The genetics of glycaemic response to GLP-1 receptor agonists in Type 2 diabetes

(1)   We will first analyse the clinical and genetic determinants of GLP-1RA response by meta-analysing across all albiglutide treatment intervention arms. We will develop a common clinical model, based upon our previous data this is likely to be:   HbA1c reduction ~ baseline HbA1c + age + diabetes duration + BMI (+SNP)We will include SNPs individually and together in the model, and will undertake meta-analysis of the model derived betas for the SNPs. As the trials differ in inclusion criteria, and background medication we anticipate using a random effects meta-analysis.(2)   We will then take any significant variants identified in (1) forward to evaluate the gene*treatment interaction in the four placebo controlled trials using a random effects meta-analysis of the gene*treatment interaction from each trial. We will also assess the impact of variants identified in (1) in the other available trials e.g. comparing against sitagliptin, pioglitazone, lispro insulin, glargine insulin.(3)   We will then examine the effects of variants identified in (1) on the secondary endpoints outlined. Again where combining across placebo controlled trials a random effects meta-analysis of the gene*treatment interaction will be undertaken.(4)   In an exploratory analysis we will undertake investigation of genetic variants that alter weight change, pulse and blood pressure change, and change in amylase and lipase in response to Albiglutide intervention. We will follow steps (1) and (2) but with HbA1c reduction replaced by these secondary outcomes.We will undertake steps 1 to 4 for five candidate genetic variants; then undertake GWAS and a gene burden test. For the candidate snp replication a p value of <0.01 will be considered statistically significant. For GWAS and gene burden test, accepted p-value thresholds of p<5x10-8 and p<2x10-6 will be used respectively. PowerFor analysis (1) there are ~2100 indviduals in whom albuglutide treatment is given. Of these we anticipate ~1400 having genetic information.For the candidate (replication SNPs), assuming we are testing 5 independent SNPs, assuming a standard deviation in treatment response of 1%, and MAF of 35% we will have 80% power to detect a difference between genotypes of 0.15%, assuming an additive model and alpha of 0.01.For the GWAS meta-analysis we will have a total n of ~ 2500. This will give us 80% power to detect a variant with an allelic impact of 0.21%, assuming an alpha of 5*10-8.Time to event analysisWe will use cox proportional hazards models of the intention to treat cohort. Our primary analysis will be on the albiglutide intervention group only as power is limited for a formal genotype*treatment interaction. We will follow up any signals identified in the albiglutide arm, in the control arm, and more formally using a test for interaction. For GWAS we will adjust for ethnicity by including the first 3 principal components. For variants that show a large variation in minor allele frequency by ethnic group we will undertake stratified and ethnic specific subgroup analyses. For continuous measures, such as weight, HbA1c, pulse rate and SBP, we will use mixed model for repeated measurements, with treatment group, genotype and genotype*treatment as fixed effects, with individuals having random intercept and slopes. We will evaluate statistically significant results by replication in independent trials and observational data sets, where response is defined as the change from baseline at ~6 months.Significance level.Where we are using SNPs well validated to impact on HbA1c reduction we will use a p-value threshold of 0.05 for the albiglutide treatment arm.For genome wide analyses we will use a p-value for the interaction term of p<5*10-8 as significant. For rare variant burden tests we will use a gene-based p-value for the interaction of p<1*10-6PowerThe primary endpoint occurred in 338 (7%) of 4731 patients at an event rate of 4·57 events per 100 person-years in the albiglutide group and in 428 (9%) of 4732 patients at an event rate of 5·87 events per 100 person-years in the placebo group (HR 0·78, 95% CI 0·68-0·90).Our primary analysis will be on the ~4187 with DNA treated with Albiglutide. For the common variants in GLP-1R, with MAF of 32%, we will have 80% power, alpha 0.05 to detect a relative risk of 1.4 in variant carriers. This equates to an endpoint occurring in 6% of wildtype individuals and 8% GLP-1R variant carriers. A gene*treatment interaction for this variant (n=8374) will have 62% power to detect an interaction effect of this size.