IMAGING, MORPHOLOGICAL KNOWLEDGE OF SPECIES, AND PHYLOGENY RECONSTRUCTION: A CASE STUDY IN PLIOMERID TRILOBITES

Many invertebrate paleontological phylogenetic analyses rely on images of specimens from the published literature as coding sources. Species are typically represented by a handful of specimens which were imaged at low magnification in plan view. New digital technology enables publication of highly resolved images of numerous specimens. Here we use silicified Lower Ordovician pliomerid trilobites from the Great Basin to investigate whether the quality of information conveyed in traditional vs current illustrations affects phylogenetic precision or accuracy. The trilobites were originally described in mid-20^th century publications widely regarded as excellent, but our field-based revision of these faunas indicates that only about 1/3 of the common species were described, and that species which initially appeared well-known were often composites of several poorly known species.

Simulations have shown that phylogenetic accuracy and resolution are reduced by inability to code characters for all taxa, not just the percentage of missing entries in any incomplete taxa. This is borne out in our analyses. Our primary dataset includes 29 species coded for 69 characters, from our photographs of specimens we collected at the type localities. Most characters are coded for all taxa. A second dataset consists of the 20 species now known to be represented in the literature (even if unrecognized at the time) coded only from the original illustrations. Dataset 2 is incomplete due to species that were known from few exoskeletal elements, or because the original illustrations did not show all relevant morphology. 25 uncodable or autapomorphic characters were removed from it. It contains no characters coded for all taxa. We analyzed both datasets using parsimony.

Analysis of Dataset 1 resulted in six well resolved, well supported trees (CI 0.64, RI 0.89, average GC bootstrap support of 74.6). Analysis of Dataset 2 resulted in nearly 1500 poorly resolved, poorly supported trees (CI 0.77, RI 0.88, average GC support of 35.2). The strict consenses reflect similar broad relationships between genera, but the low resolution of the second analysis obscures relationships within genera. These preliminary data suggest that much greater species-level phylogenetic precision is possible with wider application of modern imaging techniques.