Research Review: Gene–environment interaction (GxE) research in youth depression – a systematic review with recommendations for future research

Abstract

Background

Depression is a major public health problem among youth, currently estimated to affect as many as 9% of US children and adolescents. The recognition that both genes (“nature”) and environments (“nurture”) are important for understanding the etiology of depression has led to rapid growth in research exploring gene–environment interactions (GxE). However, there has been no systematic review of GxE in youth depression to date.

Methods

The goal of this article was to systematically review evidence on the contribution of GxE to the risk of child and adolescent depression. Though a search of PubMed and PsycINFO databases to 1 April 2010, we identified 20 candidate gene–environment interaction studies focused on depression in youth (up to age 26) and compared each study in terms of the following characteristics: research design and sample studied; measure of depression and environment used; genes explored; and GxE findings in relation to these factors.

Results

In total 80% of studies (n=16) found at least one significant GxE association. However, there was wide variation in methods and analyses adopted across studies, especially with respect to environmental measures used and tests conducted to estimate GxE. This heterogeneity made it difficult to compare findings and evaluate the strength of the evidence for GxE.

Conclusions

The existing body of GxE research on depression in youth contains studies that are conceptually and methodologically quite different, which contributes to mixed findings and makes it difficult to assess the current state of the evidence. To decrease this heterogeneity, we offer 20 recommendations that are focused on: (1) reporting GxE research; (2) testing and reporting GxE effects; (3) conceptualizing, measuring, and analyzing depression; (4) conceptualizing measuring, and analyzing environment; (5) increasing power to test for GxE; and (6) improving the quality of genetic data used. Although targeted to GxE research on depression, these recommendations can be adopted by GxE researchers focusing on other mental health outcomes.

Introduction

This article presents a systematic review of the evidence for genotype by environment interaction (GxE) in youth depression, which currently affects as many as 9% of U.S. youth (Avenevoli, Knight, Kessler, & Merikangas, 2008). The etiology of depression is complex, resulting from both genetic and environmental factors. Twin studies show that the heritability of youth-onset depression ranges from 30–80%, with the remaining variance explained by environmental factors (Rice, Harold, & Thapar, 2002). Although this suggests that depression is moderately to highly heritable, neither candidate gene nor genome-wide association studies have identified robust associations between specific genes and depression (Lopez-Leon et al., 2008; Shaikh et al., 2008; Shyn et al., 2009; Sullivan et al., 2009). In contrast, environmental risk factors for depression are well-documented and include poverty (Brooks-Gunn & Duncan, 1997; McLeod & Shanahan, 1996), negative family relationships and parental divorce (Gilman, Kawachi, Fitzmaurice, & Buka, 2003; Repetti, Taylor, & Seeman, 2002), and child maltreatment (Chapman et al., 2004; Widom, DuMont, & Czaja, 2007). However, only a minority of youth exposed to these environments develop depression, raising questions about individual differences in genetic vulnerability (or sensitivity) to adverse environments. GxE research addresses such questions by examining whether individuals with specific alleles (i.e. alternative forms of DNA sequence at a specific locus) or genotypes (i.e. the combination of alleles that an individual carries at a specific locus) are more or less sensitive to the effects of their environments (Brown & Harris, 2008; Khoury, Davis, Gwinn, Lindegren, & Yoon, 2005; Moffitt, Caspi, & Rutter, 2006; Uher & McGuffin, 2008).

The goal of the current article was to systematically identify and summarize studies that tested for GxE in relation to depression among children and adolescents. We focused specifically on the characteristics of these studies with respect to both their methods and findings. We did so in order to inventory the published literature with respect to their research designs, compliment existing meta-analyses (which have emphasized statistical findings, rather than methodological issues), and ultimately provide substantive conclusions that could guide future research in this area. We also sought to discern the level of heterogeneity in GxE findings that existed across studies specifically among youth, given that previous reviews, which included studies of multiple age ranges, concluded there were mixed findings for GxE (Munafo, Durrant, Lewis, & Flint, 2009; Uher & McGuffin, 2008, 2010). We focused on GxE research among youth, defined as age 26 and below, as this age span represents a “sensitive period” characterized by rapid social, psychological, and biological changes and marks the time when depression often first emerges (Rudolph, 2009; Steinberg & Morris, 2001). By focusing specifically on depression in youth, rather than depression in adulthood, we hoped to gain greater etiological insights into the development of this disorder. Previous reviews have also not attended to these developmental periods (Brown & Harris, 2008; Karg, Burmeister, Shedden, & Sen, 2011; Monroe & Reid, 2008; Munafo et al., 2009; Risch et al., 2009; Uher & McGuffin, 2008, 2010). For example, out of the 28 studies cited in a recent meta-analysis (Risch et al., 2009), only about one-third (10 studies) focused on youth.

Methods

Search Procedures

We systematically identified articles published as of 1 April 2010 through PubMed and PsycINFO search engines. We used a combination of database-specific index terms (i.e. “depression,” “genetics,” “environment,” “social environment,”) and individual terms located in the title or abstract (i.e. “gene,” “environment,” “interaction,” “moderation,” “modification,” OR “depress*”). We applied limits to searches in both databases to eliminate articles: (1) focused on individuals older than age 26; (2) not written in English; (3) focused on non-human animals, (4) not published in a peer-reviewed journal; or (5) based on reviews or meta analyses. We also searched for articles by examining the references pages of review articles, meta analyses, and other empirical articles published since 2005. Through these searches, we located 278 studies.

We applied the following criteria to these 278 articles to identify the articles for our review: (1) measured major depressive disorder or depressive symptoms as a unique outcome; (2) included participants age 26 or younger; and (3) focused on a specific candidate GxE interaction (as opposed to a twin or adoption study). We examined depression because it has received the most attention in the adult literature and popular press. In addition, we believed by narrowing our phenotype we would be able to make a more comprehensive assessment of the state of the literature. We used age 26 as a liberal cut-point to capture the developmental periods of childhood, adolescence, and young adulthood. This cut-point enabled us to be more inclusive and thus incorporate many articles relevant to youth as possible. After applying these criteria, 20 studies remained, which were published between 1 July 2003 and 1 April 2010.

Results

For the purposes of describing the current state of GxE research, we summarize the research design and study samples, measurement of depression, genotype, environment, and main study findings of these 20 studies (see Table 1).

Table 1.

GxE Studies Examining Depression or Depressive Symptoms in Youth (n=20)

Author (year)	Study Sample and Research Design*	Measurement of Outcome	Genetic Variant	Measurement of Environment	GxE Interaction Detected and Scale Tested
Aslund et al. (2009)	cross sectional; community-based; n = 1,482; Swedish youth ages 17–18	DSRS	5-HTTLPR	exposure to violence	Yes overall for symptoms (additive scale) and diagnosis (multiplicative scale); when stratified by sex, observed interaction with both outcomes in girls only
Benjet, Thompson, & Gotlib (2010)	cross sectional; community-based; n = 78 females; mean age 12.2	brief CDI	5-HTTLPR	relational peer victimization	Yes (additive scale)
Caspi et al. (2003)	longitudinal; n=847; assessed at age 26	DIS and survey	5-HTTLPR	stressful life events	Yes (additive and multiplicative scale)
Chipman et al. (2007)	two community-based Australian samples: (1) cross-sectional; n=2,095; 20–24 years old	GDAS	5-HTTLPR	stressful life events; childhood adversities	No (multiplicative scale)
	(2) longitudinal; n=584; age 15–16 and n=655; ages 17–18	SMFQ	5-HTTLPR	family stressors; persistent family adversity	No for family stressors when cohort aged 15–16 or 17–18 (multiplicative scale); Yes (multiplicative scale) with family adversity when cohort was aged 17–18, but not when 15–16
Chipman et al (2010)	cross-sectional; community-based; n=2,099; ages 20–24	GDAS	HTR1A	childhood adversity; stressful life events	No (multiplicative scale) when adjusted for multiple comparisons
Chorbov et al. (2007)	case-control; n=227; twins ages 13–23	diagnosis of a lifetime DSM-IV MDD	5-HTTLPR	traumatic life events	No (multiplicative scale) for s/l genotype or number or presence of s alleles; Yes (multiplicative) for number of LA alleles
Cicchetti et al (2007)	cross sectional; n=339; mean age 16.7; with/without history of maltreatment	DISC	MAOA and 5-HTTLPR	child maltreatment	Yes (additive scale) for MAOA; no for 5-HTTLPR
Cicchetti et al (2010)	cross sectional; n= 858; mean age 9.19; with/without history of maltreatment	CDI	5-HTTLPR	child maltreatment	No (additive scale)
Eley et al (2004)	cross sectional; n= 377; UK adolescents age 10–20	SMFQ	5-HTTLPR, 5HT2A, 5HT2C, MAOA, TPH1	high vs. low risk family environment	Trend (multiplicative scale) with 5-HTTLPR, 5HT2A, and TPH1 overall; among females, significant interaction with 5-HTTLPR; among males, significant interaction with TPH1; no for 5HT2C, MAOA
Gibb, Benas et al (2009)	longitudinal; n=74; mothers and their children; mean age 9.96	CDI	5-HTTLPR	maternal depressive symptoms; attentional bias for facial displays of emotion	Yes (additive scale) for maternal depression; 3-way interaction with maternal depression and attentional bias
Gibb et al (2009)	longitudinal; n=100; mothers and their children; mean age 9.97	CDI	5-HTTLPR	maternal expressed emotion; inferential styles for the causes of negative events, and styles for consequences and self-characteristics; maternal depression	Yes (additive scale) for all interactions tested except found trend when excluded youth with history of depression; No (additive scale) with inferential styles
Guo & Tillman (2009)	longitudinal; n=2,286; twins/siblings ages 13–25 (Add Health)	CES-D	DRD2 and DRD4	developmental period (e.g. adolescence versus young adulthood)	No (additive scale)
Haeffel et al (2008)	cross sectional; n=176; court-ordered into juvenile detention; mean age 16.2	BDI and K- SADS	SLC6A3	maternal rejection	No for rs6347 and rs2652511 with K-SADS (multiplicative scale); Yes for rs40184 for BDI (additive scale) and borderline for K-SADS (multiplicative scale); trend with rs40184 and BDI (additive scale)
Hammen et al (2010)	longitudinal; n = 346; mean age 23.7	BDI-II	5-HTTLPR	chronic family stress/disorder; acute stress	No with acute stress (additive scale); Yes with chronic stress, gender, and genotype (additive scale)
Kaufman et al (2004)	cross sectional; n=101; mean age 10 with/without history of maltreatment	MFQ	5-HTTLPR	exposure to maltreatment; social support	Yes (additive scale) with maltreatment; No 2-way interaction with genotype and social support; 3-way interaction with maltreatment genotype, and social support
Kaufman et al (2006)	cross sectional; n=196; mean age 9.3; with/without history of maltreatment	MFQ	5-HTTLPR and BDNF	exposure to maltreatment; social support	Trend (additive scale) with maltreatment; significant 3-way interaction with BDNF, 5-HTTLPR and maltreatment; significant 4-way interaction with BDNF, 5-HTTLPR maltreatment, and social support
Nilsson et al (2009)	longitudinal; n=180; Swedish youth; ages 19–22	DSRS	AP-2B	type of residence; parent marital status; psychosocial risk	Yes (additive scale) for type of residence overall (marginal in boys and significant in girls); Yes for marital status overall and separately for boys and girls; yes for psychosocial risk overall
Sjoberg et al (2006)	longitudinal; n=180; Swedish youth; ages 19–22	DSRS	5-HTTLPR	Type of residence; parent marital status; traumatic conflicts within the family; Psychosocial risk	Yes (additive scale) among boys, for type of residence, separated families, and psychosocial index; among girls, significant interaction for traumatic conflicts and psychosocial index; Trend for other variables
Uddin et al (2010)	longitudinal; n=1,084; twins/siblings ages 12–20; mean age 16 (Add Health)	CES-D	5-HTTLPR	county-level deprivation	Yes (additive scale) for males; not for females
Vaske et al. (2009)	longitudinal; n=2,380; twins/siblings (Add Health)	CES-D	DRD2	exposure to violence	No (additive scale) for full sample or when stratified by sex; Yes among females; effects seen for black but not white females

^*Refers to how data for the overall project were collected.

BDI: Beck Depression Inventory; CDI: Children’s Depression Inventory; CES-D: Center for Epidemiological Studies of Depression Scale; DIS: Diagnostic Interview Schedule; DISC: Diagnostic Interview Schedule for Children; DSRS: Depression Self-Rating Scale (of the DSM-IV); GDAS: Goldberg Depression and Anxiety Scales; KSADS: Schedule for Affective Disorders and Schizophrenia; MFQ: Mood and Feelings Questionnaire; SMFQ: Short Mood and Feelings Questionnaire

Additive scale: refers to when add in their effects (e.g. linear regression)

Multiplicative scale: refers to when risks multiply in their effects (e.g. logistic regression)

Research Design and Study Samples

Research Design

The research design used to test for GxE varied across studies. Slightly more than half (i=11) of the studies included in this review used data from an ongoing longitudinal study (Caspi et al., 2003; Chipman et al., 2007; Chorbov et al., 2007; Gibb, Benas, Grassia, & McGeary, 2009; Gibb, Uhrlass, Grassia, Benas, & McGeary, 2009; Guo & Tillman, 2009; Hammen, Brennan, Keenan-Miller, Hazel, & Najman, 2010; Nilsson, Sjoberg, Leppert, Oreland, & Damberg, 2009; Sjoberg et al., 2006; Uddin et al., 2010; Vaske, Makarios, Boisvert, Beaver, & Wright, 2009). Although in these studies data may have been collected longitudinally, the association between environmental exposure(s) and outcome was not always prospective. For example, out of 11 studies, five assessed some or all exposures included in the test for GxE concurrently with depression (Chipman et al., 2007; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009; Guo & Tillman, 2009; Hammen et al., 2010) and one, which was analyzed as a case-control study (Chorbov et al., 2007) did not have a clear exposure-disease association. Of the remaining nine, five collected data cross-sectionally (Aslund et al., 2009; Benjet, Thompson, & Gotlib, 2010; Chipman, Jorm, Tan, & Easteal, 2010; Eley et al., 2004; Haeffel et al., 2008), and four collected some or all exposure data (i.e. exposure to child maltreatment) prior to depression assessment (Cicchetti, Rogosch, Sturge-Apple, & Toth, 2010; Cicchetti, Rogosch, & Sturge-Apple, 2007; Kaufman et al., 2006; Kaufman et al., 2004), with two not reporting assessing the environmental exposure or outcome repeatedly (Kaufman et al., 2006; Kaufman et al., 2004) and two assessing only the environmental exposure repeatedly (Cicchetti et al., 2010; Cicchetti et al., 2007).

Sample

At least 12 unique samples were examined. Although some drew at least part of their sample from a high risk or clinical population (Cicchetti et al., 2010; Cicchetti et al., 2007; Haeffel et al., 2008; Kaufman et al., 2006; Kaufman et al., 2004), the majority (n=15) studied community-based samples. Studies varied in their sample size, with the smallest sample including 74 (Gibb, Benas et al., 2009) and the largest 2,380 (Vaske et al., 2009).

Race/ethnicity

Studies varied with respect to the amount of racial/ethnic diversity in the sample. Five studies exclusively focused on participants who were White (Aslund et al., 2009; Caspi et al., 2003; Chipman et al., 2007; Chipman et al., 2010; Chorbov et al., 2007), five included a majority of participants who were White (Benjet et al., 2010; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009; Haeffel et al., 2008; Hammen et al., 2010), and the remaining either included more diverse samples (Cicchetti et al., 2010; Cicchetti et al., 2007; Guo & Tillman, 2009; Kaufman et al., 2006; Kaufman et al., 2004; Uddin et al., 2010; Vaske et al., 2009) or did not report this information (Eley et al., 2004; Nilsson et al., 2009; Sjoberg et al., 2006).

Sex

Studies were balanced with respect to sex, though two limited their sample to males (Haeffel et al., 2008) or females (Benjet et al., 2010).

Age

Studies varied in the age of the youth included, with 20% (n=4) focusing on youth ages 12 and under (Benjet et al., 2010; Cicchetti et al., 2010; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009), 4% (n=1) on youth ages 13 to 18 (Aslund et al., 2009), 25% (n=5) on youth ages 19 and above (Caspi et al., 2003; Chipman et al., 2010; Hammen et al., 2010; Nilsson et al., 2009; Sjoberg et al., 2006), and roughly 40% (n=8) on a wide age range (Chipman et al., 2007; Chorbov et al., 2007; Eley et al., 2004; Guo & Tillman, 2009; Kaufman et al., 2006; Kaufman et al., 2004; Uddin et al., 2010; Vaske et al., 2009). Two studies provided only the mean age of their sample, though both referred to studying “adolescents” (Cicchetti et al., 2007; Haeffel et al., 2008). A total of 14 studies included youth in adolescence (between ages 10–24).

Measurement of Depression (i.e. Phenotype)

Outcome

The majority (n=15) assessed depressive symptoms (Benjet et al., 2010; Chipman et al., 2007; Chipman et al., 2010; Cicchetti et al., 2010; Eley et al., 2004; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009; Guo & Tillman, 2009; Hammen et al., 2010; Kaufman et al., 2006; Kaufman et al., 2004; Nilsson et al., 2009; Sjoberg et al., 2006; Uddin et al., 2010; Vaske et al., 2009). Of the remaining five, one assessed a depression diagnosis (Chorbov et al., 2007) and four assessed both a depression diagnosis and depressive symptoms (Aslund et al., 2009; Caspi et al., 2003; Cicchetti et al., 2007; Haeffel et al., 2008).

Symptom Measures

Seven different measures were used to capture depressive symptoms, with the most commonly used measures being a brief or complete version of the Children’s Depression Inventory (CDI; Benjet et al., 2010; Cicchetti et al., 2010; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009), a brief or complete version of the Mood and Feelings Questionnaire (MFQ or SMRQ; Chipman et al., 2007; Eley et al., 2004; Kaufman et al., 2006; Kaufman et al., 2004), and a modified version of the Center for Epidemiological Studies of Depression Scale (CES-D; Guo & Tillman, 2009; Sjoberg et al., 2006; Uddin et al., 2010; Vaske et al., 2009). Three collapsed symptoms into binary high vs. low depressed categories (Chipman et al., 2007; Chipman et al., 2010; Eley et al., 2004).

Diagnostic Measures

Seven studies used diagnostic measures to capture presence or absence of a depressive disorder or depressive symptoms. Three used the Depression Self Rating Scale (DSRS; Aslund et al., 2009; Nilsson et al., 2009; Sjoberg et al., 2006); of these, two used only symptom counts in the analyses (Nilsson et al., 2009; Sjoberg et al., 2006) and one used both symptom counts and a binary indicator (i.e. depressed/not depressed) (Aslund et al., 2009). Two used the Diagnostic Interview Schedule (DIS; Caspi et al., 2003; Cicchetti et al., 2007); in one case, the authors used only symptom counts in the analysis (Cicchetti et al., 2007) and in the other, the authors used both symptom counts and a binary indicator of a depression diagnosis (Caspi et al., 2003). One study also used an adapted version of the Diagnostic Interview for Children and Adolescents (DICA; Chorbov et al., 2007) and another used the Schedule for Affective Disorders and Schizophrenia for School Aged Children (K-SADS; Haeffel et al., 2008); in both, the authors used a binary measure for the outcome (depressed/not depressed).

Data Collection Method

Youth were the primary if not sole respondent. In eight (40%) studies, data on the outcome were collected exclusively through youth self-report (Aslund et al., 2009; Benjet et al., 2010; Chipman et al., 2007; Chipman et al., 2010; Cicchetti et al., 2010; Eley et al., 2004; Hammen et al., 2010; Sjoberg et al., 2006). The remaining 12 collected data by a trained interviewer (Guo & Tillman, 2009; Kaufman et al., 2006; Kaufman et al., 2004; Nilsson et al., 2009; Uddin et al., 2010; Vaske et al., 2009), or clinician (Caspi et al., 2003); in one case, this data was collected through a telephone interview and the background of the interviewer was unclear (Chorbov et al., 2007). Youth also self-reported information about depression for the first assessment and subsequent assessments were interviewer-administered (Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009). Some studies used a combination of methods (Cicchetti et al., 2007; Haeffel et al., 2008). In only one case did researchers use depression data reported by both youth and another informant (Caspi et al., 2003),

Psychometric Properties

30% of studies (n=6) reported information about the psychometric properties of the outcome measure in their sample (Benjet et al., 2010; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009; Guo & Tillman, 2009; Haeffel et al., 2008; Uddin et al., 2010).

Measurement of genotype

Polymorphisms examined

Eleven genetic polymorphisms were investigated. These polymorphisms included variants in genes involved in serotonergic function (SLC6A4, HTR1A, 5HT_2A, 5HT_2C, TPH1), dopaminergic function (DRD2, DRD4, SLC6A3), monoamine catabolism (MAOA), brain-derived neurotropic factor (BDNF), and a transcription factor implicated in the differentiation of neural crest cells (AP-2β). The most commonly examined polymorphism, in 75% of studies (n=15), was the 5-HTTLPR variable number tandem repeat (VNTR), which consists of the s/s, s/l, and l/l genotype. Eight studies (Chipman et al., 2010; Chorbov et al., 2007; Eley et al., 2004; Guo & Tillman, 2009; Haeffel et al., 2008; Kaufman et al., 2006; Nilsson et al., 2009; Vaske et al., 2009) either examined the 5-HTTLPR polymorphism along with other polymorphisms or focused entirely on other genetic markers.

Genotypes analyzed

Of the 15 studies focusing on 5-HTTLPR, 13 captured the biallelic variant (s/s, s/l, or l/l genotype) and three (Chorbov et al., 2007; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009) captured the triallelic version (Hu et al., 2005): (1) L_GS, L_GL_G, SS; (2) L′S: L_AS, L_AL_G;; and (3) L′L: L_AL_A. When analyzing these 5-HTTLPR genotypes, one study combined the s/s and s/l genotype (Cicchetti et al., 2010).

Measurement of environment

The measurement of environment varied considerably across studies, with respect to both the quality of the measures employed and quantity of environments considered.

Types of exposures assessed

With the exception of two studies (Kaufman et al., 2006; Kaufman et al., 2004), which examined exposure to both risky and protective environments, studies focused on exposure to risky environmental factors. The types of risk factors assessed were diverse, ranging from acute or discretely-occurring stressful life events (typically occurring in the past six months), including unemployment, housing, financial, and relationship stressors, to more potentially chronic or ongoing exposures to childhood adversity, such as experiences of child abuse and neglect, low family socioeconomic status, high levels of family stress, maternal depression, risky family structure (i.e. separated parents), and levels of maternal expressed emotion and rejection. Studies also assessed environment as developmental period (Guo & Tillman, 2009), aspects of the child’s cognition, including attentional bias for facial displays of emotion and attributional style (Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009), type of residence (Nilsson et al., 2009; Sjoberg et al., 2006), and characteristics of the child’s broader social ecology, namely county-level socioeconomic deprivation (Uddin et al., 2010)

Method

Questionnaires were the most commonly employed method for obtaining information about environmental exposures, used as the only source of data collection in 35% (n=7) of studies (Aslund et al., 2009; Benjet et al., 2010; Chorbov et al., 2007; Eley et al., 2004; Guo & Tillman, 2009; Haeffel et al., 2008; Vaske et al., 2009). The remaining studies relied on interviews (primarily with youth), review of administrative records (e.g. Census-data; maltreatment data), computer-based methods, or a combination of approaches.

Classification of exposure

There was considerable variation in how studies treated exposure status for the analysis. In 30% of studies (n=6), environment was treated as a binary variable (i.e. exposed vs. unexposed; Aslund et al., 2009; Cicchetti et al., 2010; Eley et al., 2004; Nilsson et al., 2009; Sjoberg et al., 2006; Uddin et al., 2010), in 37.5% (n=8) environment was treated as a ordinal variable (ranging from three to five categories, based on frequency of exposure; Caspi et al., 2003; Chipman et al., 2007; Chipman et al., 2010; Chorbov et al., 2007; Cicchetti et al., 2007; Gibb, Uhrlass et al., 2009; Guo & Tillman, 2009; Hammen et al., 2010), and 25% (n=5) employed continuous measures or scales (Benjet et al., 2010; Gibb, Benas et al., 2009; Haeffel et al., 2008), or a combination of approaches (Kaufman et al., 2006; Kaufman et al., 2004). In one case it was unclear (Vaske et al., 2009).

Psychometric properties

Half (n=10) reported information about the psychometric properties of the environment measure, either with an internal consistency reliability coefficient or a measure of inter-rater agreement (Benjet et al., 2010; Cicchetti et al., 2010; Cicchetti et al., 2007; Gibb, Uhrlass et al., 2009; Haeffel et al., 2008; Hammen et al., 2010; Kaufman et al., 2006; Kaufman et al., 2004; Nilsson et al., 2009; Vaske et al., 2009). The same number used an existing measure that had been psychometrically evaluated in another sample.

Main Study Findings

As described in previous sections, there was considerable heterogeneity in the methods and analyses used across studies to test for GxE. This diversity made it difficult to summarize this research, provide a synthesis of the main findings, and led us to emphasize statistical significance over magnitude of effects. We therefore provide a summary of findings on specific aspects of this research.

GxE Findings

16 studies found at least one significant (p≤0.05) GxE effect. Among the 15 studies investigating 5-HTTLPR, 13 found at least some evidence in support of GxE, with the risk allele or genotype varying depending on the analyses conducted. With respect to the ten other polymorphisms examined, two studies found no evidence of a GxE effect for HTR1A (Chipman et al., 2010), or 5HT_2C (Eley et al., 2004); one found mixed evidence for DAT1 (Haeffel et al., 2008); one found a trend for 5HT_2A and TPH (Eley et al., 2004); one found evidence for MAOA (Cicchetti et al., 2007) and another failed to find significant evidence (Eley et al., 2004); one found evidence for DRD2 (Vaske et al., 2009) and another failed to find significant evidence (Guo & Tillman, 2009); one found evidence for AP-2β (Nilsson et al., 2009); and one found evidence of three-way interaction with environment, BDNF, and 5-HTTLPR (Kaufman et al., 2006).

Main effect of genotype

Eight (40%) found significant main effects or a trend (p<0.08) for some of these specific genes (Aslund et al., 2009; Caspi et al., 2003; Chorbov et al., 2007; Eley et al., 2004; Guo & Tillman, 2009; Kaufman et al., 2004; Uddin et al., 2010; Vaske et al., 2009). Another eight (40%) did not (Benjet et al., 2010; Chipman et al., 2010; Chorbov et al., 2007; Cicchetti et al., 2010; Cicchetti et al., 2007; Hammen et al., 2010; Kaufman et al., 2006; Nilsson et al., 2009), and the remaining four (20%) did not provide sufficient information to make this determination (Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009; Haeffel et al., 2008; Sjoberg et al., 2006).

Main effect of environment

Slightly more than half (n=12) found significant main effects or a trend for at least one of the environmental variables (Aslund et al., 2009; Benjet et al., 2010; Caspi et al., 2003; Chorbov et al., 2007; Cicchetti et al., 2010; Cicchetti et al., 2007; Gibb, Uhrlass et al., 2009; Guo & Tillman, 2009; Kaufman et al., 2004; Nilsson et al., 2009; Uddin et al., 2010; Vaske et al., 2009). Of the remaining eight studies, three did not find significant effects of any environmental measure (Eley et al., 2004; Gibb, Benas et al., 2009; Hammen et al., 2010), four did not provide sufficient evidence to evaluate this association (Chipman et al., 2010; Haeffel et al., 2008; Kaufman et al., 2006; Sjoberg et al., 2006), and one had mixed results based on its inclusion of two different samples (Chipman et al., 2007).

Effect size

Although half of the studies did not provide enough information to evaluate whether the effect for genotype or environment was larger, the studies that did report this information tended to find that the effects for environment were larger.

Effects by developmental period

Among the 14 studies that focused on adolescents (defined as ages 10–24), only two (Chipman et al., 2010; Guo & Tillman, 2009) did not find at least some evidence for GxE. Among 3 studies that focused on children or early adolescents (defined as ages 13 and below), one did not find evidence for GxE (Cicchetti et al., 2010).

Overall, we found more consistent evidence in support of GxE than expected. Out of 20 studies, 16 found at least some evidence to suggest a GxE effect, with 13 out of the 15 studies examining the 5-HTTLPR polymorphism finding GxE. However, within these 13 studies, researchers found mixed results concerning the risk allele or genotype; in some cases, the s allele was protective and in others it was associated with increased risk. In addition, at least two papers provided evidence of heterosis, whereby the heterozygote genotype (e.g. s/l) conferred a protective effect beyond that observed in either homozygote (e.g. s/s or s/l) (Sjoberg et al., 2006; Uddin et al., 2010). The evidence was also mixed concerning whether main effects were detected for both genotype and environment across most of the 11 genes examined. Specifically, less than half of the studies detected a main effect of the genotype and slightly more than half detected a main effect of one or more environmental variables.

The heterogeneity in results was matched by (and likely related to) the heterogeneity in conceptual and methodological approaches used to test for GxE. Studies varied tremendously in the populations sampled, methods used to assess environmental exposures, and ultimately test for GxE. For instance, some studies presented the main effects of either environment or genotype as well as the interaction effects after controlling for sex, age, and race/ethnicity, whereas others did not include any covariates. Moreover, some studies also presented main effects of either environment or genotype in the presence of interaction terms and some did not. This heterogeneity made it difficult for us to provide the kind of synthesis we sought to at the outset. It is also a major limitation, as methodological diversity likely creates discrepancies of GxE effects across studies and prevents a deeper understanding of potential GxE interactions around youth depression. We suspect the use of disparate methods across studies is an artifact of the cross-disciplinary nature of GxE research and reflects differences in both conceptual understanding and methodological conventions adopted across disciplines.

In an effort to build bridges across disciplines and guide GxE researchers in conducting more consistent GxE research in the future, we offer in the following sections suggestions to address some of these challenges (Table 2). These recommendations are centered on: (1) reporting GxE research; (2) testing and reporting GxE effects; (3) conceptualizing, measuring, and analyzing depression; (4) conceptualizing measuring, and analyzing environment; (5) increasing power to test for GxE; and (6) improving the quality of genetic data. Our hope is that these recommendations will enable GxE researchers to conduct future research that is better methodologically aligned so that substantive conclusions can one day be drawn about GxE effects on depression among youth.

Table 2.

Recommendations for Conducting and Reporting GxE Research on Depression in Youth)

Reporting GxE Research	1. Adopt more rigorous reporting standards  (a) Follow STROBE guidelines for reporting observational studies.  (b) Follow STREGA guidelines for reporting genetic studies.
Testing and Reporting GxE Effects
Treatment of the Genotype	2. Make comparisons across all genotype groups (e.g. separately test for the effects of s/s and s/l genotypes).
Reporting Main and Interaction Effects	3. Adopt more thorough reporting standards.  (a) Report all parameters included in the regression model.
	4. Test for main effects with and without the interaction term present depending on hypotheses.  (b) When the interaction term is present, correctly interpret main effects.  (c) Present all data to enable interpretation of interaction effects.
Conducting Formal Tests for Interaction	5. Test for GxE using traditional methods (i.e. cross-product terms).
	6. Incorporate new methods (i.e. test GxE at different values of E).
	7. Report descriptive statistics to enable readers to understand the distribution of genotype and exposures and estimate GxE.
Treatment of Covariates	8. Include all relevant covariates in the analysis, including sex, age, and race/ethnicity.
Reporting Gene-Environment Correlation	9. Test and report gene-environment correlation.
Conceptualizing, Measuring, and Analyzing Depression	10. Use a measure of depressive symptoms as the outcome and note the scale (i.e. additive or multiplicative) used to test GxE.
	11. When diagnostic information is available, use this data to validate results obtained from tests of GxE on symptoms.
	12. Investigate other aspects of depression, including symptom clusters, age at onset vs. course, and comorbidity.
Conceptualizing, Measuring, and Analyzing Environment
Focus on Timing	13. Use more rigorous research designs (i.e. longitudinal, experimental, quasi-experimental).
Examine Frequency and Duration of Exposure	14. Refer to existing reviews on how to conceptualize, measure, and analyze data on life stress and depression.
Poor Measurement and Modeling of Environmental Exposures	15. Use rigorous methods to reliably and validly assess environment and causal associations of “E” to mental health outcomes.
	16. Use more specific methods, rather than indexes or counts, to statistically model aspects of the environment.
Incorporate a Multi-Level Approach	17. Examine how the broader social environment (i.e. schools, neighborhoods) modify genetic effects on depression.
Examine a Wider Array of Proximal Environments	18. Investigate how other proximal environments, such as families and peers, interact with genotypes to influence depression risk.
Increasing Power to Test for GxE	19. Test GxE in larger samples and use more valid and reliable measures of environmental exposures and depressive outcomes.
Improving the Quality of Genetic Data	20. Follow STREGA guidelines for reporting genetic studies and test for Hardy Weinberg Equilibrium (HWE).

Reporting GxE Research

Few studies fully described their methods and analyses, including how tests for interaction were conducted and the nature of the association between exposure and outcome. This lack of specificity is not unique to GxE research. However, the complexity inherent in testing for interactions, combined with the interdisciplinary nature of this work, require that future studies adopt more explicit reporting standards. Future studies should therefore follow both the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) and STREGA guidelines (STrenghtening the Reporting of Genetic Association studies) (Little et al., 2009; http://www.strobe-statement.org).

Testing and Reporting GxE Effects

Treatment of the Genotype

Although nearly all studies examining 5-HTTLPR examined the effect of each genotype (s/s, s/l, and l/l) in the analysis, there was one case where investigators combined across genotype groups (Cicchetti et al., 2010), making it impossible to detect differences that exist between s/s and s/l genotypes. This type of grouping presumes a dominant model of inheritance, whereby having at least one s allele increases risk. Given that the mode of inheritance for these candidate genes is unknown, researchers should, when possible, test an additive model (where genotype is an ordinal variable) and use all data to make comparisons across all genotype groups.

Reporting Main and Interaction Effects

The need for more thorough reporting standards especially applies to conducting tests for interaction. This includes reporting actual regression coefficients for all parameters included in a regression model and explicitly noting what variables were included, including not just covariates. Whether or not main effects are tested and reported without the interaction term will depend on the investigators’ hypotheses. However, when main effects are reported in the presence of an interaction term, they should be correctly interpreted (see online Appendix S1). Given our observations, we caution against interpreting any of the genotype or environment main effects reported in the previous studies, unless the authors explicitly described the parameters included in the regression model. Moreover, since the formal tests for interaction were often only presented for one interpretation, that is, either the environment or the genotype was the modifier, we recommend researchers report specific values for each variable included and provide actual beta coefficients so that readers can interpret interaction effects as they wish. Doing so can help lessen concerns that only the most robust GxE finding was reported.

Conducting Formal Tests of Interaction

Future research should report basic descriptive information that may be suggestive of GxE, including a basic data table (i.e. genotype by exposure by outcome or 3×2) that summarizes the distribution of the environmental exposure and genotype according to the outcome. This recommendation is based on the finding that a considerable number of studies were missing basic univariate analyses on environmental exposures and depression outcomes. This trend in reporting perhaps showcases the overemphasis on statistical significance in current thinking on GxE, rather than on magnitude of effects and meaningful observations that are consistent with theory and existing research (Caspi, Hariri, Holmes, Uher, & Moffitt, 2010). Epidemiologists have advocated for the use of a “counterfactual approach” to examine GxE, which does not emphasize statistical models and instead focuses on the joint contribution of two combined risk factors or causal effects (Greenland & Rothman, 1998; Institute of Medicine Board on Health Sciences Policy, 2006). By presenting more descriptive data, researchers will be better able to discern any patterning that may exist in the GxE effect even without knowing the results of a formal statistical test for such an interaction (Kraft & Hunter, 2009).

Treatment of Covariates

Three covariates – sex, age or developmental period, and race/ethnicity – should be included more explicitly in future GxE research. These factors are important for understanding the etiology of youth depression, the patterning of environmental exposures, and may relate to differences in genotype frequency. Although journal article space often prevents researchers from including this information, attempts should be made wherever possible to test and present the results of analyses according to these three factors (if not in the published article, then as online supplementary material).

Sex

Not all studies controlled for or stratified their results by sex, even though the higher prevalence of depression in females is one of the most consistent findings in psychiatric epidemiology. Without controlling for or stratifying tests of GxE by sex (or conducting 3-way interactions), it is unknown whether GxE effects may manifest differently for boys and girls and results may be biased. Indeed, several studies reviewed here found different GxE effects for males compared to females (see for example Eley et al., 2004; Uddin et al., 2010).

Age or Developmental Period

Many studies did not examine whether GxE effects differed across development. Exploration of the salience of age for GxE is important for several reasons. First, some environmental exposures are age-specific. For instance, the risk of exposure to certain types of maltreatment, such as sexual violence, increases over time and sharply peaks for females around early adolescence (U.S. Department of Health and Human Services & Administration on Children, 2008). Second, neuroscience research suggests that there may be sensitive periods for these environmental exposures, whereby their effects on brain structures and functioning are more pronounced during one period in development than another (Gunnar & Quevedo, 2007; Lupien, McEwen, Gunnar, & Heim, 2009). Third, the risk of depression increases with age (Kessler et al., 2005; Rudolph, 2009). Thus, without accounting for or explicitly exploring how age or developmental period influences GxE effects, research may be biased and the field misses an opportunity to better understand how GxE effects may differ across development.

Race/ethnicity

Some studies did not control for race/ethnicity. This issue relates to population stratification, a concept from population genetics that refers to the presence of different allele frequencies among different sub-populations or ancestral groups. In cases where samples are drawn from different population groups, the failure to control for race/ethnicity may lead to biased results. Ideally, research should follow STREGA guidelines (Little et al., 2009) and describe any methods used to assess or address population stratification. At a minimum, studies should control for self-reported race/ethnicity and conduct sensitivity analyses to test whether observed GxE effects vary by race.

Reporting of Gene-Environment Correlation

Given that less than half of the studies (n=9) reported the results of tests for gene-environment correlation (rGE; Caspi et al., 2003; Chipman et al., 2010; Chorbov et al., 2007; Cicchetti et al., 2007; Gibb, Benas et al., 2009; Gibb, Uhrlass et al., 2009; Hammen et al., 2010; Kaufman et al., 2006; Kaufman et al., 2004), we recommend future research test and report whether rGE is present. rGE refers to the idea that individuals select, modify, and construct their environment (Kendler & Baker, 2007). For GxE, the concern of rGE arises in making causal inferences about the effect of environmental exposures on depression; that is, are genes and environments independent or did genetic factors play some role in determining which environments an individual was exposed to? Tests for rGE can be conducted through simple tests of association between environmental exposures and genotype. However, such tests are limited by the genotypes measured; the absence of association between tested genotypes and the environment does not rule out rGE in general but only reduces the likelihood that it is of particular concern for the specific GxE tested.

If rGE is observed, researchers can conduct stratified analyses, where the risk of depression is estimated separately for each genetic subgroup, rather than using a combined group test of statistical interaction, which presumes that genetic and environmental risks are independent. Otherwise, GxE results may be biased and should be interpreted cautiously (Jaffee & Price, 2007). Prospective or cohort designs, where environmental exposures precede depression onset, are also least likely to be affected by rGE. In contrast, retrospective designs may give rise to rGE, as recall of past events may be influenced by factors under genetic influence (i.e. mood, personality) (Jaffee & Price, 2007).

Conceptualizing, Measuring, and Analyzing Depression

Future research should use a measure of depressive symptoms (i.e. continuous) rather than diagnosis (i.e. binary) as the outcome, as has already been done in many GxE studies on youth. Researchers should also explicitly note the scale (i.e. additive or multiplicative) used to detect GxE effects. These recommendations are based on the finding that the way the outcome measure is scaled and whether the GxE effect is tested on the additive (i.e. risks add in their effect, such as linear regression) or multiplicative scale (e.g. risks multiply in their effects, such as logistic regression) influences whether a GxE effect is observed (Greenland & Rothman, 1998; Institute of Medicine Board on Health Sciences Policy, 2006). In fact, changing the scale of the outcome may create interactions that may not have previously existed or eliminate interactions that were once present (Kraft & Hunter, 2009). To that end, analyses based on binary outcomes have been shown through simulations to incorrectly detect a GxE effect when none existed (Eaves, 2006), thus raising concerns about the validity of results based on diagnoses. Using dimensional (rather than categorical) approaches is also warranted given the current emphasis in genetics on understanding how all levels of liability shape complex quantitative traits (i.e. symptoms), rather than on how extremes in liability shape qualitative traits (i.e. disorders) (Plomin, Haworth, & Davis, 2009). However, when available, researchers could validate their test of GxE based on symptoms with a test of GxE using diagnoses. Some authors have argued this approach can help validate results obtained from a previous test of GxE, decreasing the chance of spurious GxE effects (Moffitt et al., 2006). Moreover, future studies can also test for GxE using a broader conceptualization of the depression phenotype (Cross-Disorder Phenotype Group of the Psychiatric GWAS Consortium et al., 2009).

Conceptualizing, Measuring, and Analyzing Environment

Focus on Timing

There was wide variation in the timing of the exposures assessed across studies, with respect to the temporal relationship between the exposure and outcome (i.e. prospective vs. cross-sectional), ability to make causal inferences (i.e. lag-time between onset and development of depression) and developmental period or stage in the life course considered (i.e. early childhood, childhood, adolescence). For instance, despite being embedded in an ongoing longitudinal study, the exposure and outcome were often measured simultaneously. As noted, little attention was also paid towards understanding the timing of exposures in relation to development. Thus, GxE research would benefit from incorporating more rigorous research designs, including experimental and quasi-experimental approaches.

Examining Frequency and Duration of Exposure

Studies differed in the frequency and duration of each exposure included (e.g. acute or discrete vs. chronic or cumulative occurrence), with some studies examining one-time events occurring close in time to the assessment of depression and others investigating repeated exposures occurring over longer periods of time. These differences not only create challenges in making comparisons, but also prevent a deeper understanding of how the frequency, duration, and persistence of the exposure influenced detection of a significant GxE interaction (Moffitt et al., 2006; Uher & McGuffin, 2008).

This variation also highlights the different theoretical traditions used to examine the association between life stress and depression (see recent reviews by Cohen, Kessler, & Gordon, 1995; Hammen, 2005; Monroe, 2008; Monroe & Reid, 2008). These approaches vary in the characteristics of the stressor examined and the psychological or biological explanations used to explain how the stressor exerts its effect. For example, in one tradition, researchers argue that depression results from exposure to acute or major, threatening, and recent life events (Brown & Harris, 1978, 1989). This approach parallels the notion of “diathesis-stress,” whereby a genetic liability (diathesis) interacts with a negative life experience (stress) to cause depression, with genes exacerbating or buffering the effects of stress (Monroe & Simons, 1991). In another tradition, researchers focus on ongoing and chronic exposures to stress or adversity, such as poverty, child maltreatment, and social deprivation, based on the idea that these stressful conditions can accumulate over time, resulting in an increased “allostatic load” or wear and tear on the body (McEwen & Seeman, 1999). Several authors have outlined the key issues to consider in conceptualizing, measuring, and analyzing data about the role of life stress in depression (Cohen et al., 1995; Hammen, 2005; Monroe, 2008). We urge researchers to consult these sources to more carefully capture characteristics of environmental stressors.

Poor Measurement and Modeling of Environmental Exposures

Studies varied considerably with respect to the quality of the measures and approaches used to capture environment or some aspect of stress. Some used reliable and valid scales, while others used single items or measures designed by their research team. Some also used self-reported measures, while others used interview-based approaches. This is an important distinction because interview-based approaches are more reliable than self-reported checklists (Monroe, 2008). These observations leave the impression that there is considerable “noise” in the assessment of environment, underscoring the need to incorporate more rigorous methods to reliably and validly assess environmental exposures.

We also found differences in how each study treated environmental exposures in the analyses. For example, some studies examined counts depicting the total number of events experienced, while others used binary indicators (exposed/unexposed) or scales to capture more complex phenomenon, not limited to stress such as maternal rejection. Greater specificity in capturing features of the environment is needed in order to better understand how specific exposures, in what combination, and to what degree, are associated with depression.

Moreover, few studies discussed whether the environmental exposures examined were selected because they had environmentally-mediated effects on the outcome, in other words that depressive symptoms in youth were caused by environmental features and that this association was not due to genetic factors. This is important because an association between an environment and an outcome may arise due to a third variable, namely common genetic liability. Evidence for environmental-mediation (i.e. that environments are associated with depression above and beyond genetic factors) is available for some “E” variables used in GxE studies (e.g. physical abuse; Jaffee, Caspi, Moffitt, Polo-Tomas et al., 2004; Jaffee, Caspi, Moffitt, & Taylor, 2004). However, providing robust evidence of environmental-mediation for many “E” variables (e.g. childhood socioeconomic status) is methodologically challenging (Purcell & Koenen, 2005; Rutter, Pickles, Murray, & Eaves, 2001; Turkheimer, D’Onofrio, Maes, & Eaves, 2005; Turkheimer & Waldron, 2000). Requiring that all “E” variables in GxE studies have demonstrated environmentally-mediated effects might unnecessarily limit the exposures that could be considered in this research. For this reason, we argue that GxE studies would be strengthened if the “E” has been shown to be environmentally-mediated, however, we do not argue that environmental mediation is a requirement for “E” to be included in a GxE study.

Incorporate a Multi-Level Approach

Although the social and physical contexts in which youth develop – their family, school, and neighborhood environments – shape their health and risk of depression (Bronfenbrenner, 1979; Gershoff & Aber, 2006; Goodman, Huang, Wade, & Kahn, 2003; Leventhal & Brooks-Gunn, 2000; Mair, Diez Roux, & Galea, 2008), nearly every study reviewed examined proximal or individual-level factors, ignoring the distal social conditions surrounding youth and thus the multi-level nature of disease causation (Diez Roux, 1998; Subramanian, Jones, & Duncan, 2003). Although proximal factors often confer larger risks for disease when compared to distal factors, the ubiquity of exposure to macro-social environmental variables suggests that their role in determining the population distribution of youth depression may be substantial.

Future GxE research would benefit from incorporating a multi-level approach to conceptualizing and measuring environments for two reasons. First, the social contexts surrounding youth play a pivotal role in determining the resources they can draw from to support their development. For instance, neighborhood and school environments can provide access to assets that affect both whether a child will engage in a behavior that increases their risk of developing depression (e.g. substance abuse) and how they respond to stressors (e.g. positive peer social networks; Berkman & Kawachi, 2000). Moreover, the fact that families are also embedded in these social contexts suggests that they can shape parents’ capacity to raise their children (Tendulkar, Buka, Dunn, Subramanian, & Koenen, 2010). Second, distal environments appear to be important for understanding GxE effects. For instance, a study included in this review found that after controlling for individual-level risks, the effect 5HTTLPR genotype on risk of depression was modified by county-level deprivation (Uddin et al., 2010). Similar GxE effects have been found for urban/rural residence on depression in adults (Jokela, Lehtimaki, & Keltikangas-Jarvinen, 2007; Xu et al., 2009).

Researchers can incorporate the social environment in future GxE studies by drawing from ecological theories of child development (Bronfenbrenner, 1979) and using multi-level modeling techniques (Raudenbush & Bryk, 2002; Subramanian et al., 2003). Measurement of the social environment often combines administrative, observational, and self-report data. Researchers can use “geocoding” procedures to link youth’s street address to data from publicly available sources (i.e. U.S. Census) that provide cost-effective information about the contexts (i.e. neighborhood, county, or state) surrounding youth. Observational data include systematic observations of social and physical markers of disorder (e.g. graffiti or public intoxication in a neighborhood; Sampson & Raudenbush, 1999). Self-report measures can also be used and when aggregated to the school- or neighborhood level, can capture features of these settings (Dunn et al., submitted; Shinn, 1990); however, brief self-report measures of these environments are lacking. Researchers will face a trade-off between sample size and measurement, with larger samples making it potentially more challenging to collect extensive measures of the environment (as well as phenotype). Regardless of measurement approach, attention should be paid towards measuring the social contexts most influential during a specific period of development (i.e. home and school with younger children; neighborhoods for adolescents).

Examine a Wider Array of Proximal Environments

Future research should focus on a broader array of proximal environments. We found limited attention to protective factors and positive environments, as every study, with the exception of two (Kaufman et al., 2006; Kaufman et al., 2004), examined negative life events, childhood adversities, or some other stressor. Moreover, other types of proximal environments that could be more deeply explored include family and peer-relationships and interventions that may play a role in influencing reactions to stress. Ideally, future research should focus on proximal environments that are unlikely influenced by genes. When selecting environments, researchers should pay attention to the salience of age or developmental period, as certain environments, such as peer relationships, may be more salient at different ages.

Increasing Power to Test for GxE

The results of this review underscore the need for larger and more diverse samples to test for GxE in depression among youth. The studies reviewed here may be too small to detect GxE effects, as the largest (n=2,380) still fell short of the estimated number needed to adequately test for GxE (Munafo et al., 2009; Munafo, Durrant, Lewis, & Flint, 2010). Existing power estimates to detect GxE are also calculated for optimal conditions and do not take account of other factors that influence power; sample size requirements will vary depending on allele frequency, the magnitude of the GxE interaction, and the strength of the association between exposure and outcome, which is influenced by the reliability and validity of the environment and depression measures (Wong, Day, Luan, Chan, & Wareham, 2003). Future research should try to test for GxE in larger samples, including ongoing cohort studies and population-based epidemiological studies, and by pooling across existing samples. It is also important for power that more valid and reliable measures of environmental exposures and depressive outcomes are used. Where tradeoffs need to be made, gaining more precise measures of environment may be a more worthwhile endeavor than trying to acquire a larger sample (Institute of Medicine Board on Health Sciences Policy, 2006).

Improving the Quality of the Genetic Data

Many studies did not report information about the quality of their genetic data. This is problematic as genotyping errors, especially if they are large in magnitude, can significantly bias the results (Little et al., 2009). Therefore, we recommend future GxE research follow STREGA guidelines and provide more specific information about the collection, storage, and analysis of DNA. Moreover, not all studies tested for Hardy-Weinberg equilibrium (HWE), which is whether genotype frequencies are consistent with random mating in the studied population. Some also tested for HWE for the entire sample and not in specific subgroups (i.e. stratified by race/ethnicity). Testing for HWE and reporting these results is important as it can provide information about deviation from its assumptions (Hartl & Clark, 2007).

Conclusion

We conducted this literature review in an attempt to understand the state of the science on GxE research in depression among children and adolescents and focus on the methodological approaches used across studies, as these features have not been previously described, including in recent meta analyses. We also sought to discern the level of heterogeneity in GxE findings that existed across studies, assuming that mixed results would be widespread, as others had concluded (Munafo et al., 2009; Uher & McGuffin, 2008, 2010). The results of this systematic review suggest that the findings of GxE are perhaps not as mixed among youth as believed: most studies reviewed did find some evidence to support GxE, though this finding may reflect publication bias. However, a more salient issue that this review illustrates is the heterogeneity that exists in the methods and conceptual approaches used to conduct studies of GxE. These variations made it difficult to interpret and summarize findings and understand the nature of GxE effects during childhood and adolescence. Research completed by interdisciplinary teams, where there are equally high levels of expertise on genetic and environmental factors, will be important for reducing this heterogeneity and adequately capturing the joint contribution of genetic and environmental factors. We hope the recommendations reported in Table 2 will provide a useful framework to guide future studies so that more research of better quality can be conducted, compared and replicated, and empirical knowledge of GxE effects among youth can advance in ways that generate new knowledge for prevention and intervention.

Supplementary Material

Supp App S1. Appendix S1.

Interpreting main effects in the presence of an interaction (Word document).

Key Findings.

• There has been a tremendous growth in research on gene–environment interaction (GxE) within the last decade.

• Most published studies on GxE in depression among children and adolescents find evidence for GxE.

• However, existing research varies tremendously in the methods and analyses used to test for GxE; this variation prevents a synthesis of the current state of knowledge of the joint contribution of genetic and environmental influences.

• The recommendations provided in this paper will enable more methodologically and conceptually consistent research on GxE in the future.