Editorial Perspective: Why is there such a mismatch between traditional heritability estimates and molecular genetic findings for behavioural traits?

Traditional family, twin and adoption studies have shown consistently that psychopathology and cognitive traits are familial and heritable. High heritability estimates have led to the conclusion that genetic factors make a substantial contribution to risk of many mental disorders and have fuelled the search for DNA variants associated with psychopathology and cognitive traits. Molecular genetic studies of behavioural traits and psychopathology involve direct assessment of variation across the whole genome using thousands of genetic markers (single nucleotide polymorphisms, SNPs). A new method of analysis assesses heritability that is attributable to the measured SNPs. Recent genome-wide association (GWA) analyses have provoked debate because they generate ‘SNP heritability’ estimates for behavioural traits that appear to be dramatically lower than those observed for other traits and when using traditional designs; although SNP heritability (h) and traditional h are not the same. Findings from a recent twin-based molecular genetic study of cognitive and behavioural traits (Trzaskowski, Dale, & Plomin, 2013) were especially striking because it was possible to directly contrast traditional twin and SNP heritability estimates in the same sample, as well as compare estimates for behavioural, cognitive and physical measures. SNP heritability estimates were extremely low for behavioural traits, notably much lower than those for cognitive ability, height and weight in the same sample and very much lower than the twin-based heritabilities. The findings imply that the low estimates for the behavioural traits reflected something different about these traits rather than a problem with the experimental design. The puzzle of the disparity between molecularand traditional behaviour genetic study findings (Vinkhuyzen, Wray, Yang, Goddard, & Visscher, 2013; see also Wray et al. (2014) in this issue for explanations) has prompted widespread discussion. Fundamental questions have been raised across the whole field of complex genetic traits as well as for behavioural traits. Did traditional approaches, such as twin studies, in fact generate misleadingly high heritability estimates for psychopathology? Do genome-wide molecular genetic techniques have value in our field? It has been questioned whether genetic contributions at a molecular level are in fact trivial for psychopathology and behaviour. Traditional behaviour genetic approaches statistically infer familial risk and heritability from observed patterns of phenotypic similarities between relatives who differ in their degree of biological relatedness (e.g. monozygotic and dizygotic twins, adoptive and biological parents’ of children). Whilst there are a number of important assumptions and limitations inherent in these designs (Tenesa & Haley, 2013), findings of consistently high h fuelled the impetus to identify genetic risk factors at a molecular level. Such investigations involve direct assessment of DNA variation (i.e., gene variants) and typically have exploited case–control comparisons or the correlation of DNA variation with trait variation in a population. How do GWA studies assess genetic contributions and heritability? The design involves assessing hundreds of thousands of DNA markers that lie across the genome. Such studies typically assess only one class of DNA marker or variation, a common frequency single nucleotide (single nucleotide polymorphism, SNP). Variation in the whole genome is not measured; this is important as there are many other classes of DNA variation. Also, the measured SNPs should be viewed as risk markers because they are selected only to ‘tag’ or provide an indirect index of DNA variation that will include the truly causal SNPs. Whole genome molecular genetic designs have utilised three key approaches [see Wray et al. (2014) for a scientific account]. First, a key aim of these GWA studies (GWAS) was to identify single SNPs associated with disorders or traits. As adjustment for multiple testing is needed, and we have now learnt the effect size of any single common SNP is small, extremely large sample sizes (tens of thousands) are required to achieve genome-wide levels of statistical significance. A second, alternative approach has involved generating aggregate SNP risk scores known as genomic profile risk or polygenic scores that include individual SNPs which, may individually, fall below the stringent threshold for genome-wide significance.