EDITORIAL Attention-Deficit/Hyperactivity Disorder Is the Extreme and Impairing Tail of a Continuum Philip Asherson,
MD, AND
F
amily, twin, and adoption studies converge to show that attention-deficit/hyperactivity disorder (ADHD) is a familial disorder influenced by genetic factors.1 However, the distinction of ADHD from the normal range remains controversial. What are the boundaries of the condition and how do we make the distinction between ADHD and normal, healthy behavior? At the heart of this debate is the evidence from genetic epidemiology that ADHD reflects the extreme of one or more traits that are continuously distributed throughout the population.2 Nearly all twin studies of ADHD have estimated heritability from within twin-pair correlations for continuous measurements of ADHD symptoms in samples with very few clinical cases. These studies establish that heritability (the proportion of phenotypic variance explained by additive genetic factors) is approximately 70% to 80% for ADHD trait scores. It is widely assumed that the same genetic influences also increase the risk for the clinical disorder. Earlier studies estimated similar high heritability from the differential regression of ADHD symptoms to the population mean among the co-twins of twin probands selected for high ADHD symptom scores, suggesting continuity of genetic influences between the extreme “clinical” group and the normal range.3 Another study showed that the siblings without ADHD of children with ADHD had higher ADHD symptom scores than population norms.2 Moreover, twin concordance rates for diagnosed cases found similar high heritabilities.4 Advances enabling the investigation of common genetic variation across the genome provide alternative approaches to genetic epidemiology. Genomewide Complex Trait Analysis applies a mixed linear model (GREML) to estimate heritability (h2GREML) from genomewide single-nucleotide polymorphism data of the relatedness of genotype to phenotype across all possible pairs in large samples of unrelated individuals. This has advantages over twin studies by using singleton series, providing information on the genetic effects of measured single-nucleotide polymorphisms and not depending on assumptions required for twin modeling. Analysis of complex traits finds that h2GREML estimates are usually approximately 50% of twin heritability (h2TWIN). This raises questions about missing heritability but demonstrates the critical role of common genetic variation for numerous phenotypes. For ADHD and other psychiatric disorders (schizophrenia, bipolar disorder, depression, and autism), h2GREML estimates are in keeping with this general rule; the h2GREML estimate for ADHD is approximately 0.3 to 0.4.5,6 However, for ADHD trait scores and other behavioral phenotypes measured in population samples, there is JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY VOLUME 54 NUMBER 4 APRIL 2015
Maciej Trzaskowski,
PhD
a discrepancy. The expected 50% of twin heritability was not detected in a sample of approximately 2,700 representative UK twins,7 suggesting a different genetic architecture for ADHD and ADHD traits. In this issue of the Journal, Stergiakouli et al.8 used the polygenic score method (PSM) to investigate the genetic relation between trait scores in a population sample and the clinical disorder. PSM evaluates the degree to which genetic variation associated with a disorder or trait in a discovery sample predicts a disorder or trait in an independent test sample. For ADHD, PSM was previously used to show that DSM-IV ADHD is associated with common genetic variation and that polygenic signals derived from ADHD case-control comparisons predict ADHD trait scores.9 In their work, Stergiakouli et al. demonstrate that polygenic signals associated with ADHD symptoms in a general population sample predict case-control status and symptom severity in a sample of children with DSM-IV ADHD. This finding supports the view that ADHD reflects the extreme and impairing tail of at least 1 continuously distributed trait. However, the findings do not address the key question raised by the h2GREML estimates, which suggest a difference in genetic architecture underlying ADHD trait scores and DSM-IV ADHD. The critical point is that h2GREML estimates the total contribution to phenotypic variation of additive genetic effects captured by single-nucleotide polymorphisms, whereas PSM estimates the strength of the association between a polygenic predictor and the phenotype. Theoretically, the 2 estimates merge with increasing sample size of the PSM discovery dataset. However, the polygenic predictor derived from the population sample in this study explained only a tiny increase in risk for ADHD (odds ratio w1.17) and variance in symptom severity (R2 w0.01). Therefore, we can conclude with a high degree of certainty only that additive genetic effects from population trait scores confer a small increase in risk for ADHD. Based on the conditional analyses included in the article, this could be further restricted to trait scores for inattention. Another point to clarify is that there is no inconsistency between the PSM and h2GREML findings. Although the earlier article had the provocative title of “No Genetic Influence for Childhood Behavior Problems from DNA Analysis,”7 this must be put into context. This study made the reasonable assumption based on results for height, weight, cognitive variables, and major psychiatric disorders that h2GREML is approximately 50% of twin heritability. However, this was not the case for behavioral traits measured in a population twin sample. Specifically, the estimated twin heritability for ADHD symptoms in the same sample was approximately
www.jaacap.org
249
ASHERSON AND TRZASKOWSKI
0.80, and there was 90% power to detect h2GREML of 0.40, which was not found. However, if true heritability is closer to 0.50, then the sample had only 56% power to detect the expected h2GREML of 0.25 (and very low power for h2GREML
MD, AND
F
amily, twin, and adoption studies converge to show that attention-deficit/hyperactivity disorder (ADHD) is a familial disorder influenced by genetic factors.1 However, the distinction of ADHD from the normal range remains controversial. What are the boundaries of the condition and how do we make the distinction between ADHD and normal, healthy behavior? At the heart of this debate is the evidence from genetic epidemiology that ADHD reflects the extreme of one or more traits that are continuously distributed throughout the population.2 Nearly all twin studies of ADHD have estimated heritability from within twin-pair correlations for continuous measurements of ADHD symptoms in samples with very few clinical cases. These studies establish that heritability (the proportion of phenotypic variance explained by additive genetic factors) is approximately 70% to 80% for ADHD trait scores. It is widely assumed that the same genetic influences also increase the risk for the clinical disorder. Earlier studies estimated similar high heritability from the differential regression of ADHD symptoms to the population mean among the co-twins of twin probands selected for high ADHD symptom scores, suggesting continuity of genetic influences between the extreme “clinical” group and the normal range.3 Another study showed that the siblings without ADHD of children with ADHD had higher ADHD symptom scores than population norms.2 Moreover, twin concordance rates for diagnosed cases found similar high heritabilities.4 Advances enabling the investigation of common genetic variation across the genome provide alternative approaches to genetic epidemiology. Genomewide Complex Trait Analysis applies a mixed linear model (GREML) to estimate heritability (h2GREML) from genomewide single-nucleotide polymorphism data of the relatedness of genotype to phenotype across all possible pairs in large samples of unrelated individuals. This has advantages over twin studies by using singleton series, providing information on the genetic effects of measured single-nucleotide polymorphisms and not depending on assumptions required for twin modeling. Analysis of complex traits finds that h2GREML estimates are usually approximately 50% of twin heritability (h2TWIN). This raises questions about missing heritability but demonstrates the critical role of common genetic variation for numerous phenotypes. For ADHD and other psychiatric disorders (schizophrenia, bipolar disorder, depression, and autism), h2GREML estimates are in keeping with this general rule; the h2GREML estimate for ADHD is approximately 0.3 to 0.4.5,6 However, for ADHD trait scores and other behavioral phenotypes measured in population samples, there is JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY VOLUME 54 NUMBER 4 APRIL 2015
Maciej Trzaskowski,
PhD
a discrepancy. The expected 50% of twin heritability was not detected in a sample of approximately 2,700 representative UK twins,7 suggesting a different genetic architecture for ADHD and ADHD traits. In this issue of the Journal, Stergiakouli et al.8 used the polygenic score method (PSM) to investigate the genetic relation between trait scores in a population sample and the clinical disorder. PSM evaluates the degree to which genetic variation associated with a disorder or trait in a discovery sample predicts a disorder or trait in an independent test sample. For ADHD, PSM was previously used to show that DSM-IV ADHD is associated with common genetic variation and that polygenic signals derived from ADHD case-control comparisons predict ADHD trait scores.9 In their work, Stergiakouli et al. demonstrate that polygenic signals associated with ADHD symptoms in a general population sample predict case-control status and symptom severity in a sample of children with DSM-IV ADHD. This finding supports the view that ADHD reflects the extreme and impairing tail of at least 1 continuously distributed trait. However, the findings do not address the key question raised by the h2GREML estimates, which suggest a difference in genetic architecture underlying ADHD trait scores and DSM-IV ADHD. The critical point is that h2GREML estimates the total contribution to phenotypic variation of additive genetic effects captured by single-nucleotide polymorphisms, whereas PSM estimates the strength of the association between a polygenic predictor and the phenotype. Theoretically, the 2 estimates merge with increasing sample size of the PSM discovery dataset. However, the polygenic predictor derived from the population sample in this study explained only a tiny increase in risk for ADHD (odds ratio w1.17) and variance in symptom severity (R2 w0.01). Therefore, we can conclude with a high degree of certainty only that additive genetic effects from population trait scores confer a small increase in risk for ADHD. Based on the conditional analyses included in the article, this could be further restricted to trait scores for inattention. Another point to clarify is that there is no inconsistency between the PSM and h2GREML findings. Although the earlier article had the provocative title of “No Genetic Influence for Childhood Behavior Problems from DNA Analysis,”7 this must be put into context. This study made the reasonable assumption based on results for height, weight, cognitive variables, and major psychiatric disorders that h2GREML is approximately 50% of twin heritability. However, this was not the case for behavioral traits measured in a population twin sample. Specifically, the estimated twin heritability for ADHD symptoms in the same sample was approximately
www.jaacap.org
249
ASHERSON AND TRZASKOWSKI
0.80, and there was 90% power to detect h2GREML of 0.40, which was not found. However, if true heritability is closer to 0.50, then the sample had only 56% power to detect the expected h2GREML of 0.25 (and very low power for h2GREML
