Concordance of X-ray and AlphaFold2 Models of SARS-CoV-2 Main Protease with Residual Dipolar Couplings Measured in Solution

Abstract

The 68-kDa homodimeric 3C-like protease of SARS-CoV-2, M^pro (3CLpro/Nsp5), is a promising antiviral drug target. We evaluate the concordance of models generated by the newly introduced AlphaFold2 structure prediction program with residual dipolar couplings (RDCs) measured in solution for ¹⁵N-¹H^N and ¹³C'-¹H^N atom pairs. The latter were measured using a new, highly precise TROSY-AntiTROSY Encoded RDC (TATER) experiment. Three sets of AlphaFold2 models were evaluated: (1) M^pro_AF, generated using the standard AlphaFold2 input structural database; (2) M^pro_AFD, where the AlphaFold2 implementation was modified to exclude all candidate template X-ray structures deposited after Jan 1, 2020; and (3) M^pro_AFS, which excluded all structures homologous to coronaviral M^pro. Close agreement between all three sets of AlphaFold models and experimental RDC data is found for most of the protein. For residues in well-defined secondary structure, the agreement decreases somewhat upon Amber relaxation. For these regions, M^pro_AF agreement exceeds that of most high-resolution X-ray structures. Residues from domain 2 that comprise elements of both the active site and the homo-dimerization interface fit less well across all structures. These results indicate novel opportunities for combining experimentation with molecular dynamics simulations, where solution RDCs provide highly precise input for QM/MM simulations of substrate binding/reaction trajectories.