The impact of alignment and scaling on biological inferences from landmark-free morphometrics

Recent advances in shape quantification techniques have revolutionised the field of evolutionary morphology by providing a time-efficient alternative to well-established methods such as geometric morphometrics. In particular, landmark-free methods are at the forefront of new developments in shape analysis, but concerns over their reproducibility and sensitivity to non-biological variation have hindered more widespread adoption. A primary source of non-biological bias in shape analysis is misalignment and incorrect scaling of specimens, which can result in capturing anatomically incongruent aspects of shape in otherwise homologous structures. Here, we compare methods for mesh alignment and scaling using a recently described landmark-free approach, Deterministic Atlas Analysis (DAA). We find similarities between morphospaces derived from meshes aligned using most alignment protocols, but with large differences in variance structure. Surprisingly, the largest differences are not found between Procrustes-based and non-Procrustes-based approaches, but between the Procrustes-based method with internal scaling and all other alignment approaches. More broadly, scaling method has the greatest impact on shape variation captured by DAA. Based on these results, we make recommendations to ensure that alignment strategy maintains homology in landmark-free morphometry. As in all morphometric analyses, violating principles of homology can result in spurious or inaccurate biological inferences in macroevolutionary studies of anatomy and must be explicitly considered in designing methodological frameworks.