Deep Mutational Scanning and Computational Modeling of SARS-CoV-2 Spike RBD: Predicting Escape from Neutralizing Antibodies
By Dr. Zubair Khalid, DVM, MS, PhD ·
Key Takeaways
- Deep Mutational Scanning (DMS) is a high-throughput experimental method that systematically quantifies the functional impact of all single amino acid mutations in a protein, such as the SARS-CoV-2 spike receptor-binding domain (RBD), on phenotypes like ACE2 binding and antibody neutralization.
- Computational modeling, including physics-based approaches (Rosetta, molecular dynamics) and deep learning (AlphaFold2, protein language models), is crucial for interpreting DMS data, predicting the structural and energetic consequences of mutations, and understanding antibody-RBD interactions.
- The integration of DMS and computational modeling creates a predictive pipeline that can identify mutations conferring antibody escape, inform the design of broadly neutralizing antibodies and vaccines, and forecast viral evolution trajectories.
- This combined methodology is directly applicable to veterinary coronaviruses, enabling the mapping of escape mutations in animal-specific spike proteins to guide the development of animal vaccines and enhance surveillance for zoonotic spillover events.
- Epistatic interactions, where the effect of one mutation depends on the presence of others, are critical for predicting the emergence of novel variants and can be captured by advanced DMS and computational modeling approaches, though shifts in these interactions over time can complicate forecasting.
Introduction
The receptor-binding domain (RBD) of the SARS-CoV-2 spike glycoprotein mediates viral attachment to the angiotensin-converting enzyme 2 (ACE2) receptor and is a dominant target of neutralizing antibodies [<a href="#ref-1">1</a>]. The continuous emergence of viral variants with mutations in the RBD has necessitated the development of high-throughput methods to prospectively identify mutations that confer escape from antibody neutralization [<a href="#ref-2">2</a>, <a href="#ref-3">3</a>]. Deep mutational scanning (DMS) has emerged as a powerful experimental technique to systematically measure the functional effects of all single amino acid mutations in a protein of interest [<a href="#ref-1">1</a>, <a href="#ref-4">4</a>]. When combined with computational modeling approaches such as Rosetta and AlphaFold2, DMS data can be used to predict escape mutations and inform vaccine design strategies for both human and veterinary applications [<a href="#ref-5">5</a>, <a href="#ref-6">6</a>].
SARS-CoV-2 has demonstrated the ability to infect a broad range of mammalian hosts, including felids, canids, mustelids, and cervids, raising concerns about reverse zoonosis and the establishment of animal reservoirs [<a href="#ref-7">7</a>, <a href="#ref-8">8</a>]. Understanding the molecular determinants of antibody escape in the spike RBD is therefore relevant not only for human medicine but also for veterinary surveillance and the development of animal vaccines [<a href="#ref-9">9</a>, <a href="#ref-10">10</a>]. This article reviews the integration of DMS experiments with computational structural biology to predict escape mutations in the SARS-CoV-2 spike RBD, with emphasis on methodologies that can be extended to veterinary coronaviruses.
Deep Mutational Scanning Methodology
DMS involves the generation of comprehensive libraries of mutant variants of a target protein, followed by a functional selection and high-throughput sequencing to quantify the fitness of each variant [<a href="#ref-1">1</a>, <a href="#ref-11">11</a>]. For the SARS-CoV-2 spike RBD, DMS libraries have been constructed using both yeast surface display and pseudovirus-based platforms [<a href="#ref-1">1</a>, <a href="#ref-7">7</a>]. In the seminal study by Starr et al., all possible single amino acid mutations in the RBD were assayed for effects on protein folding and ACE2 binding affinity [<a href="#ref-1">1</a>]. This work revealed that most mutations are deleterious, but a subset of mutations at the ACE2 interface can enhance binding [<a href="#ref-1">1</a>]. Subsequent studies extended DMS to full-length spike proteins using pseudotyped lentiviruses, enabling measurement of mutations affecting ACE2 binding, cell entry, and antibody neutralization [<a href="#ref-2">2</a>, <a href="#ref-7">7</a>, <a href="#ref-11">11</a>].
The pseudovirus DMS platform developed by Dadonaite et al. allows the simultaneous quantification of thousands of mutations in the context of the full spike trimer [<a href="#ref-7">7</a>]. This approach has been applied to Omicron BA.1, Delta, XBB.1.5, and BA.2 spikes, generating comprehensive maps of mutational effects on ACE2 binding and serum neutralization [<a href="#ref-2">2</a>, <a href="#ref-11">11</a>]. Key findings include the identification of strong escape mutations at RBD sites 357, 420, 440, 456, and 473, as well as escape mutations outside the RBD that modulate RBD conformation [<a href="#ref-2">2</a>, <a href="#ref-11">11</a>]. DMS has also been used to characterize the fusion peptide region (S2 domain) and its role in antibody escape [<a href="#ref-8">8</a>, <a href="#ref-9">9</a>].
The DMS data are typically represented as fitness landscapes or heatmaps showing the effect of each amino acid substitution on a given phenotype [<a href="#ref-1">1</a>, <a href="#ref-3">3</a>]. These landscapes can be used to identify constrained regions that are desirable targets for broadly neutralizing antibodies [<a href="#ref-1">1</a>, <a href="#ref-12">12</a>]. DMS can capture epistatic interactions between mutations, which are critical for predicting the emergence of variants with multiple mutations [<a href="#ref-3">3</a>, <a href="#ref-6">6</a>]. For example, the Q493E mutation reversed its effect on ACE2 binding from deleterious to beneficial when combined with L455S and F456L in the KP.3 variant [<a href="#ref-3">3</a>].
Computational Modeling Approaches
Computational modeling plays a central role in interpreting DMS data and predicting antibody escape. Two major classes of methods are used: physics-based approaches such as Rosetta and molecular dynamics (MD) simulations, and deep learning-based approaches such as AlphaFold2 and protein language models [<a href="#ref-5">5</a>, <a href="#ref-13">13</a>, <a href="#ref-14">14</a>].
Rosetta can be used to calculate binding free energy changes (ΔΔG) upon mutation and to model the structure of RBD-antibody complexes [<a href="#ref-15">15</a>, <a href="#ref-16">16</a>]. Chen et al. developed a neural network that maps decomposed energy terms from MD simulations to experimental binding affinity changes, achieving high accuracy in classifying mutations as improving or worsening ACE2 binding [<a href="#ref-15">15</a>]. Similarly, interface-guided computational protein design has been used to predict bebtelovimab resistance mutations in the RBD, with 69-100% correlation with patient-derived sequences [<a href="#ref-16">16</a>].
AlphaFold2 and its successors (AlphaFold3, Boltz-2, Protenix, Chai-1) have been evaluated for their ability to predict RBD-antibody complex structures [<a href="#ref-13">13</a>]. Morozova et al. found that AlphaFold3 performed best among AI-driven tools for reproducing known RBD-antibody complexes, while traditional docking tools like ClusPro performed well for small antibodies with defined interfaces [<a href="#ref-13">13</a>]. Integrative computational modeling combining MD simulations, mutational scanning, and MM-GBSA binding free energy calculations has elucidated the binding mechanisms of broadly neutralizing antibodies such as S309, S304, CYFN1006, and VIR-7229 [<a href="#ref-14">14</a>, <a href="#ref-17">17</a>]. These studies identified key residues that serve as evolutionary "weak spots" balancing viral fitness and immune evasion [<a href="#ref-14">14</a>].
Machine learning models trained on DMS data can predict the effects of combinatorial mutations and extrapolate to unseen variants [<a href="#ref-4">4</a>, <a href="#ref-5">5</a>, <a href="#ref-10">10</a>]. Durumeric et al. trained supervised models on RBD DMS libraries labeled with ACE2 binding affinity and used Markov Chain Monte Carlo simulations to predict evolutionary trajectories [<a href="#ref-5">5</a>]. Protein language models informed by DMS data have been used to predict SARS-CoV-2 evolution dynamics with spatiotemporal resolution [<a href="#ref-10">10</a>]. The integration of genomic epidemiology with DMS data in models such as CoVPF has improved the accuracy of lineage prevalence forecasting by 20.7% compared to genomic data alone [<a href="#ref-6">6</a>].
Integration of DMS and Computational Modeling
The combination of DMS and computational modeling creates a powerful pipeline for predicting antibody escape. The workflow typically involves: (1) experimental DMS to generate mutational fitness landscapes, (2) structural modeling of the RBD-antibody complex, (3) computational prediction of escape mutations using binding energy calculations or machine learning, and (4) experimental validation of predicted escape variants [<a href="#ref-1">1</a>, <a href="#ref-7">7</a>, <a href="#ref-15">15</a>]. This integrated approach has been applied to predict escape from monoclonal antibodies such as bebtelovimab, where DMS identified key residues K444, V445, and G446 as escape hotspots, and subsequent modeling confirmed the structural basis [<a href="#ref-12">12</a>, <a href="#ref-16">16</a>].
The following Mermaid diagram illustrates the integrated workflow:
graph TD
A["Deep Mutational Scanning"] --> B["Fitness Landscapes"]
B --> C["Identify Escape Hotspots"]
D["Structural Modeling"] --> E["RBD-Antibody Complex"]
E --> F["Binding Energy Calculations"]
C --> G["Machine Learning Models"]
F --> G
G --> H["Predict Escape Mutations"]
H --> I["Experimental Validation"]
I --> J["Vaccine Design & Surveillance"]Computational modeling also helps to understand the biophysical basis of escape. For example, electrostatic interactions are the primary determinant of ACE2 binding affinity in Omicron variants, with mutations such as Q493R enhancing binding through salt bridge formation [<a href="#ref-18">18</a>]. Allosteric effects are also important: class 4 antibodies that bind to a cryptic epitope can neutralize through long-range modulation of RBD dynamics rather than direct steric hindrance [<a href="#ref-19">19</a>]. Dynamic network analysis has identified a conserved "allosteric ring" in the RBD core that mediates signal propagation from the antibody epitope to the ACE2 interface [<a href="#ref-19">19</a>].
Predicting Antibody Escape
DMS combined with computational modeling has successfully predicted escape mutations for multiple classes of neutralizing antibodies. For class 1 antibodies targeting the ACE2 binding site, broad epitope coverage and distributed hotspot architecture confer resilience to escape, as seen with antibody BD55-1205 [<a href="#ref-20">20</a>]. In contrast, antibodies with localized binding modes and strong dependence on side-chain interactions are more vulnerable to single mutations such as K417N, L455M, and F456L [<a href="#ref-20">20</a>]. For class 4 antibodies, neutralization efficacy arises from the interplay of direct interfacial interactions and allosteric effects, with group F1 antibodies operating via classic allostery and group F3 antibodies combining direct competition with allosteric stabilization [<a href="#ref-19">19</a>].
The predictive power of DMS has been validated by retrospective analysis of emerging variants. The growth rates of human SARS-CoV-2 clades can be explained in substantial part by DMS-measured effects on spike phenotypes, suggesting that DMS data can enable better prediction of viral evolution [<a href="#ref-2">2</a>, <a href="#ref-11">11</a>]. However, epistatic shifts in mutational effects can hinder forecasting when measurements are made in outdated strain backgrounds [<a href="#ref-3">3</a>]. The BA.2.86 background showed modest epistatic drift from BA.2, but strong positive epistasis among subsequent mutations (e.g., Q493E with L455S and F456L) led to the emergence of the KP.3 variant [<a href="#ref-3">3</a>].
Implications for Veterinary Medicine and Surveillance
The methodologies described above are directly applicable to veterinary coronaviruses, such as feline infectious peritonitis virus (FIPV), transmissible gastroenteritis virus (TGEV), and canine respiratory coronavirus. DMS can be used to map antibody escape mutations in the spike proteins of these viruses, informing the design of broadly protective vaccines for companion animals and livestock [<a href="#ref-9">9</a>, <a href="#ref-10">10</a>]. Computational modeling of RBD-ACE2 interactions can predict host range and spillover risk, as demonstrated for SARS-CoV-2 variants binding to ACE2 orthologs from different species.
The emergence of SARS-CoV-2 variants in animals, such as the mink-associated cluster in Denmark, underscores the need for surveillance tools that can predict which mutations may arise under immune pressure in animal hosts [<a href="#ref-7">7</a>, <a href="#ref-8">8</a>]. DMS data from human SARS-CoV-2 strains can be used to infer potential escape mutations in animal-adapted variants, although species-specific differences in ACE2 sequence must be considered. Computational docking and MD simulations can assess the impact of ACE2 polymorphisms on RBD binding affinity, as shown for human ACE2 variants S19P, K26R, and K341R.
Veterinary vaccine development can benefit from the same integrated DMS-computational pipeline. By identifying conserved epitopes that are resistant to escape, researchers can design vaccines that elicit broadly neutralizing antibodies [<a href="#ref-19">19</a>, <a href="#ref-21">21</a>]. For example, the fusion peptide and stem helix regions of the S2 domain are highly conserved across coronaviruses and represent promising targets for pan-coronavirus vaccines [<a href="#ref-8">8</a>, <a href="#ref-9">9</a>]. DMS of these regions can identify mutations that confer resistance to broadly neutralizing nanobodies, as demonstrated for SARS-CoV-2 S2 [<a href="#ref-9">9</a>].
Conclusion
Deep mutational scanning and computational modeling are complementary approaches that together provide a comprehensive framework for predicting antibody escape mutations in the SARS-CoV-2 spike RBD. DMS generates high-resolution fitness landscapes that capture the effects of all single mutations, while computational methods such as Rosetta, AlphaFold2, and machine learning models interpret these data in structural and energetic terms. The integration of these approaches has successfully predicted escape from monoclonal antibodies and serum neutralization, and has been validated by the emergence of new variants. These methodologies are directly transferable to veterinary coronaviruses and can inform vaccine design and surveillance strategies for animal health. Continued development of computational tools that account for epistasis and species-specific receptor interactions will further enhance our ability to anticipate viral evolution and mitigate the impact of emerging variants in both human and animal populations.
Related Clinical & Scientific Guides
- A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data
- Computational Immunology: Modeling the Immune System
- How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices