Discovery of selective ACAT2 antagonist via a combination strategy based on deep docking, pharmacophore modelling, and molecular dynamics simulation

Abstract

Acyl-CoA: cholesterol acyltransferase (ACAT), a pivotal enzyme in the absorption and metabolism of cholesterol, is primarily responsible for intracellular esterification. ACAT inhibition is expected to diminish plasma lipid levels by impeding intestinal cholesterol absorption, thereby preventing the progression of atherosclerotic lesions. A previous study shows that selective inhibition of ACAT2 significantly mitigated hypercholesterolaemia and atherosclerosis in mouse models. Therefore, the need for ACAT2 selective inhibitors becomes particularly urgent. In this study, we established a multilayer virtual screening workflow and subjected biologically evaluated representative compounds to enzyme inhibitory assays. The experimental results indicated that the two compounds, STL565001 (inhibition rate at 25 μM: 75.7 ± 27.8%, selectivity = 6) and STL528213 (inhibition rate at 25 μM: 87.8 ± 12.4%, selectivity = 13), demonstrated robust activity against ACAT2, displaying greater selectivity for ACAT2 than for ACAT1. The molecular mechanisms governing the inhibitory activities of the selected compounds were systematically elucidated using computational approaches. In addition, hotspot residues in ACAT2 that are crucial for ligand binding were successfully identified. In summary, we devised a multilayer screening scheme to expeditiously and efficiently identify compounds with enzyme inhibitory activity, offering novel scaffolds for subsequent drug design centred on ACAT2 targets.

Graphical Abstract

Introduction

Atherosclerosis is a prevalent and insidious disease responsible for a significant burden of morbidity and mortality worldwide¹. Atherosclerosis, a pathological condition characterised by the thickening and hardening of arterial walls, can affect the majority of arteries in the body, including those in vital organs such as the heart, brain, pelvis, arms, legs, and kidneys^2–4. It is a leading risk factor for the development of cardiovascular diseases⁵. Elevated concentrations of low-density lipoprotein (LDL) within the bloodstream facilitate the creation of fatty deposits and the progression of atherosclerosis⁶. One approach for treating atherosclerosis is the inhibition of the intracellular enzyme acyl-coenzyme A: cholesterol acyltransferase (ACAT), which has two isoforms, ACAT1 and ACAT2^7–11. ACAT is an intracellular enzyme that converts free cholesterol to cholesterol esters, which can be stored within cells or transported as LDL particles¹². Before the identification of ACAT isomerases (ACAT1 and ACAT2), several artificial ACAT inhibitors were developed as potential therapeutic drugs for the treatment of atherosclerosis. However, during the 1990s, clinical trials investigating two ACAT inhibitors, pactimibe and avasimibe, yielded unfavourable results^7,13,14. The results showed that suppressing ACAT1 could result in the build-up of cholesterol and, subsequently, the manifestation of toxic effects on immune cells and cells in the vascular wall^15–18. Notably, both avasimibe and pactimibe lack abilities to inhibit ACAT2^19,20. Another study showed that selective inhibition of ACAT1 resulted in detrimental adverse effects and did not decrease the progression of atherosclerosis in a mouse model¹⁵. Compared with non-selective or ACAT1 selective inhibitors, inhibitors that selectively target ACAT2 may exhibit greater stability and efficacy in managing hyperlipidaemia²¹. Therefore, the identification of novel ACAT2 selective inhibitors. In recent years, virtual screening has become an attractive tool for discovering new drug candidates. The use of computational methods in drug discovery, such as pharmacophore modelling, molecular docking, and molecular dynamics simulations, has enabled the identification of novel inhibitors that target specific proteins with high affinity and selectivity. In this study, we combined the hypogen pharmacophore modelling technique with molecular docking and molecular dynamics simulations to virtually screen a library of compounds as potential ACAT2 inhibitors (Figure 1). Ultimately, the screened compounds underwent preliminary evaluation of their biological activities through NBD22-steryl ester fluorescence assay and cholesterol oxidase assay. This study aimed to identify novel selective ACAT2 inhibitors with improved efficacy and safety profiles.

Figure 1.

Virtual screening workflow.

Method

Deep docking

Gentile et al. introduced a Deep Docking (DD) workflow to expedite the virtual screening processes for extensive databases²². In the previous study, the DD method employs deep learning architectures trained on docking scores derived from select segments of a chemical repository to estimate the docking results for as yet unprocessed compounds, thus expunging unfavourable entities through an iterative process. The integration of DD methodology alongside the molecular docking framework facilitated swift and precise computation of docking scores for extensive arrays of compounds against pivotal target proteins, showcasing remarkable data condensation by up to 100-fold and enrichment of high-scoring molecules by a staggering 6000-fold (with negligible compromise on favourably docked entities). In the current study, the training of the model followed a workflow similar to that described in Gentile’s study. We conducted a virtual screening of all Topscience Database (https://www.tsbiochem.com/service/topscience-database) compounds against ACAT2 using the deep docking method. The Top Science database was utilised in the SMILES format to conduct the virtual screening study. In addition, the stereoisomers in the training set were expanded using Ligprep to enumerate and specify previously unspecified stereocenters. Extended Connectivity Fingerprints with a radius of 2 and consisting of 1,024 bits were calculated from the SMILES of the prepared compounds and adopted as descriptors for subsequent analysis. These fingerprints, which are widely used in chemical informatics, provide representations of molecular structures for chemical informatics studies. Model initiation involved random sampling of 15,000 molecules from the Topscience Database, which were subsequently partitioned into training, validation, and test sets with equal distribution. The computed scores of these samples were then used to initialise the deep neural network (DNN) model. The predominant SP score within the uppermost 1% echelon of compounds in the validation set served as a threshold for classifying molecules as either top-scoring (scores below the cut-off) or low-scoring.

In subsequent iterations, the training set was augmented by incorporating 5000 molecules randomly selected from positive predictions in the preceding iteration, and this iterative procedure was repeated. The score cut-off threshold was adjusted towards lower values by reducing the percentage of top-ranked molecules identified as positive by 0.1% in each successive iteration, reaching 0.01% in the final iteration. A total of 11 iterations were performed.

Molecular docking

Crystallographic structures of ACAT2 (PDB ID: 7N6Q)²³ and ACAT1 (PDB ID: 6VUM)²⁴ were used for the docking studies. Additionally, the ligands were meticulously crafted using the Ligand Preparation module in Maestro. Proteins were prepared using the Protein Preparation Wizard in Maestro. This involved the removal of water molecules from the protein, the addition of polar hydrogen, and addressing the protonation states of charged residues within ACAT1 and ACAT2. Receptor grids were prepared through the Receptor Grid Generation process, focusing exclusively on the active sites of ACAT1 and ACAT2. To evaluate the performance of the virtual screening procedure employed in ACAT2 molecular docking, an assessment was conducted using a Receiver Operating Characteristic (ROC) curve. Redocking analysis was also conducted to evaluate the performance of the docking method.

To determine the selectivity of the identified molecules, they were subsequently subjected to Glide molecular docking within the active sites of ACAT1 and ACAT2 to elucidate their interactions with these non-target ACAT isoforms. The obtained Glide XP docking poses were further subjected to molecular dynamics simulations to calculate the molecular mechanics − Poisson Boltzmann surface area (MM-PBSA).

Pharmacophore modelling

A library of 42 ACAT inhibitors from the relevant literature^25–32 was assembled to construct models for ligand-based screening. Five ACAT1-selective inhibitors (Figure S1) and six ACAT2-selective inhibitors (Figure S2), exhibiting both high activity and structural diversity, were carefully chosen as the training set compounds for constructing the pharmacophore hypotheses for ACAT1 and ACAT2 selective inhibitors. The HipHop algorithm implemented in the Discovery Studio (DS) was employed for this purpose. A set of 12 selective inhibitors targeting ACAT1, along with their corresponding 105 decoy molecules, were carefully chosen as the test set for the respective pharmacophore. Similarly, a set of 19 selective inhibitors targeting ACAT2, along with their corresponding 179 decoy molecules, was selected as the test set for the corresponding pharmacophore.

Molecular dynamics simulation

All the molecular dynamics simulations were performed using Gromacs2020.6 software³³. AmberTools22 was employed to incorporate the GAFF2 force field³⁴ into small molecules to facilitate small-molecule preprocessing. The simulation conditions were a static temperature of 300 K and atmospheric pressure (1 bar). The Amber14sb force field³⁵ was employed with water molecules modelled as the solvent (using the Tip3p water model)³⁶. The overall charge of the simulation system was neutralised by introducing a suitable quantity of Na⁺ or Cl⁻ ions. The simulation system employed the steepest descent method to minimise the energy, followed by conducting 100,000 steps of isothermal isovolumic ensemble (NVT) equilibrium and isothermal isobaric ensemble (NPT) equilibrium, using a coupling constant of 0.1 ps and a total duration of 100 ps. Finally, a free molecular dynamics simulation was conducted. The experiment consisted of 50,000,000 individual steps, with each step spanning 2 fs. Consequently, the entire process ultimately spanned a total duration of 100 ns. Subsequently, the root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), solvent-accessible surface area (SASA), principal component analysis (PCA), and number of hydrogen bonds were computed and determined. We conducted binding free energy calculations using the molecular mechanics Poisson–Boltzmann surface area (MMPBSA) approach between the ligand and protein, employing the g_mmpbsa tool³⁷.

Biological evaluations

Zhan et al. have proposed a comprehensive schema for ascertaining the selectivity of compounds in the inhibition of ACAT2²⁹. In briefly, the inhibitory effect on ACAT2 is initially evaluated through the NBD22-steryl ester fluorescence assay. Subsequently, the cumulative inhibitory impact on ACAT, encompassing both ACAT1 and ACAT2, is quantified utilising the cholesterol oxidase assay. Ultimately, the integration of compound-mediated inhibition data pertaining to both ACAT and ACAT2 facilitates the derivation of the selective inhibitory profile of the compound against ACAT2. We have used the NBD22-steryl ester fluorescence assay and the cholesterol oxidase assay to conduct biological evaluations.

The fluorescence-labeled sterol method was used to determine the inhibitory activity of the test compounds against ACAT2. HepG2 cells were cultured in a 96-well plate at an initial density of 1.5 × 10⁴ cells per well overnight. Following medium replacement, the cells were incubated for 6 h with a final concentration of 0.5 μg/mL NBD-22-labeled cholesterol and a final concentration of 25 μM of the test compound. Fluorescence analysis was performed at the excitation and emission wavelengths of 488 and 535 nm, respectively, using a fluorescence spectrophotometer (Agilent BioTek Synergy H1 Multimode Reader). Fluorescence intensity (FI) was measured and recorded as FI_SC (FI value of the compound group). The FI_bg (blank FI value) was obtained by subtracting the FI value of the cell-free medium from the FI value of the medium containing cells to adjust the baseline. In the test compound group, the test compound was replaced with DMSO, and the corresponding FI was recorded as FI_NI (FI value of the negative control group). Similarly, replacement with 5 μM PPPA in the test compound group resulted in the FI_PI (FI value of the positive control group). The inhibitory rates of test compounds against ACAT2 were determined using the following formula:

$$\text{Inhibition}\%=\left({\text{FI}}_{\text{NI}}\right)-{\text{FI}}_{\text{SC}})/\left({\text{FI}}_{\text{NI}}-{\text{FI}}_{\text{PI}}\right)\times 100\%.$$

A cholesterol oxidase assay was used to determine the overall inhibition rate of the test compounds against ACAT. HepG2 cells were cultured in a 6-well plate at an initial density of 4 × 10⁵ cells per well and incubated overnight. Following medium replacement, cells were further cultured for 6 h with the addition of cholesterol at a final concentration of 10 μg/mL and the test compound at a final concentration of 25 μM. Cell lipids were extracted using the Folch method, and cellular cholesterol levels (SE) were quantified using a Red Cholesterol Assay Kit(Amplex®). The cell protein content was determined using a BCA protein concentration assay kit, and SE values were standardised and recorded as SE_SC (the SE value of the test compound group). The test compound was replaced with DMSO in the SC group, and the corresponding SE was recorded as SE_NI (SE value of the negative control group). Similarly, a mixture of 3 μM K604 and 5 μM PPPA was used to replace the test compound in the SC group, and the resulting SE was recorded as SE_TI (SE value of the positive control group). The ACAT inhibition rates (%) of the test compounds against ACAT were calculated using the following formula:

$$\text{Inhibition }\%=\left({\text{SE}}_{\text{NI}}–{\text{SE}}_{\text{SC}}\right)/\left({\text{SE}}_{\text{NI}}–{\text{SE}}_{\text{TI}}\right)\times \text{1}00\%.$$

ACAT1 and ACAT2 each contribute 50% to the catalysis of cholesterol in cellular lipids.²⁹ Consequently, based on the inhibition rates of the test compound against ACAT2 and total ACAT at the same concentration, the selectivity of the test compound at that concentration can be calculated. The formula for calculating the Selectivity is as follows:

$$\begin{array}{c}\text{Selectivity}=\text{\hspace{0.17em}Inhibition\hspace{0.17em}}{\text{Rates}}_{\text{ACAT}2}\\ \hspace{0.17em}\times 50\%/(\text{Inhibition\hspace{0.17em}}{\text{Rates}}_{\text{ACAT}}-\text{Inhibition\hspace{0.17em}}{\text{Rates}}_{\text{ACAT}2}\times 50\%)\end{array}$$

Results and discussion

Deep Docking of Top Science database against ACAT2

Our study employed the Deep Docking method to computationally analyse 12 million entries from the Top Science database against the active site of ACAT2. The accuracy of the software in reproducing the conformations of the cocrystallized ligands was evaluated by calculating the RMSD values between the redocked and original poses. The smaller the RMSD value, the greater the similarity between the redocked and original cocrystallized poses. The RMSD results for Glide SP, and XP are 1.178Å and 0.877Å, respectively. Furthermore, we validated the docking procedure by employing the Directory of Useful Decoys-Enhanced (DUD-E)³⁸ benchmark dataset, affirming its efficacy and reliability. Briefly, 25 active ACAT2 inhibitors and 1250 decoys were obtained from DUD-E. The molecules mentioned earlier were docked onto the active site of ACAT2, employing Glide’s SP and XP protocols (with identical docking parameter settings, as detailed in the methodology section). Receiver operating characteristic (ROC) curves were generated for the SP and XP docking methods utilising the docking scores of the active molecules and decoys, respectively. As shown in Figure 2, the areas under the ROC curve (AUC) for SP and XP were 0.782 and 0.837, respectively.

Figure 2.

The ROC curve for efficiency validation of molecular docking.

The deep learning process encompassed 11 iterations, and the optimal model performance for each iteration is presented in Table 1. Generally, the ROC-AUC values for all iterations exceeded 0.8, indicating that the model of each iteration demonstrated a strong predictive capability for the docking scores of molecules in the database while effectively maintaining a low false positive rate. Following a series of 11 iterations, it was observed that the ROC-AUC values reached a state of convergence, thereby selecting the model of the eleventh round as our definitive model. The iterative training process gradually enhanced the full predicted database enrichment (FPDE) outcomes, with the recovery rate of hits in the test set displaying irregular patterns. Nevertheless, the values consistently surpassed the 80% threshold across all iterations.

Table 1.

Deep docking evaluation metrics of the optimal model performance for 11 iterations.

Iteration	Best model no.	Cut-off value	Recall	AUC	FPDE	Virtual hits
1	8	−8.4334	0.9079	0.8016	1.7035	6480029.1440
2	4	−8.5059	0.9038	0.8269	1.8578	3225684.3490
3	4	−8.5680	0.8897	0.8489	2.0111	2607179.9450
4	8	−8.6633	0.9000	0.8497	2.1000	2290100.1430
5	1	−8.7262	0.9052	0.8596	2.1921	2143482.5330
6	4	−8.8253	0.9194	0.8596	2.2526	2085496.5990
7	2	−8.9522	0.9027	0.8781	2.6754	1669649.3910
8	10	−9.0938	0.8679	0.8682	2.9448	1216780.4890
9	3	−9.3241	0.8318	0.8874	3.5665	856728.0216
10	6	−9.5929	0.7931	0.9097	7.3850	308650.6223
11	5	−10.7943	1.0000	0.9867	16.9762	78611.9832

This observation showed a comparable predefined recovery rate in the validation set, thereby implying the potential extension of all models to an external dataset. Finally, the model built in the eleventh iteration was selected as our final model for the virtual screening of ACAT2 antagonists owing to its balanced performance in all aspects. To discover novel ACAT2 antagonists, a virtual screening was conducted on the Top Science database library using the aforementioned Deep Docking model. The predicted probability of docking scores was used to rank compounds in the database. Subsequently, about twenty thousand molecules were subjected to further screening using the pharmacophore method.

ACAT2 selective pharmacophore evaluation

The top 10 pharmacophore models for ACAT2 selective inhibitors were computed using an eight-compound HipHop run. The selection of compounds for pharmacophore generation was driven by IC50 value, and structural divergence. Selective pharmacophore models for the two ACAT subtypes were validated using their respective test sets. Compounds exhibiting complete conformity with the pharmacophore were identified as hits in the test set. The results are presented in Tables 2 and 3. Based on selectivity and specificity criteria, among the 10 hypotheses for ACAT1, Hypo3 is the optimal model, while for ACAT2, Hypo1 was the optimal model within the set of 10 hypotheses.

Table 2.

The validation of 10 pharmacophore models for ACAT1.

Pharmacophore	Total actives	Total inactives	True positives	True negatives	False positives	False negatives	Sensitivity	Specificity
1	12	105	12	83	22	0	1	0.79048
2	12	105	12	83	22	0	1	0.79048
3	12	105	12	84	21	0	1	0.80000
4	12	105	12	83	22	0	1	0.79048
5	12	105	12	83	22	0	1	0.79048
6	12	105	11	79	26	1	0.91667	0.75238
7	12	105	12	79	26	0	1	0.75238
8	12	105	12	76	29	0	1	0.72381
9	12	105	11	79	26	1	0.91667	0.75238
10	12	105	12	75	30	0	1	0.71429

Table 3.

The validation of 10 pharmacophore models for ACAT2.

Pharmacophore	Total actives	Total inactives	True positives	True negatives	False positives	False negatives	Sensitivity	Specificity
1	19	179	19	156	23	0	1	0.87151
2	19	179	19	150	29	0	1	0.83799
3	19	179	19	153	26	0	1	0.85475
4	19	179	19	150	29	0	1	0.83799
5	19	179	19	146	33	0	1	0.81564
6	19	179	11	161	18	8	0.57895	0.89944
7	19	179	11	167	12	8	0.57895	0.93296
8	19	179	19	115	64	0	1	0.64246
9	19	179	19	132	47	0	1	0.73743
10	19	179	11	160	19	8	0.57895	0.89385

Database screening was performed using the Ligand Pharmacophore Mapping protocol within the DS. Compounds that correlated with the ACAT1 selective pharmacophore were eliminated, while ACAT2 selective inhibitors were retained. Finally, 460 compounds were selected for Glide XP docking.

Identification of ACAT2 selective inhibitors (docking, MD, MMPBSA)

In this study, we combined deep docking, pharmacophore modelling, molecular docking and molecular dynamics simulations methods to identify potential selective inhibitors for ACAT2. The rapid identification of promising lead compounds was achieved using refined biological activity evaluation methodologies. Specifically, we utilised a set of advanced assays and computational techniques to assess the biological activity of the selected compounds, as outlined in the work by Zhan et al.²⁹ After ACAT2 selective pharmacophore evaluation, Glide XP docking was used to perform ACAT2-based virtual screening, resulting in the selection and retention of the top 150 small molecules based on their docking scores. In this study, 150 molecules were subjected to rigorous visual inspection, resulting in the final selection of 22 compounds (Table S1) with the selection criterias as follows: (1) compounds capable of forming discernible hydrogen bonds, particularly with Gln488 and His434²³, are considered favourable candidates; (2) Each compound underwent meticulous visual inspection and individual evaluation to assess its interactions with the receptor, efficacy of engagement with target residues, complementarity with the binding-site surface, ligand strain, distorted ligand geometry, and commercial availability (Figure 3 and Figure S3). To guarantee the presence of specific subtype selection activity within the screened molecules, the binding of the acquired compounds to the representative subtype ACAT1 was subsequently assessed via Glide XP docking. All 22 small molecules possessing a docking score in ACAT1 that surpassed that of ACAT2 were retained to enhance the preference for binding with ACAT2 among the screened compounds.

Figure 3.

Binding modes of the selective inhibitors with ACAT2. Arrows represent hydrogen bonding interactions.

To evaluate the target affinity of the screened molecules, molecular dynamics (MD) simulations were performed, and the binding free energies were calculated using the MM-PBSA (Molecular Mechanics Poisson-Boltzmann Surface Area) method. A 100 ns MD simulation was conducted for each system to provide detailed insights into the molecular interactions. The binding affinities of the selected compounds with the ACAT2 subtype were assessed using MM-PBSA, focusing on comparing the computed free energies between ACAT2 and ACAT1. Specifically, nine molecules with lower ACAT2 MM-PBSA values compared to ACAT1 were retained for further analysis. This approach was intended to identify molecules with a potentially higher affinity for ACAT2 over ACAT1. The stability of all nine systems throughout the simulation time supports the favourable binding interactions of these compounds. The MD simulations analysis results for STL565001 and STL528213 are discussed in detail in the section on elucidating the binding mechanisms, while the results for the other systems are provided in supplementary materials (Table S2, Figures S4–S12). Nine specific molecules (STL565001, STL565365, STL528029, STL528213, STL528094, STL528144, STL528164, STL553262 and STL527299) were procuredfor subsequent evaluation of biological activity (Figure 4).

Figure 4.

2D structure diagram of nine compounds.

Effect of hits on ACAT2 selectivity

Based on the assessment of binding free energy, we identified the most promising compounds for subsequent evaluation of enzyme inhibitory activities. These compounds were purchased from Topscience Co., Ltd. Ten compounds, STL565001, STL565365, STL528029, STL528213, STL528094, STL528144, STL528164, STL553262, STL527299 and PPPA, were tested for ACAT2 inhibition using NBD22-steryl ester fluorescence assay. This experiment can be used to show the inhibition activity of the ACAT2 subtype. Our results strongly indicated that STL565001, STL528029, STL528213, STL528094, STL528144, STL553262, and STL527299 exhibited significant inhibitory activity against the ACAT2 subtype, demonstrating promising inhibitory potential. Of all the compounds tested, STL528144, STL565001, STL528029, and STL528213 showed the most promising inhibitory activity against ACAT2. Specifically, PPPA is a selective ACAT2 inhibitor with an IC50 of 0.191 μM (Figure S13). STL528144 exhibited an inhibition rate of 92.5 ± 36.8%, STL565001 showed an inhibition rate of 75.7 ± 27.8%, STL528029 displayed an inhibition rate of 86.9 ± 4.80%, and STL528213 presented an inhibition rate of 87.8 ± 12.4% (Figure 5). To evaluate the selectivity of these compounds further, we conducted a cholesterol oxidase assay using the Amplex Red Cholesterol Assay Kit. However, the experimental process required a significant amount of compound consumption and the available quantities for the purchased molecules of STL528029 and STL528144 were insufficient for conducting the experiment. Therefore, in this study, only STL565001 and STL528213 were used for selective validation. In this analysis, STL565001 and STL528213 demonstrated significant selectivity towards the ACAT2 isoform, as evidenced by their respective selectivity indices of 6 and 13 (Table 4). Therefore, the molecular skeleton merits crucial deliberation for the subsequent development of selective ACAT2 inhibitors.

Figure 5.

Inhibition rate of ACAT2 for nine compounds at 25 μM.

Table 4.

The MM-PBSA values, inhibition rate, selectivity, and docking score of nine inhibitors.

Compounds	MM-PBSA values of ACAT1 (kcal/mol)	MM-PBSA values of ACAT2 (kcal/mol)	Inhibition rate of ACAT2 (%)	Inhibition rate of ACAT (%)	Selectivity	ACAT1 docking score	ACAT2 docking score
STL565001	−184.682 ± 1.269	−424.556 ± 1.561	75.7 ± 27.8	44.4 ± 11.8	6	−4.783	−11.972
STL565365	−417.501 ± 1.912	−623.186 ± 2.492	−0.418 ± 2.09	/	/	−7.019	−12.464
STL528029	−127.229 ± 1.508	−602.773 ± 2.804	86.9 ± 4.80	/	/	−10.393	−12.492
STL528213	−122.294 ± 1.570	−662.680 ± 1.946	87.8 ± 12.4	47.2 ± 21.2	13	−7.462	−12.347
STL528094	−171.826 ± 1.412	−672.047 ± 2.548	65.7 ± 6.90	/	/	−9.457	−11.885
STL528144	−145.499 ± 1.556	−416.354 ± 2.096	92.5 ± 36.8	/	/	−4.972	−12.058
STL553262	−153.685 ± 1.531	−197.349 ± 1.239	5.53 ± 18.6	/	/	−9.696	−12.048
STL527299	−201.267 ± 1.323	−236.529 ± 1.436	59.4 ± 45.7	/	/	−8.688	−13.349
STL528164	−447.244 ± 2.061	−456.159 ± 1.632	−6.17 ± 16.2	/	/	−7.705	−11.899

In silico evaluation of STL565001 and STL528213 by MD simulations

For compounds STL565001 and STL528213, exhibiting commendable enzyme inhibitory activities coupled with heightened selectivity towards the ACAT2 isoform, an in-depth investigation was undertaken into their molecular mechanisms of action. This exploration holds substantial significance for subsequent receptor-based drug design efforts. First, the docking results for the two molecules were analysed. As shown in Figure 3, the analysis of molecular interactions revealed that STL565001 and STL528213 establish hydrogen bonding interactions within the primary binding pocket, indicating their propensity to adopt a preferred conformation. Our analysis revealed similar binding interactions of STL565001 and STL528213 with the binding-site residue of ACAT2, as shown in Figure 3. Notably, these compounds demonstrated comparable interactions with Gln485, closely resembling those observed with the reference ACAT2 inhibitor, PPPA (Figure 6). In the ACAT2 binding pocket, STL565001 formed four hydrogen bonds with residues Trp394, Ser430, His434, and Gln485 (Figure 3). Similarly, STL528213 exhibited a more extensive interaction pattern and engaged in seven hydrogen bonds with Tyr390, Tyr391, Ser430, His434, Gln485, and Gln488 (Figure 3).

Figure 6.

Binding mode of the PPPA with ACAT2. The interaction was shown using PLIP webserver (https://plip-tool.biotec.tu-dresden.de/plip-web/plip/index)³⁹. The blue lines represent hydrogen bonding interactions. The orange dashed line represents hydrophobic interactions.

The conformational stabilities of the two molecules were assessed by MD simulations. Various analyses were conducted to evaluate MD simulations. The stability of each system and the fluctuations in protein residues upon ligand binding were assessed using RMSD and RMSF analyses. The RMSD values of the four ACAT2 systems exhibit a gradual initial increase during the simulations (Figure 7), followed by stabilisation at approximately 25 ns (7N6Q), 35 ns (compound STL565001), and 40 ns (compound STL528213, apo).

Figure 7.

Root mean square deviation (RMSD) plots for apo, ACAT2-PPPA, ACAT2-STL565001 and ACAT2-STL528213.

The RMSF serves as an indicator of the dynamic flexibility of amino acid residues. All the investigated systems exhibited comparable trends in the RMSF, with the majority of RMSF values being below 0.2 nm. Moreover, a limited number of residues displayed heightened fluctuations exceeding 0.5 nm across all the studied systems. The loop regions exhibited pronounced fluctuations, whereas active site residues displayed comparatively lower fluctuation levels. The average RMSF per residue for the apo, STL565001, STL528213, and 7N6Q complexes were 0.177 ± 0.99, 0.162 ± 0.09, 0.173 ± 0.117, and 0.197 ± 0.111 nm, respectively (Table S3). Notably, as shown in Figure 8, certain crucial residues within the binding pocket, such as Phe438, Gln488, and Val489, maintained low RMSF values, suggesting the stability of the binding pockets.

Figure 8.

The root mean square fluctuation (RMSF) of the protein structure plots for apo, ACAT2-PPPA, ACAT2-STL565001 and ACAT2-STL528213.

Rg is an informative parameter that offers valuable insights into the global conformation, stability, and folding characteristics of a given protein entity⁴⁰. The mean values of the Rg were determined for apo as 2.154 ± 0.019 nm, ACAT2-STL565001 as 2.175 ± 0.015 nm, ACAT2-STL528213 as 2.171 ± 0.013 nm, and the 7N6Q complex as 2.163 ± 0.016 nm (Table S3). The Rg plot in Figure 9 reveals that the ACAT2 protein exhibits slightly increased Rg values upon binding to the target compounds STL565001 and STL528213, with average Rg values of 2.175 ± 0.015 nm and 2.171 ± 0.013 nm, respectively. These values are marginally higher than the Rg observed for the positive control PPPA (2.163 ± 0.016 nm) and the apo structure (2.154 ± 0.019 nm). This subtle increase in Rg suggests a minor expansion of the ACAT2 molecular structure, likely due to the accommodation of these compounds within the protein’s binding pocket. However, the relatively small difference in Rg indicates that the overall packing of ACAT2 remains largely intact, with only slight conformational adjustments upon ligand binding.

Figure 9.

The gyration radius (Rg) of the protein structure plots for apo, ACAT2-PPPA, ACAT2-STL565001 and ACAT2-STL528213.

The SASA, which signifies the region directly engaged in interactions with the surrounding solvent, was calculated for ACAT2 and the complexes ACAT2-STL565001, ACAT2-STL528213, and 7N6Q during 100 ns MD simulations. The mean SASA values for STL565001, STL528213, and 7N6Q, as well as the complexes with free ACAT2, were 195.222 ± 5.344 nm², 200.097 ± 4.399 nm², 190.91 ± 5.021 nm², and 192.589 ± 5.611 nm² (Table S3), respectively. Discernible variations in the SASA patterns were discerned within the inhibited complexes, in contrast to those observed in the apo state (Figure 10). These observations substantiate the conformational changes in protein structure induced by the binding of small-molecule inhibitors.

Figure 10.

The solvent-accessible surface area (SASA) of the protein structure plots for apo, ACAT2-PPPA, ACAT2-STL565001 and ACAT2-STL528213.

Hydrogen bonding is regarded as a pivotal interaction in molecular recognition and plays a crucial role in the interactions between proteins and ligands. The hydrogen bonds in the complexes of ACAT2-STL565001, ACAT2-STL528213, and 7N6Q were computed over a 100 ns period through MD simulations (Figure 11). As shown in Figure 11, PPPA, STL565501, and STL528213 exhibited the ability to engage in a maximum of 3 − 4, 6 − 7, and 6 − 7 hydrogen bonds, respectively, with the active site residues of ACAT2.

Figure 11.

The number of hydrogen bonds plot vs time for the ACAT2-PPPA, ACAT2-STL565001 and ACAT2-STL528213.

Principal component analysis (PCA) was employed to elucidate the large-scale conformational motions, or “essential dynamics,” of the ACAT2 protein as captured in the MD trajectories. Through the decomposition of motion into orthogonal principal components (PCs), the first 10 eigenvectors (Figure 12) were identified, representing the most substantial conformational shifts within the protein. These principal components effectively distil the intricate dynamical behaviour of ACAT2, revealing the critical motions that underpin its functional states.

Figure 12.

Fraction motions of the first ten eigenvectors for apo, ACAT2-PPPA, ACAT2-STL565001 and ACAT2-STL528213.

By projecting PC1 against PC2, we delineated the conformational landscape of ACAT2 in both its apo form and when complexed with various inhibitors. In the apo configuration, the principal motion encompassed 33.03% of the total variance, while the holo structure (7N6Q) demonstrated a slightly higher variance of 33.29%. The inhibitor-bound systems, particularly those complexed with STL565001 and STL528213, exhibited distinct dynamic profiles, with dominant motions accounting for 31.48% and 46.46% of the variance, respectively (Figure 12). These observations suggest that inhibitor binding induces pronounced conformational changes, particularly with STL528213, which significantly perturbs the protein’s dynamic equilibrium. The analysis of these essential dynamics substantiates the notion that ligand engagement within the ACAT2 active site precipitates substantial structural rearrangements, which are integral to its inhibitory mechanism (Figure 13).

Figure 13.

The first two principal components (PCs) define the projection of conformational distribution on subspaces. The principal component analysis (PCA) of ACAT2-STL565001 and ACAT2-STL528213, ACAT2-PPPA, apo.

Based on the free energy landscape (FEL) graph obtained by performing PCA on diverse systems, we identified unique conformational landscapes pertaining to multiple ACAT2-inhibitor complexes. This encompasses the ACAT2-PPPA complex (PDB: 7N6Q) as a benchmark complex, two carefully chosen ACAT2-inhibitor complexes, and the apo-ACAT2 structure. The energy-landscape plots shown in Figure 14 were derived from the principal components, PC1 and PC2, obtained through the analysis. High-energy states on the FEL plot are indicated in red, whereas stable-energy states are represented in blue. In the ACAT2 system, the apo-ACAT2 system demonstrated two energy barriers after stabilisation without any additional constraints on the energy distribution. The energy profiles of the inhibited complexes deviated significantly from the unbound state of ACAT2, consistently maintaining elevated levels throughout the simulation.

Figure 14.

PC1 and PC2 Gibbs free energy landscape for ACAT2-STL565001, ACAT2-STL528213, apo, and ACAT2-PPPA.

In this study, the binding energy values were obtained from the conformations generated through MD simulations spanning 90 − 100 ns. The binding energy (ΔGbind) was computed utilising the van der Waals energy (ΔGvdW), electrostatic energy (ΔGele), polar solvation energy (ΔGpol), and SASA energy (ΔGnonpol). As shown in Table 5, the mean binding-free energy of STL565501 and STL528213 was −424.556 ± 1.561 and −662.680 ± 1.946 kJ/mol, respectively. In the case of PPPA, the calculated average binding free energy value amounted to −160.161 ± 1.337 kJ/mol. Additionally, to delineate the impact of crucial residues on the substantial augmentation of binding affinity, a rigorous investigation encompassing comprehensive residue contribution and energy decomposition analyses was performed on a set of three systems (Figure 15). Based on the interaction maps, residues Arg238, Lys100, Lys242, Arg392, Arg379, Arg107, Arg248, Arg283, Arg419, and Arg412 participated in all ACAT2-ligand interactions (Figure 15).

Table 5.

MM-PBSA calculations of ACAT2-STL565001, ACAT2-STL528213, and ACAT2-PPPA.

Complex	MM-PBSA calculations (all units kcal/mol)
	van der Waal energy	Electrostattic energy	Polar solvation energy	SASA energy	Binding energy
STL565001	−263.826 ± 1.285	−367.445 ± 2.607	235.739 ± 2.636	−29.131 ± 0.104	−424.556 ± 1.561
STL528213	−246.179 ± 1.390	−756.494 ± 2.660	367.928 ± 2.179	−28.015 ± 0.089	−662.680 ± 1.946
PPPA	−282.379 ± 1.489	−55.958 ± 0.765	207.185 ± 0.959	−28.926 ± 0.089	−160.161 ± 1.337

Figure 15.

Residual decomposition analysis of ACAT2-STL565001 and ACAT2-STL528213, ACAT2-PPPA.

Conclusions

In summary, a multifaceted screening strategy was devised to ensure that the screened molecules exhibited the requisite pharmacophoric attributes while concurrently possessing discernible druggability and marked affinity for the target. In this study, we employed our Deep Docking platform in conjunction with two pharmacophore models and diverse consensus-filtering methodologies to conduct a virtual screening of the Top Science database against the active site of ACAT2. In addition, this investigation integrated biological experiments to directly validate the biological activity of the screened molecules, identifying two compounds (STL565001 and STL528213) as ACAT2 inhibitors with selective inhibitory activities, with their identification being partly influenced by considerations of purchase availability and other practical factors. Furthermore, the elucidation of the binding mechanisms of these two molecules in ACAT2 was expounded, providing invaluable insights for subsequent drug design rooted in the understanding of their mechanisms of action. Taken together, the current investigation fused in silico methodologies with biological experiments to pinpoint potential hit compounds that could serve as foundational structures for developing markedly potent and selective ACAT2 inhibitors.

Funding Statement

This study was supported by National Natural Science Foundation of China (No. 82305010 and No. 82374262), the Natural Science Foundation of Nanjing University of Chinese Medicine (Grant No. XZR2021035), Advanced Training Program for Leading Personnel in Traditional Chinese Medicine in Jiangsu Province (Jiangsu Traditional Chinese Medicine Science and Education [2022] no.17), Medical Research Project of Jiangsu Province Health Commission in 2023 (Grant No. K2023056 and H2023084), Project of Nanjing Lishui District Hospital of Traditional Chinese Medicine (Grant No. LZY202202).

Disclosure statement

The data that support the findings of this study are available in Topscience Database at https://www.tsbiochem.com/service/topscience-database from Topscience Co.,Ltd.

Author contributions

Yanfeng Liu and Feng Ding contributed to methodology, investigation, conducting the experiment, visualisation and writing. Liangying Deng contributed to Investigation, experimenting, visualisation and writing. Shuran Zhang contributed to visualisation. Lixing Wu and Huangjin Tong contributed to project, resources, software and supervision.

Data availability statement

All data generated or analysed during this study are included in this published article.