Computer-aided drug design for the pain-like protease (PL^pro) inhibitors against SARS-CoV-2

Abstract

A new coronavirus, known as Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), is a highly contagious virus and has caused a massive worldwide health crisis. While large-scale vaccination efforts are underway, the management of population health, economic impact and asof-yet unknown long-term effects on physical and mental health will be a key challenge for the next decade. The papain-like protease (PL^pro) of SARS-CoV-2 is a promising target for antiviral drugs. This report used pharmacophore-based drug design technology to identify potential compounds as PL^pro inhibitors against SARS-CoV-2. The optimal pharmacophore model was fully validated using different strategies and then was employed to virtually screen out 10 compounds with inhibitory. Molecular docking and non-bonding interactions between the targeted protein PL^pro and compounds showed that UKR1129266 was the best compound. These results provided a theoretical foundation for future studies of PL^pro inhibitors against SARS-CoV-2.

Graphical Abstract

1. Introduction

Over the last two years, the world experienced a viral epidemic of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), which is the third outbreak of coronavirus after Severe Acute Respiratory Syndrome Coronavirus (SARS-CoV) in 2003, and the Middle East Respiratory Syndrome Coronavirus (MERS-CoV) pathogens in 2012 [1], [2], [3], [4], [5]. The disease caused by SARS-CoV-2 was announced as “COVID-19″ by World Health Organization (WHO).

The analysis of SARS-CoV-2 at the molecular level and the elucidation of the pathogenic mechanism of the virus and the life cycle of the virus in the host cells by understanding the pathogenic mechanism of the pathogen, leading to timely and effective preventive and therapeutic countermeasures, are the keys to deal with this outbreak. Whole-genome-based molecular phylogenetic analysis of the above three coronavirus revealed that SARS-CoV-2 has high homology with SARS-CoV and MERS-CoV. Of 7 coronaviruses identified from human, HCoV-229E and HCoV-NL63 are belong toα-coronaviruses. Moreover, HCoV-OC43, MERS-CoV, SARS-CoV, and SARS-CoV-2 belong to β-coronaviruses, positive and single-stranded RNA viruses [5]. Furthermore, their conserved nature was essential in studying CoVs and limiting the virus outbreak.

Both SARS-CoV and SARS-CoV-2 firstly emerged in China. Although the genome-wide similarity was only about 79%, the similarity of the seven conserved domains used for virus identification was as high as 94.6 % [4]. This suggests that SARS-CoV-2 belongs to the same genus as SARS-CoV. For example, the amino acid sequence in spike glycoprotein (S) protein, envelop (E) protein, membeane (M) protein, and nucleocapsid (N) protein of the two CoVs represented about 81–92 % amino acid sequence homology [6]. The mechanism of infection is also the same for both CoVs, with each acting autonomously or synergistically through the various proteins mentioned above to achieve invasion of the host. The virus uses its surface S protein to bind to the host cell surface receptor and then enters the cell by endocytosis. Upon entry into the cell, the virus will immediately release its protein shell and single-stranded RNA encoding its genetic material. The released RNA immediately binds to ribosomes in the host cell and translates functional proteins necessary for its replication. The RNA of the virus synthesizes several vital enzymatic proteins on the ribosome, including RNA polymerase, master protease (M^pro), and papain-like cysteine protease (PL^pro), which are required for SARS-CoV-2 replication and are essential targets for antiviral drug development.

The primary function of PL^pro and M^pro are to process the viral polyprotein in an aynergistic way, which start replicating the virus. It is why so many researchers assumed the PL^pro and M^pro as targets to study targeted-drugs against CoVs. PL^pro performs significant role in the innate immune reponse during viral infection [7], [8], [9]. Besides, it can also strip ubiquitin (Ub) and ISG15 (interferon-induced gene 15) from cellular proteins to help CoVs evade the host innate immune response[10], [11]. Therefore, it can inhibit viral replication and dysregulation of signaling cascades in infected cells by targeting PL^pro with antiviral drugs.

Computer-aided drug design (CADD) has been widely used to predict drug-target interactions and evaluate drug safety to improve clinical efficiency and production efficiency in drug discovery and development [12], [13]. "Polypharmacology" is included in CADD, which means that small molecular compounds with the potential to binding to many proteins, not just a single protein [14]. It focused on designing new therapies for multiple proteins (such as receptors or enzymes) or specific diseases, which is a model therapy for new drug discovery and development [15], [16]. Therefore, the technology of CADD, a new method in drug discovery that focuses on multi-target drugs, has potential applications for drug repurposing, the process of finding new uses for existing approved drugs, prediction of off-target toxicities and rational design of MTDs.

Some small molecular compounds with specific head groups should be designed for the activity site of PL^pro [17]. These types of groups known to inhibit SARS-COV PL^pro include aldehydes, epoxy ketones, activated ketones and activated esters. Wioletta Rut’team from Wroclaw University of Science and Technology has revealed that small molecular compounds VIR250 and VIR251 with inhibitory against SARS-CoV-2 PL^pro [18]. Wu [19] et al. systematically analyzed all the proteins encoded by the SARS-CoV-2 gene and compared them with other coronavirus proteins to predict their structures. By performing target-based virtual ligand screening, a series of anti-viral drugs including ribavirin, valganciclovir, thymidine, chloramphenicol, cefamandole, tigecycline, exhibited high binding affinity to PL^pro, suggesting the potential utility of these compounds in the treatment of SARS-CoV-2.

Nevertheless, studies on targeted therapeutic agents to eradicate SARS-CoV-2 remain. Compounds with accurate and efficiently inhibitory need to be discovered by applying various scientific research methods [20], [21]. To this end, the pharmaophore model, virtual screening and molecular docking method were used to study the mechanism of action and interactions for small drug-molecular and target proteins, finding the compounds with high targeted binding as candidates. As some researchers believe, compounds obtained through virtual screening alone do not necessarily have the potential to develop into drugs. This requires multiple scoring functions to verify the selected compounds and molecular dynamics simulations to verify their stability [22]. Therefore, in this study, the researchers used three molecular docking programs and molecular dynamics simulations to ensure that the selected compounds have the potential to develop into drugs. Our study can provide a theoretical basis for ongoing drug development efforts.

The COVID-19 pandemic continues unabated in many countries and, despite ongoing mass vaccination efforts, remains a significant barrier to economic development, as well as human physical and mental health. To truly overcome the threat posed by the causative coronavirus (CoV), SARS-CoV-2, and its emerging variants of concern, it is paramount to generate and clinically validate additional, orthogonally acting antiviral drugs [23]. We envisage that small molecule drugs that target the viral proteins themselves, acting in concert with vaccination, will stop viral replication in cells and hence impact on virus fitness and transmission. Therefore, the development of therapeutic and prophylactic countermeasures to SARS-CoV-2 is necessary. In our innovative study, we focus on discovering PL^pro inhibitors by analyzing interactions between the natural herbal small molecules and PL^pro using Pharmacophore-based and drug design to obtain the herbal compounds with the inhibitory ability to PL^pro, to provide crucial new insight into the research on new drugs against SARS-CoV-2.

2. Materials and methods

2.1. Preparation of protein receptor

The three-dimensional structure of PL^pro (PDB ID:6WUU) [24] was obtained from the protein database (RCSB)(http://www.pdbus.org) with good resolution (2.79 Ǻ). Discovery Studio 2020 (DS 2020) Client program was used to prepare the structure: crystal water molecules and the ligands in this structrue were removed, after which an updated pdb file was generated and saved.

2.2. Quantitative structure-activity relationship (QSAR) analyses

2.2.1. Data preparation

Many small molecule compounds with sound inhibitory effects on SARS-CoV-2 PL^pro have been reported, and researchers have collected and compiled these compounds. For example, Zhao et al. screened a series of pharmacologically active compounds against SARS-CoV-2 to obtain Sepantronium Bromide (YM155), which was effective in inhibiting PL^pro (YM155:IC50 =2.47 μM) [25]. As mentioned earlier, the homology between SARS-CoV-2 (GenBank ID: MN908947.3) and SARS-CoV (GenBank ID: NC_004718.3) was as high as 83 %, and the similarity of the seven conserved structural domains used for virus identification was 94.6 % [26], [27], [28], [29]. Thanks to the similarity of the two viruses, many inhibitors against PL^pro have also been recently discovered regarding the re-emergence of older drugs. Both Copper Gluconate and Disulfiram have been shown to have inhibitory activity against PL^pro of SARS-CoV. In a recent study, Wang et al. used enzyme assay studies to find that Copper Gluconate and Disulfiram also had very superior inhibitory activity against PL^pro of SARS-CoV-2 (Copper Gluconate:IC50 =0.033 μM; Disulfiram:IC50 = 0.480 μM) [30]. Tanshinone is a compound extracted from Salvia miltiorrhiza, which was identified as an inhibitor of PL^pro of SARS-CoV-1 in 2012 [31]. With the help of previous studies, in recent years, Lim et al. identified a series of compounds that could inhibit PL^pro of SARS-CoV-2, including Dihydrotanshinone I, Tanshinone IIA, and Cryptotanshinone, with Dihydrotanshinone I showing the most potent inhibitory activity (Dihydrotanshinone I:IC50 = 0.586 μM; Tanshinone IIA:IC50 =1.57 μM; Cryptotanshinone:IC50 =1.34 μM) [32]. We collected and screened a series of compounds with inhibitory activity against PL^pro of SARS-CoV-2. All of the compounds have been confirmed the specific inhibitory value (IC50) by experimental means. We divided these compounds into the training set and the test set based on the IC50 and the structures of each compound, the criteria being as follows [33]: (1) compounds should be distributed across different orders of magnitude; (2) compounds in the same order of magnitude should have structurally diverse; (3) The activities of molecules in similar structures should differ by at least an order of magnitude; (4) compounds contained “Activ” and “Uncert” values, with the structures and active values in the training and testing sets being very similar to one another [34].

2.2.2. Data preprocessing

The 2D structures of the training set and test set compounds were drawn and saved as new preparing files. Importing and converting the 3D structures of them into Discovery Studio 2020 (DS 2020), then we need to insert "Activ" and "Uncert" columns in manual one by one because these data are necessary for our subsequent work. We were performing the operation of “Ligand prepare” and “Minimization of Ligands” to prepare the training and test sets. The CHARMm force field is minimized during the Minimize Ligands module. This study used the PDB database to find the inhibitor 3k that currently binds best to PL^pro. It was used as a control for this experiment to find the critical amino acid residue sites.

2.2.3. 3D QSAR pharmacophore generation

3D QSAR (Quantitative Structure-Activity Relationship) Pharmacophore model is an intuitive analysis of the "pharmacodynamic characteristic elements" and their spatial arrangement forms in drug-active molecules. It has been widely applied in the scope of research on chemicals, food, agriculture, and the environment [35], [36]. Besides, QSAR models are also used with increasing frequency to search for active compounds, predict the compound activity and study particular drug targets by Pharmacophore-based virtual screening [37].

Here we constructed a ligand-based Pharmacophore model to quantitatively analyze the pharmacological characteristics of the training set using BOVIA Discovery Studio 2020. The "Feature Mapping tool" was performed to get the information on pharmacophore characteristic elements to prepare for pharmacophore model construction. Then “Common Feature pharmacophore Generation” can be easily selected. Typing the appropriate parameters and runing the job.

2.2.4. Analysis and validation of the pharmacophores

According to the results of the previous step, 10 pharmacophores are generated. Based on the matching degree of the training set and the diversity, the pharmacophores ranked first were selected. What is worth noting is that the first pharmacophore model may not be the best. Hence, it is necessary to conduct a comprehensive analysis of these 10 pharmacophores and perform sufficient verification to selecting the best one.

Validation is an essential step for developing QSAR models that function in a reliable manner [38]. Herein, we validated the selected pharmacophore models using four approaches: (1) Root Mean Square deviation (RMS), correlation coefficient (Correlation coefficient), and cost difference (∆cost) verification were performed; (2) Fischer's randomization test was used for verification; (3) Verifying the activities of the compounds from the test set; (4) The Ligand profiler heat map was used for additional verification.

2.3. Virtual drug database screening

A virtual screening database can efficiently discover possible low-molecular-weight compounds that can bind a given protein within a database of commercially available compounds [39]. Following the verification steps detailed above, the optimal pharmacophore was utilized as a 3D structural query to search the Traditional Chinese Medicine database, Druglike Diverse database [40], and the MiniMaybridge database to identify small molecule compounds with high binding affinity or pharmacodynamic characteristics in order to select better drug candidates. The selected compound molecules were further screened by performing Lipiniski's "rule of five" and "Veber" operations. We set the highest activity value at 1 μM, that is, Activ ≤ 1 μM, to remove the less active (higher activity value) compound molecules and retain only the more active (lower activity value) compounds.

The 3D-RSAR model is a regression model constructed based on the steric and electrostatic fields of small molecules and can be used to predict the activity of unknown ligand small molecules and to observe favorable and unfavorable receptor-ligand interactions. In this study, a partial least squares (PLS) model was constructed using energy grids as descriptors. The energy grid points are calculated by two probes used to measure the electrostatic potential and steric effects. After compound screening, this study used the constructed 3D-QSAR model to predict the inhibitory activity of compounds against PL^pro.

2.4. Molecular docking

Molecular docking analyses can predict the binding affinity of small molecules for specific receptors by using a series of biological, mathematical, and computer-based models, enabling researchers to predict drug affinities for a given binding site and to assess interactions between drugs and targeted enzyme in order better to understand the complex pharmacological systems [41]. Dev et al. [42] evaluated several leading docking programs in a study, namely Glide, DOCK, AutoDock, AutoDock Vina, FRED, and EnzyDock. The researchers assessed the molecular docking ability of these programs by whether these programs correctly identified the binding mode of the Main protease (M^pro) and the ligand, and whether they could accurately and objectively score the binding mode. During SARS-CoV-2's replication and reproduction into host cells, M^pro, like PL^pro, is an essential functional protein for the virus. The results of this study have a certain guiding effect on our selection of docking procedures. In this study, Zev presented the success rate of different docking programs in reproducing the crystal bound poses. In the overall success of all projects, the top three are as follows, Glide and EnzyDock reproduce the correct crystal structure pose (rmsd < 2 Å) for over 50 % of the structures, with success rates of 64 % and 70 %, respectively, while for AutoDock, this rate falls to 40%. After comprehensive analysis, we chose to complete the molecular docking of compounds with PL^pro through CDOCKER, and verified the docking results through AutoDock [43], [44], [45] and Glide [46], [47], [48].

Herein, we employed the CDOCKER molecular docking strategy. CDOCKER is a precise molecular docking method based upon CHARMm that can yield highly precise docking results. The selected compounds were pretreated with "prepare or Filter Ligands" and "Minimization of Ligands", and the processed compounds were directly docked with the target proteins. We used the crystal structure of Mpro obtained in the RCSB Protein database with a resolution of 2.16 Ǻ, pre-processed the target protease through the "Protein Prepare", defined the receptor binding site and prepared the docking system: ①Define Receptor→From Receptor; ②Receptor-Ligand Interactions→Define and Edit Binding Site→show/hide Residues Outside Sphere→show/hide Spher. The pretreated compound and the positive control compound 3k were used as ligand molecules to perform molecular docking operations.

The researchers analyzed CDOCKER's molecular docking results and screened out the most suitable compound. In order to ensure the accuracy of the calculation results, the researchers used AutoDock and Glide docking methods to reconnect the selected optimal compound.

AutoDock is an open-source and free molecular docking software widely used in academic institutions, governments, and commercial institutions. AutoDock uses two methods, "rapid grid-based energy evaluation" and "efficient search of torsional freedom", to achieve a balance between the most accurate calculation possible and reasonable computational resources to predict the interaction between ligands and biological macromolecular targets. Import the selected optimal compounds and PL^pro into AutoDock software to complete the preparation of the world file. The affinity for each atomic type in the docking ligand molecule is budgeted by AutoGrid. Next, the molecular docking between the compound and the acceptor is completed by AutoDock. Finally, AutoDockTools was used to analyze the results of molecular docking.

Glide is a module of software Schrodinger for precise ligand-acceptor docking. Glide can fully consider the influence of lipophilicity, hydrogen bonding, metal ligands, as well as the rotation of unsuitable bonds and the spatial repulsion of atoms, which can effectively reduce false positives and improve the enrichment rate. Its scoring mechanism can reasonably evaluate the results of ligand and receptor docking. After the treated compounds and receptor are introduced into Schrodinger, molecular docking is completed through Glide.

2.5. Molecular dynamic simulation

Molecular dynamics (MD) simulation is a rapid development of molecular simulation methods in recent years. It is based on classical mechanics, quantum mechanics and statistical mechanics, and uses computer numerical methods to solve the equations of motion of molecular systems to simulate and study the structure and properties of molecular systems. This technique can not only obtain the motion trajectory of atoms, but also observe various microscopic details in the process of atomic motion. It is a powerful complement to theoretical calculations and experiments. In this study, the best selected compound were subjected to MD simulations to simulate the interaction between the ligand and PL^pro. MD simulations were performed with AMBER18 [49] using the ff14SB force field. The force field parameters of inhibitors were built by Antechamber [50] module of AMBER18. Three chloride ions were added using the tleap module in AMBER based on a coulomb potential grid in order to keep the whole systems electrically neutral. TIP3P explicit water boxes with an 8.0 Å distance around the solute were added to these complexes. The solvent and ions were subjected to 12,000 steps of steepest decent minimization followed by 8000 steps of conjugate gradient minimization with the protein and small molecules fixed with a 500 kcal/mol Å⁻² constraint. Then each system was totally minimized for another 20,000 steps with no restraint (12,000 steps of steepest decent minimization and 8000 steps of conjugate gradient minimization). After minimization, the three systems were heated up gradually from 0 to 310 K in the NVT ensemble, applying harmonic restraints with a force constant of 10.0 kcal/mol Å⁻² on the protein and small molecules. A Langevin thermostat was adopted. Then these systems went through 500 ps equilibrium MD simulations. Finally, a total of 30 ns was simulated for each system under NPT ensemble conditions with the cut-off at 10 Å. The time step was set to 2 fs. The researchers then conducted Root-mean-square deviation (RMSD) and Root-mean-square fluctuation (RMSF) studies and performed energy calculations.

3. Results and discussion

3.1. Results of 3D QSAR analysis

After the steps of "Data preparation", 3 compounds with poor performance were eliminated, 1 of the compounds belonging to the training set and the others belonging to the test set. 10 Pharmacophore models were generated according to the comprehensive analysis of the training set compounds. A comprehensive analysing of the scores of 10 pharmacophore models is necessary for choosing the best Pharmacophore. Here we have considered the following 6 scoring parameters as the basis for evaluating the pharmacophore: (1) Total cost: For each pharmacophore model, based on calculation, the system will show a Total cost value and Null cost value. Null cost is also known as Fixed cost, meaning the Null cost value of each pharmacophore is equal. Δcost (Null cost - Total cost) is an essential indicator for evaluating a pharmacophore model. The criteria are as follows: Δcost< 40, the confidence interval of the model is reduced to less than 50 %; 40 < Δcost< 60, the confidence interval of the model is 75 %− 90 %; Δcost> 60, it means that the probability that the pharmacophore model is greater than 90 % in a statistical sense reflects the objective situation; The results were shown in Table 1.

Table 1.

10 pharmacophore models generated by PL^pro inhibitors through the HypoGen module.

Hypo	Total cost	Cost difference	Error	RMSb	Correlation	Features
No.
1	184.77	337.533	169.10	1.93	0.91	HBD,HR,HY
2	251.75	270.556	236.36	2.66	0.82	HBD,HR,HY
3	298.93	223.37	280.95	3.05	0.75	HBD,HR,HR
4	311.16	211.139	293.59	3.15	0.73	HBD,HR,HR
5	317.92	204.38	295.40	3.17	0.73	HBA,HBD,HR,HR,HY
6	318.81	203.488	296.83	3.18	0.73	HBA,HBD,HR,HY,HY
7	325.26	197.037	301.86	3.22	0.72	HBA,HR,HR,HY,PI
8	327.39	194.916	304.69	3.24	0.71	HBD,HR,HY,HY,PI
9	333.36	188.946	316.85	3.33	0.69	HBD,HY,HY
10	336.67	185.634	94.67	3.32	0.70	HBD,HY,HY,HY,PI

HBA: hydrogen bond acceptor; HY: hydrophobic.

Cost difference between the null and the total cost, null cost= 469.44, fixed cost= 87.85, for the Hypo6 weight= 1.87, configuration cost= 13.35.

^bRMS: root mean square deviation.

The results of the best pharmacophore model showed 5 types of characteristic elements: hydrogen bond donor (HBD), hydrophobic aromatic (HR), and hydrophobic (HY). As shown in the table, “Total cost”, “cost difference”, “error”, “RMS” and “Correlation” of the first pharmacophore are better than others. Even though “features” of the first pharmacophore are not very good, in considering, we suggest that the phormacophore Hypo1 is the best one perform the next work: QSAR model validation. we after comparing the 6 scoring results.

3.2. QSAR model validation

The best phormacophore which we selected above (Hypo1) needs to be validated by 4 methods: Cost analysis, Fischer’s randomization, Heat map, and Experimental and Estimate activity (IC₅₀ (µM)) evaluation. Only if the validation results match the best one we selected, it can be used to perform all of the next works.

3.2.1. Cost analysis

The cost difference also known as ΔCost, represents the probability correlation of the data. Moreover, the value of the Cost difference depends on the Total cost values. The best pharmacophore Hypo1 is characterized by the total cost value (184.77). The Δcost value (Cost difference 337.533), Error(169.10), RMS(1.93), Correlation(0.91).

3.2.2. Fischer randomization validation

The significance of Hypo1 was further validated by conducting Fischer's randomization test according to the statistical relevance. The cost spreadsheets are automatically generated from the results of the pharmacophore construction, with 19 pharmacophore data randomly generated from cost values. These data are used to verify the best pharmacophore to perform Fischer randomization validation. Here, the confidence level was set by 95%. As shown in Fig. 7, The total cost value for the original hypothesis (the total cost value for the initial 10 pharmacophore models) was substantially higher than the values for the 1–19 randomly generated models. As the cost difference for the original hypothesis was higher than that for these randomly generated pharmacophore models, this confirmed that the Hypo1 hypothesis is the best. The total cost value for the original hypothesis Hypo1 was substantially lower than the values for the 1–19 randomly generated models, which were all superior compared with the initial hypothesis Hypo1, which indicates that none of the randomly generated hypothesis has suitable statistical parameters than Hypo1.

Fig. 7.

The total cost of the initial hypothesis (Hypo1) and 19 random spreadsheets (95% confidence level).

3.2.3. Training and test sets validation

In total, 33 and 30 compounds of PL^pro inhibitors were included in the training set and the testing set, respectively ( Fig. 1, Fig. 2, Fig. 3, Fig. 4, Fig. 5, Fig. 6). Correlation analyses between experimental and estimated activity levels (IC₅₀ values) for these compounds from the training and testing sets, the results showed high correlation coefficients (0.906, 0.924, respectively) ( Fig. 8, Fig. 9). The results are consistent with our hypothesis.

Fig. 1.

The 2D structures of training set compounds (1−12).

Fig. 2.

The 2D structures of training set compounds (13−24).

Fig. 3.

The 2D structures of test set compounds (25−33).

Fig. 4.

The 2D structures of test set compounds (1−10).

Fig. 5.

The 2D structures of test set compounds (11−22).

Fig. 6.

The 2D tructures of test set compounds (23−30).

Fig. 8.

Correlation analysis between experimental and estimated activities (IC₅₀) of Hypo1 on training set compounds.

Fig. 9.

Correlation analysis between experimental and estimated activities (IC₅₀) of Hypo1 on test set compounds.

Besides the training and testing sets were then used to analyze and assess the predictive ability of the pharmacophore Hypo1, with the predicted and experimental values shown in Table 2 and Table 3, respectively. In the tables, high activity (IC₅₀ < 1 μM), moderate activity (1 μM ≤ IC₅₀ < 100 μM), and low or ineffective activity (IC₅₀ ≥100 μM) were respectively represented by +++, ++, and + symbols. Overall, these results revealed that our predictions regarding highly active compounds were correct.

Table 2.

Experimental and Estimate activity (IC₅₀ (µM)) evaluation of the training set compounds based on the pharmacophore model Hypo1.

Compound no.	Fit valuea	Exp.IC50 µM	Estimate	Error	Experimental scaleb	Estimated scaleb
1	6.35	0.46	0.32	-1.42	+++	+++
2	6.33	0.37	0.34	-1.10	+++	+++
3	6.07	0.34	0.62	1.82	+++	+++
4	5.32	0.15	0.67	4.49	+++	+++
5	6.01	0.56	0.70	6.03	+++	+++
6	6.01	0.58	0.71	1.22	+++	+++
7	5.91	0.47	0.89	5.91	+++	+++
8	5.86	0.63	0.99	5.86	+++	+++
9	5.69	0.39	1.47	5.69	+++	++
10	5.59	1.30	1.83	1.41	++	++
11	5.33	2.70	3.37	1.25	++	++
12	5.26	1.90	3.98	5.26	++	++
13	5.24	3.70	4.12	1.11	++	++
14	4.83	13.90	10.55	4.83	++	++
15	4.60	12.70	18.14	1.43	++	++
16	4.51	8.70	22.41	2.58	++	++
17	4.51	59.20	22.41	4.00	++	++
18	4.51	13.50	22.41	1.66	++	++
19	4.51	24.80	22.42	-1.11	++	++
20	4.51	23.50	22.43	-1.05	++	++
21	4.51	11.70	22.45	1.92	++	++
22	4.51	43.20	22.45	-1.92	++	++
23	4.51	27.80	22.50	-1.24	++	++
24	4.50	32.30	22.70	4.50	++	++
25	4.50	13.20	22.84	4.50	++	++
26	4.50	10.40	23.06	2.22	++	++
27	4.48	66.20	23.74	-2.79	++	++
28	4.48	35.80	24.03	-1.49	++	++
29	4.47	20.00	24.29	1.21	++	++
30	4.46	12.10	24.95	2.06	++	++
31	4.45	14.50	25.87	1.78	++	++
32	4.43	46.10	27.06	4.43	++	++
33	4.43	21.10	27.11	1.28	++	++

^a Fit value represents the degree of overlap between the features in Hypo1 and the chemical features in the molecule.

^b Activity scale: IC₅₀＜1 μM = ++ + (highly active); 1 μM ≤ IC₅₀＜100 μM = ++ (moderately active); IC₅₀ ≥ 100 μM = + (low active).

Table 3.

Experimental and Estimate activity (IC₅₀ (µM)) evaluation of the test set compounds based on the pharmacophore model Hypo1.

Compound no.	Fit valuea	Exp.IC50 µM	Estimate	Error	Experimental scaleb	Estimated scaleb
1	6.05	0.35	0.64	1.82	+++	+++
2	6.60	0.11	0.18	1.62	+++	+++
3	6.34	0.39	0.33	-1.18	+++	+++
4	6.20	0.34	0.45	6.20	+++	+++
5	6.08	0.59	0.61	1.03	+++	+++
6	6.03	0.60	0.68	1.13	+++	+++
7	6.00	0.46	0.73	0.46	+++	+++
8	5.73	2.20	1.36	-1.62	++	++
9	5.44	1.90	2.61	1.37	++	++
10	5.31	2.60	3.50	1.35	++	++
11	5.28	5.80	3.76	-1.54	++	++
12	5.25	3.70	4.11	1.11	++	++
13	4.96	4.80	8.00	1.67	++	++
14	4.60	13.50	18.14	1.34	++	++
15	4.51	34.80	22.41	-1.55	++	++
16	4.51	23.50	22.43	-1.05	++	++
17	4.51	22.60	22.41	-1.01	++	++
18	4.51	13.20	22.41	1.70	++	++
19	4.51	11.10	22.41	2.02	++	++
20	4.51	7.30	22.41	3.07	++	++
21	4.50	31.40	22.63	-1.39	++	++
22	4.50	14.40	22.65	1.57	++	++
23	4.50	11.70	22.74	1.94	++	++
24	4.50	7.30	22.87	3.13	++	++
25	4.49	10.10	23.56	2.33	++	++
26	4.48	38.40	24.05	-1.60	++	++
27	4.47	26.00	24.78	4.47	++	++
28	4.45	8.70	25.48	2.93	++	++
29	4.44	19.30	26.15	1.35	++	++
30	4.42	29.10	27.35	-1.06	++	++

^aFit value represents the degree of overlap between the features in Hypo1 and the chemical features in the molecule.

^bActivity scale: +++ highly active (< 1 μM), ++ moderately active (1–100 μM) and + weakly active (> 100 μM).

3.2.4. Heat map validation

The heat map is also displayed as a reference for validation with the results shown in Fig. 10, Fig. 11. The color in the figure represents the degree of matching between the training and testing sets, with better matches being shown in red. Hypo1 is superior to the others. As a comprehensive consideration, Hypo1 was validated to be a reliable pharmacophore.

Fig. 10.

Heat map analysis（Ligand Profiler）of the training set compounds.

Fig. 11.

Heat map analysis（Ligand Profiler）of the test set compounds.

3.3. Virtual screening results

Virtual screening was conducted using 3 databases, including Traditional Chinese Medicine databases, Druglike Diverse database and the MiniMaybridge database. In the results, we gained 1974 compounds from Druglike Diverse database, 657 compounds from MiniMaybridge database, and 1469 compounds from Traditional Chinese Medicine databases database. In the end, 1120 compounds with better activities (active > 1 μM) were retained, while the others with Hypogen-estimated activity lower than 1 μM were eliminated. These 1120 compounds were compiled into a new file, and the “Prepare Ligand” and “Filter by Lipinski and Veber Rules” functions were then performed to support subsequent molecular docking analyses.

3.4. Molecular docking analysis

The compounds above performed “molecular docking” with the targeted PL^pro. The molecular docking results for the top 10 inhibitors ( Fig. 12) were further evaluated using the -CDOCKER_ENERGY and -CDOCKER_INTERATION_ENERGY. ( Table 4). Meanwhile, the inhibitory activity of compounds predicted by 3D-QSAR model is also shown in Fig. 12. The inhibitor 3k, which was used as a control, was also molecularly docked to PL^pro, with -CDOCKER_ENERGY values of 23 for 3k and B values of -CDOCKER_INTERATION_ENERGY.

Fig. 12.

The structure of the top 10 compounds.

Table 4.

Results from molecular docking of PL^pro inhibitors.

Name	-Cdocker_Energy	-Cdocker_Interaction_Energy	Consensus
ASI312247	28.26	31.75	4
UKR1129266	32.27	35.98	2
CDI78166	36.58	41.46	1
CDI815192	30.82	47.40	1
VIT352147	32.32	42.80	1
RJF 01858	37.15	39.78	1
ENA48729	38.11	43.01	0
ENA1233261	37.23	46.91	0
IBS161059	30.17	49.30	0
CAP04845532	36.37	43.69	0

As shown in Table 4, the results of compound ASI312247 were better than those of other compounds: the -CDOCKER_ENERGY (28.26), -CDOCKER_ INTERACTION_ENERGY (31.75), scores for the consensus (4). This is followed by compound UKR1129266, with -CDOCKER_ENERGY values of 32.27, -CDOCKER_ INTERACTION_ENERGY values of 35.98, and the scores for the consensus of 2. The other eight compounds were scored only 1 or 0 for consistency, so we considered the first two compounds the better candidates.

The amino acid binding sites obtained through molecular docking with PL^pro were compared to 3k, as shown in Table 5 .

Table 5.

The interaction amino acid in the ligand-protein for the top 10 docking compounds including the control compound.

compounds	Interaction acids
ASI312247	GLU166, MET208, MET208
UKR1129266	LEU162, ASP164, ARG166, GLY163, SER170, TRY171, MET208, GLN269
CDI78166	ASP164, ARG166, PRO247
CDI815192	LYS157, ASP164, ARG166, LEU199, MET208, PRO247, PRO248
VIT352147	CYS155, ASN156, LYS157, ARG166, GLU167, TYR171, MET206
RJF 01858	LYS157, LEU162, GLY163, ASP164, ARG166, TYR268
ENA48729	ASP164, GLU167, SER170, VAL202, MET206, TYR264
ENA1233261	LYS157, ASP164, MET206, TRY207, MET208, LYS232, TYR268
IBS161059	LEU162, GLY163, ASP164, ARG166, GLU167, MET206
CAP04845532	LEU162, GLY163, ASP164, GLU167, GLU203, MET208, TYR264
3k	SER111, LEU162, ASP164, PRO247, PRO248, TYR268

The top 10 compounds and 3k docked with targeted PL^pro in different bond sites. The 3k was seen as a control to be compared, with coincident amino acid highlighted in bold. The amino acid residues bound to PL^pro by inhibitor 3k are SER111, LEU162, ASP164, PRO247, PRO248, and TYR268. As can be seen from the table, the compound UKR1129266 in our top two compounds with better overall scores was proposed to interact with the target protein PL^pro at significantly more amino acid sites than the other compounds, which is consistent with our original hypothesis. Compound UKR1129266 interacts with the target protein PL^pro at amino acid sites LEU162, ASP164, ARG166, GLY163, SER170, TRY171, MET208 and GLN269. Therefore, compound UKR1129266 can be considered the best candidate compound.

We observed the docking conformations for each compound and assessed the binding modes for the ligand and receptor, with non-bonding interactions between the acceptor and the ligands during docking being represented using different color scales ( Figs. 13–14). It was shown that there are five types of interactions between the compound UKR1129266 and targeting PL^pro, including van der Waals, Conventional Hydrogen Bond, Carbon Hydrogen Bond, Pi-Cation, and Pi-Lone Pair. Van der Waals happened between Cysteine (CYS C:155), Asparagine (ASN C:156), Glutamate (GLU C:161), Leucine (LEU C:162), Arginine (ARG A:166), Glutamate (GLU A:167), (VAL A:202), Glutamate (GLU A:203), Methionine (MET A:206), Tyrosine (TYR A:207), Methionine (MET A:208), Methionine (MET C:208) and Tyrosine (TYR A:268). Conventional Hydrogen Bond happened between Aspartic acid (ASP A:164), Arginine (ARG C:166), Glutamate (GLU A:167) and Serine (SER A:170). Carbon Hydrogen Bond happened between Tyrosine (TYR C:171), Pi-Cation happened between Lysine (LYS C:157) and Pi-Lone Pair happened between Aspartic acid (ASP:164). These results indicated that the compound UKR1129266 exhibited the most favorable interactions with the PL^pro among all analyzed compounds, which means that compound UKR1129266 represents the best candidate compound for further research.

Fig. 13.

2-dimensional analysis of non-bonded interaction between compound UKR1129266 and PL^pro.

Fig. 14.

The docking interactions between compound UKR1129266 and PL^pro.

The accuracy of the above results was verified by molecular docking of compound UKR1129266 and PL^pro via AutoDock, and the results are shown in Fig. 15. The binding _ energy of the compound UKR1129266 and acceptor is − 4.49 kcal and the ligand _ efficiency is − 0.18. The compound UKR1129266 forms hydrogen bonds with the acceptor's Aspartic acid (ASP:164) and Glutamate (GLU:167). The AutoDock results are similar to CDOCKER, and both demonstrate that UKR1129266 can stably bind to PL^pro receptors.

Fig. 15.

The result of molecular docking of compound UKR1129266 and PL^pro by AutoDock.

The molecular docking results obtained by Glide also support our hypothesis that the compound UKR1129266 can stably bind to the acceptor. The docking results are shown in Fig. 16. The docking score is − 5.267, and Glide's scoring criterion is that the lower the score, the more stable the ligand and receptor binding, so the compound in this study is very good with PL^pro.

Fig. 16.

The result of molecular docking of compound UKR1129266 and PL^pro by Glide.

In this study, CDOCKER was first used for molecular docking to screen out the best compound UKR1129266. The researchers then used AutoDock and Glide to verify the binding stability of the compound UKR1129266 to PL^pro. All three docking methods clearly show that the UKR1129266 screened by the compound can be stably bound to PL^pro.

3.5. Analysis of molecular dynamics results

The best selected compound, UKR1129266, was subjected to molecular dynamics simulations. RMSD is a detection method which can provide a sketch of the conformational changes by comparing changes in the positions of the atoms with a reference structure. In Fig. 17, it can be seen that the fluctuation range of the compound UKR1129266 is within 2 Å, the fluctuation amplitude is weak, and the linear relationship tends to converge, which indicates that the complex remains stable throughout the simulation time. RMSF is a curve which can offer details on fluctuations of each residue over the simulation time. A high RMSF value represents that the certain residue has a large flexibility, while a low one manifests large stability. The RMSF values are displayed in Fig. 18. The residues with a high value were checked, only to find that majority of these residues locate on the edge of the complex and they are far away from the inhibitor binding pocket. The residues by ligand to PL^pro are LEU162, ASP164, ARG166, GLY163, SER170, TRY171, which have very low RMSF values and strong stability. Combined with the calculation of binding free energy, Binding Energy is − 12.6 kcal/mol, Entropic Energy is 20.0 kcal/mol. The above data indicate that the compound UKR1129266 can bind stably to PL^pro. Compound UKR1129266 has great potential to be developed as a potent inhibitor.

Fig. 17.

The RMSD results of compound UKR1129266.

Fig. 18.

The RMSF results of compound UKR1129266.

3.6. Discussion

Many recent experimental studies on PL^pro inhibitors in vivo and in vitro have been reported. Inhibitor development campaigns against SARS-CoV PL^pro have resulted in two primary chemical scaffolds the benzamide ring (“GRL-0617″ family of compounds) and the piperidine carboxamide (“5c” family of compounds) series [51], [52]. Many subsequent PL^pro inhibitors were modified from GRL-0617 and 5c. Indeed, researchers show that commonly used PL^pro inhibitors suffer from a multitude of liabilities, mostly due to the presence of a naphthyl moiety that is present in both 5c but also GRL-0617 compounds. Fortunately, the best compound UKR1129266 screened in this study does not have a naphthyl moiety in its structure. This effectively solves the problem of inhibitor shortcomings due to the naphthyl moiety and expands the PL^pro inhibitor structure type, and provides new ideas for future structural modification work.

Klemm et al. [27] adapted a ubiquitin-Rhodamine110-based high throughput screening (HTS) assay to identify small molecule PL^pro inhibitors as previously developed for human DUBs [53]. A first drug repurposing campaign was performed, in the hope of uncovering human-safe medications that could be progressed towards the clinic. Researchers ideally required nanomolar inhibitory activity, a “clean” specificity profile against human DUBs and sensible chemistry lacking reactive groups or PAINS. However, screening 5576 molecules including 3727 unique FDA approved small molecule drugs, and researchers failed to identify suitable compounds that would enable progression to the clinic. Janes et al.[54] extended these studies to include the ReFRAME library, which is a collection of 11,804 compounds, mostly approved drugs and drug candidates that had progressed to late-stage clinical trials, and hence had in-human safety data associated. Two compounds were finally screened, which are XL-999, a receptor tyrosine kinase and FLT3 kinase inhibitor, and a derivative of codeine, an opioid receptor agonist. Both compounds displayed weak in vitro inhibitory activity against PL^pro and had been optimized for their human targets (XL-999:IC50 =48 μM; FLT3:IC50 =51 μM). Weak activity against PL^pro (necessitating extreme dosing regimes) rendered both compounds unsuitable for progression toward the clinic. Compounds were also considered unsuitable as starting points for medicinal chemistry due to inferior potency and ligand efficiency compared to other scaffolds. In another study, the researchers screened the best compound from the database for Tarloxotinib, but the inhibitory activity of Tarloxotinib in subsequent activity assays still fell short of human expectations [55]. From many studies, it can be found that the discovery process of lead compounds of target PL^pro inhibitors is not very smooth, and no satisfactory compounds have been obtained. This study used a ligand-based pharmacophore model to virtually screen suitable PL^pro inhibitor lead compounds. In this study, a pharmacophore model was constructed based on 63 inhibitors with known inhibitory activities, which avoided the loss of screening results due to the significant differences in the structure of pharmacophore models with too few common features or too many combined features. Subsequently, the generated pharmacophore model was screened and verified by four validation methods: Cost difference(ΔCost), Fischer randomization validation, training and test sets validation, and heat map validation. The validation results were consistent with the original hypothesis of the researchers. The best pharmacophore model was used to screen several databases in turn, and the obtained molecules were tested for drug-like properties and finally docked with the target proteases. The binding sites and docking interactions of the compounds with the target proteases were analyzed. The results of each stage of the study have been fully verified to ensure the rigor and scientific nature of the experiment. The final compound UKR1129266 showed excellent binding activity with PL^pro, which was structurally innovative and had ideal inhibitory activity compared with the compounds selected in previous studies.

Another problem faced by virtual screening is the evaluation of drug molecular activity. The activity of many drugs and other biomolecules is manifested through interactions with receptor macromolecules, so the evaluation of binding free energy between receptors and ligands is a core issue in computer-aided drug molecular design. There are many methods to calculate the binding free energy of protein–ligand and protein–protein systems, such as free energy perturbation (FEP), thermodynamic integration (TI), and the molecular mechanics/Poisson–Boltzmann (generalized Born) surface area (MM/PB(GB)SA)) method. Among these, FEP and TI are considered as the most rigorous approaches, but they are also time-consuming as they require a large number of statistical samples and calculations. In contrast, the MM/PB(GB)SA method is used widely for the binding free energy calculations, due to its computational efficiency. Many studies have found that the relative binding free energy calculated by the MM/PB(GB)SA method can be used to efficiently rank a series of small molecules, and this method also has better performance in distinguishing the native structure from a large number of decoys than scoring functions [56], [57], [58], [59]. In one study, in order to improve the performance of the MM/PBSA and MM/GBSA methods in correctly distinguishing the native structure from the decoy structures, Zhong [56] adopted six methods using the MM/PBSA and two MM/GBSA (GBHCT and GBOBC1) models were combined with the IE method and used to calculate the binding free energy of 176 protein ligands and protein-protein structures in the Bcl-2 family. In this study, the binding free energy of the compound UKR1129266 and PL^pro docking system was calculated with reference to Zhong's method, which can more intuitively reflect the quality of the inhibitory activity of the compound.

4. Conclusion

In summary, SARS-CoV-2 remains a significant, constantly evolving threat to global public health that cannot be easily eradicated. Herein, we conducted an analysis of the molecular structure of the SARS-CoV-2 PL^pro, identifying this essential protein and the binding sites.

Many obstacles will be encountered in the development of PL^pro inhibitors. Nonetheless, the considerable efforts applied to inhibiting PL^pro and full structural enablement have significantly advanced our understanding to remain a viable drug target for the treatment of COVID-19. Once thriving, PL^pro inhibitors will have similar or even more potent anti-CoV activity, as observed with the recently approved M^pro inhibitors. Indeed, in addition to blocking the essential protein processing steps in the viral replication cycle, inhibiting PL^pro may also serve additional purposes. As a DUB and deISGylase, PL^pro prevents virus induced derailing of the cellular inflammatory and antiviral cascades affected by PL^pro mediated cleavage of ubiquitin and ISG15. It may at least partially be responsible for the observed inflammatory flares reported in COVID-19 patients. Therefore, we consider PL^pro as the ultimate drug target for treating Coronaviruses, which, although challenging, is likely to provide significant protection against future pandemics.

The CDOCKER_ENERGY score, the amino acid residues binding to PL^pro and the type of interaction were analyzed by molecular docking, and compound UKR1129266 was screened to have the best binding effect. Therefore, UKR1129266 was the most promising compound for targeted drug design. The results provided important new insight into the research on drugs against SARS-CoV-2.

CRediT authorship contribution statement

Hongwei Gao: Conceptualization, Validation, Resources, Supervision, Project administration. Renhui Dai: Formal analysis, Investigation, Software, Methodology, Writing – Original Draft, Writing – Review & Editing, Visualization. Ruiling Su: Data Curation.

Conflict of interests

All authors declare that No conflict of interest exists.