Identification of CDPK1 inhibitors using molecular modelling and QSAR analysis

a
a
b
a
Pinaki Saha , Jayaraman Valadi , Michael Krein , N. Sukumar
Shiv Nadar University, NH91, Tehsil Dadri , Gautam Buddha Nagar, Uttar Pradesh : 201304
b
Rensellaer Polytechnic Institute, 110 8th St, Troy, NY 12180, United States of America

Abstract: A combination of molecular modelling/QSAR analysis is proposed for Identification of CDPK1 inhibitors. A list of 72 inhibitors of CDPK1 belonging to T. gondii and C. parvum were compiled from literature and were

subsequenty docked against their corresponding X-ray crystal structures using Autodock Vina. In case of E. tenella no X ray crystal structure existed so to get the crystal structure, the X-ray cystal structure of 4IEB (T. gondii
mutant) was modified (the gatekeeper residue was changed from methionine to threonine). The aforementioned 72 compounds were docked into modified 4IEB crystal structure as no inhibitors were found in literature for E.
tenella CDPK1. The dockingIntroduction
scores and IC50 values of the inhibitors were utilized in creation of a QSAR model
using
YAMS online tool provided by Rensselaer Exploratory Center for Cheminformatics
Method
(contd)
Result (contd) Research. The descriptors
used for QSAR were calculated from parameter client provided by Edragon along with RECON descriptors provided by Rensselaer Exploratory Center for Cheminformatics Research.
In case of C. parvum the YAMS online tool generated an QSAR
QSAR studies are used for correlating molecular structures with
3.) The QSAR model for the T. gondii docking score was used to
screening model using SVM model.
biological activity and chemical activity. Molecular docking is
predict the docking score for the TDT test set. In this case also
used to predict the preferred orientation of a ligand to a receptor
the compounds were docked and their actual docking score was
when they are bound to form a stable complex. In our studies we
correlated with the docking score predicted by the QSAR model.
use both QSAR and molecular modelling to develop virtual
The Pearson correlation coefficient here is found to be 0.67.
screening models for the parasitology drug targets: the calciumdependent protein kinase 1 (CDPK1) of Eimeria tenella,
4.) In case of C. parvum the QSAR model was based on pIC50
Toxoplasma gondii and Cryptosporidium parvum.
values; an internal test set of random 16 compounds was
selected from the literature dataset . The QSAR model was used
E. tenella, T. gondii and C. parvum, all are parasitic protozoans
to predict the pIC50 values of the test set. The Pearson
belonging to the phylum apicomplexa. E. tenella is a major
pathogen of chickens and other birds while T. gondii and C.
correlation coefficient calculated for the actual and the predicted
parvum
are protozoans responsible for the diseases
pIC50 is 0.89 in this case.
toxoplasmosis and cryptosporidiosis respectively.

Our study focuses on development of virtual screening models

that allow for the selection of a new set of molecules that could
possibly be potent inhibitors of CDPK1 for the aforementioned
parasites.

Dataset

72 molecules were taken from relevant literature (1-4)

on the parasitic protozoans.
3-d structure of each of the 72 molecules was created using
MarvinSketch. Theoretical molecular descriptors have been
calculated using parameter client program from the website
http://www.vcclab.org and Recon program from the website
http://reccr.chem.rpi.edu/

The QSAR screening model in this case is predicting the pIc50

values of the given dataset. The 72 molecules from data set
were split into training and test set (80:20 split). The R2 value
obtained for training set is 0.96 and for test set the value
obtained is 0.72 and a low R.M.S.E of 0.63; hence this model is
suitable for predicting pIC50 values for an unknown external test
data set.
In case of T. gondii the QSAR screening model was used to
determine the docking score and not the pIC50 vaue as there are
only 34 compounds available in case of T. gondii and the
number of data points are insufficient to create a QSAR model
for pIC50 Instead we create a QSAR model for predicting
docking score. The YAMS tool created this screening model
using PLS model.

Conclusion

1.) The QSAR model for the docking scores of E. tenella and T.
gondii had respectable Pearson correlation coefficients of 0.63
and 0.67 respectively.
2.) The QSAR model for the pIC50 of C. parvum had a good
Pearson correlation coefficient of 0.89.
3.) The top ranking compounds for the three cases selected from
held out test set of 22 compounds of TDT 2014 challenge 3 are
given below:
a.) C. parvum CDPK1 inhibitor

Method

The Yams QSAR modeling tool available at the website http

://reccr.chem.rpi.edu was used for the creation of QSAR models.
Molecular docking was performed using AutoDockTools-1.5.6
and AutoDock Vina 1.1.2.
The table below shows the correlation of docking score with the
IC50 of compounds reported in literature. The table below gives
the Pearson correlation coefficient between the docking score
and the IC50 in four different cases.
CDPK1 organism

Docking score
(crystal structure
with water)

Docking score
(crystal structure
without water)

C. parvum (72
compounds)

0.5226283

0.50936159

T. gondii (34
compounds)

0.45372674

0.558084

b..) T. Gondii CDPK1 inhibitor

The R2 value obtained for training set is 0.97. The Test data in
this case has R2 of 0.81 and R.M.S.E. of 0.36. hence this model
is suitable for predicting docking scores for an unknown external
test data set.
In case of E. tenella no literature was available for inhibitors of
E. tenella CDPK1. The gatekeeper of E. tenella CDPK1 is
threonine which is larger than the glycine gatekeeper residue
present in CDPK1 of T. gondii and C. parvum. In PDB crystal
structure 4IEB is CDPK1 kinase of T. gondii whose gatekeeper
residue glycine is mutated to methionine. 4IEB sequence shares
93.3% similarity with E. tenella sequence (found by local
alignment; EMBOSS Water). Using Pymols mutagenesis
functionality the methionine gatekeeper group of 4IEB was
changed to threonine. The protein stucture so obtained was
used for docking for E. tenella inhibitors. The 72 compounds
were docked into the modified 4IEB protein. YAMS was used to
generate QSAR screening model for prediction of docking score
using PLS model

c.) E. Tenella CDPK1 inhibitor

References

1.) Bioorg Med Chem Lett. 2012 August 15; 22(16): 52645267.doi:10.1016/j.bmcl.2012.06.050.
2.) J Med Chem. 2012 March 22; 55(6): 2803-2810. doi:
10.1021/jm201725v..
One of the 72 ligands docked in crystal structure of C. parvum
(3NCG) without water, Docking score: -9.5 Kcal/mol

R2 value obtained for training set is 0.78. The Test data in this
case has R2 of 0.75 and R.M.S.E. of 0.58, hence this model is
suitable for predicting docking scores for an unknown external
test data set.
Results

1.) The QSAR models generated for the above targets are
used to rank order a hold-out test set of 22 compounds of
TDT 2014 challenge 3.

Docked ligand overlapped with the actual inhibitor (BK1) of C.

parvum (3NCG) CDPK1

2.) The QSAR model for the E. tenella docking score was
used to predict the docking score for the TDT test set,
additionally the 22 compounds were also docked into the
protein crystal. The Pearson correlation coefficient between
the actual docking score and the predicted docking score is
found to be 0.63.

3.) J Med Chem. 2012 March 8; 55(5): 2416-2426. doi:

10.1021/jm201713h.
4.) J Clin Invest.
10.1172/JCI61822.

2012;

122(6):

2301-2305.

doi:

Acknowledgements

We like to thank Shiv Nadar University for provoding the

resources to carry out my research work.