A number of CD4bs-directed neutralizing antibodies have been identified and reported. Based on their mode of recognition and B-cell ontogeny, CD4bs antibodies fall into two categories: VH-gene restricted antibodies derived from the heavy chain germline genes VH1-2 or VH1-46, and CDRH3 dominated antibodies in which the antibody binding interfaces are dominated by the complementary-determining region three (CDR3) (13, 31, 32). The CD4bs directed bNAbs used for this study included VRC01 and 8ANC131, considered to be the first identified members of the VH-gene restricted “VRC01-class and 8ANC131-class” bNAbs (33) since the co-crystal structures of these antibodies with HIV-1B and C envelopes were available. VRC01 (VH1-2) and 8ANC131 (VH1-46) are both potent bNAbs found to be capable of neutralizing about 91 and 78% of the HIV-1 strains, respectively (34). The co-crystal structures of 8ANC131 with the HIV-1 subtype B envelope YU-2 gp120 (PDB ID: 4RWY 2.13 Å resolution) and VRC01 with the HIV-1 subtype C envelope ZM176.66 gp120 (PDB ID: 4LST 2.55 Å resolution) were downloaded from PDB (Protein Data Bank) (Figure 1). Both co-crystal structures included the heavy and light chains of the respective antibodies complexed with HIV-1 envelope gp120. The Fab (Fragment antigen-binding) regions of the antibodies were bound to the CD4bs in HIV-1 gp120.

The Rosetta Peptiderive is a computational tool designed to predict possible inhibitory peptides from the crystal structures of protein complexes based on their interacting interface, was used to identify short linear peptides that would target the CD4bs and bring about virus neutralization. This tool is hosted online in ROSIE (Rosetta Online Server that Includes Everyone) web interface and can be accessed at https://rosie.rosettacommons.org/peptiderive. The antigen (HIV-1 gp120)–antibody (bNAb) complex was uploaded on the Rosetta peptiderive tool in PDB format, with optimal parameters defining the Receptor and Partner. The tool automatically refines the antigen–antibody complex by removing local clashes and extracts potential peptide fragments of specified window size. The binding energies of the identified peptide–antigen complexes were calculated using the Rosetta energy function (35). Peptides with the most significant binding scores were shortlisted, and their position, sequence, interface score and relative interface score were obtained (36). Intermolecular interactions of the identified peptide-antigen complexes were visualized in the PDBsum webserver (37) and CHIMERA (38).

To validate the binding of the identified peptides with HIV-1 gp120, peptide–antigen docking was performed using HADDOCK (High Ambiguity Driven protein-protein DOCKing) webserver (Version 2.2) in the EASY interface available at https://wenmr.science.uu.nl/haddock2.4/. The antigen and peptides were docked by generating Ambiguous Interaction Restraints (AIR) with the interface residues identified from the PDBsum analysis of the Rosetta peptiderive complexes (39, 40). The docked structures were summarized in clusters, and each cluster was assigned a HADDOCK score, cluster size, RMSD from the overall lowest energy conformations, Z-score and buried surface area along with bonding energies (Vander Waal’s, electrostatic, desolvation, and restraints violation energies). The best-docked complex (topmost cluster suggested by HADDOCK) replicating the desired residual interactions was identified and selected for further analysis. The binding affinity (ΔG) and dissociation constant (K_d) of the docked complexes were calculated using the PRODIGY webserver, available at https://wenmr.science.uu.nl/prodigy/. This webserver predicts binding affinities based on inter-molecular contacts within a distance cut-off of 5.5 Å (41, 42).

The peptide-HIV-1 envelope complexes identified using Rosetta peptiderive were subjected to Molecular dynamics (MD) simulations to deduce their dynamic behavior under physiologically simulated conditions (43). MD simulations were performed using the DESMOND software package (44) with OPLS_2005 as a force field and implemented as in Muthukumaran et al. (45). To begin with, the system was built in an auto-calculated cubic box and solvated with explicit Single Point Charge (SPC) water molecules. The solvated system was energy minimized and the MD run was carried out for 200 ns by implementing an NPT ensemble with a sampling interval of 10 ps. During the MD run, the whole system was maintained at an equilibrium of 300 K temperature and 1 atm pressure. Analytical tools available in DESMOND were used to infer the Root Mean Square Deviation (RMSD) of the protein backbone, the Root Mean Square Fluctuation (RMSF) of the residues, the radius of gyration, and other structural transitions throughout the simulations.

The binding free energy (ΔG) of the final frames of stable neutralizing peptide–antigen complexes obtained from the MD simulation was calculated by implementing MM-PBSA (Molecular Mechanics-Poisson Boltzmann Surface Area) protocol in farPPI (fast amber rescoring for Protein–Protein interaction Inhibitors) webserver, available at http://cadd.zju.edu.cn/farppi/. Precise binding energies of the docked poses were evaluated by the MM-PBSA method which combines energy calculations based on implicit solvent and molecular mechanics model (46). Among the MM-PBSA procedures, PB3 based approach was found to be highly accurate as compared to the other approaches in farPPI, as two force fields, GAFF2 and ff14SB, were applied to the peptide and antigen, respectively (47, 48). Hence, this method was adopted to score the binding free energy of the peptide–antigen complexes.

The identified peptides were subjected to alanine scanning using Bude Alanine Scan² (52, 53) and Robetta Alanine scan³ (54) webservers to infer the energetically significant amino acids at the peptide–antigen interface. This prediction helps to prioritize key residues in the identified peptides. Toxicity and physico-chemical properties of the peptides were predicted using ToxinPred⁴ (55) and the peptide analyzing tool provided by Thermo-fisher Scientific⁵.

Four hexameric peptides were derived through structure-based sequence inference from the 8ANC131 and VRC01 neutralizing antibody-HIV-1 gp120 complexes using the Rosetta peptiderive protocol as shown in Figures 2, 3. From the hot segments in the bNAbs (that contribute to the most significant binding interaction with the HIV-1 envelope gp120 protein), two peptides were identified from each of the two antigen–antibody complexes. These included the peptide Arg-Asp-Arg-Ser-Thr-Gly (RDRSTG) from the H chain of 8ANC131, which had an interface score of –9.447 and contributed to 29% of binding energy, and the peptide Glu-Tyr-Ser-Ser-Thr-Pro (EYSSTP) from the L chain, which had an interface score of –4.554 and contributed to 101% of binding energy. Two other hexamers were derived from the PDB crystal structure of VRC01-HIV-1C envelope, namely, Val-Asn-Tyr-Ala-Arg-Pro (VNYARP) from the H chain, which had an interface score of –9.982 and contributed to 30% of binding energy, and QQYEFF (Gln-Gln-Tyr-Glu-Phe-Phe) from the L chain, which had an interface score of –6.839 and contributed to 68% of binding energy (Table 1). In general, the peptides derived from the heavy chain of the antibodies gave comparatively lower interface scores than peptides derived from the light chain, signifying better binding affinity of the former. Among the four peptides, RDRSTG peptide having an interface score of –9.447 showed the most significant binding to the HIV-1 envelope.

Structural analysis of the 8ANC131-subtype B gp120 (PDB ID: 4RWY) and VRC01-subtype C gp120 (PDB ID: 4LST) complexes revealed close interaction between the antibody Heavy chains and the HIV-1 gp120 CD4-binding site, while the light chains protruded beyond the CD4bs, particularly the D Loop and V5 regions (Figure 4). Therefore, we excluded the peptides derived from the light chains as they did not engage our target, i.e., the CD4bs. Residues 365–371 of HIV-1 gp120 were found to be the key residues involved in making critical contacts with Phe43 and Arg59 residues of the CD4 receptor (56). The VRC01 antibody showed a non-bonded interaction with Ser365_gp120, Gly366_gp120, and Gly367_gp120 of the Phe43 cavity, while in 8ANC131, Gly366_gp120 and Gly367_gp120 were found to be involved in the interaction (57). Furthermore, ASP368_gp120 was observed to mediate the interaction with ARG71_8ANC131/_VRC01 by forming hydrogen bonds and salt bridges, which mimicked the natural interaction between ARG59_{CD4 RECEPTOR} and ASP368_gp120 (25, 31, 34) (Figure 5).

We also performed intermolecular interaction analysis of the Rosetta-derived peptide-antigen complexes and observed similar interactions as seen in the PDB crystal structures (Supplementary Figure 1). Among the peptides derived from the antibody heavy chains, RDRSTG was found to form two hydrogen bonds with ASP368 (2.75 Å and 2.82 Å) and MET426 (2.72 Å and 3.20 Å), and one hydrogen bond with GLY431 (2.76 Å); the other crystal structure residues were found to have non-bonded contacts in the vicinity of <5 Å (LEU122, VAL430, TRP427 and LYS432). The VNYARP peptide formed two hydrogen bonds with GLY458 (3.08 Å and 3.12 Å) and one hydrogen bond with ARG456 (2.73 Å) and THR467 (3.00 Å). In addition, salt bridges were also observed at ASP461 and GLU466. Other non-bonded contacting residues were ASN280, THR465, GLY366, SER365, ASN460 and ASP457. Based on these observations, we docked the neutralizing antibody 8ANC131-derived peptide RDRSTG with subtype C (ZM176.66) gp120, and the VRC01-derived peptides VNYARP with subtype B (YU-2) gp120, to examine the closeness of the interaction patterns (especially ASP368 and SER365) with that seen in the native crystal structures. To revalidate the observed interactions in the Rosetta derived complexes, we performed re-docking of the peptide RDRSTG with subtype B (YU-2) gp120 and VNYARP with subtype C (ZM176.66) gp120. (Positions of residues are different in PDB crystal structures and 2D LIGPLOT—Supplementary Table 1; Residues stated here are in accordance with the crystal structure but different from that in LIGPLOT).

To start with, the HIV-1 subtype B and subtype C gp120 antigens (without peptides) were subjected to a production run of 200 ns, and trajectory analysis was performed. The system of subtype B gp120 antigen comprised of 49,311 atoms with 14,688 water molecules in the neutralized state, while the subtype C gp120 antigen system comprised of 48,478 atoms with 14,405 water molecules in the neutralized state. The RMSD plot of both antigens revealed that the Cα deviations were stable and within the range of 3 Å, and were found to converge toward the final stages of simulation (Supplementary Figure 4). The RMSF plot identified the peaks which represent the regions/residues that fluctuated the most during the simulation: 175–200 (4.7 Å) and 225–250 (5.0 Å) regions in subtype B gp120, and 250–275 (4.0 Å) and 300–337 (4.2 Å) regions in subtype C gp120 (Supplementary Figure 5).

We then performed molecular dynamics simulation of the peptide–gp120 complexes. The simulation system of subtype B gp120-RDRSTG solvated complex comprised of 49,305 atoms with 14,654 water molecules, and was neutralized by adding one cl^– ion (1.241 mM). On trajectory analysis, the protein–ligand RMSD plot revealed that the complex converged at 10 ns with a 0.6 Å difference between the peptide and antigen-bound state (Figure 6). The ligand RMSD value was in the range of 3.0 Å with reference to the backbone of the antigen and was found to be well-bound to the binding regions. The RMSF plot revealed that RDRSTG (Supplementary Figure 6) interacted well at regions 50–100, 220–250, 250–300 and 300–337, despite fluctuations. Fluctuations posed by the peptide throughout the simulation were inferred from the ligand RMSF plot (Supplementary Figure 7), where it was found to be stable in the range of 4 Å. The structural compactness of the peptide was measured based on the radius of gyration (rGyr). This analysis revealed that the peptide RDRSTG maintained its compactness up to 150 ns in the range of 1 Å (Supplementary Figure 8). The bonded interactions between the antigenic residues and the RDRSTG peptide were analyzed from the ligand–protein contacts plot (Figure 7), wherein it was found that for about 79 and 52% of the duration of the run, Asp229 (ASP368) and Glu274 (GLU429) interacted by means of hydrogen bonds, ionic bonds and water bridges, respectively (Supplementary Figure 10).

Subtype C gp120-VNYARP peptide was made up of 48,414 atoms with 14,350 water molecules in the neutralized state. The antigen–peptide RMSD plot inferred that the complex converged at 75 ns with a 0.6 Å difference between the peptide and antigen-bound states (Figure 6). However, the peptide VNYARP evolved to make stable interactions between 85 and 160 ns in the vicinity of <3 Å. The ligand RMSD value was in the range of 2.0 Å with a major fluctuation at 75 ns. Residues in the region 150–180, 200–250, and 300–339 were found to sustain bonded interactions with the peptide as per the RMSF plot (Supplementary Figure 6). The ligand RMSF plot inferred that the peptide is stable as the fluctuations were within the range of 4 Å (Supplementary Figure 7). The rGyr analysis revealed a minimum deviation of 6.0–6.5 Å, indicating that the peptide sustained high compactness during the entire simulation process (Supplementary Figure 9). With regard to peptide–antigen contacts (Figure 8), Gly305 (GLY458) was found to interact 96% of the time during the entire run by means of hydrogen bonds and water bridges. Asp304 (ASP457—70%), Asp227 (ASP368—67%), Gly226 (GLY367—62%) and Ser224 (SER365—62%) formed hydrogen bond interactions and water bridges, with the exception of Asp227 (ASP368), where an additional ionic interaction featured. The least interacting residue was Arg303 (ARG456), which revealed sustained binding (hydrogen bonds and water bridges) around 53% of the 200 ns production run (Supplementary Figure 11).

The MD trajectories revealed RDRDTG and VNYARP peptides to be highly stable in terms of bonded interactions during the 200 ns of simulation. The dynamic evolution of the peptides RDRSTG and VNYARP are illustrated in Figures 9, 10. The MD trajectory analyses revealed that the peptides RDRSTG and VNYARP were stable binders, as they feature stable contacts with the key residues namely, SER365, GLY366, GLY367, and ASP368 across the production run (Supplementary Figure 12). The binding free energies (ΔG) of the complexes (RDRSTG-subtype B gp120 and VNYARP-subtype C gp120) were calculated over the MD simulation trajectory for the frames sampled at an interval of 20 ns and subjected to MM-PBSA (PB3) using the far-ppi server and binding affinity calculation using Kdeep, respectively. MM-PBSA calculations of RDRSTG-subtype B gp120 and VNYARP-subtype C gp120 complexes gave an average of –13.58 ± 2.85 (Mean ± SD) kCal/mol and –16.04 ± 8.77 (Mean ± SD) kCal/mol, respectively. Similarly, KDeep calculations gave an average of –9.32 ± 0.80 (Mean ± SD) kCal/mol for RDRSTG-subtype B gp120 and –10.18 ± 0.63 (Mean ± SD) kCal/mol for VNYARP-subtype C gp120, respectively (Supplementary Figure 17).

The Alanine scan analysis for RDRSTG-Subtype B gp120 and VNYARP-Subtype C gp120 complexes using Robetta and Bude scan identified the cumulative energetically important amino acids in the peptides across the binding interface as R, D, R, S and T in the RDRSTG peptide and V, N, Y and R in the VNYARP peptide. The binding affinities of the alanine mutated peptides are provided in Supplementary Table 2. The results of the physico-chemical analysis are provided in Table 2. Further, the peptides were found to be non-toxic.