Inferring the Nature of Missing Heritability in Human Traits Using Data from the GWAS Catalog

Abstract

Thousands of genes responsible for many diseases and other common traits in humans have been detected by Genome Wide Association Studies (GWAS) in the last decade. However, candidate causal variants found so far usually explain only a small fraction of the heritability estimated by family data. The most common explanation for this observation is that the missing heritability corresponds to variants, either rare or common, with very small effect, which pass undetected due to a lack of statistical power. We carried out a meta-analysis using data from the NHGRI-EBI GWAS Catalog in order to explore the observed distribution of locus effects for a set of 42 complex traits and to quantify their contribution to narrow-sense heritability. With the data at hand, we were able to predict the expected distribution of locus effects for 16 traits and diseases, their expected contribution to heritability, and the missing number of loci yet to be discovered to fully explain the familial heritability estimates. Our results indicate that, for 6 out of the 16 traits, the additive contribution of a great number of loci is unable to explain the familial (broad-sense) heritability, suggesting that the gap between GWAS and familial estimates of heritability may not ever be closed for these traits. In contrast, for the other 10 traits, the additive contribution of hundreds or thousands of loci yet to be found could potentially explain the familial heritability estimates, if this were the case. Computer simulations are used to illustrate the possible contribution from nonadditive genetic effects to the gap between GWAS and familial estimates of heritability.

Keywords: GWAS; big data; missing heritability; prediction of complex traits.

Figure 1

Decline of the average locus effect (α) with the number of loci found. The points represent the cumulated results of successive GWAS with increasing larger sample sizes. The first point at the left of the series is the mean effect of loci found in the GWAS with the lowest sample size (conditional on finding at least 30 loci), and the following points give the mean effect of loci as additional ones are found by studies with larger sample sizes (usually, but not always, by more recent studies). The lines are the fit of the observations to an exponential model (average $R^2$ = 0.96). Traits are colored depending on the functional domain they belong: cancer (green), dermatological (pink), endocrine (orange), gastrointestinal (brown), hematological (red), immunological (yellow), metabolic (beige), skeletal (gray). The final set of data corresponding to the last (right-hand side) points for each line are given in Table S3.

Figure 2

Increase of heritability explained by loci found $\left({\mathit{h}}_{\mathrm{gwas}}^2\right)$ as the number of these increases. The points represent the observed values, while the lines are the fit to an exponential model (average $R^2$ = 0.97). Traits are colored depending on the functional domain they belong: cancer (green), dermatological (pink), endocrine (orange), gastrointestinal (brown), hematological (red), immunological (yellow), metabolic (beige), skeletal (gray). The final set of data corresponding to the last (right-hand side) points for each line are given in Table S3.

Figure 3

Percentage of loci for different classes of effect sizes and their contributions to heritability (in %). (A) Three arbitrary classes of locus effect sizes (high, medium, and low effects) are assumed such that ∼50% of loci are within the low-effect class (high transparency), ∼36% within the medium-effect class (low transparency), and ∼14% within the large-effect class (solid colors). (B) Contribution (in percentage) of the three classes to heritability. Traits are ordered and colored by functional domain: Cancer (green), dermatological (pink), endocrine (orange), gastrointestinal (brown), hematological (red), immunological (yellow), metabolic (beige), skeletal (gray).

Figure 4

Observed and expected values of heritability. The full length of bars indicate the mean familial heritability $\left(h_{\text{fam}}^2\right)$ for the studied traits (average values are shown when there is a range of estimates from the Literature, Table S2). In solid color it is shown the heritability explained by the loci already found and available from the Catalog $\left(h_{\text{gwas}}^2\right)$. The blue error bar gives the inferred value of heritability (the dot corresponds to the median value) that approaches most to the familial heritability with a 95% confidence interval, using data from the expected distribution of locus effects. The expected number of loci for each trait required to explain the familial heritabilities within the error bars assuming an additive contribution of single loci are given over the bars. Traits are colored depending on the functional domain they belong: Cancer (green), dermatological (pink), endocrine (orange), gastrointestinal (brown), hematological (red), immunological (yellow), metabolic (beige), skeletal (gray).

Figure 5

Decline of the average locus effect (upper graph), and increase of the heritability explained $\left({\mathit{h}}_{\mathrm{gwas}}^2\right)$ (lower graph) as the number of loci found is increasing. The small points represent the observed values of the previous analyses (Figure 1 and Figure 2) and the large points those of a more recently collected set. Lines are the fit to an exponential model. Traits are colored depending on the functional domain they belong: Cancer (green), dermatological (pink), endocrine (orange), immunological (yellow), skeletal (gray).