Template-free prediction of a new monotopic membrane protein fold and oligomeric assembly by AlphaFold2.

Abstract

AlphaFold2 (AF2) has revolutionized the field of protein structural prediction. Here, we test its ability to predict the tertiary and quaternary structure of a previously undescribed scaffold with new folds and unusual architecture, the monotopic membrane protein caveolin-1 (CAV1). We recently found that CAV1 can assemble a disc-shaped structure composed of 11 symmetrically arranged CAV1 protomers, each of which assumes an identical new fold, that also contains the largest parallel β-barrel known to exist in nature. As this novel fold and assembly was not part of the PDB when AF2 was trained, it presents a blind and challenging test case for the algorithm. AF2 predicts the fold of the protomer despite CAV1's lack of structural similarity to other proteins in the PDB. It also assembles multiple copies of CAV1 into disc-shaped complexes similar in overall organization to those observed experimentally, and predicts several interfaces between protomers accurately at atomic-detail including helix-helix interactions and interactions between conserved residues in loops that stabilize the outer rim of the complex. Interestingly, AF2 predicts additional closed disc structures containing between 7 and 15 CAV1 protomers. However, some of these multimers are energetically strained compared to the experimental structure, especially the parallel β-barrel located in the center of the CAV1 complex. These findings highlight the ability of AF2 to correctly predict new protein folds and assemblies at a granular level, identify important protein interaction interfaces, and predict the overall organization of homo-oligomeric assemblies, while missing some elements of higher order complexes. This work posits a new direction for the continued development of AF2 and other deep learning protein structure prediction approaches.