Genetic prediction of myopia

The number of people developing myopia and high myopia has increased in recent years, throughout Europe and Asia. This increased incidence is set to escalate further, with reports of nearly 50% of the population likely to develop myopia by 2050. Although investigations into what is causing this ‘myopia boom’ and its associated risk factors is currently ongoing, it is still not entirely clear why certain individuals are affected and not others. It is suspected that changes in the environment within recent years, such as a global push towards education, may be responsible. However, there is a large body of evidence that demonstrates the complexity of the refractive error phenotype and human emmetropisation, with reports identifying more than 150 genetic loci associated with myopia, implying some people may have a genetic predisposition to the condition. The aim of this thesis was to investigate whether inspecting genetic predisposition to myopia would allow us to detect individuals at risk, and determine whether a genetic model to predict children at risk was feasible. This may then help identify individuals who would benefit more from early intervention, or more regular monitoring. Initially, 149 genetic variants that reached genome-wide statistical significance in a GWAS for refractive error carried out by the CREAM consortium were used to create a ‘genetic risk score’ to assess the accuracy with which incident myopia could be predicted in children from the ALSPAC cohort. Analyses were also carried out for another predictor, namely the children’s number of myopic parents. The results suggested that the number of myopic parents was a better predictor of refractive error and incident myopia than the genetic risk score (R2 = 4.8% vs. 2.6%). This was likely due to several limitations in the genetic risk score. Notably, the results also demonstrated that these two predictors were largely independent, hence prediction accuracy improved when they were used together (R2 = 7.0%). To try and increase the accuracy of genetic prediction, I took advantage of the recently released genetic data from the UK Biobank cohort, for which a proportion (23%) of individuals also had ocular measurements taken. A genome wide association study (GWAS) was performed for autorefraction-measured refractive error in European individuals with both genetic and refractive data (N = 95,505), which replicated many loci previously shown to be associated with refractive error. A regression model to impute refractive error in UK Biobank participants who did not undergo autorefraction measurement was also created (to improve the accuracy of the existing genetic risk score by means of running an additional GWAS analysis, thus expanding the effective sample size used in the creation of the genetic risk score). A multi-variable model was developed using age of onset of first spectacle wear, age and gender; the model fit was optimised objectively. The resultant model yielded an imputed refractive error that was moderately explanative (R2 = 0.30) for the variance of ‘true’ (autorefraction-measured) refractive error, as judged in an independent sample. A GWAS for imputed refractive error was carried out in 287,448 European UK Biobank participants who were not amongst the 95,505 individuals included in the original GWAS for autorefraction-measured refractive error. The genetic correlation between the 2 traits (imputed refractive error vs. autorefraction-measured refractive error) was rg = 0.92, which confirmed that the imputed refractive error phenotype was a good surrogate for the true phenotype. Summary statistics from the 2 GWAS analyses described above were combined, along with GWAS summary statistics for educational attainment taken from a published study (www.SSGAC.org). Meta-analysis was performed using ‘multi-trait analysis of genome-wide association summary statistics’ (MTAG). The accuracy of the genetic risk scores in predicting refractive error was assessed in an independent sample of European adults (the ALSPAC mothers cohort). The best prediction accuracy was achieved by combining summary statistics for all 3 traits (autorefraction-measured refractive error, imputed refractive error, and educational attainment). The resultant genetic risk score explained 11.2% of the variance of refractive error, and demonstrated an area under the receiver operating characteristics curve (AUROC) of 0.67 and 0.75 for predicting any (≤ 0.75D) and moderate (≤-3.00D) myopia, respectively. Participants from the ALSPAC mothers cohort in the top 10th percentile of the genetic risk score were found to be at 6-fold greater risk of developing high myopia (≤-5.00D) compared to the remainder of the sample. The accuracy of the genetic risk score was also tested in individuals of Asian, Chinese, and Black ancestry. Prediction accuracy was reduced by approximately 50% in Asian and Chinese individuals. Prediction accuracy was worse still in those of Black ethnicity