Design and study of bioisosteric analogues of the drug Molnupiravir as potential therapeutics against SARS-COV-2: an in silico approach

Abstract

The ongoing SARS-CoV-2 pandemic has created an urgent need for effective antiviral drugs that can be rapidly developed and utilized to treat patients infected with the virus. Molnupiravir, a direct-acting oral antiviral, has shown promising results in reducing viral infections with SARS-CoV-2. Nonetheless, there is still a need for the development of more efficacious analogues with enhanced interaction with the specific target, the RNA dependent RNA polymerase (RdRp) and better physico-chemical profile. Our study is based on a rational design strategy, known as “bioisosterism”, to design some analogues. The pool of bioisosteric structural analogues was further enriched using the “SwissBioisostere” database. Only structures with a Tanimoto score more than 0.85 (calculated using the Maximum Common Substructure scoring method) and with ΔlogP (lipophilicity) ± 1 and ΔPSA (Polar Surface Area) ± 10 Å were retained. Next, molecular docking studies were conducted using AutoDock Vina®. Ligand and receptor preparation and molecular interaction analysis were performed using UCSF Chimera® and Biovia Discovery Studio®, respectively. The three-dimensional structure of the RdRp of SARS-CoV-2 (6M71) was sourced from RCSB PDB®. Ligands were prepared in 3D, and the receptor underwent solvent removal, elimination of alternative positions, hydrogen atom addition, and partial charge assignment. Binding pocket coordinates were determined, and utilized for AutoDock Vina® docking. Parallelly, the druglikeness of our molecules was predicted using the website ADME-SWISS: http://www.swissadme.ch/, based on Lipinski and Weber scores. Docking outcomes, combined to druglikeness prediction results, identified two fluorinated analogues with superior binding affinity (lowest score and an RMSD ≤ 2 Å) and improved physico-chemical properties (no violation of Lipinsky and Veber rules). This study contributes to the development of more effective antiviral drugs by providing insights into potential uegs with enhanced interactions with RNA polymerase and better druglikeness profile.

Graphical abstract

Introduction

The emergence of the novel SARS-CoV-2 virus in December 2019 has unleashed a global health crisis all over the world. Health authorities and pharmaceutical companies worldwide have been working with extraordinary urgency to develop effective treatments for the Coronavirus disease within reduced timelines. In this critical endeavor, in silico prediction studies have risen to prominence, offering a powerful tool to expedite drug discovery, reduce development costs and optimize therapeutic interventions. The underlying principle of the in silico approach is grounded in the premise that chemically similar compounds often share similar biological properties, including both therapeutic activity and potential toxicity.

In the same context, the pharmaceutical industry has turned to drugs that have already achieved regulatory approval and are accessible on the market for the treatment of other pathologies. This strategy aims to identify rapidly molecules with the potential to interact with one or more of the well-established therapeutic targets of the virus. The most relevant examples include Angiotensin II Receptor Antagonists (ARAII) (Arcos et al. 2020; Lamas-Barreiro et al. 2020) and several antiviral agents, such as remdesivir, favipiravir, ritonavir, ivermectin and interferons alpha-, beta- and lambda. Notably, several of these drugs have been introduced as antiviral agents against SARS-CoV-2 and have received approval from the Food and Drug Administration (FDA) (National Institute of Health. Covid19 treatment guidelines. 2023. Available at: Table: Characteristics of Antiviral Agents and Antibody Products COVID-19 Treatment Guidelines (nih.gov)).

Within this dynamic landscape, Molnupiravir (MOL), an experimental broad-spectrum antiviral, has gained significant attention for its potential in the treatment of the SARS-CoV-2 disease. The specific target of MOL is the RNA-dependent RNA polymerase (RdRp). Unlike remdesivir, a previously approved antiviral medication, MOL offers the advantage of a simplified oral dosing regimen (800 mg*2 per day during 5 days) (Yasri et al. 2022). This regimen not only enhances patient adherence, but also decrease the need for hospitalization. Preliminary results from ongoing Phase II/III trials are encouraging, as they have signaled a reduced risk of hospitalization and mortality compared to control groups (p = 0.0012) (Haute Autorité de Santé. 2023. Available at: avis_ansm_aap_molnupiravir.pdf (has-sante.fr)).

From the other hand, the World Health Organization (WHO) recognized the potential of MOL and included it, in March 2022, into its guidelines for the treatment of SARS-CoV-2 (Le molnupiravir intégré dans les orientations actualisées de l’OMS sur les traitements contre la COVID-19. World Health Organisation. 2023. Available at: https://www.who.int/fr/news/item/03-03-2022-molnupiravir).

From a chemical point of view, MOL is the isopropyl ester prodrug of EIDD-1931 (Fig. 1) and is rapidly converted into EIDD-1931 in the plasma by the host’s esterase. This chemical modification was designed to address the low bioavailability associated with this latter (Hernandez-Santiago et al. 2004).

Fig. 1.

Chemical structures of Molnupiravir (A) and its precursor, (the EIDD-1931) (B)

To sum up, there is still a scope to deepen our comprehension of MOL and investigate other structural analogues, with the aim to identify compounds with enhanced affinity for the target and better pharmacological properties. In this work, we present a comprehensive investigation, encompassing in silico design and assessment of various bioisosteric analogues of Molnupiravir.

Materials and methods

Design of structural analogues

The strategy of "bioisosterism" was adopted for the rational design of structural analogues of the MOL. Two types of analogues were designed: "classic" bioisosteres (based on the classical rules of isosterism between atoms and chemical groups) and "complex" bioisosteres (using equivalences between complex structural scaffolds). For more details, see the “Spplementary information section”. In order to enlarge the panel of bioisosteric structural analogues, the "SwissBioisostere”, a free database of molecular replacements for ligand design, was used (Fig. 2).

Fig. 2.

SwissBioisostere® interface

In order to refine the search for bioisosteres and limit their number to the most similar ones, a double filter using the difference in logP (ΔlogP) and polar surface area (ΔPSA) was applied: only structures with a ΔlogP ± 1 and a ΔPSA ± 10 Å will be retained and examined.

The Tanimoto score, which is the ratio of the number of molecular fingerprints in common between two molecules, was calculated for all analogues using the freely accessible online tool: https://chemminetools.ucr.edu/. We selected those with a score greater than 0.85 (Patterson et al. 1996). Note that we specifically chose the MCS Tanimoto score (Maximum Common Substructure) to evaluate structural similarity, rather than the scoring method AP. In fact, the MCS is a similarity measure that seeks the largest common substructure between two molecules. It identifies the largest portion of molecules that is identical between the two structures. The MCS is often used to find important common substructures among different molecules.

Molecular docking study

Tools and software

UCSF Chimera® software was used for the preparation of the ligand and receptor for docking and the visualization of the interaction.

AutoDock Vina, a free access software (http://vina.scripps.edu) was used for docking.

Biovia Discovery Studio® allowed us to determine the coordinates of the active site on the receptor and perform the analysis of molecular interactions between the ligand and the target receptor.

During our study, protein database RCSB PDB (Research Collaboratory for Structural Bioinformatics Protein Data Bank) was used to extract the three-dimensional structure of the RdRp of SARS-CoV-2, and its identifier is "6M71".

Steps before docking

The ligand is first drawn in 2D using ChemDraw Professional software (version 12.0.2.1076) and converted into a SMILES code. The SMILES code is then copied into the dedicated space on the CHIMERA® software to obtain the 3D structure of the ligand. Finally, hydrogen atoms are added to the structure. The receptor should then be prepared. After downloading the "6M71" PDB file and opening it with the UCSF CHIMERA software, the DOCK PREP function of the software is used, and allows: removal of the solvent in the crystallized format, removal of alternative positions (retaining only the most occupied positions), addition of hydrogen atoms at the structural level and assignment of partial charges.

Finally, the location of the pocket is performed by the DISCOVERY STUDIO® 3.5 software using the "Edit and define binding site" function. The pocket is determined by a yellow sphere that allows us to obtain the coordinates of the active site in the form (xxx, yyy, zzz) which are then inserted into the AutoDock Vina® software.

Docking using Autodock Vina®

Docking was carried out using Autodock vina, and the results of the anchored complex were analyzed. Next, for each pose, the following parameters will be assigned: a Vina score (wich reflects the affinity energy expressed in kcal mol⁻¹), the RMSD score (upper bound, u.p and lower bound, l.b) and the number of hydrogen bonds established between the receptor and the ligand. The best pose of the ligand at its receptor is the one with the lowest binding energy and the lowest RMSD. Afterwards, this pose is inputted into the Discovery studio® software to generate a 2D image showing the various non-covalent interactions.

To validate our docking protocol, we performed docking of the MOL co-crystallized with its target five times. The results should be repeatable and all should have a similar score with an RMSD of less than 2 Å or close to 2 Å. Additionally, the best pose of each trial should also have ligand-target interactions similar to those observed during docking.

Prediction of structural analogues' druglikeness

The druglikeness of our analogues was predicted using the ADME-SWISS website: http://www.swissadme.ch/. Lipinski and Weber scores were calculated, and PAINS (Pan-Assay Interference compounds) were searched simultaneously.

Results and discussion

Design of analogues and in silico-study

All analogues were designed on the basis of bioisosteric rules and/or provided by the "SwissBioisoster" database. They are listed in (Tables 1, 2 and 3).

Table 1.

Bioisosteric analogues of Molnupiravir and their correspondent Tanimoto score

Table 2.

Bioisosteric analogues provided by BioisosterSwiss® database and their correspondent Tanimoto score

Table 3.

Analogues designed based on structural similarity with other antivirals effective against the target (such as remdesivir) and their correspondent Tanimoto score

Bioisosteric approaches are widely used as effective tools in rational drug design processes. The design of bioisosteres frequently introduces changes at the structural level, affecting key properties in target recognition such as molecular size, topology, electronic distribution, polarizability, polarity, lipophilicity, and pKa. These effects can be beneficial or detrimental, including improved efficacy, selectivity, changes in physical properties, reduction in metabolism and elimination, modification of toxicophores, etc. Some substitutions are simple. Others are more complex and are based on equivalence between scaffolds considered as "mimetics". Complex bioisosteres would thus likely be different in terms of electronic distribution, physicochemical properties, steric and topological effects. Among recent bioisosteric replacements, deuteration, which involves replacing hydrogen with deuterium, improves the chemical stability of a molecule, especially in cases where molecules have a chiral center that racemizes in vivo. Deuteration also reduces metabolism by Cytochrome P450 and aldheyde oxidase enzymes. Fluorination is commonly used to modulate metabolism or target activity, as the C-F bond is strong and resistant to metabolic cleavage. Fluorine is also used to modulate the pKa of a basic nitrogen. Replacing an OH group with an NH₂ group results in significant changes in the acid–base properties of the molecule. The replacement of an oxygen atom with a sulfur atom in an ether-oxide yields thioethers, which are less polar than the corresponding oxygen-containing compounds. The rules for replacement between chemical groups have led to the generation of analogues that are highly similar to the parent molecule. A Tanimoto similarity score of 0.85 or greater is considered a reliable indicator of similar activity between two compounds. However, the reliability of this score depends on various factors, such as the size of the database of molecules and the type of structures used for analysis (Dunkel et al. 2008; Maggiora et al. 2014; Mellor et al. 2019).

All of the designed bioisosteric analogues (compounds 2–30, except 19, 20, 21 and 22) showed a Tanimoto score lower than 0.85. These four compounds correspond to remdesivir-like compounds designed based on a non-rational approach. Interestingly, these analogues had the lowest Tanimoto scores (0.4000 and 0.5806, respectively). These results indicate that the geometrical similarity with the MOL is no longer maintained, suggesting that the approach used to design these analogues may not be optimal for developing effective treatments for COVID-19.

Results of the molecular docking

In total, we investigated 30 structural analogues through molecular docking studies. The validation of the docking protocol was performed in two parts: docking of the co-crystallized ligand with the target receptor and comparison of the chemical interactions between the best docking pose and the co-crystallized ligand. The results of the 5 trials were reproducible (standard deviation σ = 0.048 and variance σ2 = 0.0024). The mean of the 5 trials was -6.76 kcal/mol, with an RMSD ≤ 2 Å, indicating that the docking was successful overall (Hevener et al. 2009; Meli et al. 2020). The RMSD plays a fundamental role in comparing different conformations of the same ligand with respect to a given receptor. As the docking software provides different ligand poses, it is particularly important to evaluate them using RMSD calculations. The RMSD is divided into rmsd/lb (RMSD lower bound) and rmsd/ub (RMSD upper bound). For the analysis of our results, we considered the rmsd/ub. Indeed, the lower this value, the better the alignment between the two structures. In addition, we used rmsd/ub because it examines each atom in one conformation with the same atom in another conformation, ignoring any symmetry (Corbo et al. 2022).

To validate the docking method, a comparison of the chemical interactions produced by the co-crystallized MOL (PDB file 7OZV) and those produced after docking the MOL (10,563 residues) was performed. Similarity was observed between the MOL-target chemical interactions produced by docking and those existing in the MOL co-crystallized with RdRp. We detected in common: hydrogen bonds, electrostatic bonds involving the attraction between two charges of opposite sign, in this case the N11 and N13 nitrogen atoms carrying a partial positive charge (resulting from electron delocalization in the urea motif) and the COO⁻ of the aspartic acid residue; conventional carbon-hydrogen bonds; low energy Van der Waals type bonds (Figs. 3, 4, 5). Additionally, docking showed the presence of Pi-alkyl type interactions, involving the aromatic amino acid residue tyrosine. The Mol was bound to the finger subdomain of RdRp by 5 hydrogen bonds, with residues THR B 859, SER B 735, SER A 316, and GLN A 314. Furthermore, it formed an electrostatic bond with ASP B 737 and a carbon-hydrogen bond with VAL B 736. Both results presented a similar binding free energy with an average of − 6.76 kcal/mol. This average belongs to the predefined reference range of binding energy ranging from − 5 to − 15 kcal/mol, already used in the validation of a docking protocol by Shah et al. Our docking score is similar to the study of (Patil et al. 2021), where the Mol presented a score of − 7.3 kcal/mol (versus − 6.8 kcal/mol in our case). This difference of -0.5 can be explained by the difference in the docking software used, the nature of the downloaded PDB file, and finally the ligand interaction site at the target level. The same study also demonstrated an energy of − 6.9 kcal/mol for the interaction between remdesivir and RdRp. This score is almost identical to that of the Mol in our study. It can be concluded that both antivirals have the same stability towards the target in question.

Fig. 3.

Chemical interactions of Molnupiravir with amino acid residues in its target, RdRp, generated by Discovery Studio® 2021 software

Fig. 4.

Chemical interactions of Molnupiravir co-crystallized with amino acid residues of the target, RdRp, generated by Discovery Studio® 2021 software

Fig. 5.

The 3D chemical interactions of Molnupiravir with amino acid residues of the target, RdRp

The best result was obtained with analogues 7 and 9, which have the lowest score and an RMSD ≤ 2 Å. Compound 20 revealed to have the highest score, namely − 4.7 kcal/mol, but its RMSD remains ≤ 2 Å. Note that these results do not allow us to correlate docking and Tanimoto score results, since correct scores were obtained with structures that are not similar to the MOL (compound 20), and vice versa. Figure 6 summarizes the docking results.

Fig. 6.

Summary diagram of docking results

The fluorinated analogues 7 and 9 showed the best docking results with a score lower than that of MOL (Fig. 7). The different interactions identified with these analogues are shown in Figs. 8, 9, and 10.

Fig. 7.

Structure of the two fluorinated analogues 7 and 9 showing the best interactions with the RNA polymerase

Fig. 8.

Chemical interactions of analogue 7 with amino acid residues of the target, RdRp, generated by Discovery Studio 2021 software (2D structure)

Fig. 9.

Chemical interactions of analogue 9 with amino acid residues of the target, RdRp, generated by Discovery Studio 2021 software (2D structure)

Fig. 10.

Interaction site of analogues 7 and 9 on the RdRp target (3D)

These two analogues have in common the presence of a fluorine atom. Fluorine, being strongly electronegative, generates a new interaction with the target residues, called "halogen bonding." Indeed, the fluorine atom on the cycle at position 9 of analogue 7 created this bond with the GLN A: 292 residue of the target. This same fluorine interacted with the ARG A: 735 residue of the target by generating a hydrogen bond.

Fluorine can interact with both polar and hydrophobic groups within proteins. Polar interactions involve hydrogen bond donors (NH, polarized C_α-H bond, water bound to proteins, lateral residues), or hydrophobic interactions with lipophilic lateral chains. Orthogonal multipolar interactions can occur with carbonyl groups, amide functions (Asn and Gln), guanidine groups (Arg), and sulfur bridges (Cys) (Bégué and Bonnet 2006; Kirk et al. 2006). Some of these interactions were highlighted by the docking in our study. In conclusion, the presence of the fluorine atom contributed to improving the affinity with the target since the free binding energy decreased in both cases. Finally, it is worth noting that introducing fluorine on the methylene showed slightly better results than those obtained by introducing fluorine on the cytosine nucleus.

In silico prediction of structural analogues' druglikeness

The druglikeness of our molecules was predicted using the website ADME-SWISS: http://www.swissadme.ch/, based on Lipinski's Rule of Five, which states that if chemical molecules meet certain parameters, they have a good probability of being orally administered.

The MOL, the 5 deuterated analogues, analogues 6 and 7, 10, 11, 12, 13, 14, 15, 16, 18, 19, 21, and 23–30 all comply with Lipinski's rule, with 0 violations. They are all potentially good candidates for oral drugs. However, they all violate Veber's rule, with a TPSA > 140 A°. TPSA is an important parameter used to estimate the drug's bioavailability. If TPSA exceeds the value of 140 A°, poor intestinal absorption is possible. We can therefore expect poor intestinal absorption for the MOL and the 5 deuterated analogues. However, clinical studies have shown that the MOL is well absorbed orally and has linear pharmacokinetics between doses of 50–1600 mg (Pourkarim et al. 2022). In conclusion, the violation of a single parameter, such as TPSA, does not necessarily mean that the drug will not be well absorbed orally.

Analogues 8 and 9 do not violate any of these rules. They can be good candidates for oral drugs. Analogues 15, 17, 20, and 22 all violate Lipinski's rule and at least one violation of Veber's rule. The violation always concerns TPSA, which is greater than 140. They are unlikely to be good candidates for oral drugs. Finally, it should be noted that the strict application of Lipinski's pre-filtering rule can easily lead to the exclusion of molecules of interest. For example, many marketed drugs do not comply with Lipinski's rules, such as atorvastatin and cyclosporine (Oprea 2000; Sekfali 2021).

Conclusion

Our research has contributed to enrich published data while using cutting-edge technology. Second, we tested the performance of several designed bioisosters of Molnupiravir based on the "bioisosterism" strategy and ADMET properties. Molecular docking of these analogues has shown to be promising, with two analogues showing very good results in terms of interaction with viral RNA polymerase. In a potential future work, these analogues could be the subject of chemical synthesis, study of their physicochemical properties, and even clinical studies against COVID-19.

Author contributions

All authors participated to the experimental work. DG, AAEH and HS wrote the manuscript. All authors reviewed the manuscript.

Funding

Not applicable.

Availability of data and materials

All data and materials used in this work were detailed in the “Materials and methods” section.

Declarations

Conflict of interest

The author declares no conflict of interest.

Ethical approval

Not applicable.

Informed consent

Not applicable.