Protein structure characters in the light of phylogenetic systematics

Protein structure characters have great potential for improving phylogenetic inference, especially for deep nodes where amino acid sequences are highly diverged. The combination of AlphaFold structure predictions and Foldseek's "3Di" structural alphabet makes it relatively easy to conduct model-based phylogenetic inference that includes a partition of slow-evolving 3Di characters. However, we show that even identical amino acid sequences can produce substantially different 3Di characters, depending on the source of structural model and whether inter-chain interactions are considered. We argue that such variability can be addressed with key concepts from traditional organism-based phylogenetic systematics: semaphoront, hypodigm, and character ascertainment method. To illustrate this, we develop an analogy between organismal development, taphonomy, and subsequent description and character coding by a systematist, and the process of protein synthesis, folding, and interaction and subsequent extraction, experimentation, and structural modeling by a biochemist. We conclude that differences in 3Di characters between semaphoronts are not intrinsically a problem, but they do require that the researcher uses the same replicable method on all proteins in the phylogenetic analysis. The guiding principle should be to maximize the chance that character differences in the data matrix are the results of underlying evolutionary changes, rather than artefacts due to differences in the methods used for obtaining semaphoronts and coding characters.