Meta-analysis of 542,934 subjects of European ancestry identifies new genes and mechanisms predisposing to refractive error and myopia

Abstract

Refractive errors, in particular myopia, are a leading cause of morbidity and disability worldwide. Genetic investigation can improve understanding of the molecular mechanisms that underlie abnormal eye development and impaired vision. We conducted a meta-analysis of genome-wide association studies (GWAS) that involved 542,934 European participants and identified 336 novel genetic loci associated with refractive error. Collectively, all associated genetic variants explain 18.4% of heritability and improve the accuracy of myopia prediction (area under the curve (AUC) = 0.75). Our results suggest that refractive error is genetically heterogeneous, driven by genes that participate in the development of every anatomical component of the eye. In addition, our analyses suggest that genetic factors controlling circadian rhythm and pigmentation are also involved in the development of myopia and refractive error. These results may enable the prediction of refractive error and the development of personalized myopia prevention strategies in the future.

Extended Data Fig. 1:

Principal components plots of the subjects in the main participating cohorts. a) UK Biobank (including the 102,117 subjects with direct refraction measurement and the imputed 108,956 likely myopes to 70,941 likely non-myopes, for a total of 179,897 subjects) , B) Genetic Epidemiology Research on Adult Health and Aging (GERA, N=34,998 ), C) 23andMe (106,086 cases and 85,757 controls, or 191,843 subjects in total).

Extended Data Fig. 2:

Correlation of effect sizes between the discovery cohort meta-analysis. Effect sizes are from two analyses, discovery (UK Biobank analysis on spherical equivalent + GERA, spherical equivalent + 23andMe, self-reported myopia cases and controls + UK Biobank inferred myopia cases and controls, for a total of N=508,855 subjects) and the replication from the non-British CREAM Consortium participants (N=34,079), used as replication. The z-scores for the discovery are on the y-axis and those from the CREAM cohort in the x-axis.

Extended Data Fig. 3:

Distribution of the base-pair length (red) of the 449 regions associated in the meta-analysis of all available cohorts (from Supplementary Table 3), alongside the distribution of number of SNPs (blue) for each region. Numbers in each of the axes in the figure are differentially colored to match the density curve they correspond to: red for the length of the region and blue for the number of SNPs.

Extended Data Fig. 4:

Expression of genes located in the associated loci (from Supplementary Table 3) along the x-axis, across several human body tissues (y-axis). The colors represent the centile ranking of the expression level of the gene in the tissue of interest. The hotter colors represent higher ranking of the gene expression and the colder colors low expression. Both genes and tissues are clustered in accordance with their pattern similarity. The symbol of all the genes could not be visualized and therefore are removed for the sake of clarity. Eye tissues, whether fetal or adult, appear to have similar patterns of gene expressions (clustered together at the bottom of the figure). Genes that are highly expressed in eye tissues fall in three clusters, shown with a black box. These clusters are displayed in more detail in Figure 4A, B and C.

Extended Data Fig. 5:

Genes from the regions associated with RE (from Supplementary Table 3) that are particularly expressed in eye tissues, compared to non-ocular tissues. These clusters are those highlighted in Supplementary Figure 3, but for the sake of clarity they are shown in transposed orientation compared to the previous figure (here genes in the y-axis and eye tissues in the x-axis), but same color codes as before. The dendrograms represent the degree of similarity observed for both tissues and gene expressions. The clusters are given in the order in which they were clustered together, from left to right: A) genes that are expressed more in other ocular tissues (fetal and adult) but much less in the adult retina. B) genes that are highly expressed in the retina and other ocular tissues, and C) genes that are expressed in the retina, but less in the other ocular tissues tested.

Extended Data Fig. 6:

Results of the LD score regression analysis applied to specifically expressed genes (LDSC-SEG) on multiple tissue for the meta-analysis results. Each point represents one tissue or cell line (along the x-axis) and the log10 value of the p-value for the enrichment of the meta-analysis results among genes expressed in these tissues. There were 205 tests carried out, one in each tissue and cell line, therefore only tissues with a correlation p-value< 0.00025 (Log_P> 3.6 in this figure), would have been significant after multiple testing. This condition was not fulfilled for any of the available tissues.

Extended Data Fig. 7:

Mendelian randomization results on causality of IOP over refractive error. Single points in the graph represent coordinates determined by the effect of each specific SNP over IOP (x-axis, mmHg) and spherical equivalent (y-axis, Diopter units). A total of 73 SNPs associated with IOP, but not directly associated with refractive error (i.e. p> 0.05) were selected as instruments. Values of associations with IOP were obtained from a meta-analysis of 139,555 European participants (Reference 50 in the manuscript) and the refractive error associations from 102,117 UK Biobank subjects. The lines represent the regression lines from each model, as specified in the figure legend. In some cases, these lines may not visible because they overlap (please refer to the values underneath the figure).

Extended Data Fig. 8:

Venn’s Diagram of the number of SNPs considered in each of the stages of this study. The different circles represent various stages, inclusion in the meta-analysis (blue), identification of significant loci (green), conditional analysis results identifying independent effects (red) and the total number of SNPs available for inclusion in prediction and heritability estimation in the independent (i.e. not part of the original meta-analysis) EPIC-Norfolk cohort (orange).

Extended Data Fig. 9:

Prediction for the total number of SNPs and phenotypic variance explained as a function of GWAS sample size in future studies, based on the distribution of effects observed in the current meta-analysis. The plot lines show the predicted relationship between the number of loci associated with refractive error (left vertical axis, blue line) and the variance they help explain (red line, right vertical axis), as a function of the sample size (x-axis) used in future GWAS or meta-analyses. These projections are consistent with the observed results, where an effective sample of 379,227 identified 904 independent signals after a conditional analysis, explaining 12–16% of refractive error variability.

Extended Data Fig. 10:

The distribution of various natural selection test scores for SNPs associated with refractive error. The values on the x-axis represent the ranking in terms of natural selection observed and the y-axis the density of that rank. The different tests shown are iHS, XP-EHH (CEU vs YRI), XP-EHH average score, XP-EHH maximum score and Tajima scores (black, green, red, blue and yellow respectively).

Figure 1.

All GWAS-associated regions from the main meta-analysis. Each band is a true scale of genomic regions associated with refractive error listed in Supplementary Table 1 (+250kbp on each side to make smaller regions more visible). The different color codes represent the significance (p-value) for the genetic variant within that region that displays the strongest evidence for association.

Figure 2.

Receiver Operating Characteristic (ROC) curves for myopia predictions, using information from 890 SNP markers identified in the meta-analysis. The three different colors represent three different curves for each of the different definition of myopia: green – all myopia (< −0.75D), magenta – moderate myopia (< −3.00 D) and brown - severe myopia (defined as < −5.00 D).