Retrained generic antibodies can recognize SARS-CoV-2

Abstract

The dramatic impact which novel viruses can have on the human society could be mitigated without the need of vaccination if antibodies present within the population are retrained to recognize these viruses. With this idea in mind, double-faced peptide-based boosters are computationally designed to allow recognition of SARS-CoV-2 by Hepatitis B antibodies. One booster face is made of ACE2-mimic peptides that can bind to the receptor binding domain (RBD) of SARS-CoV-2. The other booster face is composed of a Hepatitis B core-antigen, targeting the Hepatitis B antibody fragment. Molecular dynamics simulations revealed that the designed boosters have a highly specific and stable binding both to RBD and the antibody fragment (AF). This approach can provide a cheap and efficient neutralization of emerging pathogens.

Graphical Abstract

In the last decades, zoonotic viral pathogens (SARS, MERS, Denge, Ebola, Zika, H1N1, etc.) have become a major global public heath problem, due to their rapid spread within the highly concentrated and mobile human population [1]. Fortunately, none of the above related diseases reached a truly global scale due to highly organized actions taken to stop their spreading, sometimes in combination with large local fatalities. However, SARS-CoV-2 took the world by surprise with its very rapid spreading and moderate mortality. It has caused a devastating COVID-19 pandemic with large fatalities and wide-ranging socioeconomic disruptions.

COVID-19 has been addressed on many parallel fronts, including the development of antiviral drugs [2–7], antibody therapies [8, 9], and vaccines [10, 11]. Ultimately, to become protected, humans can gain antibodies through therapies, vaccinations [12], or real infections. However, these approaches have various limitations. Preparation of antibodies is a complex process, their delivery is instantaneous, but such antibodies have shorter lifetimes. Vaccinations need to be repeated, it takes some time before the antibody response is robust and effective, and the vaccines might not work for everybody. Finally, the real infections can have large consequences.

To address novel viral infections in an emergency mode, we propose an alternative approach to fast redirect (train) the immune response. In particular, we show that one can design interfacial molecular boosters which allow generic antibodies preexisting in the human body to recognize novel viruses, thereby allowing their selective clearance by standard pathways [13]. Such double-faced boosters can provide highly specific binding of generic antibodies (resulting from vaccination against other diseases) with novel viruses. Hepatitis B antibodies are a good choice for recognizing new viruses, due to their long lifetimes (30 years) [14]. As a practical example of this treatment, we design and simulate boosters composed of the ACE2-based peptide inhibitors, binding to the Spike RBD of SARS-CoV-2, and segments of the Hepatitis B antigen, binding to the Hepatitis B antibody. This computational study could provide guidance in the preparation of active therapeutics against emerging pathogens with the combined advantages of small-protein and antibody therapies. However, the designed boosters should be thoroughly tested and further optimized in follow-up experimental/computational studies.

Booster design:

Each booster has two connected and outside oriented faces, including the ACE2-mimic (Face 1 - red) and the antigen of Hepatitis B (Face 2 - orange), as shown in Figs. 1a-c. Face 1 is formed by peptide inhibitors of SARS-CoV-2, very similar to those designed in our previous work [2]. Their components, such as $\alpha$-helices and $\beta$-hairpin segment, were extracted from ACE2 that were in close contact with the RBD of SARS-CoV-2. Face 2 is a region extracted from the Hepatitis B core-antigen that is recognized by the Hepatitis B antibodies.

FIG. 1:

The structure of double-faced boosters bound to the Spike RBD and AF (scFv). (a) Booster 1 is composed of Face 1 formed by the ACE2-mimic (19−102 amino acids [15]) and Face 2 formed by the Hepatitis B antigen (without 66−91 residues [16]); (b) Booster 2 has the inhibitor 3 from [2] as Face 1 and the Hepatitis B antigen [16] as Face 2; (c) Booster 3 has the same faces as Booster 1, but with a PEG linker in between (inset); (d) the amino acids of AF which initially interact with Face 2; (e) the amino acids of RBD which initially interact with Face 1. Color scale: green - antibody, orange - antigen (Face 2), red - ACE2-mimic (Face 1), blue - RBD of SARS-CoV-2, gray - C atom, red - O atom, blue - N atom. ACE2: Angiotensin-converting enzyme 2, which is the cellular receptor of SARS-CoV-2.

Booster 1:

Face 1 is composed of the ${\alpha }_1{\alpha }_2$ helices of ACE2 (19–102 amino acids), which are in close contact with the RBD of SARS-CoV-2. Face 2 is the Hepatitis B core-antigen without the 66 – 91 amino acids. Since these amino acids form a flexible random coil, they are excluded in Booster 1 to make a stable structure. The two faces are connected by a peptide bond formed between the 102nd amino acid of the ACE2 helices and the 5th amino acid of the Hepatitis B core-antigen, as shown in Fig. 1a.

Booster 2:

Face 1 is the previously-designed SARS-CoV-2 Inhibitor 3 [2]. Face 2 contains the 5–145 amino acids of the Hepatitis B core-antigen. The two faces are linked by a peptide bond formed between the 362nd amino acid of Inhibitor 3 and the 5th amino acid of the Hepatitis B core-antigen (Fig. 1b).

Booster 3:

Both faces have the same components as Booster 1, but with an extra Polyethylene Glycol (PEG) linker between the 94th amino acid of the ACE2 helices and the 5th amino acid of the Hepatitis B core-antigen, as shown in the inset of Fig. 1c. A highly bio-compatible PEG chain adds a non-peptide linkage between the faces which may maximize the conformational integrity of individual faces. Figure S1 shows the detailed Chem Draw structure of the PEG linkage in Booster 3.

As shown in Figs. 1a-c, the boosters were initially placed with Face 1 (red) binding to the RBD of SARS-CoV-2 (blue) and Face 2 (orange) binding to AF (green), the single-chain variable fragment (scFv) of the Hepatitis B antibody. The initial binding configurations were based on the (6LZG) [15] and (6CWD) [16] pdb crystal structures, respectively. The amino acids of AF and RBD which had initial contacts with the boosters (within 3 Å of boosters) are shown in licorice and listed (IDs and names) in Figs. 1d and e, respectively.

Binding conformations:

Figure 2 displays the three booster systems after 100 ns of simulations. The systems show some similarities but also remarkable differences. To analyze better the observations, we tracked the amino acids in RBD and AF which are in close contacts with the faces of boosters over the last 50 ns (500 frames) of the trajectories. We counted contact frames for amino acids where they are within 3 Å of their targets (Face 1 or Face 2). Figures S2-4 show the number of contact frames of RBD and AF amino acids which are highly involved in the interactions with the three boosters. In Table S1, we listed those hot spots (amino acids) in RBD and AF which have more than 250 contact frames out of 500 frames (last 50 ns). Most of these amino acids are polar. As the Face 1 designs are ACE2-mimics, they bind to RBD in almost the same manner as ACE2 [17]. Analogously, the Face 2 designs, based on the Hepatitis B core-antigen, should bind to AF.

FIG. 2:

Simulated booster, RBD and AF complexes. (a-c) Final conformations of Booster 1–3 systems at 100 ns. (d) Averaged RMSD for Face 1 (ACE2-mimic, blue bar) and Face 2 (antigen, orange bar); (e) Averaged free energy of binding of RBD with Face 1 (blue bar) and antibody with Face 2 (orange bar).

Figure 2a - Booster 1 reveals that the two faces stay tightly bound with both RBD and AF, while at the same time the peptide structure remains largely preserved (insets). The RBD hot spots in binding with Face 1 are formed by 16 amino acids, including two new contacts marked in red (Table S1). The AF hot spots in binding with Face 2 are formed by 14 amino acids, where half of them are new binding contacts. Here, the new binding contacts refer to amino acids which are not listed in the initially binding sets of amino acids, shown in Figs. 1d and e.

Figure 2b - Booster 2 shows that the conformations of the $\alpha$ helices and the $\beta$ hairpin remain largely intact during their binding with RBD. Although Face 1 has an extra hairpin and coil structure, compared to Booster 1, it has the same number of hot spots, with no new contacts forming. The hot spots on AF are formed by 19 amino acids, with one new contact (202T), which is less than the initial contact number shown in Fig. 1d. Although Booster 2 has the longest sequence, it doesn’t generate more contacts in the interface of RBD and Face 1.

Figure 2c - Booster 3 shows that Face 1 will likely dissociate from RBD since only 6 hot spots remain bound, but Face 2 still binds to the AF with 14 amino acid hot spots. These results reveal that when the faces are attached by a long chain with the connecting points close to the sides of faces, the multivalent binding of the faces to their targets can be released by a fluctuative pulling generated by the linker, in analogy to unzipping a double-stranded DNA from its ends. In contrast, if the same pulling was applied on the whole face or at least several of its regions, it would be difficult to disturb the multivalent binding between the faces and their targets. In Boosters 1 and 2, the faces connected by a short peptide bond also develop other binding within faces, so they behave like a rigid body which can preserve the multivalency in binding with RBD or AF. In principle, one could preserve the multivalent binding in Booster 3 by joining its faces with several PEG chains or at least having the attaching points at the centers of their faces.

RMSDs and free energies of binding:

To further quantify the booster-target binding, we calculated for each booster the Root-Mean-Square Deviation (RMSD) and the free energy of binding, $\text{Δ}{G}_{MMGB-SA}$. Figure S5 shows the time dependent RMSDs of individual faces and whole boosters. RMSDs of all three Boosters are large compared to their faces, revealing a relatively small rigidity of these Boosters. Moreover, Booster 3 has large RMSD fluctuations, which become eventually responsible for its unbinding in Face 1. Figure 2d shows the average RMSDs for the two faces obtained in the last 50 ns of simulations. Similar faces present in Boosters 1 and 3 have similar RMSDs, revealing that the instability of Booster 3 doesn’t originate in Face 1, but in its PEG linker. Booster 2 with a somewhat different (larger) Face 1 shows a larger RMSD for Face 1, but this doesn’t destabilize its binding. Overall, RMSD values reflect the complexity of the involved faces.

Figure 2e shows $\text{Δ}{G}_{MMGB-SA}$ calculated for each face coupled to its target. Booster 1 has the strongest overall binding with RBD and AF, while Booster 3 has the weakest binding in both faces, where Face 1 of Booster 3 tends to dissociate from RBD. Booster 2 has a free energy of binding (stability) positioned somewhere between Booster 1 and 3.

To better understand the booster-target binding, the interaction energies (enthalpies) of binding components were separately calculated over the last 50 ns of simulations (Figs. S6-8) and separated into Coulombic and van der Waals (vdW) contributions. The interaction energies were calculated by NAMD Energy plugin in VMD [18], where the dielectric constant was set to 1. The large electrostatic binding energy contributions in all systems could be somewhat scaled down to reflect on the presence of water around the binding regions. However, the interaction energies again reveal a relatively stable binding of Boosters 1 and 2, as compared to Booster 3, which is in line with the free energy calculations.

In summary, using classical MD simulations, we have shown that double-faced boosters provide highly promising pathways for targeting novel viruses by generic antibodies, in particular SARS-CoV-2 by the Hepatitis B antibody. By allowing the immune system to recognize new viruses by antibodies preexisting in the organisms, one can establish new generic therapeutic methods, which could be used in a fast treatment of emerging pathogens.

MD simulations:

The two faces of boosters were separately bound to RBD and AF. All structures were directly based on the crystal structure of the human ACE2 protein bound to RBD of SARS-CoV-2 (pdbID: 6LZG) [15] and Hepatitis B antigen bound to AF [16]. Snapshots were taken by VMD [18].

The systems were simulated using NAMD2 [19], the CHARMM36 protein force field [20] and the CHARMM36 general force field. The simulations were conducted with the Langevin dynamics (${\gamma }_{Lang}=1$ ps⁻¹) in the NpT ensemble at temperature of $T=310$ K and pressure of $p=1$ bar. The particle-mesh Ewald (PME) method was used to evaluate a long range Coulombic coupling, with periodic boundary conditions applied [21]. The time step was set to 2 fs. The long range van der Waals and Coulombic coupling were evaluated every 1 and 2 time steps, respectively. After 2, 000 steps of minimization, the solvent molecules were equilibrated for 3 ns, while the complexes were restrained using harmonic forces with a spring constant of 1 kcal/(mol Å). Next, the systems were equilibrated in 100 ns production MD runs with restraints on the top part of AF. All systems were simulated in 150 mM NaCl solutions with the TIP3P water model [22].

RMSD calculations:

The time-dependent RMSD for Face 1 and Face 2 (Fig. S4) were calculated from

$$RMS{D}_{\alpha }(t_j)=\sqrt{\frac{\sum _{\alpha =1}^{N_{\alpha }}{({\overrightarrow{r}}_{\alpha }(t_j)-{\overrightarrow{r}}_{\alpha }(t_0))}^2}{N_{\alpha }}},$$ (1)

where $N_{\alpha }$ is the number of atoms whose positions are being compared, ${\overrightarrow{r}}_{\alpha }(t_j)$ is the position of atom $\alpha$ at time $t_j$ and ${\overrightarrow{r}}_{\alpha }(t_0)$ is the initial coordinate. The selection of coordinates contains all the atoms in Face 1 or Face 2, excluding hydrogens. The time-dependent RMSD was averaged over the last 50 ns of simulation time, which corresponds to the last 500 frames of each trajectory as shown in Fig. 2d. The standard deviations were shown by the error bars.

MMGB-SA calculations:

We used the Molecular Mechanics Generalized Born - Surface Area (MMGB-SA) method [23, 24] to estimate the relative binding free energies between booster faces and their binders (RBD or AF). The free energies were estimated from separate MMGB-SA calculations for three systems related to the face and its binder (the face, the binder of the face, and the complex of the face and its binder) in configurations extracted from the MD trajectories of the whole complex in the explicit solvent. The MMGB-SA free energies of the extracted configurations of the three systems were calculated as

$$G_{tot}=E_{MM}+G_{solv-p}+G_{solv-np}-T\text{Δ}{S}_{conf},$$

where $E_{MM}$, $G_{solv-p}$, $G_{solv-np}$, and $\text{Δ}{S}_{conf}$ are the sum of bonded and Lennard-Jones energy terms, the polar contribution to the solvation energy, the nonpolar contribution, and the conformational entropy, respectively. The $E_{MM}$, $G_{solv-p}$ and $G_{solv-np}$ terms were calculated using the NAMD 2 package [19] generalized Born implicit solvent model [25], with a solvent dielectric constant of $\epsilon =78.5$. The $G_{solv-np}$ term for each system configuration was calculated in NAMD as a linear function of the solvent-accessible surface area (SASA), determined using a probe radius of 1.4 Å, as $G_{solv-np}=\text{SASA}$ $\gamma$, where $\gamma =0.00542$ kcal/(mol Ų) is the surface tension. The $\text{Δ}{S}_{conf}$ term was neglected, as the entropy term is often calculated with a large computational cost and low prediction accuracy, which is likely to be similar for the studied systems, which differ in the connecting part of the two faces. [26, 27]. Since the $G_{tot}$ values are obtained for configurations extracted from the trajectories of complexes, $G_{tot}$ doesn’t include the free energies of faces reorganization; the correct free energies of binding should consider configurations of separately relaxed systems. The approximate binding free energies of the studied complexes were calculated as $\langle \text{Δ}{G}_{MMGB-SA}\rangle =\langle G_{tot}(face-binder)-G_{tot}(face)-G_{tot}(binder)\rangle$, where face-binder represents the complex of face with its binder, and the $\langle \text{averaging}\rangle$ is performed over configurations within the second half of the calculated trajectories.