Aims

Human phenotypes exhibit significant levels of variation among individuals. Recently, genome-wide association studies (GWAS) have been performed to identify the genetic loci that control these variations. Despite the success of GWAS to identify a set of genetic loci, it has become widely recognized that these loci cannot fully explain the variance of a phenotype. One possible explaination is that there is phenotypic stochastic noise which has not been taken into account in the typical GWAS. Normally, only one measurement of phenotype from each individual has been used in the typical GWAS which could be considered as the average phenotype of each individual. However, if we measure one phenotype several times in one individual, we will find that in some people the phenotype is relatively stable while in other people the phenotype is relatively variable. In other words, people show different levels of phenotypic robustness (canalization). The genetic control on phenotypic robustness has been reported in plants, yeast and mice. However, there is still scarce data in human.

Hypothesis

We hypothesized that genetic components controlling phenotype robustness may exist in humans. Since typical GWAS's focus on the genetic variation impact upon average phenotypes, the findings from genetic control on phenotype stochastic noise would improve our understanding on the genotype-phenotype relationship. In TwinsUK dataset, we observed different levels of phenotypic robustness on glucose and lipids across multiple measures on each individual. In the initial GWA analysis on glucose level robustness, we found a trend of association on several genetic loci across the genome. In order to verify our finding, we will request to use the ALSPAC data to perfrom GWA on glucose level robustness. Then we will perform meta-analysis using GWAS results from TwinsUK and ALSPAC. We will carry out genome-wide meta-analysis on lipids levels as well.

Request Data

Exposure variables

The genome-wide genotypes are considered as exposure variables. Both genotyped and imputed SNPs will be used for analysis. We suggest to employ the following threshold for quality control of the SNPs:

a. Call rate greater than = 95%

b. Minor Allele Frequency greater than =1%

c. For the imputed SNPs, MACH r2greater than =3 or IMPUTE prop infogreater than =0.4

d.HWE pvalue greater than 10e-6

Outcome variables

Glucose levels and lipids levels (including HDL/LDL cholesterol, triglyceride) will be used to generate outcome variables. The individuals with greater than =2 measurements will be included in the analysis. The standard deviation of multiple measurements of phenotypic levels in each individual will be used as outcome variables. Age of phenotype collection will be included in the analysis model as covariate.