Limb proportions predict aquatic habits and soft-tissue flippers in extinct amniotes

Gordon et al. use a phylogenetic machine-learning approach incorporating linear and geometric morphometric analyses (including >11,000 original linear measurements) to characterize osteological correlates of interdigital webbing, soft-tissue flippers, and aquatic habits in amniotes. They find that relative hand length can reconstruct soft-tissue phenotypes and aquatic habits in extinct amniotes with >90% accuracy, enabling a more phylogenetically comprehensive investigation of limb evolution among bona fide aquatic and terrestrial species. This investigation reveals distinct morphometric landscapes for limb evolution in mammals and reptiles and clarifies the aquatic habits of multiple extinct groups with historically ambiguous ecologies. In addition, it debuts a versatile supervised machine-learning approach for reconstructing cryptic phenotypes from fossil material.