Inference of host-pathogen interaction matrices from genome-wide polymorphism data

Abstract

Coevolution is defined as the evolutionary change in one antagonist (host) in response to changes in the other antagonist (pathogen). At the genetic level, these changes are determined by genotype x genotype (GxG) interactions. We build on a general theoretical model of a host-pathogen interaction to derive four indices to retrieve key features of GxG interactions. The four developed indices extract relevant information from polymorphism data of randomly sampled uninfected hosts as well as infected hosts and their respective pathogen strains. Using these indices as summary statistics in an Approximate Bayesian Computation method, we can show their power to discriminate between GxG interaction matrices. Second, we apply our ABC method to a SNP data set of 451 European humans and their infecting Hepatitis C Virus (HCV) strains supplemented by polymorphism data of 503 individuals from the 1,000 genomes project. As our indices encompass and extend previous natural co-GWAs we recover many of the associations previously reported for this dataset and infer their underlying interaction matrix. We reveal a new candidate gene for resistance to HCV in the human genome, and two groups of significant GxG associations exhibiting gene-for-gene interactions. We suggest that the inferred types of GxG interactions result from the recent expansion, adaptation and low prevalence of the HCV virus population in Europe.