|Year : 2018 | Volume
| Issue : 1 | Page : 16-31
Pharmacophore Modeling and Database Mining to Identify Novel Lead Compounds Active Against the Disease Stage of Trypanosomiasis in the Central Nervous System
Kirtika Madan, Ankita N Verma, Sarvesh K Paliwal, Divya Yadav, Swapnil Sharma, Manu Sharma
Department of Pharmacy, Banasthali Vidyapith, Banasthali, Rajasthan, India
|Date of Web Publication||15-Jan-2018|
Department of Pharmacy, Banasthali Vidyapith, Banasthali - 304 022, Rajasthan
Source of Support: None, Conflict of Interest: None
| Abstract|| |
Introduction: Sleeping sickness has long been considered as a neglected disease, and very few pharmaceutical companies and research organizations are involved in the design and development of anti-trypanosomal drugs. This may be especially due to poor financial returns. Materials and Methods: In view of the dire need for new drugs for sleeping sickness, we have implemented in-silico ligand- and structure-based methods for the development of a universal pharmacophore model. The ligand-based pharmacophore models for 1,2-dihydroquinolin-6-ols and their ester derivatives were developed using Catalyst HypoGen refine algorithm. The best quantitative pharmacophore hypothesis was selected on the basis of correlation coefficient (0.92), root mean square deviation (0.97), and cost difference (76) values. The best pharmacophore model was compared with a structure-based model developed using the Protein Data Bank structure of trypanothione reductase (TR) bound to WPC inhibitor. Results and Discussion: High consistency between ligand- and structure-based models was observed, and both the approaches indicate that four-point interactions [three hydrophobic and one hydrogen bond acceptor (HBA)] are necessary for the anti-trypanosomal activity of 1,2-dihydroquinolin-6-ols. The pharmacophoric features obtained were in accordance with the binding requirement of TR binding site, indicating that these compounds can act as TR inhibitors. To further evaluate the model, an external test set comprising known trypanocidal agents were mapped on to a developed pharmacophoric model, which also showed four-point mapping and estimated values in close range to actual values. The screening of chemical database resulted in the identification of three druggable structurally diverse potent lead compounds. Conclusion: Since no pharmacophore model has been developed for this new series of compounds till date, the achieved results will allow researchers to further use this 3D pharmacophore model and hits for the design and synthesis of newer anti-trypanosomal compounds.
Keywords: Central nervous system, computer-assisted drug design, human african trypanosomiasis, quantitative structure–activity relationship, root mean square deviation, trypanothione reductase
|How to cite this article:|
Madan K, Verma AN, Paliwal SK, Yadav D, Sharma S, Sharma M. Pharmacophore Modeling and Database Mining to Identify Novel Lead Compounds Active Against the Disease Stage of Trypanosomiasis in the Central Nervous System. Int J Nutr Pharmacol Neurol Dis 2018;8:16-31
|How to cite this URL:|
Madan K, Verma AN, Paliwal SK, Yadav D, Sharma S, Sharma M. Pharmacophore Modeling and Database Mining to Identify Novel Lead Compounds Active Against the Disease Stage of Trypanosomiasis in the Central Nervous System. Int J Nutr Pharmacol Neurol Dis [serial online] 2018 [cited 2018 Feb 18];8:16-31. Available from: http://www.ijnpnd.com/text.asp?2018/8/1/16/223291
| Background|| |
Human African trypanosomiasis (HAT) or sleeping sickness is a critical disease caused by two different subspecies of Trypanosoma brucei (T. b. gambiense and T. b. rhodesiense), and becomes fatal if it remains untreated.,,,, Currently, HAT is prevalent in 36 countries especially in sub-Saharan Africa, and about 60 million people are struggling from this disease worldwide. Despite more than a century of study, there are no well-tolerated or effective drugs available for the treatment of HAT. Treatment of both the forms (early and late) of HAT has always been difficult, especially when the disease reached an advanced stage. The drugs currently available for the treatment of HAT show significant toxicity, low efficacy, and do require parenteral administration over a longer duration with compromised therapeutic efficacy. Since 1950, only one drug has been developed for the treatment of late-stage HAT.,, Recently, pentamidine (5-day regimen) and suramin are being used for early-stage disease, whereas melarsoprol is being used as the sole treatment option for late-stage HAT caused by T. b. rhodesiense. The emergence of resistance to these agents along with its adverse effects is becoming a major threat to scientists and clinicians involved in the treatment of HAT. Thus, the need for well-tolerated and affordable anti-trypanosomal therapies capable of overcoming parasite resistance makes the identification of new drug candidates an urgent priority.
Discovering and bringing one new drug to public use typically costs a pharmaceutical or biotechnology company nearly $900 million and takes an average of 10–12 years. The application of computer-assisted drug design (CADD) methodologies plays a critical role in reducing cost, time, and effort required to discover new medicines or to improve the efficacy of existing drugs. In view of this, this study applied the application of CADD to develop a new pharmacophore, which could help visualize the potential interaction between ligands and the target and could be used as a query in a 3D database search to identify new structural classes of potential lead compounds.,,
To combine the advantages of both approaches [ligand based drug design (LBDD) and structure-based drug design (SBDD)], which are very different in nature but similar in aims, we performed pharmacophore-based 3D quantitative structure–activity relationship (QSAR) and structure-based pharmacophore (SBP) modeling. The HypoGen module of discovery studio was used to generate ligand-based pharmacophore, which takes activity data into account and uses both active and inactive compounds in an attempt to identify the hypotheses that are common among the active compounds but not among the inactive compounds. The steric effect was also evaluated by applying excluded volume feature to the generated hypothesis, and this refined hypothesis was compared with the HypoGen hypothesis. The generated pharmacophore model was validated by internal test set prediction and Fischer’s randomization test. The pharmacophore model was further evaluated by using an external test set comprising known trypanocidal agents. Next, we generated a pharmacophore model through structure-based approach and then compared it with the 3D QSAR model.
| Results And Discussion|| |
Ligand-based pharmacophore generation
After the generation of conformers and the selection of relevant features through feature mapping, the next step was the generation of a hypothesis, which was attained by performing HypoGen algorithm and HypoGen refine algorithm calculations on the conformational model of 37 training set molecules. An ideal pharmacophore hypothesis was selected on the basis of cost difference (null-fixed), root mean square deviation (RMSD) values (root mean square values), correlation coefficient, weight, and configuration cost. The statistical values for the selected pharmacophoric model obtained by both the processes are listed in [Table 1] and [Table 2]. It is apparent from the statistical values that the automated refinement of the pharmacophore by the addition of excluded volumes to HypoGen algorithm generated a more selective and predictive pharmacophore having better statistical values than the HypoGen model.
Among the 10 hypotheses generated by HypoGen refine calculation, hypo1 was considered to be the best significant hypothesis because of low RMSD value (=0.97), high correlation (=0.92), and configuration cost <17. In the selected refined hypothesis, the null cost value of the top 10 hypotheses was 232.42, and all hypotheses had five features with two hydrogen bond acceptor (HBA) features and three hydrophobic features [Table 1]. The difference of 76 bits between the null cost and the total cost of hypo1 suggested that the model represents a true correlation and has more than 90% probability of correlating the data. Pharmacophoric features and corresponding weights, tolerances, and the 3D coordinates of HypoGen refined hypo1 are listed in [Table 3]. The selected hypo1 had two hydrogen bond acceptor features, three hydrophobic features, and 10 excluded volume features [Figure 1]. The pharmacophore model was color coded as follows: blue for hydrophobic, green for hydrogen bond acceptor, and gray for excluded volume feature.
|Table 3: Pharmacophoric features and corresponding weights, tolerances, and 3D coordinates of HypoGen refined hypo1|
Click here to view
|Figure 1: Top scoring pharmacophore model, that is, hypo1 of HypoGen refined hypothesis with 10 excluded volume. The pharmacophore features are color-coded with: light-blue = hydrophobic groups; green = hydrogen bond acceptor; magenta = hydrogen bond donar; grey = excluded volume|
Click here to view
Hypo1, which was identified as the best hypothesis, was then used to estimate the activity of the training set molecules. In this study, all compounds were classified by their activity as highly active (<0.1 μM, +++), moderately active (0.1–10 μM, ++), and inactive (>10 μM, +). [Table 4] represents the actual and estimated activities of the 37 training set molecules based on the best hypothesis, hypo1. Except two compounds, all the active training set compounds were predicted correctly. Out of twenty moderately active compounds, one of the compounds was overrated while one was underrated; three of the inactive compounds were estimated as moderately active. Consequently, for 30 of 37 compounds, the predicted activity values were to be within the same activity scale as the experimental values in the training set. The error value was calculated as the ratio between the predicted and experimental activities. A positive error value indicated that the predicted IC50 was higher than that which was obtained experimentally, whereas a negative error value indicated that the predicted IC50 was lower than that which was obtained experimentally. An error value of <10 represents a difference no >1 order between the predicted and experimental activities. Among the training set compounds, very few had an error value >4; hence, these values showed that the generated model was statistically significant. [Figure 2]a represents the correlation graph between the experimental and estimated activities of training set compounds.
|Table 4: The actual and estimated anti-trypanosomal activity (μM), error and activity scale of the 37 training set molecules, based on the best hypothesis hypo1|
Click here to view
|Figure 2: (a) Plot of actual versus estimated activity of training set compounds. (b) Plot of actual versus estimated activity of test set compounds|
Click here to view
Cross-validation of pharmacophore model
The Cat-scramble module in Catalyst software was used to validate the statistical relevance of hypo1, which was the preferred pharmacophore model, by the principle of Fisher’s randomization test. In this cross-validation test, 99% confidence level was selected, and thereby, 99 spreadsheets were generated. The 11 lowest correlation values resulting from the 99 hypotheses are listed in [Table 5]. The data of cross-validation clearly indicate that the statistics of hypo1 is better than other random hypotheses, as revealed by the lowest total cost and highest correlation coefficient, which verify that the hypo1 was not obtained by chance. A significance level of 99% for hypo1 further confirms that it is the best-ranked pharmacophore hypothesis.
Test set validation of pharmacophore model
To further validate hypo1, 15 molecules that were kept aside as test set molecules were used to determine the predictive ability of the generated pharmacophore hypothesis. The correlation coefficient (r) of 0.84 for the test set compounds by using hypo1 shows a good correlation between experimental and estimated activities. The actual and experimental activities for the test set compounds are listed in [Table 6]. Three out of the six active compounds were underestimated as moderately active; two moderately active compounds were underestimated as inactive, whereas one compound was overestimated; all the inactive compounds were overestimated as moderately active. Among the 15 test set compounds, 14 compounds had an error value <10, representing a not more than one order difference between estimated and experimental activities. Thus, hypo1 showed good predictive ability and, therefore, is a convincing model. The correlation graph between the experimental and estimated activities of test set is shown in [Figure 2]b.
|Table 6: Actual and estimated anti-trypanosomal activity (μM) of test set molecules based on pharmacophore model hypo1|
Click here to view
Structure-based pharmacophore generation
To test the soundness and performance of the ligand-based model, we compared it with a structure-based model developed from trypanothione reductase (TR) enzyme complex with an inhibitor., Structure-based methods rely exclusively on the prior knowledge of a protein structure to derive novel ligands, while ligand-based methods are traditionally used when no protein structure is available. However, when sufficient information is available, then both ligand-based and structure-based drug design methods can be used in conjunction to increase the accuracy of simulation and enhance the drug design process. Hence, the available 3D structure of TR-WPC complex [from Protein Data Bank (PDB)] was used to construct a SBP model to gain insight into the binding requirement for anti-trypanosomal activity.
Initially, seven features pharmacophoric models (2HBA, 2HBD, and three hydrophobic features) were generated by direct approach using the PDB entry. This SBP, which takes into account all possible interactions between a ligand and a binding pocket of the enzyme, was then used for screening the compound library of 52 molecules including different conformations of compounds 1,2-dihydroquinolin-6-ols and their ester derivatives. The hits obtained as a result of screening presented the chemical features and the shape suggested by the SBP model.
Mapping analysis of ligand- and structure-based pharmacophore
The most important feature of a validated pharmacophore hypothesis is its ability to provide valuable information regarding the physicochemical environment of target-binding site and the compatibility between this environment and functionality present on drug candidates. Therefore, after obtaining a validated predictive pharmacophoric model, the next step was the mapping of the most active and inactive compounds of the series under investigation onto the best hypothesis. The mapping results provide information about the most important features that account for the activity of the selected series of compound. The most active compounds, 9a and 10a, of the training set were mapped with the best-ranked pharmacophore hypothesis, that is hypo1, as shown in [Figure 3]a and [Figure 3]b. The results of pharmprint from the mapping studies of training set compounds are given in [Table 7]. The most active compound, 9a (IC50 = 0.007 μM), of the series mapped one hydrogen bond acceptor feature and three hydrophobic features, while the ester compound, 10a (IC50 = 0.014 μM), mapped two hydrogen bond acceptor features and three hydrophobic features. Functional group contribution analysis revealed that the benzyl ring mapped one of the hydrophobic feature, the methyl group at two positions of 1,2-dihydroquinoline mapped the second hydrophobic feature, the dihydroquinoline nucleus mapped the third hydrophobic feature, one of the hydrogen bond acceptor feature mapped the hydroxyl oxygen at the position 6 of 1,2-dihydroquinoline, and the second hydrogen bond acceptor feature mapped the acetyl oxygen attached to the hydroxyl group at position 6 in compound 10a.
|Figure 3: (a) Mapping of active compound 9a on ligand-based pharmacophore. (b) Mapping of compound 10a on ligand-based pharmacophore model|
Click here to view
The mapping of inactive compound 2a [[Figure 4]a] with hypo1 indicated that quinoline compounds are almost inactive against the trypanosomes in comparison to 1,2-dihydroquinolines, because these compounds missed one hydrophobic feature and two hydrogen bond acceptor features. In addition to this, the least active compounds belonging to 1,2-dihydroquinolin-6-ols series were also mapped with hypo1 [[Figure 4]b]. The mapping of compound 8e showed that this compound missed two hydrogen bond acceptor features.
|Figure 4: (a) Mapping of inactive quinoline compounds 2a with hypo1. (b) Mapping of 1,2-dihydroquinoline compound 8e with hypo1|
Click here to view
Further, the test set molecules were also mapped on the generated hypothesis. The results of pharmprint from the mapping studies of test set compounds are given in [Table 8]. The most potent compounds 12a and 9f in the test set were selected for the mapping on hypo1 [[Figure 5]a and [Figure 5]b]. Compound 9f missed one acceptor feature, while 12a mapped all the features. Mapping results showed that the active and inactive compounds did not map any excluded volume feature, which means that the present compounds do not impinge sterically on the enzyme-binding site.
|Figure 5: (a) Mapping of active compound 9f of test set. (b) Mapping of test set compound 12a|
Click here to view
While analyzing the mapping results of 1,2-dihydroquinolin-6-ols and their ester derivatives on the pharmacophore model derived from 3D QSAR analysis, we found that most of the 1,2-dihydroquinoline compounds mapped four features (one HBA and three hydrophobic features) while the ester derivatives mapped five features (two HBA and three hydrophobic features), as evident from the pharmacophore mapping of 9a and 9f and the ester analogues 10a and 12a. The usual interpretation of the above results will be that the activity of the most active compound 9a is underestimated by the model due to four features mapping in the place of five features suggested by the model. On the other hand, the activity of ester analogue 10a is correctly predicted due to five features mapping. The purpose of the formation of ester derivatives was to protect the unstable oxygen from auto-oxidation, so the hydroxyl group in the compounds 10a–o, 12a–e, and 13a–e of the series was protected through esterification. However, when ester derivatives will be subjected to physiological environment, the ester linkage will be hydrolyzed by the esterase, resulting in compounds with only four features required for its anti-trypanosomal activity. Thus, on this basis, we can explain that the mapping of the compound 9a with one HBA and three hydrophobic features is important for the anti-trypanosomal activity of 1,2-dihydroquinolines, while the ester substituent does not influence the binding of any of the analogs to a receptor within the parasite. This observation was further supported by the mapping of 9a and 10a on to the SBP model.
The best mapping pose of the most active compound 9a and 10a with SBP is shown in [Figure 6]a and [Figure 6]b. On analyzing the best mapping pose of the compound 9a with SBP, we found that it mapped four features, that is, one HBA and three hydrophobic features. The hydroxyl oxygen at 6th position of the 1,2-dihydroquinolines was involved in hydrogen bonding. The 1,2-dihydroquinoline nucleus was essentially involved in hydrophobic interaction with surrounding hydrophobic amino acids. One of the methyl at the 2nd position of 1,2-dihydroquinolines was involved in hydrophobic interaction, while the tail portion comprising the phenyl ring at N1 was also involved in hydrophobic interaction. Compound 10a also mapped one HBA and three hydrophobic features. Because structure-based modeling studies provide atomistic details about interactions governing the binding of drug with the binding cavity of the enzyme, we can say that the presence of one HBA and three hydrophobic features in the ligand is the absolute requirement for anti-trypanosomal activity.
|Figure 6: (a) Mapping of active compound 9a into the receptor-based pharmacophore model. (b) Mapping of ester analogue 10a on receptor-based pharmacophore model|
Click here to view
The main objective for comparing the ligand-based virtual model and the structure-based model was to examine the commonality of these two models, that is, to derive the common features from both the models that will govern the anti-trypanosomal activity of 1,2-dihydroquinolin-6-ols. On comparing the mapping results from both the studies, we found that the essential requirements for anti-trypanosomal activity were one HBA and three hydrophobic features. It was also observed that the proposed mapping modes of the most active compound 9a in ligand-based hypothesis are homogeneous and match the information from the crystallized enzyme–ligand complexes. The obtained features provide information about the kind of interactions important for the ligand–enzyme binding, so the retrieval of same features by the two models suggest a similarity of binding mode of 3,4-dihydroquinazolines (WPC) and 1,2-dihydroquinolin-6-ols. In the light of the aforementioned detailed pharmacophore study, it was concluded that we have obtained complimentary ligand- and structure-based models.
As proposed in the earlier studies, the dihydroquinolin-6-ols show their anti-trypanosomal activity by producing oxidative stress and/or the inhibition of TR enzyme. The potential of 1,2-dihydroquinolin-6-ols as TR inhibitors has been well supported by the complimentary results obtained in this study from both ligand- and structure-based models.
To further strengthen the findings that the compounds under consideration are TR inhibitors, we searched the earlier literature to find out the geometry and structural characteristics of TR-binding site. According to the previously published studies, the TR-binding site is large (2.2 × 2.0 × 2.8 nm3) and has hydrophobic patch formed by Glu18, Trp21, Ser109, Met113, and Ala343 amino acids. The binding site is characterized by several directed interactions such as hydrogen bond and hydrophobic interactions. Hence, a substrate having hydrophobic substituents and hydrogen bonding groups at specified interatomic distances will bind selectively to the TR-binding site of trypanosome. The retrieval of HBA and hydrophobic features in the ligand-based pharmacophore and SBP thus clearly complements the interaction requirement of TR binding site, which suggests that the possibility of TR inhibition cannot be ruled out. In addition, the compounds are likely to be redox-active agents that may rely on TR to account for the trypanocidal action.
The pharmacophore mapping of known trypanocidal agents with ligand-based pharmacophore features showed four-point interactions, suggesting that for designing new anti-trypanosomal agents, HBA and hydrophobic features are the important ones. The results of the mapping interactions of known anti-trypanosomal compounds and their estimated activities are listed in [Table 9]. The estimated anti-trypanosomal activity values obtained compare favorably with those of the actual values, which further confirm the predictivity of the developed model.
|Table 9: External test set compounds; their estimated and actual anti-trypanosomal activity (μM), pharmaprint values and mapping with ligand-based pharmacophore|
Click here to view
The structure–activity relationship revealed by pharmacophore-based 3D-QSAR and structure-based molecular studies is illustrated in [Figure 7], which shows the detailed information about the key interactions, the necessary features, and their location constraints for the development of newer anti-trypanosomal agents. In detail, hydrophobic groups at N1 and the 2nd position of 1,2-dihyroquinolines are favorable for hydrophobic interactions, while hydrogen bond acceptor group at the 6th position of the dihydroquinoline ring is critical for anti-trypanosomal activity, because the removal of hydroxyl group from 6th position in compound 17a drastically decreases the activity. The 1,2-dihydroquinoline ring is also important for the hydrophobic interaction, because quinolone compound 2a is almost inactive.
|Figure 7: Structure–activity relationship of 1,2-dihydroquinoline taken from 3D-QSAR and structure-based studies|
Click here to view
Herein, the term, pharmacophore feature, is used to describe the characteristic of a chemical structure that may facilitate a noncovalent interaction between a ligand and a biological target. Therefore, the analysis of the structure- and ligand-based modeling suggests that ideal high-affinity trypanocidal agents have to be developed under the following constraints: (i) three hydrophobic groups that form close contacts with hydrophobic amino acids along the binding cavity and (ii) the presence of at least one hydrogen bond acceptor functional group capable of forming hydrogen bonding.
More than 200 compounds were retrieved from the MiniMayBridge and National Cancer Institute (NCI) database. [Table 10] lists the name of the screened compounds and their estimated anti-trypanosomal activities. The parameters included in Lipinski’s rule of five were calculated for three compounds obtained from 3D database search, which indicates that there is no violation to Lipinski’s rule, and it is highly likely that these two compounds could have favorable pharmacokinetics profile. The screened lead compounds need further evaluation to produce newer compounds for the treatment of second-stage HAT. The overall work flow starting from the selection of training set to the development of statistically fit pharmacophore and database search is depicted in [Figure 8].
|Table 10: Name and estimated anti-trypanosomal activity of screened compounds obtained from MiniMaybridge and NCI database|
Click here to view
| Conclusion|| |
One of the major goals of this study was to generate predictive pharmacophore models that could be utilized as a query tool to search 3D databases of diverse drug-like compounds for the identification of new molecules that possess potent anti-trypanosomal activities. This was achieved by building a pharmacophore model by using in-vitro IC50 data, which contains all of the characteristics of the structure–activity relationships that were revealed to be essential for their activity. The best pharmacophore hypothesis, hypo1, consisted of the following five features: two hydrogen bond acceptor features and three hydrophobic features. Two validation tests, Cat-scramble method and test set prediction, were used to check the predictive ability of the developed model. Hypo1 predicted the affinity of the test set molecules with a correlation coefficient of 0.84 and showed the best statistical significance among all the generated models. This confirmed the statistical validity of our simple but effective 3D pharmacophore, excluding any possibility of a chance correlation between experimental and predicted activity values. The derived ligand-based model was merged with a receptor-based model, and a high consistency between these models was found. An external validation set of known trypanocidal agents was also used to evaluate the derived pharmacophore features necessary for the anti-trypanosomal activity of 1,2-dihydroquinolin-6-ols. Importantly, the analysis of the constructed model has shown that it can serve as a tool to scan and predict the anti-trypanosomal activity of candidate drugs before their synthesis. The validated pharmacophore model was used for searching new lead compounds, and we obtained three compounds with promising activities. There is no reported study till date combining cross-validated ligand-based QSAR and structure-based models for the development of anti-trypanosomal agents, and thus we believe that our pharmacophore model has been able to find important structural features required for finding newer molecules with improved efficacy and oral bioavailability for the second-stage HAT treatment.
The new lead candidate compounds were checked for druggable properties by applying Lipinski’s rule. Thus, our pharmacophore model was able to retrieve few leads that had good estimated anti-trypanosomal activities with acceptable calculated drug-like properties, and, therefore, they can be subjected to further optimization.
| Materials and Methods|| |
Ligand-based pharmacophore design
Data set preparation
A series of 1,2-dihydroquinolin-6-ols and their ester derivatives synthesized by Werbovetz et al. were selected for 3D pharmacophore modeling in view of the structural diversity and the wide coverage of activity range (four orders of magnitude). All the structures were built with Catalyst software (Catalyst version 2.2 software; Accelrys Inc., San Diego, CA), and energy was minimized to the closest local minimum using the generalized CHARMm force field as implemented in the program. The 52 compounds of the series were divided into training set and test set on the basis of training set selection criteria. A total of 37 compounds of the series including the most active and inactive compounds were selected as training set [Table 11], which were used for hypothesis (pharmacophore) generation, while the 15 test set compounds [Table 12] were used to quantify the validity of the proposed model.
|Table 11: Structures of training set compounds used for model development|
Click here to view
Catalyst software generates a group of representative conformational models for each compound in the training set using the Poling algorithm. Poling explicitly promotes conformational variation by forcing similar conformers away from each other. Every training set member had a collection of conformers that covered the conformational space accessible to the molecule within a given energy range. Catalyst software provides the following two types of conformational analysis: fast and best quality. The fast flexible search command considers only already existing conformers within the database, whereas the best flexible search additionally optimizes the conformational models during computation. The best option was used, specifying 255 as the maximum number of conformers under the constraint of 20 kcal/mol energy threshold above the estimated global minimum based on the use of the CHARMm force field., All other parameters used were the default values. The conformational model of the training set was used for hypothesis (pharmacophore) generation within the Catalyst software, which aims to identify the best 3D arrangement of chemical functions explaining the activity variations among the compounds in the training set.
A maximum of five features can be considered in the pharmacophore generation process using Catalyst HypoGen algorithm., Accordingly, from the 11 features available in the Catalyst features dictionary, the features necessary to explain the variance in the activity of the present 1,2-dihydroqunoline series were identified by using feature mapping protocol. The feature mapping protocol generates all possible pharmacophore features for the given input ligands. In this study, the chemical features optimized for exploring the spatial pharmacophore mapping of a series of 1,2-dihydroquinolin-6-ols and their ester derivatives were hydrophobic, ring aromatic, hydrogen bond acceptor, hydrogen bond acceptor lipid and hydrogen bond donor. Using these five features and varying the value of these parameters from minimum 0 to maximum 5, hypotheses were generated. The analysis revealed that hydrophobic (0, 5) and hydrogen bond acceptor (0, 5) features were the most important pharmacophoric features for explaining the dependence of the anti-trypanosomal activity of 1,2-dihydroquinolin-6-ols and their ester derivatives on functional groups.
Generation of pharmacophore hypotheses
The HypoGen module, was used to generate biological activity based pharmacophore hypotheses. This module evaluates a collection of conformational models of molecules and maps them to the selected chemical features (pharmacophore). A total of 10 hypotheses were generated using Catalyst module, among which the best-ranked pharmacophore was expected to identify the common binding features and the hypothetical orientation of the active compounds interacting with the target. The different control parameters employed for hypothesis generation were spacing, uncertainty, and weight variation. Spacing is a parameter representing the minimum interfeature distance that may be allowed in the resulting hypothesis. In the generated hypothesis, each feature signifies some degree of magnitude of the compound’s activity, which is controlled by the weight variation parameter. A value of 3 was assigned for the uncertainty parameter, which reflects the error of prediction, and denotes the standard deviation of a prediction error factor called the error cost.
HypoGen refine algorithm
A limitation of the pharmacophore feature hypothesis is that it does not take account of steric effect, and the activity prediction is based purely on the presence and arrangement of pharmacophoric features. The steric effect was evaluated by applying excluded volume parameters to the generated hypothesis. This feature in Catalyst software is called as HypoGen refine algorithm.
The HypoGen refine algorithm is an extension of the HypoGen algorithm, in which excluded volumes are placed in key locations to model unfavorable steric interactions. When placing an excluded volume, HypoGen refine inspects the volume space of the active and inactive molecules to locate reasonable positions for adding an excluded volume. These excluded volumes specify one or more spherical spaces in a pharmacophore that must not contain any atoms or bonds, because it might impinge sterically on a receptor. Thus, the use of excluded volumes can result in a more stringent pharmacophore, resulting in a more selective and predictive pharmacophore. In this study, we generated pharmacophore models with excluded volumes; a maximum of 10 excluded volume spheres were allowed.
Quality assessment of pharmacophore hypothesis
The statistical relevance of the best hypotheses was assessed on the basis of their cost relative to the null hypothesis, RMSD values, and their correlation coefficients (r). The overall costs of a model consist of the following three cost components: the weight cost, the error cost, and the configuration cost. The weight component is a value that increases in a Gaussian form, because this function weighs the model deviation from the ideal value of two. The error cost represents the root-mean squared difference between the estimated and measured activities of the training set. The configuration cost quantifies the entropy of the hypothesis space and should not exceed a maximum value of 17, which corresponds to a number of 217 pharmacophore models. It has been empirically determined that higher values often lead to a good correlation by chance. Two additional cost calculations performed by HypoGen are fixed cost and null cost. The fixed cost is the lowest possible cost representing a hypothetical simplest model that fits all data perfectly. Fixed costs are calculated by adding the minimum achievable error and weight cost and the constant configuration cost. Another cost parameter is the null cost, which represents the maximum cost of a pharmacophore with no features and estimates the activity to be the average of the training set molecules’ activity data. The null cost value is equal to the maximum occurring error cost. The greater the difference between null cost and total cost and the closer the total cost of the generated hypothesis is to the fixed cost, the more statistically significant is the generated hypothesis. According to randomized studies, a cost difference of 40–60 between the total cost and the null cost indicates a 75–90% chance of representing a true correlation in the data. The RMSD and correlation coefficients indicate the quality of “prediction” for the training set. In this study, the pharmacophore model with high correlation coefficients (r), lowest RMSD values, and highest cost difference was considered as the best pharmacophore model.
Validation of best pharmacophore model
To assess the statistical significance of the generated pharmacophore hypotheses, a validation procedure based on Fischer’s randomization test was performed. This was conducted by randomizing the activity data associated with the training set compounds using the Cat-scramble technique, available in the Catalyst/HypoGen module. These randomized training sets were then used to generate pharmacophore hypotheses, employing the same features and parameters as used in the development of the original pharmacophore hypothesis. The number of spreadsheets generated depends on the level of statistical significance selected. Thus, 19, 49, or 99 random spreadsheets have to be generated for attaining 95, 98, or 99% confidence level, respectively.,, If the randomized data set results in the generation of pharmacophoric models with similar or better cost values and correlation, then the original hypothesis is considered to be generated by chance.
Test set prediction
To verify the predictive power of the generated hypothesis, internal test set compounds comprising 15 molecules were used., All test set molecules were built and minimized as well as used in conformational analysis such as the training set molecules, and their activities were estimated using the best-ranked pharmacophore. The predictability of the hypothesis was assessed on the basis of the correlation coefficient values of the test set compounds.
PDB, established in 1971 at Brookhaven National Laboratory, is the single worldwide depository of information regarding the three-dimensional structures of large biological molecules, including proteins and nucleic acids. Initially, it had only seven structures, but today it contains more than 45,000 3D structures of biological molecules. For this study, the crystal structure of TR bound to 3,4-dihydroquinazoline was obtained from PDB (PDB code: 2WPC) [Figure 9]. The protein structure was monitored for valence; water molecules that may compete with compound binding were removed, and the missing hydrogen was added. The final cleaned and protonated structure of the protein target was subjected to active site identification.
|Figure 9: Trypanosoma brucei trypanothione reductase in complex with 3,4-dihydroquinazoline inhibitor (DDD00073357) from Protein Data Bank (PDB code: 2WPC)|
Click here to view
Defining active site and interaction generation
The enzyme active site was identified using a sphere, whose radius was adjusted to 9.0 Å, so as to include the active site and the key residues of the protein involved in interaction with ligand [Figure 10]. Keeping the density of lipophilic sites and the density of polar sites parameter value to 10, the interaction map was generated. The SBP method as implemented in Discovery Studio converted LUDI interaction maps within the protein-binding site into the following Catalyst pharmacophoric features: H-bond acceptor, H-bond donor, and hydrophobe. The LUDI models exhibit the protein–ligand interactions through the use of interaction sites. The generated interaction map complemented the TR enzyme-binding site. Specifically, functional features such as hydrogen bond donors/acceptors and lipophilic groups were identified in the active site, and the complementary features were placed within the binding site in chemically reasonable positions. The interaction map often displays a large number of features, especially when the receptor is capable of binding a variety of ligands and has a number of different binding modes. Thus, deriving pharmacophore models directly from the interaction map can be quite complicated. To overcome this problem, neighboring features of the same type were grouped to the same cluster. The feature that is closest to the geometric center of the cluster was selected to represent the cluster, whereas the rest of the features were omitted. However, even after the clustering, the number of features was still too high to use all of them in a single query. Therefore, multiple 3D queries, composed of fewer numbers of features, were generated from the interaction map by considering all the possible combinations. Finally, two clusters each for hydrogen bond acceptor and donor and three clusters for hydrophobe were used to construct the final pharmacophore model. The final model was subjected to nonfeature atoms exclusion. The exclusion constraint feature is an object that represents an excluded volume in space, within a given radius. The excluded volumes are placed on the regions of space that are occupied by the inactive molecules but not the active molecules. A pharmacophore with an excluded volume only matches if no atoms penetrate the excluded area. The obtained hypothesis contained the following seven features: two hydrogen bond donors, two hydrogen bond acceptors, and three hydrophobic groups describing the interactions between the TR enzyme and the ligand WPC [Figure 11].
|Figure 10: Sphere of radius 9 Å in the binding cavity of TR enzyme used for interaction generation|
Click here to view
External validation set
The external validations set comprising six compounds with known anti-trypanosomal activities were taken from the reported literatures.,, Various conformers of the compounds were generated and mapped on to the pharmacophoric features obtained from both ligand- and structure-based approaches. The anti-trypanosomal activity was estimated and compared with the actual activity. This further supported the essential binding requirement for the anti-trypanosomal activity of the 1,2-dihydroquinolin-6-ols.
Database search for new hits
After generating a validated pharmacophore model containing pharmacophore features, shapes, excluded volumes, etc., MiniMaybridge 3D database and NCI database search was performed to identify new molecules that share their features and can, thus, exhibit the desired biological response. The screened compounds were subjected to drug-like property calculation by applying Lipinski’s rule of five, which is a simple model to predict the absorption and intestinal permeability of a compound. According to the rule, compounds are well-absorbed when they possess Log P < 5, molecular weight <500, the number of H-bond donors <5, number H-bond acceptors <10, and the number of rotable bonds <10.
The authors sincerely thank Prof. Aditya Shastri, Vice Chancellor, Banasthali Vidyapith, Rajasthan, India for providing the necessary computational facilities for the completion of the study in a convenient manner.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Welburn SC, Maudlin I. The origins, dynamics and generation of Trypanosoma brucei rhodesiense
epidemics in East Africa. Parasitol Today 1999;15:399-403.
Mansand VH, Mahajan DT, Rastija V. QSAR analysis for 6-arylpyrazine-2-carboxamides as Trypanosoma brucei
inhibitors. SAR QSAR Environ Res 2017;28:165-77.
Cullen DR, Pengon J, Rattanajak R, Mocerino M. Scoping studies into the structure-activity relationship (SAR) of phenylephrine-derived analogues as inhibitors of Trypanosoma brucei rhodesiense. Chem Select 2016;15:4533-8.
Hiltensperger G, Jones NG, Niedermeier S, Jung J. Synthesis and structure-activity relationships of new quinolone-type molecules against Trypanosoma brucei
. J Med Chem 2012;55:2538-48.
Kryshchyshyn A, Kaminskyy D, Grellier P, Lesyk R. Trends in research of antitrypanosomal agents among synthetic heterocycles. Eur J Med Chem 2014;85:51-64.
WHO Expert Committee. Control and surveillance of African trypanosomiasis. In: WHO Technical Report Series. Geneva: World Health Organisation; 1998. p. 881.
Patrick DA, Bakunov SA, Bakunova SM, Kumar SE, Lombardy JR, Jones SK et al.
Synthesis and in vitro
antiprotozoal activities of dicationic 3,5-diphenylisoxazoles. J Med Chem 2007;50:2468-85.
African Trypanosomiasis (Sleeping Sickness). Fact Sheet Number 259. Geneva: World Health Organization Publications; 2001.
Legros D, Ollivier G, Paquet C, Burri C, Jannin J, Gasteullu-Etchegorry M et al.
Treatment of human African trypanosomiasis − Present situation and needs for research and development. Lancet Infect Dis 2002;2:437-40.
Barrett MP, Burchmore RJ, Stich A, Lazzari JO, Frasch AC, Cazzulo JJ et al.
The trypanosomiases. Lancet 2003;362:1469-80.
PAREXEL: PAREXEL’s Pharmaceutical R&D Statistical Source Book; 2001. p. 96.
Güner OF, editor. Pharmacophore Perception Development and Use in Drug Design. La Jolla, CA: IUL Biotechnology 2000.
Langer T, Hoffmann RD. Pharmacophores and Pharmacophore Searches. Wiley-VCH; 2006.
Kubinyi H. Success stories of computer-aided design. In: Ekins S, editor. Computer Applications in Pharmaceutical Research and Development. New York: Wiley-Interscience; 2006. p. 377-424.
Prajapati VK, Pandey RK, Kumbhar BV, Sundar S. Structure-based virtual screening, molecular docking, ADMET and molecular simulations to develop benzoxaborole analogs as potential inhibitor against Leishmania donovani trypanothione reductase. J Recept Signal Transduct Res 2017;37:60-70.
Hossain MU, Shah AR, Ahmad AI, Khan A. Identification of potential inhibitor and enzyme-inhibitor complex on trypanothione reductase to control Chagas disease. Comput Biol Chem 2016;65:29-36.
Becerra JR, Jensen LC, Ramos JH, Barrientos L. Identification of potential trypanothione reductase inhibitors among commercially available ββ-carboline derivatives using chemical space, lead-like and drug-like filters, pharmacophore models and molecular docking. Mol Divers 2017;3:697-711.
Krauth-Siegel RL, Inhoff O. Parasite-specific trypanothione reductase as drug target molecule. Parasitol Res 2003;90:S77-85.
Fotie J, Kaiser M, Delfın DA, Manley J, Reid CS, Paris JM et al.
Antitrypanosomal activity of 1, 2-dihydroquinolin-6-ols and their ester derivatives. J Med Chem 2010;53:966-82.
Brooks BR, Bruccoleri RE, Olafson BD, Sates DJ, Swaninathan S, Karplus M. CHARMM: A program for macromolecular energy, minimization, and dynamics calculations. J Comput Chem 1983;4:187-217.
Krovat EM, Langer T. Non-peptide angiotensin II receptor antgonists: Chemical feature based pharmacophore identification. J Med Chem 2003;46:716-26.
Smellie A, Kahn SD, Teig SL. Analysis of conformational coverage. 1. Validation and estimation of coverage. J Chem Inf Comput Sci 1995;35:285-94.
Smellie A, Kahn SD, Teig SL. Analysis of conformational space. 2. Applications of conformational models. J Chem Inf Comput Sci 1995;35:295-304.
Greene J, Kahn S, Savoy H, Sprague P, Teig S. Chemical function queries for 3D database search. J Chem Inf Comput Sci 1994;34:1297-308.
Catalyst version 2.2. San Diego, CA, USA: MSI Inc.; 2007.
Kurogi Y, Güner OF. Pharmacophore modeling and three-dimensional database searching for drug design using catalyst. Curr Med Chem 2001;8:1035-55.
Poptodorov K, Luu T, Langer T, Hoffmann R. Principles in medicinal chemistry. In: Hoffmann RD, editor. Methods and Pharmacophores and Pharmacophores Searches. Weinheim, Germany: Wiley-VCH; 2006. p. 17-47.
Maynard AJ. HypoGenRefine and HipHopRefine: Pharmacophore refienement using steric information from inactive compounds. HypoGenRefine algorithm. Presented at the ACS National Meeting, Spring; 2004.
Sutter J, Hoffmann R, Li H, Waldman M. Effect of variable weights and tolerances on predictive model generation. In: Guner O, editor. Pharmacophore Perception, Development, and Use in Drug Design. La Jolla, CA: International University Line; 2000. p. 499-511.
Fischer R. In: Hafner, editor. The Principle of Experimentation, Illustrated by a Psycho-Physical Experiment. Ch. 2. New York: Hafner Publishing Co.; 1966.
Funk OF, Kettmann V, Drimal J, Langer T. Chemical function based pharmacophore generation of endothelin − A selective receptor antagonists. J Med Chem 2004;47:2750-60.
Krovat EM, Langer T. Non-peptide angiotensin II receptor antagonists: Chemical feature based pharmacophore identification. J Med Chem 2003;46:716-26.
Tafi A, Costi R, Botta M, Di Santo R, Corelli F, Massa S et al.
Antifungal agents. 10. New derivatives of 1-[(aryl)[4-aryl-1H-pyrrol-3-yl]methyl]-1H-imidazole, synthesis, anti-candida activity, and quantitative structure–analysis relationship studies. J Med Chem 2002;45:2720-32.
Sakkiah S, Krishnamoorthy N, Gajendrarao P, Thangapandian S, Lee Y, Suh JK et al.
Pharmacophore mapping and virtual screening for SIRT1 activators. Bull Korean Chem Soc 2009;30:1152-6.
Lee Y, Bharatham N, Bharatham K, Lee KW. Adenosine Kinase Inhibitor Design Based on Pharmacophore Modeling. Bull Korean Chem Soc 2007;28:561-6.
Brookhaven National Laboratory. Protein Data Bank. Available from: http://www.rcsb.org/pdb/
. [Last accessed on 2017 Sep 29].
Kirchhoff PD, Brown R, Kahn S, Waldman M, Venkatachalam CM. Application of structure-based focusing to the estrogen receptor. J Comput Chem 2001;22:993-1003.
Bohm HJ. The computer program LUDI: A new method for the de novo design of enzyme inhibitors. J Comput Aided Mol Des 1992;6:61-78.
Meiering S, Inhoff O, Kramer B, Dormeyer M, Krauth-Siegel RL. Inhibitors of Trypanosoma cruzi
trypanothione reductase revealed by virtual screening and parallel synthesis. J Med Chem 2005;48:4793-802.
Salmon-chemin L, Buisine E, Yardley V, Kohler S, Debreu MA, Lander V et al.
2- and 3-substituted 1,4-naphthoquinone derivatives as subversive substrate of trypanothione reductase and lipoamide dehydrogenase from Trypanosoma cruzi
: Synthesis and correlation between redox cycling activities and in vitro
cytotoxicity. J Med Chem 2001;44:548-65.
Krauth-Siegel RL, Schoneck R. Flavoprotein structure and mechanism. 5. Trypanothione reductase and lipoamide dehydrogenase as targets for a structure-based drug design. FASEB J 1995;9:1138-46.
Lipinski CA, Lombardo F, Dominy BW, Feeney PJ. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv Drug Delivery Rev 2001;46:3-26.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6], [Figure 7], [Figure 8], [Figure 9], [Figure 10], [Figure 11]
[Table 1], [Table 2], [Table 3], [Table 4], [Table 5], [Table 6], [Table 7], [Table 8], [Table 9], [Table 10], [Table 11], [Table 12]