In a recent study posted to bioRxiv*, researchers determined the cryo-electron microscopic (cryo-EM) structures of engineered angiotensin-converting enzyme 2 (ACE2) receptor traps.
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has acquired multiple mutations throughout the coronavirus disease 2019 (COVID-19) pandemic. SARS-CoV-2 Omicron has 37 mutations in the spike protein. The N-terminal (NTD) and receptor-binding (RBD) domains of the spike protein contain 11 and 15 mutations, respectively, resulting in lower neutralization by plasma from convalescents or fully vaccinated subjects.
Previously, the authors engineered ACE2 receptor traps for SARS-CoV-2 neutralization. They were designed computationally and affinity-optimized by yeast surface display. ACE2 domains were fused with fragment crystallizable (Fc) domain of human IgG1 for additional binding avidity and to a neonatal Fc receptor for increased half-life. The structures of computationally designed (CVD293) and affinity matured (CVD313) ACE2-Fc fusion constructs bound to RBD have not been determined.
The study and findings
In the present study, researchers reduced the linker length between ACE2 and Fc to generate a new construct (CVD432) and resolved cryo-EM structures of the engineered traps (CVD293 and CVD432) bound to whole spike protein. First, the authors verified that CVD293 and CVD432 neutralize wild-type (WT) spike pseudo-typed virus.
The WT spike-CVD293 complex had a 1-RBD-up state with full ACE2 occupancy and an appreciable percentage of 1-RBD-up state with partial ACE2 occupancy, a 2-RBD-up state with 1-ACE2 occupancy, and 1-RBD-up state with no ACE2 occupancy, per trimer. In contrast, the WT spike-CVD432 complex showed a 1-RBD-up state with full ACE2 occupancy, a 2-RBD-up state with 2-ACE2 occupancy, an all-RBD-down state, and other partial- or no-ACE2 occupancy 1- or 2-RBD-up states.
Next, the team developed a multi-model workflow combining cryo-EM and Rosetta protein modeling to compute the average predicted interface energy. 10-residue overlapping stretches of ACE2 interface of each cryo-EM model (spike-CVD239 and spike-CVD432) were subjected to a CartesianSampler mover in Rosetta to generate 2000 models for each 10-residue stretch. These models were all-atom minimized in the cryo-EM map using FastRelax mover. The refinement protocol was iterated to generate nearly 8000 models.
The atomic models were ranked based on Rosetta scores, and 80 models were selected. The interface helix residues of the 80 models were superimposed to examine the convergence of side-chain conformations and intermolecular interactions. They noted that Q35 residue in CVD293 and CVD432, a high average side-chain root mean square deviation (RMSD) residue, formed a hydrogen bond with the Q493 residue of WT spike in more than 90% of atomic models.
Further, they noted that low average side-chain RMSD residues in CVD293 and CVD432 formed hydrophobic interactions with the corresponding spike residues. The authors reported that the low average side-chain RMSD hydrophobic residues engineered in the receptor traps improved the binding affinity.
The predicted interface energy for CVD293 was lower (-58 Rosetta energy units, REU) than the mean interface energy for the 80 models (-45 REU) due to differences in the side-chain-mediated interactions. Q35 residue engineered in CVD293 had the largest average side-chain RMSD per residue.
Multiple spike mutations of SARS-CoV-2 Omicron have been identified in other variants, albeit the variant has 14 unique modifications that improve its binding affinity. Next, the researchers assessed the binding of Omicron RBD to engineered ACE2 traps. To this end, they generated models of Omicron RBD-CVD293 and Omicron RBD-CVD432 complex by superimposing and substituting WT RBD in cryo-EM local refinement with Omicron RBD and minimized the complexes.
The interface energy for residues in the Omicron RBD-CVD293 complex was 10.77 REU and -8.5 REU for the Omicron RBD-CVD432 complex, the interface energy for Omicron RBD-WT ACE2 was -4.99 REU. The authors performed biolayer interferometry (BLI) of CVD293 or CVD432 with Omicron RBD to test whether the predicted interface energies corresponded to apparent binding affinities.
The dissociation constants (KD) for Omicron RBD-CVD293 (4.2 nM) and Omicron RBD-CVD432 (0.53 nM) were 10- and-100-fold lower than that for Omicron RBD-WT RBD, respectively. Moreover, neutralization assays were performed with recombinant vesicular stomatitis virus pseudo-typed with Delta or Omicron spike. CVD293 and CVD432 neutralized Delta and Omicron pseudoviruses, with 2-to-20-fold improvements in half-maximal inhibition concentrations (IC50) over WT spike.
Conclusions
In summary, the authors determined cryo-EM structures of engineered ACE2 traps complexed with WT spike. Although informative, the cryo-EM provided a limited resolution at the ACE2-RBD interface, which prompted the development of a multi-model cryo-EM:Rosetta pipeline.
This pipeline revealed that distributed binding interactions at the interface between the two proteins were more effective than one or two interactions at the interface and that the stability of individual proteins was equally as important as the stability of the complex.
Further, Omicron RBD binding to receptor traps was experimentally validated using BLI and pseudovirus neutralization, which showed that ACE2 traps designed for the WT spike were robust against the mutant spike (of Omicron). Overall, the study exemplified how cryo-EM and computation modeling could be combined to improve the design-build-test cycle to engineer biotherapeutics.
*Important notice
bioRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.