This study confirms and expands previous investigations suggesting individual genes moderate response variability in ADHD treatment. Unlike most previous reports, we prospectively examined an expanded set of candidate genes, used blinded ratings of both efficacy and side effects, assessed responses over a range of doses, analyzed statistically derived components as dependent variables rather than arbitrarily choosing outcome measures, and defined genotypes empirically based on allele frequencies without a priori consideration of potential effects. While recognizing that sample sizes for certain genotypes are too small to sustain population inferences, the overall study sample is larger than a majority of currently published ADHD pharmacogenetic studies. Apart from any significant findings, study methods described provide a strong basis for future investigations.
For efficacy outcomes, our most striking finding is the observation that differences in response profiles across genotypes, i.e. gene × dose interactions, are potentially more significant than genotype effects alone. Conflicting results in prior reports might be due in part to failures to titrate medication over a sufficient range to elicit differences in dose-response by genotype. Earlier investigations described both linear50
stimulant dose-response profiles. Our results, while requiring replication, suggest that genotype might moderate dose-response outcomes depending on the specific gene and outcome of interest. Assessing response over a full range of doses and consideration of both gene effects and gene × dose interactions are essential for future ADHD pharmacogenetic studies.
Selection of appropriate outcome measures remains a major issue. Correlations between various ADHD efficacy measures are notably weak,52
and there is particular risk for spurious findings in pharmacogenetic studies of outcome variables selected post hoc
. Our data reduction strategy allowed us to select outcome measures objectively and reduced the risk of false positive findings from multiple exploratory tests. Although our primary results did not meet a level for genome wide significance (p≤1×10-8
), the high prior probability that selected candidates are involved in stimulant medication response and the strength of demonstrated associations support the validity of our findings. Interestingly, our two component solution for efficacy outcomes is consistent with earlier observations describing two domains of ADHD treatment response, namely learning performance and behavior.51
Our results support the idea that ADHD outcomes are not unitary, but reflect multiple domains of functioning. Furthermore, our data suggest that different genes differentially influence response on these separate domains.
Our results are largely consistent with previous speculations about the neurobiological significance of these genes. The possibility remains that selected candidates are merely linked with other genes that influence medication response, or are interacting with other unidentified genes to yield observed findings. As such, attempts to provide mechanistic explanations significant effects are speculative.
One prior study of DRD4VNTR
reported that subjects with the 7-repeat required higher doses to respond on overt ADHD behaviors.18
The apparent deterioration on Math Problems Correct
at higher doses in subjects without the homozygous +4/+4 genotype is consistent with older observations that the higher stimulant doses required to control behavior lead to concomitant impairments in learning measures.51,53
For the SLC6A4VNTR
, findings that individuals with the -12/-12 genotype are less sensitive to symptom improvement are consistent with one report describing increased euphoria from dextro-amphetamine.43
The poorer Math
response at higher doses in individuals with the +L allele at 5HTTLPR
is consistent with knowledge of the interactions between serotonergic and dopaminergic systems, where serotonin is generally found to constrain dopaminergic signaling.26
Although our results only demonstrated trend effects for COMT
, several previous reports described a potential role in moderating medication response in both depression and schizophrenia.54,55
Our lack of significant findings with SLC6A3
, in spite of overall power to detect small effects, reflects the inconsistencies in earlier reports and provides further evidence suggesting that this gene plays less of a role in treatment response than previously proposed.
Examination of genetic moderators of tolerability might ultimately prove more useful than predictions of efficacy.1,3,21
Side effects findings for the DRD4PROM
are consistent with previous work.21
However, relative to investigations of symptom reduction, optimal approaches to assess genetic contributions to side effects are much less developed. For example, the current report describes a brief titration trial and used a low severity threshold to generate sufficient side effect frequencies for reasonable analysis. More clinically relevant information would be derived from a longer term study in larger numbers of patients. Long term investigations of potential genetic moderators of side effects such as growth deceleration, appetite loss, irritability, and tic emergence would, if positive and replicated, have significant impact on clinical practice.
There is an increasing emphasis in ADHD pharmacogenetics on estimating gene effect sizes,1,19-22
often based on Cohen's d
. While we view the estimation of gene effects as laudable, we are now of the opinion that using Cohen's d
for this purpose is a misapplication of that statistic that leads to an overestimation of genetic effects. Cohen's d
generally serves to estimate the differences in effect size between two treatments.49
However, given the large stimulant treatment effects demonstrated in ADHD, effects of medication dose are likely to outweigh any genetic effects. The proper question is not whether different genotypes should be considered as different treatments, but how much variance in treatment outcome is explained by genetic as opposed to medication effects. Cohen f2
provides a signal to noise ratio, indicating how much variability is attributable to the model's fixed effects and thus provides a dimensionless measure of the magnitude of gene and gene × dose signals.49
Based on convention, whereby small effects are defined by f2
>.02, moderate effects as >.15, and large effects >.35, none of the statistically significant effects demonstrated in primary analyses are likely to represent more than small clinical effects. This is consistent with results from meta-analyses of these same candidate genes for ADHD risk.8,10
Unlike the majority of published ADHD pharmacogenetic studies, we adjusted significance levels for multiple efficacy outcomes based on the Bonferroni correction. We regarded tests of each genotype as independent, but chose to report these within a single study rather than describe each finding separately. Nonetheless, some might argue that a stricter correction should have been applied, i.e. that further corrections should have been made when examining two polymorphisms within a single gene, or that testing two outcomes for each of nine genotypes suggests adjusting for eighteen tests (p<.003). Although this pilot investigation was underpowered to detect small effects at this very conservative level of significance, the fact that over 20% of primary outcomes were positive at p<.025 is highly suggestive of true findings. Interestingly, our most robust suggested finding was in the restricted sample of white/non-Hispanic subjects which demonstrated a medium sized effect of DRD4PROM × dose on Math Problems Correct, at a significance level corrected for 100 tests (p<.0005). This highlights the need to conduct future research in more homogeneous samples.
This study has several limitations. Although there is increased probability that candidate genes in the catecholamine system are involved in methylphenidate response, candidate gene investigations have high false positive rates. Our results require replication in independent samples prior to any consideration of clinical relevance. Although the majority of our positive findings remained significant when the sample was restricted to whites only, we were unable, given the size and heterogeneity of the sample, to make further assessments of potential population stratification. We did not correct for comparison of multiple polymorphisms within single genes, but treated these as independent tests. This might prove meaningful as our four positive findings are found within two genes. Future research should consider the possibility of linkage disequilibrium between these variants. While we attempted to reduce problems associated with multiple testing, our decision to define genotypes a priori
based on allele frequencies negated our ability to assess other previously described genotypes of interest, such as the homozygous 9-repeat allele at SLC6A3
, or the 7-repeat allele at DRD4
as primary dependent variables. However, it may prove advantageous to assess the role of these alleles directly after removing minor alleles in future studies of larger samples. Our work did not consider recent findings suggesting that 5HTTLPR
is functionally triallelic, with the +L polymorphism exhibiting a G versus A substitution with nearly equivalent expression as the S allele.56
However, whereas this variant was not reported until after we completed genotyping and was not considered in earlier studies involving this deletion/insertion polymorphism, full elucidation of the role of polymorphisms at 5HTTLPR
will require ongoing examination. Finally, we did not explore potential gene × gene interactions as these would have required additional sets of hypotheses and analyses. These are more suitable for a separate report.
Our investigation assessed short-term outcomes. Large scale studies document that short-term response rates, cited as high as 75-80%, often fall to 55-60% over time. Short-term data are incomplete assessments of treatment response. Long-term outcomes might prove to be much more clinically relevant for investigation. Ultimately, definitive findings in ADHD pharmacogenetics are likely to depend on a combination of candidate gene and genome wide association approaches with extended treatment outcomes in much larger samples.