Linkage Disequilibrium Information within Monozygotic Twin Pairs - a Rationale for Genotyping Twins

Abstract

Recently genome-wide association studies have become a standard gene mapping method, but MZs have been either considered to be redundant or treated as unrelated individuals after randomly selecting one cotwin. It has been a common sense in academia that monozygotic twin (MZ) pairs, although they have unique strengths in detecting non-genomic etiology, do not contribute to gene mapping studies. It is indeed true for linkage analysis, but resemblance of phenotypes between MZ cotwins does include information about linkage disequilibrium (LD) between the genetic markers and postulated disease-susceptibility loci (DSL) when the concordance/discordance rates are compared across genotypes. The authors attempted to formulate a method to detect association and suggest several ways of applying the information using triglyceride and hypertriglyceridemia as a model phenotype. Genome-wide association test findings from the Healthy Twin Study, Korea which do not utilize MZ concordance information were compared with the tests of MZ concordance only from 493 MZs with genetic markers It is natural to assume that if allele D (wild type allele is +) is in LD with true DSL, MZ twins with genotype DD or D+ should have more diseases, and thus concordance rate than those with ++ genotype. Numerically, the observed concordance rate of MZ twins will exceed the expected concordance rate under null hypothesis of non-association (Equation 1) Equation 1 H0: Obs (A*A|DD)= Obs (A*A |D+)= Obs (A*A |++) = (prevalence)2, H1: Obs(A*A|DD)>Obs(A*A|D+)>Obs(A*A|++) and Obs(U*U|DD)<Obs(U*U|D+)< Obs(U*U|++) where "A*A"-the proportion of MZ with both affected, "U*U"-the proportion of MZ with both unaffected, "Obs()"-observed rate, "prevalence"-prevalence rate among general population or study participantsThis can be further formulated by a trend test assuming additive model,or reduced tests assuming dominant model. Extended models can be constructed such as a logistic regression model allowing covariate adjustment or multiple linear regression calculating the sum of covariance between the cotwins for continuous traits.Formula : logit[p(A*A)]= α×covariates + β×num(allele D) + ε : additive modellogit[p(A*A)]= α×covariates + β×I(allele D exists) + ε : dominant model where "I(allele D exists)"-1:D exists,0:not existsWhen we applied the multiple logistic regression test with 5.83 mmol/L as a cut-off for hypertriglyceridemia, 10 markers out of 26523 SNPs in chromosome 11 which exceeded p<10-3 showed p-value of 0.0003~0.001. When we consider p-value of 0.05,0.1,0.15,0.2 and 0.3 as cut-off for the MZ test and p-value of 10-3 and 10-4 as cut-off for GWAS results, the sensitivity/specificity of the MZ tests ranged 0.26~0.56 for sensitivity and 0.69~0.95 for specificity. The MZ test results were combined with family-based GWA results. Our findings show a way to extract LD information from MZ concordance or resemblance when one of cotwin was genotyped. Because the MZ resemblance is independent of other information conventionally used for GWAS, it can be combined with other results. The test can be also used to screen markers to alleviate the burden of multiple testing with appropriate threshold. Given that the MZ will be ever popularly applied for various omics studies, the addition of genomic information will facilitate multi-omics study in twin research.