Introduction

Telomerase is a ribonucleoprotein reverse transcriptase, consisting of telomerase reverse transcriptional template RNA (hTR), telomerase reverse transcriptase (hTERT) and related protein [1, 2]. It has already been identified that 85–90% of cancer cells show overexpression of telomerase, which prevents telomere shorting and enables them to divide uncontrollably [3, 4]. Therefore, telomerase is an attractive oncological target that shows a distinctive role in cancer cell growth, high selectivity towards inhibitory mechanisms of cancer cell proliferation and multi-target ability.

Multiple anti-telomerase approaches are explored in the form of chemotherapy, gene-therapy, immunotherapy, combination and stem cell therapy [5]. Additionally, G-quadruplex ligand-based and heat shock protein (HSP-90) inhibitors are reported as the mechanistic approach for targeting telomere [6]. Following the abovementioned approaches, various natural and synthetic heterocyclic moieties are identified as an anti-telomerase agents. Among all anti-telomerase agents, only few small molecules like BIBR1532 [7], MST-312 [8] and BARCO-19 [9] have reached in the pre-clinical trial stages, as shown in Fig 1. However, none of the above telomerase inhibitors is successful in getting the FDA approval until date; it might be due to lower potency, lower selectivity or toxicity constraints [10].

Fig. 1
figure 1

Telomerase inhibitors reported in the pre-clinical trial state

Full size image

In this study, we performed structure-based and ligand-based drug design approaches for the identification of structural features of small molecules required for telomerase inhibition.

3D-QSAR is a ligand-based drug design approach used to generate a reliable statistical model for the prediction of the potency of new chemical entities [11]. It provided the correlation of molecular properties of the compound with its ligand-binding affinities [12]. CoMFA (Comparative Molecular Field Analysis) and CoMSIA (Comparative Molecular Similarity Indices Analysis) [13] are the most common and standard methods used for the development of the 3D-QSAR model [14]. In the CoMFA, columbic potential in terms of electrostatic field and Lennard-Jones potential in terms of the steric field are calculated. Like the CoMFA, the CoMSIA also helps to evaluate both the electrostatic and steric fields of the molecule, and additionally, it involves hydrophobic, hydrogen bond donor (HBD) and hydrogen bond acceptor (HBA) fields [15]. The CoMSIA differentiated from CoMFA through the Gaussian function and the steepness of the Gaussian function is controlled by an attenuation factor α at 0.3 as a standard value [16].

MDS (molecular dynamics simulations)-aided molecular docking helps to validate the QSAR model [17]. The MDS studies are performed for the identification of stable conformation of protein that helps to understand the behaviour and activity of biomolecule [18]. During the simulation process, changes in protein conformation in the presence of potential energy, pressure and temperature are calculated with respect to time. The results of MDS studies are validated by RMSD (root-mean-square deviation) value, RMSF (root-mean-square fluctuations), Rg (radius of gyration) analysis and Ramachandran plot [19, 20]. RMSD is the similarity measure used for the tracking of protein modifications during molecular dynamic studies. RMSD helps in the comparative studies of structural change of simulated and unsimulated protein molecules, while the RMSF analysis is useful to calculate the flexibility of enzyme residues of the given target throughout the simulation. The findings of the RMSF analysis suggest changes in protein flexibility pattern in the presence of analysed variants. Structural displacement of protein atoms from their common mass centre is measured by the Rg analysis and, therefore, the Rg analysis helps to describe the effect of MDS over the proteins dimension [21, 22]. Furthermore, the Ramachandran plot helps to analyse the change in the torsional angle between the amino acids of the protein due to molecular simulation. In addition to the results of MDS, molecular docking is performed to explore the possible interaction of inhibitors with the target protein [23, 24].

Materials and methods

A data set of 37 synthetic molecules, reported as telomerase inhibitors in various literature, were selected for the development of the 3D-QSAR model, as shown in Table 1 [25,26,27]. Structures of all molecules were drawn and minimised by utilising tripose force field in SYBYL-X, and minimised charges were calculated through the Gasteiger-Huckel method. The reported IC50 values of all molecules were converted into pIC50 (= −log IC50) because pIC50 values traversed a wide range that help in the preparation of modified derivatives with enhanced activity [28]. This pIC50 was utilised as the dependent variable. Data set was divided into training set for the generation of the QSAR model and test set for validation of the QSAR model. In this work, 28 molecules were used as a training set and nine molecules were used as test set [75:25]. The selection of test and training set was carried out in such a manner that compounds of test set resembled the compounds of the training set in multi-dimensional descriptor space. Thus, a test set was a true demonstrator of a training set (mol 2 structures of all compounds are provided as Supplementary Information data).

Table 1 Structure of telomerase inhibitors with biological activity
Full size table

Before starting the QSAR analysis, molecular alignments were performed, as revealed in Fig. 2. Molecular alignment is the key factor for generating a reliable model as it helps to generate a better statistical result [29, 30]. In the present study, distill- or rigid body-based, pharmacophore-based and docking-based alignments were performed to find out the best alignment method for the generation of the 3D-QSAR model. In distill alignment method (Fig. 2a), a common core moiety from the data set of all 37 molecules was identified. After the identification of common core, the highest active molecule from the data set, viz. molecule 3, was selected as a template molecule and the rest molecules of data set were aligned over this template. In pharmacophore-based alignment (Fig. 2b), different conformers of every molecule were generated with the help of DISCOtech module of SYBYL-X. Conformers were aligned over each other for the generation of a common pharmacophore. Docking-based alignment (Fig. 2c) was carried out with the help of the Surflex module of SYBYL-X, in which ligand of telomerase (PDB ID 5CQG) [31] was selected from RCSB-PDB. A binding pocket was generated into the ligand structure by protomol generation technique, and all molecules were docked in the binding pocket. From all these alignment techniques, distill alignment gave better statistical results as reported in Table 2, which was utilised for the development of a reliable 3D-QSAR model for further studies.

Fig. 2
figure 2

Different molecular alignments. a Distill alignment. b Pharmacophore alignment. c Docking alignment

Full size image
Table 2 Statistical results of CoMFA and CoMSIA models by PLS analysis using three different alignment methods
Full size table

The CoMFA and CoMSIA were reported as efficient tools in ligand-based drug design which were used for contour map generation as well as for the identification of favourable and unfavourable regions in a moiety [32]. The CoMFA and CoMSIA of five defined fields were calculated at regular grid space of 2.0 Å in Cartesian directions and the grid points were generated using +1 charge as well as Vander Waal’s radius of 1.52 Å. Energy cutoff for all fields was set as default at 30 kcal/mol. In the case of the CoMSIA study, the Gaussian function utilised for the determination of the distance between molecular atom and probe atom inside and outside of grid points [33].

For the estimation of linear correlation between structural parameters and biological activity, PLS (partial least squares) analysis was performed [34]. In the PLS analysis, pIC50 values were used as dependent variables and the CoMFA/CoMSIA descriptors were employed as independent variables for the development of the 3D-QSAR model [35]. No validation (r2), standard error of estimation (SEE), Fischer’s statistic value (f), leave one out (q2), cross-validation (r2cv) and boost strapping (r2bs) were calculated in the PLS analysis with an optimal number of components (N). After numerous trials, the best model was selected based on the highest q2, value and the results of the PLS analysis was validated through the estimation of predictive power (r2pred) of the best-generated model. The determination of r2pred was based on test set molecules by the following equation:

r2pred = 1 − (PRESS/SD)

where PRESS is the sum \of squared deviation between the actual and predicted biological activity (pIC50) of the test set molecules [PRESS = Σ(Y predicted − Y mean)2] and the SD denotes the sum of squared deviation of the actual and predicted biological activity of training set molecules [SD = Σ (Y actual − Y mean)2]. The generated 3D-QSAR model having high predictive ability was confirmed based on high q2 and r2 values (q2 > 0.5, r2 > 0.9, r2pred > 0.5) [36, 37].

Visual illustrations of the 3D-QSAR were performed through contour maps, generated in a 3D grid orientation of 80% favourable and 20% unfavourable field contributions. Contour maps give the idea about favourable and unfavourable regions around the molecules. Structural modification, according to the contour map analysis, helps in the alteration of physicochemical properties that might be responsible for increasing or decreasing the potency of molecules [38].

In MDS studies, a chain of co-crystallised ligand of telomerase (PDB: 5CQG) was stabilised in the presence of Amber14sb force field for 50 ns with a time step of 2 fs [39]. During the simulation process, protein structure was stabilised at 1-atm pressure with an average temperature of 299.98 k and potential energy of 1,145,336 kJ/m converged in 997 steps using GROMACS 2019.4 software. For the determination of molecular interactions of the highest active compound with simulated/unsimulated protein structure [40], molecular docking was carried out in SurFlex dock module of SYBYL-X. Based upon all studies, few novel quinoline and acridine derivatives were designed and docked on the same target to evaluate their comparative potentials.

Overall, in this study, we performed the 3D-QSAR and molecular dynamics/simulation-assisted molecular docking studies of known reported telomerase inhibitors to identify potential spatial features for designing novel telomerase inhibitors and also designing few novel molecules.

Results and discussion

The experimental and predicted activity values of the CoMFA/CoMSIA are listed in Table 3, and a correlation between experimental and predicted pIC50 is shown in Fig. 3. Here, the correlation analysis of experimental and predicted pIC50 value was explained by the linearity graph. In this graph, the experimental pIC50 was used as the logarithmic values of the biological activity (IC50) of selected molecules as reported in the literature and the predicted pIC50 generated through the CoMFA and CoMSIA PLS analysis. During the linearity analysis, it was observed that the CoMSIA analysis showed better linearity between both the parameters with the observed R2 value of 0.7403, but in case of the CoMFA, the linear relationship between the experiment and predicted pIC50 was observed lower with the R2 value of 0.6546. The graph helped to validate the predictive power of the generated QSAR model. In this study, statistical outcomes of the CoMFA was found to be lower than the standard limits; therefore, the statistical significance of the CoMSIA was utilised for the generation of the QSAR model [12, 17].

Table 3 The predicted and experimental activity of the training and test set compounds for CoMFA and CoMSIA
Full size table
Fig. 3
figure 3

Plot of experimental pIC50 versus predictive pIC50 of CoMFA and CoMSIA models.

Full size image

The CoMSIA models with good statistical values were generated by permutation combination of all descriptors, viz. steric, electrostatic, hydrophobic and donor as well as acceptor fields. The statistical results of all descriptors’ combination of the CoMSIA model are shown in Fig. 4 (and also in Supplementary Information Table ST-1). From all these descriptors’ combinations, SEHD (steric, electrostatic, hydrophobic and donor) showed best statistical results with q2 value of 0.667, r2 value of 0.975, r2cv value of 0.659, r2bs value of 0.988, F value of 134.336, SEE value of 0.062 and r2pred value of 0.570. Combination of SEHDA (steric, electrostatic, hydrophobic, donor and acceptor) also showed almost equivalent result as SEHD with q2 value of 0.662, r2 value of 0.974, r2cv value of 0.660, r2bs value of 0.981, F value of 129.633, SEE value of 0.063 and r2pred value of 0.560, with the field contribution value of 0.093, 0.217, 0.193, 0.384 and 0.113 for steric, electrostatic, hydrophobic, donor and acceptor field, respectively. Because SEHDA combination covered all the fields as well as it gave optimum statistical results, this combination was utilised for the generation and analysis of counter maps.

Fig. 4
figure 4

Comparison of CoMSIA descriptors’ results

Full size image

Highly potent molecule (molecule 3) overlaid on the contour maps of steric, electrostatic, hydrophobic, donor and acceptor fields as shown in Fig. 5. Based on favourable and disfavourable regions, the contour map analysis was performed. As shown in Fig. 5a of a steric contour map, the green colour referred to the favoured region and the yellow colour indicated disfavoured region for bulky group substitution which could be explained by a comparison of compounds 3 and 9. Compound 3 (IC50 value of 0.44 μM) consisted of the methyl-substituted aliphatic amide side chain at the 4th position of the quinoline ring which was found to be sterically favourable due to the presence of –CH3 group. Compared to compound 3, compound 9 with less bulky and polar –OH substitution was found to be significantly less potent (IC50 value of 12.3 μM). This was also proved by the comparison of compounds 14 and 11; where substitution of o-fluoro benzamide-oxadiazole at 2nd position of quinoline ring in compound 14 was responsible for higher potency (IC50 value of 0.8 μM) as compared to compound 11 where benzamide-oxadiazole-linked quinoline moiety showed lower potency (IC50 value of 12.8 μM). CoMSIA electrostatic contour map is shown in Fig. 5b. Here, the blue colour region favoured electropositive substitutions while red colour favoured electronegative substitutions. Substitution of electropositive –NH of indole ring at 3rd position of quinoline ring favoured the activity while electronegative oxygen of benzofuran at the same position decreased the activity. This was confirmed by all selected molecules of series one. Thus, it is concluded throughout the series that negatively charged nitrogen atom, which is present in the quinoline ring, is essential for its activity and the same is confirmed by the red colour contour map. Additionally, electronegative halide substitution is also favourable for the activity which is supported by compound 14 (IC50 value of 0.8 μM).

Fig. 5
figure 5

CoMSIA contour maps with structure of compound 3 shown inside the field. a CoMSIA steric contour map: favourable region refers to green colour while yellow colour refers to a disfavourable region; b CoMSIA electrostatic contour map: red colour represents negatively charged and blue represents positively charged substituents; c hydrophobic field: yellow colour refers to favourable and white colour refers to disfavourable substitution; d hydrogen bond donor field: cyan and purple colours refer to favourable and disfavourable, respectively; e hydrogen bond acceptor field: magenta and red colour contour map supports favourable and disfavourable substitutions

Full size image

In the hydrophobic contour map, the yellow colour region favoured hydrophobic substitution while white colour favoured hydrophilic substitution, which is shown in Fig. 5c. The contour map of the hydrophobic region was validated by compounds 1 and 10. In compound 1, the substitution of nonpolar morpholinoethanamide side chain favoured activity (IC50 value of 0.5 μM) while polar aminoethanol substitution in compound 10 disfavoured activity (IC50 value of 8.1 μM). The abovementioned compounds additionally helped in cross-validation of the electrostatic contour map, where hydrophilic contour map was shown in place of electropositive region or in the place of –NH substitution. Donor field contributors are shown in Fig. 5d, where the cyan colour represented favourable donor substitution while purple colour represented the substitution of disfavour electron donor group. Amine group of indole at 3rd position of quinoline moiety favoured electron donor property as found in compounds 1 to 5. Electropositive and hydrophilic contour maps sympathetically supported to favourable donor substitution. All selected molecules are comprised of electronegative nitrogen atom which was responsible for disfavoured donor field contribution. Contour map of the CoMSIA acceptor field is shown in Fig. 5e, where the magenta and the red colour region represented favourable and disfavourable substitutions, respectively. Acceptor region was validated by compounds 12 and 19. In the case of compound 19. p-chloromethyl aniline favoured the activity due to the presence of electron-withdrawing halide group (IC50 value of 0.9 μM). In contrast, electron-donating methyl atom of dimethylaniline in compound 12 disfavoured the activity (IC50 value of 7.3 μM). Additionally, in the case of compounds 1, 6 and 24; electron-withdrawing oxygen atom of the morpholine ring supported the acceptor region and activity.

During MDS study, effects of concerned pressure, temperature and potential energy over the stabilisation of protein at a particular time duration were observed, as shown in Fig. 6. The binding site residues, at around 6 Å of the ligand, were compared for compounds 3 and 11, before and after MD simulation and are shown in Fig. 7. In the case of compound 3, the ligand showed better interaction affinity towards the protein which was significantly changed with the change in dynamic environment of protein. The tail of the ligand displaced towards the exposed solvent and the main core of the ligand mainly lied towards the hydrophobic pocket. Before the simulation, the compound showed the pi-stacking interactions with the residue of Phe494 and Tyr551 which was absent at the 50-ns pose and it might be due to the fluctuations in amino acid sequences. The compound showed pi-cation interaction with Arg486, and the –NH of fused indole ring showed the H-bond interaction with the carbonyl group of Asn492 at the distance of 2.05 Å after the simulation. The compound 11 showed low interaction affinity towards the target as it showed only a single pi-interaction with Phe494 and none of the significant changes in the interaction was observed after the simulation. The findings suggested that the changes in the dynamic environment of the protein led the variation in ligand-protein interaction affinity, which helped to gain an idea about the phenomenon behind ligand designing for the significant inhibition.

Fig. 6
figure 6

MDS study of a chain of telomerase (PDB ID; 5CQG) concerning a potential energy versus time; b pressure versus time; c temperature versus time

Full size image
Fig. 7
figure 7

Effect of MDS analysis over the active and inactive ligand. a MDS study of compound 3; b MDS study of compound 11

Full size image

Furthermore, the reliability of the molecular dynamic studies was checked through the RMSD, RMSF and Rg analysis, as shown in Fig. 8. Resultant RMSD analysis is shown in Fig. 8a. Here, similar alignments of simulated and unsimulated protein with zero RMSD value helped to understand that the amino acid sequence of the unsimulated protein remained unchanged after simulation. Effect of active compound 3 and comparatively inactive compound 11 variants of selected QSAR series over the flexibility of simulated protein were analysed by the RMSF analysis as shown in Fig. 8b. Less fluctuations in the presence of compound 3 indicated more stable conformer, whereas more fluctuations in the presence of compound 11 indicated less stable conformer. The measure of the compactness of protein by the Rg analysis is shown in Fig. 8c where the Rg values were reasonably unchanged in their compact (folded) region over the course of 50 ns at 300 k, which defined that the structure of the protein was stable in the dynamic environment. Amino acid sequencing was rechecked with the help of the Ramachandran Plot (Supplementary Information as Figure SF-1). The resultant similarity of simulated and unsimulated protein was the number of residues in the favourable region 1103 (92.8%), the number of residues in the allowed region 68 (5.7%) and the number of residues in the outlier region 17 (1.4%). This explained that the amino acid sequence of unsimulated protein remained unchanged after the simulation.

Fig. 8
figure 8

Analysis of molecular simulation study. a Backbone RMSD of 5CQG: unstimulated protein is represented by the black colour while simulated protein is represented by the red colour; b RMSF per residue of 5CQG with its variants: black colour represents the fluctuation in presence of active variant and red colour represents the fluctuation in presence of inactive variants; c Measure of Rg after simulation

Full size image

Molecular docking studies of the active ligand with unsimulated and simulated proteins are shown in Fig. 9, and they supported cross-validation of the CoMSIA contour maps. Docking of compound 3 with unsimulated and simulated proteins showed the interactions of nitrogen atom of alkyl amide side chain with hydrophobic Gly553 and positive charge Lys416, respectively. These interactions explained the importance of the hydrogen bond acceptor and favourable hydrophobic interactions in this region which was supported by magenta colour contour map of acceptor field around nitrogen atom. After simulation, the number of hydrogen bond interactions increased as the –NH– of indole ring substituted at 3rd position and –NH– substituted at 4th position of quinoline core moiety interacted with Phe443. These interactions supported an electropositive functional group required for the activity which was validated by the blue colour counter map around these ‘N’ atom. Based on the 3D-QSAR studies and MDS-assisted molecular docking studies, the essential structural requirements of quinoline moiety is shown in Fig. 10.

Fig. 9
figure 9

Docking interactions of highest active ligand with a simulated and b unsimulated protein structure (PDB-5CQG)

Full size image
Fig. 10
figure 10

Important structural requirements of quinoline moieties as per the ligand-based 3D-QSAR and structure-based MDS-assisted molecular docking study

Full size image

By utilising overall data of the study, we have designed ten molecules, five with quinoline and five with acridine as core moiety. All designed molecules were docked in A-chain of telomerase active site (PDB-5CQG) in SurFlex Dock module in SYBYL (results are shown as Supplementary Information Table ST-2). Docking interactions of potent molecules along with design compounds are shown in Fig. 11. Comparative docking study of the designed molecules with marketed anticancer drug 5-FU and standard anti-telomerase agent BIBR1532 was also performed. The designed molecules showed better interaction affinity then marketed drug and it is similar to BIBR1532. As the same in BIBR1532, in the design molecules, negatively charged oxygen was facing positively charged Arg484.

Fig. 11
figure 11

Docking interactions of potent molecules along with design compounds

Full size image

Conclusion

This study is an attempt to design novel compounds with potent telomerase inhibitory activity. Statically significant CoMSIA with a q2 value of 0.662 and higher predictability with an r2pred value of 0.560 was utilised in the generation of contour maps. Substitution of steric, electrostatic, hydrophobic and acceptor and donor atoms is significant to inhibit the effect of telomerase enzyme. Molecular dynamic and simulation-aided molecular docking study assisted the identification of promising features of quinoline moiety for telomerase inhibition. It was also observed that the improvement of hydrogen bond interactions resultant from a conformational change in protein structure was not due to the changes in amino acid sequences. The generated 3D-QSAR model helped to introduce a new direction in designing novel quinoline and acridine derivatives while molecular docking helped to investigate the possible binding interactions of design molecules which could be explored in future.