Child behaviour problems show biggest gap between DNA-based and twin heritability

Behavioural problems in childhood, such as anxiety, depression, hyperactivity, and conduct disorder are common. In the US, the likelihood of having one or more such problems is greater than 1 in 10. Our recent study asked to what extent individual differences in behaviour problems can be explained by differences in people’s DNA. Here, Rosa [EditLab PhD student] summarises the findings and their implications.



Consider for a moment your own childhood behaviour and that of your friends — how anxious or restless you were. There are two main sources of individual differences in behaviour: genetic and environmental differences.


For decades, twin studies have offered an elegant method for investigating the relative roles of nature and nurture. We compare the behavioural similarity of identical twins, who share 100% of their genes, with the similarity of non-identical twins, who share on average 50%. From this comparison, we get an estimate of heritability — i.e., the proportion of variation in a trait (e.g. anxiety) that can be explained by inherited genetic differences. Heritability relates to population-level variation, not to individualsTwin studies have consistently confirmed that childhood behaviour problems are under significant genetic influence. Heritability estimates for autistic-like traits and hyperactivity range from about 40% for anxiety and depression to >60% .



“Twin studies have consistently confirmed that childhood behaviour problems are under significant genetic influence”

Great technological advances mean that there is now another method for estimating the importance of genetics. We can now directly examine individuals’ DNA variation and estimate heritability from measured genetic variants (this is called ‘SNP heritability’ — Single Nucleotide Polymorphism heritability).

In this study, using data from the Twins Early Development Study, we estimated SNP heritabilities of a variety of childhood behaviour problems, as rated by parents, teachers, and children themselves at ages 12 and 16. Then, we compared these with heritability estimates from the traditional twin method.

SNP heritability: the method

Unrelated individuals differ slightly genetically, and this is mainly due to chance. They also differ in their trait similarity – i.e., how similar they are in their levels of behaviour problems.

To estimate SNP heritability, we take a sample of thousands of unrelated people (in our sample, we took one twin from each pair). Then, we calculate the genetic similarity of all possible pairs of people, across millions of common DNA markers. This leaves us with an enormous genetic relatedness matrix (the left-hand box below). We also calculate the similarity of all of the pairs for childhood behaviour problems, giving a phenotypic (trait) relatedness matrix (the right-hand box). Next, we see if DNA-based similarity across common markers predicts trait similarity. I.e. How much does the left-hand genetic matrix predict the right-hand trait matrix? This method gives us the proportion of variance in each trait that can be explained by common measured DNA markers. This is different from twin heritability, which is the proportion of variance explained by any inherited genetic variants.

Screen Shot 2017-12-05 at 11.10.20


Results from our study

“This research has important implications for how research about childhood behaviour problems will benefit from the DNA revolution”

We found low, largely non-significant DNA-based heritability estimates (6% on average) for childhood behaviour problems. In contrast twin heritabilities were high (52% on average), as expected from the literature. The SNP-twin heritability gap was consistent across 37 measures of childhood behaviour problems as assessed by different reporters and at different ages.

Screen Shot 2017-12-05 at 11.16.25

Importantly, we expect SNP heritability to be lower than twin heritability for all complex traits. This is because it only looks at the effects of the common measured DNA markers detectable by current technology, rather than the effects of any DNA differences between people (including e.g. rare variants). Even so, the gap is strikingly large. How could this be? The fact that our twin heritability estimates are similar to those reported in the literature suggests that our measures are adequate. SNP-based estimates for height, weight, and intelligence were, as expected, half those of twin-based estimates, which demonstrates that the problem is not due to methodological problems, but seems to be specific to behaviour problems as distinct from physical and cognitive traits.


This research has important implications for how research about childhood behaviour problems will benefit from the DNA revolution. By ‘DNA revolution’, we mean the efforts to find specific genetic markers involved in the development of behaviour problems in childhood, and to predict individual differences in behaviour using DNA.

For the first point, studies can test whether specific genetic variations are more associated with quantitative variation in a trait. This method has exactly the same limitations as our DNA heritability estimates to the additive effects of common genetic markers. This means that, if DNA heritability of childhood behaviour problems is truly as low as 6%, we will struggle to find specific genetic variants involved in their development.

Secondly, we know that the effects of genetic markers on behaviour are small individually, but we can add them together like items on a questionnaire into a score for each person (polygenic score). We can then use these polygenic scores to predict the onset of childhood behaviour problems. This is an exciting advance towards personalised medicine, and even prevention, which is especially important given the young ages of onset of behaviour problems. However, research into child mental health cannot join this trend if the heritability based on the SNPs we are measuring is close to zero.


“we need to build larger samples and to think about how we measure behaviour problems”

Luckily, there are numerous areas for further research to help achieve these goals. Importantly, notwithstanding this study, we can assume that the genetic basis of behaviour problems is worth investigation, based on decades of twin research.

Future research should continue to investigate the reasons why SNP heritabilities for childhood behaviour problems are so low compared with twin estimates. We are looking at other mechanisms, especially involving genetic markers not well captured in the set of markers that we measure, and interplay with the environment.

We should also try to maximise SNP heritability before performing gene-finding and prediction studies. Crucially, we need to build larger samples to increase our statistical power to detect SNP heritability. We also need to think carefully about how we measure behaviour problems in the first place.



Watch this ‘TEDS talk‘ to hear me explain our findings in person.



Author Rosa Cheesman

More posts by Rosa Cheesman

Join the discussion One Comment