Inferring the multi-host fitness landscape of endive necrotic mosaic virus from cross-inoculation experiments

Fitness landscapes offer a compact representation of adaptation, yet are rarely inferred from multi-environment data. We present a Bayesian approach to infer a multi-host phenotypic fitness landscape from cross-inoculation assays by linking successful infection probabilities to Fishers geometrical model under strong selection and weak mutation. The model estimates (i) the distance matrix among host-specific phenotypic optima, (ii) host-specific permissiveness through the widths of fitness peaks on target hosts, and (iii) host-specific differences in the efficiency with which phenotypic suitability translates into successful infection. We apply the approach to an experimental evolution dataset for endive necrotic mosaic virus evolved on five Asteraceae hosts and challenged in a full cross-inoculation design. The inferred landscape can be visualized as a phenotypic map of the host community, revealing pronounced heterogeneity in host permissiveness and a geometry broadly concordant with host phylogeny. By grounding assay-derived distances in an explicit mechanistic model, the approach provides a parsimonious representation of multi-host constraints that can be used to discuss establishment barriers and potential springboard hosts in heterogeneous communities. More broadly, it offers a general method for inferring effective fitness landscapes from sparse multi-environment data.