Exploring polyamine interactions and binding pockets in SARS-CoV-2 ORF3a

Abstract

Ongoing global pandemic caused by coronavirus (COVID-19) requires urgent development of vaccines, treatments, and diagnostic tools. Open reading frame 3a (ORF3a) from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is considered to be a potential drug target for COVID-19 treatment. ORF3a is an accessory protein that plays a significant role in virus-host interactions and in facilitating host immune responses. Using putrescine, spermidine and spermine, an aliphatic polyamine for the activity suppression of ORF3a appears to be a promising approach in finding new targets for drug design. In this study, we explored the possible binding poses of polyamines to the ORF3a protein using a combination of various computational approaches i.e. pocket prediction, blind and site-specific molecular docking, molecular dynamics and ligand flooding simulations. The results showed that the tip of cytoplasmic domain and the upper tunnel of transmembrane domain of ORF3a provide a suitable binding site specific for the polyamines. MD simulations revealed the stability of spermidine binding in the upper tunnel pocket of ORF3a through salt bridge and hydrogen bond interactions between the amine groups of the ligand and negatively charged residues of ORF3a. These findings can be helpful in designing new therapeutic drugs.

Graphical abstract

1. Introduction

The global pandemic of severe acute respiratory syndrome from coronavirus-2 (SARS-CoV-2), causing the infectious disease known as COVID-19 has a huge detrimental effect on public health care around the world. SARS-CoV-2 has a large single-positive-strand RNA genome which consists of approximately ∼30 kb [1] and shares about 82% sequence identity with the 2002–2004 SARS-CoV [2]. In addition, the SARS-CoV-2 genome contains 29 open reading frames (ORFs) that encode four structural, sixteen non-structural and eleven accessory proteins [2]. The four structural proteins, spike (S), envelope (E), membrane (M), and nucleocapsid (N), are required for viral transmission and reproduction, are encoded by ORFs 2, 4, 5 and 9a, respectively [3]. These proteins, common to all members of the coronavirus family [4], are targets for the development of drugs, vaccines, and other therapeutic agents. The non-structural proteins (nsp1-16) are involved in genome replication and early transcription regulation. Roles of the accessory proteins (ORF3a, ORF3b, ORF3c, ORF3d, ORF6, ORF7a, ORF7b, ORF8, ORF9b, ORF9c and ORF10), are, however, not yet fully understood [2,5,6]. The protein ORF3a is of special interest due to availability of its three-dimensional structure [7] useful in designing drug against COVID-19 [6,7]. ORF3a is a small, non-structural protein that is involved in various processes during the virus's lifecycle, but its precise function is not fully understood [6]. It has been proposed that ORF3a plays an important role for viral release, host immune evasion and cytopathogenic effects. ORF3a is also involved in inflammasome activation, apoptosis, and necrosis and therefore, a promising target for COVID-19 treatment.

ORF3a from SARS-CoV-2 is an accessory protein with 275 amino acid residues and shares ∼73% amino acid identity with that from SARS-CoV [6]. The protein is classified as an integral membrane protein as its transmembrane domain is anchored within the membrane. The transmembrane domain of ORF3a helps facilitating virus release through the lysosomal exocytosis pathway [6]. The three-dimensional structure of SARS-CoV-2 ORF3a obtained by cryogenic electron microscopy (cryo-EM) provides insights into the structural arrangement in the transmembrane domain (Fig. 1 A) [7]. The ORF3a protein assembles to form a dimer. Each chain contains transmembrane domain (TMD; a.a. 40–128) and cytoplasmic domain (CD; a.a. 144–236). Its N-terminus is on the extracellular side and its C-terminus is in cytosol. The TMD region is composed of three transmembrane helices (TM1, TM2 and TM3) spanning across the lipid bilayer and connecting to the CD through TM-CD linker (a.a. 136–143). The three transmembrane segments are arranged in a clockwise direction. The transmembrane domain of ORF3a forms a tunnel-like structure that could provide a pathway for molecules or ions to pass through the membrane; it could also serve as a binding site for other molecules [[6], [7], [8]]. The exact function of the tunnel in the transmembrane domain of ORF3a is not yet fully understood. The CD is composes of eight antiparallel β-sheets, a short cytoplasmic loop (a.a. 175–180) and C-terminal end (a.a. 237–240). Functional studies indicated that ORF3a is permeable for Ca²⁺. Elevation of Ca²⁺ concentration at cytoplasm was detected in ORF3a-expressing cells during viral releasing process [6]. However, Kern et al. [7] demonstrated that SARS-CoV-2 ORF3a exhibits non-selective cation channel activity. Not only Ca²⁺ ion can be permeated but also other cationic species such as Na⁺, K⁺ and N-methyl-d-glucamine (NMDG⁺) ions. While ORF3a has been found to exhibit ion channel activity, different studies with conflicting results have also been reported [9]. It is known that ORF3a is a protein encoded by a viral genome and its function is to modulate the host cell's innate immune response. It is possible that the ion channel activity of ORF3a is too weak to be detectable experimentally and/or the protein has multiple functions. Thus, the functional activities and the underlying mechanism of this viral protein remains challenging and is therefore worth investigating.

Fig. 1.

(A) Cryo-EM structure of ORF3a dimer (PDB ID: 7KJR) [7] Each part of the protein is represented in different colors. (B) Structures of SPD (triamine), SPM (tetraamine) and PUT (diamine). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

The natural polyamines such as spermidine (SPD), spermine (SPM) and putrescine (PUT) that are present all living cells play important roles in a variety of biological processes such as protein synthesis, cell division and cell growth. SPD (Fig. 1B) is a triamine compound that is involved in cellular metabolism including DNA synthesis and repair [10]. It is known as autophagy activator [11]. It has been shown to have some effect on certain membrane protein ion channels including voltage-gated potassium channels, calcium channels, and glutamate receptors [7,10,12]. For example, SPD has been shown to inhibit voltage-gated potassium channels in some specific neurons that can affect their excitability. SPD has also been shown to modulate the activity of certain types of calcium channels, which can affect the intracellular concentration of calcium and thereby various cellular processes. It should be noted that the effects of SPD on ion channels are far from fully understood and require further investigations.

In recent years, several studies have been shown the effect of targeting polyamines as well as the pathway in viral infection suggesting that the important role of polyamines in the occurrence and development of viral replication, including severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), coronavirus middle east respiratory syndrome virus (MERS-CoV), influenza virus, dengue virus (DENV) and human immunodeficiency virus (HIV) [13,14]. For SARS-CoV-2, there is limited information available on the effects of spermidine on ORF3a. It has been shown that SPD and SPM hinder coronavirus replication by reducing cellular attachment and entry up to 66% at 100 μM [11]. Kern et al. [7] also found that SPD blocks the ion conduction of ORF3a protein at 10 mM almost completely. SPD, SPM and PUT are in general weak base molecules as they can accept a proton from an acid to form a cationic species (positively charged ion) [10,15]. SPD has three nitrogen atoms, each of which can act as a basic site (Fig. 1B). The presence of multiple basic sites allows SPD to interact with acidic and/or polar residues in a protein through salt-bridge or hydrogen bond (H-bond), and thus contribute to the stability of the protein-polyamine complex [10]. While the polyamines appear to be a potent inhibitor of ORF3a, specific interactions at the atomic level are unknown. In this study, we employed various computational approaches to explore interactions between the accessory protein ORF3a and polyamine molecules. We used pocket prediction, molecular docking, molecular dynamics (MD) simulations to provide a reliable structural insight into the binding poses of ORF3a-polyamine complex. This can guide the rational design of polyamine inhibitors for targeting the ORF3a accessory protein.

2. Computational details

2.1. Ligand preparations

3D structure models for spermidine (SPD), spermine (SPM), and putrescine (PUT) were downloaded as SDF files (.sdf) from the PubChem Compound (https://pubchem.ncbi.nlm.nih.gov) [16] and converted into the PDB format (.pdb) using Open Babel graphical user interface (GUI) tool [17]. Protonation state of the amine groups was predicted based on pK_a value using PROPKA software [18]. The hydrogen atoms were primarily added by visual molecular dynamics (VMD) version 1.9.3 [19]. Since the polyamines are weak base with pK_a ranging from 7.9 to 10.9, all amine groups present in the ligands were in an protonated form (Fig. 1B) [10,15]. CHARMM-GUI web server (http://www.charmm-gui.org) [20] was used to generate ligand topologies and CHARMM General Force Field (CGenFF) parameters [21].

2.2. Pocket detection analysis

The cryo-EM structure of ORF3a dimer was taken from PDB entry 7KJR as shown in Fig. 1A [7]. Missing residues in the short loops of the ORF3a protein were modeled using the program Loopy [22]. The resulting structure was subsequently refined and further used to predict the ligand binding sites of the protein. We used the Fpocket program to search putative pockets for ligands in the protein cavities [23]. The Fpocket analysis employed Voronoi tessellation algorithm and sequential clustering steps to extract pockets in proteins. The predicted pocket sites were chosen based on the druggability score derived from physio-chemical descriptors of the pocket together with careful visual inspection. The candidate pockets for ligands were further validated using molecular docking methods and MD simulations which are described in the next section.

2.3. Molecular docking

Predictions of the binding poses of polyamines to the residues of the ORF3a protein were carried out using docking approaches. We employed three commonly used docking programs: Autodock Vina (ADVN) [24], and AutoDock 4.2.6 (AD) [25] and RosettaLigand [26]. The identified pockets predicted by Fpocket were used as guideline to define the docking region of the protein (Fig. S1). For ADVN and AD, AutodockTools 1.5.7 (ADT) [25] were used to prepare docking input parameters, i.e. atom types and charges (PBDQT format), box size, grid space etc. The ORF3a protein was processed by i) adding all hydrogen, ii) merging non-polar hydrogen, and iii) adding polar hydrogen atoms using ADT. Gasteiger atomic charges were assigned for the protein and ligands. During docking simulations, each ligand molecule (SPD, SPM and PUT) was computationally docked into a pre-defined docking region of the protein with a grid spacing of 0.375 Å. For ADVN, the exhaustiveness parameter was set to 50. Other parameters such as the step size for translation, quaternion and torsion angles were set as default. Torsional degrees of freedom in the polyamine ligands were detected by ADT. We employed two algorithms for AD, Monte Carlo simulated annealing (MCSA) [27] and Lamarkian genetic algorithm (LGA) [28] with flexible ligand-rigid protein docking. Maximum numbers of steps per cycle (accepted and rejected) is 25,000 for MCSA. For LGA, the number of individuals (configurations) in the population generated in each cycle of the genetic algorithm is 150, while 2,500,000 maximum number of energy evaluations are performed with 27,000 maximum number of generations. With LGA, only the best fittest configuration in the given population was chosen to survive and become a potential candidate for producing new configurations in the next generation with a 0.02 rate of mutation with 0.8 rate of crossover [29]. 50 independent runs were used for both algorithms.

Both protein and ligand were fully flexible to perform RosettaLigand docking. The conformation ensemble of ligands was constructed using the Meiler lab's BioChemicalLibrary (BCL) [30]. 500 conformations of each ligand were generated for docking. RosettaLigand simultaneously places probable side-chain amino acid rotamers around the ligand and optimizes the randomly sampled flexible ligand pose using a Metropolis Monte Carlo simulated annealing algorithm. Remaining parameters were set to their default values. 100 independent runs were performed. The predicted binding regions were ranked and evaluated according to their lowest energy score along with structural criteria and the number of occurrences. Ligand docked poses were chosen based on binding energy score together with visual inspection and evaluated using MD simulations. The VMD program was used for visualization to assist an analysis of ligand binding poses in the protein [19]. We combined the top poses from different docking programs to obtain the final poses of the molecules in the binding pocket for further MD simulations.

2.4. Molecular dynamics simulations

The protein structure of ORF3a dimer with PDB ID code 7KJR [7] was used for MD simulations. The detailed steps for studying protein-membrane systems with MD simulations were performed as described in our previous work [31]. Briefly, prior to insertion of the protein into explicit lipid models, we determined the optimal spatial position of the protein in a membrane by calculating the solvation free energy along the z-axis [32]. The continuum electrostatic solvation free energy of protein atoms in the implicit water and lipid solvents was employed using the adaptive Poisson−Boltzmann solver (APBS) software package [33]. The missing hydrogen atoms were added to the protein structure. Atomic charges and radii with PARSE partial charges were assigned using the PDB2PQR tool [34]. The dielectric constants of 2 for the protein and membrane regions (with a thickness of 30 Å for the hydrophobic slab) and 80 for non-membrane region were used in the calculation. The Poisson−Boltzmann electrostatic potentials were solved by sequential focusing multigrid algorithm with three resolution maps, 300 × 300 × 300 Å for coarse, 200 × 200 × 200 Å for medium, and 100 × 100 × 100 Å for fine resolution. The electrostatic solvation energy profile as a function of the protein position along the bilayer normal allows us to estimate the optimal position corresponding to the minimum solvation energy. Subsequently, initial protein–membrane systems for MD simulations were built using the VMD program using TCL command scripts [19]. Ionization state of charged residues (Lys, Arg, Asp and Glu) at pH 7 were assigned to the protein on the basis of pK_a calculations using the PROPKA software [18]. The protein-ligand complex models were inserted into a pre-built and equilibrated lipid bilayer containing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and TIP3P water model [35]. The system was neutralized by adding counter ions and maintaining a salt concentration of 0.1 M NaCl using VMD's Autoionize plugin at 298 K. Periodic boundary condition was employed with a box size of ∼103 × 103 × 125 Å³. A distance cutoff of 12 Å was used for calculating nonbonded interactions. Electrostatic interactions were calculated with particle mesh Ewald summation via fast Fourier transform, including van der Waals interactions, with a switching distance of 10 Å. Langevin dynamics at a constant temperature of 298 K was used with a damping coefficient of 1 ps⁻¹. The pressure was kept constant at 1 atm using the Nose −Hoover Langevin piston method, with a piston period of 200 fs and a damping time of 50 fs. Energy minimization was performed to remove bad contacts between atoms. Restrained MD simulations were employed to relax structural strains of the model systems. In the first stage, restrained MD was conducted with the protein and lipid head group atoms kept fixed to their initial positions. A subsequent run then allowed the whole system (waters, lipids, and counterions) except for the protein to be relaxed. Finally, the equilibration and production runs were performed without any positional restraints. Unless otherwise stated, all MD simulations without restraints were performed with a time step of 2 fs. The CHARMM36 force field parameters [36] were applied for proteins and lipid atoms. The CHARMM generalized force field force parameters (CGenFF) [21] of the ligands were taken from the CHARMM GUI webserver [20]. The MD simulations were performed with the program NAMD version 2.12 [37].

2.5. Ligand flooding simulations

Ligand flooding simulations [38] were performed to help identify potential binding sites for three polyamines inhibitors; SPD, SPM and PUT, in the ORF3a protein. The polyamine molecules were considered to be fully protonated ($-\mathrm{N}{\mathrm{H}}_3^{+}$ or $-\mathrm{N}{\mathrm{H}}_2^{+}-$). For each polyamine, ligand molecules corresponding to 1 mM concentration (calculated with respect the volume of the simulation box), were explicitly added to an equilibrated membrane-embedded protein. Structure topology and force field parameters of the polyamines were generated using the CHARMM GUI webserver [20] and the CHARMM generalized force field (CGenFF) [21], respectively. Before starting the simulation, the polyamines were placed randomly in the aqueous solution. The simulations were carried out using the same condition as describe before. Each system was simulated for 400 ns. We performed three independent 400 ns MD simulations for the ORF3a-SPD system.

2.6. MD trajectory analysis

MD trajectories were used to analyze such quantities as the root-mean-square deviation (RMSD) and the root-mean-square-fluctuation (RMSF) of the Cα atom of the protein, hydrogen bond (H-bond) distance, position of SPD along the z-axis computed from center-of-mass, orientation of ligand binding in the pore region (θ_Ligand), the shape of SPD ligand (N1–N3 distance), number of chloride ions and classification of ligand interaction states using in-house TCL scripts in VMD [19], The RMSDs relative to the initial structure were calculated only on the backbone atoms. The RMSF relative to the average structure was calculated based on the Cα atom of chosen residues. The average structure of protein was generated by the WORDOM software [39]. The number of H-bond (for estimates of population statistics) were based on the distance between the donor and accepter heteroatoms within 3.5 Å while the angular distribution between the donor and accepter heteroatoms were limited by 180°. The percentage of hydrogen bonding between protein and ligand at different docking regions and poses was computed by counting the number of salt-bridge or H-bond forming events divided by the total number of frames.

3. Results

3.1. Candidate binding pockets for ligand

The pocket detection analysis identified a total of twelve binding pockets for ligand molecules in the protein (Fig. 2 ). Based on the identical structure of dimeric ORF3a, these pockets can be grouped into four regions: extracellular portal (ExPort), tunnel region, lateral side region (around β-barrel fold) at the cytoplasmic domain (LatCD), and the tip of the cytoplasmic domain (TipCD). The ligand binding site in the tunnel pocket is buried in the membrane whereas the ExPort, LatCD and TipCD pockets are relatively more exposed to aqueous environment. In addition, two sub-cavities in the tunnel region are detected by the pocket analysis: the upper cavity lies near the inner half of the membrane and the lower one runs underneath the TM1–TM2 loop above the CD. Detection of these sub-cavities in the upper tunnel (UpT) and lower tunnel (LoT) of ORF3a are consistent with the findings reported in previous studies [[6], [7], [8]]. The UpT pocket has highest druggability score (DS) of 0.79 followed by the LoT pocket (0.28) and the extracellular portal (ExPort) pocket (0.27) (Table 1 ). The zero-drug score of LatCD and TipCD pockets suggested that the cavities may not be suitable for binding a drug-like molecule. The UpT pocket has volume of 2979 Å³, hydrophobicity score of 29.9, polar score of 16 and 5 charge score as shown in Table 1. Amino acid residues of the UpT pocket are Q57, S60–K66, A72, S74–F79, I118, R122, I123, R126, P138-L142, N144, Y145 and S205-T208. The volume of LoT pocket is much smaller than that of the UpT pocket and have corresponding scores of 32.2, 7 and 0 for hydrophobicity, polar and charge respectively (Table 1). Amino acid residues in LoT pocket are L65, L71, K75, L140-N144, Y160, N161, G187 and Y189. ExPort and LatCD pocket cavities are exposed to the surface with volume of 250.4 Å³ and 277.8 Å³ respectively. Note that the ExPort pocket is located at the extracellular side of the membrane while the LatCD pocket consists of several small cavities spread out in the β-barrel fold in the cytoplasmic domain.

Fig. 2.

Candidate ligand binding pockets determined by Fpocket. The protein is shown as cartoon representation, and the five pockets namely extracellular portal (ExPort), upper tunnel (UpT), lower tunnel (LoT), lateral region (LatCD) and the tip of cytoplasmic domain (TipCD), are shown as transparent surface representation and colored in green, blue, magenta, orange and red, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Table 1.

Properties of five pockets on ORF3a determined by Fpocket.

Properties	Pocketsa
	ExPort	UpT	LoT	LatCD	TipCD
Druggability Score (DS)	0.27	0.79	0.28	0.0	0.0
Hydrophobicity Score	63.7	29.9	32.2	15.9	−8.8
Polarity Score	1	16	7	6	5
Pocket volume (Å3)	250.4	2979.0	672.6	277.8	407.7
Charge Score	0	5	0	−1	−1

Note ^aExPort, UpT, LoT, LatCD and TipCD are the potential binding pockets located at the extracellular portal, upper tunnel, lower tunnel, lateral region, and tip of cytoplasmic domain, respectively (see text for details).

3.2. Binding poses of ORF3a-polyamine complexes

The five pockets in ORF3a were identified by Fpocket and subsequently used as a guideline for site-specific molecular docking. The predicted binding poses of SPD obtained from different molecular docking approaches are shown in Fig. 3 . Docking hits from different docking programs revealed putative binding regions in the candidate pockets with different binding energy and population. Docking with ADVN showed that the SPD molecule was bound to the LoT pocket with the lowest energy docked poses (Fig. 3A and E). We found that the ligand adopted conformations in such a way that the protonated amine ($\mathrm{N}{\mathrm{H}}_3^{+}$) group of the ligand forms possible H-bond with the carboxylate (COO⁻) group of D142 from ORF3a. In a minority of cases, SPD was bound to the ExPort and LatCD pockets. However, ADVN was unable to find ligand-binding site for SPD in UpT and TipCD pockets. In contrast, the docking results with ADMC and ADGA methods revealed that the ligand can bind to four different pockets including ExPort, LoT, LatCD and TipCD, with similar probabilities (Fig. 3B and C). The SPD adopted multiple poses in which the amine group is an essential requirement for ligand recognition by providing interaction between the protonated amine and acidic (aspartic or glutamic) residues. The docking by ADMC and ADGA showed that the lowest binding energy corresponds to the docked poses in TipCD (Fig. 3E). Analysis of binding poses identified key residues responsible for possible H-bond interactions in the protein-ligand complex, including E102 in ExPort, D142 in LoT, D222 in TipCD, D173 and D183 in LatCD (Fig. 3C). In addition, results from docking with SPM and PUT showed similar ligand binding poses (Fig. S2) and the lowest docking energy in comparison to that of SPD (Fig. 3E). Analysis from three docking methods showed that the lowest docking energy of the best pose depends on the number of amine groups present in the polyamine molecules. As shown by Fig. 3E, the best pose of SPM (tetraamine) located in TipCD has the lowest energy in comparison to that of the SPD (triamine) and the PUT (diamine). Result of RosettaLigand docking is shown in a separate plot in Fig. 3F with different energy scale as it uses different scoring functions and energy scales to evaluate the interactions between the ligand and the protein. It identified polyamine binding poses located only in UpT and LoT pockets with similar lowest energy as that of the best pose (Fig. 3D, S2 and 3F). Polar or charged amino acid residues involving in possible H-bond interactions with the polyamines were Q57, S60, S74, N82 and D142 in the UpT pocket and D142, Y189 and G187 in the LoT pocket (Fig. 3D).

Fig. 3.

(A)–(D) Binding poses of SPD onto ORF3a predicted by (A) ADVN, (B) ADMC, (C) ADGA and (D) RosettaLigand. (E) and (F) Comparison of the docking energy among different candidate pockets. In (F), docking energy obtained from RosettaLigand is shown in a separate plot with different energy scale. The ORF3a structure is shown as cartoon with gray color and the Cα of acidic residues (Glu and Asp) are shown as red beads. The docked poses of SPD are shown as line with standard color coding for atom types. Residues involving with possible H-bond interactions are shown as licorice. A prime (′) denotes the residues of another protein chain. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

3.3. Stability of docked poses by MD

To improve docking results as well as to evaluate the docked poses, we performed MD simulations of the docking complexes. For constructing the simulation systems, top five docked poses of SPD in four binding pockets (ExPort, LatCD, TipCD and LoT) obtained from ADGA and the docked posed of SPD in UpT pocket obtained from RosettaLigand were used as initial configurations. These configurations represent the best docking states with possible hydrogen bonds between the ligand and protein (Table S1). All-atom MD simulations of ORF3a-SPD complexes were performed in an explicit membrane environment. Since H-bond is the important contribution to the binding of the protein and polyamine, estimates of H-bond occurrences between the polyamine and surrounding acidic residues in the binding poses were evaluated. Every pair of H-bond distance between nitrogen atoms of SPD and oxygen atoms of carboxylate group of acidic residues (${\mathrm{d}}_{\mathrm{N}-\mathrm{O}}$) of ORF3a was measured from MD trajectory with H-bond criteria described in the Method section. We have analyzed the frequency with which potential H-bond donor and acceptor heavy atoms are satisfied in the complexes as percentage. The frequency of hydrogen bonds refers to the number of times that hydrogen bonds are formed and broken within a given time frame. A plot of fraction of H-bonds of all five simulated models versus five binding pockets (ExPort, LatCD, TipCD, UpT and LoT) is presented in Fig. 4 A.

Fig. 4.

(A) Percentage of H-bond frequency between ORF3a and SPD from MD simulations in assessing the selected five docking poses in five binding pockets. (B–C) MD snapshots showing H-bond interactions between the amine groups of SPD and negatively charged residues in TipCD and UpT. The protein structure is shown with gray cartoon. Interacting residues and SPD are shown in licorice. A prime (′) denotes the residues on the opposing subunit.

MD simulations of all five ORF3a-SPD complex models from molecular docking revealed a low frequency of H-bonds (<50%) for the protein-polyamine interactions in ExPort and LoT pockets of ORF3a (Fig. 4A). This indicates that the binding affinity for SPD is low at the extracellular portal and the lower tunnel, and SPD does not bind tightly to the residues at these sites which are consistent with a low binding energy predicted by docking simulations. Residues in ExPort and LoT pockets of the protein and the polyamine ligand were not interacting strongly via hydrogen bonding. This may be due to a mismatch in the shapes and lack of sufficient electrostatic interactions that prevents them from forming stable hydrogen bonds. Consequently, ExPort and LoT pockets may not be a specific site for designing polyamine-like inhibitors as drugs. On the other hand, MD simulations of ORF3a and SPD bound in the LatCD, TipCD and UpT pockets showed the H-bond frequency of more than 50% for the complex models as shown in Fig. 4A. The LatCD, TipCD and UpT pockets have multiple acidic and polar residues that are distributed over the cytoplasmic domain of ORF3a (Fig. 3C and D) and can be a good target for inhibitors. They can electrostatically interact with positively charged amine groups on SPD via a salt bridge or H-bond. This electrostatic interaction can help to stabilize the binding of the polyamine to the protein. Particularly, the complex models where SPD was bound in the TipCD and UpT pockets showed a highly stable H-bond with a frequency up to 80%. Fig. 4B and C shows the binding of SPD in the TipCD and UpT pockets, respectively. The SPD interacts with D222 in TipCD which is consistent with the docking results. In UpT, the negatively charged sidechains of D142 and D142' interact with the protonated amine groups of the SPD through salt bridge. By binding to these specific residues (D222, D142) in these pockets (TipCD, UpT), a positively charged inhibitor may disrupt the interaction and therefore prevent the protein's response, a welcome therapeutic effect.

It is important to point out that the best docking poses in LoT and LatCD, as determined by docking algorithms, may not correspond to most relevant biologically or thermodynamically stable configuration in MD simulations. The scoring functions used in most docking algorithms do not fully take into account the complexity of the system, such as solvent effects and conformational flexibility. As a result, the docking pose may not be the global energy minimum, which is the stable state that the system would ultimately converge to. Additionally, the movement and interactions of the atoms in MD can cause the conformation fluctuation, some of which could be different from the initial docking pose. Therefore, it is important to validate docking poses using more rigorous methods such as MD simulations to ensure that they are biologically relevant and thermodynamically stable.

3.4. Ligand flooding simulations

In an attempt to find out the polyamine-binding site of ORF3a, we employed ligand flooding simulations [38] where a number of SPD molecules corresponding to 1 mM concentration [7] was randomly placed in the simulation systems. MD simulations were run for 400 ns to monitor the time evolution of the set of protein-inhibitor interactions. RMSD and RMSF of protein backbone are shown in Fig. S3. Fig. 5 A represents the distribution of SPD bound to the protein from ligand flooding simulations. From MD, SPD is widely distributed around the β-barrel in the cytoplasmic domain in a way that the positively charged nitrogen faces toward the negatively charged residues of the protein to form electrostatic interactions (Fig. 5A). However, these interactions could be relatively weak and non-specific, so SPD can bind to multiple sites on the protein, leading to a wide distribution. We found that the polyamine preferentially binds to the solvent exposed surface (LatCD and TipCD) of the CD rather than interacting deeply with the grooves of CD because the CD surface is more accessible and offers more potential binding sites than the grooves which are buried deeper in the protein structure and have limited accessibility to potential ligands.

Fig. 5.

Density map of interaction frequency of (A) positively charged nitrogen atoms of SPD (blue dots), (B) Na⁺ (orange dots) and (C) Cl⁻ (green dots) to the protein. The acidic (Glu and Asp) and basic residues (Lys and Arg) are shown in red and blue beads, A prime (′) denotes the residues on the opposing subunit. The distance cutoff for interactions is set as 3.5 Å. The interaction frequency was extracted from MD trajectory of ligand flooding simulations. Only an equilibrated MD snapshot of the protein is shown (gray cartoon). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

It should also be noted that the simulations were also unable to generate SPD binding to the UpT and LoT pockets as well as the outer face of the transmembrane domain. This is because of the hydrophobic barrier created by the phospholipid bilayer. In our simulations, SPD is considered as a polar molecule. It is not able to interact with the non-polar tail regions of the phospholipids. SPD cannot easily pass through the hydrophobic (non-polar) phospholipid bilayer and requires a specific mechanism to reach UpT and LoT that are an internal pocket formed by three transmembrane helices. The outer face of transmembrane domain of ORF3a is composed of hydrophobic amino acids, which are not compatible with the charged groups in the polyamine. Flooding simulations with SPM and PUT showed similar characteristics for the density map of polyamine interactions to ORF3a (Fig. S4). Therefore, the tested polyamines (SPD, SPM and PUT) cannot reproduce the necessary interactions with the outer surface of transmembrane domain to bind effectively. It is apparent that there are many basic residues (K61, K66, R68, K75, R122, R126 and K136) near by the LoT pocket. These basic residues can contribute to a highly positive charge in the binding pocket of ORF3a making it unsuitable for the binding of the protonated polyamine. As shown by the density map in Fig. 5A and B, SPD and the counter cations (Na⁺) did not bind closely to a highly positive charged LoT pocket, but they were lined with acidic residues instead. Because of a strong electrostatic repulsion with the positive charge environment in LoT, it is less likely for the polyamine to bind in the LoT pocket. On the other hand, we found that the counter anions (Cl⁻) frequently visited or entered to bind to residues in the LoT pocket (Fig. 5C).

We also noticed that the UpT pocket may be a potential binding site for polyamines because it has less positive charge environment than LoT. There are several polar and a negatively charged residues (Q57, S60, S74, N82 and D142 in Fig. 3D) which can form electrostatic interactions with the positively charged nitrogen atoms of the polyamine, helping to stabilize its binding to the protein. Therefore, we have further carried out MD simulations to address whether the UpT pocket can act as a sort of "trap" for the polyamine, helping to retain it in the binding site. 150 ns MD simulations were performed using the structure of ORF3a-SPD complex obtained from RosettaLigand with three independent runs.

Fig. 6 illustrated characteristics of ligand position, shape and orientation in the UpT pocket of ORF3a. During the course of simulations, SPD was remaining relatively stable in the binding pocket, as revealed by a small difference in position of center-of-mass in the ligand (Fig. 6A and B). The time-evolution profile of the distance between the two nitrogen atoms (N1–N3) of the terminal amine group in the ligand shows small fluctuation (Fig. 6A and C) indicated that the aliphatic polyamine ligand adopted a linear stable structure in the pocket. We next determined the angular distribution of SPD to investigate how the ligand orients with respect to the pore axis. Orientation of the bound ligand in the pore is defined as the angle between the unit vector of the axis connecting the two terminal nitrogen atoms of the ligand and the unit vector normal to the membrane (Fig. 6A). From the angular analysis, the major distribution of the ligand orientation ranges from ∼90 to 110^o degrees (Fig. 6D). This suggested that the bound SPD preferably lies almost perpendicular to the bilayer normal.

Fig. 6.

(A) Definition of the position and orientation of SPD with respect to the z-axis at an initial time of MD simulations. SPD was located in the UpT pocket with measured N1–N3 distance within SPD structure and θ_Ligand relative to the z-axis. Plots of (B) center-of-mass position of SPD, (C) N1–N3 distance and (D) θ_Ligand as a function of simulation time. The inset figure in (D) shows the distribution of θ_Ligand. (E) MD snapshots showing interactions of SPD, D142, Cl⁻ and K61. The protein structure is shown with gray cartoon. Residues and SPD are shown in licorice and chloride ions are shown in green. A prime ′ denotes the residues on the opposing subunit. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Detailed analysis of the interactions indicated that D142 in the UpT pocket of ORF3a has the dominant contact with the ligand by forming a hydrogen bond between the terminal protonated amine group of SPD and the carboxylate group of D142 sidechain (Fig. 6E). In addition, we found that a number (up to 3) of counter chloride ions entered the UpT pocket through the cavity in the LoT pocket and bound to the positively charged nitrogen atoms of SPD and to the protein. These counter anions mediate protein-ligand interactions by forming an ion bridge, thereby keeping the SPD in respective position in the binding pocket. Nevertheless, these chloride ions did not bind permanently to the ligand. We observed repetitive events of binding and dissociation of the anions during the simulations. Upon chloride ion binding, several positively charged residues were found to participate in the anion-mediated protein-ligand contacts. From our MD data, we identified residues involved with the contacts as follows (ranging from high percent to low percent contact frequency): K61 (65.3%), R122 (22.4%), R68 (6.0%), K66 (5.7%), R126 (5.1%) and K75 (1.5%) (Fig. S5). Particularly, the chloride ion bridge between the protonated amine group of SPD and K61 is the most stable one. The observation of interactions between basic residues and chloride ions is also consistent with results of previous simulation [40,41]. The SPD is caught into the cavity of transmembrane domain by lysine or arginine associated with anions. Thus, D142 and K61 appear to be critical residues for stabilizing the SPD at the pocket.

4. Discussion

A variety of computational methods were employed to identify potential binding sites of small polyamines to the accessory protein ORF3a of SARS-CoV-2. The experimental studies showed that putrescine, spermidine and spermine that are composed of multiple amine groups can suppress the activity of ORF3a [7,11]. The experimental observations are complemented by results of our computer simulations where the tip of cytoplasmic domain and the upper tunnel of transmembrane domain of ORF3a appear to be a suitable binding site specific for the polyamines. Particularly, MD simulations revealed the stability of binding poses of spermidine in the upper tunnel pocket of ORF3a through salt bridge and hydrogen bond interactions. We have also evaluated the most likely binding mode of polyamines in the upper tunnel pocket of ORF3a by using MD simulations. The binding of the polyamines to the sites may block the activity of proteins by altering its conformation in an active state or stabilizing its conformation to a non-active state. Our findings can be used to design new drugs or ligands that specifically target the binding site of ORF3a. For example, specific roles of the residues D222 in TipCD and D142 and K61 in UpT could be important for binding, which can help in designing new molecules that mimic the binding interactions of the polyamines. However, it is worth noting that specific experimental validation is needed to confirm the predicted binding. Furthermore, it is also possible that a significant rearrangement of the protein's structure upon ligand binding may occur. Therefore, further studies are needed to fully understand the specific protein-ligand interaction to understand the mechanism of action and the conformational changes that occur.

While the discovery of a novel inhibitor of SARS-CoV-2 is important, the concentrations of these polyamine compounds on ORF3a are of 100 μM to 10 mM [7,11], which are far beyond what is expected for high-affinity potent inhibitors. This can limit its biological relevance on SARS-CoV-2. Nevertheless, it is possible to improve the inhibition efficacy of polyamines by chemical modification. One approach to enhance the inhibition efficacy is to modify their chemical structure to improve their binding affinity and selectivity for the target molecule. For example, polyamine analogues with modifications such as the replacement of the amine group with a heterocyclic ring, or the addition of bulky substituents to the polyamine backbone, have been shown to have improved inhibition activity compared to natural polyamines [42,43]. Another approach is to modify the length and/or charge of the polyamine chain. Shortening the chain length can reduce its size and flexibility, which may decrease its degree of freedom and potentially increase its specificity and binding affinity to the target [44,45]. Moreover, reducing the overall positive charge of the polyamine might enhance its ability to penetrate the cell membrane. In addition, conjugation of polyamines with other molecules, such as lipids or targeting peptides, can improve their specificity and enhance their delivery to the desired target site. However, it's important to note that the effect of such modifications can be complex and depends on a variety of factors, such as the specific target, membrane permeability and surrounding environment. Therefore, it is important to consider each specific case individually and to perform experimental studies to assess the effect on inhibitor efficacy.

Pocket prediction, molecular docking, and MD simulations are useful to predict and analyze the binding of small molecules. Each method has its own strengths and weaknesses. For example, pocket prediction and molecular docking have been shown to be helpful in predicting the binding pose of polyamines to ORF3a, but they do not explicitly take into account the solvent effect and the dynamic nature of protein-ligand interactions. MD simulations have then been employed to further refine the predictions made by molecular docking and to provide information about the flexibility and dynamics of the interactions. We have demonstrated that the accuracy and reliability of polyamine binding sites predicted by molecular docking can be improved by using ligand flooding and MD simulations. These methods generate a set of possible conformations for a ligand molecule to identify the most likely binding region of the protein. Additionally, the stability of a docked complex can be evaluated by MD to identify the most stable binding poses. By using a combination of methods, it is possible to overcome some of the limitations of individual methods and enhance the accuracy of the predictions. Another approach is to use a combination of molecular docking and machine learning-based virtual screening to identify potential binding sites and predict the binding affinities of a ligand with a protein. Free energy calculations can also be used to predict the binding affinities by considering the entropy and enthalpy of the system. It is important to note that even though the predictions made by these computational methods are useful, they should be validated through experimental studies, for example using biochemical, biophysical or structural biology techniques.

5. Conclusion

Open reading frame 3a is an accessory protein that is encoded by the coronavirus 2 (COVID-19). It has been suggested that blocking ORF3a could be a strategy to reduce the severity and prevent the development of severe inflammation in COVID-19 patients [6,46,47]. The protein could provide additional avenues for the development of effective treatments against COVID-19. We employed a combination of various computational techniques i.e. pocket prediction, blind and site-specific molecular docking, molecular dynamics and ligand flooding simulations to identify potential binding pockets for polyamine ligands in ORF3a. Based on the pocket analysis, we found that the predicted binding sites for ligand molecules are located in the extracellular portal, the tunnel, the cytoplasmic β-barrel fold, and the cytoplasmic tip of the protein. The pocket prediction showed that the upper tunnel is a potential binding pocket with highest druggability score compared to other pockets. The binding mode and binding affinity of the polyamine ligands to ORF3a using molecular docking methods identified potential hydrogen bond interactions between the amine groups of the ligands and the acidic or polar residues in the predicted binding pockets of the protein. The docking results indicated that the polyamine molecules are most likely to bind to the cytoplasmic tip and the upper tunnel of the protein based on the lowest binding energy. This finding was supported by MD results of ORF3a-spermidine complex models. Particularly, the simulations revealed the stability of spermidine binding in the upper tunnel pocket of ORF3a through salt bridge and hydrogen bond interactions between the protonated amine groups of the ligand and negatively charged sidechain of aspartic residues of the protein. Based on our study, the upper tunnel of the transmembrane domain of ORF3 could potentially act as a binding pocket for the molecule spermidine. Nevertheless, other factors such as protein conformational changes and solvation effects could also affect the binding of the ligand to the protein. It is important to note that further experimental and computational studies would be needed to validate our hypothesis and the mechanisms of interactions between polyamines and ORF3a.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

Data will be made available on request.