Using GWAS Data for Enhanced - PowerPoint Presentation

1. Using GWAS Data for Enhanced Mendelian Randomization StudiesStefan Walter(swalter@hsph.harvard.edu)on behalf of: Laura Kubzansky, Maria Glymour, Eric J Tchetgen Tchetgen, Liming Liang, Shun-Chiao Chang, Eric Rimm, Marilyn Cornelis, Karestan Koenen, and Ichiro KawachiGenetic Risk “G”(e.g., Risk Score, Candidate Genes)Phenotype/Risk Factor “X”(e.g., Anxiety)Outcome “Y”(e.g., CHD, Diabetes

2. OutlineSettingProjects:Anxiety, Depression, Social TiesResultsMethods and Instruments: Using the Genetic Risk Score as an InstrumentLeave One Out ApproachInstrumental Inequality TestsFuture:including interactions in score (GxG, GxE)

3. SettingNHS I n = 7000 (NHST2D, NHSCGEMS, NHSCHD, NHSKS)age at blood draw = 59.5 (SD 8) HPFS n = 4125 (HPFST2D, HPFSCHD, HPFSKS)age at blood draw = 62.2 (SD 9)Genotyped on different chips (Illumina, Affy), HapMap and 1000g imputation available.

4. ProjectsAnxiety:Crown Crisp Experimental Index (phobic anxiety)Depressionlong-term composite depressive symptom score (1992-2006). Social Tiessocial isolation and ability to connect with others (continuously married versus otherwise).

5. Depression and Anxiety:No evidence of a genetic instrument from additive internal GWA risk score, candidate genes, or external GWA risk score (R2 < 0.1%, mostly not significant)But inconsistent results from chip heritability (GCTA): NHS_T2D, Affy 6.0, 0.214349NHS_CGE, Illumina 550k, 0.062074NHS_CHD, Affy 6.0, 0.305586NHS_KS, Illumina 610Q, 0.000001 HPFS_T2D, Affy 6.0, 0.214349HPFS_CHD, Affy 6.0, 0.200122HPFS_KS, Illumina 610k, 0.046558Results

6. Advantage:Known genetic relationship for BMI (Speliotes et al. Association analyses of 249,796 individuals reveal 18 new loci associated with body mass index).Replicates in NHS/HPFS explaining 2% of variance.Challenge:Analysis within nested case-control (NCC) samples (all but KS) requires IP weighting to allow unbiased inference based on the original sampling population.Currently, we are recreating the risk sets based on the published matching criteria used to derive the NCC.Genetic Risk “G”(e.g., Risk Score, Candidate Genes)Phenotype/Risk Factor “X”(e.g., BMI)Outcome “Y”(e.g., Anxiety, Depression, Social Ties)

7. Methods and Instruments (1): Genetic Risk Score applying allele scoring (Purcell et al., 2009): Derived from internal GWAS Approach: running GWAS in 7 samples, meta-analyzing 6 and scoring in set number 7. Iterative leave one out procedure and subsequent meta-analysis of results.Derived from published GWAS (Speliotes et al, 2010; Demirkan et al., 2010)Derived from Candidate Genes

8. Methods and Instruments (2): Glymour, Tchetgen Tchetgen, Robins., Credible Mendelian Randomization Studies: Approaches for Evaluating the Instrumental Variable Assumptions, Am J Epi 2012Genetic IVs cannot be proven to be valid. They can sometimes be shown to be invalid, although these tests generally rely on additional assumptions. Four empirical approaches to (in)validation:Leverage prior causal assumptions regarding the confounding of the phenotype-outcome association: 4 equivalent versions of this test.Identify factors that modify the genotype-phenotype association and compare the IV effect estimate across values of the modifier. Instrumental inequality tests: applicable only when the causal phenotype is known to be categorical. Over-identification tests with multiple IVs. Other genes or even polymorphisms of the same gene might provide additional IVs. *Instrumental Inequality Test Macros available (R, SAS) for dichtomous outcome, dichotomous phenotype, and ploytomous instruments

9. FutureTesting and including GxG, GxEKnown Protein x Protein interactions to inform underlying SNP x SNP interactions. (anxiety)Investigate gender differences. (anxiety)Quantile Regression to identify and quantify interactions with (unknown) environmental factors. Construct separate scores based on presumed mechanism (e.g., appetite, adipogenesis, cardio-pulmonary fitness) and apply over-identification tests.Get access to the Health Retirement Study genetic data from dbGaP to increase sample size for psycho-social phenotypes.

10. Acknowledgements & Contact:NIH/NIHM 1RC4MH092707 (L Kubzansky)And the entire team: Laura Kubzansky, Maria Glymour, Eric J Tchetgen Tchetgen, Liming Liang, Shun-Chiao Chang, Eric Rimm, Marilyn Cornelis, Karestan Koenen, and Ichiro KawachiPlease feel free to contact us, we are happy to share thoughts, code, macros, etc.Stefan: swalter@hsph.harvard.eduLaura: lkubzans@hsph.harvard.eduMaria: mglymour@hsph.harvard.edu