Computational Approach for the Design of Novel Tankyrase Inhibitors:

A Rational Study based on Pharmacophore and Atom based 3D QSAR

Vasudev Pai¹ , Muddukrishna B.S.², Aravinda Pai^3*.

¹Department of Pharmacognosy, Manipal College of Pharmaceutical Sciences (MCOPS), Manipal University, Manipal, Karnataka, India.

²Department of Pharmaceutical Quality Assurance,Manipal College of Pharmaceutical Sciences (MCOPS), Manipal University, Manipal, Karnataka, India.

^3*Department of Pharmaceutical Chemistry, Manipal College of Pharmaceutical Sciences (MCOPS), Manipal University, Manipal, Karnataka, India.

*Corresponding Author E-mail: aravind.pai@manipal.edu

ABSTRACT:

Pharmacophore based screening was undertaken for a set of 30 flavones exhibiting Tankyrase I inhibitory activity. The two isoenzymes Tankyrase I and Tankyrase II share 80% sequence homology. This gives rise to the fine tuning required in the designing of small molecules wchich can selectively inhibit one or the other. In a recent study it was reported that, small molecule inhibitors of Tankyrases will be benificial in certain types of cancers like colorectal cancers. In future it may be a slective target for cancers like colorectal cancer.

Four feature pharmacophores with one H bond acceptor and three aromatic rings were developped.The hypothesis ARRR2 was considered as best hypothesis ,which gave a statistically significant three dimensional QSAR model with corresponding statistical parameters (0.9028 as r²value and 0.8548 as q² value). The generated model was applied sucessfully on a set of both training and test set molecules. The squared corelation coefficient of 0.85 was observed between actual and predicted values in test set. The squared corelation coefficient of 0.96 was observed between actual and predicted value in training set. The built model will be useful in predicting the structural requirements needed for the selective inhibition of Tankyrase I

KEYWORDS: Tankyrase, Pharmacophore, QSAR, flavones.

INTRODUCTION:

Colorectal cancer is one of the major area of study in recent years owing to its higher mortality rate. Recent study proves that colorectal cancer is the second major cause of death[1]. Tankyrases are recently idenfied cancer targets getting expressed in colon cancer cell lines.Tankyrases are members of the poly ADP-ribosepolymerase (PARP) family of proteins attracted attention owing to their role in axin downregulation and stabilization of β catenin[2]. Human Tankyrases are highly conserved with a sequence homology of 89%. Between two isoforms namely, Tankyrase I and Tankyrase II[3].

The role of Tankyrases in colorectal cancer is evidenced with the overactvation of wnt β catenin signalling pathway[4].

Extensive literature search for the small molecule inhibitors of Tankyrase I arouse our interest in flavone pharmacophore due its diverse biological activity and simple scaffold. Literature also provided evidences for the Tankyrase inhibition by flavones. Flavones are naturally occuring secondary plant metabolites categorised under the broad class of ﬂavonoids, which pocess various degress of free radical scavenging properties present in a wide variety of edible plants and vegetables. Flavonoids have also been shown to pocess antitumor effect in various cancer cell lines.Inhibition of TNKS1 with ﬂavone and its antiproliferative properties were already reported. The present study is based on the work reported in the literature. The reported activity was used for generating the best fit QSAR equation based on pharmacophore approach. The generated equation will be used for designing selective inhibitors for Tankyrase.

MATERIALS AND METHODS:

Selection of data set:

In the present study, the data set of 30 compounds were taken from the literature(5) with their invitro activity on Tankyrase 1. A training set of 21 molecules (70% of total molecules) were used to generate QSAR equation. The training set molecules were selected based on their structural diversity, activity range of 3 log order difference and activities covering the entire range were used.

To test the predictive power of the model, a set of 9 molecules were choosen for the test set.The test set was selected so that , they are structurally mimicking the training set. Model was generated using training set and validated using test set.

The activity values (IC₅₀) from literature were converted to PIC₅₀ ( logarthmic scale) using options available in the calculator.

Pharmacophore hypothesis generation:

The pharmacophore models were generated using phase 3.0 module available with Schrodinger molecular modelling software. In QSAR studies, the minimum energy structural feature is required to achieve accurate 3D descriptor values. The structures were drawn using 2D skecth option available with Maestro. The 2D structures were later converted to their corresponding 3D structures. Geometry optimization was performed using OPLS force field. The whole set of molecules were divided into training and test sets.

PHASE uses a structure cleaning utility called ligprep, which adds hydrogen, generates stereoisomers and various conformers and predicts proper ionization states at a perticular P_H.PHASE provides two in built approaches, employing the MacroModel conformational search engine. Due to flexibility of ligands , all the possible conformers were observed for the each set of ligands and conformers close to the crystal structures can be sorted out. Conformational analysis was carried out using Monte Carlo Multiple Minimum method. The ligands were grouped into active and inactive sets by assigning a suitable activity threshhold value.

Creation of Pharmacophore sites:

The local chemical environment of selected ligands were defined by four point pharmacophoric features: one Hydrogen bond acceptor (A) and three aromatic rings (R).The pharmacophore was developped based on the intersite distance and angles.

Scoring pharmacophores on the basis of their acativity threshhold:

The resulting pharmacophore sets were scored and ranked. The scoring will identify the best pharmacophore hypothesis. The scoring algorithm considered the effecive contributions from the alignment of site points and magnitude of vectors, selectivity and activity with overall conformational energies.

Identifying common pharmacophores:

After the selection of best hypothesis, ARRR (Figure 1A) it was further analysed (Table 1). The pharmacophore ARRR (Figure 1B) included the following special features: One hydrogen bond acceptor (A) (pink sphere with two arrows), three aromatic rings (R) (grey circle). The 2D representation of the pharmacophore ARRR shows one hydrogen bond acceptor (A), one aromatic ring , fused gama benzopyrone ( two aromatic rings ) respectively R6 , R4 and R5 are as the key pharmacophoric elements present in the selected pharmacophore. Table 3 and 4 represents site scores of angles and distances.

Table 2: Site score distances

Entry	Site1	Site2	Distance
ARRR.2	A2	R4	3.713
ARRR.2	A2	R5	2.69
ARRR.2	A2	R6	6.159
ARRR.2	R4	R5	2.448
ARRR.2	R4	R6	6.444
ARRR.2	R5	R6	4.282

Table 3: Site score angles

Entry	Site1	Site2	Site3	Angle
ARRR.2	R4	A2	R5	41.2
ARRR.2	R4	A2	R6	77.1
ARRR.2	R5	A2	R6	35.9
ARRR.2	A2	R4	R5	46.4
ARRR.2	A2	R4	R6	68.7
ARRR.2	R5	R4	R6	22.3
ARRR.2	A2	R5	R4	92.4
ARRR.2	A2	R5	R6	122.5
ARRR.2	R4	R5	R6	145.1
ARRR.2	A2	R6	R4	34.2
ARRR.2	A2	R6	R5	21.6
ARRR.2	R4	R6	R5	12.5

Table 4 : activity prediction graphs of QSAR studies

Ligand Name	QSAR Set	Activity reported	Predicted Activity	Pharm Set	Fitness	PLS factors
A2	training	1	1.04	inactive	2.57	4
A1	training	2.519	2.11	active	2.93	4
A3	test	1	1.13	inactive	2.57	4
A4	training	1	1.08	inactive	2.71	4
A5	training	1	0.98	inactive	2.81	4
A6	training	0.699	0.62	inactive	2.87	4
A7	training	1	1.06	inactive	2.82	4
A8	test	1	1.06	inactive	2.6	4
A9	training	2.322	2.30	active	2.75	4
A10	training	2.775	2.26	active	2.91	4
A11	training	1	1.16	inactive	2.71	4
A12	training	1	1.60	inactive	2.91	4
A13	training	1	1.05	inactive	2.82	4
A14	training	2.556	2.13	active	2.74	4
A15	training	2.845	2.48	active	2.8	4
A16	test	2.367	2.25	active	2.97	4
A17	training	2.496	2.23	active	2.97	4
A18	test	2.897	2.22	active	2.97	4
A19	training	2.929	3.10	active	2.96	4
A20	training	1.82	1.82	active	2.8	4
A21	training	1.672	1.85	Active	2.97	4
A22	training	0.778	0.78	inactive	2.96	4
A23	test	1.826	1.88	active	2.8	4
A24	training	1.851	1.89	active	2.97	4
A25	training	2.435	2.47	active	2.91	4
A26	training	2.21	2.26	active	2.94	4
A27	test	2.164	2.18	active	2.77	4
A29	training	0.845	0.80	inactive	2.89	4
A28	training	2.161	2.12	active	3	4
A30	test	2.158	2.15	active	2.75	4

Building QSAR models:

QSAR model was developped based on selected common hypothesis by dividing the literature data in to training set (70%) and test set (30%) based on activity, structural similarity and functional group variation. PHASE provides dual options for the geometric alignment of 3dimensional structures. In the present study, an atom- based QSAR model was used , which is an effective tool for the study of structure–activity relationships. In atom-based QSAR, a set of overlapping van der Waal’s spheres were considered for each molecule.

Each atom is placed into one of the six categories like, hyderogen bond donor (D), hydrogen bond acceptor (H), Hydrophobic nonpolar, Positive ionic, negative ionic and miscellaneous. To develop best fit QSAR model, the aligned training set molecules were placed in regular Grid of cubes each cube is labelled in a binary bit format to account for the different atom types and their features. The resulted data can be used to generate partial least square regression model (PLS). Atom-based QSAR models were generated for the selected hypothesis using the 21-member training set using four PLS factors. The remaining molecules were taken under test set. The PLS factors selected was 20% of the total poulation of the training set.

RESULTS AND DISCUSSIONS:

The present study aimed at elucidating the 3 dimensional structural features of substitued flavones for the selective inhibition of Tankyrase I. The pharmacophore modeling and QSAR studies were carried out using Phase module of Schrodinger suite. This Phase generated hypothesis helps in predicting the relative binding of the test molecules at the active site of the receptor. Hence, pharmacophore based alignment was used for generating 3D-QSAR model to identify key structural features required for the inhibition of TankyraseI.

For the generation of pharmacophore model, we have considered 30 compounds having activity against Tankyarse I as they contain important structural features crucial for binding to the receptor binding site. We used four minimum sites and five maximum sites to have optimum combination of sites or features common to the most active compounds. One hundred and two common pharmacophore models were generated. The hypothesis ARRR2 was selected based on Survival active and inactive scores ( as shown in the table 1). The pharmacophore hypothesis was used for building 3D QSAR models. The 3D QSAR model was evaluated based on its predictive ability in both training and test sets. For a given set of PLS factors the equation was assesed based on the regression coefficient value and crossed validation coefficient value. The QSAR model was based on the special features of the atoms atached to the core ring system. The features studied are Hydrogen bond donor, hydrogen bond acceptors, hydrophobic nonpolar interactions, electron withdrawing grou interactions , positive ionic interactions and negative ionic interactions.

CONCLUSION:

A ligand-based pharmacophore model was built for a series of novel flavones inhibiting TankyraseI. The pharmacophore and atom based 3D QSAR equation was generated and validated using a set of 30 flavones reported in the literature. The hypothesis ARRR2 was considered as best hypothesis ,which gave a statistically significant three dimensional QSAR model with corresponding statistical parameters ( 0.9028 as r²value and 0.8548 as q² value). The generated model was applied sucessfully on a set of both training and test set molecules. The squared corelation coefficient of 0.85 was observed between actual and predicted values in test set. The squared corelation coefficient of 0.96 was observed between actual and predicted value in training set. The QSAR model gave us the substitution pattern required for the effective binding of flavones to Tankyrase I. The hydrophobic nonpolar and electron withdrawing groups are essential for the activity.The built model will be useful in predicting the structural requirements needed for the selective inhibition of Tankyrase I

Table 5: QSAR statistics

Factors	SD	R^2	R^2 CV	R^2 Scramble	Stability	F	P	RMSE	Q^2	Pearson-r
1	0.3466	0.7778	0.5905	0.4538	0.866	66.5	1.26E-07	0.34	0.7897	0.9323
2	0.2355	0.9028	0.6508	0.6792	0.833	83.6	7.73E-10	0.28	0.8548	0.9445
3	0.1691	0.9526	0.703	0.7755	0.767	114	1.85E-11	0.33	0.8028	0.9261
4	0.1542	0.9629	0.7231	0.8306	0.791	103.9	3.10E-11	0.33	0.7976	0.9232

Figure 1: A

Figure 1:B

Figure 1:C

Figure 1:D

Figure 1: A. common hypothesis B. common hypothesis on active molecule C. alignment of active molecules D. alignment of inactive molecules.

Figure 2: A

Figure 2: B

Figure 2: C

Figure 2: D

Figure 2: E

Figure 2: QSAR models for different types of interactions. A. positive ionic interactions B. negative ionic interactions C. hydrophobic non polar interactions D. hydrogen bonding interactions E. electron withdrawing group interactions.

Graph 1: activity prediction training set Graph 2: activity prediction test set