Discovering novel Cathepsin L inhibitors from natural products using artificial intelligence

Abstract

Cathepsin L (CTSL) is a promising therapeutic target for metabolic disorders. Current pharmacological interventions targeting CTSL have demonstrated potential in reducing body weight gain, serum insulin levels, and improving glucose tolerance. However, the clinical application of CTSL inhibitors remains limited. In this study, we used a combination of artificial intelligence and experimental methods to identify new CTSL inhibitors from natural products. Through a robust deep learning model and molecular docking, we screened 150 molecules from natural products for experimental validation. At a concentration of 100 µM, we found that 36 of them exhibited more than 50 % inhibition of CTSL. Notably, 13 molecules displayed over 90 % inhibition and exhibiting concentration-dependent effects. The molecular dynamics simulation on the two most potent inhibitors, Plumbagin and Beta-Lapachone, demonstrated stable interaction at the CTSL active site. Enzyme kinetics studies have shown that these inhibitors exert an uncompetitive inhibitory effect on CTSL. In conclusion, our research identifies Plumbagin and Beta-Lapachone as potential CTSL inhibitors, offering promising candidates for the treatment of metabolic disorders and illustrating the effectiveness of artificial intelligence in drug discovery.

Graphical Abstract

Highlights

• •We used artificial intelligence to identify new CTSL inhibitors from natural products.
• •We identified nine molecules that showed more than 90 % inhibition at a concentration of 100 µM.
• •The two most potent inhibitors, Plumbagin and Beta-Lapachone, demonstrated stable interaction at the CTSL active site.
• •Enzyme kinetics research revealed an uncompetitive inhibition pattern for CTSL by Plumbagin and Beta-Lapachone.

1. Introduction

Metabolic-related diseases refer to disorders characterized by abnormalities in material metabolism or energy metabolism, manifesting as metabolic dysregulation in multiple systems throughout the body. In recent years, the incidence of metabolic-related diseases has been on the rise globally [1]. According to the World Health Organization (WHO), the global prevalence of diabetes has nearly doubled in the past 30 years, rising from 4.7 % in 1980 to 8.5 % in 2014. In 2019, it was estimated that approximately 463 million adults worldwide were living with diabetes, a number expected to increase to 700 million by 2045 [2], [3]. Similarly, the prevalence of obesity is also steadily increasing, with about 650 million adults and 340 million children and adolescents (ages 5–19) suffering from obesity [4], further leading to a rise in the incidence of metabolism-related diseases such as atherosclerosis and non-alcoholic fatty liver disease (NAFLD) [5], [6]. Metabolic-related diseases often have severe consequences, significantly impacting the quality of life for patients and imposing a serious economic burden. For example, uncontrolled diabetes can lead to blindness, kidney failure, nerve damage, and limb amputation [7]. Obesity increases the risk of heart disease, stroke, and cancer [8]. In some cases, NAFLD can progress to a more severe liver condition, non-alcoholic steatohepatitis (NASH), eventually leading to cirrhosis and liver failure [9]. Therefore, metabolic-related diseases have become a serious public health issue and efforts to prevent, diagnose, and treat these diseases are crucial for improving patient health and reducing medical expenses.

In recent years, there has been growing interest in the role of Cathepsin L (CTSL) in the pathogenesis of metabolic-related diseases. CTSL, a lysosomal cysteine protease, is involved in the degradation of protein substrates within lysosomes, the maturation of hormones, and antigen presentation [10], [11], [12], [13]. It has been established that CTSL is associated with various metabolic processes, including glucose and lipid metabolism, insulin signaling, and inflammation. Furthermore, CTSL plays a role in numerous metabolic-related diseases, such as obesity, atherosclerosis, NAFLD, and diabetic kidney disease, making it a highly promising therapeutic target [14], [15], [16], [17], [18]. In-vitro studies have demonstrated that CTSL is crucial in the degradation of key molecules such as fibronectin, insulin receptor (IR), and insulin-like growth factor-1 receptor (IGF-1R), which are vital for adipogenesis and glucose metabolism [19], [20], [21]. The use of the CTSL inhibitor CLIK-148 has been shown to mitigate weight gain and serum insulin levels in high-fat diet-induced obese mice and ob/ob mice, as well as improve their glucose tolerance [18]. Furthermore, the cysteine protease inhibitor E-64 significantly reduces CTSL expression in the kidneys of diabetic rats and ameliorates proteinuria levels [22]. These studies suggest that CTSL inhibitors hold great potential in the development of therapeutics for metabolic-related diseases. However, to date, no CTSL inhibitors can be used in clinical practice [23], [24], [25]. Therefore, the development of novel CTSL inhibitors is critical for the treatment of metabolic-related diseases.

Natural products derived from plants and animals have long been a primary source for drug discovery, particularly in the development of anticancer and antimicrobial agents [26], [27]. In recent decades, there has been a notable increase in the use of medicinal plants for health promotion and disease treatment, with various plant extracts now being integrated into prescription drugs in several countries [28], [29], [30]. Galega officinalis is a summer-flowering perennial herb found in most temperate regions and used in diabetes treatment for centuries [31]. It was later found that guanidine compounds in Galega officinalis were the primary components responsible for lowering blood glucose levels, leading to the discovery of metformin, which is now the first-line medication for treating type 2 diabetes [32]. Additionally, biguanide compounds can recruit endogenous Zn²⁺ to inhibit the activity of cysteinyl cathepsins [33]. This has directed our research towards finding potential CTSL inhibitors from natural products. Drug development is an exceptionally time-consuming and labor-intensive process, typically requiring over a decade from target selection to clinical trial approval, and expending billions of dollars in resources [34]. However, the advancement of artificial intelligence technology has significantly accelerated the process of drug discovery [35], [36], [37]. Since the 1960s, medicinal chemists have applied artificial intelligence in various forms to the field of new drug development, achieving varying degrees of success. Among the diverse AI methodologies, supervised learning has seen the widest application. It employs labeled training datasets to train AI models, which can then predict properties of given chemical structures [38], [39]. The Message Passing Deep Neural Network Model (MPNNs) is an example of such a deep learning network based on supervised learning. Its primary framework consists of nodes and edges, with individual atomic information of compounds representing the nodes, and chemical bond information between atoms depicting the edges connecting adjacent nodes [40]. In this study, we used a combination of artificial intelligence and experimental methods to find new CTSL inhibitors from natural products. We found that nine molecules displaying activity inhibition against CTSL. The most effective molecules were subsequently chosen for advanced enzyme kinetics analyses and molecular dynamic simulations.

2. Materials and methods

2.1. Deep learning model training and predictions

We used a message passing neural network (MPNNs) to build a binary classification model that could predict the probability of whether a molecule would inhibit the activity of CTSL. The detail of model training has been described previously [41]. The prediction dataset was sourced from the Selleck L8300 Library and the Topscience L6000 Library, which comprised 6439 natural products characterized by a diversity of structures and functions. To ensure that our training and prediction datasets were completely distinct, a deduplication algorithm was employed. Canonical SMILES strings for each molecule were generated utilizing the canonicalization algorithm available in RDKit [42], with each string uniquely representing its corresponding chemical structure. This approach enabled the exclusion of test molecules that shared SMILES strings with those in the training set. Consequently, a total of 4631 unique natural products were identified and used as the prediction dataset.

2.2. Receptor protein selection and preparation

We use human CTSL X-ray structures co-crystallized with a covalent inhibitor from Protein Data Bank (PDB) database (PDB ID: 5MQY, resolution 1.13 Å) to perform molecule docking [43]. The Schrödinger protein preparation wizard was used to prepare each crystal structure [44]. We refined the crystal structure by first deleting all water molecules and artifacts, subsequently adding hydrogen atoms, and generating potential metal binding states. Hydrogen bond sampling with adjustment of active site, water molecule orientations was performed using PROPKA at pH 7.4. Finally, the protein-ligand complexes were subjected to geometry refinements using the OPLS3 force field in restrained minimizations [45].

2.3. Ligand preparation

We employed the gen3D (generate 3D coordinates) module of Open Babel (version 2.3.0) for the conversion of chemical structures in the Selleck L8300 library and Topscience L6000 Library from SMILES format to mol2 format, rendering them suitable for subsequent docking analyses [46]. Before docking, we processed ligands using the LigPrep module within the Schrödinger suite. The Epik tool was employed to generate their ionization and tautomeric states at a pH of 7.4. Subsequently, the ligands underwent minimization with the OPLSe-3 force field until achieving a root-mean-square deviation (RMSD) of 2.0 Å. These optimized ligands were then utilized for docking analysis [45], [47].

2.4. Molecule docking

The interactions of ligands and CTSL were determined using the glide SP flexible ligand mode of Schrödinger 2021–2 version [48]. The receptor grid was generated using Receptor Grid Generation module of Schrödinger suite centroid of the co-crystallized inhibitor. By default, the ligands were modeled as flexible, and only nonplanar conformations of amides were subject to penalties. Lastly, Epik state penalties were applied to the docking scores. This is used to allow for higher energy states of ligands that underwent ionization or tautomerization during the preparation stage. The number of output poses per ligand was limited to the single best pose, while post-docking minimization was also performed.

2.5. Similarity Screening

We employed the RDKit toolkit to assess molecular similarity and perform dimensionality reduction [49]. The SMILES strings of each molecule were converted to Morgan fingerprints. Similarity between molecules was quantified using the Tanimoto coefficient. Furthermore, t-Distributed Stochastic Neighbor Embedding (t-SNE) was applied to the fingerprint data to project it into a two-dimensional space, facilitating the visualization of molecular clusters and patterns. The similarity matrix was visualized as a heatmap, with each cell displaying the Tanimoto coefficient.

2.6. CTSL activity test

The assessment of the inhibition of CTSL by small molecules was conducted in a cell-free system utilizing a commercially available kit (Abcam, Cat. No. ab65306). In detail, 5 µl of CTSL protein (25 mg/L), which was prepared from human liver tissue (Sigma-Aldrich, Cat. No. C6854) was used as the enzyme in this cell-free enzymatic reaction system. For the initial screening stage, each molecule were prepared to a concentration of 5 mM. 2 µl of these 5 mM molecules or drugs was added to the set 100 µl system (1 µl of CTSL protein (25 mg/ L) + 95.5 µl of CL buffer + 1 µl of CTSL inhibitor + 0.5 µl of DTT (1 mM) + 2 µl of CL substrate Z-FR-AFC (0.5 mM)) to achieve a working concentration of 100 µM. For further confirmation and determination of the half maximal inhibitory concentration, molecules were tested at working concentrations of 100 µM, 10 µM, 1 µM, 0.1 µM, and 0.01 µM. The equivalent amount of solvent was used as a control.

2.7. Enzyme kinetics

The Michaelis constant (Km) and maximal reaction velocity (Vmax) of CTSL were determined at varying concentrations of substrate Z-FR-AFC (6.25 µM, 12.5 µM, 25 µM, 50 µM, and 100 µM), in the presence of increasing concentrations of inhibitors (0 µM, 6.25 µM, 12.5 µM, 25 µM, 50 µM, and 100 µM). Reaction velocities were calculated using a conversion factor obtained from a standard curve. Data analysis was conducted using GraphPad Prism software. All measurements were corrected by subtracting the values of the negative control (0 µM inhibitors). The enzyme kinetics were characterized by fitting the data to the Michaelis-Menten model, with Vmax representing the extrapolated maximum velocity and Km defined as the substrate concentration at which half of Vmax is achieved.

2.8. Molecular dynamics simulation studies

We performed molecular dynamics (MD) simulations with ligands (Plumbagin and Beta-Lapachone) and the receptor proteins (CTSL and CTSL with Zn²⁺ at the active site, PDB ID: 5MQY and 4AXL) using Groningen machine for chemical simulation (GROMACS) software version 2021.3 [50]. We used the protein-ligand complexes obtained from molecule docking as the starting point for simulations and implemented the CHARMM36 all-atom force field to establish the complex stability [51]. We used the CGenFF server to prepare and retrieve the ligand topology [52] and solvated the ligand-receptor complex using the TIP3P water model [53]. Since 4AXL contains Zn²⁺, We used the Amber99SB-ILDN force field [54] to prepare the receptor topology and use the Sobtop tool to generate the ligand topology file [55]. To neutralize the total charge, we added Na⁺ to the system. To minimize the initial energy of the complex for the MD simulation, we conducted two consecutive equilibrium simulations, employing a sequence of canonical (NVT) and isobaric-isothermal (NPT) ensembles, each lasting 100 ps. The system was further equilibrated to perform 10 ns MD simulations at 300 K and 1 bar using the GROMACS 2021.3 simulation package.

One of the most often used techniques for calculating binding free energy is molecular mechanics/Poisson-Boltzmann surface area (MM/PBSA) [56], which describes the structural and molecular stability of the ligands in the active site of the protein. In this study, we used the gmx_MMPBSA tool to calculate the binding free energy [57]. The MM/PBSA is estimated by Eqs. (1) – (3):

$$\text{Δ}\text{G}\text{bind}=\text{Δ}\text{H}-\text{T}\text{Δ}\text{S}\approx \text{Δ}\text{E}\text{MM}+\text{Δ}\text{G}\text{sol}-\text{T}\text{Δ}\text{S}$$ (1)

$$\text{Δ}\text{E}\text{MM}=\text{Δ}\text{E}\text{inte}+\text{Δ}\text{E}\text{ele}+\text{Δ}\text{E}\text{vdw}$$ (2)

$$\text{Δ}\text{G}\text{sol}=\text{Δ}\text{G}\text{pol}+\text{Δ}\text{G}\text{np}$$ (3)

Where ΔE_MM represents the molecular mechanics energy change, decomposed into internal energy (ΔE_inte), electrostatic energy (ΔE_ele) and van der Waals energy (ΔE_vdw). The solvation free energy change (ΔG_pol) includes both polar and non-polar (ΔG_np) contributions. To simplify the computation, the entropic term TΔS is neglected.

2.9. Statistical analysis

All statistical analyses were conducted with GraphPad Prism version 8.0.1 software. Data are presented as the mean ± SEM, and statistical details for individual experiments can be found in the figure legends.

3. Results

3.1. CTSL inhibitor prediction using MPNNs

MPNNs is a type of deep learning model designed for processing graph-structured data. In chemistry, molecules are represented as graphs where atoms are nodes and chemical bonds are edges [58]. MPNNs learn the properties of molecules by passing information between these nodes [59]. In drug discovery, MPNNs are capable of handling complex molecular structures to effectively predict the biological activity of molecules. This is crucial for screening potential drug candidates in large compound libraries. By learning the interactions between known drugs and targets, MPNNs can predict the activity of new compounds and identify promising candidates [60], [61]. In our previous study, we established a robust MPNNs model capable of predicting whether a molecule can inhibit CTSL activity [41], [62]. In this research, we applied the well-established model to discover potential CTSL inhibitors from the Selleck L8300 Library and Topscience L6000 Library, comprising 6439 natural products characterized by diverse structures and functions. We removed the molecules with the same molecular graphs as the training dataset, leaving 4631 molecules of diverse structure and function. Then, we determined the prediction scores for each molecule, and molecules were ranked based on their probability of displaying activity inhibition against CTSL (Supplementary Table 1, Fig. 1A).

Fig. 1.

CTSL inhibitors prediction and enzyme activity test. A, Rank-ordered CTSL inhibitor prediction scores, higher prediction scores indicate a greater probability of activity inhibition against CTSL. B, The molecular docking scores of natural products to CTSL (PDB 5MQY). C, The top 150 molecules were chosen for verifying the inhibition effect against CTSL in a cell-free system at a single dose of 100 μM. The data are expressed as the mean of three individual trials.

3.2. Molecule docking and CTSL activity test

In recent years, computer-aided drug discovery using a structure-based approach has emerged as a predominant trend within computational chemistry. Molecular docking enables the prediction of interactions between small molecules and their respective targets, typically proteins, facilitating an evaluation of the molecule's binding affinity and conformation. This computational approach allows researchers to identify and refine prospective drug candidates prior to their synthesis and experimental validation. Following the predictions from the MPNNs model, we conducted molecular docking using the Schrödinger Glide SP suite, employing human CTSL crystal structures obtained from the PDB database (PDB code: 5MQY). The 3D molecular structures were generated using the gen3D module of Open Babel (version 2.3.0) [46]. The docking results were ranked according to the docking score with CTSL (Supplementary Table 1, Fig. 1B). We combined the prediction score rankings and docking score rankings of these molecules to obtain an integrated ranking (Supplementary Table 1). From the molecules available for purchase, we acquired the top 100 molecules based on the integrated ranking. Additionally, we purchased the top 25 molecules based on the prediction score rankings and the top 25 molecules based on the docking score rankings from the remaining molecules, totaling 150 molecules. We administered dose of 100 μM to evaluate their potential in inhibiting CTSL activity. The assay results showed that 36 out of the 150 molecules exhibited more than 50 % inhibition of CTSL. Notably, 13 of them demonstrated inhibition efficiencies exceeding 90 % (Fig. 1C). In this study, we employed an aminomethyl coumarin (AFC)-based assay to measure CTSL enzyme activity. Although this method is a robust and widely accepted assay for cathepsin activity, it is susceptible to false positives and interference from compounds containing chromophores. Therefore, we added 13 of the most effective molecules at a concentration of 100 µM to a 10 µM AFC solution and monitored the AFC fluorescence levels. The results indicated that 4 molecules significantly affected the AFC fluorescence values, while the remaining 9 molecules, including Plumbagin, Dihydrotanshione l, Naringenin chalcone, Phenylacetaldehyde, Flavokawain B, Farrerol, Beta-Lapathone, Asperphenamate and (E)-Cardamonin, did not impact the AFC fluorescence (Supplementary Figure 1). Then, we tested a range of concentrations of these 9 molecules for further confirmation and determination of the half maximal inhibitory concentration (IC50). Notably, all 9 molecules inhibited CTSL activity in a concentration-dependent manner (Supplementary Figure 2, Fig. 2A-I). Since Asperphenamate is a well-known CTSL inhibitor [63], [64], [65], we analyzed the binding modes of the two most effective inhibitors among the remaining molecules, Plumbagin and Beta-Lapachone, with the CTSL active site and generated a three-dimensional (3D) representation to illustrate their interactions. Plumbagin forms two hydrogen bonds with Gly68 and Met161 residues of CTSL (Fig. 3A), Beta-Lapachone forms a hydrogen bond with Gln19 residue of CTSL (Fig. 3B). These findings substantiate the efficacy and reliability of the integrated deep learning and molecular docking approach employed in the identification of CTSL inhibitors.

Fig. 2.

Half maximal inhibitory concentration (IC50) determination. A-I, Nine predicted CTSL inhibitors, Plumbagin (A), Dihydrotanshione l (B), Naringenin chalcone (C), Phenylacetaldehyde (D), Flavokawain B (E), Farrerol (F), Beta-Lapathone (G), Asperphenamate (H) and (E)-Cardamonin (I), with inhibition efficiencies greater than 90 % at 100 μM were further tested for determination of the half maximal inhibitory concentration (IC50) in the cell-free system. Corresponding molecular structure was drawn by Chemdraw. Non-linear fit to a variable response curve from one representative experiment with three replicates is shown (black lines). The data are expressed as the mean ± s.e.m.

Fig. 3.

Docking results of Plumbagin and Beta-Lapachone in the crystal structure of CTSL. A, Docking interactions of Plumbagin in the active sites of CTSL. Plumbagin is represented as sticks with carbon atoms in cyan. CTSL is displayed as a cartoon in green. B, Docking interactions of Beta-Lapachone in the active sites of CTSL. Beta-Lapachone is represented as sticks with carbon atoms in yellow. CTSL is displayed as a cartoon in green.

To investigate the relationship between compound structure and CTSL inhibition, we conducted a Tanimoto similarity analysis on the 9 active compounds using RDKit and performed dimensionality reduction with t-SNE. The results indicated that (E)-Cardamonin and Flavokawain B exhibited high structural similarity, while the remaining molecules showed low structural similarity (Fig. 4A-B). These findings demonstrate that our artificial intelligence methods can effectively screen for structurally diverse inhibitors.

Fig. 4.

Molecular clustering analysis. A, The SMILES strings of each molecule were converted to Morgan fingerprints using RDKit. Similarity between molecules was quantified using the Tanimoto coefficient. similarity matrix was visualized as a heatmap, with each cell displaying the Tanimoto coefficient. B, T-distributed stochastic neighbor embedding (t-SNE) was used to visualize fingerprint similarities in a scatter plot, where each dot represents a unique hit compound. The IC50 values were represented by the dot size, with larger dots indicating lower IC50 values and smaller dots indicating higher IC50 values.

3.3. Molecular dynamics simulation

To elucidate the specific binding patterns and the time-dependent stability of the complexes between these inhibitors and CTSL, we conducted MD simulations using GROMACS software (version 2021.3) with ligands (Plumbagin and Beta-Lapachone) and the receptor protein CTSL (PDB ID:5MQY). Simultaneously, as Zn²⁺ is a natural inhibitor of Cathepsins [66], we conducted MD simulations using CTSL with Zn²⁺ at the active site as the receptor (PDB ID:4AXL). Energy minimization was achieved using a steepest descent algorithm, ensuring the absence of atomic or chemical bond collisions within the system. Subsequent equilibration facilitated 10 ns MD simulations at 300 K and 1 bar (Supplementary Figure 3). The root mean square deviation of CTSL (RMSD-P) in complexation with Plumbagin ranged from 0.04 to 0.19 nm, averaging 0.15 nm (Fig. 5A), indicating minimal receptor structure fluctuations while the ligand occupied the active site, thus affirming system stability. The RMSD-L for ligand fit in the receptor active site varied from 0.05 to 0.36 nm, averaging 0.15 nm (Fig. 5B), corroborating the ligand-receptor stability and consistent ligand orientation during the simulation. Root mean square fluctuation (RMSF) analysis revealed residual mobility and structural integrity, with all residues displaying RMSF values below 0.45 nm in the presence of Plumbagin (Fig. 5C). Next, the number of hydrogen bonds between the receptor−ligand complex throughout the simulation period were determined using Gromacs g_hbond utility. A maximum of three hydrogen bonds were observed between the ligand and the receptor, whereas one to two H-bonds were consistently present throughout the simulation period (Fig. 5D). Further, the radius of gyration of protein (Rg-P) and ligand (Rg-L) was in the range, supporting the stability of the ligand—protein complex (Supplementary Video 1, Fig. 5E-F).

Fig. 5.

Molecular dynamics simulation results. A-F, RMSD-P (A), RMSD-L (B), RMSF (C), H-bond (D), radius of gyration of protein (E) and radius of gyration of ligand (F) for ligand-receptor complex. Blue represents the Plumbagin-CTSL complex, red represents the Beta-Lapachone-CTSL complex, green represents the Plumbagin-CTSL complex in the presence of Zn²⁺, and yellow represents the Beta-Lapachone-CTSL complex in the presence of Zn²⁺. G, The binding free energy terms obtained from the MM/PBSA calculations.

Dynamic simulations revealed that Beta-Lapachone is capable of forming stable bindings at the active site (Supplementary Video 2). The RMSD-P value for CTSL with Beta-Lapachone ranged from 0.04 to 0.14 nm, averaging 0.11 nm (Fig. 5A), while RMSD-L ranged from 0.05 to 0.55 nm, averaging 0.25 nm (Fig. 5B). All residues maintained RMSF values below 0.45 nm with Beta-Lapathone in the active site (Fig. 5C). Despite limited hydrogen bonding, Rg-P and Rg-L remained stable (Fig. 5D-F).

However, after add Zn²⁺ to the active site of CTSL, the binding of both Plumbagin and Beta-Lapachone to the active site became unstable (Supplementary Videos 3–4). The RMSD-P results indicated that the protein structure remained stable throughout the entire simulation period (Fig. 5A). However, in the presence of Zn²⁺, the RMSD-L results of Plumbagin ranged from 0.04 to 1.32 nm, averaging 0.91 nm. The RMSD-L results of Beta-Lapachone from 0.04 to 12.28 nm, representing that Beta-Lapachone could not stably exist at the active site of CTSL (Fig. 5B). The initial perturbation in the Rg-P trajectory of Beta-Lapachone may indicate the spatial adjustment of the ligand in the binding site (Fig. 5E). The Rg-L results indicated that the ligand maintained stable conformations throughout the simulation, with no significant unfolding or compaction (Fig. 5F). To better understand the molecular interactions and stability of Plumbagin and Beta-Lapachone with CTSL or CTSL in the presence of Zn²⁺, we executed a detailed analysis of the binding free energy through the MM/PBSA. The results demonstrated that Plumbagin and Beta-Lapachone exhibited strong affinity towards CTSL, with ΔG_bind of − 15.93 kcal/mol and − 10.98 kcal/mol, respectively. However, in the presence of Zn²⁺, the affinity of Plumbagin and Beta-Lapachone towards CTSL was significantly reduced (Fig. 5G).

3.4. Enzyme kinetics

Enzyme kinetics research is crucial in the study of enzyme inhibitors. It helps in determining the mechanism of inhibition, whether an inhibitor binds competitively, non-competitively, or through another mechanism. Such knowledge is vital for designing inhibitors that can target specific enzymes. Hence, we performed enzyme kinetic studies of Plumbagin and Beta-Lapathone with CTSL. At concentrations of 6.25 µM to 100 µM, Plumbagin and Beta-Lapachone treatment substantially decreased both Km and Vmax, indicating that these inhibitors follow an uncompetitive inhibition mechanism (Table 1, Fig. 6A-D).

Table 1.

Kinetic parameters of CTSL in the presence of Plumbagin and Beta-Lapachone.

	Plumbagin		Beta-Lapachone
Concentration (μM)	Km (μM)	Vmax (Rlu/min)	Km (μM)	Vmax (Rlu/min)
0	8.1 ± 2.1	369.0 ± 17.9	10.1 ± 0.3	234.1 ± 40.2
6.25	7.2 ± 0.1	321.7 ± 13.9	10.8 ± 3.5	194.8 ± 9.2
12.5	6.9 ± 1.8	277.5 ± 15.6	18.9 ± 14.7	181.3 ± 32.9
25	7.2 ± 1.7	270.5 ± 11.2	9.2 ± 3.3	136.9 ± 27.6
50	4.5 ± 3.6	216.8 ± 20.5	6.9 ± 6.8	103.1 ± 2.2
100	5.1 ± 3.9	12.3 ± 2.4	6.1 ± 2.9	39.9 ± 0.2

Fig. 6.

Enzyme kinetics results. A-B, The velocity (A) and Lineweaver–burk plot (B) of CTSL was determined under different concentrations of Plumbagin at varying substrate concentrations. C-D, The velocity (A) and Lineweaver–burk plot (B) of CTSL was determined under different concentrations of Beta-Lapachone at varying substrate concentrations.

4. Discussion

CTSL plays a significant role in a variety of metabolism-related diseases such as diabetes, obesity, and NAFLD, as well as in viral infections, inflammatory states, tumor invasion, and metastasis [23]. Since the first CTSL inhibitor, cystatin, was isolated from Aspergillus in 1981, an increasing number of CTSL inhibitors have been synthesized. Pharmacological inhibition of CTSL has been shown to mitigate body weight gain, lower serum insulin levels, and enhance glucose tolerance [18]. Additionally, a CTSL inhibitor effectively prevents autoimmune diabetes in mice through the suppression of CD8⁺ T-cell activity [67]. However, due to toxicity and some unpredictable side effects, these small molecules have not advanced to clinical trials [68], [69]. Recent studies have found that K777 can reduce the infectivity of SARS-CoV-2 by inhibiting the activity of host CTSL and has completed Phase I clinical trials [70]. It was also reported that zinc pyrithione is a potent inhibitor of CTSL, which effectively inhibits SARS-CoV-2 entry and replication ex vivo [71]. However, its efficacy and safety still require further validation. Therefore, discovering more potential CTSL inhibitors is of great significance for drug development.

Drug development is one of the most complex, risky, and time-consuming fields of technical research in human development. It requires the collaboration of professionals in biology, chemistry, medicine, and other related areas. The entire process can take decades and cost billions of dollars, with a failure rate exceeding 90 % [72]. Recently, artificial intelligence has become increasingly prevalent in drug discovery, significantly impacting the early stages of the development pipeline [73], [74]. In this study, we employed a synergistic approach of artificial intelligence and experimental methods to identify new CTSL derived from natural products. We found that nine molecules exhibited concentration-dependent inhibition of CTSL activity. Notably, the two most effective molecules, Plumbagin and Beta-Lapachone, inhibited CTSL with IC50 values of 31.3 μM and 9.6 μM, respectively. Molecular dynamics simulations demonstrated stable binding of these inhibitors at the CTSL active site: Plumbagin forms two hydrogen bonds with Gly68 and Met161 residues of CTSL, Beta-Lapachone forms a hydrogen bond with Gln19 residue of CTSL. Enzyme kinetics studies have shown that these inhibitors exert an uncompetitive inhibitory effect on CTSL. Zn²⁺ is a natural inhibitor of Cathepsins [66], we further conducted MD simulations using CTSL with Zn²⁺ at the active site as the receptor. However, we found that the affinity of Plumbagin and Beta-Lapachone towards CTSL was significantly reduced. Given the absence of Zn²⁺ in the CTSL activity assay system we used, our results showed that the inhibitory effects of Plumbagin and Beta-Lapachone on CTSL are independent of the recruitment of endogenous Zn²⁺.

Plumbagin is a naturally occurring naphthoquinone primarily found in the roots of Plumbago species. It possesses anti-inflammatory and antioxidant properties that are essential for alleviating metabolic disorders such as diabetes and obesity [75]. Beta-Lapachone, another naphthoquinone sourced from the lapacho tree, is effective in enhancing glucose uptake and improving insulin sensitivity [76]. In this study, we found that these molecules can significantly inhibit CTSL’s activity, which may represent one of the key mechanisms underlying their anti-inflammatory actions. The Molecular dynamics simulations demonstrated stable binding of all these inhibitors at the CTSL active site. However, all these inhibitors exhibited an uncompetitive mode of inhibition, indicating potential additional binding sites beyond the CTSL's active site. Future investigations will aim to elucidate these specific binding sites using cryo-electron microscopy and other techniques.

Author statement

Jin-Kui Yang: conceived the idea for the study, supported the study, designed the experiments, and wrote the manuscript. Qi Li: designed, performed the experiments, and wrote the first version of manuscript. Si-Rui Zhou: performed the experiments. Hanna Kim: partially performed the experiments. Hao Wang and Juan-Juan Zhu: helped with the interpretation of the results, design of experiments, and supported the study. All authors agree to be accountable for all aspects of the work. All authors read and approved the final manuscript to be published.

CRediT authorship contribution statement

Juan-Juan Zhu: Funding acquisition, Investigation. Hao Wang: Funding acquisition, Investigation. Hanna Kim: Data curation. Si-Rui Zhou: Data curation, Formal analysis. Jin-Kui Yang: Conceptualization, Supervision, Writing – review & editing, Funding acquisition. Qi Li: Data curation, Formal analysis, Investigation, Methodology, Project administration, Writing – original draft.

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 material

Supplementary Table 1

Supplementary material.

.

Supplementary Table 2

Supplementary material.

.