SEARCH

SEARCH BY CITATION

Keywords:

  • 3D-QSAR;
  • docking;
  • pyrrolidine derivatives

Abstract

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results and discussions
  5. Conclusions
  6. Acknowledgments
  7. References
  8. Supporting Information

Docking studies of pyrrolidine derivatives indicated that Trp178, Arg371, and Tyr406 were the key residues in the active pocket of influenza neuraminidase (NA). Hydrogen bond and electrostatic factors mainly influenced interactions between pyrrolidine derivatives and NA. Moreover, there was a significant correlation between binding affinity (total scores) and the experimental pIC50. Meanwhile, 3D-QSAR models of 87 pyrrolidine derivatives were developed to understand chemical–biological interactions governing their activities toward NA. Furthermore, R2, Q2, inline image, and inline image of the models were from 0.731 to 0.830, from 0.560 to 0.611, from 0.762 to 0.875, and from 0.649 to 0.856, respectively. QSAR modeling results elucidated that hydrogen bonds and electrostatic factors highly contributed to inhibitory activity, which was unanimous in the docking results.

Neuraminidase (NA) is one of the two surface glycoproteins of influenza virus and a potential target to control influenza virus (1). Neuraminidase inhibitors (NIs) can selectively inhibit NA, prevent progeny virosome from replication and release in the host cell, and thus take precautions against influenza and the alleviation symptom effectively (2). Sialic acid analogs are the class of NIs that was reported first. Then, different series of NIs were prepared, such as cyclohexene (3), cyclopentane (4), and pyrrolidine derivates (5). At present, both zanamivir and oseltamivir are effective inhibitors for both A and B forms of NA. In addition, peramivir has been authorized for emergent treatment of 2009 H1N1 influenza virus in some countries (6–8). As an effective kind of anti-influenza drugs, there is a rising trend toward resistance of NIs because of the emergence of new influenza virus. So, it is necessary to research and develop new NIs. Moreover, docking (9) and QSAR (10) are widely used in drug design. In this study, Surflex-Dock is employed to investigate ligand–receptor interaction mechanisms. 3D-QSAR models of pyrrolidine derivatives are constructed by almond, CoMFA (comparative molecular field analysis), CoMSIA (comparative similarity indices analysis), and Topomer CoMFA to understand chemical–biological interactions governing their activities toward influenza NA.

Materials and Methods

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results and discussions
  5. Conclusions
  6. Acknowledgments
  7. References
  8. Supporting Information

Docking

Surflex-Dock (11) is a classical protein–ligand docking program associated with a specific empirical scoring function and search engine pair. Its usefulness as a drug design tool has already been demonstrated in several cases (12). Protomol is used to guide molecular docking. Protomol is a computational representation of the intended binding site to which putative ligands are aligned. Production of Protomol supplies three manners (13): (i) Automatic: Surflex-Dock finds the largest cavity in the receptor protein; (ii) Ligand: a ligand in the same co-ordinate space as the receptor; (iii) Residues: specified residues in the receptor. Surflex-Dock scores include hydrophobic, repulsive, entropic, and solvation. The strengths of individual scoring functions combine to produce a consensus that is more robust and accurate than any single function for evaluating ligand–receptor interactions. CScore (consensus scores) (14) integrates a number of popular scoring functions for ranking the affinity of ligands bound to the active site of a receptor. CScore provides several functions: D Score (15), PMF Score (16), G Score (17), and CHEM Score (18). CScore range is from 1 to 5. The best CScore is 5.

Preparation of receptors (NA)

The crystal structures of 1nnc (19), 1l7f (20), and 2qwh (21) were retrieved from RCSB Protein Data Bank. All ligands and water molecules were removed, and the polar hydrogen atoms were added (22,23).

Optimization of ligands (NIs)

The minimization method of energy was Powell. Force field and charge were in turn Tripos and MMFF94. Max iterations, termination, and RMS displacement were 1000, 0.001 kcal/(mol*Å), and 0.001 Å, respectively.

Settings of docking parameters

Surflex-Dock 2.0 was used to dock NIs into NA. Protomol was set to automatically define the receptor binding site. Two parameters determining the extent of Protomol, threshold and bloat parameters, were 0.31 and 1 Å, respectively.

Structural characterization and modeling

almond is a program for generating and handling GRIND (GRid INdependent Descriptors) based on Molecular Interaction Fields (MIFs). GRIND needs no alignment of compounds. The obtaining GRIND procedure involves three steps (24): calculation of MIF; filtration of MIF to extract the most relevant regions that define virtual receptor site (VRS); and encoding the VRS into the GRIND variables. almond 3.3.0 was used to generate GRIND. QSAR model was constructed by PLS. The default GRID probes were used, namely DRY (hydrophobic probe), O (carbonyl oxygen probe as hydrogen bond acceptor), N1 (amidic nitrogen probe as hydrogen bond donor), and TIP (shape probe). CoMFA (25) and CoMSIA (26,27) parameters were established by default. The models were generated by SAMPLS (sample-distance partial least squares) (28). Topomer CoMFA (29–31) was a 3D-QSAR tool that automates the creation of models for predicting the biological activity or properties of compounds. The core of this method was fragment selected compound. In this study, fragment method was fragment 1 – core – fragment 2.

Model validation

Rationality of docking was validated by root mean squared deviation (RMSD) between the position calculated of co-crystal ligand in NA and that observed in crystal structure (32). Consistency of different scores was evaluated by CScore. QSAR models were validated by leave-one-out cross-validation (LOO CV) and external prediction by test sets inline image and inline image (33). Meanwhile, the models were validated by different modeling methods.

  • image

Dataset

Eighty seven pyrrolidine derivatives with inhibitory influenza A virus were obtained from references (5,34–36). From structurally diverse molecules possessing activities of a wide range, 87 NIs were divided into training set with 67 samples and test set with 20 samples (Table S1, marked with asterisk).

Results and discussions

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results and discussions
  5. Conclusions
  6. Acknowledgments
  7. References
  8. Supporting Information

Docking

Comparison of binding pocket between co-crystal ligand and NA

From hydrogen bond (dashed line) interactions between co-crystal ligand zanamivir in 1nnc and the key residues in the active pocket (Figure S1) by ligplot (37), main residues in the active pocket were hydrophilic. It showed that hydrogen bond and electrostatic were main factors. 11 hydrogen bonds were formed between zanamivir and residues (Arg118, Asp151, Arg152, Trp178, Glu227, Glu276, and Arg371). Number in dashed line showed distance of every hydrogen. In addition, nine hydrogen bonds were formed between BCZ in 1l7f and residues (Arg118, Asp151, Arg152, Trp178, Glu227, Arg292, and Arg371) (Figure S2). Seven hydrogen bonds were formed between oseltamivir in 2qwh and residues (Arg118, Glu119, Asp151, Arg152, and Arg371) (Figure S3). The docking of three protein targets showed that Arg118, Asp151, Arg152, and Arg371 were critical residues. From the active pocket, hydrophobic interactions between co-crystal ligand and residues were produced.

Comparison of docking between NIs and NA

The docking results elucidated that there was a significant correlation between total scores (binding affinity) and the experimental pIC50 (Figure S4). Moreover, RMSD between the position calculated of zanamivir in 1nnc and that observed in the crystal structure was 1.14 Å. Similarity (a measure of similarity between solution co-ordinates and reference co-ordinates) was 0.77 (Figure 1). Total score and CScore were in turn 8.37 and 4. In addition, RMSD between the position calculated of BCZ in 1l7f and that observed in the crystal structure was 1.09 Å, and similarity was 0.78 (Figure S5). Total score and CScore were 9.60 and 4, respectively. RMSD between the position calculated of oseltamivir in 2qwh and that observed in the crystal structure was 0.89 Å, and similarity was 0.80 (Figure S6). Total score and CScore were 6.80 and 5, respectively. The obtained RMSD was all smaller than 2.5 Å (32), which showed that the docking results were reasonable and successful. CScore of all samples was good (Table S2).

image

Figure 1.  Comparison between the calculated position (molecule with ball and stick) of zanamivir in 1nnc and that observed (molecule with sticks) in the crystal structure.

Download figure to PowerPoint

Figure 2 illustrated hydrogen bond (dashed line) interactions between the most active sample ID 86 and the main residues in the active pocket of 1nnc. Seven hydrogen bonds of four types (including N-H···O=C, O-H···O=C, C-O···H-N, and C=O···H-N) between ID 86 and residues (Glu277, Arg292, Arg371, and Tyr406) were formed. Meanwhile, hydrophobic interactions between ID 86 and residues (Ala246, Ile275, Ile222, and Trp178) were produced. In addition, six hydrogen bonds between ID 86 and the key residues (Glu277, Arg371, and Tyr406) in the active pocket of 1l7f were formed (Figure S7). Eight hydrogen bonds between ID 86 and the key residues (Arg152, Glu277, Lys292, Arg371, and Tyr406) in the active pocket of 2qwh were formed (Figure S8). From the active pocket, main influencing factors of interactions between INs and NA were hydrogen bond and electrostatic, followed by hydrophobic interaction. So, Trp178, Arg371, and Tyr406 were the key residues in the active pocket.

image

Figure 2.  Hydrogen bond (dashed line) interactions between the most active sample ID 86 and residues in the active pocket of 1nnc.

Download figure to PowerPoint

3D-QSAR

Modeling comparison

Through almond modeling, R2, Q2, inline image, and inline image of the optimal model with four components were 0.830, 0.560, 0.765, and 0.649, respectively (Table 1). Based on common skeletons alignment, the optimum CoMFA and CoMSIA models with two components were obtained. R2, Q2, inline image, and inline image of CoMFA model were 0.780, 0.590, 0.875, and 0.848, respectively. In addition, R2, Q2, inline image, and inline image of CoMSIA model were in turn 0.731, 0.611, 0.864, and 0.856, respectively. Relative contributions of two models showed that hydrogen bond highly contributed to inhibitory activity, then hydrophobic and electrostatic factors (Table 1). Through Topomer CoMFA modeling, the optimum model with five components was obtained. R2, Q2, inline image, and inline image of the model were 0.807, 0.600, 0.762, and 0.722, respectively (Table 1). So, four 3D-QSAR models were robust and had good predictive capabilities. Relative plots between the predicted and experimental pIC50 values of 87 NIs (Figure S9) showed that all samples were almost uniformly distributed around diagonal, not obviously exceptional point had selected. It further showed that models had good exterior predictive capabilities.

Table 1.   Statistical results and relative contributions of the models
MethodsR2SEQ2SEcvinline imageinline imageAFRelative contributions
StericElectrostaticHydrophobicDonorAcceptor
  1. R2, squared multiple correlation coefficients; SE, standard error of estimate; Q2, squared cross-validated correlation coefficient; SEcv, cross-validated standard error of estimate; A, components; F, Fisher statistic; inline image and inline image are R2 and Q2 of test set, respectively.

almond0.8300.5310.5600.8560.7650.6494
CoMFA0.7800.6190.5900.8460.8750.8482113.6080.4060.594
CoMSIA0.7310.6850.6110.8230.8640.856287.0910.0890.2420.2550.2090.206
Topomer CoMFA0.8070.6000.8610.7620.7225

Model analysis

GRID plots for interaction fields (Figure S10) showed that hydrophobic interaction fields were mainly produced around acryl in site 4, propyl in site 5, and methene in site 2 (Table S1). Hydrogen bond interaction fields were mainly formed around amino in site 1, carboxy in site 2, and amide group in site 5. Shape interaction fields were mainly formed about carboxy in site 2 and methyl in site 5. Therefore, according to models, affinity is clearly driven by hydrogen bonds rather than by hydrophobic interactions or shape complementarity.

In CoMFA steric contour maps (Figure 3A), large bulk group in green regions was favorable to enhance activity, whereas small bulk group in yellow regions was favorable. Green regions were mainly distributed around sites 2 and 4 of pyrrolidine core, and there were yellow regions around sites 3 and 5 of pyrrolidine core. In CoMFA electrostatics contour maps (Figure 3B), blue and red regions were favorable to improve activities when substituent groups were more electropositivity and more electronegativity, respectively. Main distributions of red regions were around sites 1 and 5, and blue regions were mainly distributed around partial 5.

image

Figure 3.  CoMFA steric (A) and electrostatic (B) contour maps (green: steric favored; yellow: steric disfavored; blue: electrostatic favored; red: electrostatic disfavored; reference molecule: ID 86).

Download figure to PowerPoint

In CoMSIA hydrophobic contour maps (Figure 4A), large hydrophobic groups in orange regions were favorable to enhance activity, whereas red regions were unfavorable. Distribution region of favorable hydrophobic group was in partial sites 2 and 3. Unfavorable hydrophobic group region was distributed around site 5. In CoMSIA hydrogen bond donor contour maps (Figure 4B), cyan regions denoted that hydrogen bond donor was of advantage to improve activity, whereas purple regions were disadvantage. Distribution regions of favorable hydrogen bond donor were around partial site 5. In CoMSIA hydrogen bond acceptor contour maps (Figure 4C), magenta regions represented hydrogen bond acceptors were favorable to improve activity; however, violet regions were unfavorable. Favorable hydrogen bond acceptors were distributed around sites 2, 3, and 4. CoMSIA steric and electrostatic contour maps were similar to that of CoMFA.

image

Figure 4.  CoMSIA contour maps: hydrophobic (A), hydrogen bond donor (B) and acceptor (C) (orange: hydrophobic favored; red: hydrophobic disfavored; cyan: hydrogen bond donor favored; purple: hydrogen bond donor disfavored; magenta: hydrogen bond acceptor favored; violet: hydrogen bond acceptor disfavored; reference molecule ID 86).

Download figure to PowerPoint

Topomer steric and electrostatic contour maps of R1 and R2 groups in ID 86 (Figure 5) were similar to that of CoMFA. The predicted activities of R1 and R2 groups of all samples were analyzed (Table S2). The predicted best compound, ID 86, is the one with the best activity for its R2 group. It might be one of the reasons that activity of ID 86 was the highest. As a whole, the four employed modeling methods appeared to be complementary. Indeed, Topomer CoMFA could obtain different groups contribution to activity. CoMFA and CoMSIA could supply relative contributions to activity. almond could provide key groups contribution to activity. Moreover, the four generated 3D-QSAR models demonstrated robust predictive capabilities and may be further used to evaluate the bioactivity of new compounds. Hydrogen bond and electrostatic factors highly contributed to inhibitory activity, which were consistent with docking results. Furthermore, site 1 and site 5 of pyrrolidine core were the key position contribution to bioactivity. Thus, these groups and atoms were modified to obtain new compounds with high activities.

image

Figure 5.  Topomer CoMFA steric (A, C) and electrostatic (B, D) contour maps of R1 and R2 groups in ID 86 (green: steric favored; yellow: steric disfavored; blue: electrostatic favored; red: electrostatic disfavored; A and B reference molecule: R1 group; C and D reference molecule: R2 group).

Download figure to PowerPoint

Conclusions

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results and discussions
  5. Conclusions
  6. Acknowledgments
  7. References
  8. Supporting Information

The docking results of 87 NIs elucidate that the main residues in the active pocket of NA are hydrophilic and polar. Moreover, there was a significant correlation between total scores and the experimental pIC50 with correlation coefficient = 0.514∼0.656 and p < 0.0001. Trp178, Arg371, and Tyr406 are the key residues in the active pocket. NIs–NA interactions in active compounds are dominated by hydrogen bonding followed by electrostatic and hydrophobic interactions. Meanwhile, four 3D-QSAR models of 87 NIs have been developed to understand chemical–biological interactions governing their activities toward NA. Furthermore, R2, Q2, inline image, and inline image were from 0.731 to 0.830, from 0.560 to 0.611, from 0.762 to 0.875, and from 0.649 to 0.856, respectively. The QSAR results indicate that hydrogen bonds and hydrophobic interactions highly contribute to inhibitory activity, which are consistent with the docking results. So, the obtained 3D-QSAR models can design and screen new compounds.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results and discussions
  5. Conclusions
  6. Acknowledgments
  7. References
  8. Supporting Information

This work was supported by the Natural Science Foundation of Sichuan Province (Grant No. 2010JY0185) and the college project for Sichuan University of Arts and Science (Grant No. 2010A07Z).

References

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results and discussions
  5. Conclusions
  6. Acknowledgments
  7. References
  8. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results and discussions
  5. Conclusions
  6. Acknowledgments
  7. References
  8. Supporting Information

Figure S1. Hydrogen bond (dashed line) interactions between cocrystalligand zanamivir in 1nnc and residues in the active pocketby ligplot.

Figure S2. Hydrogen bonding (dashed line) interactions betweenco-crystal ligand BCZ in 1l7f and residues in the active pocket byligplot.

Figure S3. Hydrogen bonding (dashed line) interactions betweenoseltamivir (G39) crystal structure in 2qwh and residues in theactive pocket by ligplot.

Figure S4. The correlation plot between the total scores and theexperimental activities of all 87 NIs.

Figure S5. Comparison between the calculated position (moleculewith ball and stick) of BCZ in1l7f and that observed (molecule withstick) in the crystal structure.

Figure S6. Comparison between the calculated position of oseltamivir(molecule with ball and stick) in 2qwh and that observed(molecule with stick) in crystal structure.

Figure S7. Hydrogen bonding (dashed line) interactions betweenthe most active sample ID 86 and residues in the active pocket of1l7f.

Figure S8. Hydrogen bond (dashed line) interactions between themost active sample ID86 and residues in the active pocket of 2qwh.

Figure S9. Relative plots through origin exp. versus pred. pIC50values of 87 NIs.

Figure S10. GRID plots for interaction fields of the most activesample ID 86.

Table S1. Structuresl and predicted activities of 87 pyrrolidinederivatives.

Table S2. Docking scores and predicted activity of 87 pyrrolidinederivatives.

FilenameFormatSizeDescription
CBDD_1330_sm_FigS1-S10-TableS1-S2.doc1957KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.