Using disease symptoms to improve detection of linkage under genetic heterogeneity.

A major reason for the slow progress in identifying susceptibility genes for complex diseases may be that the clinical diagnoses used as phenotypes are genetically heterogeneous. This has led researchers to collect various phenotypes related to the diagnosis, such as detailed symptoms, in the hope that these measurements define more homogeneous disease sub-types, influenced by a smaller number of genes that will thus be more easily detectable. Latent class analysis can be used to define disease sub-types from multivariate symptoms under the assumption that the subjects are independent, an assumption that does not hold between members of the same family. We have recently developed a latent class model allowing dependence between the latent disease class status of relatives within nuclear families. In this paper, we propose approaches to use the resulting latent class probabilities in linkage analysis. We present results from a simulation study showing that the latent class approach can provide a substantial gain in power to detect disease genes over the standard heterogeneity approach of Smith and identity-by-descent sharing methods applied to the disease diagnosis. Taking into account familial dependence in the latent class model generally provides greater power than assuming independence. In an analysis of autism symptoms in families from the Autism Genetics Research Exchange, linkage signals obtained with latent class-derived phenotypes were stronger than those obtained using the original autism spectrum disorder diagnosis.