M for Missing Heritability

By July 11, 2018 A-Z No Comments

Decades of twin studies have yielded evidence of the heritability of many different traits. What we mean by this is the proportion of variation in a trait that can be explained by genetic differences between individuals. Research has moved towards identifying specific genetic variants associated with these traits through Genome-Wide Association Studies (GWAS). However, across most psychiatric phenotypes studied by geneticists, individual genetic variants identified by GWAS do not account for much of the heritability. This gap has been termed ‘Missing Heritability’. This A-Z blog goes through some of the possible reasons for this gap.


When the human genome was sequenced in 2003, there was a great deal of hope that individual genetic variants accounting for the long-established heritability of traits would be identified. Although this breakthrough did instigate the discovery of many variants with significant effects, the amount of heritability accounted for by these GWAS ‘hits’ was much less than that estimated by twin studies. The discrepancy between GWAS and twin estimates was named ‘missing heritability’ (Maher, 2008).

“The discrepancy between GWAS and twin estimates was named ‘missing heritability’.”

Some have taken ‘missing heritability’ to mean that no genetic variants accounting for variability have been found. In fact, thousands have been found, but not all. The lack of large effects of individual variants adds to the evidence that most diseases and traits cannot be pinned down to single genetic mutations, but instead result from the cumulative impact of many genetic variants of small effect, as well as environmental factors. It does not mean they don’t exist, just that there are issues we need to overcome in order to identify them. We will illustrate some of these issues and solutions that have been implemented to tackle them using a notoriously difficult phenotype in the missing heritability literature: depression.

Sample Size

Since the effects of the variants influencing depression are extremely small, a great deal of statistical power is required to detect them. Power can be increased in several ways, one of which is to use a larger sample size. Thanks to open source data and increasing collaboration between research groups, sample sizes have continually improved. Only 5 years ago, the largest study of depression (a mega-analysis with 9,000 cases and 9,000 controls; Ripke et al., 2013) failed to identify any genetic variants with genome-wide significant effects on depression. The most recent study by the same collaborative organisation – the Psychiatric Genomics Consortium (PGC) – had an astounding ~130,000 cases and ~330,000 controls, and identified 44 variants, accounting for ~9% depression heritability (Wray et al., 2018).

Phenotypic Heterogeneity and Measurement

Depression is a highly heterogeneous disorder: individuals with a diagnosis can experience very different symptoms. For example, one individual might report insomnia whilst another sleeps excessively. It is plausible that unique combinations of variants predict different patterns of symptoms. Therefore, when performing GWAS on individuals with depression, genetic variants associated with specific symptoms may be missed if the sample contains symptomatically heterogeneous individuals. The CONVERGE Consortium (2015) demonstrated the power of reducing heterogeneity by recruiting only Chinese females with severe, recurrent depression; despite ‘only’ 5,000 cases, two genome-wide significant loci were identified.

“…genetic variants associated with specific symptoms may be missed if the sample contains symptomatically heterogeneous individuals.”

Related to this is the way in which we define and measure the phenotype of interest. Psychiatric illnesses like depression are much harder to measure than a directly observable phenotype such as height; they largely rely on subjective self-report data. This therefore plays a role in the reliability and validity of a GWAS. For more detail on how measurement problems might affect the missing heritability for childhood psychopathology, see Rosa’s previous blog.

Rare Variants

The ‘common disease-common variant’ hypothesis is the prevailing view that has powered GWAS. This hypothesis states that the alleles that underlie susceptibility to most common diseases are commonly found in the population, but individuals with those alleles often do not show clinical symptoms. To put it simply, this means that these common alleles are ‘weak’ and having these variants does not necessarily mean one will develop the disease. However, while GWAS have been critical in our current understanding of the genetic basis of depression, it is likely that rare variants also have a role to play but are not yet being examined within most GWAS. Investigating both rare and common variants has been found to improve predictive accuracy for objectively observable phenotypes, with Yang et al. (2015) identifying an increase in DNA-based heritability estimates for height from 45% to 56% by using a technique called imputation to look at a larger number of variants across the genome.

“…it is likely that rare variants also have a role to play but are not yet being examined within most GWAS.”

Additionally, a recent study from Yu et al. (2018) used an approach called ‘Hamming distance’ to measure the genetic distance between two individuals. Using this novel computational and statistical strategy, they were able to show that low frequency and rare variants are likely to play a significant role in the genetics of depression thus should be a focus for future research. As next-generation sequencing technologies improve, the extent to which rare variants contribute to missing heritability should become clearer.

Twin estimates

Although controversial, it has been posited that missing heritability is partly attributable to problems with how the twin method calculates heritability estimates. These estimates are derived by comparing trait resemblance of monozygotic (MZ, identical) twin pairs reared together to dizygotic (DZ, fraternal) twin pairs reared together. This method relies on several key theoretical assumptions, the most controversial being that MZ and DZ twins grow up experiencing roughly equal environments, otherwise known as the ‘equal-environment assumption’.

Twin studies have drawn criticism regarding the equal-environment assumption for the past 40 years. For example, there is evidence that MZ twin pairs are treated more alike than DZ twin pairs (Loehlin & Nichols, 1976). If MZ twins are treated more similarly, MZ correlations will be inflated compared to DZ correlations, thus genetic influence may be overestimated for the trait of interest. However, increased similarity in treatment of MZ twin pairs may also occur as a result of MZ twins evoking more similar responses from their environment as a result of their genetic identity (see Thalia’s blog post for more information on evocative gene-environment correlation). These differences would therefore be genetic, not environmental, and as such the equal-environment assumption would not be violated (Eaves, Foley & Silberg, 2003).

Most investigations of the equal-environment assumption have, however, failed to find that this assumption is violated to the extent that heritability estimates are substantially biased by it (Kendler et al., 1994). A further criticism of the twin design suggests that it fails to capture the complexity of gene-environment interplay, which may further bias heritability estimates.

Other potential explanations not detailed here include the difficulty in accounting for the complex interaction effects between the environment and genetic factors, as well as between genes at shared and separate loci (epistasis and dominance), and genetic heterogeneity.

Going forward

There are clearly many avenues for research to tackle the missing heritability problem. However, there is sometimes a trade-off between strategies. For example, we need larger samples to have sufficient statistical power to detect the small effect sizes of genetic variants on traits, but it is difficult to build a large sample without aggregating many heterogeneous measures. Large national biobank samples with homogeneous phenotyping such as the UK Biobank don’t have this issue and so are particularly exciting resources going forward.


CONVERGE Consortium. (2015). Sparse whole-genome sequencing identifies two loci for major depressive disorder. Nature, 523(7562), 588.

Eaves, L., Foley, D., & Silberg, J. (2003). Has the “equal environments” assumption been tested in twin studies? Twin Research and Human Genetics6(6), 486-489.

Kendler, K. S., Neale, M. C., Kessler, R. C., Heath, A. C., & Eaves, L. J. (1994). Parental treatment and the equal environment assumption in twin studies of psychiatric illness. Psychological Medicine24(3), 579-590.

Loehlin, J., & Nichols, R. C. (1976). Heredity, environment and personality: A study of 850 sets of twins [as cited by Sicotte, Woods, & Mazziotta, 1999].

Maher, B. (2008). Personal genomes: The case of the missing heritability. Nature News, 456(7218), 18-21.

Ripke, S., Wray, N. R., Lewis, C. M., Hamilton, S. P., Weissman, M. M., Breen, G., … & Heath, A. C. (2013). A mega-analysis of genome-wide association studies for major depressive disorder. Molecular Psychiatry, 18(4), 497.

Wray, N. R., Ripke, S., Mattheisen, M., Trzaskowski, M., Byrne, E. M., Abdellaoui, A., … & Bacanu, S. A. (2018). Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nature Genetics, 50(5), 668.

Yang, J., Bakshi, A., Zhu, Z., Hemani, G., Vinkhuyzen, A. A., Lee, S. H., … & Snieder, H. (2015). Genetic variance estimation with imputed variants finds negligible missing heritability for human height and body mass index. Nature Genetics47(10), 1114.

Yu, C., Arcos-Burgos, M., Baune, B. T., Arolt, V., Dannlowski, U., Wong, M. L., & Licinio, J. (2018). Low-frequency and rare variants may contribute to elucidate the genetics of major depressive disorder. Translational Psychiatry8(1), 70.