Identification of Aloe-derived natural products as prospective lead scaffolds for SARS-CoV-2 main protease (M^pro) inhibitors

Graphical abstract

Abstract

In the past two years, the COVID-19 pandemic has caused over 5 million deaths and 250 million infections worldwide. Despite successful vaccination efforts and emergency approval of small molecule therapies, a diverse range of antivirals is still needed to combat the inevitable resistance that will arise from new SARS-CoV-2 variants. The main protease of SARS-CoV-2 (M^pro) is an attractive drug target due to the clinical success of protease inhibitors against other viruses, such as HIV and HCV. However, in order to combat resistance, various chemical scaffolds need to be identified that have the potential to be developed into potent inhibitors. To this end, we screened a high-content protease inhibitor library against M^proin vitro, in order to identify structurally diverse compounds that could be further developed into antiviral leads. Our high-content screening efforts retrieved 27 hits each with > 50% inhibition in our M^pro FRET assay. Of these, four of the top inhibitor compounds were chosen for follow-up due to their potency and drugability (Lipinski’s rules of five criteria): anacardic acid, aloesin, aloeresin D, and TCID. Further analysis via dose response curves revealed IC₅₀ values of 6.8 μM, 38.9 μM, 125.3 μM, and 138.0 μM for each compound, respectively. Molecular docking studies demonstrated that the four inhibitors bound at the catalytic active site of M^pro with varying binding energies (-7.5 to −5.6 kcal/mol). Furthermore, M^pro FRET assay kinetic studies demonstrated that M^pro catalysis is better represented by a sigmoidal Hill model than the standard Michaelis-Menten hyperbola, indicating substantial cooperativity of the active enzyme dimer. This result suggests that the dimerization interface could be an attractive target for allosteric inhibitors. In conclusion, we identified two closely-related natural product compounds from the Aloe plant (aloesin and aloeresin D) that may serve as novel scaffolds for M^pro inhibitor design and additionally confirmed the strongly cooperative kinetics of M^pro proteolysis. These results further advance our knowledge of structure–function relationships in M^pro and offer new molecular scaffolds for inhibitor design.

The global scourge of the COVID-19 pandemic has enveloped the world for two years, leaving over 5 million dead and 250 million infected in its wake.1, 2 Despite the emerging success of vaccines, a significant number of people remain unvaccinated, either due to lack of availability³ or vaccine hesitancy.4, 5 Furthermore, “breakthrough” infections are becoming more common in the vaccinated with the identification of new variants,6, 7 leaving few therapeutic options available for life threatening infections.8, 9 This illustrates the need to develop novel small molecule chemical modalities that can be used both to treat severe SARS-CoV-2 infections and as potential prophylactic agents in high-risk population groups. While the earliest small molecule therapies targeting SARS-CoV-2 directed their activities towards the RNA polymerase target (e.g., remdesivir and molnupiravir), recent attention has turned to the 3C-like viral main protease (3CL^pro or M^pro) as a possible drug target.10, 11, 12, 13, 14, 15, 16, 17 Viral protease inhibitors have previously demonstrated high clinical efficacy against other viruses, such as HIV (human immunodeficiency virus)18, 19 and HCV (hepatitis C virus),20, 21, 22 making M^pro an attractive target for antiviral drug development. Indeed, Pfizer, Inc. has a first-in-class M^pro inhibitor currently in Phase III clinical trials,¹¹ pointing to M^pro inhibition as an effective therapeutic strategy. Despite this, a major liability of viral protease inhibitors is the eventual development of resistance, which necessitates their use as part of combination therapy and also the development of second and third generation compounds. This highlights the need for the identification of new lead structures that may function as M^pro inhibitor scaffolds.12, 14, 15 Since the publication of the SARS-CoV-2 M^pro crystal structure,²³ a number of campaigns have been conducted to identify selective M^pro inhibitors11, 12, 14, 15, 16, 23, 24, 25, 26, 27, 28, 29 While most of these published accounts have exclusively involved computational approaches,16, 25, 26, 30 several have screened existing small molecule libraries for M^pro inhibition in vitro.13, 15, 31 However, these approaches have led to a number of redundant hits with similar mechanisms of action.32, 33, 34 In order to forestall eventual viral resistance, chemical diversity is an important consideration in identification of early protease inhibitor leads.35, 36, 37, 38, 39 Even small chemical libraries that are highly diverse can lead to the identification of inhibitors with unique mechanisms of action.40, 41, 42 In an attempt to expand the chemical diversity of potential scaffolds for SARS-CoV-2 M^pro inhibitor lead design, our approach focused on a high-content protease inhibitor library to filter out redundant molecules and obtain unique pharmacophores that could be further exploited by chemical optimization. Here, we report the identification of novel Aloe-derived natural product scaffolds that have the potential to be further optimized into effective clinical candidates, given their high drug-like qualities (Lipinski's rule of five 43, 44) and safety profiles. Additionally, our data confirm that SARS-CoV-2 M^pro functions as a cooperative dimer, hinting at the possibility of developing allosteric inhibitors that may target and disrupt the enzyme dimer interface.

In order to screen library compounds for inhibitory activity against M^pro in an in vitro system, we cloned, expressed, and purified the M^pro viral protein in an E. coli based heterologous expression system.⁴⁵ To facilitate production of the native protease, we retained the original viral N-terminal autocatalytic cleavage site (SAVLQ↓SGFRK) and modified the C-terminal sequence with the core amino acids of the HRV (human rhinovirus)–3C protease cleavage site (VTFQ↓GP) to permit removal of the His tag needed for the protein purification (see supporting information, Figure S1). This construct resulted in high yields of the native protein (see supporting information, Figure S2). To ascertain proteolytic activity, we relied on a FRET (fluorescence resonance energy transfer)-based assay with the peptide substrate Dabcyl-KTSAVLQ↓SGFRKME-Edans-NH₂, which emits a fluorescent signal at 460 nm when cleaved and excited at 360 nm.⁴⁶ Once we evaluated our M^pro activity assay for linearity as a function of time and protein concentration (see supporting information, Figure S3), a kinetic experiment was performed at 100 nM M^pro, 2–128 μM FRET substrate (5% final volume dimethyl sulfoxide, DMSO) in 100 μL final reaction volume with reaction buffer comprised of 20 mM Tris HCl, 100 mM NaCl, 1 mM EDTA, 1 mM DTT, pH 7.3 at 37 °C. The reaction was initiated by adding 50 μL of the FRET substrate in reaction buffer to 50 μL of M^pro in reaction buffer. Cleavage of the substrate was measured via fluorescence on a Tecan Infinite M Plex plate reader every minute for 30 min. A free Edans calibration curve from 0.1 to 25 μM was used to convert the initial velocities in RFU (relative fluorescence units) to pmol/s. Concentration of FRET substrate vs. initial velocity was plotted (Fig. 1 ) and analyzed for kinetic parameters via GraphPad Prism (v. 9.2.0.332). Both Michaelis-Menten and allosteric sigmoidal models (Hill equation) were assessed (Fig. 1A) and compared via the second order Akaike Information Criterion (AICc).⁴⁷ The resulting kinetic parameters for the Michaelis-Menten fit were K_m = 44.6 ± 8.0 μM and V_max = 359.6 ± 27.8 pmol/S/Nmol M^pro, and S₅₀ = 24.2 ± 2.0 μM and V_max = 269.6 ± 11.4 pmol/S/Nmol M^pro for the allosteric sigmoidal fit (see supporting information, Table S1).

Fig. 1.

Kinetic analysis of FRET substrate proteolysis by M^pro. A: Kinetic data fit to Michaelis-Menten and allosteric sigmoidal (Hill) models. B: Eadie-Hofstee transformation of Hill equation at h = 1 (no cooperativity) and h = 1.7 (positive cooperativity). The kinetic assay was performed at 2–128 μM FRET substrate and 100 nM M^pro. Initial velocities in RFU converted to pmol/s with the free Edans calibration curve. Points represent mean of triplicate measurements ± standard deviation.

Interestingly, the data best fit to the sigmoidal Hill plot with an R² of 0.985 and an AICc correct fit probability of 99.98%. This is dramatically apparent when the data is analyzed by the Eadie-Hofstee transformation (Fig. 1B). The presence of cooperativity in enzyme activity is readily explainable and somewhat expected as the enzyme is reported to be only active as a dimer.48, 49, 50 However, while allosteric cooperativity has been demonstrated previously for this enzyme,51, 52, 53 this is the first time that kinetic data have been quantitatively compared to determine the best model for enzyme activity, i.e. either the Michaelis-Menten or the sigmoidal Hill equations (Fig. 1A, Table S1). The clear appearance of allosteric kinetics, as represented in the Eadie-Hofstee plot (Fig. 1B), indicates that the allosteric dimer interface may also be a promising inhibitor target, as was found to be the case for the SARS-CoV-1 M^pro enzyme.⁵⁰

For our high-content library screening, a total of 236 compounds were obtained from different vendors: Protease Inhibitor Library (catalog no. L2500) and PF-00835231 from Selleck Chemicals (Houston, TX); ebselen from Sigma-Aldrich (St. Louis, MO); aloin B, aloe-emodin, calpain inhibitor II, calpain inhibitor III, calpain inhibitor VI, calpain inhibitor XII, 2-cyano-pyrimidine, E-64d, oseltamivir (phosphate), PD 150606, PSI-7977, tosyllysine chloromethyl ketone HCl, and pimodivir from Cayman Chemical (Ann Arbor, MI); aloeresin D from eNovation Chemicals (Green Brook, NJ); and 7-O-methylaloeresin A from Muse Chem (Fairfield, NJ). The inhibition screening assays were performed at 50 nM M^pro, 10 μM FRET substrate, and 100 μM inhibitor (1.5% final volume DMSO). 100 μM inhibitor concentration was selected as our goal was not to identify highly potent compounds, but rather chemically diverse leads. The Pfizer M^pro inhibitor, PF–00835231, was used as a positive control for inhibition at a concentration of 1 μM. The reactions were initialized at room temperature by the addition of 50 μL FRET substrate and inhibitor mixture, dissolved in reaction buffer, to 50 μL of M^pro also prepared in reaction buffer. The reaction buffer consisted of 20 mM Tris HCl, 100 mM NaCl, 1 mM EDTA, and 1 mM DTT, pH 7.3 at room temperature. Fluorescence was monitored for 1 h, with readings taken every 2.5 min. The first 20 min were used to calculate the initial velocity, and percent inhibition was calculated from the slope compared to the DMSO control. Background from the FRET substrate was subtracted from all samples. All experiments were performed in triplicate and the presented values represent the mean ± standard deviation. Positive “hits” were defined as any compound that inhibited M^pro activity by 50% or more. As can be seen in Fig. 2 , 27 compounds fell into this category, indicating an assay hit rate of approximately 11.4%. The structures and percent inhibition are shown in Table 1 , demonstrating the chemical diversity present in the hits obtained from the initial library screening. To assess assay robustness, the Z’-factor was calculated for each screening experiment (see supporting information, Figure S4). The resulting Z’-factors were between 0.727 and 0.969, well within the guidelines of 0.5 to 1 for a high-throughput assay.⁵⁴

Fig. 2.

High-content library screening for M^pro inhibition. A total of 236 protease inhibitors were screened at 100 μM against M^pro. Hits are defined as compounds with over 50% inhibition, represented by black circles. Compounds with less than 50% inhibition are represented by gray squares. Percent inhibition was calculated against DMSO control. Data points represent the mean of three replicates ± standard deviation (several standard deviations are too small to be discernable).

Table 1.

M^pro inhibitors identified from high content protease inhibitor library screening.

Compound	Structure	Inhibition (%)	Compound	Structure	Inhibition (%)	Compound	Structure	Inhibition (%)
Z-DEVD-FMK		100.0 ± 0.7	Z-FA-FMK		90.1 ± 0.8	Leupeptin hemisulfate		59.0 ± 1.8
Calpain Inhibitor II		99.8 ± 0.7	Telaprevir		88.8 ± 0.5	Celastrol		57.7 ± 1.3
MG-132		99.0 ± 1.7	Aloesin a (aloeresin B)		80.1 ± 2.9	Nelfinavir mesylate		56.1 ± 2.2
Calpain Inhibitor XII		98.3 ± 0.3	Aloeresin D a		72.6 ± 1.1	IU1		54.8 ± 0.1
Boceprevir		97.4 ± 0.2	PD 151,746		71.5 ± 1.4	Tanshinone IIA a		53.9 ± 3.3
Calpeptin		97.2 ± 0.1	Mocetinostat a		66.7 ± 2.1	Dibenzazepine		53.8 ± 2.0
Anacardic acid		96.1 ± 0.1	L-685,458		65.0 ± 7.1	Glecaprevir a		51.7 ± 4.9
Calpain Inhibitor I (MG-101)		95.2 ± 0.9	LDN-57444		61.4 ± 1.0	MC1568		51.4 ± 1.3
TCID a		94.2 ± 2.5	7-O-Methyl- aloeresin A		60.5 ± 1.7	Asunaprevir a		49.4 ± 4.2

Percent inhibition values represent mean of three replicates ± standard deviation. ^a Compound with background fluorescence.

Inherent limitations of the FRET-based assay led to 19 of the initial library compounds (8.1%) being eliminated from the assay screen due to either intrinsic fluorescence from the compounds themselves or issues stemming from inner filter effects. Some of the compounds we identified have been previously reported as SARS-CoV-2 M^pro inhibitors (see supporting information, Table S2), thereby increasing our confidence in the robustness of our M^pro screening assay. Interestingly, a number of the most potent compounds are natural products (e.g., anacardic acid, aloesin, aloeresin D).

Aloesin and aloeresin D come from the Aloe plant, including Aloe perryi and A. barbadensis (aka, A. vera) species.55, 56 These compounds have a long history of medicinal use as stimulant-laxatives to treat constipation,⁵⁷ anti-inflammatory compounds to promote wound healing,⁵⁸ and bittering agents in food supplements, thereby demonstrating their positive safety profile and bioavailability in humans.⁵⁶ Furthermore, aloesin is a known tyrosinase inhibitor with an IC₅₀ of 0.9 mM against mushroom tyrosinase.59, 60 Anacardic acid is a natural product from the cashew nut that exhibits anti-inflammatory and antinociceptive properties.⁶¹Due to the chemical novelty of the natural products identified and the availability of readily accessible functional groups on these compounds for further modification, we decided to focus on the top three natural products with high percent inhibition in the 100 μM screen: anacardic acid (96.1% ± 0.1), aloesin (80.1% ± 2.9), and aloeresin D (72.6% ± 1.1).

We also chose to include the non-natural product TCID (a selective ubiquitin C-terminal hydrolase-L3 inhibitor), due to its performance in the high-content screen (94.2% ± 2.5).⁶² In order to quantitatively assess the extent of M^pro inhibition, we performed IC₅₀ experiments using the screening assay that we had already developed. The dose response curves in Fig. 3 demonstrate IC₅₀′s in the micromolar range. It should be noted that 0% inhibition (100% activity) was not reached in all cases due to assay background fluorescence.

Fig. 3.

Dose response curves of anacardic acid (A), aloesin (B), aloeresin D (C), and TCID (D) against M^pro. Percent inhibition calculated against the DMSO control and analyzed in GraphPad Prism via non-linear regression with a dose–response inhibition model (three parameters). Points represent mean of triplicates ± standard deviation (several standard deviations are too small to be discernable).

TCID (IC₅₀ = 138.0 μM ± 12.9), was ruled out as a possible scaffold due to its higher IC₅₀ value and potential toxicity.⁶³ While the most potent inhibitory compound was anacardic acid (IC₅₀ = 6.8 μM ± 1.0), its toxic liabilities may also prevent it from further development as an antiviral agent, whereas the two Aloe compounds (aloesin IC₅₀ = 38.9 μM ± 8.6, aloeresin D IC₅₀ = 125.3 μM ± 24.5), have a more favorable safety and efficacy profile in humans.⁵⁶ The four hits were also pre-incubated with and without the enzyme in order to assess time-dependent inhibition (TDI) (see supporting information, Figure S5). While a single inhibitor concentration is not conclusive, the data suggest that the compounds exhibit a moderate amount of TDI. Experiments involving multiple concentrations points will need to be carried out to further delineate the TDI potential of these compounds.

Table S3 (see supporting information) illustrates Lipinski’s rule of five for the four inhibitors: aloesin and TCID meet all four criteria, while anacardic acid and aloeresin D have one and two violations respectively. Compounds meeting at least three of the four of Lipinski’s criteria generally make excellent orally active drugs.43, 44

In an effort to expand our knowledge as to the specific mode of action of these compounds, a molecular docking study was undertaken using the AutoDock Vina algorithm, v. 1.1.2.⁶⁴ Briefly, the SARS-CoV-2 M^pro crystal structure (PDB entry 6Y2E) was prepared for docking by removing ions and water molecules and adding polar hydrogens using MGL AutoDock Tools v. 1.5.7 (UCSD Molecular Graphics Lab and The Scripps Research Institute). Inhibitor (PF-00835231, anacardic acid, aloesin, aloeresin D, and TCID) structural coordinates were obtained from the Protein Data Bank (PDB: https://www.rcsb.org/) and parameterized by adding polar hydrogens and identifying rotatable bonds. The receptor docking grid was defined by the following parameters - grid box center: x-center = -16.022, y-center = –32.73, z-center = 4.648, and the total number of grid points in each dimension being: x-dimension = 80, y-dimension = 48, and z-dimension = 62. To facilitate an efficient docking routine, a configuration file docking script was prepared using MS Notepad in simple text format with the energy range set to 4 and the exhaustiveness search parameter set to 24. AutoDock Vina was executed using the configuration file with both PDBQT.out and log.out file options selected. The PDBQT.out file contains all of the ligand binding poses for any particular docking simulation. Output files were analyzed using the ViewDock function of UCSF Chimera v.1.15, and ranked according to binding energy (ΔG). To test the validity of AutoDock Vina to predict the correct binding pose for our selected inhibitors, we first attempted to dock PF-00835231 using the parameters described above. The reported IC₅₀ of Pfizer’s reference inhibitor is in the low nanomolar range (6.9 nM). Here, we found that the lower IC₅₀ value correlated with the lower docking score, both indicating higher affinity for the enzyme. The most energetically stable pose found the compound bound in the active site within close proximity of the catalytic dyad (Cys 145 and His 41) (see supporting information, Figure S6). This docking pose is strikingly similar, although not quite superimposable, with the location of the compound found in PDB entry 6HXM, where it is covalently adducted to Cys 145. The difference in position is likely due to the covalent adduct formed between PF00835231 and Cys 145 during catalysis.

As seen in Fig. 4 , all four inhibitors docked within the active site of M^pro, near the catalytic dyad of Cys 145 and His 41, although with differing binding energies. Aloesin bound with the lowest predicted free energy, −7.5 kcal/mol, followed by aloeresin D at −6.8 kcal/mol, TCID at −5.8 kcal/mol, and anacardic acid at −5.6 kcal/mol. TCID, the smallest molecule, docked near the Cys 145 at subsite S1 (Fig. 4D). Both aloesin and anacardic acid are bound in extended conformations, occupying the entirety of the active site, including subsites S1, S1′, S2, and S4 (Fig. 4 A, B). In contrast, aloeresin D (Fig. 4C) is bound in the active site in a more sterically constrained conformation, with the vinylphenol ring hovering just above the glycosyl moiety, only occupying subsites S1 and S4. This correlates well with the IC₅₀ potencies, where aloeresin D was approximately 3-fold less potent than aloesin (Fig. 3), and may indicate that the vinylphenol group of aloeresin D sterically hinders binding in the active site and in fact decreases the affinity of the ligand for M^pro. Removal and/or substitution of this group with smaller functional groups could result in more potent inhibitors that fit more snugly into the active site. In addition, removal of the vinylphenol ring of aloeresin D would likely change its Lipinski’s rules of five parameters, decreasing the number of violations and improving its drugability. We also attempted to dock vitamin K, a natural product with structural similarity to anacardic acid, but did not obtain a binding pose that met the search criteria, likely due to the unsaturated, rigid hydrocarbon tail present in the vitamin K structure. While the Aloe compounds are not extremely potent inhibitors on their own, these results illustrate their potential as molecular scaffolds that can be further developed into more efficacious M^pro inhibitors. Furthermore, the excellent safety profile and bioavailability of both aloesin and aloeresin D make it likely that minor structural modifications to improve affinity to M^pro will not result in producing significant toxic liabilities. Given the limited number of validated chemical scaffolds currently available for M^pro inhibitors, identification of these compounds further adds to the medicinal chemist’s toolkit for SARS-CoV-2 antiviral drug design.

Fig. 4.

Docking of anacardic acid, aloesin, aloeresin D, and TCID in M^pro active site. A. Anacardic acid (magenta) docked with a predicted affinity of −5.6 kcal/mol. B. Aloesin (cyan) docked with a predicted affinity of −7.5 kcal/mol. C. Aloeresin D (purple) docked with a predicted affinity of −6.8 kcal/mol. D. TCID (red) docked with a predicted affinity of −5.8 kcal/mol. The catalytic dyad of Cys 145 and His 41 are displayed in green and orange respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

In conclusion, here we have identified two natural product compounds, aloesin and aloeresin D, as novel SARS-CoV-2 M^pro inhibitors through screening of a high-content protease inhibitor library. Both of these compounds are safe in humans and have functional groups that are readily accessible for modification allowing them to serve as leads for M^pro inhibitors. Anacardic acid and TCID performed well in our study but were both ruled out due to potential toxicity. Additionally, we have confirmed that SARS-CoV-2 M^pro exhibits cooperative kinetics during catalysis, suggesting that the protein dimer interface may be an attractive target for allosteric inhibitors. Combined, these results advance our knowledge of structure–function relationships in M^pro and offer new molecular scaffolds for inhibitor design.

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 are the Supplementary data to this article: