BUILDING PHYLOGENETIC TREES FOR TRILOBITES USING MAXIMUM LIKELIHOOD

Parsimony analysis has been widely applied to phylogenetic studies of fossil and extant taxa. Sampling error can limit the accuracy of tree estimates, especially when the number of taxa analyzed exceeds the number of morphological characters that can be identified and coded. Such situations arise more frequently with fossil taxa, which typically lack data from soft tissues. Supertree approaches have frequently been used as a way around this problem. A broad phylogenetic framework can be used to identify small, overlapping clusters of taxa. Trees for each of these taxon sets can be estimated using different data matrices. Then these estimates of the trees can be stitched together to generate a higher-level phylogenetic analysis. Although such an approach is tenable, ultimately it will be very valuable to have a means of building larger phylogenetic trees using a single data matrix. Moreover, taxon-rich analyses based on a single supermatrix may provide greater potential for interfacing with phylogenetic analyses based on neontological data and using molecular approaches. In 2001, Paul Lewis introduced modifications to the calculation of the likelihood on phylogenetic trees to accommodate some forms of ascertainment bias that are inherent in morphological data. The application of maximum likelihood techniques to fossil data sets (or other matrices with substantial proportion of missing data) requires generalization of the software implementations of Lewis’ approach to tree inference. We see this not as a rejection of parsimony, but rather as the logical evolution of phylogenetic paleontology in the 21^st century, and an important means of keeping paleontology relevant to evolutionary biology. As a test case, we focus on applying maximum likelihood methods to a supermatrix generated for a group of ~80 species of Early Cambrian trilobites comprising the Olenelloidea: a diverse, morphologically complex clade. Previously, these were analyzed using supertree approaches and parsimony analysis. Thus, our study provides a means of directly comparing and contrasting results from these different phylogenetic approaches. We found that there is significant potential for likelihood-based approaches as applied to large fossil data matrices, especially when information about character change is incorporated.