Tools For Association Studies: Quantitative Trait Analysis Mark J. Rieder, PhD Department of Genome Sciences CWRU, April 11, 2008

CasesControls 40% T, 60% C15% T, 85% C Strategies for Genetic Analysis Populations Association Studies C/CC/T C/CC/T C/C C/C C/T C/C C/C C/T C/CC/TC/T C/C Single Gene Rare Variants ~600 Short Tandem Repeat Markers Families Linkage Studies Simple Inheritance Phenotype Measure Frequency Continuous Multiple Genes Common Variants 300,000 -1,000,000 SNPs Complex Inheritance

CasesControls 40% T, 60% C15% T, 85% C Approaches to Association Studies Phenotype Measure Frequency ContinuousQuantitative Directed - Candidate Gene Studies Resequencing SNP data Whole Genome Association Studies (WGAS) - tagSNPs

Candidate gene association studies Choose gene based on previous knowledge – – Gene function – – Biological pathway – – Previous linkage or association study Choose DNA variations for genotyping SNPs from HapMap, resequencing - Direct association approach - Indirect association approach

Advantages – – Detect common variation with small genetic contributions Multiple independent genes can be involved – – Association (indirect) defines a relatively small region (based on linkage disequilibrium) – – Does not require a priori knowledge of what genes or regions are involved Caveats – –Typically, requires thousands of samples to find a significant association – –Statistical issues related to multiple testing (Bonferroni correction) – –Extremely large datasets are generated (e.g., 2000 samples X 500,000 loci or more than 1 billion genotypes) – –Analysis and replication strategies are important The Hope The identified targets will lead to new biological and medical insights (hypothesis generating) Whole Genome Association Studies (WGAS)

Strategies for Genetic Analysis (WGAS) Regression Analysis y =  1 x 1 +  2 x 2 + … +  i x i +  UNIVARIATE: dose =  1 genotype1 MULTIVARIATE dose =  1 genotype1 +  2 age +  3 sex

Warfarin Pharmacogenetics 1.Background Vitamin K cycle Pharmacokinetics/Pharmacodynamics 2.Candidate Gene Approach - VKORC1 SNP discovery, tagSNP selection 3.Clinical Association Study VKORC1 and warfarin dose SNP/Haplotype approach Replication, Function 4.Whole Genome Association Study New candidates for warfarin dosing? Power, Significance, Replication

Warfarin Dosing - Background Commonly prescribed oral anti-coagulant and an inhibitor of the vitamin K cycle Prescribed following MI, atrial fibrillation, stroke, venous thrombosis, prosthetic heart valve replacement, and following major surgery Warfarin (Coumadin) >20 million US prescriptions (2007) (++) Prevents 20 strokes for each bleeding event (-) Major bleeding episodes in 1-2% of all patients 11% of all adverse events (Gurwitz et al. JAMA 2003) (-) Difficult to determine effective dosage - - Narrow therapeutic range (INR 2-3) - - Large inter-individual variation Goal: Use genetics to understand dose requirements --> fewer complications

Ave: 5.2 mg/d n = 186 European-American Stabilized warfarin dose (3 consecutive visits within INR range) UWMC warfarin patient population

Vitamin K-dependent clotting factors (FII, FVII, FIX, FX, Protein C/S/Z) Epoxide Reductase  -Carboxylase (GGCX) Warfarin inhibits the vitamin K cycle Warfarin Inactivation CYP2C9 Pharmacokinetic

Warfarin pharmaokinetics (metabolism) Major pathway for termination of pharmacologic effect is through metabolism of S-warfarin in the liver by CYP2C9 CYP2C9 SNPs alter warfarin metabolism: CYP2C9*1 (WT) - normal CYP2C9*2 (Arg144Cys) - intermediate metabolizer CYP2C9*3 (Ile359Leu) - poor metabolizer McClain, Genet Med, 2008

Vitamin K-dependent clotting factors (FII, FVII, FIX, FX, Protein C/S/Z) Epoxide Reductase  -Carboxylase (GGCX) Warfarin acts as a vitamin K antagonist Warfarin Inactivation CYP2C9 Pharmacodynamic

New Target Protein for Warfarin Epoxide Reductase  -Carboxylase (GGCX) Clotting Factors (FII, FVII, FIX, FX, Protein C/S/Z) Rost et al. & Li, et al., Nature (2004) (VKORC1) 5 kb - Chromosome 16

Warfarin Resistance VKORC1 Polymorphisms Rare non-synonymous mutations in VKORC1 causative for warfarin resistance (15-35 mg/d) NO NO non-synonymous mutations found in ‘control’ chromosomes (n = ~400) Rost, et. al. Nature (2004)

Warfarin maintenance dose (mg/day) Inter-Individual Variability in Warfarin Dose: Genetic Liabilities SENSITIVITY CYP2C9 coding SNPs - *3/*3 RESISTANCE VKORC1 nonsynonymous coding SNPs Frequency Common VKORC1 non-coding SNPs?

SNP Discovery: Resequencing VKORC1 PCR amplicons --> Resequencing of the complete genomic region 5 Kb upstream and each of the 3 exons and intronic segments; ~11 Kb Warfarin treated clinical patients (UWMC): 186 European Rieder et al, NEJM 352, , 2005

VKORC1 - PGA samples (European, n = 23) Total:13 SNPs identified 10 common/3 rare (<5% MAF) VKORC1 - Clinical Samples (European patients n = 186) Total:28 SNPs identified 10 common/18 rare (<5% MAF) 1 - nonsynonymous - single heterozygous indiv. - highest warfarin dose = 15.5 mg/d Do common SNPs associate with warfarin dose? Candidate Gene Association Study Direct Approach SNP Discovery: Resequencing Results

SNP Selection: VKORC1 tagSNPs

Five Bins to Test , 3673, 6484, 6853, , SNP Testing: VKORC1 tagSNPs Regression Analysis y =  1 x 1 +  2 x 2 + … +  i x i +  Additive = 0,1,2 (e.g. AA = 0, AG = 1, GG = 2) Recessive, Dominant = 0,1 (e.g. AA, AG = 0, GG = 1) UNIVARIATE: dose =  1 genotype1 MULTIVARIATE: dose =  1 genotype1 +  2 age +  3 sex

Five Bins to Test 1.381, 3673, 6484, 6853, , Bin 1 - p < Bin 2 - p < 0.02 Bin 3 - p < 0.01 Bin 4 - p < Bin 5 - p < C/CC/TT/T e.g. Bin 1 - SNP 381 SNP Testing: VKORC1 tagSNPs r 2 = 21% r 2 = 3% r 2 = 4.5% r 2 = 12% r 2 = 11% MULTIVARIATE: adjusted for other confounders e.g., age, sex, medication, CYP2C9 r 2 = variance explained in dose

CCGATCTCTG-H1 CCGAGCTCTG-H2 TAGGTCCGCA-H8 TACGTTCGCG-H9 (381, 3673, 6484, 6853, 7566) B A VKORC1 haplotypes cluster into divergent clades Patients can be assigned a clade diplotype: e.g. Patient 1 - H1/H2 = A/A Patient 2 - H1/H7 = A/B Patient 3 - H7/H9 = B/B Explore the evolutionary relationship across haplotypes TCGGTCCGCA-H7 Multi-SNP testing: Haplotypes

VKORC1 haplotypes show a strong association with warfarin dose Low High A/A A/B B/B * † † * * All patients2C9 WT patients2C9 VAR patients AAABBB AA AB BB AAABBB (n = 181)(n = 124)(n = 57)

* † † * * All patients2C9 WT patients2C9 VAR patients AAABBB AAABBB AAABBB Univ. of Washington n = 185 All patients2C9 WT patients2C9 VAR patients AAABBB AAABBB AAABBB † † * † * 21% variance in dose explained Washington University n = 386 Brian Gage Howard McCleod Charles Eby

SNP Function: VKORC1 Expression Expression in human liver tissue (n = 53) shows a graded change in expression.

GWAS approach for association studies 1. 1.Establish baseline genetic diversity in a discovery population (HapMap) (e.g. find all common SNPs, >5% frequency) 2. 2.Calculate correlation between SNPs to find informative SNPs (tagSNPs) 3. 3.Genotype tagSNPs in population Use commercially available whole genome chips (e.g. Illumina, Affy) QC genotype data 4. 4.Perform statistical test for association Association Results: Multivariate Regression Establish p-value cutoff (1E-7) - Bonferroni corrected ~p< Replicate in similar populations

Clinical Adoption of Dosing Algorithms NHLBI Clinical Warfarin Trial: Randomized trial of prospective genotype-guided dosing Multi-center, double-blind randomized trial (n=2,000) "standard of care" vs. clinical alg. vs. clinical + genetic alg. Total warfarin variance (r 2 ): Outcomes: % time with INR range, time to stable dose

Design - WGA Warfarin Dose 550 K Illumina (561,278 SNPs) 181 samples tested (Higashi, et al., Rieder, et al.) - (~100 million genotypes) Call rates 99% 100% concordance - VKORC (rs ) with rs (LD) 100% concordance - CYP2C9*3 (rs ) Univariate - Individual SNP effects: ln(dose) = SNP Additive = 0,1,2 (e.g. AA = 0, AG = 1, GG = 2) ** Recessive, Dominant = 0,1(e.g. AA, AG = 0, GG = 1) Multivariate - Full Model (Genetic + Clinical): ln(dose) = Age + Sex + Amiodarone + Losartan + CYP2C9 (*2 or *3) + VKORC1-3673

Warfarin Dose - Detection Power Ave = 5.2 ± 2.5 mg/d Additive effect, quantitative trait P = 1x10 -7 (Bonferroni, p=0.05) CYP2C9 VKORC1

Univariate Results - SNP Associations Bonferroni correction (p = 0.05 corrected for 500,000 SNPs)

CYP2C9 *2/*3 = rs composite CYP2C9 allele (Wadelius, et al) Warfarin Dose GWA Replication Results UF Replication (J. Johnson) Vanderbilt Replication (D. Roden)

CYP2C9 *2/*3 = rs composite CYP2C9 allele (Wadelius, et al) Warfarin Dose GWA Replication Results UF Replication (J. Johnson) Vanderbilt Replication (D. Roden)

CYP2C9 *2/*3 = rs composite CYP2C9 allele (Wadelius, et al) Warfarin Dose GWA Replication Results UF Replication (J. Johnson) Vanderbilt Replication (D. Roden) 1x10 -4 replication threshold Correct for 384 SNPs (p=0.04)

log (p-value) chr1 chrX large real effects (VKORC1) non-replicated SNPs Range for large effect genes for warfarin dosing

log (p-value) chr1 chrX smaller, validated effect CYP2C9 New Candidates small, moderate effects replication, function Range for new candidate genes with smaller effect

VKORC1 Pharmacogenetics Association Summary 1. 1.Candidate gene associations studies are used for direct hypothesis testing In the candidate gene association, VKORC1 SNPs are the major contributor to warfarin dose variability (21-25%). 2. VKORC1 SNPs replicate in independent patient populatios and show functional effects on gene expression. 3. Whole genome association studies (WGAS) have power to detect large effect size (20-25% variance) with limited sample size. Power is limited to detect small/moderate effects. 4. WGAS don’t reveal any SNPs that have a large effect on warfarin dose similar to VKORC1. 5. WGAS studies are best suited to studies with large samples sizes and able to detect smaller genetic effects.

Acknowledgments Greg Cooper Allan Rettie Alex Reiner Dave Veenstra Dave Blough Ken Thummel Debbie Nickerson Josh Smith Michelle Wong Eric Johanson Washington University Brian Gage Howard McLeod Charles Eby Replication Studies Julie Johnson, UF Dan Roden, VU