Exploring the inhibitory mechanisms of indazole compounds against SAH/MTAN-mediated quorum sensing utilizing QSAR and docking

Activity data

The biological activity data consist of 40 indazole compounds (Tab. I) designed, synthesized, and tested by Li et al (9). These compounds evaluated the biological inhibitory effect of taking SAH/MTAN enzyme expressed on the full-length E. coli pfs gene having high conservation on the bacterial species. The SAH/MTAN is the key target for the production of AHL-mediated AI-1 and SAH-mediated AI-2, which are the building blocks for the synthesis of the QS AIs (9). The enzyme inhibitory activities have been measured in terms of Ki measuring the affinity of the compound to bind the active cavity of SAH/MTAN. A negative logarithm of these Ki values (pKi) has been done for data reduction and taken as a dependent variable whereas molecular structural descriptor has been computed as independent variables for these compounds.

* Test compounds.

Structure computation and molecular optimization

The two-dimensional structures of 40 indazole compounds were drawn using Chemdraw 8.0 software. These structures were transformed into 3D files, which are minimized using the MM2 force field. The molecular energy minimization is carried out taking the convergence criterion and dielectric constant of 0.01 Kcal/mol and 1.0 respectively (10) utilizing the Chem3D Ultra window (11).

Molecular descriptors computation

All the three-dimensional mol files were incorporated into PaDEL Descriptor Computation software for the computation of structural descriptors. PaDEL is a freeware for the calculation of molecular properties (12). A total number of 1,875 2D and 3D descriptors were calculated and before the model generation, this descriptor data had been reduced to 1,055 (Tab. II). Perfectly constant and highly intercorrelated descriptors were removed taking variance and correlation coefficient cut-off values of 0.0001 and 0.9 using the V-WSP algorithm of NanoBridges software (13). The reduced descriptor data set has been used for the development of QSAR models.

QSAR model generation followed by validation

A number of QSAR models have been generated for the deliberated indazole compounds utilizing various sets of a combination of 2D and 3D descriptors using genetic algorithm coupled multiple linear regression (GA-MLR) methods (14) based on the theory of mutation and crossover of the parents’ genes to generate the new solutions, taking the most appropriate transformations of the independent variables incorporated in the NanoBridges software (15,16). A population of 100 different random combinations of the structural descriptors is generated taking default parameters as set in the NanoBridges software (17). The impact of these indazole compounds’ computed descriptors on SAH/MTAN inhibitory activities has been shown through QSAR model development by considering each parent combination of descriptors for the entire data set using MLR. The entire data set was divided randomly into test and training sets before QSAR modeling. The developed models were validated statistically. The validation parameters are denoted by R² (R is the square root of multiple R-square for regression), Q² (cross-validated r²) values for the training set, whereas external validation was carried out by calculating predictive R² (R²_pred) and the standard error of estimation (SEE) represents standard deviation measured by the error mean square, which expresses the variation of the residuals or the variation about the regression line (18). Further, the external predictability of the generated QSAR models was scrutinized by calculating modified r² (r²_m), average modified r² (r_m²), and delta modified r² (∆r²_m) respectively (19). The best training model is composed of 72.5% and the test set consists of 27.5% of the total data. The test compounds have been marked in Table I with an asterisk.

Ligand docking

All optimized ligands were docked into the receptor active cavity using molecular docking, which is a powerful structure-based drug discovery simulation for the identification of ligand-receptor complexes having minimal interaction energy. The energy of interactions between ligand and protein was calculated in terms of the score, which can predict the affinity of the compound toward active binding (20,21). The crystal structure of E. coli SAH/MTAN (PDB ID: 1JYS) in complex with adenine cocrystal was selected as a receptor for in silico molecular docking studies (22).

The protein was downloaded and prepared by removing water molecules, and hydrogen atoms in the H-depleted target molecule were added. Grid points were generated surrounding the cocrystallized ligand bound with the active cavity of the target. This cocrystallized molecule is considered a reference to make the binding site for the ligand X-ray group. A flexible docking module was incorporated in ArgusLab 4.0.1, which is a very powerful docking simulation freeware (23,24,25). In the present docking simulation, the ligand is freely rotated inside the target cavity to generate multiple 150 conformers that can produce many docked complex poses considering grid resolution (angle) of 0.4 degrees as the default value. The term pose usually designates the specific set of coordinates of a docked ligand. The coordinates of conformation will change concerning this docked pose. The ligand is docked inside the target of the active site, which is well-kept within the grid box. The best complex pose with minimal interaction energy has been taken into consideration for a better explanation of the mode of interaction between the ligand and active amino acid residues of the receptor protein (26).

QSAR modeling

In the present study, QSAR modeling of indazole compounds having inhibitory activities against SAH/MTAN-mediated QS has been carried out utilizing a different combination of 2D and 3D structural descriptors. The impact of the different classes of computed descriptors on SAH/MTAN inhibitory activities of these compounds has been discussed by the development of optimal training QSAR model formulated by 72.5% of total data and the remaining 27.5% of total data is used as a test set marked by an asterisk. These model parameters have been expressed as R² and Q² (cross-validated r²) values for the training set while the external model validation significance is carried out by calculating predictive R² (R²_pred), the SEE and modified r² (r²_m) given in Table III.

It was shown that Equation [1] can produce an explained variance of 71.4% and an internal predicted variance of 55.6% of the observed data. For a predictive QSAR model, the value of R² should be more than 0.6 (27). The external model validation parameters such as R²_pred, r²_m, and SEE are given as 0.151, 0.112, and 0.660, which do not produce significant predictability because R²_pred and r²_m must be greater than 0.5, while the SEE values should be less than 0.5 to have a significant model (28).

Therefore, outlier analyses have been carried out by testing of applicability domain of the training QSAR model, which determines its acceptance as per the Organization for Economic Cooperation and Development (OECD) incorporated in NanoBridges software (29). The training molecules 14 and 18 were detected as an outlier and again best QSAR model (2) was modeled by deleting the outliers showing the best result on the SAH/MTAN inhibition. The developed QSAR model (2) can explain and predict 85.2% and 78.1% of variances of the SAH/MTAN inhibitory activity of the deliberated compounds. This model can also produce 68.5% external predictability and r²_m (test) and SEE values of 0.636 and 0.490 respectively. This model is quite acceptable as per statistical validation. The square correlation coefficient between observed activities vs. predicted activities of the test compounds obtained from the correlation plot (Fig. 1) is calculated as 0.751, which suggests good model predictivity.

The model (2) parameters such as RDF55m, E1s, and AATSC7s have a positive impact on aromaticity toward SAH/MTAN inhibition, whereas the decrease in value of AATS1v and ATSC3s of the deliberated indazoles may increase the enzyme inhibition to stop the QS.

Indazole-SAH/MTAN docking

The resultant binding affinity along with details of amino acid residues bound with studied 40 indazole compounds toward inhibition of SAH/MTAN target is shown in Table IV. When ligands of interest are docked inside the defined target cavity of SAH/MTAN, both the conformational changes of ligand and receptor occur to make a number of ligand-receptor complexes. These ligand-receptor complexes represent the conformational rotation of the ligand inside the cavity. The ligand-receptor complex having a minimal dock score may produce maximum binding affinity toward the cavity of the target. The docking of indazole compounds having inhibition of SAH/MTAN resulted in common amino acid interaction inside the SAH/MTAN cavity for all compounds. These common amino acid residues such as ASP197, PHE151, GLU172, ILE152, and MET173 (highlighted in Tab. IV) are essential for ligand-protein interaction and help in unique conformational changing of ligand-receptor interaction complex, which is responsible for producing biological activity. If these amino acids are removed from the cavity then the binding affinity of a compound may be reduced. So these common amino acids play an important role in the suitable binding of the ligand toward the active site.

For a better interpretation of the binding mode of the indazole ligands, molecules are categorized into three highly active, intermediate active, and lower active by considering their biological activities (pKi) range as 1.50 to 2.79 µm, 0.44 to 1.4 µm, and <0.4 µm respectively. The highly active compounds such as 20, 22, 23, 26, 30-34, and 39 have more interactions with ALA150, PHE335, and VAL171 along with common amino acids bearing –14.477 to –11.191 kcal/mol dock score. The intermediate active compounds 24, 25, 27, 28, 35, 36, and 40 have –11.186 to –10.367 kcal/mol dock score with a lack of interactions with ALA150 except compound 28 which does not capture both ALA150 and PHE335. The remaining compounds are 1-18, 37, and 38 with lower active range having dock score greater than –10 kcal/mol, and these compounds lack either ALA150 and PHE335 or VAL171 amino acid interactions.

From Table IV, we can say that the binding affinity of compounds depends on a number of residues and minimal dock score involved in the ligand-receptor interactions. The highest active compound 21 shows the following pattern of interactions (Fig. 2).

The amino group of sulfonyl amino linkage interacts with VAL171 and sulfonyl group interacts with GLU172, ILE152, and PHE151 by hydrogen bonding. ILE152 also interacts with isobutyl group by the same bonding. Chlorine atom at position-3 of 3,4-dichlorophenyl interacts with PHE335, TYR337, and PRO343, and VAL332 along with PRO343 interacts with 3,4-dichlorophenyl ring by hydrophobic bonding, whereas MET173 interacts with the sulfur bonding. Some amino acid residues such as ASP197, ALA150, ALA77, and GLY78 are also present at the binding site.