Insights on cross-species transmission of SARS-CoV-2 from structural modeling

Abstract

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is responsible for the ongoing global pandemic that has infected more than 31 million people in more than 180 countries worldwide. Like other coronaviruses, SARS-CoV-2 is thought to have been transmitted to humans from wild animals. Given the scale and widespread geographical distribution of the current pandemic and confirmed cases of cross-species transmission, the question of the extent to which this transmission is possible emerges, as well as what molecular features distinguish susceptible from non-susceptible animal species. Here, we investigated the structural properties of several ACE2 orthologs bound to the SARS-CoV-2 spike protein. We found that species known not to be susceptible to SARS-CoV-2 infection have non-conservative mutations in several ACE2 amino acid residues that disrupt key polar and charged contacts with the viral spike protein. Our models also allow us to predict affinity-enhancing mutations that could be used to design ACE2 variants for therapeutic purposes. Finally, our study provides a blueprint for modeling viral-host protein interactions and highlights several important considerations when designing these computational studies and analyzing their results.

Fig 1. Interface statistics of modeled ACE2:RBD complexes.

SARS-CoV-2^pos species (in blue) generally have lower (better) HADDOCK scores (left, expressed in arbitrary units) than SARS-CoV-2^neg species (in red). A similar but less conclusive trend is observed between the sequence similarities of amino acid residues interacting with the viral RBD (derived from PDB 6m17) (right). Collectively, these results suggest that SARS-CoV-2^neg species lack specific key ACE2 amino acid residues that form strong interactions with the viral spike protein, leading to impaired binding between the two proteins. Species are ordered in increasing order of HADDOCK score. Species for which SARS-CoV-2 susceptibility is unknown or assays were inconclusive are shown in gray.

Fig 2. Correlation of HADDOCK score with individual energy terms and structural features.

Differences in electrostatics energy contribute the most towards discriminating SARS-CoV-2^pos species (blue) from SARS-CoV-2^neg species (red), supporting observations of hydrogen bonding networks and charged interactions in experimental structures. The buried surface area of the models is also strongly correlated with their HADDOCK score, suggesting larger interfaces of SARS-CoV-2^pos species might confer better binding properties.

Fig 3. HADDOCK score of individual ACE2 interface residues.

For each species (row), the blocks (columns) represent amino acid residues within 5Å of the viral RBD in any of the species’ best 10 models. The identity of the amino acid is shown in one-letter code. The colors represent the HADDOCK score of each residue, averaged over the 10 models: lower scores (dark blue) indicate more favorable interactions, while positive scores indicate steric clashes or electrostatic repulsion. The first row shows the median of the averages for each column. From this analysis, we predict that amino acid residues at positions 30, 31, and 353 contribute the most to the stability of the ACE2:RBD complex. In SARS-CoV-2^neg species (red labels), some of these residues are consistently mutated (30 and 31), which could explain their lower susceptibility to infection. S3 Fig shows the per-residue analysis for all species in the dataset.

Fig 4. Interface differences between human and SARS-CoV-2^neg models.

The top panels show key residue-residue interactions at the interface between hACE2 (white) and the viral RBD (teal), which are conserved in nearly all SARS-CoV-2^pos species: salt-bridge between D30 and K417 (left); three-body interaction between K31, E35, and RBD Q493 (middle); and the interactions of K353, an intramolecular salt-bridge with D38 and an intermolecular hydrogen bonds with G496 and G502 (right). The bottom panels highlight equivalent regions in three SARS-CoV-2^neg species: D30N mutation in mice (left) disrupts the intermolecular salt-bridge; D31K/D35R in ducks stabilizes an intramolecular salt-bridge and weakens the intermolecular hydrogen bond (middle); K353H in mice disrupts the intramolecular salt-bridge (right).

Fig 5. ΔHADDOCK score of individual ACE2 interface residues compared to hACE2.

For each species (row), the blocks (columns) represent amino acid residues within 7.5Å of the viral RBD in any of the species’ best 10 models. The identity of the amino acid is shown in one-letter code. The colors represent the ΔHADDOCK score–including intramolecular interactions–of each residue, averaged over the 10 models, compared to the average of the corresponding hACE2 residue: negative scores (dark blue) indicate a stabilizing mutation. This analysis highlights several potential affinity-enhancing mutations, namely Q24E, A25V, D30E, H34Y, F40S, Y41H, F72Y, L79H, and A387E. We note that this analysis requires further visual inspection of the models to account for additional variations in ACE2 sequence that may skew the per-residue HADDOCK score. Refer to the main text for details. S4 Fig shows the same plot for all species of the dataset.

Fig 6. Predicted affinity-enhancing mutations for hACE2.

Analyzing the residue energetics of ACE2 orthologs suggests mutations that have the potential to enhance the affinity of hACE2 (white) to RBD (teal). The top panels shows our top-scoring hACE2:RBD model and its interactions (yellow cylinders) for four such sites: residues 24, 25, 30, and 387. The bottom panels show mutations in specific species, and the resulting new or enhanced interactions: Q24E in pangolin, A25V in dog, D30E in cow, and A387E in ferret. Some of these mutations are found in multiple ACE2 orthologs.