Incorporation of covariates in multipoint model-free linkage analysis of binary traits: how important are unaffecteds?

When the mode of inheritance is unknown, genetic linkage analysis of binary trait is commonly performed using affected-sib-pair approaches. When there is evidence that some covariates influence the phenotype, incorporation of this information is expected to increase the power of the analysis since it allows (1) a better specification of the phenotype and (2) to take into account unaffected subjects. Here, we show how to account for covariates in the sibship-oriented Maximum-Likelihood-Binomial (MLB) linkage method by means of Pearson's logistic regression residuals which are computed using phenotypic and covariate information on both affected and unaffected subjects. These residuals are subsequently analysed as a quantitative phenotype with the corresponding extension of the MLB approach which can be used without any assumption on the distribution of these residuals. Then, a large simulation study is performed to study the relative power of incorporating or not unaffected sibs. To this aim, two different strategies in the multipoint analysis of family data are compared: (1) using residuals of the whole sibships (ie both covariate and genotypic information on unaffecteds is needed), and (2) using affecteds only (no information on unaffecteds is needed), under different generating models according to genetic and covariate effects. The results show that there is a clear increment in the power to detect the susceptibility locus when making use of the information carried by unaffecteds, in particular for dominant mode of inheritance and when values of the covariates influencing the disease are shared by all the members of the family.