Pharmaceutical Sciences 79 (2015) 1–12
Contents lists available at ScienceDirect
Pharmaceutical Sciences journal homepage: www.elsevier.com/locate/phasci
3D-QSAR (CoMFA, CoMFA-RG, CoMSIA) and molecular docking study of thienopyrimidine and thienopyridine derivatives to explore structural requirements for aurora-B kinase inhibition Ankit Borisa, Hardik Bhatt ⁎ Department of Pharmaceutical Chemistry, Institute of Pharmacy, Nirma University, Ahmedabad 382 481 India
a r t i c l e
i n f o
Article history: Received 23 June 2015 Received in revised form 31 July 2015 Accepted 28 August 2015 Available online 4 September 2015 Keywords: Cancer Aurora-B kinase CoMFA-RG CoMSIA Thienopyrimidine and thienopyridine Molecular docking
a b s t r a c t Aurora-B kinase plays a crucial role in cell cycle events and is identified as an important factor in regulation of spindle check point assembly. Thus, it can be proved as an important target in the field of oncology. 3D-QSAR model was generated using 54 molecules reported in literature containing thienopyrimidine and thienopyridine as scaffolds. All molecules were aligned using Distill function in Sybyl X1.2. This generated best model of CoMFARG (Region focusing) and CoMSIA were statistically significant with correlation coefficient r2ncv of 0.97, for both & Leave one out coefficient (LOO) q2 of 0.70 and 0.72, respectively. Best CoMSIA model was built up using various combination of descriptors and proved statistical significant among all models. Best CoMFA-RG and CoMSIA models were validated by 12 test set molecules giving satisfactory prediction (r2pred) values of 0.86 and 0.88, respectively. External test set validation was performed using 20 molecules and satisfactory prediction of their biological activity was found. Active compounds were docked on protein (PDB ID: 4C2V) by GOLD module and revealed important interactions with amino acids at ATP-binding region. These data explored insight requirements for Aurora-B inhibition which might be fruitful for understanding mechanisms with kinase ligand interactions. © 2015 Elsevier B.V. All rights reserved.
1. Introduction Aurora kinase is a member of the enzyme serine/threonine kinase family which controls cell division (Carmena and Earnshaw, 2003). Therapeutic inhibition of Aurora kinases shows great promise as an apparent anticancer regime because they play an important role during cell division at G2/M phase of the cell cycle and their regulation process by various substrate and co-factors which are observed throughout cell division (Katayama et al., 2003). Over expression or deregulation of Aurora kinases are observed in human cancers. In fungi, only one Aurora-1 member is identified but in mammals, 3 Aurora kinases are identified viz. Aurora-A, Aurora-B, Aurora-C (Carpinelli and Moll, 2008). They have an evolutionary conserved N-terminal domain and catalytic domain which vary in sequence and in length (Shan et al., 2012). Since its discovery in 1995, Aurora kinase is explored as a target to inhibit cancer as well as various processes involve in cell cycle. Aurora-A and B are highly expressed in majority of human cancer cell lines (Lens et al., 2010). Aurora-B is a member of the Aurora family present at chromosome 17p13.1. Aurora-B, present in mammalian cells, acts as a chromosome ⁎ Corresponding author at: Institute of Pharmacy, Nirma University, S. G. Highway, Chharodi, Ahmedabad, 382 481, India. E-mail address:
[email protected] (H. Bhatt).
http://dx.doi.org/10.1016/j.ejps.2015.08.017 0928-0987/© 2015 Elsevier B.V. All rights reserved.
passenger complex. Three co-factors, viz. inner centromere protein (INCENP), borealin and survivin are crucial for actions of Aurora-B (Teperek-Tkacz et al., 2010). Aurora-B phosphorylation is associated with co-factors viz. INCENP and survivin. Survivin is an interesting protein co-factor for inhibition of apoptosis, but still data are conflicted in various studies. Aurora-B has three important functions called Histone H3 phosphorylation, cytokinesis and spindle checkpoint kinase. Functional importance of Aurora-B also includes following but not limited to chromosome condensation, sister chromatid cohesion, mitotic spindle assembly, promoting chromosome bi-orientation, merotelic chromosome attachments and spindle assembly checkpoint (Gassmann et al., 2004). Different studies suggested that Aurora-B is a master regulator of mitosis. Though linkage of Aurora-A in cancer is strongest; inhibition of transcripts of Aurora-B, observed in multiple tumours, either by small molecules or by RNA interference, lead to cells into endoreduplication followed by cell death. Thus, Aurora-B received much more focus in cancer study (Kwiatkowski et al., 2012). Various drug like candidates targeting Aurora-B in various phases of clinical trials are Barasertib (1), PHA-739358 (2) and Hesperadin (3) which are shown in Fig. 1 (Keen and Taylor, 2004). 3D-QSAR is a very important term which calculated all possible descriptors which could be correlated in SAR studies with the proof of experimental results (Du et al., 2012). Well known two techniques
2
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
(Gasteiger and Marsili, 1980). All molecules were align on the Nmethyl-1-(4-methyltetrahydrothiophene-3-yl)methane diamine fragment using Distill alignment technique. Align dataset was shown in Fig. 2. (Statistical comparison of three different alignment methods viz. Distilled based rigid body, Flex – Docking based and Pharmacophoric multi-fit alignment was given as Table S1 in supplementary information.) 2.3. CoMFA studies
Fig. 1. Clinical trial candidates of Aurora-B.
used for 3D-QSAR study are Comparative Molecular Field Analysis (CoMFA) (Cramer et al., 1988) and Comparative Molecular Similarity Indices Analysis (CoMSIA) (Klebe and Abraham, 1999; Klebe et al., 1994). CoMFA and CoMFA-RG (Region focusing analysis) analysis are carried out using electrostatic energy field (Coulombic) and steric energy field (van der Waals) of the molecules. This analysis counts structural resemblance of positive and negative region, calculated as activity in favour (Stanton, 2012). CoMFA-RG analysis calculates it by creating specific grid space at the specific lattice points. CoMSIA also counts similarity of probe atoms, like CoMFA and analyse molecules that are counted through-out the regular grid space. In comparison with CoMFA, CoMSIA is differentiated by Gaussian functions where no cut off is required. Alignment or superimposition of the molecule is most important part because contour maps created by the system are dependent on the structural alignment of the molecules (Davis and Vasanthi, 2015). The purpose of this work is to correlate relationship between structural properties and biological activity of Aurora-B kinase inhibitors by using ligand-based 3D-QSAR approach by CoMFA, COMFA-RG and CoMSIA methods and molecular docking tools. These study might be useful to design more selective Aurora-B inhibitors. 2. Material and methods 2.1. Data set and selection of training and test set A data set of 54 thienopyrimidine and thienopyridine derivatives was used for present study (Curtin et al., 2012; McClellan et al., 2011). Their experimental structures with its inhibition of Aurora-B by Homogenous Time Resolve Fluorescence (HTRF) enzymatic assay method were listed in Tables 1 and 2. The IC50 values were reported in nM and were converted in to pIC50 (= − logIC50). The pIC50 values of the molecules in the present study traversed a wide range which helped in preparation of improved derivatives with enhanced activity. All structures of both series were divided into a training set consisting of 42 molecules to generate the QSAR model and 12 molecules were used as a test set to validate the quality of the generated model. The test set molecules were marked with asterisk in the Tables 1 and 2. Selection of training and test set was carried out in such a manner that compounds of test set resembled compounds of training set in multidimensional descriptor space and all representative compounds of training set resembled compounds of test set. Thus, a test set was a true demonstrative of a training set. This was achieved by randomly setting aside test set compounds with a distributed biological data (Patel et al., 2014). 2.2. Molecular modelling and alignment All molecular modelling studies including structure drawing were performed using Sybyl X1.2 molecular modelling software by Tripos, Inc., St. Louis, MO. Molecules having no defined value were dropped from the study. All structures of the series were drawn and energy minimization was performed using force field and Gasteir-Huckel charge
CoMFA analysis calculated steric and electrostatic fields at each lattice of a regularly spaced grid of 2.0 Å in all Cartesian directions. These grid points generated with Tripos force field using sp3 carbon with a van der Waal’s radius of 1.52 Å using net +1 charge. The default energy cut off was set at 30 kcal/mol for both steric as well as electrostatic fields which was the optimal parameter for the model (Stahle and Wold, 1988). 2.4. CoMFA region focusing (COMFA-RG) Region focusing is an advanced tool available for the refinement of the developed CoMFA model in 3D-QSAR technique. Here, the developed model was refined in COMFA-RG which showed statistically significant results along with refined contours generated by Sybyl X. Major advantage of CoMFA-RG is to improve q2 value of generated CoMFA model and to refine the CoMFA model by considering important lattice points which were most pertinent to the model (Atabati and Sharifi, 2014). 2.5. CoMSIA studies CoMSIA calculated the binding affinity of molecules which changed their molecular properties in the field. In CoMSIA method, five different fields are considered, from similar active molecules, to develop a model. These fields were steric (S), electrostatic (E), hydrophobic (H), hydrogen bond acceptor (A) and hydrogen bond donor (D). In this study, the attenuation factor, by default, was set at value of 0.3 (Zheng et al., 2011). Gaussian functions were used to determine the distance between molecule atoms and probe atoms while similarity positions inside and outside could be calculated at all grid points (Balasubramanian et al., 2015). 2.6. Partial least square analysis Analysis of 3D-QSAR results was carried out by partial least square (PLS) regression method (Bhatt and Patel, 2012; Liu et al., 2015). Evaluation of CoMFA and CoMSIA descriptors was carried out using Leaveone-out (LOO) method (q2) where one compound was removed from the structure series one by one and predicted activity of the developed model. PLS was also checked using cross-validation method which calculated maximum number of the latent variables used in this model. Optimum number of latent variables was used for correct correlation coefficient (r2cv) (Shibi et al., 2015). r 2 cv ¼ 1–
X 2 X 2 Yobs –Ypre = ðYobs –Ymean Þ
where, Yobs is observed value, Ypre is predicted value, Ymean is mean value of the given activity data or target property (pIC50). In the next step, after getting optimum number of latent variables, PLS calculations were carried out with no validation with column filtering to create maximum correlation coefficient (r2ncv). Along with this maximum correlation coefficient, Sybyl computed F value and standard error of estimation (SEE), which described explained versus unexplained variables (Ståhle and Wold, 1987) In CoMFA analysis, descriptive values were used as independent variables and pIC50 values were used as dependent variables. In CoMSIA analysis; steric, electrostatic
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
3
Table 1 Experimental and predicted pIC50 values of thienopyrimidine derivatives.
Compd. No.
R
R2
Actual pIC50
Predicted pIC50 CoMFA-RG
CoMSIA
–H
7.51
8.68
7.50
2.
–H
8.26
8.25
8.20
3.⁎
–H
8.40
8.38
8.33
1.
–H
R1
4.
–H
–H
9.30
9.41
9.25
5.
–H
–H
9.70
9.07
9.60
6.⁎
–H
–H
9.70
9.79
9.96
7.
–H
–H
9.10
9.11
9.13
8.
–H
–H
9.22
9.25
9.30
9. ⁎
–H
–H
7.06
6.82
8.77
10.
–H
–H
9.30
8.92
9.41
11.
–H
–H
9.70
9.67
9.85
12.
–H
8.11
8.11
8.12
13.
–H
8.54
8.55
8.61
14.
–H
–H
7.24
7.23
7.44
15.
–H
–H
7.46
7.89
7.36
16.⁎
–H
–H
8.42
8.22
8.07
17.
–H
–H
7.54
7.70
7.90
18.
–H
–H
9.15
9.02
8.99
19.
–H
–H
7.72
7.80
7.22
(continued on next page)
4
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
Table 1 (continued)
Compd. No.
R
R1
R2
Actual pIC50
Predicted pIC50 CoMFA-RG
CoMSIA
20.
–H
–H
8.89
8.72
8.99
21.⁎
–H
–H
9.00
8.67
7.44
22.
–H
–H
7.64
7.81
7.75
23.
–H
–H
8.34
8.29
7.48
⁎ Test set molecules.
and hydrophobic descriptors were used as independent variables and dependent variables were same as CoMFA (Vyas et al., 2013). In the evaluation of predicted values of CoMFA and CoMSIA models, the r2pred was based on the test set of molecule calculated by r2pred equation (Liu et al., 2015). r 2 pred ¼ 1–ðPRESS=SDÞ where, PRESS denotes as sum of square deviation between predictive and experimental activities of the test molecules. SD stands for sum of square deviation between its biological activity of test molecules and activity in mean of training molecules. Contour maps were generated graphically after models were developed in CoMFA, CoMFA-RG and CoMSIA and coefficients were produced using field type “STDEV*COEFF”. 2.7. Docking analysis 2.7.1. Docking protocol generation and validation To find a suitable binding alignment and conformation of thienopyrimidine and thienopyridine analogues to interact with Aurora-B, a robust and reliable computational method is required. The advanced molecular docking programme GOLD (version 5.2), with powerful genetic algorithm (GA) method for conformational search and docking programme, was used to generate and ensemble of docking conformations. Fig. 3 represented docking process and its validation protocol. Aurora-B kinase crystal structure was retrieved from the RSCB Protein Data Bank (PDB ID: 4C2V). Barasertib, extracted from the X-ray structure of the Aurora-B kinase complex, was re-docked into its binding site using GOLD docking to validate the docking protocol and binding of co-crystalized ligand in the protein. During docking of standard ligand in GOLD software, actual orientation of reference molecule Barasertib was used for validation. Once the parameters for docking were set, study was carried out to check the binding orientation of Barasertib on crystal structure of protein obtained from database. The original ligand Barasertib and water molecules were removed from the protein complex. 2.7.2. Docking protocol for thienopyridine and thienopyrimidine urea Docking experiments performed using default GOLD parameters (Balupuri et al., 2014). GOLD score was used as the fitness functions
with internal hydrogen bonds being allowed. The binding site was defined as being within 6 Å of the bound ligand which was extracted from the protein crystal. Analysis of crystal structure from literature revealed important interactions of Barasertib with various amino acids like Lue99, Lys122, Gln145, Lue154, Lue168, Phe172, Ala173, Gly176 and Lue223. This is shown in Fig. 4. From all above amino acids, Ala173 was most important amino acid for interaction and responsible to change conformation of the Aurora-B. In case of ATP binding, Ala173 showed conformation change at α-C helix region (Sessa and Villa, 2014). Compounds 3, 5 and 54 were docked on the binding site of Aurora-B kinase to obtain the GOLD score of conformations, fitness score of docked compounds and interactive amino acids. Data is shown in Table 3. Reason to choose these compounds is to access the largest substructure from the dataset to view all around binding in PDB core which resembles the common structures as well as substructures. Both compounds, 3 and 53, have substitution on thiophene ring and −1-(3-methyl-1H-pyrazol-1-yl)propan-2-ol substitution on ringA, respectively, which covered the whole dataset and both showed good activity, as reported in literature. 3. Results and discussion The predicted and experimental activity values of the training and test set compounds for CoMFA, CoMFA-RG and CoMSIA are listed in Table 4 and a correlation between predicted and experimental activity of training and test set is depicted in Fig. 5. 3.1. CoMFA and CoMFA-RG results In CoMFA-RG, the best results of steric and electrostatic fields were achieved at column filtering of 2.0 kcal/mol. The maximum number of latent variables was 5 in Leave one out mode. By this latent variables, q2 value was calculated as 0.70 in PLS analysis in region focusing CoMFA which is greater than q2 value of normal CoMFA (q2 = 0.51). The no-validation or non-crossover validate PLS calculation results were found better in CoMFA-RG as compared to CoMFA. These statistics significantly proved that generated CoMFA-RG model contour maps were taken in consideration for design. The no-validation or noncrossover validate PLS calculation results were found momentous in CoMFA-RG than CoMFA. No validation r2ncv value was 0.97, standard
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
5
Table 2 Experiment and predicted pIC50 values of thienopyridine urea derivatives.
Compd. No.
R
R1
Actual pIC50
Predicted pIC50 CoMFA-RG
CoMSIA
6.31
6.27
6.29
25.⁎
7.10
7.55
7.08
26.
7.96
7.95
7.97
27.
7.89
8.06
7.88
28.⁎
5.48
4.02
5.46
29.
7.00
7.09
7.07
30.
6.75
6.89
6.70
31.
7.17
7.15
7.22
32.
7.18
7.49
7.52
33.
8.10
7.45
7.99
34.
7.80
8.34
7.90
35.⁎
8.10
7.45
8.15
36.
7.30
7.24
7.40
37.
8.22
7.93
7.86
38.
7.74
7.73
7.58
39.⁎
8.05
7.33
7.84
40.
7.15
7.32
7.54
41.
7.32
7.57
7.48
42.
7.28
7.37
7.41
24.
-H
(continued on next page)
6
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
Table 2 (continued)
Compd. No.
R
R1
Actual pIC50
Predicted pIC50 CoMFA-RG
CoMSIA
43.⁎
6.94
6.89
6.89
44.
7.77
7.72
7.72
45.
7.64
7.55
7.74
46.⁎
7.51
7.44
7.97
47.
6.75
6.67
7.70
48.
6.28
6.29
6.22
49.
7.96
7.88
7.94
50.
8.10
8.17
8.40
51.
8.30
7.66
8.33
52.
8.15
8.02
8.20
53.
9.00
7.87
9.00
54.⁎
8.70
8.77
8.61
⁎ Test set molecules.
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
7
actual q 2 representing numbers of latent variables used in model for validation and it was not much effected if they calculated randomly. Thus results clearly showed much higher range of predictive accuracy in r2pred.The contribution of steric and electrostatic fields was 0.46 and 0.54, respectively in CoMFA-RG.
3.2. CoMSIA results
Fig. 2. Alignment of training and test set compounds using Distill.
error of estimation (SEE) value was 0.15 and F value was 199.34 in CoMFA-RG whereas in CoMFA, values were 0.97, 0.19 and 188.79 for r 2ncv, standard error of estimation and F value, respectively. Low SEE value and higher r 2ncv value represented that generated model had very good predictive ability of activity in compare with actual activity as well as low margin of standard error. Furthermore, value of cross validated r 2cv CoMFA-RG model was 0.68 (while for CoMFA, r 2cv value was 0.57) and r 2pred value was 0.86 (while for CoMFA, r 2cv value was 0.78). Cross validation of these generated model was quite near to
To evaluate best CoMSIA model, various combinations with field contribution analysis were taken into the consideration for development. All five fields were analysed and generated 15 different models which were then used for result analysis. The statistical result of various models of CoMSIA are showed in Fig. 6 and statistical values are given in supplementary information Table S2. Five field parameters were used in model 15 which was best model among all with superior in all aspect of analysis. It contained maximum number of latent variables 5. The highest q2 value was 0.72 in leave one out method at column filtering of 2.0 kcal/mol. A high no cross over validation r 2ncv value was 0.97 with standard error of estimation value of 0.17 and F value of 228.14. Furthermore, the cross over validation value r2cv was 0.74 which was very good at the group of 10 compounds. Here, cross validation results of developed model was more than q2 value which represented much more reliability on CoMSIA contour maps. Contribution of steric, electrostatic, hydrophobic, hydrogen bond acceptor and hydrogen bond donor fields are 0.16, 0.24, 0.21, 0.22, 0.17, respectively. In PLS analysis,
Fig. 3. Step wise molecular docking process with validation.
Fig. 4. Comparison of amino acid interactions of Barasertib obtained from protein databank and docking of extracted Barasertib on PDB ID: 4C2V.
8
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
Table 3 Docking results of Barasertib along with compounds 3, 5 and 54. Compound
GOLD score
GOLD Fitness score
Interactions with amino acids
RMSD⁎
Barasertib 3 5 54
−13.26 −12.50 −13.10 −12.99
78.82 75.29 80.34 91.56
Glu141, Lue223, Ala173 , Pro174 Lue99, Gly145, Ala173, Glu177, Asp234 Lys122, Glu141, Ala173, Lue223, Ala233 Lys122, Ile142, Gln145,Glu171, Ala173, Pro174
0.8 2.1 1.9 3.8
⁎ RMSD = root mean square distance in Å.
predictive correlation coefficient r2pred value was 0.88 for five latent variable of test sets. Model 1 of the CoMSIA analysis showed lowest statistical figures given in supplementary information Table S1 which depicted that combination of steric, electrostatics and H-bond donor fields didn’t match statistically. Some other models had significant values like model 11 which had r2ncv value of 0.92, q2 value of 0.66, r2cv value of 0.71, r2pred value of 0.83 but this model had higher SEE (0.21) and lower F value (187.36) than the model 15. Here, electrostatic, hydrophobic and hydrogen bond acceptor contributions are more than steric and hydrogen bond donor in the model. Statistics of CoMFA and CoMSIA represented that the predictive ability of generated model was very good. 3.3. Contour analysis In results of the developed model of CoMFA and CoMSIA, their contour maps were generated at 3D grid orientation for proper visualization. The CoMFA and CoMSIA outcomes were graphically interpreted by field contribution maps using the STDEV*COEFF type of field. Here, graphs displayed steric and electrostatic region of CoMFA-RG model in Fig. 7A–B with compound 6 in the map visualization. In the CoMSIA model, Fig. 7C–E showed contour maps of hydrophobic, hydrogen bond donor and hydrogen bond acceptor fields. All maps of contour generated in the field were 80% and 20% contributions for favourable and unfavourable regions, respectively.
similar potency, when aligned in the steric contours, exhibited green contour map all over the structures with similarity in the orientation of structures. In Fig. 7B, the electrostatic field was indicated by red and blue contours which indicated electron-withdrawing and electrondonating groups would be favourable and increased inhibitory activity, respectively. In electrostatic contours, compound 6 with amino group on ring-A as well as amide bond formed on both side of phenyl ring favoured the activity most. Compounds 52 (pIC50 = 8.15) and 53 (pIC50 = 9.00) having pyrazole substituents with alcohol and at the end of 3-position having fluorine atom on phenyl ring increased activity. Compounds 3 (pIC50 = 8.40) showed good activity because of the presence of electron withdrawing groups having positive charge, in comparison to compounds 47 (pIC50 = 6.75) and 48 (pIC50 = 6.28) where electro donating groups like butyl and iso-pentyl were present at the terminal points reduced activity. 3.3.2. CoMSIA contour map analysis CoMSIA steric and electrostatic contour maps along with the most active compound 6 showed similar results like CoMFA. So discussion for these two maps would be same. Discussion of hydrophobic contours explained that yellow and grey contours specifying hydrophobic and hydrophilic groups, respectively, were favoured by generated model as shown in Fig. 7C. At the ring B, yellow position favoured hydrophobic part containing aromatic ring which showed good activity and grey
3.3.1. CoMFA-RG contour map analysis CoMFA-RG contour maps were generated in three dimensional space around molecules and steric and electrostatic fields were predicted to determine increase or decrease in the value of activity. In Fig. 7A, with compound 6, the green contours (80% contribution) indicated that a steric contribution increased potency, while the yellow contours (20% contribution) indicated regions of steric hindrance which decreased activity. This could be explained by the fact that compounds 5 (pIC50 = 9.67) and 6 (pIC50 = 9.67) having 2-methyl and 3-methoxy substitution at terminal, respectively, favoured the activity while in compound 28 (pIC50 = 5.48), extra bulk of aromatic ring at terminal end of ring A showed less inhibitory activity. Compound 48 (pIC50 = 6.28) having bulk of pyrazole-N-alkyl substitution adversely affect the activity. Moreover, compounds 5, 6 and 11 (pIC50 = 9.67), having Table 4 Statistical results of CoMFA, CoMFA-RG and CoMSIA models by PLS analysis. PLS statistics
CoMFA
r2ncv
0.97 0.97 0.19 0.15 188.79 199.34 0.51 0.70 0.57 0.68 0.78 0.86 3 5 Field contribution 0.43 0.46 0.57 0.54 -
Standard error of estimation F value q2 r2cv r2pred N Steric Electrostatic Hydrophobic H-bond donor H-bond acceptor
CoMFA-RG
CoMSIA 0.97 0.17 228.14 0.72 0.74 0.88 5 0.16 0.24 0.21 0.22 0.17
Fig. 5. Plot of experimental pIC50 versus predictive pIC50 of all compounds; (A) CoMFA-RG model; (B) CoMSIA model.
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
9
Fig. 6. Comparative results of descriptors of CoMSIA models.
colour at amide linkage with substituted aromatic ring favoured hydrophilicity which is present in almost all molecules. The fact was satisfied after comparing compounds 50 (pIC50 = 8.10), 53 (pIC50 = 9.00) and 54 (pIC50 = 8.70), where all compounds were similar in orientation and contained hydrophilic moiety - pyrolidine-3-ol which increased activity, while hydrophobic moieties - phenyl, furan and thiophene decreased activity on same position of ring A which was explained by compounds 28 (pIC50 = 5.48), 29 (pIC50 = 7.00) and 30 (pIC50 = 6.74). The cyan contours of Fig. 7D represented hydrogen bond donating groups which increased activity and purple contours represented hydrogen bond donating group which decreased activity. Most of the compounds from the series had secondary amino group presented in the form of amide bond which would be important for the activity because compounds without amide linkage did not produce good inhibition. Examples were compounds 1 (pIC50 = 7.50), 5 (pIC50 = 8.26) and 7 (pIC50 = 8.40) having 3-methylpyrazole moiety on ring A which exhibited good activity as compared to others. Here, substitution on thiophene moiety of Ring A is preferable for good activity. Compound 43 (pIC50 = 6.94) with terminal methoxy group on phenyl ring decreased activity. The magenta contours of Fig. 7E indicated the regions where hydrogen bond accepting groups increased activity and red contours
showed hydrogen bond accepting group decreased activity. Compounds with substituents on ring A of thienopyrimidine and thienopyridine favoured activity. Compounds 2 (pIC50 = 8.26), 3 (pIC50 = 8.38) and 53 (pIC50 = 9.00) having pyrazole and 3-methylpyrazole groups on Ring A showed good inhibition activity. 3.4. External test set validation Validation of the developed model is crucial part at the last stage of the 3D-QSAR method. Final models of CoMFA-RG and CoMSIA, generated by the Sybyl X1.2, found fully significant with statistics of test set containing 20 molecules and their r2pred, however, if those models might not be predicted the activity of other similar compounds then these generated models cannot be used anymore. A true and trustworthy model should be able to predict precise activity in external test set. These external test set should contain compounds having almost similar series and their biological activity should be available in specific values. A new external test set was selected having thienopyridine series which consisted of 20 molecules described in Table 5 with their actual pIC50 value. Those test molecules were excluded in developed 3D-QSAR model and they were introduced to predict their activity using
Fig. 7. CoMFA-RG STDEV*COEFF contour maps. (A) Favourable (green) and unfavourable (yellow) steric field. (B) Electropositive (blue) and electronegative (red) fields. CoMSIA STDEV*COEFF contour maps. (C) Favourable (yellow) and unfavourable (grey) hydrophobic field. (D) Favourable (cyan) and unfavourable (purple) hydrogen bond donor fields. (E) Favourable (magenta) and unfavourable (red) hydrogen bond acceptor fields. Compound 6 was overlaid in each map.
10
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
Table 5 External test set for validation of CoMFA-RG and CoMSIA (model 15).
Sr. no
R
R1
Actual pIC50
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
–NH2 –NHCOPh –NHCONHPh –NHCONH(3-MePh) –NHCON(Me)(3-MePh) –N(Me)CONH(3-MePh) –NHCOCH2(3-MePh) –CH2CONH(3-MePh) –NHC(S)NH(3-MePh) –NHCONH(4-MePh) –NHCONH(2-MePh) –NHCONH(3-Et-Ph) –NHCONH(3-F-Ph) –NHCONH(3-MePh) –NHCONH(3-MePh) –NHCONH(3-MePh) –NHCONH(3-MePh) –NHCONH(3-MePh) –NHCONH(3-MePh) –NHCONH(3-MePh)
–H –H –H –H –H –H –H –H –H –H –H –H –H Ph 4-(OH)Ph 3-(CONHMe)Ph 3-Pyridyl 4-Pyridyl 5-Pyrimidyl 4-Isoquinolyl
5.31 4.92 7.25 8.05 5.31 7.64 6.28 7.03 6.39 7.59 6.82 7.60 7.82 7.89 7.96 7.96 8.05 8.30 8.10 6.41
developed models of CoMFA-RG and CoMSIA (Model 15). These 20 compounds (Heyman et al., 2007) were selected purposely since they have similar range of activity like in training and internal test set of actual study and also compounds of external set having steric, hydrophobic and electrostatic factors. Table 5 listed the predictive activity values in pIC50 for external data set and values were reasonably well for all compounds except compound 5, which was the biggest outlier from the external test set. Structurally, this compound has N-methyl group on the amide nitrogen which falls in un-favourable hydrophobic contour region. Along with this, it does not fit in symmetry of the aligned data set molecules. Activity of all other compounds of external test set was accurately predicted by CoMFA-RG and CoMSIA. These generated models can accurately predict activity of novel design compounds. 3.5. Docking results Barasertib was used as a reference molecule to validate the docking protocol and binding of co-crystalized ligand in the protein. Fig. 4 represented comparison of amino acid interactions of Barasertib obtained
Predictive pIC50 CoMFA-RG
CoMSIA Model 15
6.84 5.02 7.37 8.53 7.32 7.51 6.79 6.98 6.12 7.16 6.92 7.58 7.92 7.82 7.84 7.79 8.35 8.49 7.85 7.29
6.21 5.86 7.59 8.11 6.99 7.63 6.11 6.82 6.26 7.40 7.02 7.80 7.62 8.02 8.35 8.02 8.30 8.13 7.54 6.79
from protein databank and extracted Barasertib from the X-ray structure of the Aurora-B kinase complex (PDB ID: 4C2V) which was re-docked into its binding site using GOLD. RMSD (Root Mean Square Distance in Å) of the docked ligand was 0.8 which represented that current protocol is having good accuracy. Terminal phenyl ring of ligand showed two hydrogen bond interactions with Glu141 at distances of 2.872 Å and 2.576 Å; in the hydrophobic pocket, one hydrogen bond interaction with Lue223 at a distance of 2.863 Å and one hydrogen bond interaction with Pro174 at a distance of 2.698 Å. Nitrogen atom of the core quinazoline ring formed a hydrogen bond with Ala173 which was crucial in this pocket as shown in Fig. 4. GOLD and fitness scores of reference molecule Barasertib along with compounds 3, 5 and 53 are given in Table 3 with their RMSD values. GOLD docking analysis of Barasertib showed maximum docking score of − 13.26 with fitness of 78.82. Compound 3 showed GOLD score of − 12.50 and fitness of 75.29, Compound 5 showed GOLD score −13.10 with fitness of 80.34 and compound 53 showed −12.99 score with maximum fitness of 91.56. Docking of compounds 3, 5 and 53 along with their hydrogen bond interactions are shown in Fig. 8. Pyrazole moiety of compound 3 was
Fig. 8. Docking interactions of compounds 3, 5 and 53.
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
buried under the hydrophobic pocket and showed interactions with Gly177 with hydrogen bond at a distance of 2.462 Å and 2.376 Å and with Lue99 at distance of 2.43 Å and 2.607 Å. It also supports hydrophobic contours represented as grey colour in Fig. 7. Thienopyrimidine core showed interactions with Ala173 by forming two hydrogen bonds at distances of 3.009 Å and 2.584 Å. Compound 3 favoured the electrostatic red contours and this is supported by acidic amino acid interactions observed with Asp234 at a distance of a 3.024 Å. Polar interaction was also observed with Terminal phenyl ring and Gly145 at a distance of 2.698 Å. Compound 5 was most active molecule in list which showed great score and fitness. Scaffold of this compound bound in the hydrophobic pocket interacted with Ala173 at a distance of 2.458 Å. Phenyl ring at the terminal point, like compound 3, showed interactions with polar Glu141 at distance of 2.6 Å and 2.608 Å. Carbonyl group of amide interacted with hydrophobic pockets with positively charged amino acid Lys122 and nonpolar Lue223 at distances of 2.528 Å and 2.514 Å, respectively. These data supported the hydrogen bond donor contours. Another active compound 53 also showed all interactions around ATP-binding pocket. Main interactions with amino acids included Pro174 which formed hydrogen bond at a distance of a 2.284 Å with terminal pyrrolidine; thienopyrimidine moiety formed hydrogen bond with Aln173 at a distance of a 2.875 Å; amino group of scaffold formed bond with Glu171 at a distance of a 2.918 Å, carbonyl group of amidebridge formed hydrogen bond with positively charged Lys122 at a distance of 2.985 Å and amine formed hydrogen bond in hydrophilic pocket with Gln145 at a distance of 2.977 Å. Terminal floro group of phenyl ring also showed interaction in hydrophobic pocket with Ile142 at a distance of 2.550 Å and it was represented as favoured region in hydrophobic contours in CoMSIA generated model. Docking data (GOLD score and GOLD fitness) of all compounds is provided as supplementary information Table S3. Molecular docking study is also showing good correlation with actual pIC50 values of all compounds. This comparison revealed that compound 5 having best pIC50 value have highest GOLD score. Top few compounds from the list of pIC50 showed best GOLD score in docking results which proved that there is a correlation between QSAR and molecular docking studies. Similarly, compounds with low pIC50 values showed low GOLD score. Comparison of pIC50 values of all compounds with GOLD score in descending order is given in supplementary information Table S4. 4. Conclusion In conclusion, current studies have established good CoMFA as well as advanced CoMFA-RG and CoMSIA predictive models which are consistent to guide for further substitution in the molecules for designing better drug like compounds. CoMFA, CoMFA-RG and CoMSIA statistics resulted in terms of q2 and r2cv values, suggested significant correlation of molecules with their inhibitory activities. The final model had good internal validity with q2 value of 0.51, 0.70 and 0.72 for CoMFA, CoMFA-RG and CoMSIA, respectively and high predictive ability of the test set (r2pred) was 0.78, 0.86 and 0.88, respectively. Results of 3D-QSAR showed hydrophobic, electrostatic, steric, hydrogen bond donor and acceptor substitutions were significant to inhibit the effect of Aurora-B kinase. GOLD docking analysis also showed good score as well as fitness of active compounds with important binding interactions with amino acids in ATP-binding region. This generated model and docking results showed insight into to the molecular structures and their activity relationship for future design of new inhibitors which might be proved as potent and specific Aurora-B inhibitors. Acknowledgement The authors are thankful to Nirma University, Ahmedabad, India for providing necessary facilities to carry out the research work.
11
Appendix A. Supplementary data Statistical parameters of comparative study of three alignments using CoMFA and CoMSIA model by PLS analysis (Table S1); Various field combinations and their statistical parameters of CoMSIA models (Table S2); Docking data (GOLD score and GOLD fitness) of all compounds along with Barasertib (Table S3); Comparison of pIC50 values of all compounds with GOLD score in descending order (Table S4); MOL2 files of all structures used to generate 3D-QSAR models and 2 files (.tbl) generated by Sybyl are given as supplementary information in support of the research work. Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10. 1016/j.ejps.2015.08.017.
References Atabati, M., Sharifi, S., 2014. 3D-QSAR study on VEGFR kinase inhibition of aminopyrazolopyridine urea derivatives by CoMFA and CoMSIA. Med. Chem. Res. 23, 3539–3545. http://dx.doi.org/10.1007/s00044-014-0933-0. Balasubramanian, P.K., Balupuri, A., Gadhe, C.G., Cho, S.J., 2015. 3D-QSAR modeling study on 7-aminofuro [2,3-c] pyridine derivatives as TAK1 inhibitors using CoMFA and COMSIA. Med. Chem. Res. 24, 2347–2365. http://dx.doi.org/10.1007/s00044-014-1221-8. Balupuri, A., Balasubramanian, P.K., Gadhe, C.G., Cho, S.J., 2014. Docking-based 3D-QSAR study of pyridyl aminothiazole derivatives as checkpoint kinase 1 inhibitors. SAR QSAR Environ. Res. 25, 651–671. http://dx.doi.org/10.1080/1062936X.2014.923040. Bhatt, H.G., Patel, P.K., 2012. Pharmacophore modeling, virtual screening and 3D-QSAR studies of 5-tetrahydroquinolinylidine aminoguanidine derivatives as sodium hydrogen exchanger inhibitors. Bioorg. Med. Chem. Lett. 22, 3758–3765. http://dx.doi.org/ 10.1016/j.bmcl.2012.04.012. Carmena, M., Earnshaw, W.C., 2003. The cellular geography of Aurora kinases. Nat. Rev. Mol. Cell Biol. 4, 842–854. http://dx.doi.org/10.1038/nrm1245. Carpinelli, P., Moll, J., 2008. Aurora kinases and their inhibitors: more than one target and one drug. Adv. Exp. Med. Biol. 610, 54–73. http://dx.doi.org/10.1007/978-0-387-73898-7_5. Cramer, R.D., Patterson, D.E., Bunce, J.D., 1988. Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J. Am. Chem. Soc. 110, 5959–5967. http://dx.doi.org/10.1021/ja00226a005. Curtin, M.L., Frey, R.R., Heyman, H.R., Soni, N.B., Marcotte, P.A., Pease, L.J., Glaser, K.B., Magoc, T.J., Tapang, P., Albert, D.H., Osterling, D.J., Olson, A.M., Bouska, J.J., Guan, Z., Preusser, L.C., Polakowski, J.S., Stewart, K.D., Tse, C., Davidsen, S.K., Michaelides, M.R., 2012. Thienopyridine ureas as dual inhibitors of the VEGF and Aurora kinase families. Bioorg. Med. Chem. Lett. 22, 3208–3212. http://dx.doi.org/10.1016/j.bmcl. 2012.03.035. Davis, G.D.J., Vasanthi, A.H.R., 2015. QSAR based docking studies of marine algal anticancer compounds as inhibitors of protein kinase B (PKBβ). Eur. J. Pharm. Sci. 76, 110–118. http://dx.doi.org/10.1016/j.ejps.2015.04.026. Du, Q.-S., Gao, J., Wei, Y.-T., Du, L.-Q., Wang, S.-Q., Huang, R.-B., 2012. Structure-based and multiple potential three-dimensional quantitative structure-activity relationship (SBMP-3D-QSAR) for inhibitor design. J. Chem. Inf. Model. 52, 996–1004. http://dx.doi. org/10.1021/ci300066y. Gassmann, R., Carvalho, A., Henzing, A.J., Ruchaud, S., Hudson, D.F., Honda, R., Nigg, E.A., Gerloff, D.L., Earnshaw, W.C., 2004. Borealin: a novel chromosomal passenger required for stability of the bipolar mitotic spindle. J. Cell Biol. 166, 179–191. http:// dx.doi.org/10.1083/jcb.200404001. Gasteiger, J., Marsili, M., 1980. Iterative partial equalization of orbital electronegativity – a rapid access to atomic charges. Tetrahedron 36, 3219–3228. http://dx.doi.org/10. 1016/0040-4020(80)80168-2. Heyman, H.R., Frey, R.R., Bousquet, P.F., Cunha, G.A., Moskey, M.D., Ahmed, A.A., Soni, N.B., Marcotte, P.A., Pease, L.J., Glaser, K.B., Yates, M., Bouska, J.J., Albert, D.H., BlackSchaefer, C.L., Dandliker, P.J., Stewart, K.D., Rafferty, P., Davidsen, S.K., Michaelides, M.R., Curtin, M.L., 2007. Thienopyridine urea inhibitors of KDR kinase. Bioorg. Med. Chem. Lett. 17, 1246–1249. http://dx.doi.org/10.1016/j.bmcl.2006.12.015. Katayama, H., Brinkley, W.R., Sen, S., 2003. The Aurora kinases: role in cell transformation and tumorigenesis. Cancer Metastasis Rev. 22, 451–464. Keen, N., Taylor, S., 2004. Aurora-kinase inhibitors as anticancer agents. Nat. Rev. Cancer 4, 927–936. http://dx.doi.org/10.1038/nrc1502. Klebe, G., Abraham, U., 1999. Comparative molecular similarity index analysis (CoMSIA) to study hydrogen-bonding properties and to score combinatorial libraries. J. Comput. Aided Mol. Des. 13, 1–10. Klebe, G., Abraham, U., Mietzner, T., 1994. Molecular similarity indices in a comparative analysis (CoMSIA) of drug molecules to correlate and predict their biological activity. J. Med. Chem. 37, 4130–4146. Kwiatkowski, N., Deng, X., Wang, J., Tan, L., Villa, F., Santaguida, S., Huang, H.-C., Mitchison, T., Musacchio, A., Gray, N., 2012. Selective Aurora kinase inhibitors identified using a taxol-induced checkpoint sensitivity screen. ACS Chem. Biol. 7, 185–196. http://dx. doi.org/10.1021/cb200305u. Lens, S.M.A., Voest, E.E., Medema, R.H., 2010. Shared and separate functions of polo-like kinases and Aurora kinases in cancer. Nat. Rev. Cancer 10, 825–841. Liu, X., Chen, X., Zhang, L., Zhan, P., Liu, X., 2015. 3D-QSAR and docking studies on piperidine-substituted diarylpyrimidine analogues as HIV-1 reverse transcriptase inhibitors. Med. Chem. Res. http://dx.doi.org/10.1007/s00044-015-1381-1.
12
A. Borisa, H. Bhatt / Pharmaceutical Sciences 79 (2015) 1–12
McClellan, W.J., Dai, Y., Abad-Zapatero, C., Albert, D.H., Bouska, J.J., Glaser, K.B., Magoc, T.J., Marcotte, P.A., Osterling, D.J., Stewart, K.D., Davidsen, S.K., Michaelides, M.R., 2011. Discovery of potent and selective thienopyrimidine inhibitors of Aurora kinases. Bioorg. Med. Chem. Lett. 21, 5620–5624. http://dx.doi.org/10.1016/j.bmcl.2011.06.041. Patel, P., Chintha, C., Ghate, M., Bhatt, H., Vyas, V., 2014. 3D-QSAR study of 4H-chromen1,2,3,4-tetrahydropyrimidine-5-carboxylate derivatives as potential anti-mycobacterial agents. Med. Chem. Res. 23, 2955–2963. http://dx.doi.org/10.1007/s00044-013-0881-0. Sessa, F., Villa, F., 2014. Structure of Aurora B-INCENP in complex with barasertib reveals a potential transinhibitory mechanism. Acta. Crystallogr. Sect. F Struct. Biol. Cryst. Commun. 70, 294–298. http://dx.doi.org/10.1107/S2053230X14002118. Shan, W., Akinfenwa, P.Y., Savannah, K.B., Kolomeyevskaya, N., Laucirica, R., Thomas, D.G., Odunsi, K., Creighton, C.J., Lev, D.C., Anderson, M.L., 2012. A small-molecule inhibitor targeting the mitotic spindle checkpoint impairs the growth of uterine leiomyosarcoma. Clin. Cancer Res. 18, 3352–3365. http://dx.doi.org/10.1158/1078-0432.CCR-11-3058. Shibi, I.G., Aswathy, L., Jisha, R.S., Masand, V.H., Divyachandran, A., Gajbhiye, J.M., 2015. Molecular docking and QSAR analyses for understanding the antimalarial activity of some 7-substituted-4-aminoquinoline derivatives. Eur. J. Pharm. Sci. 77, 9–23. http://dx.doi.org/10.1016/j.ejps.2015.05.025. Ståhle, L., Wold, S., 1987. Partial least squares analysis with cross-validation for the twoclass problem: A Monte Carlo study. J. Chemom. 1, 185–196. http://dx.doi.org/10. 1002/cem.1180010306.
Stahle, L., Wold, S., 1988. Multivariate data analysis and experimental design in biomedical research. Prog. Med. Chem. 25, 291–338. Stanton, D.T., 2012. QSAR and QSPR model interpretation using partial least squares (PLS) analysis. Curr. Comput. Aided. Drug Des. 8, 107–127. Teperek-Tkacz, M., Meglicki, M., Pasternak, M., Kubiak, J.Z., Borsuk, E., 2010. Phosphorylation of histone H3 serine 10 in early mouse embryos: active phosphorylation at late S phase and differential effects of ZM447439 on first two embryonic mitoses. Cell Cycle 9, 4674–4687. Vyas, V.K., Bhatt, H.G., Patel, P.K., Jalu, J., Chintha, C., Gupta, N., Ghate, M., 2013. CoMFA and CoMSIA studies on C-aryl glucoside SGLT2 inhibitors as potential anti-diabetic agents. SAR QSAR Environ. Res. 24, 519–551. http://dx.doi.org/10.1080/1062936X.2012. 751553. Zheng, J., Xiao, G., Guo, J., Zheng, Y., Gao, H., Zhao, S., Zhang, K., Sun, P., 2011. Exploring QSARs for 5-lipoxygenase (5-LO) inhibitory activity of 2-substituted 5-hydroxyindole-3carboxylates by CoMFA and CoMSIA. Chem. Biol. Drug Des. http://dx.doi.org/10.1111/j. 1747-0285.2011.01146.x.
