Parsimony analysis of coevolving species associa-tions

Parsimony analysis of host-parasite associations is based on coevolutionary models specifying relevant events, and costs associated with each of these events. Several alternative models and cost assignments have been proposed but the relations between them and other conceivable models and cost assignments have not been examined. Here, I define some desirable properties of parsimony methods and use them to explore a model space, which is a superset of the discussed possibilities. I find that there are only five valid models, each associated with specific cost constraints. All five models, or close analogues of them, have been discussed in the literature. Brooks Parsimony Analysis, formerly not associated with a model, is shown to be a close approximation of an event-based method in which switches are weighted according to the relatedness of the hosts involved in the shift. I review progress in the development of algorithms and compare alternative strategies used for significance testing. I describe new solutions to the problems of treating parasites with multiple hosts, for dealing with uncertainty in phylogenetic estimates, and for optimising event costs. I show that parsimony analysis is superior to methods maximising cospeciation, both in identifying general historical constraints in host-parasite associations and in sorting out well-supported cospeciation events when cospeciation is mixed with other processes, such as independent parasite speciation or host switching. The described techniques are implemented in the freely available program TreeFitter.