Epidemiological and molecular genetic data have established that both environmental and genetic factors contribute to the risk of major depression, with heritability estimates ranging between 30% and 40%, with the remaining variance influenced by mostly individual specific environmental factors [Kendler, 1995; Levinson, 2006]. Twin and family studies have suggested that the genetic contribution is most likely oligo- to polygenetic, with many genetic variations of smaller effect contributing in an additive or interactive manner [Levinson, 2006]. Recently, two large genome-wide association studies for unipolar depression have been published, which reported more or less negative results for larger main genetic effects [Muglia et al., 2008; Sullivan et al., 2008]. Both studies were case–control associations, and neither included environmental exposure as a factor nor investigated gene × gene (G × G) interactions. Inclusion of gene × environment (G × E) and G × G interactions in such analyses might, however, be critical to detect genetic effects in unipolar depression. Genome-wide G × E or gene × gene × environment (G × G × E) studies for unipolar depression have, to the best of our knowledge, not yet been conducted and would require very large sample sizes to attain sufficient power. In this study, we therefore focus on a candidate G × G × E interaction.

The most established environmental risk factors for major depression are stressful life events, with early exposure to trauma being a particularly strong risk factor in conferring risk for the development of depression. Specifically, a history of adverse childhood experience (e.g., child abuse or early separation from parents) increases the risk for developing major depression as an adult by two- to threefold, and there is a highly significant dose–response relationship between the degree of early adverse experience and depressive disorders as adults [Chapman et al., 2004; Anda et al., 2006; Bradley et al., 2008]. Early trauma has been associated with long-lasting abnormalities in several biological systems that have also directly been implicated in the pathophysiology of major depression, including the serotonergic (5HT) and the corticotropin-releasing factor (CRF) systems [Arborelius et al., 1999; Heim and Nemeroff, 2002; Edwards et al., 2003; Nemeroff, 2004].

In fact, both rodents and non-human primates exposed to early adverse experiences exhibit evidence of hyperactivity of the CRF system as adults, as measured by cerebrospinal fluid (CSF) CRF concentrations, CRF and CRF receptor type 1 (CRF1) mRNA expression, and CRF receptor binding [Plotsky and Meaney, 1993; Coplan et al., 1996; Ladd et al., 2000; Plotsky et al., 2005]. These observations in laboratory animals are consistent with the findings in humans where early life trauma is a strong predictor of CSF CRF concentration in adults [Carpenter et al., 2004; Lee et al., 2005, 2006]. It is also associated with an enhanced stress response to psychosocial stressors, such as the Trier Social Stress Test, as well as to endocrine challenge tests, including the CRF stimulation test and the combined dexamethasone suppression/CRF stimulation test [Heim et al., 2000, 2001, 2008; Tyrka et al., 2008].

In rodents, maternal separation in the early postnatal period also leads to persistent alterations in serotonergic systems, including changes in cell firing of serotonergic raphe neurons, frontal cortical 5-HT concentration, and expression levels of certain serotonin receptors in cortex and hippocampus [Arborelius et al., 1999; Matthews et al., 2001; Gardner et al., 2005; Vicentic et al., 2006; Arborelius and Eklund, 2007; Lambas-Senas et al., 2009]. In primates, stressful rearing conditions are associated with altered CSF concentrations of the major serotonin metabolite 5-hydroxyindoleacetic acid (5-HIAA) [Mathew et al., 2002] and in humans, child abuse was shown to negatively correlate with CSF 5-HIAA concentrations [Roy, 2002] and to be associated with dysregulation in serotonergic function [Tiihonen et al., 1997; Rinne et al., 2000; Steiger et al., 2001].

The fact that early adverse experience impacts both CRF and 5-HT systems is likely due, in part, to the strong interconnection between the two neural circuits. CRF has been reported to inhibit serotonergic neurons in the dorsal raphe nuclei [Price et al., 1998; Kirby et al., 2000] and may thus modulate serotonin release in terminal areas [Price and Lucki, 2001]. Very recent data suggest that CRF leads to inhibition of raphe 5-HT neurons via increased inhibitory action of GABA [Kirby et al., 2008]. Brain regions central to the stress response such as the hippocampus and prefrontal cortex in turn receive dense serotonergic innervation, and serotonin release from the raphe leads to inhibition within the basolateral amygdala [Rainnie, 1999].

Maternal deprivation in rats is associated with an increase in CRF binding sites in the raphe nuclei [Ladd et al., 1996]. In primates exposed to early adverse environment, there is a negative correlation between CSF CRF and 5-HIAA concentrations [Mathew et al., 2002], supporting an interacting effect of these two systems in the biology of early adverse experience. Consistent with these data, a number of studies have demonstrated that serotonin reuptake inhibitor (SRI) antidepressants reverse the stress-related effects of CRF administration [Lowry et al., 2008], and human depressed patients treated with SRIs show decreased concentrations of CSF CRF [De Bellis et al., 1993; Nikisch et al., 2005]. Furthermore, dysregulation in both systems has been implicated in the pathophysiology of major depression. Child abuse-induced alteration of these systems could thus be causally involved in the increased risk for depression conferred by early life trauma.

In light of these data, it is thus not surprising that genetic variants impacting the function of genes within these systems have been shown to moderate the effects of early trauma on depression. The serotonin transporter-linked polymorphic region [Lesch et al., 1996; Caspi et al., 2003; Eley et al., 2004; Gillespie et al., 2005; Grabe et al., 2005; Kendler et al., 2005; Munafo et al., 2008; Surtees et al., 2006; Taylor et al., 2006; Uher, 2008] (5-HTTLPR) has repeatedly been shown to interact with life stress and childhood maltreatment to predict adult depression. While a meta-analysis could not validate this interaction for number of life events [Risch et al., 2009], no meta-analysis so far has been reported for the interaction with childhood maltreatment. A number of studies report a significant moderation of the effects of childhood maltreatment on adult depressive or psychiatric disorders [Caspi et al., 2003; Kaufman et al., 2004; Gibb et al., 2006; Roy et al., 2007; Richardson et al., 2008], in that the short allele of this polymorphism is associated with an increase in symptom severity with early trauma severity. Interaction of the 5-HTTLPR may thus be more robust with early trauma than with life events in general. We have recently reported an interaction of child abuse with polymorphisms in the type 1 CRF receptor gene, CRHR1, to predict adult depression in two independent cohorts [Bradley et al., 2008]. This finding has now been replicated in an independent, Caucasian cohort [Polanczyk et al., 2009], supporting the trans-ethnic relevance of the investigated interaction. This G × E interaction has also been shown to have endocrine repercussions, with the CRHR1 polymorphisms moderating childhood maltreatment-related HPA-axis hyperactivity [Tyrka et al., 2009].

Because of the above-described interaction between these two systems on a biological level, it is also possible that G × G interactions across these systems moderate the impact of early adverse experience. In fact, such G × G × E interactions have been reported for the BDNF Val66Met polymorphism as well as a functional interleukin-6 (IL6) polymorphism with the 5-HTTLPR, for example [Kaufman et al., 2006; Kim et al., 2007; Bull et al., 2008; Wichers et al., 2008]. The aim of the current study was to examine the combined interaction of the 5-HTTLPR and CRHR1 polymorphisms with childhood abuse in predicting current symptoms of depression in a sample of over 1,000 African-American men and women.

This article presents three primary findings. First, we did not find a main effect of 5-HTTPLR genotype or an interaction effect of 5-HTTPLR and child abuse on current level of depression symptoms in a sample of over 900 African-American men and women but did replicate an interaction of child abuse and 5-HTTLPR on lifetime diagnosis of major depression in a subsample of 234 individuals. Second, using a data set of 624 more individuals, we extended our previously reported interaction of CRHR1 variants with child abuse to predict adult depressive symptoms in a sample of now over 1,000 individuals. Third, we found that the 5-HTTLPR interacts with CRHR1 variants and child abuse to predict current, adult depressive symptoms: individuals carrying the risk alleles in both genes exhibit clinically relevant depressive symptoms at less severe levels of child abuse than individuals with no or only one of the risk alleles.

To the best of our knowledge this the first article to present data on 5-HTTPLR and depressive symptoms in an African-American sample—though a previous study in a similar population investigated this polymorphism as predictor of suicide attempts in a sample with a history of drug abuse [Roy et al., 2007]. The observed allele frequencies for the 5-HTTLPR, including the VL and XL alleles, are in agreement with previous publications and support that the SS genotype (6–7%) is much less frequent in individuals from African-American descent than in European Americans (20–25%) [Gelernter et al., 1997, 1998; Hu et al., 2006]. The XL allele has thus far only been observed in individuals from African descent and VL allele had previously been detected in Asian populations [Gelernter et al., 1997; Nakamura et al., 2000]. The few VL alleles observed in this sample might be related to admixture from native American or Asian ancestry that we observe in our sample [Bradley et al., 2008].

Consistent with the majority of the extant literature, we did not find a main effect of 5-HTTLPR genotype on current symptoms of depression in this sample. We replicate the previously described increased prevalence of incidence of lifetime major depression in individuals carrying the 5-HTTLPR S allele and exposed to child abuse [Caspi et al., 2003], but this interaction was not seen when using the functional classification of Hu et al. 2006. Possibly, the higher prevalence of the low-function rs25531 G allele in African-American compared to Caucasians as well as additional genetic differences of this locus depending on ethnic background might lead to this discrepancy. We, however, failed to find a significant interaction of childhood abuse and 5-HTTLPR on current depression symptoms in this sample. To resolve the conflicting data on lifetime versus current depression for this G × E interaction in our sample, we are currently gathering more comprehensive data on factors such as age of onset of abuse, age of onset, and recurrence of depression symptoms in the current sample which will allow us to more closely analyze the roles of these factors in the current sample. While a recent meta-analysis concluded that the observed interaction of the 5-HTTLPR with life events to predict adult depression is likely attributable to chance [Risch et al., 2009], these controversial results might be due to sample-related differences in such factors including age, gender, type of environmental exposure (early life trauma vs. recent life stressors), and type of depression (recent onset adult vs. chronic) [Brown and Harris, 2008].

In a sample expanded to 1,059 individuals, we were able to extend our previously reported interaction of CRHR1 variants and child abuse to predict adult depressive symptoms [Bradley et al., 2008]. Using more graded levels of child abuse as the environmental variables, we could show that the protective effects of CRHR1 variants are more pronounced with the presence of at least one type of severe abuse. However, it is important to note, as in Figure 3, that sufficiently high levels of environmental exposure to trauma may overcome any genetic effects. The additional samples not reported in the initial manuscript show an interaction of CRHR1 variants and child abuse in the same direction, but less pronounced than in the first sample. This might be due to differences of the initial sample versus the second set of recruited individuals. The more recently recruited individuals had lower BDI and CTQ scores and were on average 2 years older, while gender and CRHR1 haplotype and 5-HTTLPR genotype distribution was the same. These differences in environmental exposure and depressive symptoms might, in part, explain the differences in effect size between the first and second set of samples (see Fig. 2). We did not see a significant interaction of CRHR1 variants and child abuse on lifetime diagnosis of major depression; however, the distribution of the protective alleles in the different groups was consistent with an over-representation of the protective CRHR1 alleles in individuals with child abuse but no lifetime major depression. This analysis was further limited by a small sample size, prohibiting analysis with a more graded child abuse variable due to very small Ns/cell (see Table IV).

The main finding of this manuscript is that the 5-HTTLPR interacts with CRHR1 variants and child abuse to predict adult depressive symptoms. In fact, the protective effects of the CRHR1 TCA haplotype were only observed in individuals carrying the 5-HTTLPR LL genotype. On the other hand, the 5-HTTLPR S allele appeared to enhance the effects of lower levels of child abuse only in the additional presence of the CRHR1 risk haplotypes. Overall, individuals carrying the risk alleles in both genes exhibit more severe depressive symptoms already at lower “doses” of child abuse than individuals with no or only one of the risk alleles (see Fig. 3B). The finding that this interaction is present when child abuse is divided across three levels but not across two levels is consistent with research suggesting a dose–response relationship between the number of types of adverse experiences in early life and increased risk for a number of health and behavioral problems, including major depression, over the lifespan [Edwards et al., 2003; Chapman et al., 2004; Anda et al., 2006]. Note, however, that with sufficient levels of childhood abuse, any amount of genetic “protection” may be overcome.

Although this G × G × E interaction needs further independent replication to be confirmed, it is well in agreement with the previously reported interactions between the serotonergic and CRF system on a cellular and systems level as described in the introduction [De Bellis et al., 1993; Price et al., 1998; Rainnie, 1999; Kirby et al., 2000, 2008; Price and Lucki, 2001; Nikisch et al., 2005; Lowry et al., 2008]. This G × G interaction could also help to clarify the meaning of G × E interaction findings with the 5-HTTLPR or CRHR1 variants alone. In fact, different frequencies in the combination of 5-HTTLPR and CRHR1 risk or protective variants could alter the effect size of the individual G × E interaction, if the second variant is not considered. Likewise, our finding that this interaction is present when child abuse is divided across three levels but not across two-level points to the importance of paying attention to the ways in which broad constructs such as “stressful life events” or even more specific constructs such as “child abuse” are defined in evaluating the meaning of G × E research. Within these broad categories are varying stressful experiences for which the underlying neurobiological processes contributing to risk/resilience are likely to be different.

Our study has several limitations such as the use of a non-diagnostic, self-report measure of current depression symptoms for the main analysis. A subset of our data now has lifetime and current psychiatric diagnoses according to the SCID-DSMIV [First et al., 1995] and we have previously reported good agreement of BDI scores >16 and current major depressive episode [see Bradley et al., 2008]. Our interaction results might thus be extended at least to the presence of a current major depressive episode. Unfortunately, the sample with data on current major depression does not have sufficient power yet for meaningful G × E and even less for G × G × E interaction analyses. Likewise, our sample with data on lifetime diagnosis of major depression is still being collected and our G × E interaction analyses using this data should be considered preliminary. As with diagnosis of current depression, our current sample size for lifetime diagnosis of depression does not allow for G × G × E analyses. As with the majority of G × E studies to date, our data are cross-sectional and rely on adult self-reported history of childhood abuse and are thus likely confounded to a certain degree by recall bias. The fact that our data are gathered in sample of low income, African-American men and women presenting for primary care in public hospital also limits the generalizability of the results. However, this limitation is mitigated by the facts that exposure to traumatic events including childhood abuse occurs at a notably higher rate among low income, urban dwelling African-American populations [Alim et al., 2006; Bierut et al., 2007] with a high percentage of this exposure occurring during youth [Shakoor and Chalmers, 1991; Fitzpatrick, 1993]. This level of trauma exposure is associated with increased levels of risk for physical and mental health problems and associated healthcare costs [Fullilove et al., 1998; Schwartz et al., 2005, 2006]. Additionally, the limitation of self-report of history of abuse is a well-known one, which is important to attend to, but which does not alter the primary findings. Certainly, prospective studies are indicated in the future. Understanding the complex relationship between risk and resilience for disease among adults with a history of childhood abuse, particularly in highly traumatized high-risk populations is of high public health relevance.

By design, G × G × E interactions require large sample sizes, so that independent replication and extension in a larger sample will be needed to confirm our finding. Nonetheless, our sample represents one of the larger samples available for G × E interaction analyses and suggests that more complex analyses than case–control associations are necessary to identify the genetic variants relevant in the genetics of depression.