Early Life Stress, MAOA, and Gene-Environment Interactions Predict Behavioral Disinhibition in Children

Abstract

Several, but not all, studies have shown that the monoamine oxidase A functional promoter polymorphism (MAOA-LPR) interacts with childhood adversity to predict adolescent and adult antisocial behavior. However, it is not known whether MAOA-LPR interacts with early life (pre-birth – 3 years) stressors to influence behavior in pre-pubertal children.

The Avon Longitudinal Study of Parents and Children, U.K., is a community-representative cohort study of children followed from pre-birth onwards. The impact of family adversity from pre-birth to age 3 years and stressful life events from 6 months to 7 years on behavioral disinhibition was determined in 7500 girls and boys. Behavioral disinhibition measures were: mother-reported hyperactivity and conduct disturbances (Strengths and Difficulties Questionnaire) at ages 4 and 7 years.

In both sexes, exposure to family adversity and stressful life events in the first three years of life predicted behavioral disinhibition at age 4, persisting until age 7. In girls, MAOA-LPR interacted with stressful life events experienced from 6 months to 3 ½ years to influence hyperactivity at ages 4 and 7. In boys, the interaction of MAOA-LPR with stressful life events between 1 ½ and 2 ½ years predicted hyperactivity at age 7 years. The low activity MAOA-LPR variant was associated with increased hyperactivity in girls and boys exposed to high stress. In contrast, there was no MAOA-LPR interaction with family adversity.

In a general population sample of pre-pubertal children, exposure to common stressors from pre-birth to 3 years predicted behavioral disinhibition, and MAOA-LPR - stressful life event interactions specifically predicted hyperactivity.

INTRODUCTION

Childhood adversity is acknowledged to be a predictor of adolescent and adult antisocial behavior and substance abuse (Dodge et al, 1990; Verona and Sachs-Ericsson, 2005). Animal studies of maternal deprivation and abuse suggest that the effects of adversity are likely to be greatest if they are experienced early in life and are also likely to be persistent (Liu et al, 1997; Maestripieri et al, 2005). Likewise, studies in children have shown that maltreatment prior to age five has the greatest impact (Keiley et al, 2001).

Genetic vulnerability is also an important influence on the development of behavioral disinhibition (Jaffe et al, 2005). One gene that has been implicated is monoamine oxidase A (MAOA) (Xp11.23) that encodes an enzyme that plays an important role in the degradation of central nervous system (CNS) serotonin and norepinephrine. Variation in serotonergic activity is implicated in impulsivity and anxiety. MAOA deficient mice have increased CNS serotonin and norepinephrine levels; pups show increased fearfulness and adult males show increased aggression (Cases et al, 1995). In healthy men, MAOA activity in cortical regions correlates inversely with measures of aggression (Alia-Klein et al, 2008). The MAOA gene has a 30 base pair repeat in the promoter region (MAOA-LPR) that affects transcriptional efficiency (Sabol et al, 1998). In healthy humans, MAOA-LPR genotype predicts limbic circuitry variation for emotional regulation, including inhibitory control, and also brain response differences in tests for impulsivity (Meyer-Lindenberg et al, 2006; Passamonti et al, 2006).

Interactions between genes and environment (G × E) are likely to influence the risk – resilience balance for pathological behavior (Moffitt et al, 2005). Earlier longitudinal G × E studies have shown that severe maltreatment (Caspi et al, 2002; Kim-Cohen et al, 2006; Widom and Brzustowicz, 2006) and family adversity (Foley et al, 2004) experienced by young children interact with the MAOA-LPR low activity variant to predict childhood conduct disorder and adult antisocial behavior. A similar G × E interaction has been demonstrated in rhesus macaque monkeys (Newman et al, 2005). However studies using retrospective childhood adversity questionnaires have had mixed results (Ducci et al, 2008; Haberstick et al, 2005; Prichard et al, 2008; Prom-Wormley et al, 2009; Sjoberg et al, 2007; Young et al, 2006).

Our study was based on the Avon Longitudinal Study of Parents and Children (ALSPAC), a general population sample. Exposure to severe adversity is uncommon in ALSPAC (Sidebotham et al, 2006). We hypothesized that MAOA-LPR would interact with widely experienced early life stressors to influence behavior in pre-pubertal children. Firstly, we investigated the effects of both family adversity and stressful life events experienced at several time points from pre-birth to age seven on behavioral disinhibition: hyperactivity and conduct disturbances, assessed at ages four and seven years. Secondly, we investigated the main effects of MAOA-LPR genotype and the interactive effects of MAOA-LPR and stressors on behavioral disinhibition in boys and girls.

METHODS AND MATERIALS

Study population

The study population is an urban / rural mix with little outward migration located in Southwest England (Golding et al, 2001). 14,541 pregnant women resident in the county of Avon with a due date between April 1991 and December 1992 were enrolled in the study, representing approximately 85% of all eligible participants. ALSPAC is therefore a community-representative sample. Only Caucasian children (95% of the sample) were included in the current study in order to minimize the effects of population stratification. All children were unrelated. A total of 4034 boys and 3976 girls had data on hyperactivity at 4 years which reduced to 3182 boys and 2947 girls who also had MAOA-LPR genotypes. Approval for the study was obtained from the ALSPAC Law and Ethics Committee and Local Research Ethics Committees.

Questionnaires

A family adversity questionnaire was completed by mothers at three time points: during pregnancy and when their child was 0 – 1 year and 2 – 3 years old. Response rates ranged over time from 92% to 78%. Mothers completed questionnaires about their children's behavior at ages 4 and 7 years and exposure to stress at: 1 ½, 2 ½, 3 ½ and 7 years. Response rates ranged from 85% at 1 ½ years to 74% at 7 years.

Study dropouts

As expected, children who dropped out from the study were more socially disadvantaged and had experienced greater adversity (see Supplementary Table 1). However there is increasing evidence to suggest that such selective dropout does not invalidate the associations found in epidemiological research although it may bias estimates of prevalence (Wolke et al, in press).

Behavioral outcomes

Mother-assessed hyperactivity and conduct disturbances were derived from the Strengths and Difficulties Questionnaire (SDQ), a widely used screening instrument that has been validated for use in the community (Goodman, 1997; Goodman et al, 2003). Two of the five SDQ factors assess hyperactive and antisocial behavior. Each factor has five questions concerning behavior over the last six months. The hyperactivity domains are: restless / overactive; constantly fidgeting; easily distracted; not thinking before acting; poor attention span. The conduct domains are: frequent temper tantrums; disobedience; fighting / bullying; telling lies / cheating; stealing.

The SDQ hyperactivity scores were approximately normally distributed (Supplementary Figure 1). In contrast, conduct scores were very low and very few children had high scores (Supplementary Figure 2). Therefore a persistent conduct disturbances outcome was derived from the maternal assessments at 4 and 7 years based upon a child being in the top decile at both times (approximately 4% of the sample).

Measures of adversity

(A) Family adversity

A family adversity score was derived from the following elements: maternal psychopathology, inter-parental cruelty, parental substance misuse and at risk/taken into care (all dichotomous variables) and social deprivation (continuous variable). The total family adversity score ranged from 0 – 15 and was derived from social deprivation (max = 11, Supplementary Table 2) and a maximum score of 1 each from maternal psychopathology, inter-parental cruelty, parent substance misuse and at risk/taken into care (see Table 1).

Table 1.

Family Adversity and Stressful Life Events at Different Time-Points

	BOYS					GIRLS
	N	Mean	S.D.	% outcome	Max	N	Mean	S.D.	% outcome	Max
Assessment: pregnancy
Inter-parental cruelty	3549			4.7	1	3278			4.2	1
At risk / taken into care	3884			1.6	1	3627			1.3	1
Mat psychopathology	3791			23.6	1	3556			23.6	1
Parent substance misuse	3878			5.0	1	3622			4.8	1
Social deprivation	3855	0.6	0.9		8 / 11	3599	0.6	0.9		6 / 11
Assessment: 0 – 1 y
Inter-parental cruelty	3762			13.1	1	3537			12.6	1
At risk / taken into care	3884			0.8	1	3627			0.8	1
Mat psychopathology	3764			20.0	1	3539			19.8	1
Parent substance misuse	3763			12.0	1	3538			13.0	1
Social deprivation	3704	0.9	1.1		7 / 11	3480	0.9	1.1		8 / 11
Assessment: 2 – 3 y
Inter-parental cruelty	3420			13.2	1	3207			13.6	1
At risk / taken into care	3884			1.2	1	3627			0.8	1
Mat psychopathology	3423			14.4	1	3210			13.5	1
Parent substance misuse	3425			13.5	1	3212			14.3	1
Social deprivation	3421	0.7	1.0		6 / 11	3209	0.7	1.0		6 / 11
Stressful Life Events
1 ½ yrs	3506	1.7	2.0		28 / 60	3277	1.5	1.9		11 / 60
2 ½ yrs	3347	2.4	2.3		28 / 60	3095	2.4	2.2		14 / 60
3 ½ yrs	3310	2.7	2.3		14 / 60	3042	2.6	2.3		14 / 60
7 yrs	2969	2.8	3.1		27 / 72	2750	2.9	3.0		18/ 72

N = total number of children assessed for a particular measure of adversity Mean (SD): social deprivation = mean (SD) score; stressful life events = mean (SD) total life events score calculated by summing the perceived impact scores reported by mothers for each life event.

% outcome = % with outcome present for dichotomous variables Max = highest score achieved in this study / maximum possible score for social deprivation and for impact of stressful life events on the child.

For the dichotomous variables, the maximum possible score is 1.

The distributions of stressful life events for boys tend to have longer tails than girls although the percentages in the upper tails are similar.

A positive score for maternal psychopathology was assigned if at least one of four criteria occurred: (a) suicide attempt, (b) Crown-Crisp Experiential Index (CCEI) (Crown and Crisp, 1979) score > 10 for anxiety, (c) CCEI score > 9 for depression, (d) Edinburgh Postnatal Depression Scale (EPDS) (Cox et al, 1987) score > 12. The CCEI cut-offs were chosen to reflect the lowest 10% of derived scores. The EPDS has been validated in community samples (Carothers and Murray, 1990). A score of > 12 is generally taken to indicate the likelihood of clinical depression, but not its severity (Warner et al, 1996). The four criteria comprising maternal psychopathology had high inter-item correlations (e.g. Cronbach alpha = 0.86 for the measure during pregnancy).

Inter-parental cruelty is a measure of emotional or physical cruelty by the partner on the mother. The parental substance misuse scale was derived from measures of alcohol consumption, hard drug use and reported diagnoses of alcoholism. The major family problems scale reflected social services referral or foster care, either institutional or with relatives. Any occurrence gave a positive score. The percentage of children with this adversity was low at all time points, ranging from 0.8% to 1.7%.

A continuous measure of social deprivation was derived from the following items: (a) young motherhood; (b) family size; (c) housing adequacy and quality; (d) parental education; (e) financial difficulties; (f) partner relationship (other than cruelty); (g) criminality (see Supplementary Table 2 for full details).

(B) Stressful life events

Mothers were asked when their children were aged 1 ½, 2 ½, 3 ½ and 7 years whether their child had experienced any of 15-18 different stressful life events in the previous 12 to 18 months and rated the impact of the event on the child on a four point scale. The items included in this questionnaire were derived from previous studies (Barnett et al, 1983; Brown and Harris, 1978). Total life event scores were calculated by summing the perceived impact score reported by mothers for each life event. The perceived impact score ranged from zero (no experience of event) to 4 (very upset by event). Each of the perceived life event impact scores was weighted according to the average impact recorded by mothers in order to reduce bias in the rating of impact resulting from maternal mood. Total scores were further adjusted for missing data by pro-rating. Life event scores increased slightly as children got older, nevertheless the mean scores were low (Table 1). Few of the children had experienced severe life events. At each of the four time points, the number of children experiencing the following stressful life events was: sexual abuse: 4 - 13; taken into care: 12 – 23; physically hurt: 251 - 384. The most common stressful life events were: pet died (588 - 2331), new sibling (545 – 2127), starting creche / nursery / school (863 – 5997).

Correlation between family adversity and stressful life events

The correlation between family adversity and stressful life events was modest (r = 0.09 – 0.16) although significant (p < 0.001).

Genotyping

DNA was extracted from peripheral blood. Within the MAOA promoter there is a 30-bp repeated sequence present in 2, 3, 3.5, 4, or 5 copies. This polymorphism was amplified from 25ng genomic DNA using the following primer sequences: Forward 5′ CCC AGG CTG CTC CAG AAA C 3′ and reverse 5′ ACT CAG AAC GGA TGC TCC ATT CG 3′. Because of the high GC content around the repeat region, amplifications were performed using Invitrogen's PlatinumTaq and PCRX Enhancer System kits, according to the manufacturer's protocol (Invitrogen, Carlsbad, California), in a total reaction volume of 25 μl. Amplifications were performed on a Perkin Elmer 9700 thermocycler (Applied Biosystems, Foster City, CA) with one cycle at 96°C for 10 min followed by 30 cycles of 94°C for 15 sec, 60°C for 15 sec, 72°C for 30 sec, and a final 3-minute extension at 72°C. The forward primer was labeled with the fluorescent dye 6-FAM, and amplicons were visualized on an ABI 3100 capillary sequencer. Allele sizes (allele 2, 233 bp; allele 3, 263 bp; allele 3.5, 278 bp; allele 4, 293 bp; allele 5, 323 bp) were determined using Genotyper 2.5 (Applied Biosystems, Foster City, CA). The genotype completion rate was 96%. Genotyping discrepancy was determined by replicating 132 of the samples. The overall error rate was 5%.

Genotypes were available for 3884 boys and 3627 girls. The allele frequencies of the 30-bp repeated sequence present in 2 (0.003), 3 (0.347), 3.5 (0.019), 4 (0.616), or 5 (0.014) copies were the same in both sexes. Enzyme expression is known to be 2 – 10 times higher for the 3.5 and 4 repeats than for the 3 repeat (Sabol et al, 1998). Therefore the 3 repeat alleles were classified as low activity (L) whereas the 3.5 and 4 repeat alleles were classified as high activity (H). The 2 and 5 repeat alleles were excluded because their activity levels are not yet clear. The genotype frequencies for boys were: L: 0.338 and H: 0.662; and for girls were: LL: 0.121, LH: 0.452, HH: 0.427. All genotype frequencies conformed to Hardy-Weinberg equilibrium.

Statistical Analyses

Data were analyzed using multivariable linear / logistic regression. Logistic regression was used for the dichotomous persistent conduct problem outcome. Linear regression was used for the continuous hyperactivity scores since this method produced effect sizes which were easier to interpret and potentially more sensitive to genetic and environmental influences. We conducted stepwise regression analyses of adversity on conduct disturbances and hyperactivity in order to identify the most important factors associated with each outcome. An arbitrary 5% significance level was used to select factors. The choice of measures and time-points that were used in subsequent analyses were therefore determined by the associations found in the stepwise regression analyses. Since this is an exploratory study, no corrections were made for multiple testing. Data was analyzed separately for boys and girls since MAOA is an X-linked gene.

The stressful life event and family adversity scores were standardized to allow a more valid comparison of effect sizes between these variables and with dichotomous variables such as maternal psychopathology. Therefore regression coefficients have been provided for maternal psychopathology but standard regression coefficients (independent variables standardized) have been provided for stressful life event and family adversity scores (Tables 2 and 3). Beta values are presented with 95% confidence intervals. Environmental effects are reported either for a one SD increase in continuous measures or as the additional impact in the presence of adversity (compared to absence) for dichotomous measures.

Table 2.

GIRLS: Effects of Adversity from Pre-Birth to Age 7 Years on Behavior Assessed at Ages 4 to 7 Years

	Hyperactivity Age 4 yrs		Hyperactivity Age 7 yrs		Conduct Disturbances Age 4 – 7 yrs
	β [95% CI]	P Value	β [95% CI]	P Value	OR [95% CI]	P Value
Family Adversity Assessment
Pre-Birth:
Inter-Parental Cruelty
At Risk / Taken into Care
Maternal Psychopathology	0.34 [0.15, 0.52]	<0.001	0.36 [0.16, 0.55]	<0.001
Social Deprivation	0.16 [0.07, 0.26]	0.001	0.22 [0.14, 0.31]	<0.001
Total Family Adversity	0.22 [0.12, 0.32]	<0.001	0.24 [0.14, 0.34]	<0.001
0 – 1 Years:
Inter-Parental Cruelty			0.19 [−0.06, 0.44]	0.10
At Risk / Taken into Care					2.85 [0.82, 9.93]	0.10
Maternal Psychopathology	0.25 [0.05, 0.45]	0.02	0.57 [0.37, 0.77]	<0.001	2.67 [1.79, 3.98]	<0.001
Social Deprivation	0.13 [0.03, 0.22]	0.01			1.38 [1.17, 1.63]	<0.001
Total Family Adversity	0.11 [0.01, 0.21]	0.04	0.25 [0.15, 0.34]	<0.001	1.70 [1.46, 1.97]	<0.001
2 - 3 Years:
Inter-Parental Cruelty			0.48 [0.26, 0.71]	<0.001
At Risk / Taken into Care
Maternal Psychopathology	0.34 [0.12, 0.56]	0.003
Social Deprivation
Total Family Adversity	0.13 [0.04, 0.23]	0.007	0.10 [−0.01, 0.20]	0.08
Stressful Life Events Assessment
1.5 years
2.5 years			0.06 [−0.02, 0.14]	0.10	1.19 [0.99, 1.43]	0.07
3.5 years	0.06 [−0.02, 0.13]	0.10	0.09 [0.01, 0.16]	0.02	1.30 [1.09, 1.54]	0.003
7 years			0.06 [−0.02, 0.14]	0.10	1.16 [0.98, 1.37]	0.08

Multivariable Linear/Logistic regression analyses showing independent or marginal effect after adjusting for other factors remaining in the model. Items in bold were significant (p<0.05) and were included in the final model. Other items shown were close to significance (p≤0.1). Odds ratios (OR) and beta (β) values are given with 95% confidence intervals.

Stressful life events at 1.5, 2.5 and 3.5 years are for last 12 months. Stressful life events at 7 years are since age 5.5 years.

Social deprivation, total family adversity and stressful life events scores were standardized. Consequently their effect sizes are per SD.

A persistent conduct disturbance outcome was derived from maternal assessments at 4 years and 7 years based upon a child being in the top decile at both times (approximately 4% of the sample).

The N's for each group were: hyperactivity at age 4: 3976; at age 7: 3340. Conduct: 3316.

Parent substance misuse was not included in the table since there were no significant effects.

Table 3.

BOYS: Effects of Adversity from Pre-Birth to Age 7 Years on Behavior Assessed at Ages 4 to 7 Years

	Hyperactivity Age 4 yrs		Hyperactivity Age 7 yrs		Conduct Disturbances Age 4 – 7 yrs
	β [95% CI]	P Value	β [95% CI]	P Value	OR [95% CI]	P Value
Family Adversity Assessment
Pre-Birth:
Inter-Parental Cruelty			−0.50 [−0.91, −0.08]	0.02
At Risk / Taken into Care
Maternal Psychopathology			0.30 [0.08, 0.53]	0.007
Social Deprivation	0.21 [0.11, 0.31]	<0.001	0.34 [0.24, 0.44]	<0.001	1.30 [1.12, 1.52]	<0.001
Total Family Adversity	0.17 [0.07, 0.28]	0.001	0.27 [0.15, 0.38]	<0.001
0 – 1 Years:
Inter-Parental Cruelty			0.35 [0.08, 0.62]	0.01	1.66 [1.12, 2.47]	0.01
At Risk / Taken into Care			1.11 [0.14, 2.09]	0.03
Maternal Psychopathology	0.55 [0.35, 0.75]	<0.001	0.53 [0.30, 0.76]	<0.001	1.76 [1.19, 2.60]	0.005
Social Deprivation	0.21 [0.11, 0.31]	<0.001	0.09 [−0.03, 0.21]	0.10
Total Family Adversity	0.29 [0.18, 0.39]	<0.001	0.29 [0.18, 0.40]	<0.001	1.42 [1.18, 1.70]	<0.001
2 – 3 Years:
Inter-Parental Cruelty
At Risk / Taken into Care					2.55 [0.87, 7.42]	0.09
Maternal Psychopathology	0.27 [0.04, 0.49]	0.02	0.26 [−0.01, 0.53]	0.06	1.95 [1.31, 2.91]	0.001
Social Deprivation
Total Family Adversity	0.10 [0.00, 0.20]	0.04			1.24 [1.03, 1.49]	0.02
Stressful Life Events Assessment
1.5 years	0.10 [0.03, 0.18]	0.01	0.12 [0.03, 0.21]	0.01
2.5 years			0.12 [0.03, 0.21]	0.006
3.5 years	0.08 [0.00, 0.15]	0.04
7 years			0.11 [0.02, 0.19]	0.01	1.30 [1.14, 1.48]	<0.001

Multivariable Linear/Logistic regression analyses showing independent or marginal effect after adjusting for other factors remaining in the model. Items in bold were significant (p<0.05) and were included in the final model. Other items shown were close to significance (p≤0.1). Odds ratios (OR) and beta (β) values are given with 95% confidence intervals.

Stressful life events at 1.5, 2.5 and 3.5 years are for last 12 months. Stressful life events at 7 years are since age 5.5 years.

Social deprivation, total family adversity and stressful life events scores were standardized. Consequently their effect sizes are per SD.

A persistent conduct disturbance outcome was derived from maternal assessments at 4years and 7years based upon a child being in the top decile at both times (approximately 4% of the sample).

The N's for each group were: hyperactivity at age 4: 4034; at age 7: 3329. Conduct: 3608.

Parent substance misuse was not included in the table since there were no significant effects.

Genetic effects were assessed in girls as a dose effect (1df reflecting the linear effect of the count of high activity alleles). This approach was justified by the fact that MAOA has been shown to be among the 15% of X-linked genes that escape inactivation (Carrel and Willard, 2005). Genetic effects were assessed in boys as the difference between high/low activity alleles.

G × E interactions were explored by considering each measure of adversity (family adversity or stressful life events) separately. Interaction terms were derived as the product of genetic and adversity variables treated as continuous variables. The presence of an interaction can be interpreted as modification to the adversity trend for continuous measures or as a non-additive effect to the presence/absence of adversity for dichotomous measures. Dose response models in which the number of high activity alleles is used as a linear covariate were used to explore G × E interactions. For interactions, the estimated regression coefficient reflects a modification to the adversity trend for a single high activity allele.

In secondary analyses we parsed out the effects of stressful life events at an earlier and later time period. Interaction effects were calculated separately for cumulative stressful life events experienced from 6 months to 3 ½ years (from questionnaires at 1 ½, 2 ½ and 3 ½ years) and from stressful life events experienced from 5 to 7 years from the questionnaire at 7 years.

Figure 1 and 2 show the predicted values for the linear trend (B1 and B2). The analyses were re-run using stressful life events dichotomized as worst quartile (versus all other quartiles) in order to illustrate the variability (SE's) in these results (Figure 1 and 2, A1 and A2).

FIGURE 1. The Effects of Stressful Life Events and MAOA-LPR Genotype on Hyperactivity in Girls Aged 4 and 7 Years.

MAOA-LPR alleles: L = low activity, H = high activity.

The maximum possible hyperactivity score is 10. No difference in genotype distribution between number of stressful life events.

A1, A2: Graphs showing means (+ SEs) for the 3 genotypes derived from analyses in which cumulative life events experienced from 0.5 to 3.5 years were dichotomized into low stress (LS) and high stress (HS) (defined as the upper quartile of the sample).

A1: LL: LS, N = 228; HS, N = 88. LH: LS, N = 862; HS, N = 327. HH: LS, N = 822; HS, N = 286. Interaction p value = 0.003.

A2: LL: LS, N = 211; HS, N = 74. LH: LS, N = 810; HS, N = 298. HH: LS, N = 779; HS, N = 262. Interaction p value = 0.01.

B1, B2: graphs showing predicted values for the linear trend using the dose response model where the strongest differences are observed.

FIGURE 2. The Effects of Stressful Life Events and MAOA-LPR Genotype on Hyperactivity in Boys at Age 7 Years.

MAOA-LPR alleles: L = low activity, H = high activity. The maximum possible hyperactivity score is 10. No difference in genotype distribution between number of stressful life events.

A1, A2: Graphs showing means (+ SEs) for the H and L alleles derived from analyses using the dichotomous life event variable: low stress (LS); high stress (HS) (upper quartile of sample).

A1: L allele: LS, N = 712; HS, N = 225. H allele: LS, N = 1375; HS, N = 467. Interaction p value = 0.03.

A2: L allele: LS, N = 637; HS, N = 249. H allele: LS, N = 1268; HS, N = 502. Interaction p value = 0.1.

B1, B2: graphs showing predicted values for the linear trend using the dose response model.

RESULTS

SDQ hyperactivity scores tended to be higher in boys than girls and decreased from 4 to 7 years in both sexes (Supplementary Figure 1). Conduct scores were low and differed little between the sexes and the two time points (Supplementary Figure 2). Girls and boys were equally exposed to adversity (p > 0.203) except for stressful life events at 1 ½ years (p = 0.029) (Table 1).

Main effects of family adversity and stressful life events on behavior

Tables 2 and 3 show the components of overall adversity that had the strongest independent effects on outcomes after adjusting for all other variables in the final model. These components of adversity that had significant independent effects will now be discussed.

Hyperactivity (Tables 2 and 3)

In both girls and boys, family adversity pre-birth and during the first year of life had significant effects on hyperactivity, not only at age 4 but also at age 7 years. In terms of individual family adversity items it was found that in both sexes, pre-birth social deprivation and maternal psychopathology in the first year of life were the strongest predictors of hyperactivity at both time points. Additionally in girls, maternal psychopathology during pregnancy also had an effect. With regard to stressful life events, stressors experienced between the ages of 6 months to 2.5 years were the strongest predictors of hyperactivity in boys whereas in girls the influence of stressful life events at 2.5 to 3.5 years was maximal.

Conduct disturbances (Tables 2 and 3)

Stressor effects on conduct disturbances were not as evident. In girls, the only significant effects were for maternal psychopathology and social deprivation in the first year of life. In boys, the main effects were maternal psychopathology between 0 – 3 years, inter-parental cruelty at 0 – 1 years and pre-birth social deprivation. Stressful life events at 2.5 to 3.5 years in girls and 5.5 to 7 year in boys predicted conduct disturbances.

MAOA-LPR: main effects and interaction with stressful life events at ≤ age 3.5 years; influence on hyperactivity

Girls

MAOA-LPR genotype moderated the impact of stressful life events experienced between ages 6 months to 3 ½ years on hyperactivity in girls at 4 years (β = −0.22; 95% CI: −0.35 to −0.09; p for interaction = 0.001) and at 7 years (β = −0.18; 95% CI: −0.31 to −0.05; p for interaction = 0.009). The low activity allele was associated with increased hyperactivity in girls exposed to high stress (Figure 1).

In the presence of the interaction there were main effects of MAOA-LPR on hyperactivity in girls at age 4 (β = 0.33; 95% CI: 0.11 to 0.55; p = 0.003) and age 7 (β = 0.29; 95% CI: 0.07 to 0.51; p = 0.01) such that in girls exposed to little or no adversity, the high activity MAOA-LPR allele was associated with greater hyperactivity (Figure 1, B1, B2). There were no main effects of MAOA-LPR independent of the interaction.

Boys

In boys, there was no main effect of MAOA-LPR genotype. Unlike in girls, there was no G×E interactive effect on hyperactivity at age 4 years. Hyperactivity at 7 years was not influenced by the interaction of MAOA-LPR genotype with stressful life events experienced between ages 6 months to 3 ½ years (p = 0.1) (Figure 2, A2 and B2). However, there was a significant G × E interaction for stressful life events experienced during the narrower window of 1 ½ to 2 ½ years such that the low activity MAOA-LPR allele was associated with higher hyperactivity at age 7 years in boys who had been exposed to high stress (β = −0.21; 95% CI: −0.21 to −0.01; p = 0.03) (Figure 2, A1 and B1).

MAOA-LPR: main effects and interaction with stressful life events at ≤ age 3.5 years; influence on conduct disturbances

There was no MAOA-LPR main effect or G × E interactive effect on conduct disturbances in girls or boys.

Secondary analyses: MAOA-LPR: main effects and interaction with stressful life events at 5.5 – 7 years

MAOA-LPR had no moderating effect on the impact of stressful life events experienced between 5 ½ to 7 years for hyperactivity at age 7 years in girls (p = 0.9) or boys (p = 0.2) for conduct disturbances.

MAOA-LPR interaction with family adversity at ≤ age 3 years

There was no MAOA-LPR G × E interactive effect with family adversity (including maternal psychopathology and social deprivation) on girls' or boys' behavior at any time point or using a composite adversity score across the three time points.

Gene × environment correlation

We investigated whether the MAOA-LPR interaction with stressful life events could potentially be an artifact of gene-environment correlation (rGE). Analysis of rGE showed no such correlation with environmental measures (range r = −0.02 – 0.02).

DISCUSSION

In our longitudinal study of a large U.K. population sample, ALSPAC, we found that childhood stressful life events and family adversity, in particular maternal psychopathology and social deprivation experienced in the first few years of life, were associated with behavioral disinhibition detectable as early as age four years. The components of overall adversity and the timing had differing effects on behavior and no one component adequately explained all the variation (Tables 2, 3). In girls, the interaction of MAOA-LPR with stressful life events experienced from ages 6 months to 3 ½ years predicted hyperactivity at ages 4 and 7 years. In boys, the interaction of MAOA-LPR with stressful life events experienced between ages 1 ½ and 2 ½ years predicted hyperactivity at 7 years. Although early-life family adversity was a strong predictor of behavioral disinhibition there was no interaction with MAOA-LPR genotype. It may well be that different stressors interact with different genes at different time-points during development to influence behavior.

In line with earlier G × E studies (Caspi et al, 2002; Ducci et al, 2008; Foley et al, 2004; Kim-Cohen et al, 2006; Widom and Brzustowicz, 2006) we found that in both sexes, the low activity MAOA-LPR allele imparted risk for hyperactivity in the context of early life stress. Moreover, our results indicate a crossover of risk whereby in the 25% of girls who had been exposed to the greatest stress (≥ 9 life events from 6 months to 3.5 years), the low activity allele was associated with increased hyperactivity. In contrast, in the 75% of girls who had been exposed to less or no stress, the low activity allele was associated with lower hyperactivity. Based on our results it is tempting to speculate that the MAOA-LPR low activity allele confers risk for behavioral inhibition in the absence of stress and behavioral disinhibition in the presence of significant stress whereas modest early life stress exposure is beneficial to all genotypes (the crossover point of graphs in Figure 1, B1 and B2). Childhood inhibited and disinhibited behaviors both predict psychopathology in young adulthood (Caspi et al, 1996). However, the SDQ hyperactivity scores are not diagnostic of pathology per se and therefore the terms behavioral ‘inhibition’ and ‘disinhibition’ may be inappropriate descriptors in our study. Nevertheless, similar G × E crossover effects have been reported in three community studies of boys with longitudinal measures of maltreatment (Caspi et al, 2002; Foley et al, 2004; Kim-Cohen et al, 2006). Although our results are in line with earlier findings, it should be noted that apparently significant G×E interactions can actually be statistical artifacts (Eaves, 2006).

The results of our longitudinal study suggest that children may be most susceptible to family adversity in the first year of life and stressful life events in the first three years of life and that this adverse influence may be persistent. Indeed, earlier studies have shown that behavioral disturbances at age three are associated with psychopathology at age 21 (Caspi et al, 1996). Our study demonstrated that maternal psychopathology (anxiety / depression occurring in 20% of mothers), as experienced by boys and girls during the first postnatal year was a strong predictor of hyperactivity and conduct disturbances until at least the age of 7 years. This finding is in line with previous studies that have shown that children of depressed mothers have increased rates of antisocial behavior and depression (Burke, 2003) and are often exposed to harsh discipline and poor nurturing (Caspi et al, 2004; Kim-Cohen et al, 2005; McLearn et al, 2006). Moreover our study showed that social deprivation had the strongest impact in both sexes when experienced from pre-birth to age one. Nevertheless, another ALSPAC study suggests that the influence of family adversity at least on conduct disturbances may be relatively modest: family adversity in the first two years of life contributes 7.5% of the variance in persistent conduct disturbances between 4 and 11 years (CDS, personal communication).

Most studies of childhood behavioral disinhibition have focused on conduct disturbances. We chose to include hyperactivity firstly because hyperactivity and conduct disturbances both predict later-onset externalizing behaviors, for example, these behaviors in boys aged eight predict frequent drunkenness 10 years later (Niemela et al, 2006). Secondly, these behaviors may follow divergent paths; hyperactivity at age 6-7 years is a stronger predictor than conduct disturbances of poor social adjustment and relationship problems 10 year later (Danckaerts et al, 2000). In fact we did not detect G × E interaction effects on conduct disturbances, perhaps because this is a relatively weak phenotype in a general population sample of pre-pubertal children. Very few children had high conduct problem scores which may account for the negative G × E findings.

It is not clear why the G × E results were weaker in boys than in girls. The influence of adversity on behavior was the same in both sexes. Boys tended to be slightly more hyperactive. Conduct problem scores differed little between the sexes unlike in adulthood when the prevalence of antisocial behavior is much higher in men than women (Rutter et al, 2003). Indeed, a recent study has shown that MAOA-LPR interacts with CSF testosterone to predict antisocial behaviors in men (Sjoberg et al, 2008). Thus there may be hormonal or age-related effects that complicate the relationship between MAOA-LPR genotype and early life stress, and therefore the G × E interaction may be hard to detect in pre-pubertal boys. Only one study has detected a G × E effect in boys aged 7 years but in this case the MAOA-LPR interaction was with severe maltreatment (Kim-Cohen et al, 2006).

MAOA appears to be among the 15% of X-linked genes that escape X-inactivation (Carrel and Willard, 2005). However, partial X inactivation could conceivably result in higher brain MAOA expression levels for female homozygotes than male hemizygotes. Furthermore, one study has shown epigenetic gene regulation in the MAOA promoter region only in females, representing a possible dosage compensation mechanism that does not correlate with X-inactivation (Pinsonneault et al, 2006). This may be another explanation for why our findings were more striking in pre-pubertal girls than boys.

We regard this as an exploratory study since it was not clear a priori which stressors would have the greatest effect in this population sample. Any correction for multiple testing would render most of the results non-significant. This is a general population sample and there was not much variability in hyperactivity scores and even less variability in conduct scores. Moreover, the SDQ scales are not a measure of pathological behavior. Nevertheless, the strength of this study is that because we had a very large sample we were able to detect a modest effect of MAOA-LPR × early life stress on behavioral disinhibition in a general population sample of children in which few had been exposed to severe adversity.

In conclusion, our study has shown that within a general population sample, exposure to common stressors within the first few years of life and interaction between MAOA-LPR genotype and stressful life events predicts behavioral disinhibition in pre-pubertal girls and boys.