Predictive fitness models for influenza vaccine strain selection

点击翻译按钮获取中文摘要

Project summary The human flu virus undergoes fast antigenic evolution driven by the challenge of the host immune system. Circulating viruses are a moving quasi-species of high genetic and antigenic diversity, which is structured in distinct clades representing niches with separate avenues of escape from human herd immunity. Specifically, viral-immune co-evolution follows a Red Queen’s dynamical pattern of competing lineages, which is further modulated by global transmission patterns, host population structure, and herd-immunity acquired by previous infections and vaccination. This process poses a challenge for vaccine strain selection: to determine a single strain from each of the four seasonal lineages to provide the best protection from the viruses that are expected to dominate almost a year in advance. Here we posit that shape and dynamics of the global viral population can be understood and predicted from the underlying human population immunity. In this project, we will develop a comprehensive and objective computational approach to provide a mechanistic understanding of the influenza virus-host immune interaction on the host level, to quantify the selection imposed on the virus on the global evolutionary scales. Specifically, we will build biophysical models, both for the B-cell and T-cell driven immune recognition of epitopes in a host, to accurately characterize the immune structure of the human population to best represent the fitness effects acting on the virus on the global scale. For the model of B-cell immune recognition (Aim 1), we will leverage diverse antigenic human serology assays to prepare a detailed map of antigenic effects of mutations and epistatic interactions. The data will be cross-mapped on the dataset of sequences of globally circulating viruses and our detailed phylogenies for each of the four seasonal lineages. The T-cell immune recognition (Aim 2) will be based on computational machine learning predictions of epitopes, combined with novel biophysically motivated models for prediction of immunodominant antigens. Host population geographically diverse HLA diversity will be accounted for estimating selective pressures imposed on the population of the virus. These components, together with a component describing the selective pressure due to previous vaccinations, will be used to optimize a joint fitness model (Aim 3). Information theoretic approaches will be used to optimize and evaluate the predictive power of the combined model. We will objectively quantify the significance of each of the components and validate the predictions on the historical sequence and epidemiological data. With the resulting fitness model we will define principled criteria for vaccine strain selection, to optimize the coverage and efficacy of the vaccine in the future populations of the of seasonal human influenza viruses.