Computational Modeling of Virally-encoded Ion Channel Structure

Viroporins are ion channels encoded within a virus’s genome, that facilitate a range of devastating infectious diseases such as COVID-19, HIV, and rotavirus. The non-structural protein 4 (NSP4) from rotavirus includes a viroporin domain that disrupts cellular Ca2+ homeostasis, initiating viral replication, and leading to life-threatening vomiting and diarrhea. Though the structure of soluble segments of NSP4 has been determined, membrane-associated regions, including the viroporin domain, remain elusive when utilizing well-established available experimental methods such as x-ray crystallography. However, two recently published protein folding algorithms, AlphaFold2 and trRosetta, demonstrated a high degree of accuracy, when determining the structure of membrane proteins from their primary amino acid sequences, though their training datasets are known to exclude proteins from viral systems. We tested the ability of these non-viral algorithms to predict functional molecular structures of the full-length NSP4 from SA11 rotavirus. We also compared the accuracy of these structures to predictions of other experimental structures of eukaryotic proteins from the Protein Data Banks (PDB), and show that the algorithms predict models more similar to corresponding experimental data than what we saw for the viroporin structure. Our data suggest that while AlphaFold2 and trRosetta each produced distinct NSP4 models, constructs based on either model showed viroporin activity when expressed in E. coli, consistent with that seen from other historical NSP4 sequences.