Prediction of human serum albumin binding affinity for anti-cancer

drugs using norm indexes

Haicheng QIAN a, Kanwal, Qingzhu JIAa, Qiang WANG a*, Xiangying X U a, Huifen JI a, Jingchen

YIN a, Wenxuan, WANG a, Peisheng Ma b

a. School of Material Science and Chemical Engineering, Tianjin University of Science and

Technology, 13St. TEDA, Tianjin, 300457, People’s Republic of China

b. School of Chemical Engineering and Technology, Tianjin University, Tianjin 300072, People’s

Republic of China

* To whom correspondence should be addressed. E-mail: wang_q@tust.edu.cn

Abstract

The binding affinity constant (logKHSA) of compounds to Human Serum Albumin (HSA) is very useful in the pharmaceutical industry to speed up the design of new drugs. In this paper, a set of new norm indexes were proposed firstly, and a QSAR model based on these was developed to predict the logKHAS of 95 heterogeneous
anticancer drugs. Results showed that this norm indexes-based-QSAR model could
predict logKHSA well with r? (square correlation coefficient) and q? (Leave-one-out validation) values of 0.76 and 0.67, respectively. Also, results shown in this work demonstrated that these new norm indexes are very important descriptors which are pure chemical structural and could be further used for other properties
prediction of drugs.
Keywords: Human serum albumin;binding affinity constant; prediction;norm index;
QSAR

Introduction

As human health and life affected by the cancer seriously, the research and development of anticancer drugs have been becoming a major issue in recent decades. In additional, new anticancer drugs discovery and development processes are time-consuming, expensive and also have a high attrition rate,1,2 pharmaceutical
researchers therefore pay much attention to the early prediction of ADME (absorption, distribution, metabolism, excretion) properties of anticancer drugs.
Plasma-protein binding is an important ADME property of drugs which affects their transport and release. Usually, drugs primarily bind to three types of plasma proteins: human serum albumin (HSA), a1-acid glycoproteins and lipoproteins.3 In fact, HSA accounts for 60% of the total plasma proteins. Moreover, HSA binds a diverse array of drugs and basically influences their free concentration, solubility, transportation and metabolic clearance.4
Some QSAR studies have been performed to predict the logKHSA by using various descriptors.5-14 Based on docking descriptors, G. Colmenarejo et al. have established a QSAR model for the logKHSA prediction of 95 drugs and achieved a good result,13 and their result also agreed with X-ray structure of HAS alone or bound to ligands. Later,
using this same method and less descriptors, Chen et al. have established another QSAR model for the logKHSA prediction of the same 95 anticancer drugs and achieved better result with r2 of 0.79.14
Recently, a set of norm indexes have been proposed by our group, based on which some QSAR modes had been developed and successfully used for prediction toxicity values (log(LC50), 96 h LC50 data for Poecilia reticulata) for 190 diverse narcotic pollutants,15 the aryl hydrocarbon receptor binding affinity (pEC50) of dibenzofurans
and the mutagenic potency (lnR) of aromatic and heteroaromatic amines.16 Therefore,
it is logical to further evaluate the performance of these norm indexes for prediction of other properties of chemical compounds, especially for anticancer drugs.
Here, the objective of the present work is to establish a new QSAR model based on our norm indexes for prediction of logKHSA of 95 heterogeneous anticancer drugs. Data
In Colmenarejo et al.’s work, the retention time for 95 anticancer drugs and drug-like compounds using high-performance affinity chromatography with an immobilized HSA column were measured in detail. Then, the binding affinity constants to HSA (logKHSA) of these 95 anticancer drugs were calculated based on the
retention time. 14 In this work, logKHSA data of 95 anticancer drugs were selected from two references and used for QSAR modeling. 13, 14

Method proposed in this work

In this work, based on molecular chemical graphs of anticancer drugs, the distance matrixs Md for 95 anticancer drugs were established firstly, from which the jump distance matrixs were then denoted. In order to reflect the diversity of various 95 anticancer drugs molecular, the electronegativity, vander waals radius, atom mass and atom charge of molecular were considered and specially used to build the extended matrix Me for 95 anticancer drugs, Then, combining the extended matrix Me with the distance matrix Md, an extended distance matrix was obtained.
Here, matrixes considered in this work are as follows.

Md = (aij )

= ( )

distance matrix

1 if the path length between atomsi and j is 3

Mc aij

{0 otherwise

Me = [electronegativity vander waals radius atom mass atom charge]

MD1 = Md

T

MD2 = Md + Me [;,1 ]× Me [;,1]

T

MD3 = Md + Me [:, 2 ]× Me [; , 2]

MD4 = Md + Me [:, 3 ]× Me [; , 3] MD5 = Md + Me [:, 4 ]× Me [; , 4] MD6 = [Mc, Me]× [Mc, Me]

Based on the above, the QSAR model for logKHAS prediction of 95 anticancer
drugs is expressed as:
log K

HAS

= 1.7 - 2.9norm(MD , 2) + 0.89 ×10-3 norm(MD , 2) - 0.30 ×10-3

norm(MD , fro) - 0.60 ×10-4 norm(MD , 2) + 2.9norm(MD , 2) + 0.51×10-3

norm(MD , fro) + 0.64 ×10-4 norm(MD , 2) + 0.81×10-8 E

Where, norm(MD, 2) means the largest singular value of matrix MD, norm(MD, fro) is the frobenius-norm of matrix MD. Emin is the minimum energy of 95 anticancer drugs molecular.
In this work, the molecular structure building work and the entire modeling work were carried out by using free software of hyperchem.7.0 in this work. ab initio method was used for obtaining the molecule’s E??? and its’ atom charge. Here, the
molecule’s Geometries and charge distributions have been optimized by using ab initio methods at STO-3G level. And multiple regression analyses were used for model development.
In order to develop the model and to evaluate the predictive capacity of the model, the parameters were first optimized by leave-one-out (LOO) method. The LOO method was performed for the training set to select the optimum values of the parameters. The LOO method is carried out by removing one example from the training set to construct the decision function on the basis of the remaining training set and then training on the removed example. Once all examples of the training data are tested, the fraction of errors over the total number of training examples can then be evaluated.

Results and discussion

The predicted vs. experimental values of logKHAS are scatter plotted and shown in fig.1. Results shown in Fig.1 indicate that the predicted logKHAS agree well with the experimental results, which described that these new norm indexes for predicting the
logKHAS has good overall accuracy.

Fig.1 Scatter plot showing the correlation between the predicted by our model vs. experimental values of logKHAS
Generally, to evaluate the effect of a prediction QSAR model, usually, follow
aspects should be considered carefully.17-19

(1) n/p Ratio: n/p =4 or n =4p.

(2) Square of correlation coefficient (r ? ): It has already been suggested that the only
QSAR model having r ? > 0.6 should be considered for validation.
(3) Cross-Validation Test (q2): according to the literatures, a QSAR model must have q ? > 0.5 for their predictive ability.

(4) Standard Deviation (S): The smaller S value is always required for the predictive

QSAR model.
For our model, n/p ratio is 11.7, r ? is 0.76, q ? and S are 0.67 and 0.22, respectively. Accordingly, it further demonstrated that this norm indexes-based QSAR model has good ability of prediction for logKHAS of various anticancer drugs. Conclusions
A new QSAR model has been developed for human serum albumin(logKHAS ) prediction of 95 compounds based on the norm indexes proposed in this work. Results indicated that logKHAS can be predicted successfully with this new model by using
these norm indexes. Also, it is evident that these proposed norm indexes can be used
to predict log (logKHAS) with a significant degree of confidence.

Acknowledgements

Research reported in this work was supported by the National Natural Science
Foundation of China (No. 21306137, and No. U1162104).

References

[1]. D.S. Wishart, Drugs R. D. 2007, 8, 349-349.
[2]. I. Kola, J. Landis, Nat.Rev. Drug Discov. 2004, 3, 711–716. [3]. P. A. Routledge, Br. J. Clin. Pharmacol. 1986, 22, 499–506. [4]. G. Colmenarejo, Med. Res. Rev. 2003, 23, 275–301.
[5]. G. Colmenarejo, A. Alvarez-Pedraglio, J. L. Lavandera, J. Med. Chem. 2001, 44,
4370–4378.
[6]. O. Deeb, B. Hemmateenejad, Chem. Biol. Drug Des. 2007, 70, 19–29.
[7]. E. Estrada, E. Uriarte, E. Molina, Y. Simon-Manso, G.W. Milne, J. Chem. Inf.
Model. 2006, 46, 2709–2724.
[8]. S. B. Gunturi, R. Narayanan, A. Khandelwal, Bioorg. Med. Chem. 2006, 14,
4118–4129.
[9]. L. M. Hall, L. H. Hall, L. B. Kier, J. Chem. Inf. Comput. Sci. 2003, 43,
2120–2128.
[10]. K. Wichmann, M. Diedenhofen, A. Klamt, J. Chem. Inf. Model. 2007, 4,
228–233.
[11]. C. X. Xue, R.S. Zhang, H. X. Liu, X. J. Yao, M. C. Liu, Z. D. Hu, et al., J. Chem.
Inf. Comput. Sci. 2004, 44, 1693–1700.
[12]. P. J. Hajduk, R. Mendoza, A. M. Petros, J. R. Huth, M. Bures, S. W. Fesik, et al., J.Comput. Aided Mol. Des. 2003, 17, 93–102.
[13]. J. R. Votano, M. Parham, L. M. Hall, L. H. Hall, L. B. Kier, S. Oloff, et al., J.
Med. Chem. 2006, 49, 7169–7181.
[14]. L. J. Chen, X Chen, Journal of Molecular Graphics and Modelling. 2012, 33,
35–43
[15]. Q. Wang. et al., Chemosphere, http: dx. doi. org/ 10. 1016/ j. chemosphere.
2014.02. 030
[16]. Z. Ch. Zhu, Q Wang, Q. Z. Jia, et al. Sh, Acta Phys. -Chim. Sin. 2013, 29 (1),
30-34.
[17]. A. Golbraikh, A. Tropsha, J. Mol. Graph. Model. 2002, 20, 269–276
[18]. A. Tropsha, P. Gramatica, V.K. Gombar, QSAR Comb. Sci. 2003, 22, 69–77. [19]. S. Zhang, L. Wei, K. Bastow, W. Zheng, A. Brossi, K.-H. Lee, A. Tropsha, J.
Comput. Aided Mol. Des. 2007, 21, 97–112.