QSAR studies on pyrazole-4-carboxamide derivatives as Aurora A kinase inhibitors

Peer review under responsibility of Taibah University.

Abstract

Cancer is a cellular disease characterized by the progressive, persistent, abnormal, purposeless, and uncontrolled proliferation of tissues. 2D-QSAR was performed on a series of N,1,3-triphenyl-1H-pyrazole-4-carboxamides as anticancer agents to gain insights into the structural determinants and their structure–activity relationship. The 2D-QSAR model using 20 molecules in the training set gave an r² value of 0.8059 and a pred_r² value of 0.7861. In models, it is shown that bulky electron-withdrawing substituents at the R1 and R2 position of the chain are favoured for inhibitory activity. Hence the model proposed in this work provides important structural insights into designing novel pyrazole-4-carboxamide derivatives with specific Aurora A kinase inhibition.

Keywords:

• Triphenyl-1H-pyrazole-4-carboxamide
• 2D-QSAR
• Aurora A kinase
• Anticancer agents

1 Introduction

The mammalian Aurora kinases comprise a family of serine/threonine kinases that are essential for coordinated mitotic progression [1]. As a potential target in anticancer therapy, these kinases have generated significant interest in the cancer research field due to their elevated expression profiles in many human cancers [2]. Safe and effective anticancer agents are still one of the challenges in the field of drug discovery. Small molecule kinase inhibitors have been successfully introduced into the drug market as selective anticancer agents with lower side effects [3]. The Aurora kinase family consists of highly homologous Ser/Thr protein kinases, which play a key role in regulating mitosis. Three members of the Aurora kinase family have been identified up to now in mammalian cells, namely Aurora A, B and C [4–6]. The biological roles of Aurora A and B have been identified, and the role of Aurora C is still unclear. Aurora kinases are strongly associated with cancer and are frequently detected to be over-expressed in a wide range of human tumours [7]. The expression and activity of Aurora kinases are tightly associated with the cell cycle. The two major Aurora kinases, Aurora A and B, have distinct roles in mitosis. Along with its cellular binding partner TPX2 (target protein of Xenopus kinesin-like protein 2), Aurora A plays an essential role in mitotic spindle formation, centrosome maturation, and segregation; meanwhile, Aurora B is involved in chromosomal condensation, alignment and separation as well as kinetochore-microtubule attachment and cytokinesis with its INCENP binding partners [8–12]. Aurora C is required for spermatogenesis, and it may complement the functions of Aurora B; however, its function in cell cycle progression is still unknown [13]. Aurora A kinase plays an important role in mitosis. It is associated with centrosome maturation and separation and thereby regulates spindle assembly and stability [7]. Over-expression or deregulation of Aurora A has been associated with a high occurrence of cancer [14]. Since its discovery, Aurora A kinase has been identified as a colon cancer-associated kinase that is over-expressed in a wide range of human tumours such as breast, colorectal, ovarian and glioma. [15–17] Thus, targeted inhibition of Aurora A kinase has become an attractive therapeutic strategy in cancer treatment, and more than 10 Aurora inhibitors have entered into early clinical assessment [18]. Currently, cancer is the most dominant cause of death worldwide [19,20]. Apoptosis is a cellular process that organisms use for the digestion of excessive cells to control cell numbers. In recent years, a number of small-molecule inhibitors of Aurora kinase, with wide structural diversity, have been reported. Among these compounds, clinical trials have shown that ZM447439, VX-680, PHA-739358 and other potential inhibitors suppress the growth of tumour cells both in-vitro and in-vivo [21]. Quantitative structure–activity relationship (QSAR) models have been demonstrated to be an effective computational tool in understanding the correlation between the structures of molecules and their activities [22,23]. In a QSAR study, one seeks to find a mathematical relationship between the activity and one or more descriptors. Using this approach, scientists could predict the activities of a series of newly designed drugs before making the final decision on whether or not to synthesize and assay them. The prediction is based on the structural descriptors of the molecular features that account for variations in the biological activity to the greatest degree [22]. To investigate the key structural features required to design potential Aurora A inhibitor drug candidates from pyrazole-4-carboxamide derivatives [24], in the current approach, an attempt has been made to understand the structural and physicochemical requirements of a set of Aurora A kinase inhibitors through the help of a regression QSAR method.

2 Experimental

2.1 Data set

A dataset comprised of 25 pyrazole-4-carboxamide derivatives was used in the present study. All compounds and associated data were taken from the literature [24] for the development of the QSAR models. The Aurora A kinase inhibitory activity data were reported as IC₅₀ values in the same biological assay conditions. The dataset consists of some highly active and inactive compounds along with some compounds with moderate activities. The IC₅₀ (μM) values reported in the literature were converted to negative logarithmic values to get pIC₅₀ values, which were used for the QSAR study. The structures of these inhibitors and their pIC₅₀ values are given in Table 1.

Table 1 Structures and activity of pyrazole-4-carboxamide derivatives as anticancer agents..

S. No.	R1	R2	HCT116 (IC50)	pIC50	MCF-7(IC50)	pIC50	QSAR set
1	H	H	2.31	5.6363	2.34	5.6307	Training
2	H	Cl	3.52	5.4534	3.86	5.4134	Training
3	H	Br	4.47	5.3496	5.75	5.2403	Test
4	H	OMe	1.35	5.8696	1.71	5.7670	Training
5	H	OEt	0.62	6.2076	0.88	6.0555	Test
6	Me	H	2.36	5.6270	.45	6.3467	Training
7	Me	Cl	3.88	5.4111	4.01	5.3968	Training
8	Me	Br	5.35	5.2716	6.08	5.2160	Training
9	Me	OMe	1.45	5.8386	1.62	5.7904	Training
10	Me	OEt	0.78	6.1079	1.03	5.9871	Test
11	OMe	H	2.54	5.5951	2.53	5.5968	Training
12	OMe	Cl	4.06	5.3914	4.35	5.3615	Training
13	OMe	Br	5.87	5.2313	8.21	5.0856	Training
14	OMe	OMe	1.52	5.8181	1.77	5.7520	Training
15	OMe	OEt	0.97	6.0132	1.10	5.9586	Test
16	Cl	H	2.18	5.6615	2.34	5.6307	Training
17	Cl	Cl	2.99	5.5243	3.75	5.4259	Training
18	Cl	Br	4.36	5.3605	5.31	5.2749	Training
19	Cl	OMe	1.33	5.8761	1.55	5.8096	Training
20	Cl	OEt	0.42	6.3767	0.56	6.2518	Training
21	NO2	H	2.01	5.6968	2.26	5.6458	Test
22	NO2	Cl	2.92	5.5346	3.24	5.4894	Training
23	NO2	Br	4.08	5.3893	5.30	5.2757	Training
24	NO2	OMe	1.02	5.9914	1.21	5.9172	Training
25	NO2	OEt	0.39	6.4089	0.46	6.3372	Training

The sphere exclusion method [25] was divided into a training set and a prediction set. The training set was comprised of 20 compounds. The test set consists of five compounds, which are randomly selected and structurally diverse molecules possessing a wide range of activities. The IC₅₀ values employed in this work were measured under the same experimental conditions, a fundamental requirement for QSAR studies. We have used some 2D-descriptors to demonstrate the relationship between the structural properties and activity.

2.2 Generation of structures

The three-dimensional structures of these compounds were built by using the Sketch Molecule module in the VLife science molecular modelling package [26]. Structural energy minimization was performed using the standard Merck molecular force field (MMFF), and the maximum number of iterations for the minimization was set to 1000. The minimization was terminated when a distance-dependent dielectric constant of 1.0 and the convergence criterion of 0.01 kcal/mol Å were reached [27].

2.3 2D-QSAR studies

A large number of theoretical 2D individual descriptors such as the molar refractivity (MR), molecular weight (Mol wt.), partition coefficient (LogP), volume, XlogP, and smr; physiochemical descriptors such as the estate numbers, element count, dipole moment, XlogP, and SlogP; and the topological type have been computed. The E-state descriptor is calculated for each atom (e.g., OH, N, and Cl) and also separately for each H-atom. Physicochemical descriptors including 239 physicochemical parameters, 700 alignment type parameters and 99 atoms type count descriptors were calculated. The QSAR worksheet was generated using biological activity as the dependent variable and various 2D descriptors as independent variables. The data set was divided into a training set and a test set. In this study, more than 300 calculated descriptors (2D) were subjected to partial least squares coupled with a stepwise forward–backward variable selection method in order to establish a correlation between the physicochemical parameters and activity. The QSAR model is considered to be predictive if the following conditions are satisfied: r² > 0.6, q² > 0.6 and pred_r² > 0.5 [25].

2.4 Cross validation

To calculate q², each molecule in the training set was sequentially removed, the model was refit using the same descriptors, and the biological activity of the removed molecule was predicted using the refit model [28]. The q² was calculated using Eq. (1)(1) $q^2=1-\frac{{\sum (y_i-{\hat{y}}_i)}^2}{\sum {(y_i-y_{\text{mean}})}^2}$(1) .(1) $q^2=1-\frac{{\sum (y_i-{\hat{y}}_i)}^2}{\sum {(y_i-y_{\text{mean}})}^2}$(1) where y_i and ${\hat{y}}_i$ are the actual and predicted activities of the ith molecule in the training set, respectively, and y_mean is the average activity of all of the molecules in the training set. For external validation, the activity of each molecule in the test set was predicted using the model generated from the training set. The pred_r² value is calculated as follows (Eq. (2)(2) $\text{Pred}\_r^2=1-\frac{{\sum (y_i-{\hat{y}}_i)}^2}{\sum {(y_i-y_{\text{mean}})}^2}$(2) ):(2) $\text{Pred}\_r^2=1-\frac{{\sum (y_i-{\hat{y}}_i)}^2}{\sum {(y_i-y_{\text{mean}})}^2}$(2) where y_i and ${\hat{y}}_i$ are the actual and predicted activities of the ith molecule in the test set, respectively, and y_mean is the average activity of all of the molecules in the training set. Thus, the pred_r² value is indicative of the predictive power of the current model based on the external test set.

3 Results and discussion

A series of triphenyl-1H-pyrazole-4-carboxamide derivatives as anticancer agents was selected for QSAR studies. QSAR investigations of the triphenyl-1H-pyrazole-4-carboxamide derivative series resulted in several QSAR equations, considering the term selection criterion as r², q² and pred_r². In the present study, training and test sets were generated using the sphere exclusion method followed by partial least squares regression coupled with a stepwise forward–backward variable selection method.pIC₅₀ (HCT116-cell-lines) = 0.6457 (±0.2833); SsCH₃E-index, +0.1137 (±0.0150); SsBrE-index, +0.0345 (±0.0016); T_2_Cl_1, −0.1743 (±0.0257); 2PathCount

N = 25, r² = 0.8059, q² = 0.7621, F test = 37.8701, r²_se = 0.3197, q²_se = 0.3738, pred_r² = 0.7861

2D-QSAR model 1 shows a coefficient of determination (r²) of 0.8059 and a cross-validated correlation coefficient (q²) of 0.7621. Table 2 shows the correlation matrix for the descriptors used in generating Eq. (1)(1) $q^2=1-\frac{{\sum (y_i-{\hat{y}}_i)}^2}{\sum {(y_i-y_{\text{mean}})}^2}$(1) . Fig. 1a shows the contribution chart representing the contributions of descriptors in the 2D-QSAR model developed by the PLS method. The descriptors contribute to the 2PathCount parameter, which can signify the total number of fragments with an order of two (two bond path). It may be inferred that increasing the branching of the compound is favourable for anticancer activity. The presence of the descriptor SsBrE-index [signifies the electrotopological indices and the number of bromine atoms connected with one single bond] showed that an increase in the value of this descriptor is beneficial for the activities of the pyrazole-4-carboxamide derivatives. Compounds 3, 8, 13, 18, and 23 bearing a relatively bromo-substituent at the R2 position were the potential derivatives. The next important descriptor is the SsCH₃E-index indices for the number of eCH₃ groups connected with one single bond, and it showed an increase in the activities of the pyrazole-4-carboxamide derivatives with the presence of a methyl group at the R1 position, such as in compounds 6, 7, 8, 9 and 10. The present QSAR model reveals that Baumann’s alignment independent topological descriptor T_2_Cl_1 has a major contribution, which means that the count of the number of double bonds separated from a chlorine atom by single bond distances in a molecule indicates that the activity will increase with the presence of a relative chloro-substituent at the R1 and R2 positions (compounds 2, 7, 12, 16, 17, 18, 19, 20 and 22). Further optimization of the phenyl ring of inhibitor compounds using common aromatic constants showed electron-withdrawing substituents at the ortho position to be essential for Aurora A kinase inhibition. A preliminary analysis with pyrazole-4-carboxamide derivatives identified certain structural features, such as Cl, OCH₃, and CH₃ groups around the pyrazole ring, as the most favourable substitution for Aurora A kinase inhibition. The plots of observed vs. predicted values of pIC₅₀ are shown in Fig. 1b.pIC₅₀ (HCT116-cell-lines) = 0.1200 (±0.0159); SssCH₂count, +0.3605 (±0.0174); SsFcount, +0.0449 (±0.0077); hydrogen count −0.3918 (±0.0862) SlogP

Table 2 Correlation matrix between descriptors present in the best QSAR model-1(HCT116-cell-lines).

Parameter	SsCH3E-index	SsBrE-index	T_2_Cl_1	2PathCount
SsCH3E-index	1.0000
SsBrE-index	0.6328	1.0000
T_2_Cl_1	0.4328	0.7123	1.0000
2PathCount	0.5241	0.6874	0.8436	1.0000

Fig. 1 (a) Contribution charts of the descriptors for the 2D model-1. (b) Plot of observed versus predicted activity by best 2D QSAR model 1. (c) Contribution charts of the descriptors for the 2D model-2. (d) Plot of observed versus predicted activity by best 2D QSAR model 2. (e) Contribution charts of the descriptors for the 2D model-3. (f) Plot of observed versus predicted activity by best 2D QSAR model 3. (g) Contribution charts of the descriptors for the 2D model-4. (h) Plot of observed versus predicted activity by best 2D QSAR model-4. (i) The designed scaffold new chemical entity.

N = 25, r² = 0.7759, q² = 0.6784, F test = 13.1278, r²_se = 0.5314, q²_se = 0.5874, pred_r² = 0.7061

Model 2 produced from the training set obtained from the sphere exclusion method shows r², q², and pred_r² values much lower than model 1. The descriptors SlogP signify the log of the octanol/water partition coefficient compounds. In the model, the lipophilicity suggests that a bulky substituent would be favourable at the R1 position. The model reveals that the descriptor SssCH₂count suggests that the activity of pyrazole-4-carboxamide derivatives may be increased by increasing the number of –CH₂ groups present in the molecules. The model indicated the SsFcount of the total number of fluorine atoms connected with a single bond, suggesting that the activity of pyrazole-4-carboxamide derivatives may be increased by increasing the number of fluorine atoms present in the moiety at the R2 position. The hydrogen count descriptor suggests that an increased number of hydrogen atoms will increase the activity. The contribution chart of selected descriptors is represented in Fig. 1c, and plots of the observed vs. predicted values of pIC₅₀ are shown in Fig. 1d. The observed and predicted activity values of the training and test sets of compounds selected for the model are given in Table 4.pIC₅₀ (MCF-7-cell-lines) = 0.2323 (±0.0273); chlorine count, +0.1138 (±0.0173); T_N_F_3, −0.1118 (±0.0097), SddsN (nitro) count

Table 4 Comparative observed and predicted activities of carboxamide derivatives.

Com	QSAR model-1 (HCT116)		QSAR model-2 (HCT116)		QSAR model-3 (MCF-7)		QSAR model-4 (MCF-7)
	Pred.	Res.	Pred.	Pred.	Res.	Res.	Pred.	Res.
1	5.6906	−0.0543	5.3562	0.2801	5.6492	−0.0185	5.6188	0.0119
2	5.3687	0.0847	5.3947	0.0587	5.4384	−0.025	5.4017	0.0117
3	5.3758	−0.0262	5.3566	−0.007	5.3199	−0.0796	5.2165	0.0238
4	5.8861	−0.0165	5.8674	0.0022	5.7931	−0.0261	5.6863	0.0807
5	6.4197	−0.2121	6.4039	−0.1963	6.3609	−0.3054	6.2685	−0.213
6	5.6165	0.0105	5.6369	−0.0099	6.2984	0.0483	6.2017	0.145
7	5.4331	−0.022	5.4171	−0.006	5.2947	0.1021	5.1945	0.2023
8	5.4327	−0.1611	5.4160	−0.1444	5.3360	−0.12	5.2249	−0.0089
9	5.8224	0.0162	5.8296	0.009	5.6426	0.147	5.7360	0.0544
10	6.2941	−0.1862	6.0195	0.0884	5.9553	0.031	5.8615	0.1256
11	5.5570	0.0381	5.5722	0.0229	5.5565	0.0403	5.5895	0.0073
12	5.3942	−0.0028	5.3905	0.0009	5.3845	−0.023	5.3074	0.0541
13	5.3875	−0.1562	5.3664	−0.1351	5.1662	−0.0806	5.1983	−0.1127
14	5.8554	−0.0373	5.8437	−0.0256	5.7674	−0.0154	5.7036	0.0484
15	6.1224	−0.1092	6.2096	−0.1964	5.9213	0.0373	5.8990	0.0596
16	5.6542	0.0073	5.6405	0.021	5.6940	−0.0633	5.6135	0.0172
17	5.4550	0.0693	5.3380	0.1863	5.3523	0.0736	5.3320	0.0939
18	5.3920	−0.0315	5.3802	−0.0197	5.3138	−0.0389	5.2086	0.0663
19	5.7120	0.1641	5.7911	0.085	5.8272	−0.0176	5.7818	0.0278
20	6.4197	−0.043	6.3639	0.0128	6.2147	0.0371	6.1645	0.0873
21	5.6365	0.0603	5.6726	0.0242	5.6403	0.0055	5.6215	0.0243
22	5.5672	−0.0326	5.5483	−0.0137	5.2975	0.1919	5.3882	0.1012
23	5.4331	−0.0438	5.4171	−0.0278	5.3317	−0.056	5.2194	0.0563
24	5.8327	0.1587	5.8160	0.1754	5.9692	−0.052	5.8873	0.0299
25	6.3188	0.0901	6.3988	0.0101	6.3302	0.007	6.2197	0.1175

N = 25, r² = 0.7601, q² = 0.7188, F test = 39.7513, r²_se = 0.4921, q²_se = 0.4608, pred_r² = 0.6821

Model 3 shows a good internal predictive power (q² = 0.7188) of 72% and a predictivity for the external test set (pred_r² = 0.6821) of approximately 68%. The descriptor chlorine count, which is the number of chlorine atoms connected with a single bond, showed that an increase in the values of this descriptor would be beneficial for the activity of the MCF-7 cell lines. SddsN (nitro) count suggests that the presence of a nitro group at the R1 position revealed that a bulky substituent at this site would increase the activity. The alignment independent descriptor T_N_F_3 (nitrogen atoms separated from any other fluorine atom by three bonds) showed that an increase in the value of this descriptor is beneficial for the activity. Compounds with an electron-withdrawing group at R1 and R2 exhibited significantly improved potencies compared to derivatives bearing an electron-donating methyl substituent at R1. Table 3 shows the correlation matrix for the descriptors used in generating model-3. The contribution chart of selected descriptors is represented in Fig. 1e, and plots of the observed vs. predicted values of pIC₅₀ are shown in Fig. 1f.pIC₅₀ (MCF-7-cell-lines) = 0.1586 (±0.0373); SdsCHE-index, +0.0052 (±0.0017); T_T_Cl_1, +0.1586 (±0.0611); SssOE-index

Table 3 Correlation matrix between descriptors present in the QSAR model-3(MCF-7-cell-lines).

Parameter	Chlorines count	T_N_F_3	SddsN (nitro) count
Chlorines count	1.0000
T_N_F_3	0.5931	1.0000
SddsN (nitro) count	0.6854	0.7593	1.0000

N = 25, r² = 0.7669, q² = 0.6933, F test = 10.0479, r²_se = 0.4658, q²_se = 0.5061, pred_r² = 0.6217

The significant model 4 using the PLS method has 0.7669 as the coefficient of determination (r²), an internal predictive power (q² = 0.6933) of 69% and a predictivity for the external test set of approximately 62%. From the above equation, the alignment independent descriptor T_T_Cl_1 contributes positively (approximately 14%) for the activity, which corresponds to the number of double bonds separated from a chlorine atom by a single bond distance. A pyrazole-4-carboxamide moiety at the R1 or R2 position suggests that a less bulky group at this position would benefit the potency of the compound. The SdsCHE-index descriptor indicates the presence of a number of –CH groups connected with one double and one single bond, showing an increase in the R1 position of the pyrazole-4-carboxamide nucleus and suggesting the presence of alkyl group at the R1 position. The descriptor SssOE-index indicated that the activity will increase with the presence of methoxy groups at the R1 position such as in compounds 11,12,13,14 and 15. The contribution chart of selected descriptors is represented in Fig. 1g, and plots of the observed vs. predicted values of pIC₅₀ are shown in Fig. 1h.

4 Conclusion

In this paper, we propose a new method for the generation of 2D-QSAR models for the prediction of activity values of chemicals. In the above discussion, 2D-QSAR models identified the important structural requirements for triphenyl-1H-pyrazole-4-carboxamide derivatives exhibiting activity against Aurora A kinase. The detailed analysis of these four models enabled us to better understand the structure activity relationship of pyrazole-4-carboxamide derivatives. The QSAR models indicate that the descriptors of the SssCH₂count, SsFcount, chlorines count, SddsN (nitro) count, T_T_Cl_1, T_N_F_3 and SssOE-index influenced the activity. The models indicated that a chlorine or methoxy substituent at the R1 site would benefit the activity of the compound. All of the derivatives involved in the present study possessed a halogen substituent (e.g., chloro-, bromo) at R1 (Fig. 1i). The results of the present study may be useful in the design of more potent Aurora A kinase inhibitors that may be used as potential anticancer agents.

Acknowledgement

The author wishes to express gratitude to V-life Science Technologies, Pvt. Ltd. for providing the software for the study.

Notes

Peer review under responsibility of Taibah University.