A novel consensus-based computational pipeline that reliably predicts hACE2-RBD binding affinity

In a recent study posted to the bioRxiv* preprint server, researchers developed a novel computational model to predict the binding affinity between the human angiotensin-converting enzyme 2 (hACE2) receptor and the receptor-binding domain (RBD) of spike (S) protein of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Additionally, they demonstrated the binding affinity of SARS-CoV-2 RBD to a panel of neutralizing antibodies (NAbs) in correlation with the experimental neutralization data.

Study: A novel consensus-based computational pipeline for rapid screening of antibody therapeutics for efficacy against SARS-CoV-2 variants of concern including omicron variant. Image Credit: Kateryna Kon/Shutterstock

Within the RBD of the Alpha, Beta, Delta, and Omicron variants of concern (VOCs), there are one, three, two, and ten mutations, respectively. Studies have shown that these mutations increase transmissibility by increasing the binding affinity of SARS-CoV-2 RBD to the hACE2 receptor. Additionally, they reduce viral neutralization by NAbs and post-immunization serum.

In the event of the emergence of new SARS-CoV-2 VOCs, all NAbs will have to be tested experimentally against the new VOCs to discover the NAbs that have retained their neutralizing ability. Experimentally, this would be resource-intensive, time-consuming, and costly. 

In such a scenario, computational methods that mimic the experimental binding affinity of SARS-CoV-2 S protein to the hACE2 receptor could prove to be a boon by aiding rapid screening of a pool of current NAbs. Further, these could help select NAbs for further neutralization testing in the lab.

About the study

In the present study, researchers selected 67 NAbs from the Coronavirus Antibody Database that targeted SARS-CoV-2 S RBD, whose experimental structures were known and did not have any missing or non-standard amino acid residues at the RBD-NAb interfaces. They extracted the protein structures of all the NAbs that met the eligibility criteria from the protein databank.

Genomic sequences encoding the S RBD were aligned using MAFFT from multiple complete genomic sequences representing distinct SARS-CoV-2 VOCs, viz., Alpha, Beta, Delta, and Omicron. The researchers sourced these from the EpiCoVTM database maintained by the global initiative on sharing all influenza data (GISAID).

Subsequently, they used RBD protein sequences corresponding to SARS-CoV-2 VOCs for computational modeling of their RBD structures. Further, they predicted a preliminary model of RBD protein structures using four state-of-the-art algorithms, viz., I-Tasser, Modeller, Rosetta, and AlphaFold2. Of the 20 model structures for a single RBD sequence, they selected the best model structure using a consensus approach employing protein structure quality evaluation by ModFold8, ProTSAV, and ProFitFun. Finally, they developed an improvised model structure for each RBD protein using GalaxyRefine, a comprehensive protein structure refinement approach.

For developing five model structures of the RBD-NAb complex for each NAb, the authors used the template-based Assembly of Complex Structures (TACOS) tool and performed a template-based protein-protein docking. Using Modeller, five more RBD-NAb complex structures were modeled by the researchers. In all, there were 10 RBD-NAb complex structures for each NAb.

However, the best RBD-NAb protein complex, selected using GalaxyRefineComplex, was used to predict binding affinity to assess their ability to neutralize SARS-CoV-2 VOCs and identify the molecular interactions at the RBD-NAb interface.

Study findings

The authors compared the computationally modeled RBD structures with their experimentally solved structures to show that the root mean squared deviation was very low, with values of 0.36 Å, 0.55 Å, 0.27 Å, 0.43 Å, and 0.85 Å for Alpha, Beta, Gamma, Delta, and Omicron, respectively.

Thus, establishing the computational pipeline as a reliable method for RBD protein structure prediction and refinement for emerging and new SARS-CoV-2 variants, whose experimental structure remains undetermined. It may also be used for downstream applications in the future.

The results showed a substantial correlation between the predicted dissociation constant (K_d) and the experimental binding affinity data, thus making K_d the most reliable metric to assess the binding affinity in both hetero-dimeric and heterotrimeric protein complexes.

Subsequently, consistent with the experimental binding affinity estimates by surface plasmon resonance, the predicted binding affinity to hACE2 receptor (via K_d) estimated in the current study was ~3.5, ~1.2, ~2.5, and ~5.3–fold higher for Alpha, Beta, Gamma, and Delta VOCs, respectively. Notably, Omicron RBD showed the highest predicted binding affinity to the hACE2 receptor, ~9.5–fold greater than wild-type (WT) SARS-CoV-2 strain, concordant with experimental binding affinity estimates.

The authors noted that 90% of NAbs (60/67) were ineffective against SARS-CoV-2 Omicron VOC, none acted broadly against all VOCs; likewise, Regeneron's REGN-COV2 antibody cocktail failed to neutralize Omicron. All these observations were in line with the results of experimental neutralization assays, thus, suggesting that the computational method was also suitable for screening the NAbs for their efficacy against SARS-CoV-2 VOCs.

Interestingly, four NAbs, GH-12, P2B-1A1, Asarnow_3D11, and C118, showed increased predicted binding affinity to Omicron RBD than WT SARS-CoV-2, indicating that further experimental research is needed to determine their neutralization potential.

Conclusions 

To conclude, the study demonstrated the utility of a novel three-fold validated two-step computational pathway for predicting binding affinity in hetero-dimeric and hetero-trimeric RBD-hACE2 complexes. This method could help identify a range of customizable NAbs, useful as anti-SARS-CoV-2 therapy, for combating current and emerging SARS-CoV-2 variants.

*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.

Written by

Neha Mathur

Neha Mathur has a Master’s degree in Biotechnology and extensive experience in digital marketing. She is passionate about reading and music. When she is not working, Neha likes to cook and travel.