Categorical versus geometric morphometric approaches to characterizing the evolution of morphological disparity in Osteostraci (Vertebrata, stem Gnathostomata)

Morphological variation (disparity) tends to be evaluated through two non-mutually exclusive approaches: (i) quantitatively, through geometric morphometrics, and (ii) in terms of discrete, ‘cladistic’, or categorical characters. Uncertainty over the comparability of these approaches diminishes the potential to obtain nomothetic insights into the evolution of morphological disparity, and the few benchmarking studies conducted so far show contrasting results. Here, we apply both approaches to characterising morphology in the stem-gnathostome vertebrate clade Osteostraci, in order to assess congruence between these alternative methods as well as to explore the evolutionary patterns of the group in terms of temporal disparity and the influence of phylogenetic relationships and habitat on morphospace occupation. Our results suggest that both approaches yield similar results in morphospace occupation and clustering, but also some differences indicating that these metrics may capture different aspects of morphology. Phylomorphospaces reveal important convergence towards a generalised ‘horseshoe’-shaped cranial morphology and two strong branching trends involving different major groups of osteostracans (benneviaspidids and thyestiids), which probably reflect adaptations to different lifestyles. Temporal patterns of disparity recorded by categorical and morphometric approaches differ considerably, capturing disparity maxima at very different times of the evolutionary history of the group. Disparity patterns recorded by the categorical approach parallel taxonomic diversity dynamics, likely reflecting a bias in facies representation rather than a real biological signal. This work provides evidence supporting that categorical and continuous data do not always capture morphological disparity in equivalent ways and that discrepancies reflect differences in the potential of each data type for characterizing more or less inclusive aspects of overall phenotype.