Attention deficit hyperactivity disorder (ADHD) is a complex disorder characterized by extremes of inattention, hyperactivity and impulsivity, by clinical and genetic heterogeneity and by high heritability whether it is measured as a discrete diagnostic entity or as a continuous trait in the general population (Swanson et al.
2009; Thapar et al.
2007). Despite high heritability, no gene conferring large relative risk has been identified. Meta-analyses of ADHD linkage studies supports genome-wide significance on 16q21-16q24 (Zhou et al.
2008) but the largest and most comprehensive genome-wide association study did not identify any genome-wide significant findings (Franke et al.
2009; Neale et al.
2010). Identification of the genetic contributions to the disorder is complicated by phenotypic and genetic heterogeneity, low penetrance and low statistical power of existing studies (van der Sluis et al.
2010). One way to enhance the power of genetic discovery is to reduce clinical and genetic heterogeneity by use of endophenotypes (Kendler and Neale
2010). Endophenotypes are biological traits that mediate the association between some of the genetic risks for a disease and the disease phenotype and which have a genetic architecture that is less complex than that of the disorder itself (Crosbie et al.
2008; Gottesman and Gould
2003; Szatmari et al.
2007). Potentially useful endophenotypes are those that are associated with the disorder, evident in relatives because of their increased genetic risk, heritable and share genetic risk with the disease phenotype (Crosbie et al.
2008; Gottesman and Gould
2003; Kendler and Neale
2010).
Considerable research into ADHD endophenotypes has focused on executive function processes that are involved in planning, decision making and action. Among the most extensively studied candidate endophenotypes are motor response inhibition, response latency and response variability (Frazier-Wood et al.
2012). Meta-analysis indicates that patients with a diagnosis of ADHD have deficient inhibitory control compared with controls with a medium effect size of 0.63 (Lipszyc and Schachar
2010) and that inhibitory control is worse in non-ADHD siblings and in parents of ADHD probands than it is in normally developing individuals (Goos et al.
2009; Schachar et al.
2005). Twin studies indicate that inhibitory control is heritable (Friedman et al.
2008; Schachar et al.
2011). Several risk alleles are associated with poor inhibition (Bellgrove and Mattingley
2008; Langley et al.
2004; Swanson et al.
2000) and with brain activity associated with response inhibition (Barnes et al.
2011; Cummins et al.
2011). Response speed and variability reflect consistency of attention and effort. Meta-analysis shows a small-to-medium effect size for response latency and a medium effect size for variability (0.71) (Lipszyc and Schachar
2010; Sergeant
2000) in distinguishing individuals with and without an ADHD diagnosis. Twin studies have found that response variability was heritable (Frazier-Wood et al.
2012; Friedman et al.
2008; Young et al.
2009).
Endophenotypes derived from simple laboratory tests such as the SST have the potential advantage of being suitable for general population research where they could help identify individuals at genetic risk for disorder without a time-consuming and costly diagnostic assessment. Despite this potential utility, there has been little research of validity of candidate endophenotypes in the general population (c.f. Cornish et al.
2005). In this study, we use the stop signal task (SST) (Logan et al.
1997; Verbruggen and Logan
2009a) to investigate the relationship of response inhibition, latency and variability and ADHD traits in the general population. If response inhibition, latency and variability are genetically informative endophenotypes, they should be correlated with ADHD traits even after controlling for age, gender, socioeconomic status and ethnicity. They should also be heritable and share genetic risk with ADHD traits. We investigated these questions using the SST and the Strengths and Weaknesses of ADHD-symptoms and Normal-Behavior (SWAN) rating scale as a measure of ADHD traits. In related individuals, we used Sequential Oligogenic Linkage Analysis Routines (SOLAR) to estimate heritability of individual traits and shared genetic risk of pairs of traits.
Analyses
Regression analysis was conducted to identify the best combination of variables making a unique contribution to SSRT, GoRT and GoRTSD while considering family as a random effect (Singer
1998). In order to normalize the residuals from each model, we modeled the log of the stop task variables. Because of the high correlation of latency and variability, we calculated a new variable, GoRTSD/GoRT. However, the distribution of the residuals violated the assumptions of the model and no transformation of the derived variable improved the result. Therefore, we chose to control GoRT in models of GoRTSD by entering GoRT into regression models. Models included gender, ADHD traits and age which were added using a forward stepwise approach. The higher order terms, age*age, gender*age, age*age*gender were also assessed when main effects were significant. We plotted predicted values obtained from models with and without the higher order terms in order to compare the predicted values between the simpler and more complex models to assess the clinical significance of the higher order terms.
In order to evaluate the effect of age, gender, and ADHD trait on the stop task values, we calculated an approximation of Cohen’s effect size using the equation
$$ {{{\left( {\mathrm{pred}1-\mathrm{pred}2} \right)}} \left/ {{\mathrm{sqrt}\left( {\mathrm{stderrpred}1\hat{} 2-\mathrm{stderrpred}2\hat{} 2} \right)}} \right.} $$
where stderrpred1 and stderrpred2 are the standard deviations of a single observation as obtained from the regression model. Effect sizes were calculated for specific scenarios, such as the effect of gender at age 6 in children with a Swan score of 54.
To determine if extremely low SWAN scores might also be maladaptive, we compared participants with high, medium and low SWAN scores on rates of parent-reported ADHD, anxiety, depression, learning disability and other disorders.
We estimated SSRT for 414 (2.6 %) participants who reported taking medication for ADHD within 48 h of testing by adjusting SSRT by 0.75 effect size based on the results of published studies of the effect of a moderate dose of stimulant medication on SST performance (Tannock et al.
1995). Results did not differ when analyses were conducted with and without these participants; consequently they were excluded from the subsequent analyses. The final sample (14,388) consisted of 7,176 boys (49.9 %) and 7,212 girls (50.1 %).
Heritability of each SST variable and of ADHD traits with age and sex as covariates was estimated using SOLAR (Almasy and Blangero
1998) in the subset of 3,507 families that included multiple full siblings (7,483). Of these families, 3,081 were from families with two children or adolescents, 387 with 3, 35 with 4, and 4 with 5. SOLAR decomposes the total variance of the phenotype into components that are the result of genetic effects (i.e. polygenic, additive genetic variance), measured covariates and random environmental effects (i.e. measured environmental factors and random unmeasured factors). The relative contribution of genetic factors to each trait is then estimated by heritability (h
2): the ratio of the genetic variance component to the residual (after removal of variance explained by covariates) phenotypic variance. We calculated confidence intervals for heritability estimates under the asymptotic normality assumption of the Maximum Likelihood estimator where sample size is ~ 7,000 using the formula
$$ 100\left( {1-\alpha } \right)\%\;\mathrm{asymptotic}\;\mathrm{confidence} \operatorname {int}\mathrm{ervals}\;\left[ {{{\mathrm{h}}^2}-\mathrm{Z}\left( {{\alpha \left/ {2} \right.}} \right)*\mathrm{SE},\;{{\mathrm{h}}^2}+\mathrm{Z}\left( {{\alpha \left/ {2} \right.}} \right)*\mathrm{SE}} \right] $$
(Neale and Miller
1997).
Bivariate analyses were conducted on pairs of neuropsychological measures that demonstrated significant univariate heritability to partition the phenotypic correlation between two traits (RhoP) into genetic (Rhog) and environmental (Rhoe) correlations according to the equation
$$ RhoP=Rhog\sqrt{{h_1^2}}\sqrt{{h_2^2}}+Rhoe\sqrt{{\left( {1-h_1^2} \right)}}\sqrt{{\left( {1-h_2^2} \right)}} $$
where h
1
2 and h
2
2 correspond to heritabilities of traits 1 and 2, respectively. Evidence of pleiotropy (a common set of genetic influences affecting more than one trait) is indicated by a genetic correlation significantly different from zero (Rhog = 0: no shared gene effect; Rhog = 1 or –1: complete pleiotropy). We estimated heritability of SST variables with and without SWAN ADHD traits as covariates in order to control for the influences of ADHD traits on cognitive performance. Heritability estimates were essentially unchanged and are presented with ADHD controlled.
Discussion
In the largest general population study reported to date we found support for the validity of response inhibition as an endophenotype of ADHD. Response inhibition, latency and variability were significantly correlated with ADHD traits even after controlling for age and gender. Analysis of the family data indicated that inhibition, latency and variability were heritable; however, of these three measures only inhibition shared genetic risk with ADHD traits. Current results have implications for understanding the genetic contributions to ADHD traits and to higher-order executive functions as well as for the design of genetic studies.
Participants with greater ADHD traits had longer SSRTs indicating inferior inhibition, slower response latencies and greater response variability than did those with fewer ADHD traits. The relationship of ADHD traits and inhibition was strong. Participants with highest ADHD trait scores had SSRTs that were substantially slower indicating less efficient inhibition than among those with the lowest trait scores across age and sex. The magnitude of these differences was substantial for both girls (e.g., age 6; 138 ms, ES = 1.48) and boys (e.g., age 6; 129 ms, ES = 1.48). These effects are comparable to those found in a meta-analysis of studies comparing ADHD clinic cases and healthy controls (Lipszyc and Schachar
2010). In that review, ADHD cases had a mean SSRT of 330 ms and controls had a mean score of 254 ms. Participants in those studies were approximately 10–12 years of age. Estimates of SSRT in the current data for children of that age were 329 ms for those with greatest ADHD traits scores and 231 ms for those with lowest ADHD trait scores. These converging data provide considerable support for the current results. We conclude that this general population sample included individuals at the extremes of ADHD trait and SST performance distributions that are equivalent to cases and controls typical of clinic-based case–control studies. Indeed, we found that 6.3 % of all participants self-reported a diagnosis of or treatment for ADHD supporting the conclusion that the sample was quite representative of the general population. Participants with high ADHD trait scores were substantially more variable than those with low ADHD trait scores with effect sizes that are comparable to those observed for response time variability in previous studies (Frazier-Wood et al.
2012). There was far less of a difference in response latency between those with high and those with low ADHD trait scores. We also note that the relationship of ADHD traits and each cognitive measure was consistent across the entire distribution of ADHD indicating that associations were essentially linear and not driven by those at the extremes of the ADHD trait.
Univariate heritability estimates were significant for SSRT, GoRT, and GoRTSD. The moderate magnitudes of these heritability estimates (from 0.26 to 0.31) are comparable to heritability estimates from prior studies of executive functions, specifically those on SSRT (Andreou et al.
2007; Frazier-Wood et al.
2012; Friedman et al.
2008; Kuntsi et al.
2010; Schachar et al.
2011). Moreover, SSRT, GoRT and GoRTSD were heritable even with adjustment for the well-documented effect of ADHD traits on these parameters indicating that heritability of cognitive traits are similar across the range of ADHD traits. We interpret the observed interfamilial correlation as indicative of genetic effects because previous twin studies show that familial influences on ADHD traits (Faraone and Doyle
2000) and on SST parameters (Schachar et al.
2011) are largely due to genetic effects with minimal influence of shared environment.
There was evidence of shared genetic risk between poor response inhibition (longer SSRT) and high ADHD trait scores whereas there was little evidence for shared genetic risk for ADHD traits and either response latency (GoRT) or variability (GoRTSD) (c.f. Kuntsi et al.
2010; Frazier-Wood et al.
2012). Shared genetic risk implies the potential for common genetic contributors. The significant although modest phenotypic and genetic correlation of SSRT and SWAN scores indicates that the ADHD phenotype is complex and is likely to be the final common outcome of multiple biological pathways only some of which involve deficient motor response inhibition. Significant genetic correlation indicates that genetic factors influence the extent to which these traits overlap. However, significant genetic correlation does not prove that a purported endophenotype has a simpler genetic architecture. Ultimately, the utility of an endophenotype will be proven in gene discovery. The role of this paper was to assess the potentially utility of response inhibition, latency and variability as measured in the SST as candidate endophenotypes.
The results of the current study are consistent with previous research on the link between ADHD and deficient response inhibition, latency and variability in clinical samples. Current results also confirm the findings of the two studies that have been conducted in the general population; Cornish et al. (
2005) in an unselected general population sample studied with a different measure of inhibition and Friedman et al. (
2008) in a twin sample studied with the SST. In the later sample as in the current study, there was evidence of shared genetic risk between a composite measure of inhibitory control and impulsiveness (Young et al.
2009).
Older participants exhibited better response inhibition than did younger participants as has been found in most (Williams et al.
1999) but not all (Friedman et al.
2008; Young et al.
2009) studies of inhibition and other executive functions. Older participants were also faster and less variable. Males were significantly although minimally better inhibitors and were faster and less variable in their responses than were girls, but the male advantage was not evident in older participants. There was no interaction between gender and ADHD trait scores indicating that gender effects were similar for those with high and low ADHD traits. Neither ethnicity nor socioeconomic status affected SST performance significantly. Consistent with prevalence rates for the disorder, ADHD traits were more marked among males. In contrast to the supposed increase prevalence of inattentive subtype of ADHD in females, we found that males had higher scores for both inattention and hyperactivity-impulsivity as measured by the SWAN.
In contrast to the results of twin studies, age had no significant effect on heritability of the cognitive measures studied here (c.f. Friedman et al.
2008). The heritability of ADHD traits as measured by the SWAN is comparable to some (Hay et al.
2007) but not other (Polderman et al.
2007) twin studies and is lower than typically reported using other measures of ADHD. Hay et al. (
2007) speculated that the increased rating options available in the SWAN afford a greater opportunity than other rating scales for informants to rate their siblings as similar.
The study revealed details of heritabilities of ADHD subcomponents. Both inattention and hyperactivity-impulsivity were associated with inhibition, latency and variability. Both dimensions were heritable even after controlling for age and sex and the two trait dimensions shared substantial genetic risk as has been found in previous twin studies (Hay et al.
2007; Polderman et al.
2007; Swanson et al.
2001). Both dimensions share genetic risk with response inhibition. The genetic risks that are shared by hyperactivity-impulsivity and response latency seem to increase the behavioral trait and decrease response latency.
One limitation of evaluating executive function in the general population is that many participants took medication for ADHD within 48 h of participation in the study. On one hand, these observations support the conclusion that the community sample was typical of the population thereby enhancing the generality of our observations and conclusions. On the other hand, stimulants are known to improve response inhibition (Tannock et al.
1995) raising the possibility that our data could be censored at the high end of the distribution. The problem of studying disorders in the general population with a substantial proportion under treatment for the disorder of interest is common to many disorders (Tobin et al.
2005). The effect of therapy in epidemiological studies could distort trait scores causing reduction in the estimated effect of etiological determinants and a marked loss in statistical power. Various strategies have been used to correct for this effect such as exclusion of treated individuals and use of treatment as a covariate. We addressed this potential confound with what is thought to be the optimal strategy, namely by estimating the effect of stimulant treatment on inhibition from a prior study in order to introduce a correction among medicated individuals, thereby reducing the bias arising from treatment or from their exclusion (Rice et al.
2009; Tobin et al.
2005). We applied this correction for medication effect to the scores of the 414 medicated participants based on published studies that showed a 0.75 effect size effect of stimulant medication (Tannock et al.
1995). Multivariate model parameters and heritability estimates were unaltered when models were rerun following this correction. This type of correction allows for the inclusion of treated participants in genetic studies.
Although study of individuals in the general population is a powerful strategy for the cost-effective collection of samples for cognitive and genetic research, it is important to note that it is far more difficult to collect data on multiple cognitive measures in this setting. Friedman et al. (
2008) has argued that combining multiple cognitive measures into a latent factor can increase the heritability of clusters of executive control measures by minimizing the effect of variance not attributable to the particular executive control construct. They argued that latent factor scores might be even more powerful in genetic studies than any single measure.
Another limitation of using cognitive tasks to measure ADHD endophenotypes is evident in the participants who had to be excluded because of invalid stop task performance. Many of these participants had invalid performance because of administration errors such as pushing the wrong buttons on the game pad device used to collect responses or leaving the task before it was complete because their family wished to continue their visit to the Science Centre. These errors may have been minimized through more intensive participant supervision, although it is unlikely that they can be eliminated totally in a busy general population setting. However, a portion of invalid stop task data is non-random. Individuals with invalid performance had significantly higher ADHD trait scores suggesting a systematic bias, which would serve to exclude those participants of greatest interest. These participants were also younger and more likely to be males. Given knowledge of participants’ age, gender and ADHD trait scores, it would be possible to estimate their performance using a regression model. Although the distribution of household income among Science Centre visitors was skewed toward higher incomes compared to the surrounding community, social class did not make a significant contribution to models of inhibition, latency, or variability or to ADHD traits.
The current results support the strategy of using endophenotypes in a general population for selection of individuals at high and low genetic risk for ADHD for genetic study. The “extreme trait” (also known as “selective genotyping”) approach using endophenotypes is a particularly powerful and cost-effective design for detecting genetic association (Huang and Lin
2007; Van Gestel et al.
2000). The approach has been successful in studies of QT interval (Arking et al.
2006), obesity (Herbert et al.
2006), foetal haemoglobin level (Menzel and Thein
2009), and in one previous candidate gene study for ADHD traits (Cornish et al.
2005). Another advantage of this approach is increased ease of collection due to the ready availability of general population samples compared to clinically ascertained subjects. We estimated that the cost of accruing participants from the general population was 10 % of the per subject cost in clinic samples.
Based on the current results and on previously published research, we conclude that inhibition is a heritable trait which is sensitive to variation across the range of ADHD traits and shares genetic risk with ADHD traits. Identification of genes involved in risk for inhibition may be a powerful way to identify ADHD related genetic risks.
Acknowledgments
This research was funded by grants from Canadian Institutes of Health Research MOP 64277 (RS), 44070 (RS), 74699 (RS, PA, JC, AP). The authors would like to thank the staff and visitors of the Ontario Science Centre, Toronto, Canada and 20 enthusiastic summer students who collected the data. RS holds Toronto Dominion Bank Chair in Child and Adolescent Psychiatry.
Disclosures
Dr. Schachar receives funding from the Canadian Institutes of Health Research, Ontario Research Fund, and the Ontario Brain Institute and has consulted to Eli Lilly, Purdue Pharma, and Highland Therapeutics.
Dr. Crosbie has received funding from the Canadian Institutes of Health Research and the Ontario Mental Health Foundation; and reports no biomedical financial interests or potential conflicts of interest.
Dr. Arnold has received funding from the Canadian Institutes of Health Research, the National Institutes of Health, and the Ontario Research Fund; and reports no biomedical financial interests or potential conflicts of interest.
Dr. Paterson has received funding from the Canadian Foundation for Innovation, the Juvenile Diabetes Research Foundation, the Canadian Institutes of Health Research, Genome Canada, the Ontario Research Fund, Autism Speaks, the Medical Research Council (UK), the Irish Health Research Board, the Canadian Foundation for Innovation, the National Cancer Institute of Canada, the Canadian Breast Cancer Research Alliance, and the National Institutes of Health; has consulted to Millennium Pharmaceuticals and Hemax Genome.
Dr. Swanson has received research support from Alza, Richwood, Shire, Celgene, Novartis, Celltech, Gliatech, Cephalon, Watson, CIBA, Janssen, and McNeil; has been on the advisory board for Alza, Richwood, Shire, Celgene, Novartis, Celltech, UCB, Gliatech, Cephalon, McNeil, Lilly, and Noven; has been on the speaker’s bureau for Alza, Shire, Novartis, Celltech, UCB, Cephalon, CIBA, Janssen, and NcNeil; and has consulted to Alza, Richwood, Shire, Celgene, Novartis, Celltech, UCB, Gliatech, Cephalon, Watson, CIBA, Janssen, McNeil, Lilly, and Noven.
Dr. Dupuis has received research funding from Institute of Educational Sciences, Social Sciences and Humanities Research Council of Canada, Autism Speaks, and Alva Foundation.
Dr. Strug has received research funding from Canadian Institutes of Health Research, the Ontario Ministry of Research and Innovation Early Researchers Award, and the Natural Sciences and Engineering Research Council of Canada.