Over the past decades, emergence of SARS-CoV-1 (Severe acute respiratory syndrome coronavirus, MERS-CoV (Middle east respiratory syndrome coronavirus), and SARS-CoV-2 (Severe acute respiratory syndrome coronavirus 2), has posed a significant threat to global public health. In particular, the COVID-19 pandemic caused by SARS-CoV-2 variants has resulted in over 772 million confirmed cases and nearly seven million deaths globally [1]. Monoclonal antibodies (mAbs) with high target specificity and prolonged circulating serum half-lives, are considered a promising therapeutic avenue for combating SARS-CoV-2 variants [2,3]. Several therapeutic mAbs have been developed in the fight against SARS-CoV-2 variants [4].

Structural understanding of the interactions between mAbs and SARS-CoV-2 variants at the molecular level plays an important role in the design of new therapeutic antibodies as well as in the further advancement of the immunological studies. Crystallography is the most common methodology in the structural study of protein-protein complexes. However, it needs a time and cost, which makes it inefficient in the study on various mutational effects in protein-protein binding [5]. Crystallographic study on the binding of mAb and SARS-CoV-2 variants is the representative case. There are continuous viral mutation and evolution, which makes it difficult to perform experimental structural study for the respective variants and developed antibodies.

Computational modeling is an alternative approach in the structural study of protein-protein complexes. It allows us to study the structures of protein complexes without the laborious or sometimes difficult procedure of crystallization [6]. It can be efficiently employed in the study of large-scale investigation on various mutational effects in protein-protein binding. Hence, computational modeling can be helpful in the structural study of the interaction between SARS-CoV-2 variants and antibodies, which may provide valuable insights in the understanding of their binding mechanisms in atomic level and foster new therapeutic applications. Protein-Protein docking simulation is a representative computational modeling method to identify a protein-protein binding mode at the molecular level [6]. In general, it searches the stable binding modes of protein pairs using the scoring functions based on binding energies, shape complementarity and other information related to protein-protein interaction.

m396 is a representative antibody which binds to the receptor binding domain (RBD) of SARS-CoV-1 spike protein and expected to be a therapeutic antibody against SARS-CoV-1 infection [7,8]. The crystal structure of the m396 and SARS-CoV-1 RBD complex was determined, which identified the critical residues in the binding interface of the complex [7]. Interestingly, it has been known that the m396 has no binding affinity to the RBD of SARS-CoV-2 wild type despite the high sequence and structural similarity between SARS-CoV-1 RBD and SARS-CoV-2 RBD [9,10]. In this study, we demonstrate that the m396 scFv has a binding affinity against the RBD of SARS-CoV-2 Beta variant. Additionally, the binding mode of m396 and RBD of SARS-CoV-2 Beta variant was investigated through HADDOCK protein-protein docking tool.

2.1. Synthesis and expression of m396 scFv and SARSCoV RBDs

m396 scFv was synthesized using the oPools™ Oligo Pools service from Integrated DNA Technologies Inc. (Coralville, IA, USA). The scFv genes were subcloned into the pMopac12-NlpAFLAG vector [11] at SfiI sites to create the pMopac12-NlpA-m396 scFv-FLAG construct, which was then transformed into Escherichia coli Jude1(F’ [Tn10(TetR) proAB+lacIq Δ(lacZ)M15] mcrA Δ(mrr-hsdRMS-mcrBC) ϕ80 dlacZΔM15 ΔlacX74 deoR recA1 araD139 Δ(ara leu)7697 galU galK rpsL endA1 nupG) [12]. The spike receptor binding domains (RBD) of SARS-CoV-1 and SARS-CoV-2 were synthesized via assembly PCR and fused with superfolder GFP (sfGFP). The SARS-CoV-2 Beta (B.1.351) RBD was constructed using QuikChange™ site-directed mutagenesis and fused with streptavidin. These RBD plasmids were transfected into Expi293F cells for expression and purified using Ni-NTA agarose resin (Qiagen, Hilden, Germany). The SARS-CoV-2 Beta RBD was fluorescently labeled using an Alexa Fluor 488 protein labeling kit, following the manufacturer's instructions.

2.2. Binding analysis of m396 scFv to SARS-CoV RBDs

The previously constructed m396 scFv-expressing cells were cultured overnight and induced with 1 mM IPTG at 25°C for 5 hours, and then converted into spheroplasts to prepare the periplasmic proteins of E. coli for analysis. The SARS-CoV-1 RBD-sfGFP, SARS-CoV-2 RBD-sfGFP, and SARS-CoV-2 Beta RBD-Streptavidin-Alexa 488 were respectively incubated with the spheroplasts at room temperature for 1 hour, after which their binding affinities were analyzed using a flow cytometer. During the binding analysis of SARS-CoV-2 Beta RBD, the expression level of m396 scFv was monitored by additional labeling with Anti-FLAG-647, which detects the C-terminal FLAG tag. Finally, the binding affinity results were analyzed using the FlowJo 7 program (BD, Ashland, US).

2.3. HADDOCK protein-protein docking and structural analysis

SARS-CoV-1-m396 complex structure (PDB ID: 2DD8) was downloaded from the Protein Data Bank (PDB) (https://www.rcsb.org/). The original crystal complex was separated and the separated structures were then prepared by removing water molecules and pseudo atoms for further analysis using the PyMOL tool. The structures were then checked for imperfections and missing atoms in the SPDBV tool (http://www.expasy.org/spdbv/). To stabilize the structure and reduce steric hindrances, energy minimization was performed in GROMACS to get the most stable conformation for the mutated Fc structure. AMBER96F4 force field was applied, and energy minimization was performed in 50,000 steps. The prepared protein structures were docked using HADDOCK (High Ambiguity Driven protein-protein DOCKing), an information-driven flexible docking tool for the modeling of biomolecular complexes. The two prepared PDB files were uploaded and the active residues were specified for m396 and RBDs, respectively in the HADDOCK web server (https://haddock.science.uu.nl/). DIMPLOT generates schematic diagrams of protein-protein interactions for a given PDB file. The information from DIMPLOT were defined as active residues in the docking. Based on these binding site residues, AIR (Ambiguous Interaction Restraints) data was introduced in docking. The docked structures were aligned with the crystal structure using VMD (Visual molecular dynamics) [13] and their RMSDs were calculated. The molecular interactions in the final model were analyzed using the DIMPLOT program in LIGPLOT+ software [14].

3.1. Binding analysis of m396 to RBDs of SARS-CoV-1 and SARS-CoV-2 wild-type

As described in the Introduction section, it is known that m396 exhibits strong binding affinity to SARS-CoV-1, but does not bind to SARS-CoV-2 [9,10]. To confirm this, the m396 single-chain variable fragment (scFv) was displayed on the periplasmic side of the E. coli inner membrane. Due to challenges in producing and purifying soluble scFv in E. coli, direct measurement of KD values was not feasible. Instead, flow cytometry was employed as an alternative method to evaluate the binding affinity of m396 scFv to sfGFP-fused SARS-CoV-1 and SARS-CoV-2 RBDs. Although flow cytometry does not provide quantitative KD values, it enables the comparative assessment of binding interactions by measuring fluorescence intensity, where a higher mean fluorescence intensity (MFI) indicates stronger binding affinity. Human IgG1 Fc derived from trastuzumab which does not have any binding affinity to both RBDs was employed as a negative control. In the analysis, the m396 scFv exhibited very strong binding affinity to the SARS-CoV-1 RBD (Fig. 1(a) and Fig. 1(b)), consistent with previous results. On the other hand, m396 scFv, similar to the negative control, did not bind to the SARS-CoV-2 wild-type RBD (Fig. 1(c) and Fig. 1(d)). These results confirm that, despite the high sequence and structural similarity between the SARS-CoV-1 RBD and the SARS-CoV-2 wild-type RBD, m396 can be used as a SARS-CoV-1 neutralizing antibody but cannot be used as a SARS-CoV-2 neutralizing antibody.

Figure 1. Binding affinity analysis of m396 to SARS-CoV-1 RBD and SARS-CoV-2 wild-type RBD.
(a) Median fluorescence intensity (MFI) comparison between m396 scFv and the negative control (human IgG1 Fc) for SARS-CoV-1 RBD binding. (b) Flow cytometry histogram showing the binding of m396 scFv to SARS-CoV-1 RBD compared to the negative control. (c) MFI comparison between m396 scFv and the negative control for SARS-CoV-2 RBD binding. (d) Flow cytometry histogram illustrating m396 scFv binding to SARS-CoV-2 RBD versus the negative control.

3.2. Binding analysis of m396 to RBD of SARS-CoV-2 Beta (B.1.351) variant

To further investigate the binding capabilities of m396 scFv, we evaluated its interaction with the SARS-CoV-2 Beta variant RBD which includes the K417N, E484K, and N501Y mutations [15]. To confirm the binding of m396 scFv displayed on the E. coli spheroplasts to the SARS-CoV-2 Beta RBD and simultaneously assess m396 scFv expression, we used Alexa 488-labeled SARS-CoV-2 Beta RBD-streptavidin (Fig. 2, green) along with anti-FLAG-647, which binds to the C-terminal FLAG tag (Fig. 2, red), in a combined binding assay (Fig. 2, cyan). A human IgG1 Fc region, which does not bind to either RBD, was used as a negative control. Flow cytometry analysis showed that the negative control exhibited no binding affinity to the SARS-CoV-2 Beta RBD (Fig. 2(a)). Although it was expected that m396 scFv would not bind to the SARS-CoV-2 Beta RBD, similar to its lack of binding to the SARS-CoV-2 wild-type RBD (Fig. 1(c) and Fig. 1(d)), m396 scFv demonstrated significantly higher binding affinity to the SARS-CoV-2 Beta RBD (Fig. 2(b)). Furthermore, m396 scFv showed even stronger binding affinity than CR3022 scFv, a well-known antibody with high affinity to the SARS-CoV-2 Beta RBD (Fig. 2(c)). This finding suggests that m396 scFv can effectively recognize and bind to the RBD of the SARS-CoV-2 Beta variant, indicating its potential therapeutic applicability against this variant.

Figure 2. Binding analysis of m396 scFv and SARS-CoV-2 Beta (B.1.351) RBD.
(a) Flow cytometry analysis of the negative control (human IgG1 Fc) showing minimal binding to SARS-CoV-2 Beta RBD (x-axis) versus protein expression levels (y-axis). (b) Flow cytometry analysis of m396 scFv showing significant binding to SARS-CoV-2 Beta RBD (x-axis) versus protein expression levels (y-axis), indicating a strong affinity of m396 scFv for the SARS-CoV-2 Beta RBD. (c) Flow cytometry analysis of the positive control (CR3022 scFv) binding to SARS-CoV-2 Beta RBD (x-axis) versus protein expression levels (y-axis).

3.3. Optimization of HADDOCK docking condition for m396-RBD complex

It is very interesting that the m396 shows high binding affinity to SARS-CoV-2 Beta RBD but not to SARS-CoV-2 wild type RBD because there are only three mutations (K417N, E484K, and N501Y) in the RBD protein. To understand this result structurally, the HADDOCK protein-protein docking tool was employed to model the protein complex structure. We first optimized the docking condition for m396 scFv and RBD protein. HADDOCK (High Ambiguity Driven protein-protein Docking) is one of the popular molecular docking tools used in the study of protein-protein interaction. HADDOCK makes use of experimentally determined biophysical and biochemical interaction data as information for protein-protein docking [16]. Specifically, the information of residues involved in protein-protein interaction is derived from experimental data and introduced as ambiguous interaction restraints (AIRs) to drive the docking. The determination of such residues called “active residues” in the HADDOCK protocol is known to be a major parameter to generate efficient AIRs, which eventually affects the protein-protein docking efficiency [17].

Here, we determined the active residues of m396 scFv and RBD which showed the higher docking efficiency by comparing crystal structure of m396-RBD complex and docked complex. Since SARS-CoV-2 and m396 do not have a complex structure, SARS-CoV-1-m396 complex structure (PDB ID: 2DD8) was downloaded and redocked to identify optimized active residues. Four different sets of residues in the binding sites were specified for RBD(CoV-1) and m396. Set 1 was the residues that were selected based on the hydrogen bond interactions. Set 2, Set 3 and Set 4 were the residues that were selected based on 4 Å, 5 Å, and 6 Å distances in the binding interface. The docking was performed using the four active residue sets and the modeled structures were compared with the crystal structure by measuring their RMSD values. Table 1 lists the RMSD values of the modeled structures based on the different set of active residues. The RMSD value is a structural deviation between X-ray structure and docked structure. Therefore, the lower the value is, the lower the deviation is. In general, the RMSD less than 1Å is accepted as a good model. In the analysis, all sets produced good models but we selected set 4 which showed the least deviation. Based on this result, the Set 4 residues were used in the modeling of m396-RBD(CoV-2 Beta) complex using HADDOCK.

Table 1 RMSDs of modeled structure for m396-RBD(CoV-1) complex

Active resdiues	Set 1	Set 2	Set 3	Set 4
RMSD (Å)	0.87	0.69	0.69	0.55

3.4. Analysis of binding mode of m396-RBD(CoV-2 Beta) complex

To model the complex structure of m396-RBD(CoV-2 Beta), m396 and RBD(CoV-2 Beta) structures were docked using the Set 4 active residues. Specifically, RBD sequences of SARS-CoV-1 and SARS-CoV-2 which share 80% percent sequence similarity, were aligned and the CoV-2 RBD residues specified for the Set were selected for the docking procedure. Six clusters were obtained and the structure with the best HADDOCK score was selected as a model.

In the analysis of binding mode, we first analyzed two electrostatic interactions i.e hydrogen bonding and salt-bridge in the binding interface. There are 24 residue pairs of hydrogen bonds (Table 2). 15 residues of RBD were involved in the hydrogen bonds, in which R408 was identified to form hydrogen bonds with 5 different residues in m396. 17 residues of m396 were involved in the hydrogen bonds, in which N58(H, heavy chain) and S93(L, light chain) were identified to form hydrogen bonds with 3 different residues in RBD. Two salt-bridge residue pairs were identified (Table 3). The three mutated residues in RBD (417N, 484K, 501Y) did not form any hydrogen bonds and salt-bridges in this analysis. We presume that the N501Y mutation is the critical mutation which induces the binding of m396 to CoV-2 Beta RBD. It was previously predicted that T501N mutation (based on SARS-CoV-2 RBD sequence) might disrupt the hydrophobic interaction with m396 and lead to the loss of binding affinity to SARS-CoV-2 RBD [9]. Hence, there is a possibility that the N501Y mutation in the Beta variant may restore the hydrophobic interactions with m396 residues. In the docked model, 5 residues in m396 were determined to form hydrophobic interactions with Y501 residue of RBD(CoV-2 Beta) (Table 4). Y498Q mutation of RBD was also previously predicted to be a critical mutation which induces a different binding affinity of m396 to CoV-1 RBD and CoV-2 RBD [9]. In this analysis, 498Q showed a hydrogen bond with S31(H) of m396 (Table 2). Fig. 3(a), (b) and (c) illustrate the interacting residues in the Table 2, 3, and 4, respectively.

Figure 3. Structure of the m396 binding to the SARS-CoV-2 Beta RBD.
(a) Overall structure of the m396 (light chain in cyan, heavy chain in yellow) bound to SARS-CoV-2 Beta RBD (green). Key interacting residues forming hydrogen bonds between m396 and RBD are labeled, corresponding to Table 2. (b) Close-up of the m396-RBD interface showing residues forming salt-bridges between the antibody and the RBD, corresponding to Table 3. (c) Hydrophobic interactions involving Y501 of the RBD and m396 antibody residues, corresponding to Table 4.

Table 2 Residues forming hydrogen bond between RBD (CoV-2 Beta) and m396

RBD (CoV-2 Beta)	m396
ALA(372)	ASN(27,L)
PHE(374)	ASN(27,L)
SER(375)	SER(30,L), LYS(31,L)
THR(376)	SER (93 L)
LYS(378)	SER(93,L), SER(94,L)
ASP(405)	ASN(58,H)
ARG(408)	ASN(58,H), TYR (59,H), GLN(61,H), SER(93,L), ASP (95,H)
ASN(437)	MET(98,H)
ASN(439)	VAL(97,H)
GLN(498)	SER(31,H)
THR(500)	SER(31,H), THR(33,H)
GLY(502)	THR(33,H)
VAL(503)	THR(96,H), TYR(96,L)
GLY(504)	TYR(96,L)
TYR(505)	GLY(50,H), ILE(56,H), ASN(58,H)

Table 3 Residues forming salt-bridge between RBD (CoV-2 Beta) and m396

RBD (CoV2-Beta)	m396
LYS(378)	ASP(92,L)
ARG(408)	ASP(95,H)

Table 4 Residues of m396 forming hydrophobic interaction with Y501 of RBD(CoV-2 Beta)

RBD (CoV-2 Beta)	m396
TYR(501)	SER(31,H), THR(52,H), ILE(53,H), LEU(54,H), VAL(97,H)

In this study, we identified the interacting residues and molecular interactions of the best docked mode, but the docked structure may include some incorrect information caused by the fundamental limitation of HADDOCK docking. Representatively, HADDOCK doesn’t accept glycan group in the modeling process, and thus the glycan groups in the proteins were not included in our computational modeling. It is known that the presence of glycan groups affects the protein-protein binding, and therefore it is difficult to exclude the effect of glycan group in the modeling completely. Experimental structural studies may be further needed to verify our computational modeling results.

m396 is the representative antibody which binds to the RBD of SARS-CoV-1. In this study, we have demonstrated that the m396 scFv which does not have a binding affinity to RBD of SARS-CoV-2 wild type shows binding affinity to RBD of SARS-CoV-2 Beta variant. Docking condition of HAADOCK protein-protein tool was optimized to model the structure of m396-RBD(CoV-2 Beta) complex. The analysis of molecular interactions in the modeled complex structure identified hydrogen bonds and saltbridges in the binding interface. It was presumed that the mutation of N501 in RBD(CoV-2 wild-type) to Y501 in RBD(CoV-2 Beta) might be the critical mutation which induces the binding of m396 to RBD(CoV-2 Beta) by restoring the hydrophobic interaction. This study is meaningful in following aspects. First, this study suggests that m396 can be a template antibody for SARS-CoV-2 variants. Since it has a binding affinity to the Beta variant of SARS-CoV-2, engineering of m396 may allow us to develop antibodies for neutralizing other SARS-CoV-2 variants, Second, the prediction of binding mode between m396 and RBD of CoV-2 Beta provides more insight in the binding of mAb and SARS-CoV-2 variants. Finally, the optimized HADDOCK docking condition is expected to be utilized in the various computational studies on the binding of m396 and its variants to the RBDs of SARS-CoV-2 variants.

This work was supported by a 2-Year Research Grant for Pusan National University.