against The Interaction. However, because none of the studies reviewed here obtained over 80% power (c.f. Munafo et al., 2008b), it seems that type II errors in these studies are more likely to have occurred than type I errors. The relatively strong support for the existence of The Interaction in this literature despite this predisposition toward null findings suggests that The Interaction does indeed exist.

Due to the likelihood that The Interaction exists and that variations in power account for the discrepancies between the findings of studies detailed above, future research examining The Interaction should make three major changes to the norms observed in this review. First, future research on The Interaction should be sure to include detailed power analyses. This will help scientists to more easily parse the role of power in producing discrepant findings across studies. Second, future research should attempt to acquire larger samples in order to achieve at least 80% power. This will help to ensure that future null findings are more meaningful and will also limit how much variation in power could account for differences in the results of future research. Finally, future research should take greater care to obtain samples exposed to homogenous stressors. This will not only help these studies to achieve adequate power, but will also aid them in more accurately estimating the effect size of The Interaction (Caspi et al., 2003). Following these guidelines should lead to a more thorough assessment of the role of the 5HTTLPR gene in MDD. As noted at the start of this paper, this outcome is valuable because identifying genetic risk factors for MDD may improve our understanding of the mechanisms involved in its pathophysiology as well as our ability to target prophylactic treatment for this disorder at high risk individuals.


Thanks for reading. Check out Sip 


References

Beevers, C., Wells, T., Ellis, A., & McGeary, J. (2009). Association of the Serotonin Transporter Gene Promoter Region (5-HTTLPR) Polymorphism With Biased Attention for Emotional Stimuli. Journal of Abnormal Psychology, 118, 670-681.

Caspi, A., Hariri, A.R., Holmes, A., Uher, R., & Moffitt, T.E. (2010). Genetic sensitivity to the environment: the case of the serotonin transporter gene and its implications for studying complex diseases and traits. Am. J. Psychiatry, 167, 509–527.

Caspi, A., Sugden, K., Moffitt, T.E., Taylor, A., Craig, I.W., Harrington, H., McClay, J., Mill,     J., Martin, J., Braithwaite, A., & Poulton, R. (2003). Influence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science, 301, 386–389.

Cervilla, J., Molina, E., Rivera, M., …& Gutierrez, B. (2007). The risk for depression conferred by stressful life events is modified by variation at the serotonin transporter 5HTTLPR genotype: evidence from the Spanish PREDICT-Gene cohort. Molecular Psychiatry, 12, 748-755.

Chipman, P., Jorm, A., Prior, M., Sanson, A., Smart, D., Tan, X., & Easteal, S. (2007). No interaction between the serotonin transporter polymorphism (5-HTTLPR) and childhood adversity or recent stressful life events on symptoms of depression: results from two community surveys. American Journal of Medical Genetics, 144, 561-565.

Clarke, H., Flint, J., Attwood, A., & Monafo, M. (2010). Association of the 5-HTTLPR genotype and unipolar depression: a meta-analysis. Psychological Medicine, 40, 1767-1778.

Culverhouse, R., Bowes, L., Breslau, N., … & Bierut, L. (2013). Protocol for a collaborative meta-analysis of 5-HTTLPR, stress, and depression. BMC Psychiatry, 13, 304.

Drury, S. S., Gleason, M. M., Theall, K. P., Smyke, A. T., Nelson, C. A., Fox, N. A., & Zeanah, C. H. (2012). Genetic sensitivity to the caregiving context: The influence of 5HTTLPR and BDNF val66met on indiscriminate social behavior. Physiology & behavior, 106, 728-735.

Gillespie, N.A., Whitfield, J.B., Williams, B., Heath, A.C., & Martin, N.G. (2005). The relationship between stressful life events, the serotonin transporter (5- HTTLPR) genotype and major depression. Psychol Med, 35, 101–111.

Hariri AR, Drabant EM, Munoz KE, Kolachana BS, Mattay VS, Egan MF, et al. (2005). A susceptibility gene for affective disorders and the response of the human amygdala. Arch Gen Psychiatry, 62, 146 –152.

Heim, C., & Binder, E. (2012). Current research trends in early life stress and depression: Review of human studies on sensitive periods, gene–environment interactions, and epigenetics. Experimental Neurology, 233, 102-111.

Kaufman J, Yang BZ, Douglas-Palumberi H, Houshyar S, Lipschitz D, Krystal JH, et al. (2004). Social supports and serotonin transporter gene moderate depression in maltreated children. Proc Natl Acad Sci U SA, 101, 17316–17321.

Karg, K., Burmeister, M., Shedden, K., & Sen, S. (2011). The serotonin transporter promoter variant (5-HTTLPR), stress, and depression meta-analysis revisited: evidence of genetic moderation. Arch. Gen. Psychiatry, 68, 444–454.

Kim, J., Stewart, R., Kim, S., …& Yoon, J. (2007). Interactions between life stressors and susceptibility genes (5-HTTLPR and BDNF) on depression in Korean elders. Biological Psychiatry, 62, 423-428.

Koster, E., Lissnyder, E., Derakshan, N., & Raedt, R. (2011). Understanding depressive rumination from a cognitive science perspective: The impaired disengagement hypothesis. Clinical Psychology Review, 31, 138-145.

Lasky-Su, J.A., Faraone, S.V., Glatt, S.J., Tsuang, M.T. (2005). Meta-analysis of the association between two polymorphisms in the serotonin transporter gene and affective disorders. Am. J. Med. Genet. B Neuropsychiatr. Genet., 133B, 110–115.

Levinson, D.F. (2006). The genetics of depression: a review. Biol. Psychiatry, 60, 84–92.

Lunetta, K. (2008). Genetic Association Studies. Circulation, 118, 96-101.

Lyubomirsky, S., Layous, K., Chancellor, J., & Nelson, S. (2015). Thinking About Rumination: The Scholarly Contributions and Intellectual Legacy of Susan Nolen-Hoeksema. Annual Review of Clinical Psychology.

Munafo, M.R., Brown, S.M.,& Hariri, A.R. (2008a). Serotonin transporter (5-HTTLPR) genotype and amygdala activation: a meta-analysis. Biol. Psychiatry, 63, 852–857.

Munafo, M.R., Durrant, C., Lewis, G., & Flint, J., (2008b). Gene×environment interactions at the serotonin transporter locus. Biol. Psychiatry, 65, 211–219.

Power, T., Stewart, R., Ancelin, M., Jaussent, I., Malafosse, A., & Ritche, K. (2010). 5HTTLPR genotype, stressful life events and late-life depression: No evidence of interaction in a French population. Neurobiology of Aging, 31, 886-887.

Ressler, K.J., Bradley, B., Mercer, K.B., Deveau, T.C., Smith, A.K., Gillespie, C.F., Nemeroff, C.B., Cubells, J.F., & Binder, E.B. (2010). Polymorphisms in CRHR1 and the serotonin transporter loci: gene×gene×environment interactions on depressive symptoms. Am. J. Med. Genet. B Neuropsychiatr. Genet., 153, 812–824.

Risch, N., Herrell, R., Lehner, T., Liang, K.Y., Eaves, L., Hoh, J., Griem, A., Kovacs, M., Ott, J., & Merikangas, K.R. (2009). Interaction between the serotonin transporter gene (5-HTTLPR), stressful life events, and risk of depression: a meta-analysis. JAMA, 301, 2462–2471.

Sharpley, C. F., Palanisamy, S. K., Glyde, N. S., Dillingham, P. W., & Agnew, L. L. (2014). An update on the interaction between the serotonin transporter promoter variant (5-HTTLPR), stress and depression, plus an exploration of non-confirming findings. Behavioural brain research, 273, 89-105.

Surtees, P., Wainwright, N., Willis-Owen, S., Luben, R., Day, N., & Flint, J. (2006). Social Adversity, the Serotonin Transporter (5-HTTLPR) Polymorphism and Major Depressive Disorder. Biological Psychiatry, 59, 224-229.

Uher, R., Caspi, A., Houts, R., Sugden, K., Williams, B., Poulton, R., & Moffitt, T.E. (2011). Serotonin transporter gene moderates childhood maltreatment's effects on persistent but not single-episode depression: replications and implications for resolving inconsistent results. J. Affect. Disord., 135, 56–65.

Vos T, Flaxman AD, Naghavi M, et al. (2012). Years lived with disability (YLDs) for 1160 sequelae of 289 diseases and injuries 1990-2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet, 380, 2163-2196.

Wilhem, K., Mitchell, P., Niven, H., … & Shofeld, P. (2006). Life events, first depression onset and the serotonin transporter gene. British Journal of Psychiatry, 188, 210-215.