CC BY
20
Scientific Research of the Union of Scientists in Bulgaria - Plovdiv, series G. Medicine, Pharmacy and Dental medicine, Vol. XVII, ISSN 1311-9427, International Conference of Young Scientists, 11 - 13 June 2015, Plovdiv
IN SILICO MODELLING TO PREDICT TRANSCELLULAR PERMEABILITY OF BIOACTIVE COMPOUNDS A. Diukendjieva, L. Marinov, P. Alov, I. Tsakovska, I. Pajeva Institute of Biophysics and Biomedical Engineering, Bulgarian Academy of Sciences, Sofia, Bulgaria
Abstract
The parallel artificial membrane permeation assay (PAMPA) is a high throughput in vitro assay system that evaluates transcellular permeation of small drug-like molecules [2]. PAMPA is used in the pharmaceutical research to screen for human intestinal absorption because PAMPA permeability has been shown to correlate with both Caco-2 cell permeability and human intestinal absorption. In the present study we report highly predictive QSAR models for a data set of nearly 300 diverse drugs with PAMPA permeability coefficients measured at pH 6.5 and 7.4 [1]. The best QSAR models included the apparent partition coefficient, the topological polar surface area, and the molecular weight of the compounds. The models were implemented in the open source knowledge-mining platform KNIME and can be easily applied for screening of chemical libraries to select compounds with suitable permeability.
Introduction
Parallel artificial membrane permeability assay (PAMPA) has been introduced by Kansy et al. (1998) [2] to predict the oral absorption of new therapeutic agents in a simple, reproducible and high-throughput manner. The assay measures effective / apparent or intrinsic permeability coefficients (Pe / Papp, P0), and flux / transport (%F / %T). Permeability coefficients are defined as number of molecules (mol) diffusing through a unit cross-section of the membrane (cm2) per unit of time (s) under a unit of concentration (mol-cm-3) gradient. The assay is performed in assemblies, consisting of: (1) donor compartment containing aqueous solution of the test compound; (2) acceptor compartment containing aqueous buffer initially free of the test compound. (3) lipid / hydrocarbon membrane used to separate the donor and acceptor compartments; (4) filter used for immobilization and stabilization of the membrane. A number of PAMPA variants have been developed based on modifications of the common experimental setting. The most frequently used PAMPA models for intestinal absorption that differ in the membrane composition are: HDM-PAMPA (n-hexadecane membrane, Wohnsland and Faller, 2001) [3], egg-PAMPA (egg lecithin, Kansy et al., 1998) [2], DOPC-PAMPA (dioleoyl phosphocholine in dodecane membrane, Avdeef et al., 2001) [4], BM-PAMPA (or BAMPA, biomimetic lipid mixture membrane, Sugano et al., 2001) [5] and DS-PAMPA (Double-Sink experimental setting, lipid mixture membrane, Avdeef, 2012) [1].
It has been shown that PAMPA permeability correlates well with CaCo-2 permeability and human intestinal absorption in vivo. The correlation was confirmed studies reported by Ano et al., 2004; Fujikawa et al., 2005, 2007; Verma et al., 2007 [6-9]. Therefore, in silico modelling of PAMPA permeability is identified as a tool to aid estimation of bioavailability of drugs and other bioactive compounds after oral administration. QSAR (quantitative structure-activity relationship) models are especially valuable in this context with their ability to predict physico-chemical, biological (incl. toxicological) and environmental fate properties of compounds from knowledge
of their chemical structures. QSAR models of PAMPA permeability may provide practical estimations of passive gasteo-intestinal ahsorption (GIA) of low molecular weight compounds and can give a deeper insrght irte the mechanisms of the membrane transport. A limited number of QSAR models for PAM1PA permeability exist in the scifntifir literature, probafly berause of the hssaf's novelty.
In this study we report QSAR models derived from a data set of nearly 300 diverse drugs with PAMPA permeability coefficients measured at pH 6.5 ¡and 7.4. The models have high predictihity and are implemented in the open souree knowledge-mining platform KNIME.
Materials and methods
For the modelling purposes we used PAMPA permeability coefficients measured by double-sink PAMPA assay gor 276 cempounds drom a dataset reported by Avdeef (22012) [S]. The ^tasee includes Pm (membrane permerbility) -values of the compounds mehsured at pH 6.5 and pH 7.4 and their logP0 (intrinsic; membrane permeability) values, collected from different papers of Avdeef's group. The compounds includpd are mainly commercial drugs and some pestipides.
Our modele ate developed improving on the multipte lineau regression model reported by Nakao et al. (2009) [10], whkh relptes tSe permeaMity parameter with calculated seructural descripterrs[ namely logP, |pKa - pH | and TPSA (Topologica[ Polar burface Aeea). We usud instead the apparent partition coefficient (logD) and TPSA divided by the molecular weight of the compounds (MW) as physrcocfemical descriptors to account for permeability. The rationale for changing the descriptors was that la) log D ~ logP - | pKa - pH |, and (b) MW accounts for the iotal surface tr estimate the ratio oh polar to totaS surface area. For the calculation of the descriptors we used ACD/Labs' PhysChem suite [11] and ChemAxon's Marvin calculutor plugins for logD [12] and CDK chemical properties KNIME node for TPSAand MW. The derived models were subjected to internal (Leave-One-Out cross-validation, LOO) and external validation and proved to have high predicitivity. The modee were implemented as a workflow in the; open source data analytics platform KNIME.
Results and discussion
With a view to the free and open use oo the models and lack of such took fof logD estimation, two models were fmplemented baaed on logD estimations readidy obtainable through free online services ChemSpider.com (calculated by ACD/Labs tools) and chemicaCize.org (calculated by hhemAxon tools).
After removal of the applicability domain and response outliers (Gfamatica, 2007) [13], the implemented mo^ls are:
Till £ tl lining irl«
■ ■•■ a »' i
i J m
■•S ■ •
ACDhLabs-calculated-(ogD model, Fig. 1:
logPm= M.945(±0 .2 288) +0.600(±0.046) logDpH74 □ 7.655(±0.811)TPSA/M W n=246, r25=0.7334, SEE=l.108,
adj
F,
d=j 338.861
q2 =0.729, external validation
J- cv 7
q2ext =0.696 (196 compounds in the training set, 50 - in the test set)
Fig. 1 Correlation between measured and calculated logPm by ACD/Labs-calculated-logD model. Dark dots represent test set data.
Test A ualning sits
ChemAxon-calculated-logD model. Fig. 2:
logPm=D2.280(±0.240) +0.494(±0.050) logDpH74 □10.063(±0.844)TPSA/MW n=247, r2, =0.711, SEE= 1.164,
7 adj 7 7
F2 244=303.713 q2 =0.705, external validation
1 cv 7
q2ext =0.722 (197 compounds in the training set, 50 - in the test set)
Fig. 2. Correlation between measured and calculated logPm by ChemAxon-calculated-logD model. Dark dots represent test set data.
The workflow that implements the QSAR models in KNIME consists of three parts -input (data submission and preparation), worker (calculation of the properties, logPm prediction and re liability of prediction ctlculations), and output (data preparation for report and download). In the input part tht user lias to submit the chemical structure(s) (as SMILES code), logD rnd tha source(s) of submitted logD. The data could de submitted as CSV file, or through the provided text input frrmu molecular editor interfaces. In the workea part the rest o° thr necessary molecular properties aae calculated using CDK KNIME nodes and, depending of the logD source, data are distributed in two nranches where logPm is predicted by the respective QSAR models. Additionally reliability of prediction (by extent of extrapolation, ar implemented sn Enalos Domain-Leverage nohe) is estimated in the worker paat of the workflow. In tide nutput part of the workflow the sorting order of the input data and obtained predictions aae restored (for the case were dhe input file contains dato passmg through different brancher of the worker), and two rets of kaia are prepared, for the report and for the output CSV file, which are reported to the user.
Fig. 3. Schematic diagram of the KNIME workflow Summary
This study reports highly ptedictve QSAR models for PAMPA permeability, whose implementation inan or>tn source platform provides a valuablu and aaeily accessible tool for prediction of gastrointestinal abuorption of bioactive compounds and dof screening of chemical libraries to select compounds with suitable permeability.
Acknowledgment
The funding from the European Community's 7th Framework Program COSMOS Project (grant n°266835) and from the Ministry of Education, Youth and Science, Bulgaria (grant n°D01-169/14.07.2014) is gratefully acknowledged
References
1. Avdeef, A., Absorption and Drug Development: Solubility, Permeability, and Charge State, 2nd edition, John Wiley and Sons, Inc., 2012, 744.
2. Kansy, M., F. Senner, K. Gubernator, Physicochemical high throughput screening: parallel artificial membrane permeation assay in the description of passive absorption processes, J. Med. Chem., 1998, 41, 1007.
3. Wohnsland F., Faller B., High-throughput permeability pH profile and high-throughput alkane/water log P with artificial membranes, J. Med. Chem., 2001, 44, 923.
4. Avdeef A., Strafford M., Block E., Balogh M.P., Chambliss W., Khan I., Drug absorption in vitro model: filter-immobilized artificial membranes 2. Studies of the permeability properties of lactones in piper methysticum forst, Eur. J. Pharm. Sci., 2001, 14, 271.
5. Sugano K., Hamada H., Machida M., Ushio H., Saitoh K., Terada K., Optimized conditions of bio-mimetic artificial membrane permeation assay, Int. J. Pharm., 2001, 228, 181.
6. Ano R., Kimura Y., Shima M., Matsuno R., Ueno T., Akamatsu M., Relationships between structure and high-throughput screening permeability of peptide derivatives and related compounds with artificial membranes: application to prediction of CaCo-2 cell permeability, Bioorg. Med. Chem., 2004, 12, 257.
7. Fujikawa M., Ano R., Nakao K., Shimizu R., Akamatsu M., Relationships between structure and high-throughput screening permeability of diverse drugs with artificial membranes: Application to prediction of CaCo-2 cell permeability, Bioorg. Med. Chem., 2005, 13, 4721.
8. Fujikawa M., K. Nakao, R. Shimizu and M. Akamatsu. QSAR study on permeability of hydrophobic compounds with artificial membranes, Bioorg. Med. Chem., 15, 37563767, 2007.
9. Verma R.P., Hansch C., Selassie C.D., Comparative QSAR studies on PAMPA/modified PAMPA for high throughput profiling of drug absorption potential with respect to CaCo-2 cells and human intestinal absorption, J. Comput. Aided Mol. Des., 2007, 21, 3.
10. Nakao K., Fujikawa M., Shimizu R., Akamatsu M., QSAR application for the prediction of compound permeability with in silico descriptors in practical use, J. Comput. Aided Mol. Des., 2009, 23, 309.
11. ACD/LogD , version 12.01, Advanced Chemistry Development, Inc., Toronto, Canada, www.acdlabs.com (2009)
12. Marvin 14.8.25.0, ChemAxon, www.chemaxon.com (2014)
13. Gramatica P. Principles of QSAR models validation: internal and external. QSAR Comb. Sci. 26: 694-701 (2007).
