To investigate similarities and differences in the serotonergic diathesis for mood disorders and suicide attempts, we conducted a study in a cohort followed longitudinally for 22 years. A total of 1255 members of this cohort, which is representative of the French-speaking population of Quebec, were investigated. Main outcome measures included (1) mood disorders (bipolar disorder and major depression) and suicide attempts by early adulthood; (2) odds ratios and probabilities associated with 143 single nucleotide polymorphisms in 11 serotonergic genes, acting directly or as moderators in gene–environment interactions with childhood sexual or childhood physical abuse (CPA), and in gene–gene interactions; (3) regression coefficients for putative endophenotypes for mood disorders (childhood anxiousness) and suicide attempts (childhood disruptiveness). Five genes showed significant adjusted effects (HTR2A, TPH1, HTR5A, SLC6A4 and HTR1A). Of these, HTR2A variation influenced both suicide attempts and mood disorders, although through different mechanisms. In suicide attempts, HTR2A variants (rs6561333, rs7997012 and rs1885884) were involved through interactions with histories of sexual and physical abuse whereas in mood disorders through one main effect (rs9316235). In terms of phenotype-specific contributions, TPH1 variation (rs10488683) was relevant only in the diathesis for suicide attempts. Three genes contributed exclusively to mood disorders, one through a main effect (HTR5A (rs1657268)) and two through gene–environment interactions with CPA (HTR1A (rs878567) and SLC6A4 (rs3794808)). Childhood anxiousness did not mediate the effects of HTR2A and HTR5A on mood disorders, nor did childhood disruptiveness mediate the effects of TPH1 on suicide attempts. Of the serotonergic genes implicated in mood disorders and suicidal behaviors, four exhibited phenotype-specific effects, suggesting that despite their high concordance and common genetic determinants, suicide attempts and mood disorders may also have partially independent etiological pathways. To identify where these pathways diverge, we need to understand the differential, phenotype-specific gene–environment interactions such as the ones observed in the present study, using suitably powered samples.
Suicidal behaviors (SBs) and mood disorders (MDs) cause considerable psychosocial and economic difficulties for individuals, families and public health systems.1, 2, 3 SBs are closely, but not exclusively, associated with MDs and represent their most serious complications.4 Personal and family histories of MDs predict both attempted and completed suicide.5, 6, 7
Heritability figures indicate a multifactorial mode of inheritance, with genes explaining 17–55%5 of the variance in suicide attempts and 30–80% in major depression and bipolar disorder.8, 9, 10 Several lines of evidence point toward emotional, cognitive and behavioral deficits in both SBs and MDs.11, 12 These deficits have, in turn, been linked to alterations in the activity of serotonin—a master regulator of other neurotransmitters and a contributor to early brain development.13, 14 Lower concentration, binding, neurotransmission, and reuptake of serotonin and its metabolites are markers of the risk for suicidality and MDs.15, 16
Several serotonergic genes (5HTT—also known as SLC6A4, TPH1, TPH2, HTR1A, HTR2A, HTR2C and HTR7) have been examined as candidates for bipolar disorder, major depression and suicide attempts, most often with inconsistent findings.17, 18, 19 High concordance of MDs and SBs has made it difficult to disentangle shared and unique genetic contributions to these phenotypes.5 For example, it is unclear if variation in the gene coding for the serotonin transporter—an extensively investigated candidate gene whose protein product controls the magnitude and duration of the serotonergic neurotransmission20—shows specificity for the diathesis for major depression, bipolar disorder or suicide attempts, or whether it is relevant for all three.21, 22 Similar questions could be raised regarding all other serotonergic genes.
Poor replicability of significant findings, methodological and population heterogeneity, low statistical power and limited knowledge about the precise roles of candidate genes23, 24, 25 have been invoked as reasons behind the relative lack of success and poor understanding of the diathesis of SBs and MDs despite two decades of active molecular research. In addition, most candidate genes are rarely investigated within the context of other genes (for example, epistasis or gene–gene interactions), environments (gene–environment interactions and correlations) or possible endophenotypes.
We26, 27, 28 and others24, 29, 30, 31, 32, 33 have shown that behavioral dysregulation reflected in impulsive aggression and related traits may be an endophenotype because it has been linked to increased risk for SBs and to the activity and variation of the serotonergic system. Anxiousness, neuroticism and internalizing behaviors have also been linked to serotonergic variants as well as to MDs.34, 35, 36 Personality traits such as these have been suggested as candidate endophenotypes33 because of their quantitative nature, moderate-to-high heritability and a closer temporal relationship to genes relative to psychiatric outcomes. However, very few candidate-gene studies have been able to take comprehensive analytical approaches and to adequately investigate the function of endophenotypes. This is because the vast majority of these studies have been cross-sectional, and as such, not useful for investigating mediating effects and temporal relationships.
Together with gene–environment correlation (genetically influenced exposure to selected environments),37 gene–environment interactions (genetically influenced sensitivity or resistance to environmental effects) may be responsible for the probabilistic rather than deterministic character of the relationship between genes and complex diseases37 and the poor replicability of genetic effects. Childhood abuse is one important early environmental risk factor for SBs and MDs.38, 39, 40, 41 For example, suicide attempts and bipolar disorder severity have been linked to childhood sexual (CSA) and physical abuse (CPA).42, 43, 44 Primate studies suggest that early life adversity alters stress sensitivity—a feature of both SBs and MDs—compromising serotonergic neurotransmission.45, 46 Stress is, for example, associated with SLC6A4 mRNA level47 and harsh parenting to HTR2A receptor density.48
Recognizing the interconnectedness of genes and environments in susceptibility to MDs and SBs, we investigated single nucleotide polymorphisms (SNPs) in 11 serotonergic genes in a cohort of young French Canadians followed since kindergarten. Our main objective was to identify variants associated with MDs and attempted suicide—a strong predictor of completed suicide49, 50—that act either directly, as moderators, or through mediators, that is, endophenotypes. Our main theoretical premise was that SNPs involved in susceptibility to SBs and MDs would, directly or through linkage disequilibrium (LD), lead to a reduction in serotonergic activity in a way that is conditional on the function of the particular gene (through dysregulation in uptake, binding or the removal of serotonin from the synaptic cleft) and its interactions with two salient environmental factors and with other examined serotonergic genes.
We also set out to resolve differences in the diatheses for MDs and suicide attempts, which, despite their relatedness, may exist at the level of genes, environmental factors and endophenotypes. We hypothesized that childhood disruptiveness, a construct related to impulsive aggression, and anxiousness, respectively, mediate genetic diatheses for suicide attempts and MDs. Several lines of evidence point toward emotional, cognitive and behavioral deficits in both SBs and MDs.11, 12 These deficits have, in turn, been linked to alterations in the activity of serotonin—a master regulator of other neurotransmitters and a contributor to early brain development.13, 14 Lower concentration, binding, neurotransmission, and reuptake of serotonin and its metabolites are markers of the risk for suicidality and MDs.15, 16
In terms of the specific moderating effects, guided by literature and theoretical evidence, we tested selected epistatic and gene–environment interactions involving parental CPA and contact CSA. These environmental factors were chosen to increase our knowledge of stressors beyond stressful life events and also to understand the differences in their effects on MDs and SBs.51, 52
Participants were selected from a larger cohort of children followed up for 22 years. This cohort was recruited in 1986–87 using a multistage sampling procedure53, 54, 55 and included children attending kindergarten in Quebec's public francophone schools. The cohort was enriched for children exhibiting disruptiveness. Of the total 3017 children, 2000 (1000 girls) were selected randomly and are considered representative of the young, French-speaking population. The initial sample in this study encompassing 1255 respondents was defined by their DNA availability (Supplementary Figure S1). Compared to the representative sample of 2000, respondents were not different in the parental age at birth of first child, maternal socioeconomic status or proportion living with both biological parents (Supplementary Table S1). The sample of respondents, however, had fewer males, and less family adversity in terms of parental education and paternal income (Supplementary Table S1). (Family adversity is a composite index described in more detail below in Measures: environmental factors section.) As both family adversity and gender were related to attrition in our previous studies in this cohort, we used them to construct weights for multivariate analyses. These weights represent the inverse of each individual's probability of being in the original sample conditional on gender and family adversity.
The assessment schedule had four stages: (1) childhood, yearly assessments from age 6 through 12 (n=3017); (2) mid-adolescence; average age 15.7 (n=1715); (3) early adulthood, average age 21.4 (n=1684); (4) mid-adulthood (the DNA collection stage), average age 27.3 (n=1255).
The study was approved by the research ethics boards of the University of Montreal and McGill University. Written informed consent was obtained from all subjects.
Eleven serotonergic genes were selected on the basis of a literature review, existing knowledge about their physiological roles, and availability of suitable and informative genetic markers. We chose common tag SNPs (minor allele frequency (maf) >5%) and SNPs located up to 5 kbp upstream of the transcription site (maf >5%) (Supplementary Table S2). Tag SNPs were obtained using HapMap for the Utah residents with ancestry from northern and western Europe56 and Tagger's57 multimarker-tagging procedure (r2>0.8). This procedure examines multiallelic LD (correlation of physically close alleles due to co-transmission) to determine the most parsimonious set of SNPs capturing all common genetic variants in the specified gene. Tagger failed to identify tagging SNPs in HTR1A where we relied on common (maf >5%), nonsynonymous SNPs instead. Ensemble, Pupa and NCBI databases were used to identify upstream SNPs. In addition to 143 serotonergic SNPs, for detection of population stratification we genotyped 42 anonymous markers spread across the genome and located outside gene-coding regions. LD is represented with r2 and D′ (Supplementary Figure S4).
We used a high-throughput, 768-SNP Illumina platform and GoldenGate panel.58 Marker density was about 13 SNPs per gene across the 11 genes. Several initially selected tag SNPs had to be replaced by related SNPs, where possible, due to poor design scores (<0.5). SNPs less than 60 bp apart were eliminated. The initial genotyping success rate for the SNPs was 95.7%. After we excluded 14 samples with suboptimal call rates (<95%), the average genotyping rate in the remaining sample was 99.9%.
Family adversity: Family adversity index was based on maternal reports on: (1) family structure (two parent or single), (2) educational level of both parents (or the parent with whom the child was living), (3) occupational status of both parents (or occupation of the parent with whom the child was living) based on the Blishen's59 occupational prestige scale and (4) mother's and father's age at birth of the first child. A score of 0 corresponds to family structure where the child was living with both biological parents and a score of 1 to all other cases. Parental educational level, parental occupational status and mother's or father's age at birth of the first child were scored 1 when the individual scores were in the lower quartile of the respective variable distribution. A score of 0 was given to scores above the first quartile of the distributions. Therefore, higher values correspond to higher family adversity levels at the time when the participants were about 6 years of age.
Contact childhood sexual abuse: Adverse Childhood Experiences Study Questionnaire.60 CSA was assessed with a subset of three yes-or-no questions in early adulthood, with each question focusing on one type of sexual violence involving physical contact before age 18 (fondling or attempted/realized vaginal, anal or oral sexual act), the identity of the perpetrator (family members, school peers/personnel, short/long-term romantic acquaintances or strangers) and the responder's own evaluation of whether the incident was sexually abusive. The test–retest κ coefficient of the original scale was 0.5.61
Childhood physical abuse: Revised Conflict Tactics Scales.62, 63 This 32-item scale assesses self-reported childhood incidences of nonviolent discipline, psychological aggression, and physical assault in parent–child and other family relationships. We administered a subset of 14 items in early adulthood. These items evaluate the presence/frequency of severe/very severe physical aggression, abuse and injuries perpetrated by each parent (biological, adaptive or any person in a parental role) against the respondent as a child. The test–retest κ coefficient of the original scale was 0.7.61
The Social Behavior Questionnaire:64 It assesses several childhood traits using teacher reports. These scores were used to identify behavioral trajectories as described in the Methods (Multivariate analyses). Disruptiveness (Cronbach's α, 0.90) was tested as an endophenotype for suicide attempts. It encompasses aggressive, antisocial, oppositional, hyperactive traits evaluated with 13 items: destroys one's own things or those of others; fights with other kids; is not liked by peers; irritable; disobedient; lies; mistreats, intimidates peers; does not share material used for a particular task; blames others; inconsiderate of others; hits and kicks others; fidgets, squirms, cannot keep still; agitated, always running and jumping, restless. Anxiousness (Cronbach's α, 0.74) was tested as an endophenotype for MDs and assessed with six items: fearful or afraid of things or new situations; is worried, worries about many things; cries easily; has a tendency to work alone; looks sad, unhappy, tearful; easily distracted.
Covariates and outcomes
Diagnostic Interview Schedule for Adults (DIS): This, using DSM-III-R criteria,65 assesses mood (major depression, bipolar disorder and dysthymia), anxiety (generalized anxiety, panic and phobias) and substance abuse disorders (abuse and/or dependence on drugs, alcohol and nicotine). In addition to being one of our final outcomes, personal MD history was a covariate in the suicide attempts model. Another DIS-derived covariate—history of psychopathology—was a count variable summarizing the number of diagnoses in each individual. The DIS was also used for information on parental suicide attempts (positive history was coded as ‘1’ and negative as ‘0’) and parental MDs (a count variable with scores ranging from 0 to 4).
Suicide attempts. Suicide attempt status was based on adolescent and adult assessments. Adolescent history was obtained from parental/adolescent responses to a question from the Diagnostic Interview Schedule for Children:66 ‘Have you already attempted suicide?’ Either parental or self-report was sufficient for a person to be classified as an attempter. Adult suicide attempts were assessed with a question from the Suicidal Intent Scale:67 ‘Have you already attempted suicide?’ Along with substance and anxiety disorders, suicide attempts were covariates in the model of MDs.
Even though French Canadians appear to have descended from a small number of founders,68 we examined evidence of population outliers—individuals displaying different allele frequency distributions from the rest of the sample (Supplementary Figure S2).69 We used the genotype log likelihood test statistic with a cutoff of P=0.01, identifying 12 outliers (Supplementary Figure S3). Hardy–Weinberg equilibrium was not rejected for any of the SNPs.
We first investigated direct, that is, main effects exerted by SNPs and haplotypes. Haplotype blocks were identified with ‘Entropy_blocker’70 in R.16 Associations of individual haplotypes (frequency >5%) with MDs and suicide attempts were examined following significant global tests using the ‘haplo.score’ R function71 in the entire sample and also in the CSA and CPA subsamples.
χ2-Test and Fisher's exact test (used when the expected cell count was 5) in conjunction with a false detection rate (FDR) cutoff of 0.20 helped us to identify significant SNPs under allelic, recessive and dominant genetic models with respect to the minor allele. For every association test, we estimated the corresponding FDR using an empirical null distribution of the z-transformed P-values.72 The FDR is the proportion of false positives expected among tests with lower or equal P-value. SNPs with FDR 0.20 were further tested in adjusted multivariate models. Serotonergic effects were examined in the total sample and in CSA and CPA subsamples.
Multivariate analyses: main/moderating effects
Using a series of regression-based analyses adjusted for gender and psychopathology, we first identified the best combination of SNPs within genes by forcing together all the SNPs in a given gene that were selected at the univariate step (Supplementary Figure S2). Keeping the SNPs that remained significant within genes (P<0.05), we then included them together in a model across genes. We also tested if the results in the final models identified in this way changed if we applied weight adjusting for the probability of remaining in the sample conditional on the variables related to attrition: gender and family adversity. To impute the missing covariate values, we used the expectation-maximization method (EM).
In addition to main effects, we investigated two types of moderating effects:
Gene–gene interactions, that is, moderation of selected effects of serotonergic loci on suicide attempts and MDs by other serotonergic loci, and
gene–environment interactions, that is, moderation of childhood abuse effects on suicide attempts and MDs. Gene–environment interaction tests interrogated SNPs that showed association at the screening step in the CSA or CPA subsamples, along with their corresponding main effects, in both abused and nonabused groups. We expected that some SNPs may be related to susceptibility only in subjects also exposed to CSA or CPA. This multivariate step allowed us to validate the effects identified at the screening stage.
We also controlled for two types of confounding: (1) passive gene–environment correlation (by adjusting models for parental history of suicide attempts and MDs) and (2) evocative gene–environment correlation (by demonstrating that genotypes involved in gene–environment interactions did not influence exposure to CSA and CPA). Significant moderating effect was followed by post hoc tests to quantify regression slopes and examine their statistical significance.73
We investigated 15 two-way gene–gene interactions (Supplementary Table S3). The genes were selected on the basis of theoretical or empirical evidence (literature and epistatic interaction databases: www.genenetwork.nl; genecards.org, Entrez Gene), functional importance of a gene product in the system and evidence for synergistic or antagonistic functional effects between any two genes. We examined allelic–allelic, dominant–dominant, dominant–recessive, recessive–dominant and recessive–recessive model combinations under the interaction testing framework proposed by Millstein et al.74
Finally, in the presence of covariates, anxiousness and disruptiveness trajectories were investigated as endophenotypes mediating the significant main effects identified in the final models of suicide attempts and MDs. These trajectories are identified with semiparametric group-based modeling (SGM), a type of growth-mixture modeling. SGM assumes that the population is composed of a mixture of groups of youth following distinct developmental trajectories described by both the shape (‘low’, ‘increasing’) and estimated proportions of individuals following them. The procedure is described in more detail elsewhere.75
Mediators account for a portion or all of the association between a predictor and an outcome.76 Mediation testing consists of four regression steps76 designed to demonstrate associations between (1) predictor (for instance, a risk genotype) and outcome (in our case, suicide attempts or MDs); (2) predictor and mediator (here, trajectories of childhood disruptiveness or anxiousness); (3) mediator and outcome and (4) predictor and outcome, while controlling for the mediator of interest. If the requirements 1–3 are met and there is a drop in the association between the predictor and outcome in step 4, suggesting mediation, statistical significance of this indirect effect may be established with Sobel and Goodman tests.77
After exclusions (Supplementary Figure S1 lists exclusion criteria), the total analyzed sample consisted of 1121 individuals, including 117 suicide attempters (83 females) and 107 individuals (84 females) with MDs (82 with major depression, 25 with bipolar disorder, none with dysthymia). CSA victims (n=230) had a higher prevalence of suicide attempts (n=53, 23%) and MDs (n=44, 19.1%) than those without CSA histories (suicide attempts: n=61, 7.1%, χ12=48.9, P<0.001; MDs: n=63, 7.4%, χ12=28.1, P<0.001). Similarly, CPA victims (n=316), had higher rates of suicide attempts (n=55, 17.5%) and MDs (n=47, 15.0%) than controls without CPA (suicide attempts: n=59, 7.7%, χ12=22.9, P<0.001; MDs: n=60, 7.8%, χ12=12.9, P<0.001).
Mood disorders: The HTR2C receptor gene had four significant SNPs (rs2428706, rs2497530, rs4911874 and rs6579511) exhibiting dominant mode of inheritance in both the total sample and in the CPA subsample (Table 1). Significant HTR2A SNPs were recessive with two (rs927544 and rs9316235) identified in the total sample and four (rs9316235, rs9526240, rs9534496 and rs927544) in the CPA subsample. In contrast to SLC6A4 (rs2020942, rs3794808, rs4583306, rs3813034, rs1042173, rs12449783 and rs4325622), TPH2 (rs4760820) and HTR1A SNPs (rs878567 and rs6295), which were significant among CPA victims, HTR5A variation (rs1657268) contributed to MDs in the total sample. No gene–gene interactions were significant at this level, precluding further analyses. The most promising, although not statistically significant, was the interaction between SNPs rs5952799 and rs4760820 (P=0.0082, FDR=0.99).
Suicide attempts: TPH1 (rs10488683) and HTR2A (rs1885884) were each represented by one significant SNP in the total and CPA samples, respectively. Among CSA victims, HTR2A had three significant SNPs (rs1885884, rs7997012 and rs6561333) and HTR5A had one (rs1440449). No gene–gene interactions were found, precluding further analyses. The most promising, although not statistically significant, was the interaction between rs1042173 and rs17110532 (P=0.0104, FDR=0.28).
Mood disorders: HTR5A's global haplotype test was significant in the total sample and CSA subsample (Table 2). One of its three analyzed haplotypes (AATATACCGAA at rs1440454, rs2581842, rs2873379, rs1017488, rs1881691, rs6320, rs2241859, rs6597455, rs731107, rs1657268, rs1440449) was nonsignificantly overrepresented in cases with CSA (41 vs 30%, P=0.09). In the total sample, the AACCTTCCGAA haplotype was more frequent in cases than controls (27 vs 23%, P=0.08), although not significantly.
Suicide attempts: A statistically significant global test in the entire sample was obtained only in TPH1. Of the four sufficiently frequent TPH1 haplotypes containing rs10741734, rs1800532, rs10488683, rs10832876, rs685657, rs10488682, rs623580, rs652458 and rs546383, in this order, two were significantly associated with suicide attempts: haplotype TGATCTATG was more frequent in controls (17 vs 12%, P=0.043) and TGGCCATTG in cases (32 vs 26%, P=0.047). These haplotypes contained rs10488683, which was statistically significant on its own as shown below.
Main/moderating effects: The final models explained about 25% of the variance in both suicide attempts and MDs (Table 3). Detailed characteristics of the implicated genotypes are presented in Supplementary Tables S5a–c. Cook's statistic identified no outliers at a cutoff of 1.078 (the largest outlier value in MDs was 0.42 and suicide attempts 0.56). Multicollinearity was unlikely, as the highest correlation coefficient among our variables was 30%. To correct for the right skew, the two count variables representing history of psychopathology were square-root transformed. The final estimates obtained with the EM missing-value imputation method had narrower confidence intervals but were otherwise similar to unimputed estimates. Weighting for gender and family adversity did not change estimates of significance levels and odds ratios (data not presented).
Mood disorders: The final model implicated four genes (HTR1A, HTR5A, HTR2A and SLC6A4), with two (HTR1A and SLC6A4) moderating the relationship of CPA and MDs through gene–environment interactions: homozygosity for the T allele in rs878567 (HTR1A) was associated with a lower risk for MDs in CPA victims relative to homo/heterozygosity for allele C (Figure 2). A-allele homozygosity at rs3794808 (SLC6A4), in contrast, increased the likelihood of MDs, relative to hetero/homozygosity for the G allele. These two gene–environment interactions were not confounded by evocative gene–environment correlation, as none of the SNPs significantly predicted exposure to CPA (rs878567, P=0.055; rs3794808, P=0.66). In terms of passive gene–environment correlations, parental MDs showed weak associations with CSA (r2=5.6%, P=0.05) and CPA (r2=8.1%, P=0.005).
The two remaining SNPs contributed through recessive main effects: rs1657268 locus GG carriers had a 79% reduction in the risk of MDs relative to A-allele carriers. rs1657268 maps to intron 1 of HTR5A and is strongly correlated with two other SNPs (rs6597455 and rs1440449) (Supplementary Figure S4d). In contrast, AA carriers at rs9316235, an SNP located in intron 2 of HTR2A, had 2.4 times the risk of G carriers.
Suicide attempts: Only one SNP (TPH1, rs10488683) made a statistically significant contribution (Table 3) independently of gender (OR=1.2), parental suicide attempts (OR=2.8) and axis I diagnoses (OR=2.3). Representing an A–G transition, it maps to intron 2 and shows moderate LD with other TPH1 SNPs (Supplementary Figure S4c).
Of the four SNPs that showed the required univariate evidence, three HTR2A SNPs remained significant at the multivariate level: (1) rs7997012 T-allele homozygosity carried a lower risk for suicide attempts in the presence of CSA (Figure 3). This intron-2 SNP was in moderate LD with several SNPs: rs7322347, rs977003, rs6561333 and rs1923886 (Supplementary Figure S4b). (2) Carriers of at least one minor A allele at the rs6561333 locus had higher risk for suicide attempts. This SNP (rs6561333) is part of an LD block with rs956773757 (not typed in this study) and also shows strong correlation with rs1923886. (3) The rs1885884 GG genotype increased the risk for suicide attempts relative to genotypes containing allele C.
We found no evidence for evocative gene–environment correlation for any of these SNPs as none predicted exposure to childhood abuse (rs1440449, P=0.08; rs7997012, P=0.28; rs6561333, P=0.46; rs1885884, P=0.76). Parental attempts showed moderate associations with CSA (r2=10.5%, P=0.0001) and CPA (r2=9.1%, P=0.002).
Mediating effects: candidate endophenotypes
Trajectories of childhood anxiousness and disruptiveness, respectively, did not mediate main genetic effects on MDs or on suicide attempts. Anxiousness levels were not statistically different between MD cases carrying risk (HTR2A, rs9316235) or protective (HTR5A, rs1657268) genotypes compared to controls, making further analyses unnecessary.
Although correlated with childhood disruptiveness (r2=12%, P<0.001), the TPH1 risk genotype effect was not mediated by disruptiveness in the models adjusted for gender and personal history of psychopathology (Sobel statistic: 1.60, P=0.11; Goodman statistic: 1.68, P=0.093).
We examined serotonergic diathesis for MDs and suicide attempts, identifying associated similarities and differences at the level of serotonergic gene variants and environmental interactions. As four of the five implicated genes exhibited phenotype-specific effects, the risk for MDs and suicide attempts may involve somewhat different pathways, a possibility consistent with previous family-based empirical work.79
A TPH1 gene variant (rs10488683) was specific to suicide attempts, with its G allele exerting a direct effect that was independent of gender and psychopathology and unmediated by childhood disruptiveness.80 As rs10488683 is a tagging, intronic locus unlikely to be involved in splicing, it may act as a proxy for one of the five SNPs it captures through LD. One of these (rs684302) was, in fact, linked to SB in borderline women,81 as well as to depression,82 although only in the analyses uncorrected for multiple testing.
The TPH1 protein product catalyzes a rate-limiting step in the synthesis of serotonin. TPH1 has 21 kbp, 8 introns and 11 exons, and maps to chromosome 11p15.1. It shares 71% homology with TPH2.83 TPH1 has had a long history as a candidate gene in association studies of SBs and MDs, with intron 7 polymorphisms being investigated the most often.81, 84, 85, 86
Three of our candidate genes, each represented by a single SNP, contributed exclusively to the risk of MDs. A variant in HTR5A (rs1657268) was associated with a risk reduction of 80%. Unlike other HTR5A variants studied in reference to antidepressant response,87 autism,88 attention deficit hyperactivity disorder89 or suicide,90 rs1657268 has not been, to our knowledge, associated with psychiatric risk. Contrary to our hypothesis, and in accordance with some studies,91 its effect was not mediated by childhood anxiousness.
In this study, SLC6A4—the most widely investigated candidate gene for both SBs and MDs20—showed an effect on MD risk through an interaction with CPA. By going beyond its promoter insertion/deletion variant and stressful life events, we add new knowledge to the field that has focused largely on this variant.89, 92, 93 Our finding is also consistent with the suggested involvement of the proximal region of SLC6A4. Our SLC6A4 candidate (rs379808) is in LD with two SNPs in introns 3 and 9, and tags an intron-1 variant (rs12150214).
The CPA effect on MDs was also moderated by HTR1A. Its protein product has been associated with major depression,94 antidepressant response95 and hippocampal neurogenesis.96 The hippocampus is a part of the limbic system involved in memory and an important target of the hypothalamic–pituitary–adrenal (HPA) axis effectors. Memory deficits seen in MDs may be related to HPA axis dysregulation and adverse influence of stress hormones on the hippocampus.97, 98 One possible interpretation of our results is that CPA, through interactions (and possibly correlations) with rs878567, influences hippocampus-mediated memory deficits in MDs. The support for this hypothesis requires more research into the functional correlates of the rs878567's T allele, which in the present study buffers the effect of CPA on MDs.
We also show that HTR2A variability contributes to the risk for both MDs and suicide attempts, although through different mechanisms: gene–environment interactions in the case of suicide attempts and main genetic effects in the case of MDs. Earlier research has reported higher HTR2A receptor density in suicides and depressed suicides.99 Our finding is also supported by the evidence relating harsh parenting to HTR2A receptor density.48 Gene–environment interactions may explain the lack of consistent research support for its contribution to suicidality.90, 100
The only HTR2A SNP (rs9316235) contributing to MDs did not operate through gene–environment interactions or childhood anxiousness. This is in contrast to a recent study, which has found one of its variants (rs6313) to moderate urban–rural residency association with depressive symptoms,101 and may be attributed to the young age and nonclinical nature of our sample, or to insufficient statistical power.
The present gene–environment interaction findings are also strengthened by our adjusting for possible confounding by gene–environment correlations. The distinction between the two processes is important because each implies a different mechanism of action and type of child/parental genotype interplay with environmental factors.102 In agreement with previous studies, parental histories of MDs or suicide attempts increased the risk of MDs and suicide attempts up to three times.6, 103, 104 Also, given the significant associations between these variables and CSA and CPA, there is some evidence for gene–environment correlations.
Our findings should be considered in light of several limitations. The assessment of gene–environment correlations may have been negatively affected by the use of childhood information collected by means of parental interviews, given that parents share half of their genes with their children.37 Recall bias may have affected our measurement of CSA. Furthermore, genetic moderators may not affect association of childhood abuse and psychiatric risk directly but represent proxies for yet another moderating factor.105 Other potential stressors are related to the young age of our sample, as some participants may develop MDs or SBs later in life.
Limited statistical power prevented us from identifying genetic effects of relevance to individuals reporting both CSA and CPA. Similar limitations precluded tests of higher-order moderation by sex and other environmental factors. In addition to low power, limited knowledge of the interactions within the serotonergic system may have affected our ability to detect gene–gene interactions.
At the phenotypic level, we were unable to examine differences in serotonergic diatheses between major depression and bipolar disorder due to low numbers of individuals with the latter diagnosis. Recent years have seen an increase in research interest in phenotypes that could mediate the relationships between genetic effects and psychiatric disorders. Although our study was methodologically well suited to investigating these intermediate phenotypes, we did not find supportive evidence for the two investigated childhood behaviors. This lack of support may be related to several factors. Underestimation of the effect of the mediator and overestimation of the effect of the predictor is directly related to how accurately the mediator is measured.76 Childhood anxiousness and disruptiveness are relatively undifferentiated, broad and probably quite heterogeneous trait composites,106 and only two of many internal and external mediating factors.105, 107, 108 Moreover, as expected, individual SNPs explained only a small fraction of the total variance in our outcomes (for example, 2% in the case of TPH1). Also, the a priori probability of an association between genetic polymorphisms with psychiatric diagnoses and intermediate phenotypes is very low.105 Our personality-based endophenotypes themselves, furthermore, may depend on small-size gene–environment and epistatic interactions.101 We would have also benefited from a suitable measure of impulsive aggression administered in adulthood. Finally, one important disadvantage to taking a comprehensive approach such as ours is elevated risk of false positives.109
Our study also has a number of strengths. Methodologically, we benefited from a longitudinal, population-based cohort study, a sampling design associated with higher ‘prior association probabilities,’ weighting for variables related to attrition and reduction in several types of bias.110 Heeding calls for multidimensional investigations of neurotransmitter gene candidates,25 we examined diatheses for two significant psychiatric phenotypes discovering the importance of the environmental context for some serotonergic genes.
Independent replications of our results are necessary, although, for at least three reasons, this may be an even more elusive goal than it is with main genetic effects. First, standardized, well-defined, comprehensive measures of ‘environments’—in terms of their onset, frequency, temporality and severity—are lacking. Second, phenotypic heterogeneity is of special concern in psychiatry, where subjective variable ascertainment is the norm. Isolating well-defined subphenotypes or endophenotypes with higher genetic loading may help us identify genetic influences more effectively. Third, a better knowledge on the developmental trends in additive and interactive genetic effects can help us identify the times of high sensitivity to specific gene–gene interactions, gene–environment interactions and mediators.25, 111, 112
In sum, our investigation of direct, mediating and moderating influences by selected genetic, personality and environmental risk factors suggests several shared and unique influences on MDs and SBs. The final validation of these findings requires independent replications, a better molecular characterization of candidate SNPs, and, in general, a clearer understanding and better-defined hypotheses regarding the relationships of genetic, behavioral and environment factors with psychiatric risk.
We thank the families for their participation in the study as well as Dr Heather Cordell and Mr Derek Daws for their helpful comments on the analytical approach in the planning stage. We also thank the anonymous reviewers for their valuable comments.
About this article
Supplementary Information accompanies the paper on the Molecular Psychiatry website http://www.nature.com/mp)
Nature Reviews Neuroscience (2014)