Testing for gene × environment interaction effects in attention deficit hyperactivity disorder and associated antisocial behavior†
Please cite this article as follows: Langley K, Turic D, Rice F, Holmans P, van den Bree MBM, Craddock N, Kent L, Owen MJ, O'Donovan MC, Thapar A. 2007. Testing for Gene × Environment Interaction Effects in Attention Deficit Hyperactivity Disorder and Associated Antisocial Behavior. Am J Med Genet Part B 147B:49–53.
Abstract
Gene × environment (G × E) interactions are increasingly thought to have substantial influence on the aetiology and clinical manifestations of complex disorders. In ADHD, although main effects of specific genetic variants and pre- or peri-natal variables have been reported and replicated using pooled analyses, few studies have looked at possible interactions. In a clinical sample of 266 children with ADHD, we tested for interaction between gene variants (in DRD4, DAT1, DRD5, and 5HTT) found to be associated with ADHD in pooled analyses and maternal smoking, alcohol use during pregnancy and birth weight. First, G × E effects on a diagnosis of ADHD were tested using conditional logistic regression analyses. Second, possible modifying effects of G × E on symptoms of associated conduct disorder and oppositional defiant disorder (ODD) were investigated using linear regression analysis. The sample size associated with each of the analyses differed as not each variant had been genotyped for each individual. No effects of G × E on ADHD diagnosis were observed. The results suggest that lower birth weight and maternal smoking during pregnancy may interact with DRD5 and DAT1 (birth weight only) in influencing associated antisocial behavior symptoms (ODD and conduct disorder). These preliminary findings showed no evidence of interaction between previously implicated variants in ADHD and specific environmental risk factors, on diagnosis of the disorder. There may be evidence of G × E on associated antisocial behavior in ADHD, but further investigation is needed. © 2007 Wiley-Liss, Inc.
Abbreviations used:
ADHD, attention deficit hyperactivity disorder; ODD, oppositional defiant disorder; G × E, gene × environment interaction.
INTRODUCTION
Attention deficit hyperactivity disorder (ADHD) is one of the most prevalent childhood mental health disorders [Leibson et al., 2003] and the most common reason for attendance in child mental health services [Woodward et al., 1997]. ADHD is a multifactorial disorder with good evidence that both genetic and environmental factors contribute to its development and modify clinical presentation [Thapar et al., 2005a].
A number of association findings between specific gene variants and ADHD have been replicated in multiple samples and meta- or pooled-analyses [Lowe et al., 2004; Faraone et al., 2005; Thapar et al., 2005a]. These include the 7-repeat allele of a 48 bp VNTR in the Dopaminergic Receptor D4 (DRD4) gene [Faraone et al., 2001]; the 10-repeat allele of a 480 bp VNTR in the Dopamine Transporter gene (DAT1) [Faraone et al., 2005]; a microsatellite marker in the Dopamine Receptor D5 (DRD5) gene [Maher et al., 2002; Lowe et al., 2004]; and a 44 bp insertion/deletion polymorphism in the promoter region of the Serotonin Transporter gene (5HTTLPR) [Faraone et al., 2005].
Despite convergence of findings, a number of non-replications have been found in studies of these and other genetic variants for which there are several potential explanations. First, non-replication across different samples may in part be explained by the clinical heterogeneity of ADHD. Second, differing levels of exposure to environmental risk factors between samples may also be important where gene × environment interaction (G × E) effects contribute [Moffitt, 2005]. G × E is defined as genetically influenced individual differences in the sensitivity to specific environmental factors/pathogens [Rutter and Silberg, 2002; Moffitt, 2005]. Thus, specific gene variants may only exert risk effects on disorder where an individual is also exposed to a particular environmental risk factor. There is increasing evidence to suggest interactions between genetic and environmental factors are common and contribute substantially to the aetiology of psychiatric disorders [Caspi et al., 2002, 2003; Rutter and Silberg, 2002; Moffitt, 2005]. There is also emerging evidence to suggest that G × E may influence clinical manifestations and course of a disorder [Thapar et al., 2007].
There are a number of different environmental risk factors thought to influence ADHD and have a modifying effect on its clinical presentation. However, the most consistent evidence, supported by meta-analysis or pooled analysis is for association between maternal smoking in pregnancy, low birth weight, and ADHD. A recent pooled-analysis of case-control studies [Langley et al., 2005] found an odds ratio of 2.39 (95% C.I. 1.61, 3.52) for a diagnosis of ADHD in those individuals whose mothers smoked during pregnancy. A meta-analysis of case-control studies [Bhutta et al., 2002] pooled data from 16 studies of ADHD and low birth weight revealing a combined calculated risk ratio of 2.64 (95%C.I. 1.85, 3.78, P < 0.001).
To date there have been relatively few published studies testing for G × E effects in ADHD, although at least five studies have reported significant results. Kahn et al. [2003] investigated interactions between the DAT1 480 bp VNTR and maternal smoking during pregnancy on ADHD symptoms, in a population-based sample of 161 5-year-olds. Significant increases in maternally reported hyperactive–impulsive symptoms were observed for individuals whose mothers smoked during pregnancy and who were homozygous for the DAT1 10 allele (t = 7.5, P < 0.01). Secondly, Seeger et al. [2004] found a significant interaction between season of birth (spring and summer births) and DRD4 48 bp VNTR on diagnosis of ADHD with comorbid conduct disorder. A third study by Brookes et al. [2006] found association between ADHD and an interaction between maternal alcohol consumption during pregnancy and a previously uninvestigated haplotype (combination of alleles across different variants) in DAT1. Recently, in a population-based sample of twins, Neuman et al. [2006] looked at possible G × E between maternal smoking during pregnancy and genotypes in the DAT1 480 bp VNTR and the DRD 48 bp VNTR on ADHD symptoms and diagnoses as defined by DSM-IV and through latent class analysis. Interactions between maternal smoking during pregnancy and the DRD4 7-repeat allele presence and presence of the DAT1 440 bp allele (9-repeat allele) were observed. Finally, analysis by our own group [Thapar et al., 2005b] revealed significant interaction between birth weight and COMT Val158Met genotype, modifying the number of conduct disorder symptoms in a clinical ADHD sample.
The aim of this article is to investigate G × E effects in ADHD and associated antisocial behaviors (conduct disorder symptoms and oppositional defiant disorder (ODD) symptoms). Specifically, we tested for interactions between the four gene variants found to be most robustly associated with ADHD; DRD4 48 bp VNTR, DAT1 480 bp VNTR, DRD5 CA(n) microsatellite and the 5HTTLPR, and the two environmental factors most strongly associated with ADHD; maternal smoking during pregnancy and birth weight. Because of previous findings [Brookes et al., 2006], we also investigated maternal alcohol consumption during pregnancy.
METHODS
A sample of 266 children diagnosed with ADHD and their biological parents was utilized for these analyses. Children, with a mean age of 9 years 4 months (SD 2 years 0 months) were referred by child and adolescent psychiatrists or pediatricians to two centers (Cardiff and Birmingham) in the United Kingdom. Research diagnoses of DSM-IV, ICD-10, and DSM-III-R ADHD, as well as other psychiatric disorders including conduct disorder, were obtained from parental information using the Child and Adolescent Psychiatric Assessment (CAPA) [Angold et al., 1995] with data on the pervasiveness of symptoms obtained from teachers using the Child ADHD Teacher Telephone Interview (ChATTI) [Holmes et al., 2004] in the Cardiff sample and teacher Conners questionnaires in the Birmingham sample. Children with full-scale IQ scores below 70, who had Autism, Tourette Syndrome or any neurological condition were excluded from the study. Written informed consent from parents and assent from children was obtained prior to participation. The study was approved by the North West Multicentre Research Ethics Committee, UK.
Genotyping
All probands and their biological parents were genotyped for each of the genetic variants. Genotyping was performed at various time points during sample collection, and this results in differing sample sizes for different sets of analysis.
For the DRD4 48 bp VNTR conditions and primers described by Payton et al. [2001] were used. Those with one or more copies of the 7-repeat allele were considered to have the risk genotype. For the DAT1 480 bp VNTR, conditions and primers described by Daly et al. [1999] were used. Those homozygous for the 10 allele were considered to have the risk genotype. For the DRD5 (CA)n microsatellite, conditions and primers described by Sherrington et al. [1993] were used. Those with one or more copies of the 5 allele were considered to have the risk allele. Finally, for the 5HTTLPR, conditions and primers described by Lesch et al. [1996] were used. Those homozygous for the long allele were considered to have the risk genotype.
Environmental measures
Information regarding pregnancy and birth complications was collected using maternally completed questionnaires with questions from the Lewis and Murray Pregnancy and birth complications scale [Lewis and Murray, 1987].
Maternal smoking during pregnancy was retrospectively assessed by maternally rated questionnaire. A yes/no answer was required to the question “Did you smoke during pregnancy.” One hundred eighteen mothers reported having smoked during pregnancy (42.5%).
The question “What did your child weigh at birth?” was also answered retrospectively by mothers. Weight in grams provided a continuous measure of birth weight. The mean birth weight was 3,275 g (SD 675 g) which is in line with the UK average for singleton births (www.hefa.gov.uk). Maternal retrospective report of birth weight has been found to be highly reliable, with a significant correlation of over 0.98 with birth records [Olson et al., 1997].
Mothers stated whether or not they drank alcohol during pregnancy as well as the quantity. To ensure consistency with the definition used by Brookes et al. [2006], those mothers who drank about one drink a month or more (n = 63, 23.7%) were considered to have drunk alcohol during their pregnancy.
Statistical Analysis
Testing G × E for diagnosis of ADHD
Conditional logistic regression analysis enables the analysis of G × E in case only samples where parental genotypes are known [Cordell, 2004; Cordell et al., 2004]. Conditional logistic regression analysis of cases and pseudocontrols is equivalent to matched case-control analysis and enables the detection of G × E in family-based trios [Cordell, 2004; Cordell et al., 2004]. The presence or absence of the risk allele in the proband (considered as the “case”) is matched to three pseudocontrols, derived from the parental genotypes which were not transmitted to the child. Thus, there are four sources of information (one “case” derived from the observed transmitted genotype and three “controls” derived from the possible genotypes which were not transmitted to the child) for each individual. For example, if two heterozygous parents (genotypes mother 1,2 and father 1,2) transmit the genotype 1,1 to their child, which is the genotype for the “case.” The other genotypes that could have been passed to the child become the pseudocontrols (i.e., 1,2 1,2 and 2,2). Environmental information for each case and pseudocontrol can be included in analyses of G × E. Because the pseudocontrols are derived from the proband information, environmental information is the same for all four observations (each case and its three pseudocontrols). Therefore, it is not possible to test for main environmental effects using conditional logistic regression. Using these data, a Chi-square test was performed to investigate whether or not the risk variables (G and G × E) were present significantly more frequently in probands than in the pseudocontrols so that it could be concluded they were associated with the disorder.
For this analysis, observations with the reported risk allele were assigned the value of 1 and those without the risk allele 0. Similarly, individuals who were exposed to the risk environment (e.g., whose mothers smoked during pregnancy) were given the value of 1 and those who were not exposed to environmental risk (e.g., whose mothers did not smoke during pregnancy) 0. Birth weight was used as a continuous variable. Genetic and environmental variables were multiplied to provide the G × E term.
We used the enter method to assess the contribution to the explained variance of the G × E term as follows: the difference in the variance explained (r2) when the main genetic effects alone were entered into the model was compared to that when the G × E term was added as well. All statistical tests were two-tailed. This analysis was performed using the Cox's regression function in SPSS version 11 (Norusis/SPSS, Inc., (2001), Chicago, IL).
G × E modifying effects on antisocial symptoms
Linear and logistic regression were performed using the enter method. Again, the risk allele and exposure to the environmental risk factor were given the value of 1 and the non-risk allele and non-exposure to the environmental risk factor 0. The product of the genetic and environmental factors provided the G × E term. All analyses were performed on both allele and genotype data, where feasible. Here, we present the results of the analyses of the alleles. The results for genotype analyses (where having two copies of this risk allele was given the risk value of 1) remained the same (available from first author). In this analysis, unlike conditional logistic regression, main effects of both genetic and environmental factors can be considered.
For regression analysis, in the first block, main effects of the genetic and environmental measures were entered into the model. In the second block, the G × E term was added. The change in proportion of variance explained by the two models is reported. Any models where the interaction term was significant were then retested with the addition of any covariates that we found to be significantly associated with the outcome measure added to block one of the model, to test whether the interaction term remained significant after taking variation explained by these covariates into account.
RESULTS
Testing G × E Effects for the Diagnosis of ADHD
Conditional logistic regression analysis revealed no evidence for significant interaction between any of the genetic markers studied and the environmental factors influencing diagnosis of ADHD. These findings are summarized in Table I.
| Main genetic effects | Interaction with smoking during pregnancy | Interaction with birth weight (in grams) | Interaction with alcohol during pregnancy | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| n | χ2 (df 1) | P-value | n | χ2 (df 1) | P-value | n | χ2 (df 1) | P-value | n | χ2 (df 2) | P-value | |
| DRD4 48 bp VNTR | 112 | 0.29 | 0.59 | 110 | 1.07 | 0.30 | 110 | 0.005 | 0.94 | 92 | 0.32 | 0.57 |
| DAT1 480 bp VNTR | 180 | 0.11 | 0.74 | 180 | 1.08 | 0.30 | 176 | 0.87 | 0.35 | 118 | 0.02 | 0.89 |
| DRD5 CA(n) microsatellite | 78 | 0.002 | 0.96 | 75 | 2.78 | 0.10 | 73 | 2.22 | 0.14 | 53 | 0.07 | 0.79 |
| 5HTT LPR | 131 | 0.07 | 0.77 | 130 | 0.03 | 0.87 | 128 | 0.20 | 0.66 | 79 | 0.22 | 0.64 |
Testing G × E Effects on Associated Conduct Disorder and ODD Symptoms
Univariate regression analyses revealed significant evidence of G × E for associated antisocial behavior (see Table II). First, there was evidence of interactions between DAT1 and birth weight on CD symptoms (r2 change = 0.02, t = −2.16, P = 0.03). Second DRD5 appeared to interact with both maternal smoking during pregnancy and birth weight to modify ODD symptoms (r2 change = 0.08, t = 3.25, P = 0.002 and r2 change = 0.07, t = −2.91, P = 0.004, respectively). There were trends for interactions between 5HTT and maternal smoking on conduct disorder symptoms (r2 change = 0.02, t = 1.94, P = 0.05) and between DRD4 and birth weight, influencing ODD symptoms (r2 change = 0.02, t = −1.89, P = 0.06). As described in the Methods Section, all these interactions were over and above any main effects of genotype and environmental factors.
| Interactions with smoking in pregnancy | Interactions with birth weight | Interactions with alcohol in pregnancy | |||||||
|---|---|---|---|---|---|---|---|---|---|
| r2 change | t | P-value | r2 change | t | P-value | r2 change | t | P-value | |
| DRD4 | |||||||||
| CD symptoms | 0.005 | 1.07 | 0.29 | 0.004 | −0.91 | 0.36 | 0.003 | 0.78 | 0.44 |
| ODD symptoms | <0.001 | −0.28 | 0.78 | 0.02 | −1.89 | 0.06 | 0.005 | −0.99 | 0.33 |
| DAT1 | |||||||||
| CD symptoms | 0.006 | −1.25 | 0.21 | 0.02 | −2.16 | 0.03a | 0.002 | −0.76 | 0.45 |
| ODD symptoms | 0.002 | 0.74 | 0.46 | 0.001 | −0.44 | 0.66 | <0.001 | −0.14 | 0.89 |
| DRD5 | |||||||||
| CD symptoms | 0.01 | −1.23 | 0.22 | 0.001 | −0.24 | 0.81 | 0.002 | −0.46 | 0.65 |
| ODD symptoms | 0.08 | 3.25 | 0.002b | 0.07 | −2.91 | 0.004b | 0.001 | 0.26 | 0.80 |
| 5HTT | |||||||||
| CD symptoms | 0.02 | 1.94 | 0.05 | <0.001 | 0.13 | 0.90 | 0.003 | 0.55 | 0.58 |
| ODD symptoms | 0.005 | 0.97 | 0.33 | 0.003 | 0.64 | 0.52 | 0.01 | −1.20 | 0.23 |
- a Significant at the P < 0.05 level.
- b Significant at the P < 0.01 level.
Multivariate analysis was subsequently performed for each of these findings, allowing for inclusion of covariates independently associated with the outcome variable. The significant covariates included were age at time of assessment, gender, full-scale IQ, total number of ADHD symptoms, and the other environmental variables (i.e., alcohol and smoking during pregnancy for interactions with birth weight and vice versa). Following this more stringent, multivariate analysis, none of the associations remained significant.
DISCUSSION
There was no evidence of significant G × E effects on ADHD in our sample for any of the genetic variants tested together with maternal smoking during pregnancy, birth weight or maternal alcohol consumption during pregnancy.
There was evidence of G × E for associated antisocial behavior symptoms. Specifically, we found evidence for maternal smoking during pregnancy × DRD5 for ODD symptoms; birth weight × DRD5 for ODD symptoms; and birth weight × DAT1 for conduct disorder symptoms. We found no evidence of G × E for maternal alcohol use in pregnancy. These findings, taken together with our previous findings of a COMT × birth weight interaction for conduct disorder symptoms [Thapar et al., 2005b], suggest G × E may be especially important in influencing the manifestation of antisocial behavior (ODD and/or conduct disorder symptoms) in children with ADHD. However, these modifying effects on antisocial behavior did not remain significant after including covariates in the analysis. There are number of potential explanations for our failure to find G × E for ADHD.
First, it may be that the gene variants examined do not interact with the selected environmental factors in increasing susceptibility for ADHD. Certainly to date, the evidence from twin and adoption studies for G × E effects is stronger for phenotypes other than ADHD, notably depression, antisocial behavior, and substance dependence [Caspi et al., 2002, 2003].
However, we adopted a cautious approach in testing for G × E effects. First, gene variants were selected for which there was good a priori evidence of association. Previous studies reporting evidence of G × E with measured genotypes have not detected significant main effects of the genetic variants although there were strong main effects of the environmental variables [Cutrona et al., 1994; Cadoret et al., 1995; Silberg et al., 2001]. Second, independently significant covariates were added to the analysis of G × E where significant G × E effects were found for associated antisocial behavior. This was a cautious approach: researchers have argued that risk factors do not act independently [Rutter et al., 1997] and previous studies [Caspi et al., 2002, 2003] have either not considered covariates or have added them individually to the statistical model.
It is possible that methodological aspects of the present analyses may have influenced results. The use of family-based analysis for G × E interactions with ADHD, meant only parent-proband trios could be utilized, which have reduced the sample size in these analyses (since mother-proband duos with or without unaffected siblings are excluded). Moreover, ADHD duos had a slightly different clinical presentation to trios [West et al., 2002]. Finally, insufficient statistical power and/or multiple testing resulting in type I and type II errors must also be considered. The number of cases for each analysis was modest, especially considering that larger samples may be required to detect G × E [Chronbach, 1991] although accurate measurement reduces the need for very large samples [Wong et al., 2003]. However, we must also consider the fact that our group has previously found evidence for interaction between the COMT Val158Met variant and birth weight on conduct disorder symptoms in an ADHD sample, with covariates included [Thapar et al., 2005b]. This indicates that our sample does have the power to detect interactions in some instances.
Large samples and accurate measurement of both environmental and genetic risk factors, as well as the outcome measure, are essential. Accuracy of measurement is improved in our study by the use of a standardized and widely reliable psychiatric interview (the CAPA) and the use of multiple informants to obtain diagnosis. In this sample, inter-rater agreement from the CAPA for ADHD and antisocial behavior symptoms was high with Cohen's Kappa values of between 0.80 and 1.00 [classified as being “good” to “very good”—Landis and Koch, 1977]. Moreover, the use of measured genotypes meant that genetic risk could be assessed directly without the need for the useful, but indirect methods required by family and twin studies [Rutter and Silberg, 2002].
Environmental variables are difficult to measure but maternal reports of birth weight and maternal smoking during pregnancy have been shown to be accurate, showing strong agreement (correlations over 0.90) with measurements documented in birth records [Olson et al., 1997]. Furthermore, there is some evidence that alcohol use during pregnancy may be better obtained from maternal questionnaire reports than from antenatal records in that this information is not routinely recorded in antenatal records [Rice et al., 2006; Delgado-Rodriguez et al., 1995]. It could, however, be argued that our environmental measures may cover a range of behaviors. For example, a mother who smoked infrequently and stopped after the second trimester, would have be classified as having the same environmental risk as a mother who smoked 20 cigarettes a day throughout pregnancy, although the actual effect on the fetus may not be the same. Again, however, it is worth noting that the previous G × E interaction findings reported by our group [Thapar et al., 2005b] indicate that our measures are sufficiently accurate to identify interactions.
It is interesting to note also, our lack of G × E interaction findings may be due to the fact that this study used a sample of relatively modest size. It is possible that we did not have the statistical power to detect some interactions and therefore, these findings of no association should be replicated in a larger sample before the possibility of interactions can be dismissed.
A reported interaction between DAT1 and maternal alcohol consumption during pregnancy predicting ADHD [Brookes et al., 2006] influenced our decision to analyze alcohol consumption during pregnancy as a potential environmental influence. However, the Brookes et al. [2006] article reports interaction findings for a two marker haplotype of DAT1 (including the 480 bp VNTR). As the second marker in this haplotype (a VNTR in intron 8) was not genotyped in this sample, replication was not possible. Brookes et al. [2006] do not report interaction analyses for only the 480 bp VNTR alone.
These preliminary findings represent, to our knowledge, the first systematic analysis of G × E in ADHD. Looking at the most robustly associated genetic and prenatal risk factors, we investigated G × E effects on both diagnosis of the disorder and modifying effects on associated antisocial behavior. No significant G × E effects were revealed for ADHD, which may reflect either small sample size, or a lack of interactions between these specific genetic and significant environmental measures. G × E effects were found for associated antisocial behavior, although these did not remain significant after inclusion of covariates. However, we may have been overly stringent by including covariates. These findings, as well as our own previous results, suggest the importance of examining the modifying effects of G × E on antisocial behavior in ADHD. Finally, these should be treated as preliminary findings, whilst replication of initial findings and pooled analysis are always needed.
Acknowledgements
The work in this article was supported by grants from The Wellcome Trust and Action Research, Kate Langley is funded by a Wellcome Trust Value in People Award.
