Computational Study of Antibody Binding to SARS-CoV-2 Variants

Background/Objectives. The unprecedented structural and binding data for antibodies to 1 the SARS-COV2 virus taken together with the mutations for the spike protein allows for a broad sim- 2 ulation study of antibody-spike protein binding. This provides an understanding of the co-evolution 3 of human immunity and viral immunity escape. Methods: We utilized the YASARA molecular 4 dynamics program to generate initial antibody-spike structures and simulate to equilibration for 5 six SARS-COV2 variants and 10 different antibodies sampling two different binding regions to the 6 receptor binding domain of the spike (especially for the Class I antibodies in the same part of the 7 spike which attaches to the ACE2 receptor protein) and one to the N-terminal of the spike. Starting 8 structures for antibody binding to variant spike proteins are perturbatively achieved through point 9 mutations and insertions/deletions in the YASARA program. We employed YASARA to measure 10 interfacial hydrogen bound counts between antibodies and variant spike proteins, and the HawkDock 11 MMGBSA program to characterize trends in binding energies with mutation for four of the antibodies. 12 We utilized the VMD program to analyze the time course of hydrogen bond populations. Results: As 13 seen in previous studies, interfacial hydrogen bond counts serve as an excellent proxy for binding en- 14 ergies without the large systematic error inherent in the latter. We find that there is generally a decline 15 in antibody binding strength, as measured by interfacial hydrogen bond counts, with viral evolution, 16 but that a modest re-entrance of binding strength is present for most antibodies studied. Generically, 17 the antibody heavy chain binds more strongly to the spike protein, through for approximately half the 18 antibodies the light chain binding strength converges to the heavy chain strength with viral evolution. 19 Conclusions. The key conclusion is that the identified re-entrant immunity, speculatively arising 20 from a balancing of maintenance of ACE2-spike binding while escaping antibodies through mutation, 21 allows for some maintenance and even strengthening of immunity for later viral strains from early 22 infection or vaccination. 23

SAS-CoV2,antibody binding,immunity escape,simulation 24 0. Introduction 25 The SARS-CoV2 virus engendered a worldwide pandemic from March 2020 to May 26 2023 which killed tens of millions worldwide. The success of the virus in escaping immune 27 responses through mutations of the SARS-CoV2 spike protein, which bound both to human 28 angiotensin converting enzyme 2 (ACE2) receptor on the surface of cells and to potentially 29 neutralizing antibodies, contributed to the lethality of disease over time. 30 Accordingly, there has been considerable experimental and computational interest 31 in understanding which mutations through new strains of SARS-CoV2 were key to anti- 32 body escape. Many of the computational studies were oriented to specific antibody-spike 33 interactions for particular strains of the SARS-CoV2 virus. 34 As has been noted elsewhere[1], there is a difficult evolutionary dance the virus plays 35 to escape immunity: most of the effective neutralizing antibodies attach in the same receptor 36 Version March 4, 2026 submitted toAntibodies https://www.mdpi.com/journal/antibodies .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 2 of 10 binding domain to which the virus binds to ACE2. Accordingly, a massive immune escape 37 is likely to lead to a virus with weaker cell binding, and presumably lower cell lethality. 38 We have adopted a different perspective in this study, which computationally examines 39 the binding strength of ten antibodies covering a broad range of binding sites to the spike 40 protein, and examines six variants of the virus from the original strain of 2020 through 41 the BA.2.86 variant predominant near the end of the pandemic in 2023. The antibodies 42 studied include Class I (binding to the same set of residues in the spike receptor binding 43 domain [RBD] as the ACE2), Class III (binding to the RBD but away from the ACE2 binding 44 domain), and spike N-terminus binding. We are unaware of any study in the literature 45 which has covered such a comprehensive range of antibodies and variants. 46 By monitoring the number of interfacial hydrogen bonds between the antibody and 47 the spike protein, we are able to discern some new results relevant to the understanding of 48 the pandemic trajectory. 49 First, while some antibodies display a monotonic decrease in binding strength with 50 strain evolution, many show a partial reentrance, that is, the binding strength rebounds 51 at least partially with time so that the antibodies retain some neutralizing capability. We 52 believe this reflects the difficult evolutionary competition between immune escape and 53 maintaining sufficient ACE2 binding. Second, quite uniformly the heavy chains of the 54 antibody bind more strongly than the light chains. Third, in general, the Class I antibodies 55 bind more strongly than the Class III or N-terminal antibodies studied. Fourth, those Class 56 I antibodies alleged to show higher efficacy for omicron and descendant strains were not 57 found to bind more strongly than the earlier delta or original (wild-type [WT]) strains. 58 Additionally, as we found in previous studies, the interfacial hydrogen bond count 59 serves as a strong proxy for binding free energy which we evaluated separately for a 60 representative subset of the antibodies. 61 The most important emerging qualitative picture from our study is that the viral 62 evolution may provide immune escape from the current extant antibodies, but a global 63 escape from all previous antibodies is likely impossible given that the most efficacious ones 64 bind in the same region as the ACE2, so that high immune escape means weak cellular 65 binding. Also because immune escape is relative to current extant antibodies, reentrance in 66 which at there is at least some restoration of immunity from previous antibodies can lead 67 to a persistent robust population immunity to the evolving virus. 68 1. Materials and Methods 69 1.1. Molecular Models 70 A summary of all the mutations relative to the original (WT) strain in the RBD and 71 N-terminus of the spike protein from the six variants (Delta, BA.1, BA.2, XBB.15, BA.2.86) 72 is found in Table 1. 73 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 3 of 10 Table 1. Mutations in variants relative to original (WT) strain including the D614G mutation. Column 1 lists the variant, Column 2 mutations in the receptor binding domain (RBD) and Column 3 mutations in the N-terminus. Mutations in the ACE2 binding region of Column 2 are identified in bold print. Point residue mutations were represented by XNY, where X is the WT residue, N the sequence number in the WT, and Y the variant residue. Deletions were represented by ∆(N-M), where N is the starting sequence number, M is the ending sequence number, and additions were represented by Σ(N − M) similarly. Sequences are listed at Ref. [2], with the original spike sequence at Ref. [3] Variant RBD Mutations N-Terminus Mutations Delta (B.1.617.2) L452R,T478K T19R,T95I,G142D,Y145H, ∆(156-157),F158G,A222V , ,W258L BA.1 G339D,S371L,S373P ,S375F, A67V ,∆(69,70),T95I,G142D, K417N,N440K,G446S,S477N, ∆(143-145),N211K,∆(212), T478K,E484A,Q493R,G496S, Σ(R214) Q498R,N501Y,Y505H BA.2 G339D,S371F,S373P ,S375F, T19I,L24S,∆(25-27),G142D, T376A,D405N,R408S,K417N, V213G N440K,S477N,T478K,E484A, Q493R,G496S,Q498R,N501Y, Y505H XBB.15 G339H,R346T,L368I,S371F, T19I,L24S,∆(25-27),G142D, S373P ,S375F,T376A,D405N, ∆(144),H146Q,Q183E,V213E, R408S,K417N,N440K,V445P , G252V E484A,G446S,N460K,S477N, T478K,F486P ,F490S,Q498R, N501Y,Y505H BA.2.86 G339H,K356T,S371F,S373P , T19I,R21T,L24S,∆(25-27), S375F,T376A,R403K,D405N, S50L,∆(69,70),V127F,G142D, R408S,K417N,N440K,V445H, ∆(144),F157S,R158G,N211I, G446S,N450D,L452W,N460K, ∆(212),V213G,L216F,H245N, S477N,T488K,N481K,∆(483), A264D,I332V E484K,F486P ,Q498R,N501Y, Y505H BA.2.75 G339H,S371F,S373P ,S375F, T19I,L24S,∆(25-27),G142D, T376A,D405N,R408S,K417N, K147E,W152R,F175L,I210V , N440K,G446S,N460K,S477N, V213G,G257S T478K,E494A,Q498R,N501Y We drew starting structures for RBD-ACE2 binding from the Protein Data Bank. Class 74 I antibodies bind in the same region of the RBD as the ACE2. Here we preferentially 75 use the antibody designations in the literature, but we also refer parenthetically to the 76 Protein Data Bank (PDB) files where the bound structures to relevant RBDs were displayed, 77 P4A1 (7CJF)[4], C1A-B12 (7KFV)[5], 2-15((7L5B)[6], C1A-C2((7KFX)[5], C1A-F10(7KFY)[5], 78 C1A-B3(7KFV)[5], S2X234(8ERQ)[7], and Omi3(7KF3)[ 8] were selected to represent the 79 spectrum of Class I antibodies. Class III antibodies bind to the RBD away from where 80 the ACE2 binds and is represented by CR.3022(6YOR)[9]. For antibodies that bind to the 81 N-terminal domain, we used 4A8(7C2L)[10]. Fig. 1 shows the structures of representative 82 spike domain-antibody complex types studied in this paper. The chosen antibodies were 83 not comprehensive of all the known neutralizing antibodies for the SARS-CoV-2 spike but 84 summarize a variety of antibodies that target the SARS-CoV-2 virus. We did not study 85 T-cell binding. Antibodies were summarized in Table 2. 86 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 4 of 10 Figure 1. Structures of WT spike protein complexes studied. Binding of RBD (red) to Class I Ab P4A1 (binds in ACE2 interface region) and Class III Ab CR3022 (binds away from ACE2). For Class I, III binding, residue away from ACE2 (F347) shown in gray, residues in ACE2 binding region (R403,S477,Y505) shown in yellow. For CR3022 we also highlight K386 in gray Binding of NTD (purple) to Ab 4A8. Ab heavy chain green, light chain cyan. For N-terminal binding, residues near interface (K147,D253) shown in yellow.‘ Graphic representations were created with YASARA[11] Table 2. Antibodies by class studied here, with class in column 1, antibody nomenclature in column 2, relevant PDB entry in column 3 Antibody Class Antibody Label PDB Entry Class I P4A1 7CJF Class I C1A-B12 7KFV Class I C1A-C2 7KFX Class I C1A-F10 7KFY Class I C1A-B3 7KFW Class I 2-15 7L5B Class I S2X234 8ERQ Class I Omi3 7KF3 Class III CR.3022 6YOR N-term 4A8 7C2L To adopt starting model structures for studying binding with molecular dynamics, 87 we picked the relevant earliest bound variant spike-antibody structure from the PDB and 88 mutated the residues point by point within the YASARA modeling suite[11]. hree of these 89 structures are shown in Fig. 1. When deletions arose, particularly in the N-terminus of the 90 spike, we grafted the corresponding ends within YASARA[11]. Insertions and mutations 91 were built upon starting structures using YASARA’s BuildLoop and SwapRes commands, 92 respectively. While this is basically a perturbative approach biased to the starting structures, 93 unlike an unbiased docking approach, it is unlikely to lead to systematic errors of the kind 94 known in docking protocols. 95 1.2. Molecular Dynamics 96 Simulations of the protein-protein interactions were completed with the molecular- 97 modeling package YASARA[11] by searching for minimum-energy conformations of the 98 SARS-CoV-2-Ab complexes. For each structure, we carried out a energy minimization (EM) 99 routine, which includes steepest descent and simulated annealing minimization to remove 100 clashes and stabilize starting energies to within 50 J/mole. 101 All molecular-dynamics simulations were run using the AMBER14 force field with 102 [12] for solute, GAFF2 [ 13], AM1BCC [ 14] for ligands, and TIP3P for water. The cutoff 103 was 8 Å for Van der Waals forces (AMBER’s default value [15]) and no cutoff was applied 104 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 5 of 10 for electrostatic forces (using the Particle Mesh Ewald algorithm [16]). The equations of 105 motion were integrated with a multiple timestep of 1.25 fs for bonded interactions and 2.5 106 fs for non-bonded interactions at T = 298 K and P = 1 atm (NPT ensemble) via algorithms 107 described in [17]. Prior to counting the hydrogen bonds and calculating the free energy, 108 we carry out several pre-processing steps on the structure including an optimization of 109 the hydrogen-bonding network [18] to increase the solute stability and a pKa prediction to 110 fine-tune the protonation states of protein residues at the chosen pH of 7.4 [17]. Simulation 111 data is collected every 100 ps after at least 2 ns of equilibrium time, observed via the 112 stabilization of: the number of hydrogen bonds, the root mean square deviations (RMSDs), 113 and the interfacial surface area. For all simulations we require approximately 10 ns or more 114 of equilibrated time as observed by stable values of root mean square deviation (RMSD) 115 from the starting structure. 116 The total hydrogen bond (HBond) counts were tabulated using a distance and angle 117 approximation between donor and acceptor atoms as described in [18] and averaged over 118 the equilibration time series of the simulation. Results are shown in Fig. 2 119 Different equilibrium runs were generated by changing the starting random number 120 seed within YASARA[11]. 121 1.3. Endpoint Free Energy Analysis 122 Binding free energy for the energy-minimized structures from molecular dynamics 123 simulations were calculated with the generalized Born surface area (MM/GBSA) method 124 on the HawkDock server[ 19]. For Type I, Type III, and NTD antibodies, we average 125 five snapshots of equilibrium conformations for binding to each SARS Cov-2 variant. 126 The MM/GBSA approximations overestimate the magnitude of binding free energy in 127 comparison to in-vitro experimental estimates, but correlate strongly with hydrogen bond 128 counts. Correlation plots for endpoint free energy analysis against interfacial hydrogen 129 bond counts are displayed in Fig. 3 130 1.4. Statistical Significance 131 T-tests were performed on every combination of two hydrogen bond means with the 132 same antibody and different spike protein variant. 15 combinations were compared for 133 each antibody. For a given combination, their means, standard deviations, and number 134 of points were input into the online T-test calculator by GraphPad[20] with the unpaired 135 T-test selection. The difference in hydrogen bond counts were statistically significant when 136 the generated p-value is less than 0.05. The p-value represents the probability that any 137 difference between the two observed groups is due to random chance. A spread sheet of 138 the t-test results are incluced in the Supplemental Materials. 139 1.5. Interfacial hydrogen bonds population analysis 140 YASARA[11] is effective at counting hydrogen bonds overall, which is a correlate to 141 binding energy. To analyze the population of individual interfacial hydrogen bonds over 142 the course of simulations, we employed a different strategy. First, we transformed the 143 simulation snapshots to a GROMACS[21] file using the mdconvert macro[11]. Second, we 144 uploaded the GROMACS trajectory to the Visual Molecular Dynamics (VMD) viewer[22]. 145 VMD hydrogen bond analyses criteria differ in detail from YASARA. To provide 146 the best match between the two separate programs we did the following. In the VMD 147 menu, we chose the hydrogen bond analysis macro, with a fixed donor-hydrogen-acceptor 148 (D − H − A) angle cutoff of θc = 35o (θc is actually 180o-⟨D − H − A). YASARA softly cuts 149 off for D-H-C anything for θc < 80o. YASARA imposes a distance criterion dependent 150 upon the H-A distance[18], while VMD measures the D-A distance. Accordingly, we begin 151 with a D-A default distance of 3.5 and vary the distance so that the average hydrogen bond 152 count over the equilibrium trajectory matches that determined for YASARA. A full table of 153 resultant hydrogen bond occupancies is available in online Supplemental Materials. 154 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 6 of 10 2. Results 155 2.1. Interfacial Hydrogen Bond Counts 156 The axis of variant is, effectively, an epidemiological timeline of COVID19 through the 157 human population. 158 The most striking aspect of the interfacial hydrogen bond count vs variant for all 159 but antibodies C1A-B12 and P4A1 is the non-monotonic "reentrant" behavior of the total 160 hydrogen bond counts. Namely, after some initial decline from the earlier WT and Delta 161 variants, there is subsequently some partial return in interfacial binding efficacy for later 162 variants. 163 The results of t-tests show that this reentrance is statistically significant. We summarize 164 as follows: 165 • P4A1 Differences between BA1 and BA2 are not statistically significant (p > 0.05) nor 166 are the differences between XBB15 and BA2.86. All other differences are statistically 167 significant. 168 • C1A-B3 Differences between WT and Delta, between BA1 and BA2, and between 169 XBB15 and BA2.86 are not statistically significant, but all other differences are. Hence, 170 the observed reentrance is statistically significant. 171 • C1A-B12 Differences between BA1 and BA2, and differences between XBB15 and 172 BA2.86 are not statistically significant, but all others are. 173 • S2X234 Differences between WT and Delta are not statistically significant, but all other 174 differences are, so that the observed reentrance for BA2.86 is statistically significant. 175 • 2-15 Differences between Delta and XBB15 are not statistically significant, but all other 176 differences are, so the observed reentrance is statistically significant. 177 • Omi3 Differences between WT and BA1, BA2 are not statistically significant, but all 178 others are, sot he observed reentrance is statistically significant. 179 • CA1-C2 Differences between BA2 and XBB15 are not statistically significant, all others 180 are. Hence the observed reentrance for BA2, XBB15, and BA2.86 are statistically 181 significant. 182 • CR 3022 Differences between XBB15 and BA2.86 are not statistically significant, but 183 all others are. Hence the reentrance for BA2.86 and XBB15 is statistically significant. 184 • C1A-F10 Differences between BA2 and BA2.86 are not statistically significant but all 185 others are. Hence the reentrance for BA2, XBB15, and BA2.86 is statistically significant. 186 • 4A8 Differences between Delta and BA2 are not statistically significant but all others 187 are. Hence, the reentrance observed for BA2, XBB15, BA2.86 is statistically significant. 188 2.2. Binding Free Energy 189 We performed endpoint free energy analysis for the P4A1, C1A-C2, CR.3022, and 4A8 190 antibodies. As shown in Figure 3, with the exception of the C1A-C2 antibody, there is a high 191 degree of correlation between the interfacial hydrogen bond counts and the endpoint free 192 energy analysis. Hence, this continues the observation made in Refs. [1,23] that interfacial 193 hydrogen bond counts are good proxies for endpoint free energy analyses. 194 The point is important because it can be seen that the binding free energies in Fig.3 are 195 quite large compared to values inferred from typical binding affinity data. As an example, 196 we can take KD ≈ 5 nM for ACE2-RBD binding from the literature[24]. The dissociation 197 constant KD is given by 198 KD = KD0 exp(∆GB/(RT)) (1) where standard estimates put KD0 ≈ 1 M (see, e.g., Dill and Bromberg[25]). Solving for 199 ∆GB gives -11.3 kcal/mole, clearly small in magnitude compared to the values found from 200 GBSA analysis. Assuming the tighter binding of antibodies giving KD = 0.1 nM changes 201 the estimated ∆GB to -14.9 kcal/mole, still far below the estimated magnitudes here. 202 The overestimates of the ∆GB magnitudes derive from the GBSA approximation itself 203 where large energies of opposite signs for the entire complex must cancel out to provide 204 the binding energy. For example, for the CR 3022 WT binding, the GBSA contributions are, 205 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 7 of 10 Figure 2. SARS CoV-2-Ab interfacial hydrogen bond counts. SARS Cov-2 variants WT, Delta, BA1 (omicron), BA2, XBB15, and BA.2.86 plotted for each Ab. BA.2.75 represents BA2 in 4A8 Ab graph. Note the reemergence effect for most variants where the binding strength rises after falling for subsequent variants. Graphing was performed with was performed using GraphPad Prism version 10.0.0 for Windows, GraphPad Software, Boston, Massachusetts USA, www.graphpad.com .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 8 of 10 Figure 3. Binding free energy estimate in kcal/mole from GBSA analysis of molecular dynamics equilibrium conformations. Type I, Type II, and N-terminal Ab represented. Straight lines were from linear regression, with coefficients of determination for P4A1 R2 = 0.8713, C1A-C2 R2 = 0.3731, CR.3022 R2 = 0.7525, 4A8 R2 = 0.7056. Graphing was performed using GraphPad Prism version 10.0.0 for Windows, GraphPad Software, Boston, Massachusetts USA, www.graphpad.com respectively: Van der Waals: -109.6 kcal/mole, Electrostatic: -302.7 kcal/mole, Generalized 206 Born: 334.7 kcal/mole, and Surface Area: -14.6 kcal/mole. We anticipate the trends of the 207 GBSA binding energy estimates to be accurate, but clearly the absolute values arising from 208 the cancellation of opposing large energies are not. 209 2.3. Population Analysis of Hydrogen Bonds 210 Population analyses presented comprehensively in the Supplementary material, show 211 that in every case, the majority of the interfacial hydrogen bonds are to the heavy chain 212 of the antibody, although in a few cases, the bond strength to the light chain becomes 213 comparable with viral evolution (for CR 3022, S2X234, C1A-B3, and 2-15 the light chain 214 interfacial hydrogen bond count for the BA.286 is comparable to the heavy chain count). 215 There are no clear systematics about changing of hydrogen bonds with viral evolution. 216 The predominant occupancies for each antibody and each variant are presented in the 217 online supplemental materials. 218 3. Discussion 219 Unsurprisingly, this work shows a general diminution of binding by antibodies devel- 220 oped at a given time to the SARS-COV2 spike protein. Surprisingly, for many antibodies 221 we have studied here there is a modest re-entrant behavior to the binding strength utilizing 222 interfacial hydrogen bond count as a proxy per the correlation between the computed bind- 223 ing energy and the interfacial hydrogen bond count. Speculatively, this can be attributed to 224 an volutionary drive to achieve antibody escape while maintaining reasonable binding of 225 the receptor binding domain to the ACE2. However, the N-terminal antibody 4A8 and the 226 Class III antibody CR 3022 show modest re-entrance This is subject to the caveat that our 227 studies approached binding perturbatively from the original Wuhan strain of SARS-COV2 228 rather than entertaining a fully new binding motif. In all cases observed in simulation here, 229 the reentrant behavior is statistically significant as measured by t-test p values. 230 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 9 of 10 This re-entrant immunity result is testable experimentally in affinity studies of an- 231 tibody binding to SARS-COV2 variant spike proteins. When coupled with structural 232 determinations from which interfacial hydrogen bond counts can be made an experimental 233 test can be made of the correlation between binding free energy (log of the dissociation 234 constant) and interfacial hydrogen bond count. 235 The significance of the re-entrant immunity is clear. Immunity gained from vaccination 236 or earlier infection is not wholly surrendered, and newer antibodies developed to later 237 variants or later vaccinations can maintain some efficacy against subsequent viral mutants. 238 Meanwhile, since it is virtually impossible to strongly evolve away from Class I antibodies 239 while maintaining sufficient ACE2 receptor binding to enter cells, there is general expecta- 240 tion for the potential damage by the virus to diminish with time, as discussed in previous 241 work[1]. 242 4. Acknowledgments 243 We acknowledge useful conversations with Rick Davis, Victor Muñoz, Javier Arsuaga, 244 and Mariel Vazquez. Aspen Drake and Rustin Mahboubi-Ardakani contributed to some of 245 the simulations. 246 Author Contributions: The study design was initiated by M.Z. Jawaid and D.L. Cox, and later 247 work by C. Chiu. Simulations were carried out by C. Chiu (60%), D.L. Cox (30%), and M.Z. Jawaid 248 (10%). Statistical analyses and endpoint free energy analyses were performed by C. Chiu. H-bond 249 population analyses were performed by D.L. Cox. The manuscript was written by C. Chiu and D.L. 250 Cox, with editing by M.Z. Jawaid. 251 Funding: This research received no external funding. 252 Data Availability Statement: Data for simulations, analyses, and hydrogen bond populations is available at https://drive.google.com/drive/folders/1ZZg0VOnmag9k8El5af459iiHAzPapdz ?usp = sharing Acknowledgments: We thank J. Solana, M. Vazquez, and R.L. Davis for useful discussions in 253 early phases of this work. A. Drake and R. Mahboubi-Ardakani contributed to early work on the 254 simulations. 255 Conflicts of Interest: The authors declare no conflicts of interest. 256

257 1. Jawaid, M.Z.; Baidya, A.; Mahboubi-Ardakani, R.; Davis, R.L.; Cox, D.L. SARS-CoV-2 omicron spike simulations: broad 258 antibody escape, weakened ACE2 binding, and modest furin cleavage. Microbiology Spectrum, 0, e01213–22. https://doi.org/doi: 259 10.1128/spectrum.01213-22. 260 2. SARSCOV2 Sequences, https://viralzone.expasy.org/9556, 2025. 261 3. Sequence of original SARS-CoV-2 Spike Protein, https://viralzone.expasy.org/resources/Coronav/Wuhan-Hu-1 262 4. Guo, Y.; Huang, L.; Zhang, G.; Yao, Y.; Zhou, H.; Shen, S.; Shen, B.; Li, B.; Li, X.; Zhang, Q.; et al. A SARS-CoV-2 neutralizing 263 antibody with extensive Spike binding coverage and modified for optimal therapeutic outcomes. Nature Communications 2021, 264 12, 2623. https://doi.org/10.1038/s41467-021-22926-2. 265 5. Clark, S.A.; Clark, L.E.; Pan, J.; Coscia, A.; McKay, L.G.; Shankar, S.; Johnson, R.I.; Griffiths, A.; Abraham, J. Molecular basis for a 266 germline-biased neutralizing antibody response to SARS-CoV-2. bioRxiv 2020, p. 2020.11.13.381533. https://doi.org/10.1101/20 267 20.11.13.381533. 268 6. Rapp, M.; Guo, Y.; Reddem, E.R.; Yu, J.; Liu, L.; Wang, P .; Cerutti, G.; Katsamba, P .; Bimela, J.S.; Bahna, F.A.; et al. Modular 269 basis for potent SARS-CoV-2 neutralization by a prevalent VH1-2-derived antibody class. Cell Reports 2021, 35. https: 270 //doi.org/10.1016/j.celrep.2021.108950. 271 7. Park, Y.J.; Pinto, D.; Walls, A.C.; Liu, Z.; Marco, A.D.; Benigni, F.; Zatta, F.; Silacci-Fregni, C.; Bassi, J.; Sprouse, K.R.; et al. 272 Imprinted antibody responses against SARS-CoV-2 Omicron sublineages. Science 2022, 378, 619–627. https://doi.org/10.1126/ 273 science.adc9127. 274 8. Nutalai, R.; Zhou, D.; Tuekprakhon, A.; Ginn, H.M.; Supasa, P .; Liu, C.; Huo, J.; Mentzer, A.J.; Duyvesteyn, H.M.; Dijokaite- 275 Guraliuc, A.; et al. Potent cross-reactive antibodies following Omicron breakthrough in vaccinees. Cell 2022, 185, 2116–2131.e18. 276 https://doi.org/10.1016/j.cell.2022.05.014. 277 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint Version March 4, 2026 submitted toAntibodies 10 of 10 9. Huo, J.; Zhao, Y.; Ren, J.; Zhou, D.; Duyvesteyn, H.M.E.; Ginn, H.M.; Carrique, L.; Malinauskas, T.; Ruza, R.R.; Shah, P .N.M.; 278 et al. Neutralization of SARS-CoV-2 by Destruction of the Prefusion Spike. Cell Host & Microbe 2020, 28, 445–454.e6. https: 279 //doi.org/10.1016/j.chom.2020.06.010. 280 10. Chi, X.; Yan, R.; Zhang, J.; Zhang, G.; Zhang, Y.; Hao, M.; Zhang, Z.; Fan, P .; Dong, Y.; Yang, Y.; et al. A neutralizing 281 human antibody binds to the N-terminal domain of the Spike protein of SARS-CoV-2. Science 2020, 369, 650–655. https: 282 //doi.org/10.1126/science.abc6952. 283 11. Krieger, E.; Vriend, G. YASARA View—molecular graphics for all devices—from smartphones to workstations. Bioinformatics 284 2014, 30, 2981–2982. 285 12. Maier, J.A.; Martinez, C.; Kasavajhala, K.; Wickstrom, L.; Hauser, K.E.; Simmerling, C. ff14SB: improving the accuracy of protein 286 side chain and backbone parameters from ff99SB. Journal of chemical theory and computation 2015, 11, 3696–3713. 287 13. Wang, J.; Wolf, R.M.; Caldwell, J.W.; Kollman, P .A.; Case, D.A. Development and testing of a general amber force field.Journal of 288 computational chemistry 2004, 25, 1157–1174. 289 14. Jakalian, A.; Jack, D.B.; Bayly, C.I. Fast, efficient generation of high-quality atomic charges. AM1-BCC model: II. Parameterization 290 and validation. Journal of computational chemistry 2002, 23, 1623–1641. 291 15. Hornak, V .; Abel, R.; Okur, A.; Strockbine, B.; Roitberg, A.; Simmerling, C. Comparison of multiple Amber force fields and 292 development of improved protein backbone parameters. Proteins: Structure, Function, and Bioinformatics 2006, 65, 712–725. 293 16. Essmann, U.; Perera, L.; Berkowitz, M.L.; Darden, T.; Lee, H.; Pedersen, L.G. A smooth particle mesh Ewald method. The Journal 294 of chemical physics 1995, 103, 8577–8593. 295 17. Krieger, E.; Vriend, G. New ways to boost molecular dynamics simulations. Journal of computational chemistry 2015, 36, 996–1007. 296 18. Krieger, E.; Dunbrack, R.L.; Hooft, R.W.; Krieger, B. Assignment of protonation states in proteins and ligands: Combining pK 297 a prediction with hydrogen bonding network optimization. In Computational Drug Discovery and Design; Springer, 2012; pp. 298 405–421. 299 19. Weng, G.Q.; Wang, E.C.; Wang, Z.; Liu, H.; Li, D.; Zhu, F.; Hou, T.J. HawkDock: a web server to predict and analyze the structures 300 of protein-protein complexes based on computational docking and MM/GBSA. Nucleic Acids Research 2019, 47, W322–W330. 301 20. Graph Pad unpaired t-test calculator, https://www.graphpad.com/quickcalcs/ttest1/, 2025. 302 21. BEKKER, H.; BERENDSEN, H.; DIJKSTRA, E.; ACHTEROP , S.; VONDRUMEN, R.; VANDERSPOEL, D.; SIJBERS, A.; Keegstra, 303 H.; RENARDUS, M. GROMACS - A PARALLEL COMPUTER FOR MOLECULAR-DYNAMICS SIMULATIONS. In Proceedings 304 of the PHYSICS COMPUTING ’92; DeGroot, R.; Nadrchal, J., Eds. World Scientific Publishing, 1993, pp. 252–256. 4th International 305 Conference on Computational Physics (PC 92) ; Conference date: 24-08-1992 Through 28-08-1992. 306 22. Humphrey, W.; Dalke, A.; Schulten, K. VMD: Visual Molecular Dynamics. J. Mol. Graphics Modell. 1996, 14, 33–38. Humphrey, W 307 Dalke, A Schulten, K Schulten, Klaus/D-5561-2009, https://doi.org/10.1016/0263-7855(96)00018-5. 308 23. Jawaid, M.Z.; Baidya, A.; Jakovcevic, S.; Lusk, J.; Mahboubi-Ardakani, R.; Solomon, N.; Gonzalez, G.; Arsuaga, J.; Vazquez, M.; 309 Davis, R.; et al. Computational study of the furin cleavage domain of SARS-CoV-2: delta binds strongest of extant variants. 310 bioRxiv 2022. https://doi.org/10.1101/2022.01.04.475011. 311 24. Walls, A.C.; Park, Y.J.; Tortorici, M.A.; Wall, A.; McGuire, A.T.; Veesler, D. Structure, function, and antigenicity of the SARS-CoV-2 312 spike glycoprotein. Cell 2020, 181, 281–292. e6. 313 25. Gilson, M.K.; Given, J.A.; Bush, B.L.; McCammon, J.A. The statistical-thermodynamic basis for computation of binding affinities: 314 a critical review. Biophysical Journal 1997, 72, 1047–1069. doi: 10.1016/S0006-3495(97)78756-3, https://doi.org/10.1016/S0006-34 315 95(97)78756-3. 316 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.03.709420doi: bioRxiv preprint