Peptides of HA-tag, c-MYC-tag, VSV-tag, Flag-tag, and their mutants (more than 98% purity) were synthesized commercially by Nanjing Yuanpeptide Biotech (Nanjing, China) (Tables S1-S4). The selection criteria for the universal protein tags were to get the peptide sequence as short as possible, which made it easier and more cost-effective to synthesize. Secondly, the amino acid sequence of each tag was diverse as it would cover 20 amino acids. As a result, the residues in the HA-tag were not duplicated when replacing the c-MYC-tag, VSV-tag, or Flag-tag. Additionally, mAbs that recognized the tags were easy to purchase and evaluate and were sensitive to the wild type and its mutations. To increase the solubility and binding ability of the universal protein tags in the 96-well plate, random peptides of 18 amino acids were respectively fused with the target tag at the C-terminus via the GSGSGS linker. Mouse anti-HA-tag mAb labels were purchased from Qrigene cat# TA180128, MD, USA. Mouse anti-c-MYC-tag and anti-Flag-tag mAb labeled were purchased from Proteintech cat# 6003-2-lg/66008-3-lg, Wuhan, China. Mouse anti-VSV-tag mAb labeled were purchased from Abbkine Scientific cat# A02180, Wuhan, China. HRP-conjugated AffiniPure goat anti-mouse IgG from Proteintech, cat# SA00001-15, Wuhan, China, and DAB color development kit cat# AR1026 were purchased from a Biological company, and the other additional buffers were prepared in our laboratory.

The peptides of tags and their mutants were diluted to 2.0 μg/ml and 50 μl each was added into 96 well plates for 1 hour to adsorb (n=3). After washing with phosphate buffer saline containing 0.05% Tween-20 (PBST) three times, the plate was blocked with a rapid-blocking buffer for 10 min. After washing with PBST three times, mouse anti-tags mAbs were added into each well (40 μl/well) and incubated at 37°C for 1 h. Next, after washing and blocking as described above, 100 μl of horseradish peroxidase (HRP)-labeled goat anti-mice IgG (1:2500) were added and incubated at 37°C for 1 h. The results were detected by the plate reader after color-developing with TMB single-component substrate solution (Solarbio, cat# PR1200, Beijing, China) and stopping. Peptide-free and mAb-free wells were performed as negative controls. The absorbance value at 450 nm (OD₄₅₀) was measured for analysis.

A literature search was performed using the keywords: ‘protein ‘interaction’, ‘antigen-antibody ‘interaction’, and ‘crystal structure of binding ‘surface’ in the NCBI PubMed database (https://www.ncbi.nlm.nih.gov/pubmed/) and Web of Science (https://www.webofscience.com/wos/alldb/basic-search) to collect the amino acid recognition binding data documented in the literature. First, each letter was carefully read and the computer-predicted Ag-mAb model was extracted. Only the models established by X-ray or cryo-electron microscopy with clear textual identifications were included in the statistical analysis along with specific interaction patterns between amino acids.

All collected data were entered into Microsoft Excel™ and visualized with Cytoscape and GraphPad Prism 8.02 software. After partial mutation, the mutant polypeptide sequence and the original protein-tag sequence of IVM and IEM were generated using the online prediction software PEP-FOLD3 (https://bioserv.rpbs.univ-paris-diderot.fr/services/PEP-FOLD3/) and AlphaFold (https://github.com/google-deepmind/alphafold/tree/main/alphafold), and a 3D structure was obtained. The best coupling model was selected according to the score. At the same time, the PDB file was downloaded to UCSF Chimera X, and the Matchmaker program was performed on the 3D structure of the mutant and the original protein tag to obtain the structure distance RMSD (root mean square deviation) value and the 3D conformational map.

We screened the key residues in HA-tag to mAb TA180128. At first, the HA-tag (Y1P2Y3D4V5P6D7Y8A9) was substituted one by one from Y1 to A9 with G, E, or H, respectively. Except for the A9 residue, mAb TA180128 reacted to the other residues from Y1 to Y8 in this study. After substituting with other 19 amino acids from Y1 to Y8, we built a total of 84 HA-tag mutants. The 32 (38.1%) mutants were categorized as IVM (P6V, P6S, P6T, V5I, Y3N, Y3F, Y3Q, V5T, Y3R, V5S, Y3W, Y3S, V5L, V5M, Y3C, Y8H, Y3G, P6G, D4H, V5K, A9H, P2H, Y3H, Y1H, A9G, Y3E, Y3L, P2G, D4G, Y1E, P6L, and Y1G) for maintaining more than 80% affinity to mAb TA180128. The 15 mutants (17.8%) were EQM (P6M, V5P, Y3M, P6E, A9E, D4E, V5H, P2E, V5Q, D7E, D7G, V5N, Y3T, P6I, and V5C) for maintaining 43%- 77% affinity (Figure 1A). The HA (V5R) mutant was ADERM for maintaining 35% mAb-binding affinity. The other 36 mutants (42.9%) were IEM (V5G, D7H, V5F, V5A, Y3V, D4C, P6A, D4S, V5Y, Y3I, Y3K, V5E, D4L, Y8G, P6Y, D4A, D4M, Y3D, V5D, P6C, P6Q, D4W, D4Y, P6K, P6N, P6D, P6R, P6W, D4F, V5W, P6F, D4P, Y3P, P6H, D4I, and Y8E) for losing the affinity to mAb TA180128 completely (Figure 1B). However, P6V, P6S, and P6T mutants were found to have 16%-21% superior affinity compared to their parental HA-tag (Figure 1A). Therefore, in 84 HA-tag mutants, approximately 55.9% of antigenic drift (IVM, 38.1% and EQM, 17.8%) did not affect the immunity of mAb TA180128 (Figure 1A).

Since IVM and EQM did not affect the protective effect of the current vaccine, the aim was to reveal the IEM of Y3, D4, V5, P6, D7, and Y8 substitutions replaced by the other 19 amino acids, respectively. In summary, V5, P6, D4, and Y3 were IEM hotspots, accounting for 24.1%, 37.9%, 31%, and 17.2% of IEM, respectively (Figure 1B). Most of the D7 and Y8 residues were IVM and EQM hotspots, and only D7H and Y8E mutations were IEM (Figure 1).

To cover 20 amino acid substitutions, we added the MYC-tag, Flag-tag, and VSV-tag to the experimental catalog. Those amino acids that did not repeat on the HA-tag were selected for mutation. Specifically, we replaced the Q2, S6, and E8 residues of MYC-tag with G and its L4 and I5 with other 19 amino acids, respectively. While the other 19 amino acids replaced the K3 of Flag-tag, G replaced the N7, R8, and I4 in VSV-tag, and the T2 and M6 were mutated to the other 19 amino acids, respectively.

For 39 c-MYC-tag mutants, except for the affinity of Q2G, I5S, and E8G > 80%, which belonged to IVM, I5L and S6G belonged to EQM for maintaining 65% and 47% affinity. The other 34 mutants had less than 20% affinity or could not be combined with mAb and were classified as IEM (Figure 2A).

Among the 20 mutants of Flag-tag, only K8G shared 95.9% mAb-binding affinity with Flag-tag and was defined as IVM, and K3D maintained 44% mAb-binding affinity and was defined as EQM. The other 18 K3 mutants were IEM for having less than 6% affinity (Figure 2B).

Similarly, among the 41 mutants of Flag-tag, M6C, M6H, M6K, and M6R of VSV-tag were IVM for maintaining more than 110% affinity compared to VSV-tag. N7G, T2R, M6Q, M6N, and R8G were EQM for maintaining 47%-77% affinity. The other 32 mutants were IEM for having less than 17% affinity (Figure 2C).

Unexpectedly, in total, the 184 universal protein tags, HA(P6V), HA(P6S), HA(P6T), VSV (M6C), VSV (M6H), VSV (M6K), VSV (M6R), and MYC (Q2G) replacement were enhanced by mAb-binding affinity from 112% to 121%, respectively (Figures 1A, 2).

To display the mutant interaction visually, we used Microsoft Excel to draw the heatmap of amino acid residue substitution in this study. From the selected and substituted nine amino acids, I, K, L, and T tended to lose their mAb affinity after substituting them with the other 19 amino acids (Figure 3) except that the binding affinity substituted by C, H, K, and R could be maintained, and substituting M by other amino acids led to IEM. Similarly, when E, H, and G replaced D, it could still recognize and bind to mAb. The other substitutions of D led to IEM. Conversely, approximately 70% of substitutions of V and Y showed IVM. Most substitutions of V and Y retained their original immune affinity. Only V/A, V/D, V/F, V/Y, V/W, V/E, V/G,Y/D, Y/P, Y/V, Y/I, and Y/K replacement models produced IEM (Figure 3). Nevertheless, only 30% of substitutions of P showed IVM. P/E, P/I, and P/M showed EQM, and the other 11 substitutions were IEM (Figure 3). Notably, all amino acids replaced by A, D, F, P, W, and Y led to IEM, and 98% of amino acids replaced by V and I might be IEM (Figure 3). However, 66% of amino acids replaced by G might be IEM (Figure 4).

Seventy articles related to Ag-Ab interaction were screened through the NCBI PubMed (https://pubmed.ncbi.nlm.nih.gov) and Web of Science (https://access.clarivate.com). The interactions of amino acid residues at the Ag-Ab binding interface were extracted from the 3D structures reported in the literature, and the amino acid binding matrix was drawn and colored according to the number of relevant literature (Figure 5A). The X-axis is the amino acids of the Ab, and the Y-axis is antigenic residues. According to the matrix, the binding rules of amino acids in Ags and Abs tended to be consistent. Among the 20 types of amino acids, Y, S, D, and R had an affinity with other amino acids and could recognize almost all amino acids in the literature. However, C and M were practically silent and only interacted with relatively few amino acids, coinciding with some of our ELISA experiments’ findings. Explicitly, residue M was not found in the Ab interface and only four were found in the Ag interface (Figures 5B, C). Residue C was found only one in the Ab interface and only three were found in the Ag interface (Figures 5B, C). Otherwise, P was rarely found in Ab interfaces and V was rarely found in Ag interfaces (Figures 5B, C).

Next, the histogram of the interaction capacity of the 20 amino acids was plotted according to the literature. The top 10 amino acids with the highest Ab detection rate were Y, R, D, N, S, K, T, F, L, and E, respectively (Figure 5B). However, the top 10 detection amino acids in Ags were R, Q, D, N, E, Y, L, A, H, K, S, P, and I (Figure 5C), which explains that H, K, S, P, and I shared the 10th rank for their same detection rate (Figure 5C).

The data were imported into Cytoscape to draw an interaction network. According to the physical and chemical properties, 20 amino acids were divided into eight categories, which were aliphatic amino acids, aromatic amino acids, acidic amino acids, alkaline amino acids, amide amino acids, sulfur-containing amino acids, hydroxyl amino acids, and imino acids. Then, the interaction diagram was constructed (Figure 6A). Except for a few aliphatic amino acids and all sulfur-containing amino acids, the side chains of hydroxyl-amino acids S and T could interact with 14 amino acids in the literature (Figure 6B). The aliphatic amino acids included A, V, I, and L, and only I could interact with C (Figure 6C). The alkaline amino acids K, R, and H could interact with 17 kinds of amino acids in Abs (Figure 6D). Notably, antigenic P interacted with 13 amino acids in Abs, and P in Abs only interacted with antigenic N, Q, and A (Figures 6A, C, E, F). The antigenic C (interacted with D, T, and V) and M (interacted with L, R, W, and Y) contained Sulphur in their side chains, causing them not to cross-link (Figure 6G). Otherwise, the residue of Ab C only recognized antigenic I and N (Figures 6A, C, F). Acidic amino acids could interact with 16 amino acids (Figure 6H). Similarly, antigenic aromatic amino acids seemed to interact with more than 10 amino acids (Figure 6I). To quickly view the amino acid interaction between Ag-Ab, single interaction networks were plotted individually in the alphabetical order of amino acid abbreviations. Amino acids without interaction were listed below the main figure in boldface (Figure 7).

To reveal the effect of mutation on the peptide structure and verify the experimental results, we used the computer ab initio folding algorithm to predict the 3D structure of peptide tags and their mutants. We predicted the 3D structure of HA-tag, MYC-tag, and Flag-tag and their mutants (HA-Y1G, HA-Y3G, HA-Y8G, MYC-E8G, MYC-E7G, Flag-K8G, and Flag-K3G), respectively, by online tools PEP-FOLD3 and AlphaFold. After obtaining the PDB file, we compared the mutant with the original structure. The fit degree of HA-Y8G and MYC-E7G to the original tag structure was lower than other mutants (Figure 8). Surprisingly, the Flag-K3G mutation disrupted the original α-helix and curled into a semi-O-ring in the opposite direction of the original Flag-K3 peptide, affecting Ab binding capacity (Figures 2B, S2). Consequently, the low fit mutation resulted in IEM due to loss of interaction. The same IEM mutations happened in HA-D4F, HA-D4W, HA-D4Y, HA-V5F, HA-V5W, HA-V5Y, HA-P6F, HA-P6W, HA-P6Y, VSV-M6F, VSV-M6Y, and VSV-M6W (Figures 10).

As a broadly known concept, protein tags mutation experiment found that the substitution of amino acids with similar physical and chemical properties residue tended to IVM, e.g., aliphatic amino acids V, I, and L substituted each other (HA-V5I, HA-V5L; MYC-I5L) (Figures 1, 2A, C). Similarly, aromatic amino acids Y, W, and F mutual substitution were also IVM, e.g., HA-Y3F and HA-Y3W (Figure 1). The results indicated that the amino acids with the same physical and chemical properties of the side chain group shared high similarity in 3D structures (Figure 9). In addition, the non-aromatic amino acids were replaced by aromatic amino acids (F, Y, and W) and often lost their affinity to antibodies (Figures 1, 2). It might be related to the benzene ring on the side chain of aromatic amino acids. We consequently predicted the structure of some non-aromatic amino acid mutants replaced by aromatic amino acids and compared their structural changes (Figure 10). To visually describe the structural changes of the mutated tags, the RMSD values were obtained in UCSF ChimeraX for the fitted structures of the above-mentioned mutants and the original protein tags were summarized (Table 1). If the RMSD was less than 3 Å, the two structures were considered similar. Conversely, if the RMSD was more significant than 3 Å, the two structures did not match. Thus, our test suggested that the protein-tag mutants with RMSD > 3.0 Å tended to IEM (Table 1).

Although supported by experimental and literature data, our study still has some shortcomings. First, the Ag-Ab amino acids interaction pattern did not cover all 20 amino acids due to the absence of certain amino acids in linear epitopes, but the integrated 3D database analysis made up for this shortcoming. Second, the Ag-Ab interaction data derived from the linear protein tag epitopes may not adequately reflect the true spatial information of complete antigen molecules. Third, on account that antigens no longer exactly reflected viruses visible within the actual world, four short protein tags were used. This study did not set out to develop a quick technique or methodology for coping with real-world viral immune escape. However, the fundamental concept of amino acid interaction between Ag and Ab might help us to predict the binding or escape of viruses and Abs in the real world. Based on this approach, our study could draw more attention to combat life-threatening viruses. Therefore, the main task was to verify the validity of this hypothesis in the future by testing contrast antibodies.