Frontiers in Drug Design & Discovery Bentham Science Publishers Ltd. http://www.bentham.org/fddd
Volume 4, 2009
Contents In Vivo PK/PD and In Vitro ADME Techniques Editorial: Biopharmaceutical & Pharmacokinetic Considerations in Drug Design & Discovery G.W. Caldwell, Atta-ur-Rahman, Zhengyin Yan and M.I. Choudhary
i
Present and Future Mass Spectrometry-Based Approaches for Exploratory Drug Metabolism and Pharmacokinetic Studies Y. Hsieh
1
Glutathione Transferases in Drug Discovery and Development: Towards Safer and Efficacious Drugs K. Skopelitou, D. Platis, I. Axarli and N.E. Labrou
23
New Sampling Techniques for PharmacokineticPharmacodynamic Modeling C. Höcht, M. Mayer, J.A.W. Opezzo, G.F. Bramuglia and C.A. Taira
43
Role of Inflammatory Biomarkers in Establishing PK/PD Relationships and Target Organ Toxicity S. Pillarisetti and I. Khanna
81
Important Drug Interactions for Clinical Oncologists H. Ishiguro, I. Yano and M. Toi
97
Pharmacogenomic Considerations in Breast Cancer Management H. Ishiguro, I. Yano and M. Toi
122
More Than Skin Deep: The Human Skin Tissue Equivalent as an Advanced Drug Discovery Tool A.S. Bause, S.D. Lamore and G.T. Wondrak
135
Across Skin Barrier: Known Methods, New Performances K. Cal
162
The Tape Stripping Method as a Valuable Tool for Evaluating Topical Applied Compounds J.J. Escobar-Chávez, L.M. Melgoza-Contreras, M. López-Cervantes, D. Quintanar-Guerrero and A. Ganem-Quintanar
189
Methylphenidate Extended-Release Capsules: A New Formulation for Attention-Deficit Hyperactivity Disorder P. García-García, F. López-Muñoz, J.D. Molina, R. Fischer and C. Alamo
228
Osmotic-Controlled Release Oral Delivery System (OROS Technology) in Chronic Pain Management F. Coluzzi and C. Mattia
247
Drug Delivery Systems Prepared by Membrane Emulsification C. Charcosset and H. Fessi
273
In Silico ADME Approaches N.E. Campillo and J.A. Páez
291
Advances in ADMET Predictions and Modeling: Rapid Drug Discovery Efforts in 21st Centuries M.T.H. Khan
333
Computational Intelligence Methods for ADMET Prediction D. Hecht and G.B. Fogel
351
Data Modeling and Chemical Interpretation of ADME Properties Using Regression and Rule Mining Techniques K. Hasegawa and K. Funatsu
378
A Review on Virtual Reality and Haptics Approaches in Drug Design and Discovery S.K. Lai-Yuen
429
Contributors
454
Editorial
Frontiers in Drug Design & Discovery, 2009 Vol. 4 i
Editorial: Biopharmaceutical & Pharmacokinetic Considerations in Drug Design & Discovery The goal of all pharmaceutical scientists is to bring affordable, safe and effective drugs to patients globally. The Frontier in Drug Design and Discovery series is dedicated to contributed by leading researchers present comprehensive reviews with fresh new ideas on drug design and drug discovery. The first volume (2005) brought together experts to review and discuss the advantages and limitations of modern screening techniques used in the drug discovery process to identify potential drug candidates. The second volume (2006) discussed new technological and conceptual approaches to accelerate and to improve the predictability of the discoveries made in the laboratory into clinical testing. The third volume (2007) reviewed ways of applying structure-based design to identify potent lead drug candidates for a variety of diseases using techniques such as in-silico screening, peptidomimetics, fragment-based approaches, protein crystallography, and NMR spectroscopy. In the fourth volume of this series (2009), reviews and discussions are presented regarding biopharmaceutical and pharmacokinetic techniques to identify potent lead drug candidates. Pharmaceutical companies have embraced new technologies, such as analytical instrumentation, chemistry and biological screening robotic systems, computerized data handling systems, and computational and simulation software. While many successes have arisen from these changes, the cost to discover and market new drugs remains staggeringly high. There are many reasons why pharmaceutical drug development costs remain high. One reason is the continuing high attrition rates of drugs during costly phase II and III human clinical trials. One decision-making strategy used by pharmaceutical companies is to eliminate high-risk drug candidates as early as possible in the drug development process. By shifting the attrition of drug candidates to an earlier stage in development, pharmaceutical companies are aiming to cut the costs associated with developing and regulatory approval of a drug. The strategy used by most companies to identify unsuccessful likely drugs involves using biopharmaceutical and pharmacokinetic principles early in the discovery process to improve decision-making in drug selection. While biopharmaceutical principles are used to understand the physical properties of drug candidates, pharmacokinetic principles are used to quantitatively model the time course for drug concentration in the body and predict various parameters such as absorption, distribution, metabolism, and excretion (ADME) of the drug. Physical and ADME properties can be used to optimize drug design in relationship to pharmacological effects, such as efficacy and toxicity. Since efficacy and toxicity deficiencies are related in part to biopharmaceutical and pharmacokinetic properties, detecting problems in drug candidates as early as possible would be highly valuable in making go/no-go decisions. We have carefully selected authors to review biopharmaceutical, in-vitro ADME and pharmacokinetic applications in drug design. Hsieh has reviewed the current mass
ii Frontiers in Drug Design & Discovery, 2009, Vol. 4
Editorial
spectrometry-based approaches and their future potential in supporting exploratory invitro and in-vivo drug metabolism/pharmacokinetic studies, that include pharmacokinetic profiling, physical property, metabolite identification, and molecular imaging tests. The chapter by Labrou and colleagues gives a detailed description of several biotransformation reactions catalyzed by glutathione transferases (GSTs). This information is useful in the early phases of drug development to eliminate unsuitable drug candidates in the drug discovery process. Chapter by Höcht and colleagues gives an excellent review of sampling techniques, such as microdialysis and ultrafiltration. In addition, a review is presented on imaging techniques, such as positron emission topography and magnetic resonance spectroscopy. These techniques allow the measurement of drug concentration at the target site. The chapter by Pillarisetti and Khanna gives an overview of biomarkers, such as the tumor necrosis factor (TNF), interleukins (IL-1, IL-6, IL-8 and IL-18), vascular cell adhesion molecules, and markers of macrophage inflammation (e.g. MMPs). Their potential applications in monitoring inflammatory responses in normal animals and case histories linking these biomarkers to PK/PD correlation from preclinical and clinical studies are discussed. Ishiguro and colleagues give an overview of the importance of ADME properties in the understanding of antineoplastics drugs. Their second paper describes the important of pharmacogenomic factors associated with drugs for the treatment of breast cancer. Wondrak and colleagues wrote an excellent review of the development of in-vitro intact stratum corneum human skin reconstruct from matrixembedded dermal fibroblasts and epidermal primary keratinocytes. These human skin reconstructs, which closely resemble the complex architecture and functional complexity of skin, can be used to screen skin care products in a 96-well format. Cal contributed a well-balanced review of transdermal drug delivery principles. Escobar-Chávez and colleagues describe the applications of the tape stripping technique to evaluate drug penetration through the skin, as well as stratum corneum composition and physiology, underlining its versatile application in the area of topical and transdermal drugs. The chapter by López-Muñoz and colleagues describe the pharmacokinetics of new extended-release formulation of methylphenidate (Medikinet®) for the treatment of Attention Deficit Hyperactivity Disorder (ADHD). They compare of the characteristics of Medikinet® with those of other formulations used for treating. Coluzzi and Mattia review the osmoticcontrolled release oral delivery system (OROS®). This drug delivery technology, which employ osmotic pressure as the driving force to deliver pharmacotherapy, is demonstrated for hydromorphone. Chapter by Charcosset summarizes the progress made in drug delivery systems based on membrane emulsification. Campillo and Páez review the progress in ADME in-silico methods, focusing on oral absorption, blood-brain barrier, metabolism, and some aspects on excretion. Khan reviews in-silico ADMET predictions of the potential drug candidates based on Quantitative Structure-Activity Relationship (QSAR) modeling approaches. The chapter by Hecht and Fogel reviews the utility of some of the more popular applications of computational intelligence to Quantitative Structure-Property Relationship (QSPR) modeling approaches including: artificial neural
Editorial
Frontiers in Drug Design & Discovery, 2009 Vol. 4 iii
networks, fuzzy logic, and evolutionary computing. Hasegawa and Funatsu review ADME modeling techniques including regression, multiple linear regressions (MLR), partial least squares (PLS), artificial neural networks (ANN), and support vector machines (SVM). Finally, S.K. Lai-Yuen introduces the reader to virtual reality and haptics techniques for in-silico drug design. We are grateful to all contributors and also wish to express our gratitude of staff and editorial assistants for their untiring efforts in beinging this volume in good shape and in timely manner.
Atta-ur-Rahman M. Iqbal Choudhary
Gary W. Caldwell Zhengyin Yan
Editorial
Frontiers in Drug Design & Discovery, 2009 Vol. 4 i
Editorial: Biopharmaceutical & Pharmacokinetic Considerations in Drug Design & Discovery The goal of all pharmaceutical scientists is to bring affordable, safe and effective drugs to patients globally. The Frontier in Drug Design and Discovery series is dedicated to contributed by leading researchers present comprehensive reviews with fresh new ideas on drug design and drug discovery. The first volume (2005) brought together experts to review and discuss the advantages and limitations of modern screening techniques used in the drug discovery process to identify potential drug candidates. The second volume (2006) discussed new technological and conceptual approaches to accelerate and to improve the predictability of the discoveries made in the laboratory into clinical testing. The third volume (2007) reviewed ways of applying structure-based design to identify potent lead drug candidates for a variety of diseases using techniques such as in-silico screening, peptidomimetics, fragment-based approaches, protein crystallography, and NMR spectroscopy. In the fourth volume of this series (2009), reviews and discussions are presented regarding biopharmaceutical and pharmacokinetic techniques to identify potent lead drug candidates. Pharmaceutical companies have embraced new technologies, such as analytical instrumentation, chemistry and biological screening robotic systems, computerized data handling systems, and computational and simulation software. While many successes have arisen from these changes, the cost to discover and market new drugs remains staggeringly high. There are many reasons why pharmaceutical drug development costs remain high. One reason is the continuing high attrition rates of drugs during costly phase II and III human clinical trials. One decision-making strategy used by pharmaceutical companies is to eliminate high-risk drug candidates as early as possible in the drug development process. By shifting the attrition of drug candidates to an earlier stage in development, pharmaceutical companies are aiming to cut the costs associated with developing and regulatory approval of a drug. The strategy used by most companies to identify unsuccessful likely drugs involves using biopharmaceutical and pharmacokinetic principles early in the discovery process to improve decision-making in drug selection. While biopharmaceutical principles are used to understand the physical properties of drug candidates, pharmacokinetic principles are used to quantitatively model the time course for drug concentration in the body and predict various parameters such as absorption, distribution, metabolism, and excretion (ADME) of the drug. Physical and ADME properties can be used to optimize drug design in relationship to pharmacological effects, such as efficacy and toxicity. Since efficacy and toxicity deficiencies are related in part to biopharmaceutical and pharmacokinetic properties, detecting problems in drug candidates as early as possible would be highly valuable in making go/no-go decisions. We have carefully selected authors to review biopharmaceutical, in-vitro ADME and pharmacokinetic applications in drug design. Hsieh has reviewed the current mass
ii Frontiers in Drug Design & Discovery, 2009, Vol. 4
Editorial
spectrometry-based approaches and their future potential in supporting exploratory invitro and in-vivo drug metabolism/pharmacokinetic studies, that include pharmacokinetic profiling, physical property, metabolite identification, and molecular imaging tests. The chapter by Labrou and colleagues gives a detailed description of several biotransformation reactions catalyzed by glutathione transferases (GSTs). This information is useful in the early phases of drug development to eliminate unsuitable drug candidates in the drug discovery process. Chapter by Höcht and colleagues gives an excellent review of sampling techniques, such as microdialysis and ultrafiltration. In addition, a review is presented on imaging techniques, such as positron emission topography and magnetic resonance spectroscopy. These techniques allow the measurement of drug concentration at the target site. The chapter by Pillarisetti and Khanna gives an overview of biomarkers, such as the tumor necrosis factor (TNF), interleukins (IL-1, IL-6, IL-8 and IL-18), vascular cell adhesion molecules, and markers of macrophage inflammation (e.g. MMPs). Their potential applications in monitoring inflammatory responses in normal animals and case histories linking these biomarkers to PK/PD correlation from preclinical and clinical studies are discussed. Ishiguro and colleagues give an overview of the importance of ADME properties in the understanding of antineoplastics drugs. Their second paper describes the important of pharmacogenomic factors associated with drugs for the treatment of breast cancer. Wondrak and colleagues wrote an excellent review of the development of in-vitro intact stratum corneum human skin reconstruct from matrixembedded dermal fibroblasts and epidermal primary keratinocytes. These human skin reconstructs, which closely resemble the complex architecture and functional complexity of skin, can be used to screen skin care products in a 96-well format. Cal contributed a well-balanced review of transdermal drug delivery principles. Escobar-Chávez and colleagues describe the applications of the tape stripping technique to evaluate drug penetration through the skin, as well as stratum corneum composition and physiology, underlining its versatile application in the area of topical and transdermal drugs. The chapter by López-Muñoz and colleagues describe the pharmacokinetics of new extended-release formulation of methylphenidate (Medikinet®) for the treatment of Attention Deficit Hyperactivity Disorder (ADHD). They compare of the characteristics of Medikinet® with those of other formulations used for treating. Coluzzi and Mattia review the osmoticcontrolled release oral delivery system (OROS®). This drug delivery technology, which employ osmotic pressure as the driving force to deliver pharmacotherapy, is demonstrated for hydromorphone. Chapter by Charcosset summarizes the progress made in drug delivery systems based on membrane emulsification. Campillo and Páez review the progress in ADME in-silico methods, focusing on oral absorption, blood-brain barrier, metabolism, and some aspects on excretion. Khan reviews in-silico ADMET predictions of the potential drug candidates based on Quantitative Structure-Activity Relationship (QSAR) modeling approaches. The chapter by Hecht and Fogel reviews the utility of some of the more popular applications of computational intelligence to Quantitative Structure-Property Relationship (QSPR) modeling approaches including: artificial neural
Editorial
Frontiers in Drug Design & Discovery, 2009 Vol. 4 iii
networks, fuzzy logic, and evolutionary computing. Hasegawa and Funatsu review ADME modeling techniques including regression, multiple linear regressions (MLR), partial least squares (PLS), artificial neural networks (ANN), and support vector machines (SVM). Finally, S.K. Lai-Yuen introduces the reader to virtual reality and haptics techniques for in-silico drug design. We are grateful to all contributors and also wish to express our gratitude of staff and editorial assistants for their untiring efforts in beinging this volume in good shape and in timely manner.
Atta-ur-Rahman M. Iqbal Choudhary
Gary W. Caldwell Zhengyin Yan
Frontiers in Drug Design & Discovery, 2009, 4, 1-22
1
Present and Future Mass Spectrometry-Based Approaches for Exploratory Drug Metabolism and Pharmacokinetic Studies Yunsheng Hsieh* Department of Drug Metabolism and Pharmacokinetics, Schering-Plough Research Institute, 2015 Galloping Hill Road, K-15-3700, Kenilworth, NJ 07033, USA Abstract: For more than a decade, mass spectrometry (MS) has played an important role in absorption, distribution, metabolism, excretion and toxicology (ADMET) studies for drug discovery to help convert lead compounds into drug candidates. Drug discovery efforts have been focused on identifying drug metabolism and pharmacokinetic (DMPK) issues at the earliest possible stage to reduce attrition rate of drug candidates throughout the drug development process by applying cutting edge MS-based techniques. These emerging techniques have proven to be extremely valuable to accelerate the lead optimization and characterization processes by eliminating potentially unpromising candidates. In this article, the current MS-based approaches and their future perspectives in supporting exploratory DMPK studies including in vitro and in vivo pharmacokinetic profiling, physical property, metabolite identification and molecular imaging tests are reviewed.
INTRODUCTION Combinatorial chemistry is an efficient way of synthesizing and testing very large libraries for bioactivity to discover leads more quickly than was formerly possible. The selected leads with an appropriate in vitro potency are further screened to provide medicinal chemists relevant pharmacokinetic parameters such as tissue penetration, protein binding, intestinal absorption, drug-drug interaction potential, metabolic stability and pathway to embark on a lead optimization campaign. Drug design is a fast moving and iterative process for creating a chemical compound with drug-like properties suitable for development as a therapeutic agent (Fig. 1). The early evaluation of pharmacokinetics, pharmacodynamics and toxicological effects should lead drug candidates with optimum performance characteristics to proceed further into development. The major purposes of obtaining DMPK knowledge during the early drug discovery stage are to identify DMPK issues earlier to resolve problems arising from lead optimization, to ensure sufficient exposure multiples achievable in all toxicology test species and to reduce the attrition rate for drug development [1-10].
*Corresponding Author: E-mail:
[email protected]
Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
2 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
Modified drug discovery compounds Biological activity Protein binding Drug-drug interaction
Physical properties
Metabolite identification
In vitro screening
Permeability
Metabolic stability
In vivo pharmacokinetics / pharmacodynamics In vivo toxicology Nomination for preclinical development Fig. (1). Strategy for use of early ADME screens in the lead optimization and characterization processes.
Drug-drug interactions can occur when one drug inhibits the biotransformation of another if patients take more than one drug. Cytochrome (CYP) P450s are a family of enzymes primarily responsible for metabolizing xenobiotics [11-15]. For CYP enzyme inhibition screening, direct and metabolism/ mechanism-based inhibition potentials of CYP of new leads are frequently quantified in terms of 50% inhibitory concentration (IC50) values for prioritizing [11]. For direct CYP inhibition measurement, the individual compounds are incubated with human liver microsomes, probe substrate and NADPH. The IC50 is calculated using certain substrate assays by fitting the inhibition data to the following equation; y = 100/ [1+(x / IC50)S] where y is % of control, x is concentration of the inhibitor and S is a slope factor [12]. Metabolite production from each enzyme inhibition assay can also be quantified to reflect the ability of the probe compounds to
Present and Future Mass Spectrometry-Based
Frontiers in Drug Design & Discovery, 2009, Vol. 4 3
directly inhibit the activity of P450s. For metabolism/mechanism-based inhibition, the test compounds are generally pre-incubated first with liver microsomes in the presence of NADPH so that metabolites might be produced before the addition of the test compounds. Compounds with low IC50 values under the pre-incubation conditions compared with co-incubation conditions are suspected to be metabolism/mechanism-based inhibitors. Oral administration is the preferred drug-dosing route for the drug development in the pharmaceutical industry. Various in vitro assays such as Caco-2, a human intestinal epithelia cell line, as a rapid permeability screening tool have been developed to predict the extent of absorption of drug candidates after oral administration in early discovery [16-20]. As an example, the Caco-2 immortal cell line forming a monolayer presents a barrier to ion flow due to the tight junctions between cells. Compounds with sufficient rates of passive diffusion across membranes will demonstrate good permeability in the Caco-2 experiments. The analyte concentrations from both sides of the cell were used to calculate the permeability and recovery of the drug candidates as follows [16]: Permeability = (VR/(S*7200)) (CR,2 hr / CD,0 hr) % Recovery = [(CD,2 hr *VD *CR,2 hr *VR) / ( CD,0 hr *VD)]*100 where S is the membrane surface area; CD, 0hr and CD, 2hr are the donor concentration immediately after dosing and after two hours incubation, respectively; CR, 2hr is the receiver concentration after two hours incubation; VD and VR are the volume of donor and receiver compartments, respectively. In addition to oral absorption, first-pass metabolism, a process of converting a dosed compound to more water soluble metabolites to be easily eliminated from the human and animal body, is the other factor to contribute the bioavailability of a drug after oral administration. In general, drug metabolism is divided into phase I and phase II reactions. Phase I reactions involve oxidation reduction and hydrolysis reactions primarily catalyzed by CYP and flavin-containing monooxygenases. Phase II reactions known as conjugation reactions involve the interactions of the polar functional groups of phase I metabolites [21-27]. Metabolic stability refers to the susceptibility of compounds to biotransformation. In the context of drug discovery screens, intrinsic metabolic clearance (CLint), the proportionality constant between drug concentrations at the enzyme site (Ce) and rate of metabolism (vo), is often referred to as metabolic stability: CLint = vo / Ce [28]. The intrinsic half-life is determined by incubating a test compound with liver subcellular preparations for a certain period of time and plotting ln % remaining versus times as t1/2 = 0.693 / m where m is the slope of the plot. Integration of metabolic stability assessment with identification of the site (soft spot) of metabolism is valuable in leading toward a modified compound with an appropriate rate of metabolism to achieve an ideal pharmacokinetic profile. Meanwhile, characterization of reactive and pharmacologically active metabolites following incubation in microsomal systems for a predetermined period is also critical in designing new drug candidates with an improved toxicological profile and efficacy [29-37]. Detection of glutathione conjugates in both in vitro and in vivo suggests the production of potentially reactive intermediates. Acyl-glucuronides metabolites are susceptible to both hydrolysis and intramolecular acyl migration to form reactive intermediates for covalently binding to proteins which may implicate in various drug-induced toxicities [29, 38]. The other objectives of metabolite profiling in drug research settings are to characterize major me-
4 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
tabolites across species and to identify human-specific metabolites to allow for assessment of potential human safety issue. Animal PK results are considered the most predictive of human PK parameters and used to select compounds with no potential PK deficiency such as poor bioavailability, short half-life, circulating reactive metabolites and so on for drug development [39-47]. A variety of in vivo PK parameters such as oral bioavailability, volume of distribution, clearance, and mean resident time calculated from the drug concentrations in animal plasma samples can sometimes be employed to predict human PK via allometric scaling. Animal studies can also be used to assess the single-dose and multiple-dose non-clinical toxicity liability in the final lead characterization of drug candidates. Also, to conduct an Animal Safety program to meet IND requirements for enabling Phases I clinical trials, it is important to achieve sufficient exposure multiples following drug administration to the rodent and non-rodent toxicological species. In all aforementioned cases, what have enabled these in vitro and in vivo experiments to drug discovery to be efficient are the development of compound-specific analytical methods. Among these methods, MS-based approaches as described in the following sections have evolved to become irreplaceable techniques in the analysis of drug- related molecules in biological samples (Table 1) [48-55]. HIGH PERFORMANCE LIQUID CHROMATOGRAPHY-MS/MS The combination of high performance liquid chromatography and tandem mass spectrometry (HPLC-MS/MS) has rapidly matured to become a very powerful analytical tool applied in many DMPK areas for qualitative and quantitative determination of new chemical entities (NCEs) and the marketed pharmaceuticals (Fig. 2) [56-66]. Due to its inherent selectivity, the HPLC-MS/MS system normally requires neither labor-intensive sample preparation procedures nor extensive chromatographic run times to avoid matrix interference. This allows researcher to shorten chromatographic times and in turn to increase sample throughput. Micro-column [67-69], monolithic column [70-76], high temperature [77] and ultra high pressure [78-85] HPLC-MS/MS methods (Fig. 3) are four popular approaches to achieve faster HPLC cycle times for small molecule assays prior to tandem mass spectrometric detection. 200
100
0 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Fig. (2). Swift growth of HPLC-MS/MS applications for drug assays in the PubMed database.
Present and Future Mass Spectrometry-Based
Table 1.
Frontiers in Drug Design & Discovery, 2009, Vol. 4 5
Representative MS-Based Methods Used for DMPK Studies
Study
Rationale
MS-Based Method
Reference
P450 enzyme inhibition
To evaluate direct and mechanism-based inhibition of CYP enzymes that are responsible for metabolism of most drugs to assess the potential for drug-drug interactions.
HPLC-MS/MS Direct-injection HPLC-MS/MS LDTD-MS/MS
[11, 188] [189]
P450 enzyme profiling
To determine which P450 is primarily responsible for metabolizing the compound to anticipate drug-drug interactions and polymorphisms.
HPLC-MS/MS
[190]
Caco-2 permeability and efflux substrate screening
To evaluate permeability in human intestinal cell line to project in vivo absorption potential.
HPLC-MS/MS Nano-ESI-MS/MS
[19] [16]
Plasma protein binding
To determine the magnitude of plasma protein binding in different species. The free drug concentration can be used to interpret PK/PD relationships.
HPLC-MS Affinity chromatography-MS AMS
[191] [144, 192]
To evaluate compound stability following incubation in plasma. Results can assist in understanding an in vivo clearance value that exceeds hepatic blood flow.
Direct-injection HPLC-MS/MS HPLC-MS/MS
[112]
Intrinsic clearance (hepatocyte stability)
To evaluate the magnitude of hepatic clearance to predict in vivo clearance and assess species differences.
HPLC-MS/MS SFC-MS/MS Pulsed ultrafiltration-MS
[195] [126] [151]
In vitro and in vivo metabolite identification
To examine the metabolic profile of a compound across species with particular emphasis on the identification of any human specific metabolites.
HPLC-MS/MS nano-ESI-MS
[119, 196, 197] [156]
Animal PK study
To access in vivo PK parameters and to support the safety pharmacology cardiovascular study.
HPLC-MS/MS SFC-MS/MS nano-ESI-MS
[198-200] [140] [157, 201]
Blood-Brain barrier
To access the potential of compounds to penetrate the blood-brain barrier.
HPLC-MS/MS
[202, 203]
Mass balance
To evaluate the amount of drug that is recovered over time via the different elimination routes of the body.
AMS
[179]
Tissue Distribution
To access the distribution of a drug and its metabolites in various tissues from laboratory animal experiments.
MALDI-IMS
[185, 187]
Plasma stability
[163]
[193]
[194]
6 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
Higher-Throughput LC-MS Strategies in Drug Discovery Fast Chromatography Micro-column HPLC Monolithic column HPLC High-temperature HPLC Ultra performance liquid chromatography (UPLC) Parallel Instrumentation micro parallel liquid chromatography (uPLC) Staggered parallel liquid chromatography Multiple Ionization Sources APPI, ESI, APCI Ion-pairing HPLC Porous graphite carbon HPLC Mixed-mode HPLC Hydrophilic interaction chromatography (HILIC) Direct injection HPLC
Fig. (3). Strategies of higher throughput HPLC-MS approaches in drug discovery.
According to chromatographic concepts, one effective way to improve the HPLC column efficiency is to reduce the particle size of the packing materials that indicate better diffusion and mass transfer in the mobile and the stationary phases. Micro-column technique adopts a shorter (less than 5 cm length) narrow-bore (~ 2 mm i.d.) column packed with small particles and operated at higher than optimal flow rates to provide for fast chromatographic separations while still maintaining satisfactory chromatographic resolution. Monolithic silica columns made from a single piece of porous silica gel can be operated at higher flow-rates than conventional HPLC columns due to their lower back pressure. The low back pressure despite higher mobile phase flow rates is due to the higher permeability of monolithic silica versus particulate silica columns to make higher speed separation possible without a noticeable effect on chromatographic resolution. The viscosity of the mobile phases used in reverse-phase chromatography decreases as the temperature increases. Consequently, the column pressure drop, P, decreases significantly to allow higher flow rates within the normal HPLC pressure limits when elevated column temperature chromatography is employed. Ultra high-pressure liquid
Present and Future Mass Spectrometry-Based
Frontiers in Drug Design & Discovery, 2009, Vol. 4 7
chromatography (UHPLC) is another technical advance that allows the liquid handling system to handle the high backpressure resulting from the stationary phase with sub-2 m particles. UHPLC offers theoretical advantages in chromatographic resolution, speed and sensitivity over conventional HPLC systems. The concept of using supercritical porous silica packing through fused-core technology was further extended for the design of a novel stationary phase for separation of small molecules [79]. The columns packed with sub-2 m stationary phases require expensive ultra-pressure equipment to achieve optimum performance. The fused-core silica columns generate backpressure slightly higher than the columns normally packed with 3- and 5-m particles which are still compliable with most existing HPLC pumps (Fig. 4). 6 0.74
4.2e5
Intensity, cps
A 1
4 2
3
5 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
6
Time, min
B
Intensity, cps
4.4e5
4 2
3
1 5 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Time, min
Fig. (4). Fast separation of six pharmaceutical compounds using (A) a 2.7-m fused-core silica column (the first injection- (solid line) and the 250th injection- (dotted line)) and (B) 1.7-m porous silica column at 40 oC. Compounds 1 through 6 are ketoconazole, thioridazine, clofazimine, amiodarone, felodipine and rimonabant, respectively [79].
Due to its inherent selectivity and sensitivity, HPLC-MS/MS method allows the simultaneous determination of multiple components sharing the same retention times. An alternative approach for higher-throughput assay without sacrificing chromatographic integrity is the use of parallel HPLC where samples were injected alternately onto each of two analytical columns in parallel at specified intervals within a single chromatographic run time [86]. The parallel system makes use of unused chromatographic time windows and can be controlled by the computer where the analyst defines the acquisition
8 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
windows placed into HPLC timetable to provide a boost in throughput. To further speed up assays, the concept of using a staggered sample injection technique has been extended to multiple parallel HPLC-MS/MS methods for quantitative applications [87, 88]. The exponential growth in HPLC-MS applications is mainly due to the introduction of the atmospheric pressure ionization (API) interfaces between HPLC and a mass spectrometer. Electrospray ionization (ESI) and atmospheric pressure chemical ionization (APCI) sources are the major ionization interface designs for the HPLC-MS/MS systems for qualitative and quantitative analysis of small molecules over the last decade. However, there is no single ionization source for HPLC-MS/MS system to cover all target compounds. New instrumental developments have resulted in combined sources such as APCI/ESI and APPI/APCI sources for on-line HPLC-MS/MS [89-91]. For example, the ESI/APCI source allows alternate on-line ESI and APCI scans with polarity switching within a single analysis. The new combined sources have reduced the analysis time of sample plates by eliminating the need for a source hardware change and repeat analyses. One of the common goals in the exploratory PK area is to develop a generic HPLCMS/MS method to simultaneously monitor a wide range of drug candidates and their metabolites in biological fluids. Reversed-phase HPLC is the most widely employed technique in pharmaceutical analysis due to its extensive application to most small drug molecules which are separated by their degree of hydrophobic interaction with the stationary phase. However, it is very challenging to establish a reliable bioanalytical method for the determination of NCEs with lower octanol-water partition coefficient, log P, because these polar compounds show little or no retention on traditional reversed-phase columns prior to mass spectrometric detection. Ion-pairing chromatography was reported as an effective way for obtaining satisfactory retention of polar analytes under reversedphase conditions [92-95]. The ion-pairing reagents added into mobile phase are used to improve chromatographic retention on the lipophilic stationary phase through the formation of neutral ion pairs. It was proven that the porous graphic carbon (PGC) column was able to provide more efficient retention than other kinds of the reversed-phase columns designed to trap and separate very polar compounds [96, 97]. The overall retention on PGC columns involves two major mechanisms: (1) dispersive interaction between analyte-mobile phase and analyte-graphite surface and (2) dipolar and ionic interaction of a polar analyte with the polarizable graphite. The mixed-mode column with an embedded ion-pairing group in the reversed-phase stationary providing the capability for ionexchange and hydrophobic interactions requires no ion-pairing reagent in the mobile phase to retain and to separate ionizable polar compounds. The mixed-mode HPLC allows for retaining hydrophobic compounds by the reversed-phase mechanism and hydrophilic compounds by the ion exchange mechanism at higher organic contents in the mobile phase [98]. Hydrophilic interaction chromatography (HILIC) using low aqueous/high organic mobile phase is emerging as another valuable supplement to the reversed-phase HPLC–MS/MS for the retention of polar pharmaceuticals [99-105]. The solvents used for HILIC-MS/MS systems, such as methanol, provide low column backpressure but also are MS-favorable for greater ionization efficiency. A simplified sample preparation step using protein precipitation technique to remove proteins from biological samples is normally sufficient for small molecule determination when HPLC-MS/MS methods are employed. This single clean-up step is required to prevent the HPLC column from clogging in reversed-phase chromatography but also to
Present and Future Mass Spectrometry-Based
Frontiers in Drug Design & Discovery, 2009, Vol. 4 9
avoid ion source contamination to reduce matrix ionization suppression in the mass spectrometer. The working principle of on-line extraction phases is to isolate macromolecules from the targeted small molecules in biological fluids based upon their sizes. The large macromolecules such as proteins, which are unable to penetrate into the hydrophobic pores, are first eluted to waste. The small molecules such as drug compounds that penetrate the pores are retained through the hydrophobic forces [106-111]. The directinjection HPLC-MS/MS method had been applied for the automatic measurement of plasma stability of drug candidates to eliminate the traditional labor-intensive sample preparation procedures [112]. The procedure makes use of a thermostatic autosampler as an incubator combined with the direct-injection HPLC-MS/MS method (Fig. 5). The untreated plasma samples from several species containing the test compound was directly and sequentially injected into a mixed-function column for on-line protein removal and chromatography. The injection-to-injection time was set at a certain time period. The peak responses of the test article were repeatedly monitored after each injection cycle which can be related to a constant incubation time period for individual plasma samples. The stability results of the test compound in all plasma samples obtained by the direct-injection HPLC-MS/MS method was found to be consistent with those obtained by the traditional HPLC-MS/MS approach (Fig. 4). Mouse Rat
Human Monkey Dog
120
% Remaining
100 80 60 40 20 0
Direct-injection HPLC-MS/MS
0
30
60
90
120 150 180 210 240 270 300
Time (min)
Result Spiking + Autosampler Fig. (5). Schematic of a direct-injection HPLC-MS/MS system for semi-automatic plasma stability measurement.
Tandem mass spectrometry covering a variety of scanning techniques such as precursor ion, neutral-loss and product ion scanning provides complementary information on small molecular structures [113, 114]. Triple-quadrupole (QqQ) is a tandem mass spectrometer commonly used for conducting product ion scan (MS2), precursor ion scan, neutral loss scan modes for structure elucidation and selected reaction monitoring (SRM) for the quantitation of trace components. The QqQ mass spectrometer is the most popular instrument for both quantitative pharmacokinetic (PK) assays and qualitative DMPK studies [50-52]. Linear ion trap (LIT) mass spectrometers perform multiple MS/MS (MSn) with a much faster trap scan speed than the QqQ mass spectrometer to generate more detailed structure information. A hybrid quadrupole linear ion trap (QqLIT) combines the features of both QqQ and LIT mass spectrometers with improved performances for supporting DMPK studies under one system [115-125]. With the QqLIT technology, collision-induced dissociation (CID) occurs in a quadrupole collision
10 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
cell and fragment ions are trapped and analyzed in the LIT mode to simultaneously perform multiple experiments without the need to repeat the runs on different MS platforms and substantial sensitivity loss. The hyphenation of HPLC and tandem mass spectrometry has widely become a powerful analytical tool for the characterization of metabolites of pharmaceuticals. SUPERCRITICAL FLUID CHROMATOGRAPHY-MS Supercritical fluid chromatography (SFC) is a hybrid of gas and liquid chromatography that eases the resolution of a mixture of compounds not conveniently resolved by either gas or liquid chromatography. The mobile phases for SFC have low viscosities and high diffusion coefficients properties compared to those for HPLC to allow for high efficiency separations. The commercial API sources used for HPLC-MS/MS system were proven to be applicable to the SFC-MS/MS system with no modification [126,127] (Fig. 6). According to the US Food and Drug Administration’s (FDA) policy statement for the development of new stereoisomeric drugs, in order to evaluate the pharmacokinetics of a single enantiomer or mixture of enantiomers, manufacturers should develop quantitative assays for individual enantiomers in in vivo samples early in drug development. SFC, offering higher resolution per unit of time and faster column re-equilibration, has been becoming the top choice among enantioselective chromatographic techniques such as capillary electrophoresis (CE) [128-130] and HPLC [131-135]. Although the MS/MS detection provides little selectivity for stereoisomeric drugs, it can provide complete resolution of the administered drugs from endogenous materials and their metabolites. By taking advantage of the inherent selectivity and sensitivity of MS/MS detection, implementing chiral SFC-MS/MS method together with sample pooling technology allows us to simultaneously monitor samples containing multiple pairs of enantiomers with different masses within the same chromatographic run to further increase assay productivity [136]. The coupling of chiral SFC to MS/MS detection was shown to be comparable for most bioanalytical attributes such as specificity, linearity, accuracy and ruggedness but advantageous for higher sample throughput and peak resolution power with respect to the HPLC-MS/MS approach [137-139]. Oven & analytical column Autosampler SFC
Tandem mass spectrometer
Fig. (6). Schematic of SFC-MS/MS system.
SFC best emulates normal-phase chromatography, but has the capability to separate a much broader range of analyte polarities amenable to SFC-MS if appropriate mobilephase modifiers, additives and columns are utilized [140-142]. As a general strategy,
Present and Future Mass Spectrometry-Based
Frontiers in Drug Design & Discovery, 2009, Vol. 4 11
CO2 combined with more polar solvents containing either acidic or basic additives as modifier to enhance the solvent strength of the mobile phase will allow for the elution of the analytes on packed column with high efficiency. SFC is not limited to relatively nonpolar compounds but could be extended to allow for the elution of polar and ionic compounds. In HPLC, the direct serial coupling of achiral and chiral columns for the simultaneous achiral/chiral separation of a range of components in a complex mixture was not feasible due to either the increased backpressure or the different mobile-phase requirement. However, this concept is not restricted for SFC because the supercritical fluid has a significantly lower viscosity than a liquid, but also is an effective eluent for both chiral and achiral stationary phases [140]. AFFINITY CHROMATOGRAPHY-MS Affinity chromatography makes use of affinity-based interactions for the analysis of specific analytes within a sample. One of the examples of these interactions includes the binding of a serum albumin with drugs where albumin is the major contribution to drugprotein binding in the plasma [143-147]. Such binding processes are performed by placing a serum albumin as a known ligand onto a solid support within a column to interact with the desired analyte. Immobilized serum albumin affinity chromatography coupled to tandem mass spectrometry has been applied for chiral separation and drug-protein binding determinations [146, 147]. In this approach, the test compounds are injected to immobilized serum albumin column and the expression of 100 [k’ / (k’+ 1)] is used to estimate the % protein binding of the analytes where k’ is the capacity factor. Good correlation between the protein binding results of the analytes obtained from the affinity chromatography-MS methodology and those obtained from ultrafiltration or dialysis methods were reported [144]. The compounds with the high non-specific binding to artificial membrane which are not applicable to ultrafiltration assay can be analyzed by this affinity chromatographic method. The immobilized serum albumin column-MS method enabled the simultaneous estimation of the individual protein-binding affinities of a mixture of compounds but also was proved to be reproducible to rank order a series of molecules according to their protein-binding affinities in less than an hour [147]. PULSED ULTRAFILTRATION-MS Ultrafiltration is a separation process using membranes with a variety of pore sizes to remove high molecular-weight substances. Pulsed ultrafiltration interfaced with MS was proved to be a powerful method with multiple uses for combinatorial libraries, metabolic stability and metabolites screening [148-151]. As an example, in the pulsed ultrafiltration (PU) experiment for metabolic stability screening, a pulse containing a library of compounds is loaded and pumped through a stirred ultrafiltration chamber fitted with an appropriate molecular weight cut-off ultrafiltration membrane (Fig. 7). Liver microsomes from various species are trapped in an ultrafiltration chamber. A continuous flow of a buffer is pumped through the chamber and into an API interface and a mass spectrometer. Substrates for cytochrome P450 together with cofactors are flow injected through the chamber and the targeted metabolites are then monitored on-line by using mass spectrometry [149]. Pulsed ultrafiltration in conjunction with LC-MS-MS can be extended to screen mixtures for compounds that might be activated metabolically by cytochrome P450. The pulsed ultrafiltration-MS method provides comparable qualitative results of a metabolic stability test with those obtained via a microsomer off-line incubation but has
12 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
the potential for increasing the throughput of such an analysis by carrying out multiple ultrafiltration reaction chambers in parallel with a HPLC injector/autosample and a single mass spectrometer [149].
Fig. (7). Scheme for a parallel pulsed ultrafiltration mass spectrometry system to screen drug candidates for drug metabolism [149].
COLUMNLESS MS MS-based methods frequently demonstrate no chromatographic interference from endogenous components. The attempt to further increase sample throughput has led some bioanalytical laboratories to replace chromatographic separation with flow injection analysis (FIA) [152-154], nanoelectrospray infusion [155-162], laser diode thermal desorption (LDTD) techniques [163], or matrix-assisted laser desorption/ionization (MALDI) [164-166] that eliminate the HPLC component prior to mass spectrometric detection. Nano-electrospray (nanoESI) ionization refers to an electrospray emitter consists of a glass capillary with a tip pulled to a narrow internal diameter operated at a flowrate between 10 to 100 nL/min. NanoESI technology initiated by microfluidic devices offers the possibility of increased sensitivity over conventional ESI. NanoESI system using a robotic platform for sample handling can automatically infuse 96 samples in approximately an hour to microfabricated nanoESI nozzles for unattended quantitative and qualitative analysis [117, 167]. For each analytical run, a new tip and a new nozzle is used to eliminate the potential carry-over issue associated with HPLC autosamplers. LDTD-MS is a direct sample introduction technique combining thermal desorption and mass spectrometry. The sample plate is held in an X-Y stage that positions it in alignment with the laser. The flare-shaped portion of the nozzle abuts on the 96-well plate. Thermal desorption by an infrared laser diode at the back of each well without addition of any matrix generates the gaseous molecules to be introduced directly into the mass spectrometer.
Present and Future Mass Spectrometry-Based
Frontiers in Drug Design & Discovery, 2009, Vol. 4 13
Laser power and duration and the carrier gas flowrate are the major parameters for optimization of the vaporization efficiency of analytes from the target samples [163]. Matrix-assisted laser desorption/ionization (MALDI) involves mixing the analyte with a large molar excess of a matrix solution. Once the mixture is applied to a stainless steel target, the sample is allowed to air evaporate for co-crystallization. The sample plate is then inserted into the mass spectrometer for laser bombardment. These crystals containing matrix materials are subject to absorb energy of the laser resulting in desorption and ionization of the analytes. The coupling of MALDI to a mass spectrometer has extended its application for the direct analysis of pharmaceuticals [165, 166]. The major concerns for direct detection of low molecular weight compounds extracted from biological fluids using MALDI-MS are poor shot-to-shot reproducibility due to the inhomogeneous distribution of analytes within the sample spots and ionization suppression [165]. In addition to a higher-throughput potential, these “columnless” direct MS-based methods offer the additional advantages over the hyphenated-MS methods which are no carry-over between samples, low sample consumption and no chromatographic method development time. However, extensive optimization of the sample clean up process would be normally required to eliminate matrix ion suppression and interferences from endogenous interferences or the metabolites of the dosed compound when attempting to perform these “columnless” MS-based assays. ACCELERATOR MS Accelerator mass spectrometry (AMS) is a nuclear physic technique to allow the remarkably sensitive means for counting elemental isotopes at the individual atom level in less than 1 milligram samples [168-171]. AMS uses a particle accelerator in conjunction with ion sources, magnets, filters and mass spectrometers to separate out interferences and detect single atoms in the presence of stable atoms. Briefly, various negative elemental or molecular ions are first created through bombardment by a beam of Cs+ ions to sputter material from the sample. The ion source is then introduced to the gas phase and enters an electrostatic accelerator. In the middle of the accelerator, all molecular species are destroyed and the outer valency electrons are stripped. The resulting positively charged species continue their acceleration toward a magnetic quadrupole lens. The desired isotope ions are separated and counted by a mass spectrometer [172, 173]. For the direct measurement of 14C atoms, AMS is at least 1000 times more sensitive than radioactivity decay counting and any other methods. This makes it possible to use for a wide variety of tracing applications for drug compounds enriched with 14C in the DMPK, pharmacology and toxicology areas [174-176]. For microdosing and mass balance studies, animals and human are administrated with 14C-labelled drug components [173, 177, 178]. At certain post-dosing time intervals, blood, urine and other samples are collected for AMS analysis. The specimen for AMS experiment must be converted to a thermally and electrically conductive solid form. As an example, carbon samples are reduced to graphite prior to each AMS experiment. Ultilization of ultrasensitive AMS technique permits drug administration with dose at least 1000 times lower than that used in conventional radioactive studies to reduce radiation exposure. Although AMS is not able to provide chemical identity information, it can be used to access the extent of metabolism through HPLC-MS metabolite profiling [179].
14 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
MALDI-IMAGING MASS SPECTROMETRY During the drug discovery and development stages, often the question is raised as to whether the drug can reach the site of action which helps researchers better assess the potential value of that compound as a pharmaceutical product and the potential for toxicological outcomes. Most MS-based approaches are not able to provide the answer to certain questions regarding the distribution of a drug in various organs or tissues from in vivo animal experiments. Whole body autoradiography (WBA) normally provides a standard way to answer this question on the time course of the drug candidates [180]. However, the major disadvantage in the autoradiographic technique is that it provides for visualization of total drug-related materials, and provides no discrimination of the administrated drugs vs. metabolites. In addition, the availability of radiolabeled compounds is another concern at the drug discovery stage. To overcome these issues, matrix-assisted laser desorption/ionization-mass spectrometric method (MALDI-MS) has been developed to directly and simultaneously determine the distribution of pharmaceuticals and their metabolites in tissue sections which might unravel their disposition or biotransformation pathway for new drug development [181-186]. The concept of MALDI-IMS was introduced to direct profiling of the analytes within a tissue section or an organ. In this process, matrix materials are uniformly deposited over various tissue sections to extract analytes into the surface of the tissue and to produce crystals. A raster of organ sections containing the compounds of interest from a small animal under a stationary laser beam is then performed over a predetermined twodimensional array to generate ion plumes directly from the tissue sections in a large (5 cm x 5 cm) sample plate sufficient to contain whole tissue slices. The movement of the sample stage is automatically accomplished in the x and y directions to locate the edges of the tissue sample and to define the exact region of interest. During an imaging mass spectrometric experiment, MALDI-MS signals of small molecules from tissue sections within a user-defined area are first obtained as a function of acquisition times which are associated with the location of an array of pixels. Thus, two-dimensional ion maps of biological tissues are reconstructed with drug signals of a given m/z value monitored in each spectrum from each pixel to provide specific molecular images. The reliability of using MALDI-IMS for direct intact analysis of pharmaceuticals in tissues had been cross-validated by autoradiographic techniques [187]. For a fair parallel comparison between these molecular imaging technologies to visualize the accumulation of a well-established drug in rat brains, the animals had been administrated by direct infusion with 3H-clozapine into rat brain to avoid any possible biotransformation. Rat brain was chosen as a tissue model due to its unique symmetry and well-defined anatomy. The optical image of a sagittal section of the brain defined the presence of the cortex, limbic system, cerebellum, brain stem, and ventricles (Fig. 8). Results from the auto-radiography suggest that clozapine is distributed throughout the brain, with the highest concentration found in the lateral ventricle. The result from MALDI-IMS was in a good agreement with that obtained from autoradiography with regard to the distribution of clozapine in rat brain after intracerebral ventricular injection. MALDI-IMS could also be beneficial as a diagnosis, screening or discovery tool where tissue sections can be explored without knowing in advance what specific molecules have changed in a comparative study. Image comparison between control tissues and study tissues from small molecules will allow researchers to identify differences in resulting changes induced by drug candidates or their metabolites that could yield important toxicological information.
Present and Future Mass Spectrometry-Based
Frontiers in Drug Design & Discovery, 2009, Vol. 4 15
(a)
(b)
(c)
Fig. (8). (A) Optical images, (B) radioautographic images and (C) MALDI-MS/MS images from the study rat brain tissue section.
FINAL REMARKS The role of the DMPK screens is to provide an efficient paradigm for improving the drug likeness of the NCEs by using various in-vitro and in-vivo experiments for lead optimization and characterization. Hyphenation of the high-resolving HPLC to a tandem mass spectrometer (MS/MS) provides straightforward method development capabilities with excellent analytical linearity, sensitivity and selectivity for monitoring drug-related
16 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Yunsheng Hsieh
compounds. HPLC-MS/MS techniques will continue to be the mainstream as a bioanalytical tool in DMPK screens for the next decade. The search for universal API interfaces to cover more NCEs and to look for strategies to ease the ever-increasing demand for faster sample turnaround time using HPLC-MS/MS are currently in progress. Any improvement in HPLC–MS/MS instrumentation should enhance not only throughput but also the chance of success in attempting the structural identification of metabolites in biological matrices. ABBREVIATIONS AMS
=
Accelerator Mass Spectrometry
APCI
=
Atmospheric Pressure Chemical Ionization
API
=
Atmospheric Pressure Ionization
CE
=
Capillary Electrophoresis
CLint
=
Intrinsic Metabolic Clearance
DMPK
=
Drug Metabolism and Pharmacokinetics
ESI
=
Electrospray Ionization
FDA
=
United State Food and Drug Administration
FIA
=
Flow Injection Analysis
HILIC
=
Hydrophilic Interaction Chromatography
HPLC
=
High Performance Liquid Chromatography
IMS
=
Imaging Mass Spectrometry
IND
=
Investigational New Drug
LDTD
=
Laser Diode Thermal Desorption
LIT
=
Linear Ion Trap
MALDI
=
Matrix-Assisted Laser Desorption/Ionization
MS
=
Mass Spectrometry
MS/MS
=
Tandem Mass Spectrometry
NCEs
=
New Chemical Entities
P450
=
Cytochrome P450
PGC
=
Porous Graphic Carbon
PU
=
Pulsed Ultrafiltration
QqLIT
=
Hybrid Quadrupole Linear Ion Trap
SFC
=
Supercritical Fluid Chromatography
UHPLC
=
Ultrahigh-Pressure Liquid Chromatography
WBA
=
Whole body autoradiography
Present and Future Mass Spectrometry-Based
Frontiers in Drug Design & Discovery, 2009, Vol. 4 17
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39]
Kola, I.; Landis, J. Nat. Rev. Drug Discov., 2004, 3, 711. White, R.E. Annu. Rev. Pharmacol. Toxicol., 2000, 40, 133. Caldwell, G.W.; Ritchie, D.M.; Masucci, J.A.; Hageman, W.; Yan, Z. Curr. Top Med. Chem., 2001, 1, 353. Korfmacher, W.A. Curr. Opin. Drug Discov. Devel., 2003, 6, 481. Di, L.; Kerns, E.H. Curr. Opin. Drug Discov. Devel., 2005, 8, 495. Di, L.; Kerns, E.H. Curr. Opin. Chem. Biol., 2003, 7, 402. Thompson, T.N. Curr. Drug Metab., 2000, 1, 215. Ekins, S.; Ring, B.J.; Grace, J.; McRobie-Belle, D.J.; Wrighton, S.A. J. Pharmacol. Toxicol. Methods, 2000, 44, 313. Bertrand, M.; Jackson, P.; Walther, B. Eur. J. Pharm. Sci., 2000, 2, S61. Lin, J.; Sahakian, D.C.; de Morais, S.M.; Xu, J.J.; Polzer, R.J.; Winter, S.M. Curr. Top. Med. Chem., 2003, 3, 1125. Chu, I.; Favreau, L.; Soares, T.; Lin, C, Nomeir, A.A. Rapid Commun. Mass Spectrom., 2000, 14, 207. Nakamura, K.; Watanabe, A.; Okudaira, N.; Okazaki, O.; Sudo, K. Drug Metab. Pharmacokinet., 2007, 22, 113. Di, L.; Kerns, E.H.; Li, S.Q.; Carter, G.T. Int. J. Pharm., 2007, 335, 1. Bu, H.Z.; Magis, L.; Knuth, K.; Teitelbaum, P. Rapid Commun. Mass Spectrom., 2001, 15, 741. Li, X.; Chen, X.; Li, Q.; Wang, L.; Zhong, D. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 852, 128. Van Pelt, C.K.; Zhang, S.; Fung, E.; Chu, I.; Liu, T.; Li, C.; Korfmacher, W.A.; Henion, J. Rapid Commun. Mass Spectrom., 2003, 17, 1573. Smalley, J.; Marino, A.M.; Xin, B.; Olah, T.; Balimane, P.V. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 854, 260. van Breemen, R.B.; Li, Y. Expert Opin. Drug Metab. Toxicol., 2005, 1, 175. Fung, E.N.; Chu, I.; Li C.; Liu, T.; Soares, A.; Morrison, R.; Nomeir, A.A. Rapid Commun. Mass Spectrom., 2003, 17, 2147. Wang, Z.; Hop, C.E.; Leung, K.H.; Pang, J. J. Mass Spectrom., 2000, 35, 71. Baranczewski, P.; Stanczak, A.; Kautiainen, A.; Sandin, P.; Edlund, P.O. Pharmacol. Rep., 2006, 58, 341. Baranczewski, P.; Stanczak, A.; Sundberg, K.; Svensson, R.; Wallin, A.; Jansson, J.; Garberg, P.; Postlind, H. Pharmacol. Rep., 2006, 58, 453. Masimirembwa, C.M.; Bredberg, U.; Andersson, T.B. Clin. Pharmacokinet., 2003, 42, 515. Masimirembwa, C.M.; Thompson, R.; Andersson, T.B. Comb. Chem. High Throughput Screen., 2001, 4, 245. Singh, S.S. Curr. Drug Metab., 2006, 7, 165. Roberts, S.A. Xenobiotica, 2001, 31, 557. Kassel, D.B. Curr. Opin. Chem. Biol., 2004, 8, 339. Thompson, T.N. Using Mass Spectrometry for Drug Metabolism Studies, Chapter 2, Korfmacher, W.A. Ed.; CRC Press: Boca Raton, FL. 2005; pp. 35-81. Ma, S.; Subramanian, R. J. Mass Spectrom., 2006, 41, 1121. Ma, S.; Chowdhury, S.K.; Alton, K.B. Curr. Drug Metab., 2006, 7, 503. Prakash, C.; Shaffer, C.L.; Nedderman, A. Mass Spectrom. Rev., 2007, 26, 340. Hop, C.E.; Tiller, P.R.; Romanyshyn, L. Rapid Commun. Mass Spectrom., 2002, 16, 212. Baillie, T.A. Chem. Res. Toxicol., 2006, 19, 889. Doss, G.A.; Baillie, T.A. Drug Metab. Rev., 2006, 38, 641. Prueksaritanont, T.; Lin J.H.; Baillie, T.A. Toxicol. Appl. Pharmacol., 2006, 217, 143. Clarke, N.J.; Rindgen, D.; Korfmacher, W.A.; Cox, K.A. Anal. Chem., 2001, 73, 430A. Castro-Perez, J.M. Drug Discov. Today, 2007, 12, 249. Samuel, K.; Yin, W.; Stearns, R.A.; Tang, Y.S.; Chaudhary, A.G.; Jewell, J.P.; Lanza, T. Jr.; Lin, L.S.; Hagmann, W.K.; Evans, D.C.; Kumar, S. J. Mass Spectrom., 2003, 38, 211. Mei, H.; Korfmacher, W.; Morrison, R. AAPS J., 2006, 8, E493.
18 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80]
Yunsheng Hsieh
Caldwell, G.W.; Masucci, J.A.; Yan, Z.; Hageman, W. Eur. J. Drug Metab. Pharmacokinet., 2004, 29, 133. Feng, M.R.; Lou, X.; Brown, R.R.; Hutchaleelaha, A. Pharm. Res., 2000, 17, 410. Mahmood, I.; Balian, J.D. Clin. Pharmacokinet., 1999, 36, 1. McGinnity, D.F.; Collington, J.; Austin, R.P.; Riley, R.J. Curr. Drug Metab., 2007, 8, 463. Kitamura, R.; Yamamoto, Y.; Nagayama, S.; Otagiri, M. Drug Metab. Dispos., 2007, 35, 1611. Kramer, J.A.; Sagartz, J.E.; Morris, D.L. Nat. Rev. Drug Discov., 2007, 6, 636. Pritchard, J.F.; Jurima-Romet, M.; Reimer, M.L.; Mortimer, E.; Rolfe, B.; Cayen, M.N. Nat. Rev. Drug Discov., 2003, 2, 542. Sadaghiani, A.M.; Verhelst, S.H.; Gocheva, V.; Hill, K.; Majerova, E.; Stinson, S.; Joyce, J.A.; Bogyo, M. Chem. Biol., 2007, 14, 499. Hopfgartner, G.; Bourgogne, E. Mass Spectrom. Rev., 2003, 22, 195. Rossi, D.T.; Sinz, M.W. Mass Spectrometry in Drug Discovery, Marcel Dekker, Inc., NY, 2002. Niessen, W.M.D. The Encyclopedia of Mass Spectrometry, Vol. 8, Elsevier, Oxford, UK, 2006. Lee, M. Integrated Strategies for Drug Discovery Using Mass Spectrometry, Marcel Dekker, Inc., NY, 2002. Korfmacher, W.A. Using Mass Spectrometry for Drug Metabolism Studies, John Wiley and Sons, Inc., Hoboken, NJ, 2005, pp. 1-34. Roddy, T.P.; Horvath, C.R.; Stout, S.J.; Kenney, K.L.; Ho, P.I.; Zhang, J.H.; Vickers, C.; Kaushik, V.; Hubbard, B.; Wang, Y.K. Anal. Chem., 2007, 79, 8207. Staack, R.F.; Hopfgartner, G. Anal. Bioanal. Chem., 2007, 388, 1365. Chu, I.; Nomeir, A.A. Curr. Drug Metab., 2006, 7, 467. Niessen, W.M. J. Chromatogr. A, 2003, 1000, 413. Lee, M.S.; Kerns, E.H. Mass Spectrom. Rev., 1999, 18, 187. Ackermann, B.L.; Berna, M.J.; Murphy, A.T. Curr. Top. Med. Chem., 2002, 2, 53. Ermer, J.; Vogel, M. Biomed. Chromatogr., 2000, 14, 373. Jemal, M. Biomed. Chromatogr., 2000, 14, 422. Lim, C.K.; Lord, G.; Biol. Pharm. Bull., 2002, 25, 547. Tiller, P.R.; Romanyshyn, L.A.; Neue, U.D. Anal. Bioanal. Chem., 2003, 377, 788. Kostiainen, R.; Kotiaho, T.; Kuuranne, T.; Auriola, S. Mass Spectrom., 2003, 38, 357. Hsieh, Y.; Fukuda, E.; Wingate, J.; Korfmacher, W.A. Comb. Chem. High Throughput Screen., 2006, 9, 3. Hsieh, Y.; Korfmacher, W.A. Curr. Drug Metab., 2006, 7, 479. Xu, R.N.; Fan, L.; Rieser, M.J.; El-Shourbagy, T.A. J. Pharm. Biomed. Anal., 2007, 44, 342. Hsieh, Y.; Brisson, J.M.; Wang, G.; Ng, K.; Korfmacher, W.A. J. Pharm. Biomed. Anal., 2003, 33, 251. Witters, E.; Laukens, K.; Deckers, P.; Van Dongen, W.; Esmans, E.; Van Onckelen, H. Rapid Commun. Mass Spectrom., 2003, 17, 2188. Hsieh, Y.; Chintala, M.; Mei, H.; Agans, J.; Brisson, J.M.; Ng, K.; Korfmacher, WA. Rapid Commun. Mass Spectrom., 2001, 15, 2481. Kirchherr, H.; Kuhn-Velten, W.N. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2006, 843, 100. Batycka, M.; Inglis, N.F.; Cook, K.; Adam, A.; Fraser-Pitt, D.; Smith, D.G.; Main, L.; Lubben, A.; Kessler, B.M. Rapid Commun. Mass Spectrom., 2006, 20, 2074. Huang, M.Q.; Mao, Y.; Jemal, M.; Arnold, M. Rapid Commun. Mass Spectrom., 2006, 20, 1709. van de Merbel, N.C.; Poelman, H. J. Pharm. Biomed. Anal., 2003, 33, 495. Peng, S.X.; Barbone, A.G.; Ritchie, D.M. Rapid Commun. Mass Spectrom., 2003, 17, 509. Hsieh, Y.; Wang, G.; Wang, Y.; Chackalamannil, S.; Brisson, J.; Ng, K.; Korfmacher, W.A. Rapid Commun. Mass Spectrom., 2002, 16, 944. Wu, J.T.; Zeng, H.; Deng, Y.; Unger, S.E. Rapid Commun. Mass Spectrom., 2001, 15, 1113. Hsieh, Y.; Merkle, K.; Wang, G. Rapid Commun. Mass Spectrom., 2003, 17, 1775. Wang, G.; Hsieh, Y.; Cheng, K.C.; Morrison, R.A.; Venkatraman, S.; Njoroge, F.G.; Heimark, L.; Korfmacher, W.A. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 852, 92. Hsieh, Y.; Duncan, C.J.; Brisson, J.M. Anal. Chem., 2007, 79, 5668. Hsieh, Y.; Duncan, C.J.; Lee, S.; Liu, M. J. Pharm. Biomed. Anal., 2007, 44, 492.
Present and Future Mass Spectrometry-Based [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [120]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 19
New, L.S.; Saha, S.; Ong, M.M.; Boelsterli, U.A.; Chan, E.C. Rapid Commun. Mass Spectrom., 2007, 21, 982. Ma, Y.; Qin, F.; Sun, X.; Lu, X.; Li, F. J. Pharm. Biomed. Anal., 2007, 43, 1540. Mensch, J.; Noppe, M.; Adriaensen, J.; Melis, A.; Mackie, C.; Augustijns, P.; Brewster, M.E. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 847, 182. Wang, G.; Hsieh, Y.; Cui, X.; Cheng, K.C.; Korfmacher, A. Rapid Commun. Mass Spectrom., 2006, 20, 2215. Kalovidouris, M.; Michalea, S.; Robola, N.; Koutsopoulou, M.; Panderi, I. Rapid Commun. Mass Spectrom., 2006, 20, 2939. Korfmacher, W.A.; Veals, J.; Dunn-Meynell, K.; Zhang, X.; Tucker, G.; Cox, K.A.; Lin, C.C. Rapid Commun. Mass Spectrom., 1999, 13, 1991. Jemal, M. Rapid Commun. Mass Spectrom., 2001, 15, 994. Briem, S.; Pettersson, B.; Skoglund, E. Anal. Chem., 2005, 77, 1905. Gallagher, R.T.; Balogh, M.P.; Davey, P.; Jackson, M.R.; Sinclair, I.; Southern, L.J. Anal. Chem., 2003, 75, 973. Short, L.C.; Hanold, K.A.; Cai, S.S.; Syage, J.A. Rapid Commun. Mass Spectrom., 2007, 21, 1561. Syage, J.A.; Hanold, K.A.; Lynn, T.C.; Horner, J.A.; Thakur, R.A. J. Chromatogr. A, 2004, 1050, 137. Hsieh, Y.; Duncan, C.J. Rapid Commun. Mass Spectrom., 2007, 21, 573. Fuh, M.R.; Haung, C.H.; Wu, T.Y.; Lin, S.L.; Pan, W.H. Rapid Commun. Mass Spectrom., 2004, 18, 1711. Quintana, J.B.; Reemtsma, T. Rapid Commun. Mass Spectrom., 2004, 18, 765. Gao, S.; Bhoopathy, S.; Zhang, Z.P.; Wright, D.S.; Jenkins, R.; Karnes, H.T. J. Pharm. Biomed. Anal., 2006, 40, 679. Hsieh, Y.; Duncan, C.J.; Brisson, J.M. Rapid Commun. Mass Spectrom., 2007, 21, 629. Xia, Y.Q.; Jemal, M.; Zheng, N.; Shen, X. Rapid Commun. Mass Spectrom., 2006, 20, 1831. Hsieh, Y.; Duncan, C.J.; Liu, M. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 854, 8. Hsieh, Y.; Chen, J. Rapid Commun. Mass Spectrom., 2005, 19, 3031. Chang, Y.W.; Yao, H.T.; Chao, Y.S.; Yeh, T.K. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 857, 195. Naidong, W.; Eerkes, A. Biomed. Chromatogr., 2004, 18, 28. Shou, W.Z.; Naidong, W. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2005, 825, 186. Li, W.; Li, Y.; Francisco, D.T.; Naidong, W. Biomed. Chromatogr., 2005, 19, 385. Naidong, W. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2003, 796, 209. Deng, Y.; Zhang, H.; Wu, J.T.; Olah, T.V. Rapid Commun. Mass Spectrom., 2005, 19, 2929. Santos-Neto, A.J.; Markides, K.E.; Sjoberg, P.J.; Bergquist, J.; Lancas, F.M. Anal. Chem., 2007, 79, 6359. Mao, Y.; Huang, M.Q.; Xia, Y.Q.; Jemal, M. J. Pharm. Biomed. Anal., 2007, 43, 1808. Hsieh, Y.; Wang, G.; Wang, Y.; Chackalamannil, S.; Korfmacher, W.A. Anal. Chem., 2003, 75, 1812. Hsieh, Y.; Bryant, M.S.; Brisson, J.M.; Ng, K.; Korfmacher, W.A. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2002, 767, 353. Hsieh, Y.; Brisson, J.M.; Ng, K.; White, R.E.; Korfmacher, W.A. Analyst, 2001, 126, 2139. Hsieh, Y.; Brisson, J.M.; Ng, K.; Korfmacher, W.A. J. Pharm. Biomed. Anal., 2002, 27, 285. Wang, G.; Hsieh, Y.; Lau, Y.; Cheng, K.C.; Ng, K.; Korfmacher, W.A.; White R.E. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2002, 780, 451. Kamel, A.; Prakash, C. Curr. Drug Metab., 2006, 7, 837. Lee, M.S.; Yost, R.A. Biomed. Environ. Mass Spectrom., 1988, 15, 193. Hopfgartner, G.; Husser, C.; Zell, M. J. Mass Spectrom., 2003, 38, 138. Shou, W.Z.; Magis, L.; Li, A.C.; Naidong, W.; Bryant, M.S. J. Mass Spectrom., 2005, 40, 1347. Hopfgartner, G.; Varesio, E.; Tschappat, V.; Grivet, C.; Bourgogne, E.; Leuthold L.A. J. Mass Spectrom., 2004, 39, 845. Xia, Y.Q.; Miller, J.D.; Bakhtiar, R.; Franklin, R.B.; Liu, D.Q. Rapid Commun. Mass Spectrom., 2003, 17, 1137. Li, A.C.; Gohdes, M.A.; Shou, W.Z. Rapid Commun. Mass Spectrom., 2007, 21, 1421. Li, A.C.; Alton, D.; Bryant, M.S.; Shou, W.Z. Rapid Commun. Mass Spectrom., 2005, 19, 1943.
20 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160]
Yunsheng Hsieh
Hager, J.W.; Yves, Le Blanc, J.C. Rapid Commun. Mass Spectrom., 2003, 17, 1056. Zhang, M.Y.; Pace N.; Kerns E.H.; Kleintop, T.; Kagan, N.; Sakuma, T. J. Mass Spectrom., 2005, 40, 1017. Zheng, J.; Ma, L.; Xin, B.; Olah, T.; Humphreys, W.G.; Zhu, M. Chem. Res. Toxicol., 2007, 20, 757. Hakala, K.S.; Kostiainen, R.; Ketola, R.A. Rapid Commun. Mass Spectrom., 2006, 20, 2081. Peterman, S.M.; Duczak, N. Jr.; Kalgutkar, A.S.; Lame, M.E.; Soglia, J.R. J. Am. Soc. Mass Spectrom., 2006, 17, 363. Hsieh, Y.; Favreau, L.; Schwerdt, J.; Cheng, K.C. J. Pharm. Biomed. Anal., 2006, 40, 799. Hsieh, Y.; Favreau, L.; Cheng, K.C.; Chen, J. Rapid Commun. Mass Spectrom., 2005, 19, 3037. Heinemann, U.; Blaschke, G.; Knebel, N. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2003, 793, 389. Kindt, E.K.; Kurzyniec, S.; Wang, S.C.; Kilby, G.; Rossi, D.T. J. Pharm. Biomed. Anal., 2003, 31, 893. Iwata, Y.T.; Kanamori, T.; Ohmae, Y.; Tsujikawa, K.; Inoue, H.; Kishi, T. Electrophoresis, 2003, 24, 1770. Martens-Lobenhoffer, J.; Reiche, I.; Troger, U.; Monkemuller, K.; Malfertheiner, P.; Bode-Boger, S.M. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 857, 301. Wickremsinhe, E.R.; Tian, Y.; Ruterbories, K.J.; Verburg, E.M.; Weerakkody, G.J.; Kurihara, A.; Farid, N.A. Drug Metab. Dispos., 2007, 35, 917. Nobilis, M.; Jira, T.; Lisa, M.; Holcapek, M.; Szotakova, B.; Lamka, J.; Skalova, L. J. Chromatogr. A, 2007, 1149, 112. Nikolai, L.N.; McClure, E.L.; Macleod, S.L.; Wong, C.S. J. Chromatogr. A, 2006, 1131, 103. Lausecker, B.; Fischer, G. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2006, 835, 40. Wang, Z.; Li, S.; Jonca, M.; Lambros, T.; Ferguson, S.; Goodnow, R.; Ho, C. Biomed. Chromatogr. 2006, 20, 1206. Chen, J.; Hsieh, Y.; Cook, J.; Morrison, R.; Korfmacher, W.A. Anal. Chem., 2006, 78, 1212. Coe, R.A.; Rathe, J.O.; Lee, J.W. J. Pharm. Biomed. Anal., 2006, 42, 573. Hoke, S.H.; Pinkston, J.D.; Bailey, R.E.; Tanguay, S.L.; Eichhold, T.H. Anal. Chem., 2000, 72, 4235. Hsieh, Y.; Li, F.; Duncan, C.J. Anal. Chem., 2007, 79, 3856. Pinkston, J.D.; Wen, D.; Morand, K.L.; Tirey, D.A.; Stanton, D.T. Anal. Chem., 2006, 78, 7467. Zhao, Y.; Woo, G.; Thomas, S.; Semin, D.; Sandra, P. J. Chromatogr. A, 2003, 1003, 157. Buchholz, L.; Cai, C.H.; Andress, L.; Cleton, A.; Brodfuehrer, J.; Cohen, L. Eur. J. Pharm. Sci., 2002, 15, 209. Cheng, Y.; Ho, E.; Subramanyam, B.; Tseng, J.L. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2004, 809, 67. Hage, D.S. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2002, 768, 3. Beaudry, F.; Coutu, M.; Brown, N.K. Biomed. Chromatogr., 1999, 13, 401. Tiller, P.R.; Mutton, I.M.; Lane, S.J.; Bevan, C.D. Rapid Commun. Mass Spectrom., 1995, 9, 261. Johnson, B.M.; Nikolic, D.; van Breemen, R.B. Mass Spectrom. Rev., 2002, 21, 76. van Breemen, R.B.; Nikolic, D.; Bolton, J.L. Drug Metab. Dispos., 1998, 26, 85. van Breemen, R.B.; Huang, C.R.; Nikolic, D.; Woodbury, C.P.; Zhao, Y.Z.; Venton D.L. Anal. Chem., 1997, 69, 2159. Geun Shin, Y.; Bolton, J.L.; van Breemen, R.B. Comb. Chem. High Throughput Screen., 2002, 5, 59. Carducci, C.; Santagata, S.; Leuzzi, V.; Carducci, C.; Artiola, C.; Giovanniello, T.; Battini, R.; Antonozzi, I. Clin. Chim. Acta, 2006, 364, 180. Weinmann, W.; Svoboda, M. J. Anal. Toxicol., 1998, 22, 319. Morand, K.L.; Burt, T.M.; Regg, B.T.; Chester, T.L. Anal. Chem., 2001, 73, 247. Trunzer, M.; Graf, D.; Kiffe, M. Rapid Commun. Mass Spectrom., 2007, 21, 937. Wickremsinhe, E.R.; Singh, G.; Ackermann, B.L.; Gillespie, T.A, Chaudhary, A.K. Curr. Drug Metab., 2006, 7, 913. Wickremsinhe, E.R.; Ackermann, B.L.; Chaudhary, A.K. Rapid Commun. Mass Spectrom., 2005, 19, 47. Kapron, J.T.; Pace, E.; Van Pelt, C.K.; Henion, J. Rapid Commun. Mass Spectrom., 2003, 17, 2019. Dethy, J.M.; Ackermann, B.L.; Delatour, C.; Henion, J.D. Schultz, G.A. Anal. Chem., 2003, 75, 805. Hop, C.E. Curr. Drug Metab., 2006, 7, 557.
Present and Future Mass Spectrometry-Based [161] [162] [163] [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [193] [194] [195]
[196] [197]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 21
Corkery, L.J.; Pang, H.; Schneider, B.B.; Covey, T.R.; Siu, K.W. J. Am. Soc. Mass Spectrom., 2005, 16, 363. Qi, L.; Danielson, N.D. J. Pharm. Biomed. Anal., 2005, 37, 225. Wu, J.; Hughes, C.S.; Picard, P.; Letarte, S.; Gaudreault, M.; Lévesque, J.F.; Nicoll-Griffith, D.A.; Bateman, K.P. Anal. Chem., 2007, 79, 4657. Gobey, J.; Cole, M.; Janiszewski, J.; Covey, T.; Chau, T.; Kovarik, P.; Corr, J. Anal. Chem., 2005, 77, 5643. Sleno, L.; Volmer, D.A. Rapid Commun. Mass Spectrom., 2005, 19, 1928. Hatsis, P.; Brombacher, S.; Corr, J.; Kovarik, P.; Volmer, D.A. Rapid Commun. Mass Spectrom., 2003, 17, 2303. Van Pelt, C.K.; Zhang, S.; Kapron, J.; Huang, X.; Henion, J.D. Am. Lab., 2003, 6, 14. Brown, K.; Dingley, K.H.; Turteltaub, K.W. Meth. Enzymol., 2005, 402, 423. Vogel, J.S. Biotechniques, 2005, 6, 25. Budzikiewicz, H.; Grigsby, R.D. Mass Spectrom. Rev., 2006, 25, 146. Garner, R.C. Curr. Drug Metab., 2000, 1, 205. Fifield, L.K. Rep. Prog. Phys., 1999, 62, 1223. Lappin, G.; Garner, R.C. Nat. Rev. Drug Discov., 2003, 2, 233. Papac, D.I.; Shahrokh, Z. Pharm. Res., 2001, 18, 131. Turteltaub, K.W.; Vogel, J.S. Curr. Pharm. Des., 2000, 6, 991. Lappin, G.; Rowland, M.; Garner, R.C. Expert Opin. Drug Metab. Toxicol., 2006, 2, 419. Sandhu, P.; Vogel, J.S.; Rose, M.J.; Ubick, E.A.; Brunner, J.E.; Wallace, M.A.; Adelsberger, J.K.; Baker, M.P.; Henderson, P.T.; Pearson, P.G.; Baillie, T.A. Drug Metab. Dispos., 2004, 32, 1254. Beumer, J.H.; Garner, R.C.; Cohen, M.B.; Galbraith, S.; Duncan, G.F.; Griffin, T.; Beijnen, J.H.; Schellens, J.H. Invest. New Drugs, 2007, 25, 327. Garner, R.C.; Goris, I.; Laenen, A.A.; Vanhoutte, E.; Meuldermans, W.; Gregory, S.; Garner, J.V.; Leong, D.; Whattam, M.; Calam, A.; Snel, C.A. Drug Metab. Dispos., 2002, 30, 823. Solon, E.G.; Balani, S.K.; Lee, F.W. Curr. Drug Metab., 2002, 3, 451. Hsieh, Y.; Chen, J.; Korfmacher, W.A. J. Pharmacol. Toxicol. Methods, 2007, 55, 193. Drexler, D.M.; Garrett, T.J.; Cantone, J.L.; Diters, R.W.; Mitroka, J.G.; Prieto Conaway, M.C.; Adams, S.P.; Yost, R.A.; Sanders, M. J. Pharmacol. Toxicol. Methods, 2007, 55, 279. Signor, L.; Varesio, E.; Staack, R.F.; Starke, V.; Richter, W.F.; Hopfgartner, G. J. Mass Spectrom., 2007, 42, 900. Reyzer, M.L.; Caprioli, R.M. Curr. Opin. Chem. Biol., 2007, 11, 29. Reyzer, M.L.; Hsieh, Y.; Ng, K.; Korfmacher, W.A.; Caprioli, R.M. J. Mass Spectrom., 2003, 38, 1081. Khatib-Shahidi, S.; Andersson, M.; Herman, J.L.; Gillespie, T.A.; Caprioli, R.M. Anal. Chem., 2006, 78, 6448. Hsieh, Y.; Casale, R.; Fukuda, E.; Chen, J.; Knemeyer, I.; Wingate, J.; Morrison, R.; Korfmacher, W. Rapid Commun. Mass Spectrom., 2006, 20, 965. Yao, M.; Zhu, M.; Sinz, M.W.; Zhang, H.; Humphreys, W.G.; Rodrigues, A.D.; Dai, R. J. Pharm. Biomed. Anal., 2007, 44, 211. Bu, H.Z.; Knuth, K.; Magis, L.; Teitelbaum, P. Rapid Commun. Mass Spectrom., 2000, 14, 1943. Benetton, S.A.; Fang, C.; Yang, Y.O.; Alok, R.; Year, M.; Lin, C.C.; Yeh, L.T. Drug Metab. Pharmacokinet., 2007, 22, 78. Wan, H.; Rehngren, M. J. Chromatogr. A, 2006, 1102, 125. Wan, H.; Ostlund, A.; Jonsson, S.; Lindberg, W. Rapid Commun. Mass Spectrom., 2005, 19, 1603. Lappin, G.; Garner, R.C. Anal. Bioanal. Chem., 2004, 378, 356. Wang, M.; Miksa, I.R. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 856, 318. Drexler, D.M.; Belcastro, J.V.; Dickinson, K.E.; Edinger, K.J.; Hnatyshyn, S.Y.; Josephs, J.L.; Langish, R.A.; McNaney, C.A.; Santone, K.S.; Shipkova, P.A.; Tymiak, A.A.; Zvyaga, T.A.; Sanders, M. Assay Drug Dev. Technol., 2007, 5, 247. Shou, W.Z.; Magis, L.; Li, A.C.; Naidong, W.; Bryant, M.S. J. Mass Spectrom., 2005, 40, 1347. Liu, D.Q.; Hop, C.E. J. Pharm. Biomed. Anal., 2005, 37, 1.
22 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [198] [199] [200] [201] [202] [203]
Yunsheng Hsieh
Korfmacher, W.A.; Cox, K.A.; Ng, K.J.; Veals, J.; Hsieh, Y.; Wainhaus, S.; Broske, L.; Prelusky, D.; Nomeir, A.; White, R.E. Rapid Commun. Mass Spectrom., 2001, 15, 335. Chang, Y.W.; Chen, W.C.; Lin, K.T.; Chang, L.; Yao, H.T.; Hsieh, H.P.; Lan, S.J.; Chen, C.T.; Chao, Y.S.; Yeh, T.K. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 2007, 846, 162. De Nardi, C.; Bonelli, F. Rapid Commun, Mass Spectrom., 2006, 20, 2709. Chen, J.; Yang, L.; Kapron, J.T.; Ma, L.; Pace, E.; Van Pelt, C.K.; Rudewicz, P.J. J. Chromatogr. B Analyt, Technol. Biomed, Life Sci., 2004, 809, 205. Chu, I.; Liu, F.; Soares, A.; Kumari, P.; Nomeir, A.A. Rapid Commun. Mass Spectrom., 2002, 16, 1501. Chen, Y.F.; Chang, C.H.; Wang, S.C.; Tsai, T.H. Biomed. Chromatogr., 2005, 19, 402.
Frontiers in Drug Design & Discovery, 2009, 4, 23-42
23
Glutathione Transferases in Drug Discovery and Development: Towards Safer and Efficacious Drugs Katholiki Skopelitou, Dimitris Platis, Irene Axarli and Nikolaos E. Labrou* Laboratory of Enzyme Technology, Department of Agricultural Biotechnology, Agricultural University of Athens, 75 Iera Odos Street, GR-11855-Athens, Greece Abstract: Glutathione transferases (GSTs) are enzymes involved in cellular detoxification by catalysing the nucleophilic attack of glutathione (GSH) on the electrophilic centre of a number of electrophilic compounds of both endogenous and exogenous origins. This conjugation reaction usually makes the electrophilic substrates more water soluble and, thereby, facilitates their excretion from the body. Determination of metabolic properties of a new chemical entity (NCE) is one of the most important steps during the drug discovery and development process. Nowadays, in vitro methods are used for early estimation and prediction of in vivo metabolism of NCEs. In this review detailed descriptions are given of several biotransformation reactions catalyzed by GSTs that can be used at very early phases of drug development, thereby enabling unsuitable candidates to be eliminated from consideration much earlier in the drug discovery process. Knowledge of the structure-function relationships in classes of compounds that are substrates for GSTs enables the design of molecules that can be stable, or labile which has potential applications in drug and prodrug design.
Key Words: Detoxification, enzyme-activated prodrug, glutathione transferase. 1. INTRODUCTION GSTs are a family of detoxification enzymes that catalyse the conjugation of GSH to a wide variety of endogenous and exogenous electrophilic compounds. The GST superfamily can be subdivided into a number of classes on the basis of their amino acid sequence [1]. Within mammals, the following classes have been defined: alpha, mu, pi, sigma, theta, zeta, kappa and omega [2]. Analysis of the GST gene family in the Human Genome Organization database showed 21 putatively functional genes [3]. Upon closer examination, however, GST-kappa 1 (GSTK1), prostaglandin E synthase (PTGES) and three microsomal GSTs (MGST1, MGST2, MGST3) were determined as encoding membrane-bound enzymes having GST-like activity, but these genes are not evolutionarily related to the GST gene family [3]. Therefore, the complete GST gene family comprises 16 genes in six subfamilies: alpha (GSTA), mu (GSTM), omega (GSTO), pi *Corresponding Author: Tel: +30 (210) 5294308; Fax: +30 (210) 5294308; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
24 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
(GSTP), theta (GSTT) and zeta (GSTZ). In other organisms additional soluble GST classes have been reported. For example, in insects: delta and epsilon [4]; in plants: phi, tau, lambda, dehydroascorbate reductase [5]; and in bacteria: beta [6] and chi [7]. In general the catalytic efficiency of GSTs towards xenobiotics is relatively low [17]. As in the case with other xenobiotic metabolizing enzymes (e.g., cytochrome P450, glucuronosyl transferases, etc.) low catalytic efficiency appears to be a trade off with regard to broad substrate specificity [8,9]. On the other hand, constitutive levels of GSTs are high (3-10% of total cytosolic protein) and therefore, although catalytic efficiency is relatively low, the overall capacity due to the high constitutive expression is important. The biological function of the GSTs is far from established [8,9]. Various electrophilic xenobiotics can be substrates for GSTs. Electrophilic centres for GSH conjugation are found in areneoxides, aliphatic and arylic halides, in carbonyls, organonitro-esters and organic thiocyanates. Industrial substrates for GST are haloalkanes, chlorobenzenes, thiocarbamates, diphenylethers, triazines, chloracetanilide. Acrolein, propenals, lipid hydroperoxides, chlorambucil and fosfomycin are additional substrates. Identification of the GST-mediated pathway for drug cleavage has been useful for elucidating the mechanism of metabolic biotransformation of compounds that have been brought forward for clinical studies [10,11]. Knowledge of the structure-function relationships in classes of compounds that are cleavage by GSTs enables design of molecules an be stable, or labile which has potential applications in drug and prodrug design. In the next sections of this chapter the different reactive groups and reaction types operating under catalysis by GSTs will be discussed. 2. STRUCTURE AND CATALYTIC MECHANISM OF CYTOSOLIC GSTs 2.1. Structure of Cytosolic GSTs Cytosolic GSTs are dimeric proteins as shown in Fig. (1). Each subunit is composed 200-250 amino acid residues with typical molecular masses ranging from 20–28 kDa. Each GST subunit adopts a canonical GST fold of seven to nine -helices and four sheets to produce two distinct domains, the N- and C-terminal domains [1,2,8,12]. The interface between the two subunits can be hydrophobic or hydrophilic, and interactions between residues in both subunits are essential for dimer stability. For example, the subunit interface for the alpha class GSTs is that of a ‘lock-and-key’ type joint with the Phe52 serving as the ‘key’ and the hydrophobic ‘lock’ residues residing between 4 and 5 helices of C-terminal [8]. All alpha, mu, phi, zeta, tau and pi class GSTs have such a ‘lock-and-key’ style interface. On the other hand, in theta class GSTs the main subunit interface lacks a ‘lock-and-key’ motif. Structurally it seems that the theta class GSTs are missing the loop feature that holds the Phe residue for interaction with the 4 and 5 helices and the hydrophobic pockets within the helices themselves [8]. In this case subunit interface rely mainly on electrostatic and hydrogen bond interactions. Incompatibility in interfacial residues prevents the formation of heterodimers between two GST subunits from different classes. However, within a class, the formation of heterodimers can expand the range of functional proteins formed. Each GST is known to contain a G-site capable of binding the GSH substrate and an H-site that has xenobiotic compound binding capabilities, (Fig. 1A). he G-site share a highly conserved amino acid sequence. The G-site is mainly composed of amino acids in
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 25
A
B
Fig. (1). A: Subunit structure of human GSTA1-1 in complex with an ethacrynic acid-GSH conjugate (PDB code 1GSE). The G- and H-site and the N- and C-terminal are labeled. B: Ribbon diagram of the dimeric GSTA1-1 structure.
the N-terminal, including the active site residue that interacts with and activates the sulfhydryl group of GSH to generate the catalytically active thiolate anion [2,8,12,13]. In the alpha, mu, pi, sigma the active site residue is a Tyr. In the delta, epsilon, theta, and zeta GSTs, the active site residue is a Ser, and in omega class GSTs, it is a Cys [14]. Several
26 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
crystal structures have shown that this active site residue is in hydrogen bond with the sulfur atom of GSH and is located at a position that would allow it to stabilize the thiolate anion of GSH and enhance its nucleophilicity. In addition, this residue may also contribute to the correct orientation of the sulfhydryl group of GSH in catalysis. Ser 11 and Tyr 9, the active site residues of rat GSTT2-2 [15] and rat GSTA1-1 [16] respectively, have been reported to play important roles in efficient product release and in controlling the C-terminal dynamics. The xenobiotic hydrophobic H-site is mainly found in the C-terminal. The H-site is hydrophobic but much less specific for substrate types allowing numerous substrates to bind [8]. This is the reason why initial attempts to classify GSTs according to substrate types (e.g. aryl-, alkyl-transferases) failed. Interclass amino acid sequence identity is rarely greater than 35% in the H-site region. Besides catalytic binding sites, GSTs have frequently been shown to possess noncatalytic binding sites. Numerous hydrophobic compounds have been identified to bind to these pockets such as bilirubin, heme, steroids, hormones and bile salts. This function has led to the name ‘ligandins’ for several of the enzymes [17]. The precise functions of ligandin GST binding to non-substrate ligands remain unclear. One possibility is that binding of non-substrate ligands to GSTs prevents oxidation of the molecules in vivo [17]. Another possibility is that GSTs prevent cellular damage from cytotoxic and genotoxic compounds that can oxidize protein and intercalate into DNA [18]. The third possibility is that GSTs facilitate delivery of the ligands to specific receptors or cellular compartments [17,18]. Some general structural features, (Fig. 2), of the main mammalian GST classes (alpha, mu, pi, theta, zeta and omega) will be discussed in the next paragraphs. Compared to the pi and mu GSTs the C-terminal of the alpha class GSTs is longer by some 4 to 8 amino acid residues [19]. The longer alpha C-terminal also forms an -helix (9), which comprises a portion of the smaller H-site [20]. This helix is thought to be important to dimer stabilization and affects both the GSH-binding rate and ionization state of the catalytically essential residue Tyr 9 [21]. Thus the alpha class C-terminal has a positive effect on catalytic activity. A special recognizable feature of the mu class GSTs is the socalled mu loop, which is the result of an insertion in N-terminal domain. Another structural attribute, which is recognizable in both mu and pi classes of GST, is the C-terminal wall [21]. On average, the C-terminal end is 4 to 8 amino acid residues shorter than their alpha counterpart and forms a wall that results in a partially blocked access to the xenobiotic binding site [22]. In addition, the pi and mu classes share a larger H-site that is more accessible to solvent entry than the alpha class GSTs. The theta class enzymes are quite distinct from alpha, mu, or pi – GST classes. The G-site of the theta class GSTs is, in general, much deeper than that of alpha, mu, and pi GST classes [8,12]. Several features set omega GSTO1-1 apart from the other members of the GST superfamily. Its N-terminal region has a unique extension of approximately 19 residues when compared with other cytosolic GSTs. This contains a proline-rich segment that in conjunction with the C-terminus forms a distinct structural unit with an unknown function. Unlike other mammalian GSTs, GSTO1-1 appears to have an active site cysteine that can form a disulfide bond with GSH. There are many structural features of the zeta class enzymes found in other GST classes, although it does not appear that GSTZ resembles more closely any one class in
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 27
particular. GSTZ has a truncated C-terminus and lacks the typical V-shaped dimer interface and hence resembles the theta class in this regard. The large loop preceding helix 2 is reminiscent of the mu loop previously seen in the mu class structure.
Fig. (2). The common chain fold of the GST superfamily. Protein data bank codes for the structures used are: alpha, 1gse; theta, 1ljr; mu, 1hna; pi, 1glp; omega, 1eem and zeta, 1fw1. Images in this figure were drawn using the program PyMol.
2.2. Kinetic Mechanism Conjugation reactions with GSH have been reported for a vast number of compounds and the kinetic mechanism has been clarified [22-27]. In general, kinetic mechanism of the GST-catalyzed conjugation reaction is very complex and class dependent. For example, several catalytic mechanisms, including random, ping-pong, and sequential, have been proposed [13,22-25], but random binding order of substrates seems to prevail. For instance, the maize GST I-catalyzed conjugation reaction between GSH and CDNB follows a rapid equilibrium random sequential Bi Bi kinetic mechanism [13], whereas a steady state sequential rapid Bi Bi mechanism was proposed for octopus GST, rat GSTs M1-1, M1-2, and A3-3 [26,27]. From a physiological point of view GSH binding should occur first because of the availability of GSH in millimolar concentrations in the cells. This value is about three magnitudes larger than the dissociation constant between GSH and GSTs [28]. 3. CATALYTIC FUNCTION The different reaction types operating under catalysis by GSTs are:
28 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
3.1. Nucleophilic Displacement of an Alkyl or Aryl Halogen or a Nitro-Group Nucleophilic displacement of an alkyl or aryl halogen or a nitro-group, (Fig. 3), seems to be the most frequently observed reactions. Halogens or nitrogroups of these molecules are soft electrophiles and react readily with the GSH. In fact, the standard enzyme assays for GST activity use 1-chloro-2,4-dinitrobenzene or l,2-dinitro-4chlorobenzene as substrates. These are nucleophilic aromatic substitution reactions that occurs via an addition-elimination sequence involving a short-lived -complex intermediate [8,9]. The reaction of GSH with nitroalkanes involves attack of an electrophilic carbon atom leading to the formation of a thioester and to the release of nitrite as shown in Fig. (3). A Cl
SG +
+ GSH O2N
NO2
O2N
HCl
NO2
B Cl
SG
Cl
NO2
GS NO2
NO-2
NO2 CDNB
NO2
-Cl -
GS -
NO2
-complex
C RCH2NO2 + GSH
RCH2SG + HNO2
Fig. (3). A: The reaction of 1-chloro-2,4-dinitrobenzene with GSH catalyzed by GSTs. B: The nucleophilic aromatic substitution reaction occurs via an addition-elimination sequence involving a short-lived -complex intermediate. C: The reaction of GSH with nitroalkanes catalyzed by GSTs.
Several GSTs catalyze the conjugation of the GSH with alkyl halides and related compounds [29]. For example, conjugation of the drug chlorobucil is an example for this type of reaction. Examples of the GSH conjugations catalyzed by GSTs are presented in Fig. (4). GSH conjugations of alkyl halides can be detoxication or bioactivation reactions. In the later case the substrate is activated leading to reactions with other nucleophiles, particularly DNA and proteins. The theta class GST enzymes are the most active [30], towards alkyl halides although other mammalian GSTs also have some activities. 3.2. Addition of GSH to Epoxides and Arene Oxides GSH transferases catalyze the addition of GSH to epoxide and arene oxide substrates, as illustrated (Fig. 5) [8]. Carcinogenic intermediates have been identified as both
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 29
bay- and fjord-region diol epoxides (DEs) [31,32]. GSTA1-1, an alpha class enzyme, has efficient catalytic activity towards stereoisomers of a series of bay region diol epoxides fjord region diol epoxides such as (±)-anti-B[]PDE, (±)-syn-B[]PDE and (±)-antiB[c]PhDE [33,34]. It is important to mention that the chemically more reactive syn diastereomers are better substrates for GSTA1-1 and this is propably due to differences in the lipophilicity of the compound. The activity and enantioselectivity of human GSTs towards diol epoxides depend on the absolute configuration and the geometry of the aromatic structure. Human mu class enzymes catalyse, in most cases, the conjugation of the stereoisomeric bay-region DEs, in particular, the syn-diastereomers [35]. OH GSH + RX
GS-R (+HX)
SG H2O
X GSH +
X
_
HX
X SG
_ -
X
+ SG
SG
GSH
SG Nucl
Nucl SG
Fig. (4). The conjugation of GSH with alkyl halides catalyzed by GSTs.
In contrast, human pi class enzyme, showed, in most cases, a noticeable activity towards bay-region anti-DEs and a high preference for conjugation of enantiomers with (R,S,S,R)-configuration [35-37]. Whereas the alpha class has no catalytic activity towards the optical isomers of benzo[c]phenanthrene DE B[c]PhDE, both the mu and the pi class, have a lower catalytic activity using as substrate these metabolites than the other bay-region DEs [35]. One relatively large, bulky substrate for certain GSTs is the epoxide of aflatoxin B1 (AFB), (Fig. 5C) [38]. AFB, one of a group of related mycotoxins produced by the common fungal mold Aspergillus flavus, is a well documented rat and human carcinogen [15,16]. AFB is activated to the highly reactive aflatoxin 8,9-exo-epoxide by certain cytochromes P450. AFBO then serves as a substrate for some, but not all, GST isoenzymes [39]. 3.3. Addition of GSH to Carbon-Carbon-Double Bonds Addition of the thiolate to carbon-carbon-double bonds is a special type of reactions on compounds with reactive carbon-carbon double bonds neighboured by an electronwithdrawing group [40]. The conjugation on these bonds is a so-called Michael addition reaction and leads to a labile conjugate that may be sensitive to pH changes. Other similar substrates include alkenals [4] and particularly the (E)-4-hydroxyalk-2-enals (4hydroxy-alkenals), illustrated in Fig. (6). 4-Hydroxy-2,3-trans-nonenal (4-HNE), a racemic lipid peroxidation product, is the most abundant hydroxyl-alkenal formed during the peroxidative breakdown of n-6 polyunsaturated fatty acid residues of membrane lipids, such as linoleate and arachidonate [41,42] and also the most cytotoxic and genotoxic product, involved in a various cellular pathologies [42,43]. GSTs are involved in the predominant mechanism for protection against 4-HNE toxicity in mammalian liver [44,45], however, despite the fact that hGSTA1-1 is the major human liver GST isoform
30 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
A GSH + SG
O
OH
OH phenanthrene 9,10 - oxide
SG
(9S, 10S) - GSPhen
(9R, 10R) - GSPhen
B Bay-region PAH Diol Epoxides
Diol Epoxides O S S HO
OH
CH3
R
HO R
S R S R
CH3 H3C
OH chrysene O
HO
RS R S
O
CH3 O
CH3
S R CH3
HO
OH
CH3
R S
CH3
OH benzo[a]pyrene
O
C O
dibenzo[a,h]anthracene
O O
HO O Glutathione
S
Fig. (5). A: GSTs catalyze the addition of GSH to epoxide and arene oxide substrates, such as phenanthrene 9,10-oxide. B: Structures of bay-region polycyclic aromatic hydrocarbons (PAHs). The stereochemistry of the syn- and anti-diol epoxides are shown to the right and left, respectively. The prefix syn indicates that the oxirane ring and the benzylic hydroxyl group are located on the same face of the molecule, while anti indicates that these groups are located on opposite faces. C: The structure of aflatoxin B1-GSH conjugate.
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 31
O H2C G-SH +
HC
O
CH H2C
CH
OH H2C
CH G
OH
S
O
CH CH
CH R
4-hydroxy-alkenal
G
CH
S
CH R
CH
R
OH
cyclic hemimercaptal
glutahione conjugate
Fig. (6). Reaction between GSH and 4-hydroxy-alkenals catalyzed by GSTs. The primary product is further transformed to a cyclic derivative.
it is relatively inefficient at conjugating 4-HNE [46,47] as well as other alkenal substrates. GSTA4-4 comprises the basic detoxification pathway for the removal of 4-HNE in the human liver mitochondria [48-50], but its contribution in other tissues is unknown due to its relatively low expression in human tissues [51]. 3.4. Addition of GSH to Organic Isothiocyanates and Thiocyanates Organic isothiocyanates, which are abundant in edible plants, undergo conjugation with GSH enzymatic as well as non-enzymatic to form dithiocarbamates [52,53], as presented in Fig. (7A). These dithiocarbamates are degraded in vivo to the corresponding S(N-acetyl)cysteinyl conjugates (mercapturates), which undergo rapid renal excretion. A number of isothiocyanates have also been shown to be inducers of GSTs in animal tissues and to display pronounced anticarcinogenic activity. In general, GSTM1-1 appears to be the most efficient catalyst towards isothiocyanates, closely followed by GST P1-1, whereas GSTAl-1 is less active by one and GSTM4-4 by even two orders of magnitude [53]. All four enzymes exhibit higher catalytic rate with increasing aliphatic chain length, and, of all the aliphatic substrates with linear carbon chains tested, hexyl-NCS gives the highest rates for all of them. The aromatic substrates, benzyl-NCS and phenethyl-NCS, tend to be the best ones. Moreover, GSTM4-4 has a preference for allyl-NCS among the aliphatic compounds. GSTs also catalyze the attack of the GSH thiolate ion on the electrophilic sulfur atom of several organic thiocyanates, resulting in the formation of an asymmetric glutathionyl disulfide and cyanide as shown in Fig. (7B). A R
C N
B
S
S
RSCN + GSH
+ GSH
HN R
C S
G
RSSG + HCN
Fig. (7). A: Organic isothiocyanates undergo enymatic conjugation with GSH to form dithiocarbamates. B: GSTs catalyze the attack of the GSH on the electrophilic sulfur atom of organic thiocyanates.
32 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
3.5. Sulfonamidase Activity The mechanism of sulfonamide cleavage by GST has been investigated using as a model substrate the PNU-109112, a potent HIV-1 protease inhibitor [10,11]. The mechanism of reaction, shown in Fig. (8), could involve initial GS- attack on either the sulfonyl S atom or the adjacent carbon atom in the cyanopyridine ring. The net result is either direct or indirect sulfonamide N-S bond cleavage. C
N
O RNH2
N
+
C
S
S O
G
GSH O S
H N
N C
O
R RNH2
+
SO2
+
G
S
Fig. (8). GST/GSH-mediated sulfonamide leavage of PNU-109112.
The GST activity toward sulfonamide bond cleavage has been characterized with respect to the molecular mechanism of the enzymatic reaction and substrate structureactivity relationships [10,11]. For example, studies of GST-mediated cleavage of PNU109112 and a wide variety of other sulfonamides reveal two striking structural features common to sulfonamide substrates [10,11]. Groups capable of withdrawing electron density from the carbon atom to the sulfonyl group are an absolute requirement. Such groups positioned ortho and/or para on aromatic or heteroaromatic rings were shown to activate the corresponding sulfonamides. The electrophilic substructure of the sulfonyl group is solely responsible for activation of the sulfonamide bond toward cleavage. On the other hand, the amine portion has little or no impact on the cleavability of sulfonamide substrates [10,11]. Recombinant forms of GSTA1–1, GSTM1–1, and GSTP1–1 were tested for sulfonamidase activity. Activity was observed for all three isoenzymes; the M1-1 isoenzyme was 5.4 and 16 times more active than the A1-1 and the P1-1 forms, respectively [10,11]. 3.6. Peroxidase Activity The release of oxygen radicals, such as O2-, H2O2, singlet oxygen (1O2) and hydroxyl radical (.OH) during oxidative stress conditions can initiate an autocatalytic chain of lipid peroxidation which can breach membrane integrity [54,55] or even cause DNA damage [56]. In fact, it has been reported that peroxidation of membrane lipids in cells and tissues is known to produce various aldehydic compounds [41,42], seriously suspected to cause a variety of chronic diseases, including atherosclerosis [57], Alzheimer's disease [58], cataractogenesis [59], Parkinson's disease [60] and cancer [42]. Glutathione peroxidase (GPx, EC.1.11.1.9) is a selenoenzyme that functions as an antioxidant by catalyzing the reduction of H2O2, lipid hydroperoxides and other organic
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 33
peroxides with GSH. The selenium atom in the enzyme catalytic site undergoes a redox cycle involving the selenolate anion as the active form which reduces hydrogen peroxides and organic peroxides. The selenolate, which is oxidized to selenenic acid (EnzSeOH), reacts with reduced GSH to form a selenosulfide adduct (Enz-SeSG). Finally, the nucleophilic attack of a second GSH to Enz-SeSG regenerates the active form of the enzyme by attacking the selenosulfide to form oxidized GSH (GSSG), as illustrated in Fig. (9) [61,62]. Regeneration of the reduced enzyme is accomplished via a two-step process involving reduced GSH as reductant. In the present of high concentrations of hydroperoxide, selenenic acid (Enz-SeOH) can be futher oxidized to a seleninic acid (Enz-SeO2H) [63].
Fig. (9). Scheme of the mechanism of the peroxidase activity.
Several GSTs were reported to display GPx activity towards organic hydroperoxide. The peroxidase activity associated with the GSTs is referred to as NonSe-GPx activity, which represents one of the important antioxidant mechanisms that exist in cells for protection against hydroperoxides [64,65]. In humans, the NonSe-GPx activity of GSTs towards lipid hydroperoxides is predominantly associated with alpha class [66, 67]. In addition, NonSe-GPx activity is a common characteristic of mammalian theta class and zeta class GSTs. GSTs belonging to the alpha class have been implicated in the detoxification of the produced peroxides and shown to exhibit high peroxidase activity towards these compounds [46,68]. In this case there is a nucleophilic attack of the GSH thiolate on an oxygen atom of the peroxy group and the unstable sulfenic acid derivative is subsequently reduced by a second GSH molecule as presented in Fig. (9). In the human liver, this Se-independent glutathione peroxidase activity is expressed primarily by isoenzymes hGSTA1-1 and hGSTA2-2 [56,69,70], and although their activity is much lower that the Se-dependent human GPx-I [71], the high amount of GSTs in the liver compared to other GPx activities [72] compensates for the low activity. Both hGSTA1-1 and hGSTA2-2 exhibit high peroxidase activity towards fatty acid hydroperoxides, phospholipid hydroperoxides, and cumene hydroperoxide [73]. Similar phospholipid hydroperoxide activities have been detected in
34 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
the human lung [74]. In mammalian cells, besides cytosolic GSTs, microsomal GSTs also express peroxidase activity [75,76]. 3.7. Transthiolation GSTs are induced in conditions of oxidative stress where it is known that sulfhydryl groups of cellular proteins can form S-thiol adducts (protein-thiol mixed disulfides, protein-S-S-glutathione (PSSG) and protein-S-S-cysteine (PSSC)) [77,78] with cysteine and GSH, through a process called S-thiolation. This result to deactivation, changes in conformation of even aggregation of the proteins involved. S-glutathionylation is a reversible post-translational modification with critical roles in sulfhydryl homeostasis acting as a protection mechanism of the cysteine residues from irreversible oxidation during periods of oxidative stress or a transport/storage form of cysteines [79]. Transthiolation reactions are not uncommon to proceed in cells without the aid of enzymes, however, cells do possess specific enzymes that catalyse the reduction of protein disulphides. In many cases this role has been attributed and accomplished by glutaredoxin (Grx). Modification of cysteine residues with GSH has also been reported for many enzymes, such as carbonic anhydrase [80] and fatty acid synthetase [81]. The transthiolation reaction consists of two steps. The first step involves the formation of a mixed disulfide with GSH which is subsequently reduced further by GSH with the production of GSH disulfide followed by complete reduction of the previously conjugated thiol residues. It is interesting that this reaction cannot distinguish between catalysis of the glutathionylation step (R-SSG formation, characteristic of a GST-mediated conjugation) and the deglutathionylation step (R-SSG reduction, characteristic of a deglutathionylase-like GRx). GST-dependent thioltransferase activity is generally very uncommon among GSTs but has been reported in mu, omega and pi classes. For example, thioltransferase activity has been observed in mu class GST from bovine lens, in three omega class GSTs in Saccharomyces cerevisae [82,83], and in human omega class GST (GSTO1-1) [14]. In addition, a pi class member has been shown to mediate the reactivation 1CysPRx, a protein that protects cells against membrane oxidation through GSH-dependent reduction of phospholipid hydroperoxides to corresponding alcohols. 3.8. Isomerase Activity It is well known that GSTs catalyze several types of isomerization reactions, with the conversion of cis double bonds to the trans configuration being the most well-studied [84-86], as in the case of isomerization of diethyl maleate to diethyl fumarate. Interestingly, GST-dependent isomerization reaction can proceed in a GSH-independent manner, as in the case of the cis-trans conversion of retinoic acid catalyzed by hGSTP1-1 [87]. The available information suggests that the prementioned cis-trans isomerization reactions are thermodynamic driven by the stability of the trans isomer, proceeding via the addition of GSH to the double bond, rotation around the resulting single bond, and elimination of GSH [86]. Positional isomerization of double bonds by GST-dependent reactions has also been been reported. The most well-characterized example of positional isomerization is the reaction in which endogenous 5-3-ketosteroids are converted to 4-3-ketosteroids [88,85,86]. For these reactions, the driving force is the stability of the conjugated double bonds in the product. In fact, GSTs appear to play an important role in the biosynthesis
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 35
of steroid hormones. All steroid hormones in humans are derived from cholesterol and in recent years it has been shown that GSTs are involved in the biosynthetic path of testosterone and progesterone [89,90]. Benson and Talalay [91] have initially discovered a GSH-dependent ketosteroid activity in rat liver, which was later identified as a GST [88]. The spontaneous conversion of 5-3-ketosteroid androstene-3,17-dione to the 4isomer product proceeds at a slow rate, which is markedly raised by the addition of GSH along with GSTA1-1, the first human GST discovered that exhibited double bind isomerase activity [92]. Another member of the alpha class, GSTA3-3 was found to exhibit approximately 20 fold higher catalytic isomerase activity towards 5-ketosteroids in human tissues [89,90] and is selectively expressed in gonads, placenta and the adrenal gland. GSTA3-3 efficiently catalyzes double-bond isomerizations of 5-androstene3,17-dione (5-AD) and of 5-pregnene-3,20-dione, intermediates in the biosynthesis of the steroid hormones progesterone and testosterone [8]. GST-catalyzed isomerization has also been observed during the biosynthesis of prostaglandins. Prostaglandin (PG) D synthase (PGDS) is the enzyme responsible for the formation of PGD2 and the J series of PGs [93]. In addition, prostaglandin H2 (PGH2) is obtained from prostaglandin D2 (PGD2) in a reaction catalyzed by a sigma class GST in vivo and alpha class GSTs in vitro [94]. The sigma class GST, isomerizes PGH2 to PGD2 selectively, whereas other GST isoenzymes catalyze the conversion of PGH2 nonselectively to produce PGD2, PGE2, and PGF2 [95]. A zeta class GST has also been identified as a maleylacetate isomerase [96], catalyzing the penultimate step in tyrosine degradation pathway, a pathway that has been associated with many disorders. [97,98]. This enzyme converts maleylacetoacetate into fumarylacetacetate, and maleylacetone into fumarylacetone. 3.9. Ester and Ether Hydrolysis Aromatic carboxylic acid esters and nitrate esters serve as substrates for GSTs as shown in Fig. (10). In the case of nitrate esters, the attack is on the electrophilic nitrogen. Experimental evidence supports the GSH sulfenyl nitrite intermediate, which is attacked nonenzymatically by a second molecule of thiol. GSTs also catalyze ‘reverse’ reactions, such as the hydrolysis of GSH thiol esters [99-101]. Reverse reactions are of interest as potential tumor-directed pro-drug activation strategies and as mechanisms for tissue redistribution of carboxylate containing drugs. Examples of ‘reverse’ GST reactions have been limited to retro-Michael additions [101], and hydrolysis of activated carbamate thiol esters and thiocarbamate thiol esters formed from GSH conjugation with isocyanates and isothiocyanates, respectively [53]. The reverse reactions are of therapeutic and toxicological importance, because the generation of GSH and an electrophile from GSH conjugates represents a potential pathway for delivery of ‘latent’ toxins or drugs to tissues remote from the initial conjugation reaction, whether release of the conjugate is mediated by GSTs or by nonenzymatic processes. The GSH thiol esters of ethacrynic acid (E-SG) (Fig. 11) and several nonsteroidal anti-inflammatory agents such as, sulindac, flurbiprofen, and diclofenac, have been tested as substrates with human GSTs [99]. Because ethacrynic acid produced by hydrolysis of E-SG contains a reactive --unsaturated ketone, it could form adducts with
36 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
protein nucleophiles, as is well established for Cys-47 of GSTP1-1. Adduction of Cys-47 by ethacrynic acid results in a slowly reversible inhibited complex. O
A
OH O
O +
+ GSH
C
GS
CH3 NO2
NO2
RCH2ONO2 + GSH
B C
RCH2OH + HNO2 + GSSG
RCH2OH + [GSNO2]
RCH2ONO2 + GSH [GSNO2] + GSH
GSSG + HNO2
Fig. (10). GSTs catalyze the attack of the GSH thiolate ion on the electrophilic carbon of carboxylic (A) and nitrate (B) esters. The later reaction takes place through a GSH sulfenyl nitrite intermediate (C), which is attacked nonenzymatically by a second molecule of thiol.
E-SG
EA
GS-EA + GSH
Cl Cl
O
Cl
O
Cl
SG O
O
O O
Cl Cl
OH
O O
OH
O
SG
Fig. (11). Reaction pathways of ethacrynic acid (EA) following its release from thiol ester (E-SG). The released EA may further react with GSH to form GS-EA.
Other example of S-ether hydrolysis involves the biotransformation of the prodrug azathioprine, to generate its active form 6-mercaptopurine as illustrated in Fig. (12) [102]. The mechanism of action of GSTs with azathioprine involves a nucleophilic attack of the deprotonated form of GSH on the electrophilic 50 carbon in the imidazole moiety of azathioprine. Thus, 5-mercaptopurine and a GSH–imidazole conjugate are released [103]. Among several human GSTs tested, the alpha class GSTA2-2 had the
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 37
highest specific activity, whereas a second alpha class enzyme GSTA1-1, as well as the mu class GSTM1-1, also demonstrated prominent catalytic activity. NO2
N
Electrophilic center N
S
H3C N
S
N
N
N H
Fig. (12). GST catalysis of S-ether hydrolysis of azathioprine, to generate its active form 6mercaptopurine.
3.10. GST-Activated Pro-Drugs Pro-drug development using the platform of GST has recently produced a number of lead compounds [104-106]. GSTs are frequently overexpressed in neoplastic tissues, with the GSTP1-1 isoform most commonly overexpressed in cancers resistant to drugs. Therefore GSTP1-1 proved to be an attractive molecular target for pro-drug activation. Two such agents that are in preclinical development will be discussed in the next section. High GST-expression levels in human cancers together with the knowledge that nitric oxide (NO) has therapeutic potential, provided the rationale for the design of the NO-releasing GST-activated pro-drug such as O2-arylated diazeniumdiolates derivatives (Fig. 13) [104,106]. The general mechanism of action of such prodrugs involves nucleophilic aromatic substitution by GSH, generating the nitric oxide-releasing diazeniumdiolate ion. Nitric oxide prodrugs of the ionic diazeniumdiolate class that spontaneously dissociate at definitive rates to form NO are useful as possible therapeutic agents in the treatment of several disease states, such as cancer. Due to the involvement of NO in diverse physiological processes, site-directed delivery of therapeutic nitric oxide is essential to avoid any undesirable side-effects. SG NR2
NR2 O
N+
O2N
NO2
N+ N
O GSH
O
NR2
N SG
N
O
O O2N
NO2
O2N
N+
H+ 2 NO
O
R2NH
NO2
Fig. (13). Structure of an NO-releasing GST-activated pro-drug, and its reaction with GSH.
38 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
Another prodrug, illustrated in Fig. (14), activated by GSTP1-1 is TLK286 [105]. This drug is activated through a -elimination reaction that cleaves the drug into a phosphorodiamidate (the eventual alkylating moiety) and a GSH analogue. Drug design ensured that this reaction was preferentially catalyzed by the Tyr 7 in the active site of GSTP 1-1.
Tyr 7 Cl Cl O
O~
O
P
Cl
N
O
H
Cl
S O H N
O H2N
N
COOH
N H
COOH
O Cl Cl CH2
O
H N
O H2N COOH
N
O
S O
P
+ COOH
Cl
N
O
Cl
N H O
Cl N
O
P O
Cl
N Cl
Fig. (14). Structure of TLK286 and its activation by human GSTP1-1. Tyr 7 is the catalytic residue.
3.11. The Negative Side of the GST-Detoxification System Despite the fact that detoxification is generally considered a significant positive aspect of GST function, in certain cases, chemically reactive metabolites produced by GST can have potentially harmful effects. A characteristic example is the case of the drug
GSTs in Drug Discovery and Development
Frontiers in Drug Design & Discovery, 2009, Vol. 4 39
acetaminophen (paracetamol) where the reactive metabolite N-acetyl-p-benzoquinoneimine is formed when the drug is ingested in high doses. In this case, the action of GST results in hepatic failure [107]. In fact, Henderson et al. were able to show that complete absence of pi class GSTs in mice was sufficient to alleviate any signs of hepatotoxicity and significantly improve GSH recovery rates [108]. 4. CONCLUSIONS The development process of new drugs is constantly being improved by defining new strategies and implementing new techniques. The human body contains a variety of enzymes that are involved in the metabolism of several chemicals that comprise today’s pharmaceuticals. One of the most important classes of metabolic enzymes is GSTs which are directly involved in the clearance of drugs from the body. During this process, drug metabolites are generated, some of which are biologically inactive or active. GSTs catalyze a wide range of reactions that can be described as accompanying the many varieties of catalysis in which the GSH thiolate anion participates as a nucleophile. The substrates are rather diverse and include important endogenous modulators, as well as xenobiotic chemicals and the oxidation products of the transformation of xenobiotics by other enzymes. In addition, GST-activated pro-drugs provide viable platform for the development of novel cancer drugs. Investigation of these biotransformation reactions in vitro can be very useful at very early phases of drug development, thereby enabling unsuitable candidates to be eliminated from consideration much earlier in the drug discovery process. ABBREVIATIONS AFB
=
Aflatoxin B1
CDNB
=
1-Chloro-2,4-dinitrobenzene
DE
=
Diol epoxide
EA
=
Ethacrynic acid
GPx
=
Glutathione peroxidase
Grx
=
Glutaredoxin
GSH
=
Glutathione
G-site
=
Glutathione binding site
GSSG
=
Oxidised GSH
GST
=
Glutathione transferase
hGST
=
Human glutathione transferase
4-HNE
=
4-Hydroxy-2,3-trans-nonenal
H-site
=
Hydrophobic binding site
NCE
=
New chemical entity
PG
=
Prostaglandin
PGDS
=
Prostaglandin D synthase
ROS
=
Reactive oxygen species
40 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Skopelitou et al.
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40]
Mannervik, B.; Danielson, U.H. CRC Crit. Rev. Biochem., 1988, 23, 283. Frova, C. Biomol. Eng., 2006, 23, 149. Nebert, D.W.; Vasiliou, V. Hum. Genomics, 2004, 1, 460. Alias, Z.; Clark, A.G. Proteomics, 2007, 7, 3618. Dixon, D.P.; Davis, B.G.; Edwards, R. J. Biol. Chem., 2002, 277, 30859. Allocati, N.; Favaloro, B.; Masulli, M.; Alexeyev, M.F.; Di Ilio, C. Biochem. J., 2003, 373, 305. Wiktelius, E.; Stenberg, G. Biochem. J., 2007, 406, 115. Armstrong, R.N. Chem. Res. Toxicol., 1997, 10, 2. Hayes, J.D.; Flanagan, J.U.; Jowsey, I.R. Annu. Rev. Pharmacol. Toxicol., 2005, 45, 51. Koeplinger, K.A.; Zhao, Z.; Peterson, T.; Leone, J.W.; Schwende, F.S.; Heinrikson, R.L.; Tomasselli, A.G. Drug Metab. Dispos., 1999, 27, 986. Zhao, Z.; Koeplinger, K.A.; Peterson, T.; Conradi, R.A.; Burton, P.S.; Suarato, A.; Heinrikson, R.L.; Tomasselli, A.G. Drug Metab. Dispos., 1999, 27, 992. Oakley, A.J. Curr. Opin. Struct. Biol., 2005, 15, 716. Labrou, N.E.; Mello, L.V.; Clonis, Y.D. Biochem. J., 2001, 358, 101. Board, P.G.; Coggan, M.; Chelvanayagam, G.; Easteal, S.; Jermiin, L.S.; Schulte, G.K.; Danley, D.E.; Hoth, L.R.; Griffor, M.C.; Kamath, A.V.; Rosner, M.H.; Chrunyk, B.A.; Perregaux, D.E.; Gabel, C.A.; Geoghegan, K.F.; Pandit, J. J. Biol. Chem., 2000, 275, 24798. Nieslanik, B.S.; Atkins, W.M. J. Biol. Chem., 2000, 275, 17447. Jemth, P.; Mannervik, B. J. Biol. Chem., 2000, 275, 8618. Axarli, I.A.; Rigden, D.J.; Labrou, N.E. Biochem. J., 2004, 382, 885. Oakley, A.J.; Lo, B.M.; Nuccetelli, M.; Mazzetti, A.P.; Parker, M.W. J. Mol. Biol., 1999, 291, 913. Cameron, A.D.; Sinning, I.; L'Hermite, G.; Olin, B.; Board, P.G.; Mannervik, B.; Jones, T.A. Structure, 1995, 3, 717. Sinning, I.; Kleywegt, G.J.; Cowan, S.W.; Reinemer, P.; Dirr, H.W.; Huber, R.; Gilliland, G.L.; Armstrong, R.N.; Ji, X.; Board, P.G. J. Mol. Biol., 1993, 232, 192. Dirr, H.W.; Wallace, L.A. Biochemistry, 1999, 38, 15631. Salinas, A.E.; Wong, M.G. Curr. Med. Chem., 1999, 6, 279. Caccuri, A.M.; Ascenzi, P.; Antonini, G.; Parker, M.W.; Oakley, A.J.; Chiessi, E.; Nuccetelli, M.; Battistoni, A.; Bellizia, A.; Ricci, G. J. Biol. Chem., 1996, 271, 16193. Caccuri, A.M.; Antonini, G.; Nicotra, M.; Battistoni, A.; Lo, B.M.; Board, P.G.; Parker, M.W.; Ricci, G. J. Biol. Chem., 1997, 272, 29681. Caccuri, A.M.; Lo, B.M.; Nuccetelli, M.; Nicotra, M.; Rossi, P.; Antonini, G.; Federici, G.; Ricci, G. Biochemistry, 1998, 37, 3028. Ivanetich, K.M.; Goold, R.D.; Sikakana, C.N. Biochem. Pharmacol., 1990, 39, 1999. Jakobson, I.; Warholm, M.; Mannervik, B. Biochem. J., 1979, 177, 861. Wilce, M.C.; Parker, M.W. Biochim. Biophys. Acta, 1994, 1205, 1. Kurtovic, S.; Shokeer, A.; Mannervik, B. Protein Eng. Des. Sel., 2008, 21, 329. Meyer, D.J.; Coles, B.; Pemble, S.E.; Gilmore, K.S.; Fraser, G.M.; Ketterer, B. Biochem. J., 1991, 274(Pt 2), 409. Dipple, A. In Polycyclic Hydrocarbons and Carcinogenesis ACS Symposium Series 283; Harvey, R.G., Ed.; American Chemical Society: Washington, D.C, 1985, pp. 1-17. Jerina, D.M.; Chadha, A.; Cheh, A.M.; Schurdak, M.E.; Wood, A.W.; Sayer, J.M. Adv. Exp. Med. Biol., 1991, 283, 533-53. Dreij, K.; Sundberg, K.; Johansson, A.S.; Nordling, E.; Seidel, A.; Persson, B.; Mannervik, B.; Jernstrom, B. Chem. Res. Toxicol., 2002, 15, 825. Jernstrom, B.; Funk, M.; Frank, H.; Mannervik, B.; Seidel, A. Carcinogenesis, 1996, 17, 1491. Sundberg, K.; Widersten, M.; Seidel, A.; Mannervik, B.; Jernstrom, B. Chem. Res. Toxicol., 1997, 10, 1221. Sundberg, K.; Seidel, A.; Mannervik, B.; Jernstrom, B. FEBS Lett., 1998, 438, 206. Sundberg, K.; Johansson, A.S.; Stenberg, G.; Widersten, M.; Seidel, A.; Mannervik, B.; Jernstrom, B. Carcinogenesis, 1998, 19, 433. Van Ness, K.P.; McHugh, T.E.; Bammler, T.K.; Eaton, D.L. Toxicol. Appl. Pharmacol., 1998, 152, 166. Esaki, H.; Kumagai, S. Toxicon, 2002, 40, 941. Talalay, P.; De Long, M.J.; Prochaska, H.J. Proc. Natl. Acad. Sci. USA, 1988, 85, 8261.
GSTs in Drug Discovery and Development [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59]
[60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 41
Esterbauer, H.; Zollner, H.; Schaur, R.J. Membrane Lipid Oxidation; Vigo-Pelfery, E., Ed.; CRC Press: Boca Raton, 1990, pp. 239-268. Esterbauer, H.; Schaur, R.J.; Zollner, H. Free Radic. Biol. Med., 1991, 11, 81. Lovell, M.A.; Xie, C.; Markesbery, W.R. Neurology, 1998, 51, 1562. Gardner, J.L.; Doi, A.M.; Pham, R.T.; Huisden, C.M.; Gallagher, E.P. Toxicol. Appl. Pharmacol., 2003, 191, 95. Hartley, D.P.; Petersen, D.R. Alcohol. Clin. Exp. Res., 1997, 21, 298. Coles, B.F.; Kadlubar, F.F. Methods Enzymol., 2005, 401, 9. Hubatsch, I.; Ridderstrom, M.; Mannervik, B. Biochem. J., 1998, 330 (Pt 1), 175. Canuto, R.A.; Ferro, M.; Muzio, G.; Bassi, A.M.; Leonarduzzi, G.; Maggiora, M.; Adamo, D.; Poli, G.; Lindahl, R. Carcinogenesis, 1994, 15, 1359. Cheng, J.Z.; Singhal, S.S.; Sharma, A.; Saini, M.; Yang, Y.; Awasthi, S.; Zimniak, P.; Awasthi, Y.C. Arch. Biochem. Biophys., 2001, 392, 197. Zimniak, L.; Awasthi, S.; Srivastava, S.K.; Zimniak, P. Toxicol. Appl. Pharmacol., 1997, 143, 221. Gallagher, E.P.; Gardner, J.L. Biochem. Pharmacol., 2002, 63, 2025. Fenwick, G.R.; Heaney, R.K.; Mullin, W.J. Crit. Rev. Food Sci. Nutr., 1983, 18, 123. Kolm, R.H.; Danielson, U.H.; Zhang, Y.; Talalay, P.; Mannervik, B. Biochem. J., 1995, 311 (Pt 2), 453. Kagan, V.E. Lipid Peroxidation in Biomembranes, CRC Press: Boca Raton, 1988. Slater, T.F. Biochem. J., 1984, 222, 1. Hayes, J.D.; Pulford, D.J. Crit. Rev. Biochem. Mol. Biol., 1995, 30, 445. Uchida, K.; Toyokuni, S.; Nishikawa, K.; Kawakishi, S.; Oda, H.; Hiai, H.; Stadtman, E.R. Biochemistry, 1994, 33, 12487. Harris, M.E.; Hensley, K.; Butterfield, D.A.; Leedle, R.A.; Carney, J.M. Exp. Neurol., 1995, 131, 193. Garland, D.; Russell, P.; Zigler, J.S. The oxidative modification of lens proteins. In Oxygen Radicals in Biology and Medicine: M.G. Simic, K.S. Taylor, J.F. Ward and C. von Sonntag, Eds; Plenum Press: New York, 1998, pp. 347-353. Yoritaka, A.; Hattori, N.; Uchida, K.; Tanaka, M.; Stadtman, E.R.; Mizuno, Y. Proc. Natl. Acad. Sci. USA, 1996, 93, 2696. Epp, O.; Ladenstein, R.; Wendel, A. Eur. J. Biochem., 1983, 133, 51. Ursini, F.; Maiorino, M.; Valente, M.; Ferri, L.; Gregolin, C. Biochim. Biophys. Acta, 1982, 710, 197. Ren, B.; Huang, W.; Akesson, B.; Ladenstein, R. J. Mol. Biol., 1997, 268, 869. Eshdat, Y.; Holland, D.; Faltin, Z.; Ben-Hayim, G. Physiol. Plant, 2008, 100, 234. Mannervik, B. Methods Enzymol., 1985, 113, 490. Yang, Y.; Cheng, J.Z.; Singhal, S.S.; Saini, M.; Pandya, U.; Awasthi, S.; Awasthi, Y.C. J. Biol. Chem., 2001, 276, 19220. Prabhu, K.S.; Reddy, P.V.; Gumpricht, E.; Hildenbrandt, G.R.; Scholz, R.W.; Sordillo, L.M.; Reddy, C.C. Biochem. J., 2001, 360, 345. Hurst, R.; Bao, Y.; Jemth, P.; Mannervik, B.; Williamson, G. Biochem. J., 1998, 332, 97. Awasthi, Y.C.; Sharma, R.; Singhal, S.S. Int. J. Biochem., 1994, 26, 295. Gallagher, E.P.; Gardner, J.L.; Barber, D.S. Biochem. Pharmacol., 2006, 71, 1619. Awasthi, Y.C.; Beutler, E.; Srivastava, S.K. J. Biol. Chem., 1975, 250, 5144. Awasthi, Y.C.; Dao, D.D.; Saneto, R.P. Biochem. J., 1980, 191, 1. Zhao, T.; Singhal, S.S.; Piper, J.T.; Cheng, J.; Pandya, U.; Clark-Wronski, J.; Awasthi, S.; Awasthi, Y.C. Arch. Biochem. Biophys., 1999, 367, 216. Singhal, S.S.; Saxena, M.; Ahmad, H.; Awasthi, S.; Haque, A.K.; Awasthi, Y.C. Arch. Biochem. Biophys., 1992, 299, 232. Morgenstern, R.; DePierre, J.W. Eur. J. Biochem., 1983, 134, 591. Morgenstern, R.; Lundquist, G.; Mosialou, E.; Anderson, C. Glutathione S-Transferases and Drug resistance; Hayes, Pickett, and Mantle, Eds.; Taylor and Francis: New York, 1990, pp. 57-64. Hanson, S.R.; Chen, A.A.; Smith, J.B.; Lou, M.F. J. Biol. Chem., 1999, 274, 4735. Seres, T.; Ravichandran, V.; Moriguchi, T.; Rokutan, K.; Thomas, J.A.; Johnston, R.B., Jr. J. Immunol., 1996, 156, 1973. Kosower ,N.S.; Kosower, E.M. Int. Rev. Cytol., 1978, 54, 109. Chai, Y.C.; Jung, C.H.; Lii, C.K.; Ashraf, S.S.; Hendrich, S.; Wolf, B.; Sies, H.; Thomas, J.A. Arch. Biochem. Biophys., 1991, 284, 270. Walters, D.W.; Gilbert, H.F. J. Biol. Chem., 1986, 261, 13135. Garcera, A.; Barreto, L.; Piedrafita, L.; Tamarit, J.; Herrero, E. Biochem. J, 2006, 398, 187. Raghavachari, N.; Qiao, F.; Lou, M.F. Exp. Eye Res., 1999, 68, 715.
42 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108]
Skopelitou et al.
Boyland, E.; Chasseaud, L.F. Biochem. J., 1967, 104, 95. Keen, J.H.; Habig, W.H.; Jakoby, W.B. J. Biol. Chem., 1976, 251, 6183. Kubal, G.; Meyer, D.J.; Norman, R.E.; Sadler, P.J. Chem. Res. Toxicol., 1995, 8, 780. Chen, H.; Juchau, M.R. Biochem. J.; 1998, 336 (Pt 1), 223. Benson, A.M.; Talalay, P.; Keen, J.H.; Jakoby, W.B. Proc. Natl. Acad. Sci. USA, 1977, 74, 158. Johansson, A.S.; Mannervik, B. J. Biol. Chem., 2001, 276, 33061. Johansson, A.S.; Mannervik, B. J. Biol. Chem., 2002, 277, 16648. Benson, A.M.; Talalay, P. Biochem. Biophys. Res. Commun., 1976, 69, 1073. Pettersson, P.L.; Mannervik, B. J. Biol. Chem., 2001, 276, 11698. Ito, S.; Narumiya, S.; Hayaishi, O. Prostaglandins Leukot. Essent. Fatty Acids, 1989, 37, 219. Atkins, W.M.; Wang, R.W.; Bird, A.W.; Newton, D.J.; Lu, A.Y. J. Biol. Chem., 1993, 268, 19188. Ujihara, M.; Tsuchida, S.; Satoh, K.; Sato, K.; Urade, Y. Arch. Biochem. Biophys., 1988, 264, 428. Fernandez-Canon, J.M.; Penalva, M.A. J. Biol. Chem., 1998, 273, 329. Blackburn, A.C.; Tzeng, H.F.; Anders, M.W.; Board, P.G. Pharmacogenetics, 2000, 10, 49. Blackburn, A.C.; Coggan, M.; Tzeng, H.F.; Lantum, H.; Polekhina, G.; Parker, M.W.; Anders, M.W.; Board, P.G. Pharmacogenetics, 2001, 11, 671. Dietze, E.C.; Grillo, M.P.; Kalhorn,T.; Nieslanik, B.S.; Jochheim, C.M.; Atkins, W.M. Biochemistry, 1998, 37, 14948. Hederos, S.; Broo, K.S.; Jakobsson, E.; Kleywegt, G.J.; Mannervik, B.; Baltzer, L. Proc. Natl. Acad. Sci. USA, 2004, 101, 13163. Ibarra, C.; Grillo, M.P.; Lo, B.M.; Nucettelli, M.; Bammler, T.K.; Atkins, W.M. Arch. Biochem. Biophys., 2003, 414, 303. Kurtovic, S.; Moden, O.; Shokeer, A.; Mannervik, B. J. Mol. Biol., 2008, 375, 1365. Kurtovic, S.; Grehn, L.; Karlsson, A.; Hellman, U.; Mannervik, B. Anal. Biochem., 2008, 375, 339. Saavedra, J.E.; Bohle, D.S.; Smith, K.N.; George, C.; Deschamps, J.R.; Parrish, D.; Ivanic, J.; Wang, Y.N.; Citro, M.L.; Keefer, L.K. J. Am. Chem. Soc.; 2004, 126, 12880. Morgan, A.S.; Ciaccio, P.J.; Tew, K.D.; Kauvar, L.M. Cancer Chemother. Pharmacol., 1996, 37, 363. Findlay, V.J.; Townsend, D.M.; Saavedra, J.E.; Buzard, G.S.; Citro, M.L.; Keefer, L. K.; Ji, X.; Tew, K.D. Mol. Pharmacol., 2004, 65, 1070. Gillam, E.M.J. Trends Pharmacol. Sci., 2001, 22, 11. Henderson, C.J.; Wolf, C.R.; Kitteringham, N.; Powell, H.; Otto, D.; Park, B.K. Proc. Natl. Acad. Sci. USA, 2000, 97, 12741.
Frontiers in Drug Design & Discovery, 2009, 4, 43-80
43
New Sampling Techniques for PharmacokineticPharmacodynamic Modeling Christian Höcht1,2,*, Marcos Mayer1,2, Javier A.W. Opezzo1,2, Guillermo F. Bramuglia1 and Carlos A. Taira1,2 1
Cátedra de Farmacología, 2Instituto de Fisiopatología y Bioquímica Clínica, Facultad de Farmacia y Bioquímica, Universidad de Buenos Aires, Junín 956, (C1113AAD) Buenos Aires, Argentina Abstract: Considering that pharmacokinetic–pharmacodynamic (PK–PD) modeling describes the relationship between tissue concentrations of drugs and their corresponding pharmacological response, an important issue of PK-PD studies is the availability of powerful sampling techniques that allow measurement of tissue concentrations of drugs at multiple time points. Traditional sampling techniques, including biopsy, blood and saliva sampling, and skin blister sampling, have several limitations for drug monitoring during PK-PD studies, considering that these techniques did not allow the measurement of drug concentrations at the site of action. In the last decades, new sampling techniques, including membrane based techniques (microdialysis and ultrafiltration) and imaging techniques (positron emission tomography and magnetic resonance spectroscopy), have been available for measurement of drug concentration at the target site. The possibility of simultaneous monitoring of target site concentrations of drugs and their pharmacological effect with these new sampling techniques have significantly improves current knowledge of PK-PD modeling. In addition, membrane based techniques also allow simultaneous monitoring of endogenous compounds and therefore permit the study of the relationship between drug target site concentrations and their effect on biochemical markers, making these techniques highly useful for PK-PD modeling studies. The aim of this chapter is to describe the principles of membrane based techniques and imaging techniques, and their applicability for drug monitoring in PK-PD modeling.
INTRODUCTION The demand for efficacious pharmacological agents is increasing due to higher lifestyle expectations and changes in demographic profiles. The discovery of the existence of a large number of orphan receptors will increase the number of new therapeutic agents to be tested. The availability of new techniques, such as the Human –Genome Project and the High throughput screening and combinatorial chemistry, will also *Corresponding Author: Tel: +(54-11)-4964-8265; Fax: +(54-11)-4508-3645; E-mail:
[email protected]
Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
44 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
identify an exploding number of new molecular targets as well as potential new therapeutic agents [1]. Therefore, a more rigorous selection process will be needed at early stages of drug development. Integration of pharmacokinetic-pharmacodynamic (PK-PD) concepts through PK-PD modeling is a potential tool to enhance the information gain and the efficiency of drug selection during drug development [2-4]. PK-PD modeling became as an important tool in clinical practice due to the possibility of an adequate dose regimen selection and the early detection of non-responders to the pharmacological therapy [2, 4]. Another important trend of drug properties to evaluate at early stages of drug development is its capability to reach the target site at sufficient amount to exert its therapeutic effect. It has been estimated that up to 40% of new chemical entities fail in the early phases of drug development because of inappropriate pharmacokinetic properties, such as inadequate drug distribution [5]. Target site distribution is also an important factor in the clinical use of pharmacological agents, especially antiinfective, antineoplastic and central acting drugs, to ensure success of pharmacological interventions. Although pharmacokinetic properties of drugs can be studied by means of traditional blood sampling, this technique is limited by the impossibility to evaluate variability in drug distribution at the target site. Therefore, in the last years, regulatory authorities have been emphasizing the evaluation of drug distribution at the target site during drug development (http://www.fda.gov/cder/guidance/2580dft.pdf). Although other traditional sampling techniques, such as biopsy and skin blister, allow measurement of tissue concentrations of drugs, the utility of these methodologies is strongly limited by the existence of serious limitations, including the inability of serial sampling for the study of the complete tissue pharmacokinetic profile of the drug [6]. In the last decades, new sampling techniques, including microdialysis and imaging techniques, have been available and validated for the study of drug distribution at the target site and PK-PD modeling [6]. Use of these techniques in preclinical and early clinical drug development might reduce failing rates of new chemical entity, also improving cost and the time required until a new chemical entities reaches the market. In addition, these new sampling techniques have also been used in clinical practice to ensure adequate tissue distribution of drugs and for the study of PK-PD models of drugs. Taking into account these antecedents, the aim of this chapter is to: i) describe the concepts of drug distribution at the target site and PK-PD modeling; ii) compare the available techniques for the study of tissue distribution of drugs and PK-PD modeling and iii) to comment the methodological and practical aspects of microdialysis and imaging techniques, and their application for the study of drug distribution and PK-PD modeling of drugs and new chemical entities during drug development and clinical practice. THE IMPORTANCE OF MEASURING TARGET SITE CONCENTRATIONS OF DRUGS The selection of the adequate dose and dosing timing is a fundamental step to achieve therapeutic success with any pharmacological treatment. Selection of the best drug and dosing regimen does not only increase the likelihood of therapeutic success but also reduces the appearance of adverse drug reactions. However, the selection of the most adequate drug and dosing interval of the same is not an easy issue. For most thera-
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 45
peutic agents, drug and dose regimen are empirically selected based on trial and error and by adjusting drug dosing taking into account patient weight. However, this strategy is subject to high clinical failure because of large interindividual variability in pharmacokinetic and pharmacodynamic properties of therapeutic agents in patient population [7, 8] (Fig. 1). Currently, therapeutic drug monitoring (TDM) (Fig. 1) is applicated for drugs with small therapeutic window, including antiepileptic drugs, antimicrobials, inmunesupressors, antiviral agents, teophylline and others, to individualize pharmacological interventions [9]. TDM guides patient dosing according to plasma concentrations of therapeutic agents. In other words, measurement of plasma concentrations has the objective to consider the relationship between dose and attained plasma concentrations in the large interindividual variability of drug response in a treated population [9]. Nevertheless, TDM has several limitations and its cost-benefit relationship is controversial [10]. It must be taken into account that most therapeutic agents exert their pharmacological effect in peripheral tissues rather than the blood compartment, and tissue and plasma pharmacokinetics may differ substantially in patients and healthy volunteers [11-13]. Large interindividual variability also exists between tissue concentrations and therapeutic response because of different receptor sensitivity along the patient population [7]. In conclusion, although TDM has been established as a valuable therapeutic tool for some drugs, measurement of plasma concentrations of drugs is a surrogate parameter with several drawbacks, since it does not exclude interindividual variability in drug distribution at the target site and pharmacodynamic variability.
Fig. (1). Sources of interindividual variability in drug response and methods available for their assessment.
Based on the assumption that unbound plasma concentrations and free tissue levels are equal at equilibrium, it is frequently considered that total plasma concentrations and plasma protein binding can be used to predict free tissue levels of therapeutic agents. However, many studies have shown lower tissue unbound levels than plasma concentrations [14-17]. Tissue distribution is also affected by anatomic barriers, such as blood-brain barrier (BBB), the presence of active transport systems, like P-glycoprotein, and tissue metabolism. For instance, BBB prevents water soluble xenobiotics from entering the brain [18]. BBB is mainly formed by brain capillary endothelial cells, pericytes, astrocytes, and
46 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
neuronal cells. Endothetial cells are different from peripheral cells because of the presence of tight junctions, which prevent paracellular transport of water-soluble compounds from blood to the brain [19]. It has also been discovered that many transport systems play an important role in determining tissue concentrations of drugs. These comprise carrier- and receptormediated transport systems, including cationic and anionic influx and efflux systems such as P-glycoprotein [20], multidrug-resistance (MDR) proteins [21], nucleoside transporters, organic anion transporters, organic cation transporters, large amino acid transporter and the scavenger receptors SB-AI and SB-BI [22]. These transporters modulate drug disposition process, with particular importance in terms of the clinical implications of transporters for drug-drug interactions, drug toxicity, interindividual variability in drug response, and disease [8]. Efflux transporters, such as P-glycoprotein, are expressed in several tissues, especially in the central nervous system, hepatocytes, kidney and the small intestine. At these sites, efflux transporters reduce drug bioavailability, tissue distribution and enhance drug elimination in order to reduce drug exposure in a given patient [23]. Moreover, an overexpression of efflux transporters exists in several pathologies affecting therapeutic success because of limited distribution of the drug at the target site [24]. For instance, reduction of central bioavailability of antiepileptic drugs by overexpression of efflux transporters has been established for several drugs in different models of experimental epilepsy as a mechanism of pharmacoresistance (for review see [25, 26]). The need to study antiepileptic drug distribution at the target site is emphasized by the fact that overexpression of efflux transporters seems only to affect drug distribution in the biophase and not in other central nuclei. The finding that resistant epileptic patients have similar central adverse effect to drug treatment than responders supports this conclusion [22]. Success of cancer treatment is also affected by the existence of an enhanced efflux transporter activity. In this case, efflux transporters reduce intracellular concentrations of cytotoxic drugs compromising the activity of chemotherapeutic agents [27]. Tissue metabolism also affects the relationship between plasma unbound concentrations of therapeutic agents and their levels at the biophase. Although most part of drug metabolism takes place in hepatocytes, other tissues present relative high metabolic activity, including small intestine, central nervous system and kidney [28]. Therefore, tissue biotransformation could also influence target site concentrations and pharmacological effects. Taking into account these antecedents, the Food and Drug Administration (FDA) encourages the study of tissue distribution of antimicrobial agents in unaffected and infected target sites and the relationship of unbound drug concentrations at the site of action to the in vitro susceptibility of the infecting microorganism (http://www.fda.gov/ cder/guidance/2580dft.pdf). In conclusion, the utility of therapeutic tools actually used for rational selection of the best drug and dosing regimen, such as TDM, is limited by the fact that this approach allows the determination of plasma drug concentrations rather than their levels at the target site. Several pharmacokinetic factors affect tissue distribution of drugs and their pharmacological effect. Therefore, an important fact in drug design and clinical practice
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 47
is the evaluation of the in vivo pharmacokinetics of new chemical entities and drugs at the biophase. In the following sections of the present chapter, available sampling techniques for the study of drug distribution at the target site, including imaging techniques and microdialysis, are discussed. Variability in drug response is also due to pharmacodynamic factors [7] (Fig. 1). Transporters activity, concentration of endogenous agonist and expression of receptors and their sensitivity to drug action vary among different patients and during the course of drug treatment on a single patient. Therefore, selection of the adequate dose of a therapeutic agent and its dose regimen also depends on pharmacodymanics of the drug at the target site. In the last years, PK-PD modeling have become an interestingly tool of pharmacological sciences for the study of factors influencing drug response [3] (Table 1). Table 1.
Information Obtained from Pharmacokinetic-Pharmacodynamic Modeling During Preclinical and Clinical Phases of Drug Development and Clinical Practice
Preclinical phase of drug development Precise definition of the dose–concentration–pharmacological effects and dose–concentration–toxicity relationship Determination of the appropriate dosing regimen for Phase I studies Identification of biomarkers and animal models for efficacy and toxicity Explore any dissociation between plasma concentration and duration and onset of pharmacological effect Provide information on drug effects that would be difficult to obtain in human subjects Provide insights regarding the mechanism of action of the new chemical entity Allows the study of the influence of the physiopathological state on PK-PD parameters Clinical phase of drug development Understanding the dose-concentration–pharmacological effects and dose–concentration–toxicity relationship in healthy volunteers. Characterization of pharmacokinetics and pharmacodynamics in special population Study of tolerance development Determination of the dosing regimens for Phase II studies. Confirms and explores the relationship between dose–concentration–effect in patients. Examination of a variety of therapeutic endpoints to understand the most adequate for further modeling. Determination of the dosing regimens for Phase III studies Prediction of the probability distribution of further clinical trial outcomes Clinical practice Early detection of poor responders or non-responders. Optimization of antihypertensive drug regimen, in terms of dose, sampling interval and time of dosing. Evaluation of the clinical impact of drug interactions.
UTILITY OF PHARMACOKINETIC-PHARMACODYNAMIC MODELING IN DRUG DISCOVERY AND CLINICAL PRACTICE Two different approaches exist for characterizing the pharmacodynamics of a drug, namely the classical dose response trial and PK–PD modeling [4]. PK–PD modeling has
48 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
several advantages over classical dose–response studies. PK–PD modeling allows not only better pharmacodynamic characterization of drugs, but also permits screening and dosage–regimen selection [4]. More recently, the introduction of mechanism-based PKPD models have allowed the study of the influence of physiological parameters on PKPD properties of the drug. Information obtained from PK-PD modeling during drug development and clinical practice is resumed in Table 1. PK-PD relationships build a bridge between the time course of drug concentrations in the organism, as assessed by pharmacokinetics (PK), and the intensity of the observed pharmacological response, as quantified by pharmacodynamics (PD). The link between PK and PD of a drug is established by the use of mathematical models, allowing the estimation of parameters such as effective concentration to yield half-maximal response (EC50) and maximal efficacy (Emax). PK-PD modeling also provides information about the onset, magnitude and duration of the pharmacological effect [4]. A drawback of PK–PD modeling is the need of simultaneous measurement of drug tissue levels and their corresponding pharmacological effect at multiple time points [4]. Blood sampling, which has traditionally been used for this purpose, has the disadvantage that the removal of samples by themselves may interfere with PK and PD drug behavior, especially in preclinical studies with small animals [29]. Furthermore, traditional sampling techniques allow the measurement of plasma concentrations of pharmacological agents rather than their levels at the target tissue. These limitations could be resolved by the application of new sampling techniques, including in vivo microdialysis and imaging techniques. Accurate measurement of the intensity of the pharmacological effect of the active compounds is also necessary for a PK-PD modeling design. A drug effect could be considered as any change in physiological parameters induced by the administration of a drug, compared to respective baseline values. Measurement of the effect should meet validation parameters, such as continuity, sensitivity, objectivity and repeatability [30]. To obtain the greatest precision in estimating PK-PD relationships, the number of measurements of drug tissue levels and its corresponding effect must be as large as possible [30]. However, multiple time point sampling is not always possible in the clinical setting. To overcome this limitation, population PK and PK-PD modeling is increasingly being introduced [31, 32]. PK-PD relationships have been described using diverse mathematical models depending on the nature of drug administration, the magnitude of the pharmacological effect and the time dependency of pharmacodynamics of the tested drug [3, 31]. Relative simple PK-PD models are needed for describing PK-PD relationship after multiple doses or long-term infusion, because the system is kinetically at steady state [31]. The most common mathematical equations employed at steady state condition are the linear, log linear and the Emax model. Although the linear and log linear model allow easy parameter estimation, these models erroneously assume that the effect can increase with concentrations without limits [30]. Therefore, the Emax model is the most broadly applied to characterize a myriad of pharmacological effects. This model derives from the classical theory of drug-receptor interaction, relating the effect to drug concentrations as in the following equation:
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 49
E = E0 +
E max C EC 50 + C
where E0 is the baseline effect, Emax the maximal effect and EC50 the effective concentration yielding half maximal response. Conversely, more complex PK-PD models are needed for describing PK-PD relationship after single dose administration or when time dependency in the pharmacodynamics of the drug is present [31]. Plotting drug effects as a function of drug concentrations and connecting data in a chronological order allows the determination of possible delays between drug response and its tissue levels [4]. A hysteresis loop appears in the plotting when the magnitude of an effect corresponds to more than one drug concentration. Anticlockwise hysteresis loop could be explained by the disequilibrium between biophase and plasma compartment [33], appearance of active metabolites [34] or indirect mechanism of action [35]. On the other hand, tolerance to the pharmacological effect is suggested when a clockwise hysteresis loop is observed [36]. In theses cases, plasma concentrations can not be directly linked to drug effect, and more complex PK-PD models, such as an effect compartment model and a physiological indirect response model, are needed [31]. The effect compartment model considers a hypothetical effect compartment as an additional compartment of a pharmacokinetic compartment model, representing the drug concentration at the effect site. The time-dependent aspects of the equilibrium between plasma concentration and effect are characterized by the first-order rate constant ke0 that represents the irreversible disappearance of the drug from the effect compartment and is in equilibrium with the rate constant of entrance of the drug to the effect compartment. The time course of drug concentration in the effect compartment is described by the following equation:
dC e = k e 0 (C p C e ) dt where Ce and Cp represent the concentration in the effect compartment and plasma, respectively and ke0 the equilibration rate constant. This approach has been successfully applied to predict the PK-PD relationship of diverse drugs [37]. Physiologic indirect response models have been employed in a variety of studies regarding biological responses such as muscle relaxation, synthesis and secretion of endogenous compounds, mediator flux, cell trafficking, enzyme induction, or inactivation, among others [35]. The physiological indirect response can be described as an inhibitory or stimulatory function applying the following equations:
I (t ) = 1
C (t ) IC50 + C (t )
S (t ) = 1 +
E max C (t ) EC 50 + C (t )
where IC50 and EC50 are the concentrations eliciting 50% of inhibition or the maximal effect (Emax), respectively.
50 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
PK–PD modeling should cover the complete pharmacodynamic range of a drug after a single administration [4]. Often, in clinical pharmacology, it is not possible to determine the maximal effect of a drug because of the appearance of adverse drug reactions, in which case an alternative PK–PD model must be applied [38]. In this model, the authors replaced the parameter EC50 with S0 in the Emax equation, which represents the initial sensitivity to the drug. The following equation describes the relationship between drug concentration and its pharmacological effect for this PK-PD model:
E = E0 +
S 0 E max C E max + S 0 C
Considering that the selection of an inadequate PK–PD model according to the PK– PD study characteristics might lead to an erroneous interpretation, it is extremely important to determine which PK–PD model is going to be applied for the analysis of the data. When selecting the PK–PD model, the investigator must keep in mind experimental variables, including type of drug administration, type of pharmacological effect measured, the existence of time dependency in the pharmacological effect of a drug and the possibility to reach a maximal response in their experimental design. More recently, development of mechanism-based models has improved knowledge of the interaction between PK-PD properties of drugs and clinical response (for review see [39, 40]). Mechanism-based PK-PD modeling integrates parameters for describing drug-specific properties with biological system-specific properties, and therefore characterizes the causal path between drug exposure and drug response. By estimating drug target-site distribution, target binding and activation and transduction process, mechanism-based PK-PD models will also characterize the interaction of the drug effect with disease processes and disease progression [39, 40]. OVERVIEW OF AVAILABLE TECHNIQUES FOR THE STUDY OF DRUG DISTRIBUTION AND PK-PD MODELING Target site concentrations of drugs and new chemical entities can be evaluated by means of traditional techniques, including biopsies, skin blisters, or using more recently available methodologies, such as imaging techniques, ultrafiltration and microdialysis [6]. Biopsy consists of homogenized tissue samples those results in cellular lysis. Therefore, biopsy gives information about analyte concentrations in the homogenate without distinguishing between blood, intra- or extracellular drug levels. Moreover, it is an invasive method associated with the risk of infection due to cross-contamination and scarring. As tissue has to be removed, biopsy is not suitable for the study of the concentration–time course of a drug and, therefore, PK-PD modeling of drugs [41]. Advantages of biopsy are that no special equipment is needed, there is no limitation with the size of the drug and no calibration is necessary [41]. Suction blisters have been an established method for more than 30 years for the study of pharmacokinetics in the dermis [42]. The method principle relies on the separation of the epidermis from the dermis along the lamina lucida due to the application of prolonged suction to the skin surface (suction blister technique) or due to the adverse reaction effect of cantharidin (cantharides blister technique). The fluid drawn into the sepa-
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 51
rated epidermis can be sampled and analyzed for drug content [41]. Skin blister technique has been compared to other sampling techniques, such as microdialysis, concluding that skin blister seems to be only valid for the dermal study of low protein bound drugs. Other limitations of skin blister sampling is the production of a great discomfort in the patients and is only applicable for studies at a certain time point not allowing continuous monitoring of drug skin penetration to the same test area. Traditional techniques have several drawbacks that limit their applicability for the evaluation of tissue distribution and PK-PD modeling during drug development. These methodologies have largely been replaced by modern analytical techniques, including ultrafiltration, imaging techniques and microdialysis sampling (Table 2). Table 2.
New Sampling Techniques for Pharmacokinetic and PK-PD Studies
Technique
Advantages
Microdialysis Simultaneous determination of the bioactive concentration of drugs, metabolites and endogenous compounds in the biophase. Good temporal and spatial resolution. Online coupling of analytical determination Monitoring of drug concentrations in different tissues by multiprobe microdialysis. Monitoring of drug effect on endogenous compounds Possibility of local drug administration. Cheap technique.
Drawbacks Semi-invasiveness technique Diluting effect of the microdialysis procedure Need of highly sensitive analytical methods In vivo calibration of the microdialysis probe during the experiment Sticking of lipophilic drugs to tubing and probe components Low recovery of large molecules
PET
Non-invasive technique High spatial and time resolution No time consuming Allows estimation of in vivo drug-receptor interactions
Short physical half-life of the most used radioisotope. Did not discriminate between drugs and its metabolites. Determinates total tissue concentrations of drugs. Expensive technique.
MRS
Allows discrimination between drugs and its metabolites. Non-invasive technique No time consuming
Poor spatial resolution. Low sensitivity Expensive technique
Ultrafiltration is an alternative membrane sampling technique to microdialysis. This technique collects a sample by the application of negative pressure as driving force. The rate of fluid collection is determined by the amount of negative pressure applied, the membrane surface area and the hydraulic resistance [43]. The mayor advantage of ultrafiltration with regards to other membrane sampling techniques, such as microdialysis, is that in vivo calibration of the technique is not necessary because the in vivo recovery is greater than 95% for small molecules [43]. However, continuous tissue sampling with
52 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
ultrafiltration depends on rapid replacement of interstitial fluid by blood vessels. In tissues with limited flow rate and low replenishment of interstitial fluid, such as subcutaneous tissue, only low sampling rates are possible [43]. Brain is also unsuitable for ultrafiltration sampling because of the limited extracellular space. In addition, ultrafiltration does not allow the administration of a substance into the extracellular space through the probe. Although the applicability of ultrafiltration has been demonstrated for drug kinetic studies and glucose and lactate monitoring [43], the scope of application of ultrafiltration technique is limited in clinical pharmacology. Microdialysis is a membrane sampling technique that allows continuous monitoring of extracellular concentrations of new chemical entities and drugs withouth fluid loss [29]. As discussed later, microdialysis has several advantages compared to other sampling techniques, including the capability to monitor unbound tissue concentrations of drugs and their metabolites, the simultaneous study of drug concentrations in several tissues and the possibility of concomitant sampling of endogenous compounds or biochemical markers [37]. However, it must be pointed out that microdialysis is a semiinvasive methodology that often requires calibration during the experiment [29]. Alternatively, several imaging techniques such as planar -scintigraphy, single photon emission computed tomography (SPET), positron emission tomography (PET), and magnetic resonance spectroscopy (MRS) have been developed for the study of drug distribution in basic and clinical settings [13]. PET is a new nuclear imaging technique that employs molecules labeled with positron-emiting radioisotopes [44]. Advantages of this imaging technique reside in its non-invasive nature, high spatial resolution (1-5 mm) and time resolution (30 seg.). However, the physical half-life of the most used radioisotope 11 C (20.4 min) does not allow monitoring of tissue levels of radiolabeled drugs over several elimination half-lifes as desired in pharmacokinetic studies [13]. Another drawback of PET is that this methodology samples total tissue concentrations of drugs and their metabolites and, consequently, does not allow to differenciate them [13]. MRS utilizes nuclear magnetic resonance phenomenon, which consists in the emission of a radiosignal equal to the respective resonance frequency with amplitude proportionally to the number of nuclei present in the examinant objective [13]. MRS can be performed serially with a temporal resolution of minutes. Contrary to other imaging techniques, MRS is capable of identifying drug metabolites in tissues [13]. However, an important limitation of MRS is its low sensitivity, allowing only the evaluation of drug distribution of xenobiotics that are present in large concentrations. In addition, spatial resolution of this imaging methodology is low [13]. Although imaging techniques are non-invasive, they are only applicable for a small group of compounds with special functional groups. Moreover, imaging techniques are very expensive and labor-intensive and, therefore, not suitable for clinical routine settings [13]. In conclusion, to date, modern techniques exist for the evaluation of tissue distribution of drugs at the target site and simultaneous PK-PD modeling, providing information that has been inaccessible with previously available methods. In the next sections, the principles and applicability of microdialysis, PET and MRS for evaluation of pharmacokinetic and pharmacodynamic properties of drugs will be discussed.
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 53
PRINCIPLES OF MICRODIALYSIS SAMPLING In the past twenty years, the microdialysis technique has become a method of choice for the study of tissue concentrations of both endogenous and exogenous substances. Microdialysis is a powerful and versatile sampling technique based on the dialysis principle [45]. In this technique, a probe that is inserted into a tissue mimics the function of a capillary blood vessel (for review see [29]). The probe has a hollow fiber that is permeable to water and small molecules, and when the perfusate solution passes through the dialysis membrane, molecules diffuse into (recovery) or out of (delivery) the perfusion fluid depending on the concentration gradient. Thus, microdialysis can be used for both collecting a substance in the dialysate as well as delivering it into the periprobe fluid. The latter is referred to as reverse microdialysis [46]. Then, dialysate samples are analyzed using quantitative techniques like liquid chromatography (LC) or capillary electrophoresis (CE). The basic setup for a microdialysis experiment consists of a microdialysis probe, a perfusion pump, and an analytical method with the required sensitivity to quantify small concentrations of substances [29]. As microdialysis sampling is not performed under equilibrium conditions because the perfusate is constantly being pumped through the probe, concentrations of the drug in the sample are some fraction of that in the surrounding tissue. Therefore, limit of quantification of analytical methods should be extremely low [47]. On the other hand, to obtain true tissue concentrations the factor by which these concentrations are interrelated needs to be determined. This factor, which is obtained during an in vivo or an in vitro calibration procedure, is called relative recovery [29]. Microdialysis sampling offers several advantages over conventional methods and imaging methods of studying the pharmacokinetics and pharmacodynamics of drugs. Microdialysis technique allows continuous tissue sampling without removing liquid, achieving higher temporal resolution than traditional techniques, without interfering with the PK and PD behavior of the drug. An economical and ethical advantage is that 5-10 times fewer animal experiments have to be performed to determine the time profile of a drug [29]. Placement of multiple microdialysis probes in different tissues allows monitoring of the drug time course in different organs in the same animal, supplying information about the distribution process of xenobiotics. The fact that the microdialysis technique provides protein free samples permits the analysis of the sample without pretreatment and LC or CE on-line coupling. Conversely, traditional blood sampling requires clean up procedures prior to analysis with the possibility of analyte loss during protein precipitation and the need of an internal standard for an accurate determination of the drug. On the other hand, tissue enzymes are also excluded from the dialysate sample without further enzymatic degradation of the drug [29]. The microdialysis technique not only allows sampling of extracellular levels of drugs but also endogenous compounds such as neurotransmitters, metabolites, glucose, lactate and low molecular wheighing peptides [29]. Therefore, this technique is highly useful for the simultaneous study of the pharmacokinetic behavior of drugs and their effects on extracellular levels of endogenous compounds [37].
54 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
However, the microdialysis sampling technique has some disadvantages compared to traditional sampling methods. Drug tissue concentrations often are in the picomolar to micromolar range and the limit of detection of the analytical methods should be for some drugs extremely low. On the other hand, in vivo recovery of the microdialysis probe must be determined during the experiment to permit conversion of microdialysis concentration into extracellular levels if an absolute concentration is needed [29]. The calibration aspects of the microdialysis probe will be discussed in following sections. In addition, a minimal lesion of the tissue surrounding the probe is produced by implantation of the microdialysis probe, causing tissue responses such as a compromise of BBB [48, 49] and acute inflammation in different tissues [47, 50, 51]. Nevertheless, several studies have demonstrated that baseline conditions are reached following periods of 60 min after probe implantation [29]. Methodological Aspects of Microdialysis Sampling The microdialysis probe is perhaps the nucleus of the microdialysis experiment. Many types are described and they are used in different experiments. The different geometry of microdialysis probes enables their use in virtually any tissue and any fluid of the body [52]. Many laboratories have designed their own probe considering that its construction takes no longer than several minutes. To date, there are also many commercial approved probes for studies in soft tissues and brain of humans and animals. Probes will have a longitudinal, a semicircular or an I-shape design. Various designs have been described, including concentric cannula probe, linear probe, and shunt probe. In addition, several modified probe designs have been reported, such as spinal loop dialysis catheter, flexible intravenous probe, and shunt intraarterial microdialysis probe [29]. For soft peripheral tissues like muscle, skin, liver, tumor, and fluids like blood and bile, flexible probes can be used. Linear probes are useful for monitoring transdermal drug delivery. However, an increase in skin blood flow, erythema and skin thickness was demonstrated after insertion of a microdialysis probe [53]. Rigid microdialysis probes, such as the concentric cannula, are placed in hard tissues, including the brain. In this case, guide cannulas can also be implanted, opening up the possibility to insert the probe after surgical recovery, thereby decreasing negative effects of anesthesia [29]. The choice of the membrane type is an essential element to optimize the probe for a particular microdialysis protocol. Conventional microdialysis probes are constructed with 20 kDa molecular weight cut-off membranes enabling measurement of small molecules [29]. Recently, a 100 kDa molecular weight cut-off microdialysis catheter have been used to sample larger molecules such as cytokines [54]. These probes can be constructed with polyethylenesulphone [55]. A technical problems encountered with the 100 kDa microdialysis membrane is the issue of poor sample volume retrieval, considering that it enables passage of high molar mass molecules. One way of addressing this problem is to equalize the pressure differences between in-flowing and out-flowing perfusate at the membrane by increasing the colloidal pressure of the perfusate using dextran [56]. Molecular weight of the substance of interest must be taken into consideration for selection of the most adequate dialysis membrane. Only substances with a molecular weight lower than the membrane cut-off are able of passing the membrane. However, even if the molar mass falls below the molecular weight cutoff, an acceptable relative
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 55
recovery will only be attained with substances having a molar mass lower than approximately one-fourth of the membrane cutoff [41]. The size of the microdialysis membrane influences the relative recovery. According to Fick’s law, perfusion across a membrane is proportional to its area. Increasing the length of the microdialysis membrane will lead to an increase in relative recovery [41]. Increasing the outer diameter of the inner cannula may also enhance relative recovery of the probe [57, 58]. An important key is that the membrane does not interact with the surrounding tissue or with the perfusate. Low recoveries of acid aminoacids have been described due to the presence of surface charge [59]. Recovery of neuropeptides can vary as much as 20% with different dialysis membranes [60]. Different perfusion media have been used in microdialysis experiments and they vary widely in their composition and pH. Composition, ion strength, osmotic value and pH of the perfusion solution should be as close as possible to those of the extracellular fluid of the dialyzed tissue. In most experiments, the perfusate is an aqueous solution of sodium and potassium salts and other ions in a minor proportion, without proteins or a very small concentration of them. Addition of dextran to the perfusion solution prevents fluid loss. Rosdahl et al. [61] have demonstrated that the estimated concentrations of dextran in the perfusion solution at which no net loss of perfusion fluid occurred produces a colloid osmotic pressure similar to the reported values for plasma. This implies that the plasma colloid osmotic pressure contributes to a mass transfer of fluid from microdialysis catheter to capillaries. In some cases, proteins should be added to the perfusion medium to prevent sticking of drugs to the microdialysis probe and tubing connection [62]. When microdialysis probes are implanted into a tissue, it is necessary to keep asepsis in the area immediately around the site of the insertion. Huff et al. [63] reported a preliminary evaluation of several disinfection/sterilization techniques used with experimental microdialysis probes. These authors found that two disinfection methods, 70% ethanol and a commercial contact lens solution, and two sterilization methods, hydrogen peroxide plasma, and e-beam radiation, did not affect the functionallity of the probes. However, hydrogen peroxide plasma and contact lens solution groups reduced extraction efficiencies of microdialysis probes. Gamma irradiation has also been used for sterilization of microdialysis probes [64]. Probes were sterilized by gamma irradiation (32 kGy) prior to use. Subsequent in vitro experiments confirmed that the irradiated probes behaved identically to native probes from the supplier. In this work, no sign of infection at the implantation sites was found, probably due to the strict attention to sterility during the probe implantation as well as to prophylactic treatment with a broad-spectrum antibiotic [64]. However, one of the most popular methods to sterilize microdialysis probes is the gas sterilization method with ethylene oxide [65, 66]. An important aspect of microdialysis in pharmacokinetic studies is the selection of an adequate analytical method for drug concentration determination. Microdialysis generates small volume samples (1-10 l), because of the need of slow perfusion rates (0.12 l/min) to obtain high recoveries of the drug maintaining an adequate temporal resolution. In addition, microdialysis samples often contain the analyte at low concentration (pM-M range) needing of a high sensitive analytical method [29]. A wide range of analytical methods can be used for the analysis of microdialysis samples. Non-separation-based methods allow the detection of one analyte at a time, in
56 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
contrast to separation-based methods that can be used for the detection of multiple analytes (original drug and metabolites) in each sample. Therefore, its applicability for microdialysis studies is scarce [29]. Separation-based methods available for analysis of microdialysis samples include LC, microbore LC, capillary LC and CE. A wide range of different detectors can be coupled to the separation method. Immunoassay, ultraviolet absorbance, electrochemical detection, fluorescence and mass spectrometry (MS) are the most common detectors for LC analysis [29]. For CE, electrochemical and laser induced fluorescence detection have been most commonly employed [29]. The drawback of the separation-based methods is the dilution of the microdialysis sample. Nevertheless, LC with tandem MS/MS detection, due to its low limit of quantification, is a powerful analytical technique for quantification of drug concentrations in microdialysis samples. In comparison to the non-separation-based methods, these methods allow the detection of many compounds in each sample allowing simultaneous quantification of drug metabolites. The great diversity of detection methods provides to the researcher the chance to select the more adequate analytical system for the determination of the compound of interest [67]. The most important aspect of the analytical system is its sensitivity in order to improve the temporal resolution of a microdialysis protocol. Importance of Microdialysis Probe Calibration Use of microdialysis sampling for determination of target site concentrations of drugs needs accurate calibration of the microdialysis probe, considering that the desired information is the absolute drug concentration in the tissue. The relationship between concentration of the drug in the dialysate and its concentration in the sample matrix may be thought of as the recovery of the probe. Several factors influence recovery of the microdialysis probe, including temperature, perfusate flow rate, chemical and physical properties of the dialysis membrane, probe geometry, membrane surface area, chemical properties of the drug and the diffusion rate of the drug within the matrix [68]. All of these parameters remain constant during the experiment under normal conditions. Few parameters, such as membrane surface area and perfusate flow rate, can be modified by the researcher to improve recovery of the microdialysis probe. If high microdialysis probe recovery is needed to obtain concentrated samples, the researcher can opt among increasing the size of the surface area or decreasing the perfusate rate. However, increase of membrane surface area implies loss of spatial resolution and decrease of the perfusate rate implies loss of temporal resolution, because of prolongation of the sampling interval [29]. Recovery of the microdialysis probe can be determined by both in vitro and in vivo assays. The in vitro estimation of the probe recovery, serves only to prove if the microdialysis probe works, since in vitro recovery values are often an overestimation of the in vivo recovery [29]. In vivo recovery of the microdialysis probe can be determined through different methods, including flow-rate or stop- flow method [69], zero-net-flux method [45], retrodialysis method [70] and delivery method [71]. Because the zero-net-flux method and the flow rate method require that examination of the study subject under steady state conditions prior to the experiment, the total study time is extended. This limits the application of these methods for pharmacokinetics purposes [29].
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 57
Determination of the in vivo recovery with the retrodialysis or delivery method is more appropriately. Both methods are based on the principle that recovery is independent of the direction of analyte diffusion. Therefore, loss of the analyte from the perfusion media to the extracellular matrix should be equal to gain of the analyte from the tissue to the perfusate. This equality must be determined by the determination of in vitro recovery of the analyte both by loss and gain methods [29]. The retrodialysis method consists in the diffusive loss of molecules of a calibrator from the perfusate into the environment surrounding the probe. The dialysance or permeability-area product (PeA) of the calibrator must be similar to that of the compound of interest [29]. The in vivo recovery of the calibrator is calculated with the following equation: R= (Cin - Cdial)/ Cin where R is the calibrator in vivo recovery, Cin is the concentration of the calibrator in the perfusate and Cout is the concentration of the calibrator in the dialysate. Unbound concentrations of the compound of interest in the surrounding environment (C) are calculated using the following equation: C = Cout / R where Cout is the compound of interest concentration in the dialysate and R is the in vivo recovery of the microdialysis probe. The delivery method consists in the determination of the in vivo recovery of the compound of interest by retrodialysis before the actual experiment. In vivo recovery of the compound of interest is determined before drug administration by perfusing microdialysis probe with a solution of the compound of interest, taking the proportion of loss across the dialysis membrane as an estimate of the recovery [29]. Comparing the last two methods, a limitation of the retrodialysis is the necessity of an analyte with similar physico-chemical characteristic as the compound of interest. On the other hand, if the analyte is added to the perfusate at high concentrations, it could introduce unwanted perturbations in the surrounding tissue area [29]. The delivery method overcomes the uncertainty introduced by using a standard to mimic the compound of interest. The shortcoming of this approach is that recovery changes resulting from the experiment are not detected. However, recovery of the microdialysis probe generally remains constant in pharmacokinetic studies [29]. Another issue of microdialysis sampling in pharmacokinetic studies is the fact that microdialysis generates data that are the integral of the concentration surrounding the probe during the sampling interval, conversely to tissue samples, which provides a point measurement during the same interval. To estimate pharmacokinetic parameters, microdialysis data must to be transformed from a series of integrals to a series of points corresponding to the same time of tissue samples collection [72]. CURRENT APPLICATIONS OF MICRODIALYSIS SAMPLING FOR EVALUATION OF TISSUE DISTRIBUTION AND PK-PD MODELING OF DRUGS Microdialysis sampling has been extensively used for the study of target site distribution and PK-PD modeling of established drugs and new chemical entities in both pre-
58 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
clinical and clinical protocols (for review see [12, 29, 37, 41]). Although pharmacokinetic properties and PK-PD relationship of different therapeutic groups have been evaluated by means of microdialysis sampling, most of microdialysis studies addressed drug distribution and PK-PD modeling of antiinfective, antineoplastic and central acting drugs [12, 37]. Microdialysis sampling is also useful for PK-PD modeling of cardiovascular drugs in laboratory animals and for the evaluation of dermal pharmacokinetics of drugs after its topical application [12, 36]. In this section, most important microdialysis studies in this area are discussed. Antiinfective Drugs Traditionally, pharmacokinetic assessment of antimicrobial agents was largely based on the measurement of total plasma concentrations. Nevertheless, the use of plasma antibiotic levels is not ideal, because most infections occur in tissue sites, and therefore the ability of antibiotics to reach the target site is a key determinant of clinical outcome. It is considered that total plasma concentrations and plasma protein binding can be used to predict free tissue levels of antibiotics, based on the assumption that unbound plasma concentrations and free tissue levels are equal at equilibrium, considering that tissue distribution is generally mediated only by passive diffusion. However, many studies have shown lower tissue unbound levels than plasma concentrations [14-17]. As discussed above, tissue distribution is also affected by anatomic barriers, presence of active transport systems and tissue metabolism. On the other hand, time to reach equilibrium between plasma and tissue concentrations of antibiotics may range from minutes to days [73]. Therefore, measurement of unbound drug concentrations in the interstitial fluid of the target tissue should be considered a gold standard for improvement of antimicrobial therapy and dose adjustment. Microdialysis has been used to measure various antimicrobials agents in human and laboratory animal tissues, including aminoglucosides, penicillins, cephalosporines, fosfomycin, fluoroquinolones and antiviral agents (for review see [74-76]). These studies have served to evaluate drug distribution in several organs, including infective tissues, and to develop in vivo PK- in vitro PD models at the target site using the same parameters calculated in plasma: time (T) above the minimum inhibitory concentration (MIC) (T>MIC), the ratio of the maximum concentration of drug in serum (Cmax) to the MIC (Cmax/MIC), the area under the inhibitory curve or the area under the curve (AUC)/MIC ratio [76]. Using microdialysis sampling, the relevance of protein binding in therapeutic failure was demonstrated. A relationship was found between the failure in cefoperazone treatment in serious illness and the degree of drug bounded to proteins [77]. In addition, cefpodoxime, with lower protein binding than cefixime (25 vs 65%) showed higher peak concentration and tissue penetration than cefixime [78], suggesting a favorable efficacy of cefpodoxime, and this is supported by clinical trial data provided by a study in paediatric acute otitis media [79]. Microdialysis sampling was also used for the determination of antimicrobial agents in the interstitial fluid of non-infected soft tissues of healthy volunteers and rats. Several studies have demonstrated that tissue concentrations of different antibiotics, including ertapenem [80], gemifloxacin [81], telithromycin [82], and imipenem [83] among others, are similar in soft tissues with regards to unbound plasma concentrations, indicating a good distribution of antimicrobial in non-infected tissues.
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 59
In addition, drug distribution of antimicrobial drugs in infected tissues was also evaluated by means of microdialysis sampling both in basic research and clinical setting. Experiments conducted in rats showed that lung and peritoneal penetration of imipenem is not reduced in rats with pneumonia and peritonitis, respectively [84, 85]. In another study, lung penetration of meropenem was determined in patients with pneumonia and metapneumonic pleural emphysema by means of microdialysis [86]. Although meropenem rapidly penetrated infected lung tissue, interstitial lung fluid levels were lower than serum concentrations with target site drug concentrations above the MIC90 threshold for many clinically relevant pathogens for up to 6 hours [86]. In another clinical protocol, moxifloxacin exhibited similar penetration into the interstitial space fluid in normal subcutaneous tissue and infected decubitus ulcers in patients with spinal cord injury [87]. Conversely, penetration of the fluorquinolone levofloxacin into tissues appears to be unaffected by local inflammation. Bellman et al. [88] observed that administration of a standard dose of levofloxacin reached adequate levels at target site, although the extent of tissue penetration showed a high interindividual variability. Zeitlinger et al. [89] have also demonstrated a reduced distribution of levofloxacin in infected lung with regards to interstitial fluid of non-infected soft tissues obtained from healthy volunteers. Microdialysis sampling was also extensively employed for PK-PD modeling of antiinfective agents (for review see [75, 76]). A three step approach has been used for the in vivo PK- in vitro PD modeling by means of microdialysis. Firstly, interstitial fluid concentrations of the antibacterial drug at the target site are measured by means of microdialysis. Secondly, time versus drug concentration profile measured in vivo is simulated in an in vitro setting on bacterial cultures. Thirdly, unbound antibiotic concentrations are linked to bacterial kill rates by means of a PK-PD model [90]. Delacher et al. [90] have demonstrated a significant correlation between the maximal bactericidal effect and several pharmacokinetic surrogate parameters, such as AUC/MIC, Cmax/MIC and T>MIC. The authors concluded that the therapeutic success or failure in antibacterial therapy depends on the target site concentrations of the antimicrobial agent. Moreover, in vivo PK- in vitro PD modeling provides valuable guidance for drug and dose selection of antibacterial drugs [90]. In vivo PK – in vitro PD modeling of antimicrobial drugs was also studied in critical ill patients by means of microdialysis. Zeitlinger et al. [91] have applied an in-vivo pharmacokinetic/in-vitro pharmacodynamic method to simulate bacterial killing in plasma and the interstitium of skeletal muscle tissue after intravenous administration of cefpirome and fosfomycin alone and in combination to patients with sepsis. The in vitro simulation of in vivo plasma and tissue pharmacokinetics of cefpirome and fosfomycin has shown that both antimicrobial agents kill Staphylococcus aureus and Pseudomonas aeruginosa strains after single dose administration, observing a synergic antimicrobial effect by the combined use. Therefore, this data confirms antimicrobial strategies of simultaneous administration of cefpirome and fosfomycin in patients with severe soft tissue infection [91]. It must be pointed out that all PK–PD studies of antimicrobial drugs by means of microdialysis have used a combined in vivo PK–in vitro PD simulation without applying mathematical PK–PD models in their analysis. Recently, (Liu et al.) [92] demonstrated that a PK–PD model based on unbound antibiotic concentrations at the site of infection,
60 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
and a sigmoid Emax relationship, effectively described the antimicrobial efficacy of both cefpodoxime and cefixime. This approach offers a more detailed information than the MIC does about the time course of antibacterial efficacy of antibiotics [92]. In this way, (Huang et al.) [93] have evaluated the simultaneous estimation of amoxicillin influx/efflux in chinchilla middle ear fluid and its antibacterial effect. The authors found that the microdialysis procedure did not interfere with the bacterial growth-kill profile, thereby enabling pharmacokinetic and pharmacodynamic evaluation concomitantly. In summary, several microdialysis studies have demonstrated that anti-infective levels at the target site are sub-inhibitory, although effective concentrations are attained in plasma. Therefore, study of target site concentrations and PK-PD modeling of antiinfective agents in clinical drug development and in critically ill patients by means of microdialysis should enhance knowledge on adequate antibiotic drug dosing and improve patient's outcome. Preliminary studies demonstrated that microdialysis sampling did not interfere and therefore might be an interesting approach for the simultaneous estimation of target site concentrations of the antimicrobial agent and their in vivo antibacterial effect. Antineoplastic Drugs Tumor drug exposure, a marker linked to clinical outcome, may be dramatically reduced due to diffusion barriers in solid tumors [27]. Differences in tumor drug distribution do not allow to predict the antineoplastic response from plasma profiles [94], thus measurements of drug exposure into tumor interstitium by microdialysis may help to develop clinical PK-PD models with the aim of individualize drug therapy [95]. Microdialysis has been employed for the characterization of different antineoplastic drugs. Methotrexate, cisplatin, capecitabine, 5-fluorouracil, dacarbacine and melphalan have been measured into the tumor using clinical microdialysis in several types of malignancies such as breast cancer, melanoma, osteosarcoma and malignant fibrous histiocitoma (for review see [95, 96]). Measurements of plasma drug concentration have been employed during high-dose methotrexate treatment [97]. Although this practice has served to identify individuals with impaired renal function in order to avoid serious side effects, it assumed a relationship between pharmacological response and circulating levels of drug. Müller et al. [98] have studied interstitial tumor pharmacokinetics and plasma-to-tumor transfer rates of methotrexate in breast cancer patients by insertion of microdialysis probes into the primary tumor and the periumbilical subcutaneous adipose layer of previously chemotherapy-naive breast cancer patients. Absence of correlation between plasma AUC and the AUC in the interstitial space of tumor tissue was found, together with a high interindividual variability in transendothelial methotrexate transfer. Therefore, plasma levels of methotrexate were not predictive of intratumor levels. Similar results were obtained for 5-fluorouracil (5-FU), showing that plasma or subcutaneous levels of 5-FU failed to predict tumor response. Conversely, high interstitial tumor concentrations of 5-FU were associated with increased tumor response. This information may explain drug resistance in some patients and help to optimize dosing and administration schedules [99].
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 61
Microdialysis has also been used to assess the pharmacodynamics of chemotherapeutic agents [96]. Castejon et al. [66] have determined plasma concentrations of serotonin and 5-hydroxyindoleacetic acid during cisplatin treatment by means of microdialysis. Serotonin is involved in the production of emesis associated to antineoplastic treatment. Microdialysis has also been used for the monitoring of extracellular levels of growth factors, such as the vascular endothelial growth factor (VEGF), during treatment with tamoxifen in a mouse model of human breast cancer [100]. Although to date scarce PK-PD modeling studies applying microdialysis were made, integration of the pharmacological response with tumor PK profiles of the corresponding drug would help to define PK-PD relationship, which is essential for the rational design of drug administration regimens in cancer patients. Thompson et al. [101] studied the clinical and biochemical responses to the time course of melphalan in the subcutaneous interstitial space and in tumor tissue from patients with various limb malignancies. The authors showed a significant correlation between the melphalan mean concentration in subcutaneous microdialysate and tumour response [101]. In another PK-PD protocol, (Müller et al.) [102] have determined the unbound interstitial drug pharmacokinetics of 5-fluorouracil and methotrexate in solid tumour lesions of patients by means of in vivo microdialysis. The authors then made a pharmacodynamic simulation of the time versus drug concentration profile in an in vitro setting by exposing breast cancer cells to interstitial tumour concentration of the antineoplastic drugs. The authors concluded that in vivo PK–in vitro PD models might provide a rational approach for describing and predicting pharmacodynamics of cytotoxic drugs at the target site [102]. In summary, high interindividual variability in intratumoral drug distribution exists, indicating lack of correlation between plasma concentration of antineoplastic drugs and their interstitial tumoral levels. Plasma measurements do not serve as surrogates for intratumoral concentration and microdialysis may help to design optimal treatment schedules and to select appropriate drug, doses and dosing intervals for antineoplastic agents. Surprisingly, in the last three years, only few clinical microdialysis protocols were conducted for the study of intratumoral distribution of anticancer drugs. A possible explanation is that microdialysis experiments in cancer patients must be conducted in strict compliance with regulatory demands and need to be based on appropriate ethical conditions. In addition, there are other factors that limit the use of microdialysis in cancer research. Puncture of solid tumors by microdialysis catheter implantation may induce metastasis. However, the estimated incidence of metastasis by puncture ranged from 0.003% to 0.005 % and there is no evidence that puncture of tumor lesions affected the course or prognosis of the underlying disease [103]. Another limitation of the applicability of microdialysis sampling in oncology is the fact that the majority of the antineoplastic drugs act within cells. The relationship between extracellular drug concentrations and intracellular drug levels remains unknown. This drawback could be overcome by applying an attractive approach consisting in the simultaneous study of drug distribution with microdialysis and PET. PET measures total (intracellular, extracellular, and intravascular) concentrations of radiolabeled drugs in tissue, and microdialysis determines unbound drug concentrations in the extracellular space fluid of tissue. Therefore, combination of both techniques allows the description of the intracellular drug pharmacokinetics. Langer et al. [104], using 18F-labeled-cipro-
62 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
floxacin as a model drug, have found that in vivo intracellular ciprofloxacin pharmacokinetics was in accordance with previous in vitro data describing cellular ciprofloxacin uptake and retention. Therefore, PET/microdialysis combination might be useful during research and development of new drugs, for which knowledge of intracellular concentrations is of interest. Finally, some antineoplastic drugs, such as 5-fluorouracil, requires intracellular enzymatic conversion in order to exert their cytotoxic activity. In addition, other aspects like tumor location and accessibility for microdialysis probe implantation, and the possibility of variation in interstitial concentrations of cytotoxic drugs in different metastases in a patient restrict the utility of microdialysis for studies of antineoplastic drug distribution [12]. Central Acting Drugs Knowledge of brain uptake is an important issue for drugs acting in the central nervous system, such as anticonvulsivants, antidepressants, anaesthetics and anticancer drugs. Concentration-time profile of the drug in the central nervous system determines the intensity and duration of the effects of a central acting drug. As in the clinical setting, the measurement of brain concentrations of the drug is highly restricted; alternative determination of lumbar or ventricular cerebrospinal fluid is sometimes used. Nevertheless, drug concentrations in the cerebrospinal fluid provides only limited information with respect to drug distribution into the brain, because brain parenchyma distribution of a central acting agent is determined by multiple factors, including active biotransformation, active transport at the BBB and intracellular-extracellular exchange [105]. Intracerebral microdialysis would be a significant improvement, considering that it allows the study of the pharmacokinetic profile of the unbound drug fraction at a specific region within the brain. However, as discussed previously, microdialysis is a semi-invasive technique and may bear risks. Therefore, microdialysis sampling is restricted in the clinical setting for the study of brain drug distribution used in critical care patients subjected to surgery. Another drawback of microdialysis for the study of central acting drugs is the fact that this technique records the average drug concentration over a sampling interval of at least 5 min, being this sampling frequency not suitable to assess central pharmacokinetics of drugs that rapidly penetrate the brain (e.g. anesthetic induction agents) [106]. Nevertheless, microdialysis can be used in the early phase of clinical trials to establish brain penetration of drugs. A phase II trial of topiramate in severe traumatic brain injury demonstrated different free-drug concentrations in the extracellular space from that measured in cerebrospinal fluid (CSF) [107]. An interesting field of intracerebral microdialysis in the clinical setting is the study of brain distribution of anticonvulsivant agents. Despite the existence of a large variety of antiepileptic drugs, almost 30% of epileptic patients are resistant to treatment [108]. There is increasing evidence that over-expression of multidrug transporters such as Pglycoprotein is involved in the generation of pharmacoresistance to antiepileptic drugs because of a greater efflux from the brain of the central acting agent [24]. Therefore, monitoring of target site concentration of antiepileptic drugs by means of intracerebral microdialysis could elucidate the involvement of the efflux transport in the generation of the pharmacoresistance to antiepileptic drugs.
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 63
In the clinical setting, microdialysis protocols have demonstrated that intracerebral concentrations of valproic acid [109] and phenytoin [110] were slightly lower than plasma levels, suggesting that both anticonvulsivant drugs may be substrate for an energy-dependent carrier transport out of the central nervous system. Conversely, carbamazepine and carbamazepine-10,11-epoxide concentrations in the extracellular brain fluid closely mirror their unbound serum concentrations [111]. Intracerebral microdialysis was also used for measurement of brain concentration of other central acting drugs. Ederoth et al. [112] studied the BBB transport of morphine in patients with severe brain trauma by simultaneous microdialysis of morphine levels in "better" and "worse" brain tissue. The authors found that unbound brain morphine levels were lower than plasma concentrations suggesting the existence of an efflux transport system for morphine across the human BBB. In addition, the results suggested an increase of BBB permeability to morphine in the "worse" brain tissue. Brain penetration of anticancer drugs was also evaluated using microdialysis sampling in basic research. Zhuang et al. [113] have shown that P-glycoprotein and breast cancer resistance protein 1 (BCRP-1) limit topotecan distribution in brain parenchyma. Microdialysis sampling also allows the study of drug distribution to brain tumors. A significant increase in the penetration of methotrexate into the tumor tissue was found [114]. A hybrid physiologically-based pharmacokinetic model was used by the authors to characterize the mechanisms that may be responsible for the high methotrexate tumor concentration. Although microdialysis sampling is restricted in the clinical setting, this membranebased sampling technique is highly attractive for the preclinical evaluation of central acting drugs, considering the possibility of the simultaneously monitoring of brain drug concentrations and their effect on neurotransmission. Both traditional neurotransmitters and neuropeptides have a low molecular weight and diffused through the membrane of the probe, and therefore microdialysis allows the evaluation of drugs actions on biophase concentration of virtually all neurochemical substance [29]. Intracerebral microdialysis sampling has been extensively used for the study of antiepileptic drug concentrations at the target site in non-epileptic animals and experimental models of refractory epilepsy. Involvement of drug efflux transporters in brain parenchyma distribution of antiepileptic drugs was demonstrated in non-epileptic animals by enhancement of drug central levels by the application of drug efflux transporters inhibitors [24, 25]. Only few studies evaluated central pharmacokinetics of antiepileptic drugs in refractory epilepsy models with conflicting results. Whilst Rizzi et al. [115] using the Kainate model of epilepsy showed that overexpression of P-glycoprotein in the hippocampus is associated with reduced brain concentrations of phenytoin compared with control animals, Potschka and Löscher [116] did not find altered levels of phenytoin in hippocampus and amygdala of kindled when compared to non-kindled rats. Recently, we have demonstrated a critical role of P-gp overexpression in the development of pharmacoresistance to phenytoin in a model of epilepsy induced by 3-mercaptopropionic acid chronic administration, suggesting that administration of efflux transporters inhibitors could be an effective strategy to decrease pharmacoresistance to phenytoin antiepileptic treatment [117]. Comparing pharmacokinetics of anticancer agents in tumor and non-tumor extracellular fluid of brain might be an attractive approach to establish the therapeutic potential
64 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
of antineoplastic therapies. Apparaju et al. [118] have found, by means of intracerebral microdialysis, that gemcitabine levels in the tumor extracellular fluid was 2.2-fold greater than the corresponding value in the tumor-free extracellular fluid of the brain, suggesting that the greater tumoral distribution of gemcitabine region may facilitate selectively higher cytotoxicity against brain tumor cells. Pharmacodynamics and PK-PD modeling of antidepressant drugs have also been evaluated using intracerebral microdialysis. In an elegant study, (Bundgaard et al.) [119] validated an integrated microdialysis rat model for multiple pharmacokinetic/pharmacodynamic investigations of serotonergic agents, such as antidepressant drugs. Concomitant with brain microdialysis, serial blood sampling was conducted by means of an automated blood sampling device. Pharmacokinetics of escitalopram have been characterized simultaneously in plasma and the hippocampus of conscious, freely moving rats. Concomitantly, the modulatory and functional effects of escitalopram could be monitored as increases in brain 5-HT and plasma corticosterone levels following drug administration. Therefore, microdialysis sampling allows simultaneous monitoring of hippocampal escitalopram levels and their effect on serotonin extracellular concentrations and therefore the assessment of PK-PD modeling of neurochemical actions of antidepressant drugs [119]. Neurochemical effects of desvenlafaxine, a novel antidepressant drug, on cathecholamine hypothalamic levels were also studied in freely moving rats by microdialysis. Administration of desvenlafaxine induced an increase in noradrenaline levels in hypothalamic dialysate without any effect on dopamine and serotonin concentrations, suggesting its utility in a variety of central nervous system-related disorders [120]. In another report, using in vivo microdialysis, the protective effect of NGP1-01, a dual blocker of neuronal voltage- and ligand-operated calcium channels, was evaluated by monitoring choline release during N-methyl-D-aspartic acid (NMDA) infusion as a measure of excitotoxic membrane breakdown [121]. Intraperitoneal administration of NGP1-01 reduced NMDA-induced membrane breakdown demonstrating that NGP1-01 simultaneously blocks both major neuronal calcium channels and is sufficiently brainpermeable. Therefore, NGP1-01 is a promising lead structure for a new class of dualmechanism neuroprotective agents. Microdialysis was also used to describe, by means of PK–PD modeling, the effect of drug candidates on dopaminergic activity at different nuclei of the central nervous system (CNS) [122, 123]. The effect of benzatropine analogues on dopamine concentration in the nucleus accumbens after its intravenous administration was evaluated [123]. The authors fitted plasma concentration of the analogues and their effects on extracellular dopamine levels to two different PK–PD models, such as an effect compartment model and a model with indirect physiological response. The authors demonstrated that the indirect model is more suitable for PK–PD modeling of benzatropine analogues than the linked PK–PD model. These results are in accordance with the mechanism of action of the analogues because these drugs bind to the dopamine transporter inhibiting the dopamine re-uptake and consequently elevate dopamine extracellular levels. Brain microdialysis has also been used for PK-PD modeling of therapeutic agents at the central nervous system. PK–PD modeling allows the study of the mechanism responsible for the time delay of central actions of drugs. In an elegant study, (Bouw et al.) [124] have simultaneously determined blood and brain concentrations of morphine-6-
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 65
glucuronide and its antinociceptive effect by means of microdialysis sampling. By applying a PK–PD model with an effect compartment, the authors found a greater delay in the onset of the effect when antinociceptation was related to plasma morphine-6glucuronide concentrations with regard to brain levels. Therefore, it was concluded that half of the effect delay could be explained by transport across the blood–brain barrier, suggesting that the remaining delay is a result of drug distribution in the brain parenchyma [124]. Microdialysis was used to describe the relationship between norfloxacin concentrations in the CNS and its adverse reactions, such as convulsive effect [125]. Brain extracellular concentrations of norfloxacin by means of microdialysis and a quantitative electroencephalogram (EEG) were simultaneously determined. Blood samples were also collected to determine norfloxacin plasma levels. Although norfloxacin brain concentrations peaked early after its intravenous administration, the effect on the EEG measurement was delayed. By applying a PK–PD model with an effect compartment, the authors demonstrated that the delayed EEG effect of norfloxacin is not due to BBB transport [125]. In conclusion, microdialysis sampling is a promising technique for drug development of central acting drugs, especially in the preclinical phase. The most attractive application of brain microdialysis seems to be the evaluation of brain parenchyma distribution of developing antiepileptic drugs and its ability to overcome pharmacoresistance in chronic epileptic animal models. In addition, impact of therapeutic strategies to enhance brain penetration of antiepileptic, such as efflux transporters inhibitors, could also be evaluated during preclinical phase of drug development by means of microdialysis sampling. Antihypertensive Drugs A poor concentration-response relationship with regard to the blood pressure effect of antihypertensive drugs has been found. Taking into account the pharmacodynamic properties of antihypertensive drugs, the suggestion of absence of relationship between plasma levels of antihypertensive drugs and its blood pressure lowering effect could reflect an inadequacy or failure in the approaches designed to detect such correlation. A number of factors have hampered the possible identification of a correlation, including failure to study individual patients, inability to collect sufficient pharmacodynamic data, failure to identify and account for temporal delay in the onset of the pharmacological effect, the use of restricted concentration ranges and the use of dose rather than concentration [126, 127]. Using a ‘shunt’ intra-arterial microdialysis probe, a good relationship was found between metoprolol concentration in the effect compartment and its hypotensive and chronotropic effect [128-130]. Moreover, the maximal response was significantly greater in hypertensive animals, such as spontaneously hypertensive rats and animals with aortic coarctation, with regards to their respective control animals. Therefore, this data suggests that the proposed lack of relationship between plasma levels of -blockers and its antihypertensive effect is probably a consequence of an inadequate experimental design and data analysis. More recently, by using in vivo microdialysis, we have demonstrated that the modified Emax designed by Schoemaker et al. [38] allows an accurate and precise estimation
66 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
of diltiazem and verapamil sensitivity to the hypotensive effect in conditions when maximal pharmacological response can not be attained [131, 132]. In conclusion, microdialysis sampling is a powerful tool for PK–PD modeling of cardiovascular drugs in basic research, taking into account that it allows continuous and simultaneous determination of antihypertensive drugs plasma levels and their corresponding effect on blood pressure and heart rate in the same animal. Moreover, PK–PD modeling not only allows a better pharmacodynamic characterization of blood pressure lowering agents, but also permits the study of the physiopathological mechanisms of the hypertensive stage in different experimental models. APPLICABILITY OF PET FOR TISSUE DRUG DISTRIBUTION STUDIES AND PK-PD MODELING Principles of PET Imaging In the last years, several imaging techniques have been developed for the evaluation of the pharmacokinetic and pharmacodynamic properties of therapeutic agents. PET, SPECT and MRS are the most popular imaging techniques used for this purpose [6]. However, SPECT and MRS have limited applicability in drug distribution studies because of the existence of technical drawbacks, including low sensitivity, impressive quantification and low spatial resolution [133]. Conversely, the main problems inherent to SPECT and MRS are not important issues with PET, and therefore this imaging technique became the most applied method for non-invasive pharmacokinetic studies of drugs. Principles of PET have been recently reviewed in several works and therefore only a brief description will be included in the present chapter (for review see [133-135]). PET is based on the emission of positrons (positive electrons, b+) of radionuclides with low atomic number and relative proton excess. After positron emission, it rapidly loses kinetic energy and sometimes interacts with an electron, generating an annihilation event. In this event, a positron-electron pair is converted to energy, resulting in the generation of two photons that emitted in opposite directions. These photons are then sensed by detector pairs positioned at 180° to each other [133]. Measurement of tissue drug levels of drugs by means of PET imaging requires radiolabelling of the compound of interest with an isotope that emits radiation that can be detected by imaging. It is important to mention, that a strict condition of PET imaging is that radiolabelling of the drug does not modify the pharmacokinetic and pharmacodynamic properties of the pharmacological agent [133]. One strategy is the radiolabelling of atoms that form part of the native structure of the drug, such as carbon, hydrogen, nitrogen or oxygen. However, there are great limitations in the use of isotopes of these atoms for PET imaging of drugs. Firstly, drugs labeled with 14C or 3H generates negative electrons that only diffused few millimeters from their site of generation. Therefore, although these isotopes could be used for the evaluation of tissue distribution of drugs in small laboratory animals, their applicability in human studies is neglected [133]. Secondly, the half-life of the decay of 15O, 13N, 11C, is extremely short to perform pharmacokinetic studies of drugs using PET imaging. Conversely, 18F, 76Br and 124I have long half-life that allows their used for PET imaging of pharmacological agents. However, only few therapeutic agents contain one or more atoms of F, Br or I that could be subjected for radiolabelling [133].
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 67
Alternatively, positron emitting halogen isotopes could also be substituted for hydrogen of hydroxyl groups. Although these analogues are not guaranteed to retain pharmacological properties of the original drug, fluoride substitution are generally hlghly conservative and have shown to be useful probes in many situations [133]. It is important to take into account that radiolabelled compounds need to be synthesized, isolated, purified and formulated as a sterile, pyrogen-free solution within 2 halflife of the radionuclide. Therefore, drug precursors that can be labelled in a single step are needed for most therapeutic agents. In addition, PET imaging studies requires both a cyclotron and a PET scanner in close proximity [106]. Compared with traditional methods for tissue sampling of drug concentrations, PET imaging has several advantages, including its non-invasive nature, high sensitivity and its universal nature, considering that theoretically the availability of PET radionuclides of nitrogen, carbon and fluorides makes possible to prepare for almost any drug [133]. In addition, PET has a high spatial resolution, allowing determination of drug tissue distribution in small volumes of tissue and also permits differentiation of physiological alterations that occurs nanometers apart [106]. Nevertheless, PET imaging studies have some drawbacks that could limit its applicability for pharmacokinetic studies. As comment above, most radiolabelled drugs need to be prepared in close proximity to the imaging laboratory [106]. In addition, due to rapid decay of most radionuclides, the time interval of the pharmacokinetic study is limited. For instance, pharmacokinetic study needs to be completed in 40 min with tracers containing 13N, 80 min with 11C and 8 hours with 18F [133]. Another drawback of PET imaging is that this methodology did not discriminate within positron emission from the parent drug and their metabolites, and therefore quantification is only accurate if drug biotransformation is negligible [133]. Additionally, PET imaging monitors total drug tissue concentrations rather than extracellular levels. Although anatomical resolution of PET imaging is high, it is much lower than other imaging techniques, including computed tomography, MRI and ultrasound [133]. Applicability of PET Imaging for Pharmacokinetic Studies In the field of pharmacology, the most attractive application of PET is the noninvasive evaluation of tissue distribution of drugs, especially in the treatment of infection and cancer. For these studies, the drug is radiolabelled with a positron emitter, preferentially with 18F, and injected in the experimental subject. Thereafter, PET images are acquired at different time points and quantified by means of maps of radioactivity concentration in the tissue [133]. When negligible biotransformation of the drug occurs, regional levels of the parent drug could be obtained by simply dividing each pixel in the images by the specific activity of the radiotracer. Therefore, repeating this procedure at different time points after drug administration allows the estimation of classical pharmacokinetic parameters, including peak concentration, plateau concentration and area under the curve of the drug. In the following section, we reviewed the most relevant PET based pharmacokinetic studies of different therapeutic agents [133]. Antiinfective Agents PET imaging was firstly used to describe tissue distribution of erythromycin in infected tissues [136]. Radiolabelled erythromycin is obtained by reductive methylation of
68 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
N-demethyl-erythromycin with [11C]formaldehyde yielding a product with similar chromatographic properties. By using this radiotracer, (Wollmer et al.) [136] showed that drug distribution in infected lung tissue is similar to normal pulmonary tissue. Although radiolabelling of erythromycin with carbon isotopes allows a successfull estimation of eryhtromycin tissue distribution, it must be taken into account that most antimicrobial agents have a much slower tissue equilibration, requiring the design of radiolabelled agents with longer lived radionuclides than 11C [136]. For instance, the presence of fluorine atoms in the molecular structure of fluconazole made this drug an ideal candidate for PET imaging. PET imaging of fluconazole is also justified considering that the drug undergoes minimal hepatic biotransformation and therefore measurement of tissue radioactivity reflects levels of the parent drug [133]. In an interestingly preliminar study [137], 18F-labelled fluconazole was used to compare pharmacokinetic properties of a microdose of the agent with regard to a pharmacological dose. Coadministration of 18F-fluconazole with a pharmacological dose of the unlabelled parent drug shows a rapid equilibration and an uniform distribution in most organs of the rabbit [137]. Conversely, application of the labelled drug in a “carrier-free” solvent accumulation of fluconazole was decreased in the heart, spleen and muscle and increased in the liver. This study clearly demonstrates the limitation of microdosing in the human phase 0 of drug development [137]. Pharmacokinetic properties of fluconazole were also evaluated in healthy volunteers by means of PET imaging [138]. The administration of 18F-fluconazole with unlabelled parent drug showed a non-uniform tissue drug distribution with accumulation of the drug in the spleen and lower levels in the bone [138]. Therefore, PET imaging demonstrated that fluconazole, at a dose of 5 mg/kg, is effective in the treatment of candidiasis at the hepatoesplacnic and urinary tract, but required higher doses for treatment of infections of the bone and central nervous system [138]. Fluoroquinolones also contain fluorine atoms in their chemical structure, and therefore are candidates for radiolabelling and PET imaging for in vivo pharmacokinetic studies. Fischman et al. [139] studied fleroxacin pharmacokinetics after administration of [18F]fleroxacin in 12 healthy volunteers. The authors demonstrated that fleroxacin rapidly distributed in most tissues, including heart, liver, lung myocardium and spleen, achieving concentrations two-fold above the MIC for 90% of Enterobacteriaceae strains tested. Conversely, a limited distribution of distribution of [18F]fleroxacin into the brain was observed. Additionally, concentrations of fleroxacin in tissue were similar in males and females [139]. Langer et al. [140] also studied the pharmacokinetic properties of radiolabelled [18F]ciprofloxacin in patients with soft tissue infections. Distribution of [18F]ciprofloxacin increased in infected tissues with peak ratios between infected and uninfected tissue ranging from 1.8 to 5.5. Nevertheless, radioactivity was not retained in infected tissue withouth differences in the elimination half-life of [18F]ciprofloxacin when comparing infected and uninfected tissues. The authors concluded that pharmacokinetics of [18F]ciprofloxacin in infected tissue are governed by increased blood flow and vascular permeability due to local infection [140]. Previously, the same authors have shown, by using [18F]ciprofloxacin, high tissue levels of this fluoroquinolone in several tissues, including kidney, heart, spleen, liver, musclea and lung [141]. Whilst [18F]ciprofloxacin rapidly wash out from the kidney, heart and spleen, prolonged readiotracer retention was
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 69
observed in liver, muscle and lung. Brain radioactivity concentrations of [18F]ciprofloxacin were under the limit of quantification suggesting that ciprofloxacin is not suitable for the treatment of infection at the central nervous system [141]. Briefly, PET imaging was also used for the study of trovafloxacin pharmacokinetics, showing that this fluoroquinolone will be useful in the treatment of a broad range of infections at diverse anatomic sites [142]. Drugs Acting at the Central Nervous System PET imaging have also been used for the study of different pharmacokinetic and pharmacodynamic properties of central nervous system acting drugs. Applicability of PET can be divided in direct and indirect studies. In direct studies, dynamic imaging is performed after administration of the radiolabelled drug, obtaining as information the distribution of the drug into different nuclei of the central nervous system [133]. However, PET imaging is also a valuable tool for the estimation of parameters describing in vivo drug-receptor interactions. For these studies, PET protocols are much more complex than for the study of drug distribution. Experimental design depends on the pharmacodynamic properties of the drug under study. For instance, for drugs that interact with a single subtype of receptors, time-activity curves from a single or multiple administration of the drug can be analysed by compartmental modeling to estimate receptor density (Bmax), binding affinity (KD), or binding potential (Bmax/K D). Conversely, in the case of drugs that bind to multiple receptor subtypes, kinetic parameters for a specific receptor can be estimated by blocking other receptors with unlabelled ligands [133]. Central distribution of zolmitriptan has been studied by means of PET imaging, considering the lack of studies regarding central nervous system access of this drug [143]. The authors found that [11C]zolmitriptan rapidly reached therapeutic concentrations at the central nervous system after its intravenous application of the triptan. Therefore it could be concluded that zolmitriptan enters the brain parenchyma in humans, achieving an uptake rate and concentration compatible with a central mechanism of action [143]. Brain distribution of vinpocetine, a compound used in the prevention of cerebrovascular disease, was evaluated by means of PET imaging after intravenous administration of [11C]vinpocetine [144]. The drug rapidly reached the central nervous system and distributed heterogeneously in different brain nuclei, suggesting the presence of specific binding site. Additionally, the authors found that the brain regions which showed increased uptake in the human brain correspond to those in which vinpocetine has been shown to induce elevated metabolism and blood flow [144]. Nevertheless, the most attractive application of PET for drugs with an action on the central nervous system is the in vivo study of receptor binding interaction. In this regard, a large number of radiotracer with specific affinity for different subtypes of receptors have been developed and validated, allowing the evaluation of in vivo binding properties of different pharmacological agents, including antipsychotics, antidepressants, anxiolytics, antiepileptics, among others (for review see [145]). On the other hand, these radiotracers also allow the study of brain receptor distribution. Table 3 summarizes actually available radiotracers for the evaluation of in vivo binding of different therapeutic drugs.
70 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 3.
Höcht et al.
Radiotracers Designed for Study of Drug-Receptor and Drug-Transporters Binding Studies Radiotracer
[11C]NNC 112, [11C]SCH 23390
Target D1 dopamine receptor
11
D2 dopamine receptor
18
[ F]fallypride
D2/D3 dopamine receptor
[11C]CFT
Dopamine uptake trabsporter
[11C]carfentanil, [11 C]diprenorphine
Opiate receptor
[ C]racloporide
18
Opiate sigma-1 receptor
11
5-HT1A serotonin receptor
11
5-HT2A serotonin receptor
[ F]FPS [ C]WAY 100635, [18F]MPPF [ C]MDL 100907 11
[ C]DASB 11
Serotonin transporter 11
[ C]flumazenil, [ C]Ro15-4513 11
Benzodiazepine receptor
[ C]TMSX
A2A adenosine receptor
NCFHEB
Nicotinic acetylcholine receptors
18
[ F]MK-9470
CB1 cannabinoid receptor
11
Abreviattions: [ C]TMSX: [7-methyl-(11)C]-(E)-8-(3,4,5-trimethoxystyryl)-1,3,7-trimethylxanthine ([(11)C]TMSX); NCFHEB: Norchloro-fluoro-homoepibatidine; [11C]NNC 112: (+)-8-Chloro-5-(7-benzofuranyl)-7-hydroxy-3-[11C]methyl-2,3,4,5tetrahydro-1H-3-benzazepine; [11C]SCH 23390: ((R)-(+)-8-chloro-2,3,4,5-tetrahydro-3-methyl-5-phenyl-1H-3-benzazepin-7ol); [11C]CFT: 2ß-carbomethoxy-3ß-(4-fluorophenyl)tropane; [18F]FPS: 3-fluoropropyl-4-((4-cyanophenoxy)methyl)-. Piperidine; [11C]WAY 100635: N-[2-[4-(2-methoxyphenyl)-1-piperazinyl]ethyl]- N-(2-pyridyl)cyclohexanecarboxamide trihydrochloride; [18F]MPPF: 2'-Methoxyphenyl-(N-2'-pyridinyl)-p-18F-fluoro-benzamidoethylpiperazine; [11C]MDL 100907: ((R)-(+)-4 -(l-hydroxy-1-(2,3-dimethoxyphenyl)methyl)-N-2-(4-fluorophenylethyl)piperidine; [18F]MPPF: 2'-methoxyphenyl-(N-2'-pyridinyl)-p-18F-fluoro-benzamidoethylpiperazine; [11C]MDL 100907: (R)-(+)-4 -(l-hydroxy-1-(2,3-dimethoxyphenyl)methyl)-N -2-(4-fluorophenylethyl)piperidine; [11C]DASB: [(11)C]-3-amino-4-(2-dimethylaminomethyl-phenylsulfanyl)-benzonitrile.
Antineoplastic Agents PET imaging of antineoplastic drugs provides information regarding tumor pharmacokinetics and treatment response, making this sampling technique highly attractive both in drug development and clinical use of anticancer drugs (for review see 146-150]). PET can provide information that is often difficult to measure in the intact animal or patient. Several antineoplastic drugs were radiolabelled with the intention to study their tissue distribution and pharmacological response, including 5-fluorouracil (5-FU), N-[2(dimethylamino)ethyl]acridine-4-carboxamide (DACA), temozolomide, docetaxel, ifosfamide, gemcitabine, carboplatin, among others [133]. An example of the applicability of PET in drug development is the evaluation of tissue distribution properties of DACA in a pre-phase I study, using a radiotracer dose equivalent to 1/1000 of the phase-I starting dose [151, 152]. Radioactivity was shown to accumulate in vertebra < brain < tumour < kidney < lung < myocardium < spleen < liver. Tissue accumulation profile of DACA suggested that myelotoxicity and neurotox-
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 71
icity were less likely to be dose-limiting, and probably cardiotoxicity limited dose escalation in the clinical practice. PET study also showed a variable tumor uptake of [11C]DACA that was moderately correlated with blood flow [151-153]. Saleem et al. [154] evaluated the tumor, normal tissue, and plasma pharmacokinetics of temozolomide in vivo, and also determined whether such pharmacokinetics resulted in tumor targeting. Considering the postulation that temozolomide undergoes decarboxylation and ring opening in the 3-4 position to produce the highly reactive methyldiazonium ion that alkylates DNA, the authors used a dual radiolabelling strategy, with [11C]temozolomide separately radiolabelled in the 3-N-methyl and 4-carbonyl positions, to establish the mechanism of action. Both radiolabelled forms of [11C]temozolomide undergo rapid systemic clearance. A decrease in tissue exposure to [11C]temozolomide was also observed with [4-11C-carbonyl] temozolomide compared with [3-N-11Cmethyl]temozolomide [154]. Another interesting result of this study was the higher [11C]radiotracer and [11C]temozolomide exposure (AUC(0-90 min)) in tumors compared with normal tissue. The significantly higher amounts of plasma and exhaled [11C]CO(2), in addition to the lower normal tissue and tumor [11C]temozolomide AUC(0-90 min) observed with [4-11C-carbonyl]temozolomide, confirmed the mechanism of metabolic activation of temozolomide postulated by the authors. Considering these results, the authors concluded that the higher tumor [11C]temozolomide exposure compared with normal tissue and the tissue-directed metabolic activation of temozolomide may confer potential therapeutic advantage in the activity of this agent [154]. The work of (Saleem et al.) clearly demonstrates that PET imaging not only allows the noninvasive study of tumor durg distribution, but also permits the elucidation of the mechanism of action of anticancer agents. PET imaging also allows comparing drug uptake of primary tumors with regard to metastatic lesions. Jayson et al. [155], using iodinated radiotracer of HuMV833, a humanized anti-VEGF antibody, showed that primary tumors exhibited clearance rates that were three 3 times faster compared to metastatic tumors. The authors suggested that the heterogeneous tumor accumulation might explain the lack of response to antitumoral therapy in some patients. In addition, the study also found discrepancies between plasma and tumoral pharmacokinetics, indicating that traditional plasma sampling cannot be used to predict intratumoral drug concentrations [155]. PET imaging is also useful for monitoring antitumoral response to antineoplastic drugs. Tumor response to therapy has been assessed by determining the glycolytic response of tumors to chemotherapy with 18F-fluorodeoxyglucose (FDG) [147]. This tracer allows obtaining a mapping of tumor glucose use, and therefore establishing activity of malignant tumors. It was found that changes in FDG uptake by tumors after the first chemotherapy cycle are highly correlated with patient survival, suggesting that FDGPET might become an early indicator for treatment efficacy in drug development and clinical studies [156]. More recently, 18F-3-deoxy-3-fluorothymidine (FLT), a marker of DNA replication and cell proliferation, was used as an early imaging biomarker to assess antitumoral response in patients with glioblastoma [157, 158]. It was found that patients with an adequate metabolic response to bevacizumab and irinotecan regimen, defined as a decrease of at least 25% in FLT uptake by the tumor, showed a significantly longer survival than patient without response. Moreover, responders were identified only 2 weeks after the start of the therapy [157, 158].
72 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Höcht et al.
Utility of PET in PK-PD Modeling Although PET imaging has been extensively used for the study of tissue pharmacokinetics (eg antiinfective, antineoplastic and central acting drugs) and for monitoring drug-receptor interactions and pharmacological response, to the best our knowledge, this sampling technique has not been applied with the intention to integrate pharmacokinetic and pharmacodynamic concepts. A possible explanation for the lack of PK-PD modeling studies with PET imaging is the uncertainty of tissue concentration of the parent drug, considering that PET imaging did not allow discriminating between the original drug and its metabolites. Another limitation of PET for PK-PD studies is the impossibility to study multiple PET tracers in a single experiment. Therefore, this technique only allows the estimation of the pharmacokinetic or the pharmacodynamic component of a PK-PD modeling study. PRINCIPLES AND APPLICABILITY OF MRS FOR PHARMACOKINETIC STUDIES MRS is an alternative imaging technique to PET that allows discrimination between parent drug and metabolites. However, a strong limitation of MRS is its low sensitivity [133]. MRS employs magnetic field gradients to produce images generated by the relaxation of nuclei with magnetic spins. Considering that individual atoms of all compounds have specific resonance frequencies, it is theoretically possible to obtain specific MR images of any drug and drug metabolites [133]. However, considering its low sensitivity, in most applications of MRS images cannot be recorded, and therefore field gradients are used to select a specific tissue site for a spectrum record. Nevertheless, with this approach, sensitivity of MRS is much lower than PET imaging. MRS measurements can be performed serially, giving the possibility of pharmacokinetic analysis with a high temporal resolution [133]. Another advantage of MRS over PET imaging is that the number of tracers available for MRS studies is potentially unlimited. Isotopes with non-zero are available for all atoms in drugs, including 1H, 13C, 15N, 17O, 19F and 31P. Majority of MRS studies have been performed with 19F isotope, considering its high intrinsic sensitivity and zero natural background [133]. The existence of high background represents a limitation for the use of 1H.In addition, 13C and 15N show lower intrinsic sensitivity. MRS tracers are stable compounds and therefore the time interval of pharmacokinetic studies is unlimited with MRS. In other words, matching half-life of the tracer to the time scale of the pharmacokinetic study is not an issue with MRS [133]. In the next section, examples of MRS applications for pharmacokinetic studies of drugs will be described. Tumor distribution of several antineoplastic drugs has been evaluated using MRS imaging. One of the most anticancer drugs studied with this methodology is 5-FU. Estimation of tumoral half-life of 5-FU is highly important in anticancer therapy, considering that trapping of 5-FU is a necessary condition for pharmacological response [159]. It was found that patients with larger tumoral half-life of 5-FU assessed by MRS showed a greater pharmacological response. Therefore, 19F-MRS might become a useful technique for early detection of non-responders to 5-FU therapy [159]. 19 F-MRS ahs also been used for the study of liver and muscle distribution of fleroxacin in healthy volunteers [160]. Compared with PET imaging, this study has demonstrated that spatial resolution of MRS was low requiring larger volumes of tissue for
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 73
quantification of fleroxacin tissue levels [160]. Nevertheless, combining PET and MRS imaging could be an attractive approach, considering that PET imaging provides precise measurement of total drug levels in small volumes of tissue, and MRS can additionally discriminates the contributions of different molecular species (parent drug and metabolites) and differentiates between intra and extracellular drug concentrations. Whole brain concentrations of psychotropic drugs have been also monitored by MRS imaging. In a pioneer study, by using 7Li-MRS, it was found that a relative slow lithium accumulation may explain the delay in therapeutic response to antimaniac therapy in some patients [161]. In another work, (Renshaw et al.) [162] demonstrated that brain concentrations of fluoxetine and its active metabolite, norfluoxetine, were 2.6-fold higher with regards to plasma levels, which may have implications for understanding therapeutic and toxic effects of fluoxetine. CONCLUSIONS The development of new sampling techniques greatly enhances our current knowledge regarding tissue drug distribution and PK-PD models. Membrane-based sampling and imaging sampling techniques monitor with high spatial and temporal resolution drug levels at the target site. Also, these methodologies allow the simultaneous assessment of drug-receptor interactions and the effect of drugs on endogenous compounds. Therefore, considering that microdialysis and imaging techniques, especially PET, provide information about tissue pharmacokinetics and pharmacodynamics, they could be considered as the reference techniques in PK-PD modeling studies. Table 4 shows the most attractive application of microdialysis and imaging techniques for pharmacokinetic, pharmacodynamic and PK-PD modeling studies. However, as pointed out in the above mentioning sections, these techniques have some drawbacks that could limit their applicability. Whilst intracerebral microdialysis is restricted to use in critical care patients in the clinical setting, PET imaging does not allow discriminating between concentration of the parent drug and its metabolites. Nevertheless, it is important to take into account that microdialysis sampling and PET imaging are complementary techniques and the combination of the same could greatly improve the information gained in PK-PD studies. For instance, as comment above, whilst microdialysis sampling assesses extracellular tissue drug concentrations, PET imaging gives information regarding total tissue levels. Therefore, simultaneous microdialysis/PET studies allow a precise estimation of intracellular drug levels that may be highly relevant for drugs acting within the cellule. Another example of the benefits of microdialysis/PET combination is the assessment of PK-PD properties of drugs that affect central neurotransmission. As shown by Schiffer and collaborators [163], whilst PET imaging allows the study of the dopamine receptorsbinding properties of 11C-raclopride, microdialysis assesses the effect of the drug on dopamine extracellular levels. In addition, combining microdialysis sampling with PET imaging could be attractive for the evaluation of the impact of metabolite generation in PET. As comment above, PET imaging does not discriminate between parent drug and metabolites. Simultaneous intravenous microdialysis may give insights regarding drug biotransformation by assessing circulating levels of the drug and its degradation products.
74 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 4.
Applications of New Sampling Techniques for Pharmacokinetic and PK-PD Studies
Technique
Therapeutic Group
Antimicrobial agents
Applications
Microdialysis
Central acting drugs
Cardiovascular drugs Antimicrobial agents
PET imaging
Comments
Antibiotic penetration in non-infected soft Highly attractive for PKtissues PD modeling of antimicroDistribution of antimicrobial agents in infected bial agents tissues In vivo PK- in vitro PD modeling of antibiotics In vivo evaluation of prodrugs
Anticancer drug distribution in solid tumors Compartmental distribution within a tumor Antineoplastic Pharmacodynamics of antineoplastic drugs on drugs biochemical markers In vivo PK-PD modeling of anticancer agents
MRS imaging
Höcht et al.
Limited by the fact that most antineoplastic drugs act within cells. Tumor distribution is not homogenous Ethical considerations
Brain penetration of anticonvulsant agents Distribution of anticonvulsivants at the epileptic focus Evaluation of central pharmacokinetics of developmental anticonvulsivant drugs Evaluation of efflux transporters inhibitors as a therapeutic strategy to overcome antiepileptic drug pharmacoresistance In vivo PK-PD modeling of antiepileptic drugs In vivo evaluation of neurochemical effects of antidepressants and antipsychotic agents
Use of clinical microdialysis limited to critical care patients. Highly attractive for preclinical drug evaluation
In vivo PK-PD modeling of antihypertensive agents
Limited to preclinical drug evaluation
Antibiotic penetration in non-infected soft Limited to specific antimitissues crobial agents Distribution of antimicrobial agents in infected tissues
Anticancer drug distribution in solid tumors Antineoplastic Pharmacodynamics of antineoplastic drugs on drugs biochemical markers
Allows evaluation of heterogeneous drug distribution in different tumors
Central acting drugs
Estimation of in vivo drug receptor interaction Allows simultaneous monitoring of drug accuCentral distribution of drugs mulation in different nuclei
Antimicrobial agents
Antibiotic penetration in non-infected soft Limited to specific antimitissues crobial agents Distribution of antimicrobial agents in infected tissues
New Sampling Techniques
Frontiers in Drug Design & Discovery, 2009, Vol. 4 75 (Table 4) contd....
Therapeutic Group
Technique
Applications
Comments
Anticancer drug distribution in solid tumors Antineoplastic Pharmacodynamics of antineoplastic drugs on drugs biochemical marker
Only allows estimation of tumor half-life
Central distribution of drugs
Only allows estimation of whole brain distribution Differentiates between parent drug and metabolites
MRS imaging Central acting drugs
These few examples strength out the potential of combined microdialysis/PET for PK-PD studies. To the best our knowledge, no studies have applied simultaneous microdialysis/PET for the evaluation and development of mechanism-based PK-PD model. Nevertheless, considering that the combination of these sampling methodologies allow the simultaneous assessment of target site drug distribution and receptor binding properties, simultaneous microdialysis/PET may became an attractive approach for mechanism-based PK-PD modeling. In conclusion, membrane-based and imaging techniques, both used alone or simultaneously, will became the standard methodology for PK-PD modeling, enhancing gain of information during both drug development and clinical practice. ABBREVIATIONS 5-FU
=
5-Fluorouracil
AUC
=
Area under the curve
BBB
=
Blood-brain barrier
CE
=
Capillary electrophoresis
Cmax
=
Maximum concentration of drug in serum
CNS
=
Central nervous system
DACA
=
Dimethylamino)ethyl]acridine-4-carboxamide
EC50
=
Effective concentration to yield half-maximal response
EEG
=
Electroencephalogram
Emax
=
Maximal efficacy
FDG
=
18
F-fluorodeoxyglucose
FLT
=
18
F-3-deoxy-3-fluorothymidine
LC
=
Liquid chromatography
MIC
=
Minimum inhibitory concentration
76 Frontiers in Drug Design & Discovery, 2009, Vol. 4
MRS
=
Magnetic resonance spectroscopy
MS
=
Mass spectrometry
NMDA
=
N-methyl-D-aspartic acid
PET
=
Photon emission tomography
PK-PD
=
Pharmacokinetic-pharmacodynamic
SPET
=
Single photon emission computed tomography
TDM
=
Therapeutic drug monitoring
Höcht et al.
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38]
Debouck, C.; Metcalf, B. Ann. Rev. Pharmacol. Toxicol., 2000, 40, 193. Curtis, G.; Colburn, W.; Heath, G.; Lenehan, T.; Kotschwar, T. Appl. Clin. Trials, 2000, 9, 52. Derendorf, H.; Lesko, L.J.; Chaikin, P.; Colburn, W.A.; Lee, P.; Miller, R.; Powell, R.; Rhodes, G.; Stanski, D.; Venitz, J. J. Clin. Pharmacol., 2000, 40, 1399. Toutain, P.L. AAPS J., 2002, 4, 1. Prentis, R.A.; Lis, Y.; Walker, S.R. Br. J. Clin. Pharmacol., 1988, 25, 387. Brunner, M.; Langer, O. AAPS J., 2006, 8, E263. Levy, G. Clin. Pharmacokinet., 1998, 34, 323. Ho, R.H.; Kim, R.B. Clin. Pharmacol. Ther., 2005, 78, 260. Gogtay, N.J.; Kshirsagar, N.A.; Dalvi, S.S. Br. J. Clin. Pharmacol., 2001, 52, 103S. Schumacher, G.E.; Barr, J.T. Expert Rev. Pharmacoeconomics Outcomes Res., 2002, 2, 89. Schuck, E.L.; Derendorf, H. Expert Rev. Antiinfect. Ther., 2005, 3, 361. Höcht, C.; Opezzo, J.A.W.; Bramuglia, G.F.; Taira, C.A. Curr. Clin. Pharmacol., 2006, 1, 163. Langer, O.; Müller, M. Curr. Drug. Metab., 2004, 5, 463. Nordbring, F. Scand. J. Infect. Dis. Suppl., 1978, 14, 21. Fischman, A.J.; Babich, J.W.; Bonab, A.A.; Alpert, N.M.; Vincent, J.; Callahan, R.J.; Correia, J.A.; Rubin R.H. Antimicrob. Agents Chemother., 1998, 42, 2048. Jain, E.K. Nat. Med., 1994, 4, 655. Heikkinen, T.; Laine, K.; Neuvonen, P.J.; Ekblad, U. Br. J. Obstet. Gynaecol., 2000, 107, 770. Begley, D.J. Pharmacol. Ther., 2004, 104, 29. de Boer, A.G.; Gaillard, P.J. Annu. Rev. Pharmocol. Toxicol., 2007, 47, 323. Cordon-Cardo, C.J.; O’Brien, J.P.; Casals, D.; Rittman-Grauer, L.; Biedler, J.L.; Melamed, M.R.; Bertino, J.R. Proc. Natl. Acad. Sci. USA, 1989, 86, 695. de Boer, A.G.; van der Sandt, I.C.; Gaillard, P.J. Annu. Rev. Pharmacol. Toxicol., 2003, 43, 629. Löscher, W.; Potschka, H. Nat. Rev. Neurosci., 2005, 6, 591. Fromm, M.F. Trends Pharmacol. Sci., 2004, 25, 423. Löscher, W.; Potschka, H. Prog. Neurobiol., 2005, 76, 22. Remy, S.; Beck, H. Brain, 2006, 129, 18. Lazarowski, A.; Czornyj, L.; Lubieniecki, L.; Vazquez, S.; D’Giano, C.; Sevlever, G.; Taratuto, A.L.; Brusco, A.; Girardi, E. Curr. Drug Ther., 2006, 1, 291. Lin, J.H. Curr. Drug Metab., 2006, 7, 39. Gibbs, M.A.; Hosea, N.A. Clin. Pharmacokinet., 2003, 42, 969. Höcht, C.; Opezzo, J.A.W.; Taira, C.A. Curr. Drug Discov. Technol., 2004, 1, 269. Perez-Urizar, J.; Granados-Soto, V.; Flores-Murrieta, F.J.; Castañeda-Hernandez, G. Arch. Med. Res., 2000, 31, 539. Csajka, C.; Verotta, D. J. Pharmacokinet. Pharmacodyn., 2006, 33, 227. Bauer, R.J.; Guzy, S.; Ng, C. AAPS J., 2007, 9, E60. Sheiner, L.B.; Stanski, D.R.; Vozeh, S.; Miller, R.D.; Ham, J. Clin. Pharmacol. Ther., 1979, 25, 358. Mandema, J.W.; Tuck, B.; van Stevenink, A.L.; Breimer, D.D.; Cohen, A.F.; Danhof, M. Clin. Pharmacol. Ther., 1992, 51, 715. Dayneka, N.L.; Garg, V.; Jusko, W. J. Pharmacokinet. Biopharm., 1993, 21, 457. Bauer, J.A.; Balthasar, J.P.; Fung, H.L. Pharm. Res., 1997, 14, 1140. Höcht, C.; Opezzo, J.A.W.; Bramuglia, G.F.; Taira, CA. Expert Opin. Drug Discov., 2006, 1, 289. Schoemaker, R.C.; van Gerven, J.M.A.; Cohen, A.F. J. Pharmacokinet. Biopharm., 1998, 26, 581.
New Sampling Techniques [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 77
Danhof, M.; de Lange, E.C.; Della Pasqua, O.E.; Ploeger, B.A.; Voskuyl, R.A. Trends Pharmacol. Sci., 2008, 29, 186. Danhof, M.; de Jongh, J.; De Lange, E.C.; Della Pasqua, O.; Ploeger, B.A.; Voskuyl, R.A. Annu. Rev. Pharmacol. Toxicol., 2007, 47, 357. Plock, N.; Kloft, C. Eur. J. Pharm. Sci., 2005, 25, 1. Benfeldt, E.; Serup, J.; Menne, T. Acta Derm. Venérelo., 1999, 79, 338. Vogt, G.L.; Janle, E.; Ash, S.R.; Venema, K.; Korf, J. Life Sci., 2003, 73, 2005. Nutt, R. Mol. Imaging Biol., 2002, 4, 11. Lönnroth, P.; Jansson, P.A.; Smith, U. Am. J. Physiol., 1987, 253, E228. Höcht, C.; Opezzo, J.A.; Taira, C.A. J. Pharmacol. Toxicol. Methods, 2007, 55, 3. Davies, M.I.; Lunte, C.E. Drug Metab. Dispos., 1995, 23, 1072. Benveniste, H.; Hüttemeier, P.C. Prog. Neurobiol., 1990, 35, 195. de Lange, E.C.; Danhof, M.; de Boer, A.G.; Breimer, D.D. Brain Res. Rev., 1997, 25, 27. Ault, J.M.; Riley, C.M.; Meltzer, N.M.; Lunte, C.E. Pharm. Res., 1994, 11, 1631. Palsmeier, R.K.; Lunte, C.E. Life Sci., 1994, 55, 815. de Lange, E.; de Boer, A.G.; Breimer, D. Adv. Drug Deliv. Rev., 2000, 45, 125. Groth, L.; Serup, J. Acta Derm. Venereol. (Stockh), 1998, 78, 5. Hutchinson, P.J.; O’Connell, M.T.; Nortje, J.; Smith, P.; Al-Rawi, P.G.; Gupta, A.K.; Menon, D.K.; Pickard, J.D. Physiol. Meas., 2005, 26, 423. Sjögren, F.; Svensson, C.; Anderson, C. Br. J. Dermatol., 2002, 146, 375. Rosdahl, H.; Ungerstedt, U.; Henriksson, J. Acta Physiol. Scand., 1997, 159, 261. Wisniewski, N.; Torto, N. Analyst, 2002, 127, 1129. Torto, N.; Mikeladze, E.; Gorton, L.; Csoregi, E.; Laurell, T. Anal. Commun., 1999, 36, 171. Sandberg, M.; Lindström, S. J. Neurosci. Methods, 1983, 9, 65. Kendrick, K.M. Methods Enzymol., 1989, 168, 182. Rosdahl, H.; Ungerstedt, U.; Henriksson, J. Acta Physiol. Scand., 1997, 159, 261. Maidment, N.T.; Brumbaugh, D.R.; Rudolph, V.D.; Erdelyi, E.; Evans, CJ. Neuroscience, 1989, 33, 549. Huff, J.K.; Bresnahan, J.F.; Davies, M.I. Life Sci., 2003, 73, 275. Ekstrøm, P.O.; Andersen, A.; Saeter, G.; Giercksky, K.E.; Slørdal, L. Cancer Chemother. Pharmacol., 1997, 39, 267. Mantovani, V.; Kennergren, C.; Berglin, E.; Moratti, R.; Lonnroth, P.; Hamberger, A.; Vigano, M. Scand. Cardiovasc. J., 2002, 36, 308. Castejon, A.M.; Paez, X.; Hernandez, L.; Luigi, X.; Cubeddu, L.X. J. Pharmacol. Exp. Ther., 1999, 291, 960. Davies, M.I.; Cooper, J.D.; Desmond, S.S.; Lunte, C.E.; Lunte, S.M. Adv. Drug Deliv. Rev., 2002, 45, 169. Johnson, R.D.; Justice, J.B. Brain Res. Bull., 1983, 10, 567. Jacobson, I.; Sandberg, M.; Hamberg, A. J. Neurosci. Methods, 1985, 15, 263. Wang, Y.; Wong, S.L.; Sawchuk, R.J. Curr. Sep., 1991, 10, 87. Benveniste, H. J. Neurochem., 1989, 52, 1667. Ståhle, L. Eur. J. Clin. Pharmacol., 1992, 43, 289. Liu, P.; Derendorf, H. Infect. Dis. Clin. N. Am., 2003, 17, 599. Joukhadar, C.; Derendorf, H.; Müller, M. Eur. J. Clin. Pharmacol., 2001, 57, 211. Schuck, E.L.; Derendorf, H. Expert Rev. Antiinfect. Ther., 2005, 3, 361. Brunner, M.; Derendorf, H.; Müller, M. Curr. Opin. Pharmacol., 2005, 5, 495. Wise, R. Clin. Pharmacokinet., 1986, 11, 470. Liu, P.; Müller, M.; Grant, M.; Webb, A.; Obermann, B.; Derendorf, H. J. Antimicrob. Chemoter., 2000, 50, 19. Cohen, R.; de La Rocque, F.; Boucherat, M. Med. Mal. Infect., 1994, 24, 844. Burkhardt, O.; Brunner, M.; Schmidt, S.; Grant, M.; Tang, Y.; Derendorf, H. J. Antimicrob. Chemother., 2006, 58, 632. Islinger, F.; Bouw, R.; Stahl, M.; Lackner, E.; Zeleny, P.; Brunner, M.; Muller, M.; Eichler, H.G.; Joukhadar, C. Antimicrob. Agents Chemother., 2004, 48, 4246. Gattringer, R.; Urbauer, E.; Traunmuller, F.; Zeitlinger, M.; Dehghanyar, P.; Zeleny, P.; Graninger, W.; Muller, M.; Joukhadar, C. Antimicrob. Agents Chemother., 2004, 48, 4650. Marchand, S.; Dahyot, C.; Lamarche, I.; Mimoz, O.; Couet, W. Antimicrob. Agents Chemother., 2005, 49, 2356.
78 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119]
Höcht et al.
Dahyot, C.; Marchand, S.; Pessini, G.L.; Pariat, C.; Debaene, B.; Couet, W.; Mimoz, O. Antimicrob Agents Chemother., 2006, 50, 2265. Lefeuvre, S.; Marchand, S.; Lamarche, I.; Mimoz, O.; Couet, W. Antimicrob. Agents Chemother., 2006, 50, 34. Tomaselli, F.; Maier, A.; Matzi, V.; Smolle-Jüttner, F.M.; Dittrich, P. Antimicrob. Agents Chemother., 2004, 48, 2228. Burkhardt, O.; Derendorf, H.; Jager, D.; Kumar, V.; Madabushi, R.; Rohl, K.; Barth, J. Scand. J. Infect. Dis., 2006, 38, 90. Bellmann, R.; Kuchling, G.; Dehghanyar, P.; Zeitlinger, M.; Minar, E.; Mayer, B.X.; Muller, M.; Joukhadar, C. Br. J. Clin. Pharmacol., 2004, 57, 563. Zeitlinger, M.A.; Traunmuller, F.; Abrahim, A.; Muller, M.R.; Erdogan, Z.; Muller, M.; Joukhadar, C. Int. J. Antimicrob. Agents, 2007, 29, 44. Delacher, S.; Derendorf, H.; Hollenstein, U.; Brunner, M.; Joukhadar, C.; Hofmann, S.; Georgopoulos, A.; Eichler, H.G.; Muller, M. J. Antimicrob. Chemother., 2000, 46, 733. Zeitlinger, M.A.; Marsik, C.; Georgopoulos, A.; Müller, M.; Heinz, G.; Joukhadar, C. Int. J. Antimicrob. Agents, 2003, 21, 562. Liu, P.; Rand, K.H.; Obermann, B.; Derendorf, H. Int. J. Antimicrob. Agents, 2005, 25, 120. Huang, Y.; Yang, Z.; Cartier, L.; Cheung, B.; Sawchuk, R.J. Antimicrob. Agents Chemother., 2007, 51, 4336. Presant, C.A.; Wolff, W.; Walush, V.; Wiseman, C.; Kennedy, P.; Blayney, D.; Brechner, R.R. Lancet, 1994, 343, 1184. Brunner, M.; Müller, M. Eur. J. Clin. Pharmacol., 2002, 58, 227. Zhouh, Q.; Gallo, J.M. AAPS J., 2005, 7, E659. Evans, W.E.; Relling, M.V.; Rodman, J.H.; Crom, W.R.; Boyett, J.M.; Pui, C.H. N. Engl. J. Med., 1998, 338, 499. Müller, M.; Brunner, M.; Schmid, R.; Mader, R.M.; Bockenheimer, J.; Steger, G.G.; Steiner, B.; Eichler, H.G.; Blochl-Daum, B. Cancer Res., 1998, 58, 2982. Müller, M.; Mader, R.M.; Steiner, B.; Steger, G.G.; Jansen, B.; Gnant, M.; Helbich, T.; Jakesz, R.; Eichler, H.G.; Blochl-Daum, B. Cancer Res., 1997, 57, 2598. Garvin, S.; Dabrosin, C. Cancer Res., 2003, 63, 8742. Thompson, J.F.; Siebert, G.A.; Anissimov, Y.G.; Smithers, B.M.; Doubrovsky, A.; Anderson, C.D.; Roberts, M.S. Br. J. Cancer, 2001, 85, 157. Müller, M.; Bockenheimer, J.; Zellenger, U.; Klein, N.; Steger, G.G.; Eichler, H.G.; Mader, R.M. Breast Cancer Res. Treat., 2000, 60, 211. Lundstedt, C.; Stridbeck, H.; Andersson, R.; Tranberg, K.G.; Andren-Sandberg, A. Acta Radiol., 1991, 32, 518. Langer, O.; Karch, R.; Muller, U.; Dobrozemsky, G.; Abrahim, A.; Zeitlinger, M.; Lackner, E.; Joukhadar, C.; Dudczak, R.; Kletter, K.; Muller, M.; Brunner, M. J. Nucl. Med., 2005, 46, 1835. de Lange, E.C.M.; Danhof, M. Clin. Pharmacokinet., 2002, 41, 691. Upton, R.N. Clin. Exp. Pharmacol. Physiol., 2008, 34, 695. Alves, O.L.; Doyle, A.J.; Clausen, T.; Gilman, C.; Bullock, R. Ann. N. Y. Acad. Sci., 2003, 993, 25. Regesta, G.; Tanganelli, P. Epilepsy Res., 1999, 34, 109. Lindberger, M.; Tomson, T.; Wallstedt, L.; Stahle L. Epilepsia, 2001, 42, 256. Scheyer, R.D.; During, M.J.; Hochholzer, J.M.; Spencer, D.D.; Cramer, J.A.; Mattson, R.H. Epilepsy Res., 1994, 18, 227. Scheyer, R.D.; During, M.J.; Spencer, D.D.; Cramer, J.A.; Mattson, R.H. Neurology, 1994, 44, 1469. Ederoth, P.; Tunblad, K.; Bouw, R.; Lundberg, C.J.; Ungerstedt, U.; Nordstrom, C.H.; HammarlundUdenaes, M. Br. J. Clin. Pharmacol., 2004, 57, 427. Zhuang, Y.; Fraga, C.H.; Hubbard, K.E.; Hagedorn, N.; Panetta, J.C.; Waters, C.M.; Stewart, C.F. Cancer Res., 2006, 66, 11305. Devineni, D.; Klein-Szanto, A.; Gallo, J.M. Cancer Chemother. Pharmacol., 1996, 38, 499. Rizzi, M.; Caccia, S.; Guiso, G.; Richichi, C.; Porter, J.A.; Aronica, E.; Aliprandi, M.; Bagnati, R.; Fanelli, R.; D´Incalci, M.; Maman, R.; Vezzani, A. J. Neurosci., 2002, 22, 5833. Potschka, H.; Löscher, W. Neuroreport, 2002, 13, 167. Höcht, C.; Lazarowski, A.; Gonzalez, N.N.; Auzmendi, J.; Opezzo, J.A.; Bramuglia, G.F.; Taira, C.A.; Girardi, E. Neurosci. Lett., 2007, 413, 168. Apparaju, S.K.; Gudelsky, G.A.; Desai, P.B. Cancer Chemother. Pharmacol., 2008, 61, 223. Bundgaard, C.; Jørgensen, M.; Mørk, A. J. Pharmacol. Toxicol. Methods, 2007, 55, 214.
New Sampling Techniques [120] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140]
[141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152]
[153] [154]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 79
Deecher, D.C.; Beyer, C.E.; Johnston, G.; Bray, J.; Shah, S.; Abou-Gharbia, M.; Andree, T.H. J. Pharmacol. Exp. Ther., 2006, 318, 657. Kiewert, C.; Hartmann, J.; Stoll, J.; Thekkumkara, T.J.; Van der Schyf, C.J.; Klein, J. Neurochem. Res., 2006, 31, 395. Weikop, P.; Egestad, B.; Kehr, J. J. Neurosci. Methods, 2004, 140, 59. Raje, S.; Cornish, J.; Newman, A.H.; Cao, J.; Katz, J.; Eddington, N.D. Pharm. Res., 2005, 22, 603. Bouw, M.R.; Xie, R.; Tunblad, K.; Hammarlund-Udenaes, M. Br. J. Pharmacol., 2001, 134, 1796. Chenel, M.; Marchand, S.; Dupuis, A.; Lamarche, I.; Paquereau, J.; Pariat, C.; Couet, W. Br. J. Pharmacol., 2004, 142, 323. Meredith, P.A. J. Hypertens., 1997, 15, S27. Brynne, L.; Karlsson, M.O.; Paalzow, L.K. J. Pharmacol. Exp. Ther., 1998, 286, 1152. Höcht, C.; Di Verniero, C.; Opezzo, J.A.W.; Taira, CA. Naunyn-Schmiedeberg’s Arch. Pharmacol., 2004, 370, 1. Höcht, C.; Di Verniero, C.; Opezzo, J.A.W.; Taira, CA. J. Pharmacol. Toxicol. Methods, 2005, 52, 244. Höcht, C.; Di Vereniero, C.; Opezzo, J.A.W.; Bramuglia, G.F.; Taira, C.A. Naunyn-Schmiedeberg’s Arch. Pharmacol., 2006, 373, 310. Bertera, F.M.; Mayer, M.A.; Opezzo, J.A.W.; Taira, C.A.; Höcht, C. J. Pharmacol. Toxicol. Methods, 2008, 57, 213. Bertera, F.M.; Mayer, M.A.; Opezzo, J.A.W.; Taira, C.A.; Bramuglia, G.F.; Höcht, C. J. Pharmacol. Toxicol. Methods, 2007, 56, 290. Fischman, A.J.; Alpert, N.M.; Rubin, R.H. Clin. Pharmacokinet., 2002, 41, 481. van den Hoff, J. Amino Acids, 2005, 29, 341. Martínez, M.J.; Ziegler, S.I.; Beyer, T. Cancer Res., 2008, 170, 1. Wollmer, P.; Pride, N.B.; Rhodes, C.G.; Sanders, A.; Pike, V.W.; Palmer, A.J.; Silvester, D.J.; Liss, R.H. Lancet, 1982, 2, 1361. Fischman, A.J.; Alpert, N.M.; Livni, E.; Ray, S.; Sinclair, I.; Elmaleh, D.R.; Weiss, S.; Correia, J.A.; Webb, D.; Lissm R. J. Pharmacol. Exp. Ther., 1991, 259, 1351. Fischman, A.J.; Alpert, N.M.; Livni, E.; Ray, S.; Sinclair, I.; Callahan, R.J.; Correia, J.A.; Webb, D.; Strauss, H.W.; Rubin, R.H. Antimicrob Agents Chemother., 1993, 37, 1270. Fischman, A.J.; Livni, E.; Babich, J.; Alpert, N.M.; Liu, Y.Y.; Thom, E.; Cleeland, R.; Prosser, B.L.; Correia, J.A.; Strauss, H.W. Antimicrob. Agents Chemother., 1993, 37, 2144. Langer, O.; Brunner, M.; Zeitlinger, M.; Ziegler, S.; Müller, U.; Dobrozemsky, G.; Lackner, E.; Joukhadar, C.; Mitterhauser, M.; Wadsak, W.; Minar, E.; Dudczak, R.; Kletter, K.; Müller, M. Eur. J. Nucl. Med. Mol. Imaging, 2005, 32, 143. Brunner, M.; Langer, O.; Dobrozemsky, G.; Müller, U.; Zeitlinger, M.; Mitterhauser, M.; Wadsak, W.; Dudczak, R.; Kletter, K.; Müller, M. Antimicrob. Agents Chemother., 2004, 48, 3850. Fischman, A.J.; Babich, J.W.; Bonab, A.A.; Alpert, N.M.; Vincent, J.;Callahan, R.J.; Correia, J.A.; Rubin, R.H. Antimicrob. Agents Chemother., 1998, 42, 2048. Wall, A.; Kågedal, M.; Bergström, M.; Jacobsson, E.; Nilsson, D.; Antoni, G.; Frändberg, P.; Gustavsson, S.A.; Långström, B.; Yates, R. Drugs R D, 2005, 6, 139. Gulyás, B.; Halldin, C.; Sandell, J.; Karlsson, P.; Sóvágó, J.; Kárpáti, E.; Kiss, B.; Vas, A.; Cselényi, Z.; Farde, L. Acta Neurol. Scand., 2002, 106, 325. Whitcher, B.; Matthews, P.M. Int. J. Pharm. Med., 2006, 20, 167. Aboagye, E.O.; Price, P.M. Invest. New Drugs, 2003, 21, 169. Weber, W.A.; Czernin, J.; Phelps, M.E.; Herschman, H.R. Nat. Clin. Pract. Oncol., 2008, 5, 44. Weber, W.A. J. Clin. Oncol., 2006, 24, 3282. Gupta, N.; Price, P.M.; Aboagye, E.O. Eur. J. Cancer, 2002, 38, 2094. Hutchinson, O.C.; Collingridge, D.R.; Barthel, H.; Price, P.M.; Aboagye, E.O. Curr. Pharm. Des., 2003, 9, 931. Saleem, A.; Harte, R.J.; Matthews, J.C.; Osman, S.; Brady, F.; Luthra, S.K.; Brown, G.D.; Bleehen, N.; Connors, T.; Jones, T.; Price, P.M.; Aboagye, E.O. J. Clin. Oncol., 2001, 19, 1421. Harte, R.J.A.; Tilsley, D.W.O.; O’Reilly, S.M.; Byrne, H.; Cunningham, V.; Osman, S.; Luthra, S.K.; Brown, G.; Brady, F.; Baguley, B.; Denny, W.; Connors, T.; Jones, T.; Price, P. Ann. Oncol., 1994, 5, 176. Meikle, S.R.; Matthews, J.C.; Brock, C.S.; Wells, P.; Harte, R.J.; Cunningham, V.J.; Jones, T.; Price, P. Cancer Chemother. Pharmacol., 1998, 42, 183. Saleem, A.; Brown, G.D.; Brady, F.; Aboagye, E.O.; Osman, S.; Luthra, S.K.; Ranicar, A.S.; Brock, C.S.; Stevens, M.F.; Newlands, E.; Jones, T.; Price, P. Cancer Res., 2003, 63, 2409.
80 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [155]
[156] [157] [158] [159] [160] [161] [162] [163]
Höcht et al.
Jayson, G.C.; Zweit, J.; Jackson, A.; Mulatero, C.; Julyan, P.; Ranson, M.; Broughton, L.; Wagstaff, J.; Hakannson, L.; Groenewegen, G.; Bailey, J.; Smith, N.; Hastings, D.; Lawrance, J.; Haroon, H.; Ward, T.; McGown, A.T.; Tang, M.; Levitt, D.; Marreaud, S.; Lehmann, F.F.; Herold, M.; Zwierzina, H.; European Organisation for Research and Treatment of Cancer Biological Therapeutic Development Group. J. Natl. Cancer Inst., 2002, 94, 1484. Shankar, L.K.; Hoffman, J.M.; Bacharach, S.; Graham, M.M.; Karp, J.; Lammertsma, A.A.; Larson, S.; Mankoff, D.A.; Siegel, B.A.; Van den Abbeele, A.; Yap, J.; Sullivan, D.; National Cancer Institute. J. Nucl. Med., 2006, 47, 1059. Vesselle, H.; Grierson, J.; Muzi, M.; Pugsley, J.M.; Schmidt, R.A.; Rabinowitz, P.; Peterson, L.M.; Vallières, E.; Wood, D.E. Clin. Cancer Res., 2002, 8, 3315. Chen, W.; Cloughesy, T.; Kamdar, N.; Satyamurthy, N.; Bergsneider, M.; Liau, L.; Mischel, P.; Czernin, J.; Phelps, M.E.; Silverman, D.H. J. Nucl. Med., 2005, 46, 945. Wolf, W.; Presant, C.A.; Waluch, V. Adv. Drug Deliv. Rev., 2000, 41, 55. Jynge, P.; Skjetne, T.; Gribbestad, I. Clin. Pharmacol. Ther., 1990, 48, 481. Renshaw, P.F.; Wicklund, S. Biol. Psychiatry, 1988, 23, 465. Renshaw, P.F.; Guimaraes, A.R.; Fava, M. Am. J. Psychiatry, 1992, 149, 1592. Schiffer, W.K.; Alexoff, D.L.; Shea, C.; Logan, J.; Dewey, S.L. J. Neurosci. Methods, 2005, 144, 25.
Frontiers in Drug Design & Discovery, 2009, 4, 81-96
81
Role of Inflammatory Biomarkers in Establishing PK/PD Relationships and Target Organ Toxicity Sivaram Pillarisetti* and Ish Khanna* Reddy US Therapeutics Inc., 3065 Northwoods Circle, Norcross GA 30071, USA Abstract: High levels of inflammatory cytokines and adhesion molecules are associated with many inflammatory disorders [e.g. rheumatoid arthritis, inflammatory bowel disease and lupus] as well as metabolic and cardiovascular diseases including diabetes and obesity. Examples of such markers include tumor necrosis factor [TNF], interleukins [IL-1, IL-6, IL-8 and IL-18], vascular cell adhesion molecules and markers of macrophage inflammation [e.g. MMPs]. In many preclinical disease models, levels of these markers are significantly elevated relative to normal animals. Modulation of these biomarkers with pharmaceutical agents in preclinical and clinical studies can be effectively used for concept validation, effective dose selection, and establishing good pharmacokinetics/pharcodynamics correlation. On the other hand, elevation of these markers in normal animals, following treatment with an agent, could indicate safety concerns leading to potential tissue damage. Monitoring of inflammatory markers in normal animals can be diagnostic and of high value in evaluating safety or efficacy of new molecules. The levels of biomarkers can be monitored either by high throughput microarrays/proteomics or by specific ELISA based assays. Since many of the biomarkers appear in systemic circulation, these can be monitored in blood/plasma without interference with tissues. This makes the approach particularly attractive for clinical studies. An overview of the biomarkers, potential applications and case histories linking biomarkers to PK/PD correlation from preclinical and clinical studies are discussed.
1. INTRODUCTION Inflammation is normally a localized, protective response to harmful stimuli such as pathogens or irritants. The process usually starts with an injury/insult to blood vessel wall endothelium. Exposure to a variety of pathogens can cause endothelial activation resulting in secretion of chemoattractant signals (Fig. 1). Depending on the nature of signal, circulating monocytes and other leukocytes are attracted to the site of injury and interact/bind to endothelium through a variety of receptors [e.g. selectins and vascular adhesion molecules] [1, 2]. Bound leukocytes transmigrate to the subendothelial space where they become tissue resident macrophages. The recruitment, accumulation and
*Corresponding Author: Tel: 770-446 9500; Fax: 770-446 1950; E-mail:
[email protected] or
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
82 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pillarisetti and Khanna
subsequent activation of leukocytes are central events in the pathogenesis of most forms of inflammation [1, 2]. Host-mediated inflammatory response destroys, dilutes or wards off the injurious agent and the injured tissue. Commonly, inflammation occurs as a defensive response to invasion of the host by foreign, particularly microbial, agents. Responses to mechanical trauma, toxins and neoplasia can also cause inflammatory reactions. Circulating monocytes
Tethering and rolling of monocytes
Transmigration
Chemotactic signals Endothelium Activated/injured endothelium
Maturation to macrophages
Cytokines (TNF, IL-1) Effects on surrounding cells in tissue
Inflammatory diseases e.g. RA, IBD, OA
Organ toxicity
Fig. (1). Pathways leading to tissue inflammation (see text for details).
There are other forms of inflammation that often don’t involve infection and host defense. These include autoimmune responses and subclinical inflammation. In autoimmune inflammation body proteins are recognized as foreign and antibodies are produced against body’s own proteins. This is the underlying cause of autoimmune diseases such as rheumatoid arthritis and systemic lupus erythematosus. Subclinical inflammation can occur under normal physiological conditions and is often attributed to nutritional status and related oxidative stress. Excess plasma glucose [hyperglycemia] or plasma lipids [cholesterol or free fatty acids] or obesity can induce oxidative stress [reactive oxygen species or ROS] within cells via various mechanisms [3-6]. Excess oxidative stress activates ROS-sensitive transcription factors which drive the expression of inflammatory cytokines [7, 8]. Prolongation of the inflammatory process can damage tissue of origin or can lead to inflammatory diseases as diverse as diabetes, diabetic nephropathy, atherosclerosis, Alzheimer’s, fatty liver, pancreatitis and cancer [9-18]. Many cytokines and the high-sensitivity C-reactive protein [CRP] are considered markers of inflammation and have been studied in both disease progression as well as organ toxicity. This review
Role of Inflammatory Biomarkers
Frontiers in Drug Design & Discovery, 2009, Vol. 4 83
will focus on selective markers of inflammation that are useful in monitoring the efficacy and safety of pharmacological agents. 2. KEY INFLAMMATORY CYTOKINES AND THEIR ROLES Cytokines encompass a large number of glycoproteins involved in cell-to-cell signaling. Cytokines can be classified based on their mode of action [pro or antiinflammatory] or based on cell type of origin. e.g., T helper cell type I [Th1; for TNF and interleukin-2] and T helper cell type II [Th2; for interleukins -4, -5, -6, -10, and -13. Cytokines may also be identified based on their function such as those effecting chemoattraction [chemokines, including monocyte chemoattractant protein-1 [MCP-1]]. Although leukocytes/lymphocytes are the major sources of these cytokines, other cell types and tissues including vascular cells, liver and adipose express one or more of these cytokines and their receptors, making them targets of down stream signaling and toxicities. A brief discussion on inflammatory cytokines is outlined below. 2a. Tumor Necrosis Factor (TNF) Tumor necrosis factor-alpha [TNF] often described as the master cytokine is the major mediator of inflammation in many tissues [19-21]. It was originally described as a circulating factor that can cause necrosis of tumors, but has since been identified as a key regulator of the inflammatory response. It is also a major mediator of cell apoptosis and plays key role in immune defense against bacterial infections. Elevated levels of TNF or sustained activation of TNF signaling has been implicated in the pathogenesis of many autoimmune diseases including rheumatoid arthritis, inflammatory bowel disease, multiple sclerosis and psoriasis. In addition more recent data suggest that TNF along with other cytokines contribute to the pathogenesis of diabetes and cardiovascular disease [atherosclerosis]. TNF binds to two specific receptors, TNF-receptor type I [TNF-R1 aka p55/60] and TNF-receptor type II [TNF-R2, aka p75/80]. Signaling through these receptors is extremely complex, leading to induction of other inflammatory mediators as well as cell death and survival signals. Although signaling involves many players, nuclear transcription factor NFB appears to be critical for regulation of both inflammatory as well as cell death genes [22]. Vascular endothelial cells are highly susceptible to TNF challenge and undergo a number of pro-inflammatory changes, which increase leukocyte adhesion, transendothelial migration and promote thrombosis [23]. Thus all vascularized tissues are prone to TNF-mediated toxicity. TNF-signaling in cells also leads to induction of other pro- and anti-inflammatory cytokines including interleukins-1, -4, -6, -8 and -10 and monocyte chemoattractant protein MCP-1. The relative levels of these cytokines may determine the extent of damage/toxicity to the tissue. The central role of TNF in inflammation has been demonstrated in clinic. A number of agents that block the action of TNF have proven highly beneficial to treat a range of inflammatory conditions, including rheumatoid arthritis, ankylosing spondylitis, inflammatory bowel disease and psoriasis [24-26]. 2b. Interleukins IL-1 The IL-1 family consists of four proteins that share considerable sequence homology IL-1, IL-1, IL-1 receptor antagonist [IL-1Ra], and IL-18 [27, 28]. Mature IL-1 is deri-
84 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pillarisetti and Khanna
ved proteolytically from pro-IL-1 by intracellular IL-1-converting enzyme [ICE or caspase-1]. IL-1 binds and induces signaling through IL-1 receptor [IL-1R]. Minute amounts [1ng/kg] of IL-1 cause significant inflammatory response in humans. The signaling cascade results in activation of nuclear factor kappa B [NF-B] and activating protein-1 [AP-1] and the transcription of a variety of pro-inflammatory genes, including autocrine amplification of IL-1 [28-30]. In addition to the IL-1RI, IL-1 may also bind to the type II interleukin-1 receptor. Binding of IL-1 to this receptor does not result in cellular activation, and IL-1RII is therefore presumed to act as a decoy that negatively regulates IL-1 activity. IL-18 has been assigned to the IL-1 family on the grounds of sequence homology [26% with IL-1] and similarity of the IL-18 receptor to IL-1R [31]. Like IL-1, IL-18 is dependent on ICE for proteolytic processing, and on nuclear translocation of NFB for transcriptional activation. IL-1 like TNF induces vessel wall inflammation. IL-1 plays a major role in a wide range of inflammatory and autoimmune diseases. These include RA, OA, chronic obstructive pulmonary disease [COPD], asthma, inflammatory bowel disease [IBD] [[both Crohn's disease [CD] and ulcerative colitis [UC]], atherosclerosis and diseases of the central nervous system such as multiple sclerosis [MS], Alzheimer's disease and stroke [32]. IL-6 The IL-6-type cytokines IL-6, IL-11, LIF [leukemia inhibitory factor], OSM [oncostatin M], ciliary neurotrophic factor, cardiotrophin-1 and cardiotrophin-like cytokine are an important family of mediators involved in the regulation of the acute-phase response to injury and infection [33]. Besides their functions in inflammation and the immune response, these cytokines also play a crucial role in hematopoiesis, liver and neuronal regeneration, embryonic development and fertility. IL-6-type cytokines exert their action via the signal transducers gp [glycoprotein] 130, LIF receptor and OSM receptor leading to the activation of the JAK/STAT [Janus kinase/signal transducer and activator of transcription] and MAPK [mitogen-activated protein kinase] cascades [34]. Although gp130 is ubiquitously expressed, the number of cells that respond to IL-6 is limited, since the expression of the other receptor subunits especially of the -receptors, is more restricted and tightly regulated. The function of the -receptors to render cells sensitive to the respective cytokinecan also be taken over by the soluble form of the -receptors. This is one of the rare situations in which a complex of cytokine and soluble receptor can act agonistically instead of antagonistically. Soluble forms of cytokine receptors in vivo are formed either by limited proteolysis [shedding] of membrane-bound receptors or by translation from an alternatively spliced mRNA [4]. In the case of IL-6, the scenario is more complex, since soluble forms for IL-6R [sIL-6R] and gp130 [sgp130] are both present in human serum. It has been demonstrated that sIL-6R potentiates the antagonistic activity of sgp130. Thus the naturally occurring combination of sIL-6R and sgp130 might act as a kind of buffer to modulate systemic responses to circulating IL-6. Because IL-6 has a wide range of biological activities on various target cells, deregulated overproduction of IL-6 causes various clinical symptoms. IL-6 plays important roles in the regulation of immune response and inflammation, and overproduction of IL-6 is involved in the pathology of inflammatory diseases such as rheumatoid arthritis [RA], Castleman’s disease, juvenile idiopathic arthritis and Crohn’s disease [35].
Role of Inflammatory Biomarkers
Frontiers in Drug Design & Discovery, 2009, Vol. 4 85
2c. Other Inflammatory Cytokines/Markers MCP-1 MCP-1 is a chemoattractant protein that specifically attracts blood monocytes and tissue macrophages to its source, via interaction with CCR2, its cell surface receptor [36]. Many cells produce MCP-1 in response to a variety of pro-inflammatory stimuli, and predictably, its expression has been identified in diseases which involve significant macrophage inflammation. These include atherosclerosis, rheumatoid arthritis and kidney diseases such as diabetic nephropathy. CRP CRP is an acute phase reactant that was discovered over 70 years ago. CRP is a blood protein that binds to the C-polysaccharide of pneumococci. CRP is a pentamer of 23 kDa subunits and is mainly produced by liver. CRP levels are usually low in normal individuals but can rise 100- to 200-fold or higher with acute systemic inflammation. CRP levels are high in patients with rheumatoid arthritis [RA] and cardiovascular disease [CVD]. In patients with CVD levels of CRP predict future myocardial infarction, stroke and peripheral vascular disease. IL-1, IL-6, and TNF, all can stimulate liver production of CRP. It is not entirely clear if CRP is just a biomarker for inflammation or CRP could have a more direct pro-inflammatory role. Modulation of CRP levels by drugs appear to correlate with disease severity [37, 38]. 3. CYTOKINES IN DISEASE Although elevated cytokines are seen in almost all inflammatory diseases the causal relationship has been established in autoimmune diseases such as rheumatoid arthritis [RA], inflammatory bowel disease [IBD], and psoriasis. 3a. Cytokines and Autoimmune Diseases Rheumatoid arthritis [RA] is a chronic disease characterized by synovial inflammation that leads to the destruction of cartilage and bone. Proinflammatory cytokines such as TNF, IL-1 and IL-6 play important role in initiating and perpetuating inflammatory and destructive processes in the rheumatoid joint [20, 27, 31, 34]. These cytokines regulate many NFB inducible genes that control expression of other cytokines, cell adhesion molecules, immunoregulatory molecules, and proinflammatory mediators. The expression of cyclooxygenase-2 and inducible nitric oxide synthase [iNOS] and thereby production of prostaglandins [PG] and NO are regulated by these cytokines. The prostaglandin PGE2 and nitric oxide [NO] further promote inflammation and likely participate in destructive mechanisms in the rheumatoid joint. Although the actions of IL-1 and TNF-alpha show a large degree of overlap, some differences have been observed in animal models [45, 47]. As indicated in Table 1 [39], inflammatory cytokines [TNF-, IL-1, IL-6] are significantly elevated in synovial fluids of RA patients relative to OA [40-44]. This seems particularly relevant for IL-6 where high levels in synovial fluid and serum correlate with disease progression. The preclinical studies suggest that IL-1 expression occurs later in the disease relative to TNF- [45, 46]. The reported data suggests that IL-1 has significant effect on cartilage destruction where as TNF- influences joint swelling [46, 47]. In RA, the cytokines such as IL-4, IL-10 and IL-13, detected in synovial fluid of
86 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pillarisetti and Khanna
patients, have protective, anti-inflammatory effects. In patients with active rheumatoid arthritis, blockade of TNF-, IL-1 or IL-6 results in improvement in clinical and radiographic scores. TNF-alpha antagonists are currently the most effective treatments for early and late RA. Use of these agents provides relief of RA symptoms and improvements in American College of Rheumatology [ACR] criteria. ACR criteria is indicated as ACR 20, ACR 50, or ACR 70 and measures improvement in tender or swollen joint counts and improvement in three of five parameters [acute phase reactant such as erythrocyte sedimentation rate, patient assessment, physician assessment, pain scale and disability questionnaire]. Table 1.
Cytokine Levels in Patients with RA and OA
Cytokine
Disease
Concentration [pg/ml]
Fold Change
Reference
TNF
RA OA
95 ± 25 39-69±6-32
2.4
[40, 41]
IL-1
RA OA
130±22 27±4.5
4.6
[42]
IL-6
RA OA
1610 89-398
18
[41, 43]
IL-1 antagonism also has additional benefits on cartilage and bone erosion [48]. In vitro studies suggest that IL-1 can cause cartilage destruction by stimulating the release of matrix metalloproteinases and other degradative products, and it can increase bone resorption by stimulating osteoclast differentiation and activation. In animal models of RA, blocking the effects of IL-1 with either IL-1 receptor antagonist [IL-1Ra; endogenous], anti-IL-1 monoclonal antibodies significantly reduced cartilage destruction and bone erosion. More recently, IL-6 antagonists have shown good efficacy in autoimmune diseases. Tocilizumab, a humanized antihuman IL-6 receptor antibody recognizes both the membrane-bound and the soluble form IL-6R and specifically blocks IL-6 actions [49]. Tocilizumab has been shown to be effective not only for improving signs and symptoms but also for preventing joint destruction of RA. Significantly elevated levels of TNF-, and other cytokines were observed in other autoimmune diseases such as Crohn’s disease, psoriasis and systemic lupus erythematosus and TNF antagonists are effective in relieving symptoms. In summary, these data strongly show that elevated cytokines not only serve as a biomarker to predict disease severity but blockingtheir function can effectively reduce disease severity in many autoimmune diseases. 3b. Cytokines and Cardiovascular Disease Cardiovascular disease is currently the leading cause of death in developed countries. Atherosclerosis - a progressive disease characterized by the accumulation of lipids in the
Role of Inflammatory Biomarkers
Frontiers in Drug Design & Discovery, 2009, Vol. 4 87
large arteries - constitutes the single most important contributor to this growing burden of cardiovascular disease [50, 51]. Macrophage-mediated inflammation plays an important role in the initiation and progression of atherosclerosis and the development of atherosclerotic events. The macrophages can avidly take up lipids within the lesion leading to lipid loaded foam cells. Inflammatory cytokines can weaken the protective fibrous cap of the atheroma by stimulating the expression degradative enzymes such as MMP-2 and -9. Loss of fibrous cap results in thrombosis and the occurrence of acute coronary syndromes such as unstable angina pectoris and myocardial infarction [MI]. Serum levels of cytokines are elevated in many cardiovascular pathologies including coronary heart disease, aortic aneurysm and acute MI [52-60]. Of these cytokines, IL-6 has been identified as an important marker of inflammation in coronary atherosclerotic plaques. Serum levels of IL-6 increase in response to acute MI, unstable angina, percutaneous coronary intervention, and late restenosis [58-60]. IL-6 levels, which were undetectable [i.e., <3 pg/mL] in healthy volunteers is elevated in majority of patients with unstable angina. Median levels of IL-6 were 5.25 pg/mL [range, 0 to 90 pg/dL] in patients with unstable angina and were below the detection limit [range, 0 to 7 pg/dL] in patients with stable angina [60]. The importance of IL-6 is also highlighted by the fact that CRP, one of many human acute-phase reactants, is produced in the liver mainly in response to IL-6. It activates the classic complement cascade, mediates phagocytosis, regulates inflammation, and is a nonspecific but sensitive marker of infection and tissue inflammation. CRP has been shown to predict future cardiovascular events in individuals with and without established cardiovascular disease. Although direct evidence for causal relationship between cytokines and CVD has not been established with specific antagonists, drugs such as statins which substantially reduce cardiovascular morbidity and mortality have been shown to lower inflammatory markers independent of their effects on plasma cholesterol [61, 62] 3c. Cytokines and Metabolic Syndrome The metabolic syndrome, also known as syndrome X, represents a collection of metabolic abnormalities that includes central obesity, insulin resistance, glucose intolerance, dyslipidemia, and hypertension. Each of these features is known to augment the risk of developing diabetes mellitus [DM] and cardiovascular disease. In addition to genetic and other exogenous factors [diet, sedentary life, and alcohol consumption] in the past decade the role of inflammation in the development of the metabolic syndrome has been clearly documented [63, 64]. Adipose tissue secretes inflammatory cytokines, TNF, MCP-1 and IL-6 which in turn contribute to impaired glucose tolerance, insulin resistance, and type 2 diabetes [65]. In animal models TNF impairs insulin action on peripheral glucose disposal and hepatic glucose output resulting in severe impairment of glucose tolerance and insulin sensitivity [66, 67] while these effects are absent in animals that lack TNFR [68, 69]. In human studies the correlation is better established between IL-6 and diabetes. The levels of circulating C-reactive protein and interleukin-6 [IL-6] independently predict the risk of developing diabetes [70, 71]. 4. CYTOKINES AND ORGAN TOXICITY Cytokines in addition to inducing inflammatory pathways enhance pro and antiapoptotic pathways. For example TNF-induced apoptosis-signaling receptors include death receptor 3 [DR3], DR4, and DR5, their ligand TRAIL [72]. Unresolved inflamma-
88 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pillarisetti and Khanna
tion leads to sustained activation of signaling pathways leading to cell death, which will result in tissue [organ] damage and organ failure. 4a. Liver Toxicity The role of TNF in liver injury has been studied in several animal models [73-83]. By using neutralizing anti-TNF antibodies or knockout mice for TNF, TNF-R1, or TNF-R2, it has become evident that TNF triggers apoptosis and/or necrosis of hepatocytes in vivo. In different animal models of liver injury, TNF plays a central or an additive role in the pathogenesis of acute liver injury. In endotoxin/galactosamine induced model of liver injury an increase in TNF levels in the wild-type C3H mice precedes liver failure. A role for TNF was also established in concavalin A [ConA] model of liver toxicity. ConA binds with high affinity the hepatic sinusoidal endothelial cells and triggers an increased influx of circulating lymphocytes into the hepatic sinus. Cytokine levels increase before infiltration of lymphocytes occurrs, showing that the early increase in cytokine levels is pivotal in triggering liver cell damage. TNF has been shown to directly contribute to liver cell damage, since anti-TNF antibodies protect against liver cell damage in this model. Additionally, it has been demonstrated that TNF null mice are protected from ConA-induced liver cell damage, further supporting the role of TNF in the pathogenesis. IL-10 inhibits liver cell damage in this model by reducing the serum levels of TNF. In humans, attempts to use TNF as an anti-cancer therapy have failed due to side effects especially in liver. One of the side effects of TNF treatment was an elevation in serum levels of transaminases and bilirubin levels, indicating a direct cytotoxic role of TNF in human hepatocytes. TNF may also be involved in viral hepatitis, alcoholic liver disease, and fulminant hepatic failure. TNF serum levels are clearly elevated in patients with fulminant hepatitis. In addition, it was found that serum TNF levels were significantly higher in patients who died than in patients who survived. A role for TNF in the pathogenesis of chronic hepatitis B and C viral infection has been suggested. Both viruses induce TNF expression in human liver and human hepatoma cell lines. Patients with chronic hepatitis B have elevated plasma TNF levels, and their peripheral blood mononuclear cells show enhanced TNF production in vitro. Serum levels of soluble TNF-R1 and TNF-R2 are also significantly elevated in chronic hepatitis B infection. The serum levels of soluble TNF-R2 correlate closely with the extent of inflammation and hepatocyte death in the liver. During interferon therapy, the response and the increase in transaminases are associated with an increase in soluble TNF-R2 serum levels. TNF serum levels are increased in patients with alcoholic hepatitis, and the levels correlate inversely with patient survival. Monocytes isolated from patients with alcoholic hepatitis spontaneously produced higher amounts of TNF compared with healthy controls. Increased TNF- levels are also observed in the liver following exposure to hepatotoxic agents. This typically leads to simultaneous generation of other cytokines such as IL-6 and acute phase reactants worsening the liver inflammation. 4b. Kidney Toxicity Patients with chronic kidney disease [CKD] of all stages experience high mortality and chronic inflammation is one of the key triggers to the process [84-86]. Macrophages are known to infiltrate into tubulointersitium in animal models of CKD. Exposure of
Role of Inflammatory Biomarkers
Frontiers in Drug Design & Discovery, 2009, Vol. 4 89
kidney epithelial cells to TNF/IFN results in a marked sustained elevation of transepithelial electrical resistance as well as elevated paracellular permeability thus compromising barrier function [87]. Pharmacological inhibition of specific signaling enzymes blocked the deleterious effects of the proinflammatory cytokines on barrier function [86, 88]. The role of inflammation and inflammatory cytokines has also been documented in acute renal failure [ARF]. Chemotherapy [Cisplatin]-induced ARF has been shown to be associated with an increase in the cytokines IL-1 and IL-6, and neutrophil infiltration in the Kidney and inhibition of cytokines in cisplatin-induced ARF is consistently associated with a reduction in renal neutrophils [89, 90]. 4c. Brain [CNS] Toxicity Inflammation is major player in diseases of the brain, such as multiple sclerosis or meningitis [91, 92]. In these cases, immune cells invade the central nervous system compartments, accompanied by a massive breakdown of the blood-brain barrier inducing changes to the cerebrospinal fluid. In addition, inflammation within the central nervous system is a common phenomenon even in diseases such as Alzheimer disease, Parkinson disease, or stroke. In these cases, inflammation is a frequent occurrence but displays different, more subtle, patterns compared with, for example, multiple sclerosis. Brain levels of IL-1 and TNF increase dramatically in neuropathological states and increased levels of these cytokines are associated with breakdown of the blood–brain barrier and recruitment of neutrophils into the CNS. Neutralizing antibodies to the two cytokines reduce the CNS accumulation of leukocytes in neuropathology. Agents that specifically inhibit the IL-1 pathway [endogenous IL-1 receptor antagonist or neutralizing IL-1 antibodies] show considerable reduction in neuronal loss in the various rodent models. 5. METHODS TO MONITOR CYTOKINES While measurement of plasma cytokines by ELISA based methods is the most widely used especially in clinical setting, PCR based methods may be more sensitive when employing tissues or blood cells. Several companies offer ELISA kits to measure rodent, monkey and human cytokines, including TNF, IL-1, IL-8, IL-6 and MCP-1. Following drug treatment of animals, plasma or serum is collected at various time points for cytokine estimation. In disease models basal cytokine levels are generally elevated and effect of drug could be seen in matter of hours to a few days. A popular model to monitor efficacy of an anti-inflammatory drug is the lipopolysaccharide-induced inflammation in rodents. Injection of low doses of LPS produces a robust increase in plasma TNF. This effect could be seen within 1 h of treatment. Anti-inflammatory drug could be given either 30 min to 1 h prior to the administration of LPS or along with LPS. In other models e.g. antigen or antibody-induced arthritis models it may take 7-14 days to show a significant reduction in plasma cytokines. [see case examples below]. In addition to monitoring plasma cytokines [TNF, IL-1 and IL-6] it would also be useful to measure cytokines in the affected tissue [e.g. synovial fluid and joint tissue in RA models] Induction of cytokine expression in a normal animal following the administration of drug is a reflection of drug toxicity. This can be monitored in repeat dose safety studies. Oral administration of a test substance in rodents for 28 and 90 days is used to evaluate chronic toxic effects, effects on organs, and no –observed-adverse effect-level [NOAEL]. The liver is the primary site for the metabolism of many chemicals and drugs by the body, and is also the primary site of potential toxic injury [hepatotoxicity]. Like all
90 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pillarisetti and Khanna
organs, the liver is composed of various cell types; predominately hepatocytes. Liver cells contain phase 1 and phase 2 drug metabolizing enzymes which convert many chemicals and drugs to other forms that can be readily used or excreted by the body [biotransformation]. In some cases, the biotransformed chemical is the one that causes the toxicity [to the liver or to another organ]. Initial studies can be performed in isolated primary cultures of livers cells to determine if the drug induces the expression of cytokines. However longer treatment times may be required for the drug effect to set in, thus may only be monitored under in-vivo animal studies. 6. CASE STUDIES OF MARKETED DRUGS THAT INHIBIT INFLAMMATORY MARKERS AND MODULATE DISEASE SEVERITY In this section we will review some case examples where modulation of cytokine activity in animal models translate into modification of disease activity in animals and humans Anti-TNF Agents: Currently there are three marketed products that inhibit TNF action. They are infliximab, etanercept andadalimumab. Infiximab and adalimumab are antibodies that bind TNF and prevent its interaction with receptor while enbrel is a soluble form of TNF receptor which can bind TNF but can not transducer signals Case 1 - Infliximab and Ulcerative Colitis Currently infliximab is the only approved anti-TNF agent for treatment of ulcerative colitis. Infliximab is a chimeric IgG1 monoclonal antibody with an approximate molecular weight of 149,100 daltons. It is composed of human constant and murine variable regions. Infliximab binds specifically to human TNF with an association constant of 10-10 M. Infliximab neutralizes the biological activity of TNF by binding with high affinity to the soluble and transmembrane forms of TNF and inhibits its binding to its receptors. In adults, single intravenous [IV] infusions of 3 mg/kg to 20 mg/kg showed a linear relationship between the dose administered and serum concentration. The volume of distribution at steady state was independent of dose and indicated that infliximab was distributed primarily within the vascular compartment. Pharmacokinetic results from single dose studies at 3 mg/kg to 10 mg/kg in rheumatoid arthritis, 5 mg/kg in Crohn's disease, and 3 mg/kg to 5 mg/kg in plaque psoriasis indicate that the median terminal half-life of infliximab is 7.7 to 9.5 Preclinical Studies -Infliximab in Rat Models Of Colitis To investigate the influence of infliximab on experimental colitis Triantafillidis et al. used 2,4,6,trinitrobenzene sulfonic acid [TNBS]-induced colitis model in rats [93]. Colitis was induced by intracolonic installation of 25 mg of TNBS dissolved in 0.25 mL of 50% ethanol and infliximab was subcutaneously administered at doses of 5 -15 mg/kg BW. Plasma and tissue levels of TNF were measured and disease activity was monitored by fixing colon in 10% buffered formalin and examined by light microscopy for the presence and activity of colitis and the extent of tissue damage. There was a significant reduction of the tissue levels of TNF-alpha in all groups of treated animals as compared with the untreated. Better reduction in tissue TNF was seen at the lowest dose [5 mg/kg]. Consistently 5 mg/kg BW dose achieved better histological results than a dose of 10 mg/kg BW.
Role of Inflammatory Biomarkers
Frontiers in Drug Design & Discovery, 2009, Vol. 4 91
Clinical Studies - Infliximab in Ulcerative Colitis Cytokines [TNF, IL-6], cytokine receptor [TNFR] and other inflammatory marker [CRP] levels are increased in patients with ulcerative colitis [94]. The effect of infliximab on circulating levels of these markers as well as disease activity has been assessed in many clinical trials [95]. Studies by Gustot et al. showed that infliximab induced a rapid decrease in CRP at one week [0.7 vs 4.3 mg/dl; p<0.01] and four weeks [2.9 mg/dl; p<0.05] Similarly, IL-6 plasma levels decreased significantly at one week [6.2 v 24.3 pg/ml; p<0.05] and four weeks [4.5 v 24.3 pg/ml; p<0.05] after infliximab administration. Soluble TNFRII levels were also decreased at one week [1924.4 v 2570 pg/ml; p<0.05] and four weeks [2055 v 2570 pg/ml; p<0.05] treatment. In larger randomized, double-blind, placebo-controlled clinical studies in patients with moderate to severe active ulcerative colitis, infliximab decreased disease severity within 4 weeks. At week 30, clinical response, clinical remission and mucosal healing for infliximab treatment group [5 mg/kg and 10 mg/kg] groups were significantly higher than in the placebo group [p < 0.01] regardless of disease duration. These effects were maintained through the end of each trial [week 54]. Thus these data correlate reduced inflammatory responses with disease severity and demonstrates efficacy of infliximab for patients with active ulcerative colitis. Case 2 - Etanercept and Rheumatoid Arthritis Collagen-induced arthritis in rat or mouse is the most widely used animal model for the evaluation of drugs targeted for rheumatoid arthritis [96]. The disease is induced by immunization of genetically susceptible strains of mice or rats with type II collagen in adjuvant. As in human rheumatoid arthritis, a number of inflammatory cytokines are expressed in the joints of mice with collagen-induced arthritis, including TNF, IL-1 and IL-6. Similar to human disease, histopathological assessment of the joints of animals with collagen-induced arthritis reveal a proliferative synovitis with infiltration of polymorphonuclear and mononuclear cells, the formation of an erosive pannus, cartilage degradation, and fibrosis. In addition to the arthritis model, animals owhen injected with LPS [bacterial lipopolysaccharide, with or without drug treatment], casuse robust induction of inflammatory cytokines [TNF, MCP-1 and IL-6] within 1-2 h. Drug efficacy can be monitored by ability to modulate plasma or tissue levels of cytokines, post LPS challenge. Etanercept is currently the most widely used anti-arthritic drug. It is a fusion protein human soluble TNF receptor fused to the Fc component of human immunoglobulin G1 [TNFR:Fc]. In animal models injection of soluble TNF receptor completely blocks induction of TNF from 19509 ± 4682 pg/ml to no detectible TNF levels wheninjected with LPS alone or with etanercept [97]. Consistent with this, treatment with TNFR:Fc also prevents disease onset [when injected along with collagen] or reduces disease severity in established disease models [98]. In human trials, anti-TNF therapy has been linked to reduced cytokine levels [TNF-, IL-1, IL-6] in joints and serum of RA patients [99]. In clinical trials in patients with active rheumatoid arthritis, etanercept [25 mg, subcutaneous, twice a week] as a monotherapy or in combinations with methotrexate [up to 20 mg per week, oral] showed marked improvement in ACR 20 [85% and 76%], ACR50 [69% and 48%] and ACR70 [43% and 24%] scores. Etanercept has also proven to be effective when given as once a week regimen [50 mg, sc].
92 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pillarisetti and Khanna
Case 3 - Methotrexate and Rheumatoid Arthritis Methotrexate, USP [formerly Amethopterin] is an antimetabolite used in the treatment of rheumatoid arthritis and severe psoriasis. Chemically methotrexate is N-[4[[[2,4-diamino-6-pteridinyl] methyl]methylamino]benzoyl]-L-glutamic acid. Although the mechanism of anti-arthritic effects of methotrexate is not completely understood among other activities methotrexate also inhibits cytokine production by T-cells [100]. A recent study by Haroon et al. [101] showed that anti-arthritic activity of methotrexate in patients is strongly correlated with its ability to suppress T-cell cytokine production. The drug is widely used, particularly in combination with anti-TNF therapy such as etanercept, infliximab, for treatment of rheumatoid arthritis. These data suggest that monitoring plasma cytokines is a valuable tool in assessing the efficacy of anti-inflammatory agents. Several other approaches that target other inflammatory cytokines [such as IL-1 and IL-6] either as antibodies, synthesis inhibitors [such as p38, TACE, JNK-1/2, MEK] or receptor antagonist have or are being evaluated in clinic to treat arthritis. Improved understanding on the importance of cytokines in the disease process led to a paradigm shift in the treatment algorithm for RA (Table 2). Table 2.
Paradigm Shift in Rheumatoid Arthritis Treatment over 25 Years
1983
2008
Hydroxychloroquine
Infiximab
Sulfasalazine
Etanercept
Injectable gold salts
Adalimumab
Azathiopurine
Methotrexate
D-penicillamine
Leflunamide Anakinara
7. CASE STUDIES OF DRUGS THAT INDUCE INFLAMMATORY CYTOKINES AND CAUSE ORGAN TOXICITY In this section we will review examples where increased cytokine levels translated into organ toxicity in animals and humans Case 1 - Role of TNF in Cisplatin-Induced Nephrotoxicity: Cisplatin is an chemotherapeutic agent used in the treatment of a wide variety of cancers. Dose-dependent and cumulative nephrotoxicity is the major toxicity of this compound, sometimes requiring a reduction in dose or discontinuation of treatment [102]. Approximately 25–35% of patients develop evidence of nephrotoxicity following a single dose of cisplatin. Oxidant stress, caused by cisplatin-treatment is an activator of the NFB transcription factor, which, in turn, promotes the production of proinflammatory cytokines, including TNF. TNF mRNA levels are increased in cisplatin renal
Role of Inflammatory Biomarkers
Frontiers in Drug Design & Discovery, 2009, Vol. 4 93
injury [103, 104]. Injection of mice with cisplatin leads to upregulation of a number of cytokine transcripts [104], e.g., TNF-, RANTES, MCP-1. The upregulation was evident 24–48 hours after injection and sustained to 72 hours. In addition, serum and tissue level of TNF protein was induced by 8-10 fold. Urine TNF- was undetectable in salinetreated mice and was markedly increased in mice injected with cisplatin. Inhibitors of TNF- production [e.g. pentoxifylline] and TNF antibody reduced serum and kidney TNF protein levels and also blunted the cisplatin-induced increases in TNF-, MCP-1 and IL-1 [105]. In addition, the TNF inhibitors also ameliorated cisplatin-induced renal dysfunction and reduced cisplatin-induced structural damage. Likewise, TNF– deficient mice were resistant to cisplatin nephrotoxicity [106]. These results indicate cisplatin nephrotoxicity is characterized by activation of proinflammatory cytokines and chemokines. TNF appears to play a central role in the activation of this cytokine response and also in the pathogenesis of cisplatin renal injury. Case 2 - Role of Cytokines in Acetaminophen-Induced Liver Injury Acetaminophen, 4'-hydroxyacetanilide, is a non-opioid, non-salicylate analgesic and antipyretic. Intake of an overdose of acetaminophen frequently causes severe acute liver injury. Acetaminophen is metabolized by cytochrome P450 to generate a toxic metabolite, N-acetyl-p-benzoquinone imine [NAPQI], which can reduce glutathione [GSH] in the liver [107]. Thus, an overdose of acetaminophen depletes hepatic GSH, and NAPQI covalently binds to cysteine residues on proteins, resulting in the formation of 3-[cysteineS-yl] adducts, which lead to liver injury. Several lines of observations demonstrate that TNF is produced after acetaminophen challenge [108, 109]. When wild-type or TNFreceptor null mice were intraperitoneally injected with a lethal dose of acetaminophen [750 mg/kg], the mortality of TNF-receptor null mice was significantly reduced compared with WT mice. Upon treatment with a nonlethal dose [600 mg/kg] of acetminophen, WT mice exhibited an increase in serum transaminase levels. Histopathologically, centrilobular hepatic necrosis with leukocyte infiltration was evident at 10 and 24 h after acetaminophen challenge. On the contrary, serum transaminase elevation and histopathological changes were attenuated in TNF-receptor null mice injected with acetaminophen. Moreover, anti-TNF-alpha neutralizing antibodies alleviated liver injury when administered at 2 or 8 h after but not at 1 h before acetaminophen challenge to WT mice. Thus TNF mediates the pathogenic effects in acetaminophen-induced liver failure. 8. EMERGING TRENDS AND SUMMARY During the last two decades, there have been important advancements in defining role of inflammatory cytokines such as TNF, IL-1, IL-6 in disease manifestations. The methods to analyze and monitor cytokines from serum or tissues have gotten increasingly sophisticated. Pharmaceutical industry is adopting biomarker monitoring as an essential tool in pre-clinical research for target validation, proof of concept and efficacy assessment. The biomarkers are also being investigated early in clinical trials for dose selection prior to delving in long, expensive efficacy trials. The technology and advances made above have resulted in paradigm shift and identification of new treatments for inflammatory diseases such as rheumatoid arthritis. This is well illustrated in Table 2. Identification of TNF as the master cytokine in inflammatory conditions is emerging a desirable option to manage, curtail and potentially reverse disease progression. The research in areas continues to progress and emerging trends include identification of cyto-
94 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pillarisetti and Khanna
kines or immune modulators to treat patients that do not respond to anti-TNF therapy. In addition combination of anti-TNF biologics with complementary mechanism [such as leflunamide, methotrexate, Cox-2 inhibitors] are being used aggressively to treat disease from multiple angles. Orally active TNF- modulators with better tissue penetration are areas of intense investigation. Increasing examples of applications of biomarkers, showing good PK/PD correlation across therapeutic areas are emerging. The impact of this technology in drug discovery and development is likely to generate new opportunities. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41]
Springer, T.A. Cell, 1994, 76, 301. Pober, J.S. Arthritis Res., 2002, 4, 109 Lin, Y.; Berg, A.H.; Iyengar, P.; Lam, T.K.; Giacca, A.; Combs, T.P.; Rajala, M.W.; Du, X.; Rollman, B.; Li, W.; Hawkins, M.; Barzilai, N.; Rhodes, C.J.; Fantus, I.G.; Brownlee, M.; Scherer, P.E. J. Biol. Chem., 2005, 280, 4617. Kleemann, R.; Kooistra, T. Curr. Drug Targets Cardiovasc. Haematol. Disord., 2005, 5 441 Berg, A.H.; Scherer, P.E. Circ. Res., 2005, 96, 939 Natali, A.; Toschi, E.; Baldeweg, S.; Ciociaro, D.; Favilla, S.; Saccà, L.; Ferrannini, E. Diabetes, 2006, 55, 1133. Karin, M. Proc. Am. Thorac. Soc., 2005, 2, 386 Nakano, H.; Nakajima, A.; Sakon-Komazawa, S.; Piao, J.H.; Xue, X.; Okumura, K. Cell Death Differ., 2006, 13, 730. Lin, W.W.; Karin, M. J. Clin. Invest., 2007, 117, 1175. Alikunju, S.; Pillarisetti, S. Med. Chem., 2006, 6, 461. Dandona, P.; Aljada, A.; Chaudhuri, A.; Mohanty, P.; Garg, R. Circulation, 2005 111, 1448. Wellen, K.E.; Hotamisligil, G.S. J. Clin. Invest., 2005, 15, 1111. Libby, P. Nature, 2002, 420, 868. Pradhan, A.D.; Ridker, P.M. Eur. Heart J., 2002, 11, 831. Pillarisetti, S.; Saxena, U. Expert. Opin. Ther. Targets, 2004, 8, 401. Fornoni, A.; Ijaz, A.; Tejada, T.; Lenz, O. Curr. Diabetes Rev., 2008, 4, 10. Wyss-Coray, T. Nat. Med., 2006, 12, 1005. Weninger, S.C.; Yankner, B.A. Nat. Med., 2001, 7, 527. Locksley, R.M., Killeen, N.; Lenardo, M.J. Cell, 2001, 104, 487. McDermott, M.F. Cell. Mol. Biol., 2001, 47, 619. Popa, C.; Netea, M.G.; van Riel, P.L.; van der Meer, J.W., Stalenhoef, A.F. J. Lipid Res., 2007, 48, 751. Garg, A.K.; Aggarwal, B.B. Mol. Immunol., 2002, 39, 509. Madge, L.A.; Pober, JS. Exp. Mol. Pathol., 2001, 70, 317. Zink, A.; Strangfeld, A.; Schneider, M.; Herzer, P.; Hierse, F.; Stoyanova-Scholz, M.; Wassenberg, S.; Kapelle, A. Arthritis Rheum., 2006, 54, 3399. Akobeng, A.K.; Zachos, M. Cochrane Database Syst. Rev., 2004, CD003574. Cordoro, K.M.; Feldman, S.R. Skin Therapy Lett., 2007, 12, 4. Dinarello, C.A. Clin. Exp. Rheumatol., 2002, 20, S1. Ringwood, L.; Li, L. Cytokine, 2008, 42, 1. Solt, L.A.; Madge, L.A.; Orange, J.S.; May, M.J. J. Biol. Chem., 2007, 282, 8724. Brigelius-Flohé, R.; Banning, A.; Kny, M.; Böl, G.F. Arch. Biochem. Biophys., 2004, 423, 66. Sims, J.E. Curr. Opin. Immunol., 2002, 14, 117. Braddock, M.; Quinn, A. Nat. Rev. Drug Discov., 2004, 3, 330. Rose-John, S.; Scheller, J.; Elson, G.; Jones, S.A. J. Leukoc. Biol., 2006, 80, 227. Rose-John, S.; Waetzig, G.H.; Scheller, J.; Grötzinger, J.; Seegert, D. Expert. Opin. Ther. Targets., 2007, 11, 613. Nishimoto, N.; Kishimoto, T. Curr. Opin. Pharmacol., 2004, 4, 386. Daly, C.; Rollins, B.J. Microcirculation, 2003, 10, 247. Szalai, A.J. Clin. Dev. Immunol., 2004, 11, 221. Yeh, E.T. Circulation, 2004, 109, 1. Read, S.J.; Dray, A. Expert Opin. Investig. Drugs, 2008 17, 619. McNerney, T.; Baethge, B.A.; Cao, S. Clin. Exp. Immunol., 2004, 137, 621. Schumacher, R.H.; Meng, Z.; Sieck, M. J. Rheumatol., 1996, 23, 1774.
Role of Inflammatory Biomarkers [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 95
Westacott, C.I.; Whicher, J.T.; Barnes, I.C. Ann. Rheum. Dis., 1990, 49, 676. Bertazzolo, N.; Punzi, L.; Stefani, M.P. Agents Actions, 1994, 41, 90. Kutukculer, N.; Caglayan, S.; Aydogdu, F. Clin. Rheumatol., 1998, 17, 288. Marinova-Mutafchieva, L.; Williams, R.O.; Mason, L.J.; Mauri, C.; Feldmann, M.; Maini, R.N. Clin. Exp. Immunol., 1997, 107, 507. Zvezdanovi, L. Jugoslov Med. Biohem., 2006, 25, 363. Kagari, T.; Doi, H.; Shimozato, T. J. Immunol., 2002, 169, 1459. Abramson, S.B.; Amin, A. Rheumatology, 2002, 41, 972. Nishimoto, N.; Kishimoto, T. Handb. Exp. Pharmacol., 2008, 181, 151. Libby, P. Nature, 2002, 420, 868. Ross, R. N. Engl. J. Med., 1999, 340, 115. Pai, J.K.; Pischon, T.; Ma, J.; Manson, J.E.; Hankinson, S.E.; Joshipura, K.; Curhan, G.C.; Rifai, N.; Cannuscio, C.C.; Stampfer, M.J.; Rimm, E.B. N. Engl. J. Med., 2004, 351, 2599. Ridker, P.M.; Hennekens, C.H.; Buring, J.E.; Rifai, N. N. Engl. J. Med., 2000, 342, 836. Ridker, P.M.; Cushman, M.; Stampfer, M.J.; Tracy, R.P.; Hennekens, C.H. Circulation, 1998, 97, 425. Volpato, S.; Guralnik, J.M.; Ferrucci, L.; Balfour, J.; Chaves, P.; Fried, L.P.; Harris, T, B. Circulation, 2001, 103, 947. Ridker, P.M.; Rifai, N.; Stampfer, M.J.; Hennekens, C.H. Circulation, 2000, 101, 1767. Wang, Y.N.; Che, S.M.; Ma, A.Q. Chin. Med. Sci. J., 2004, 191, 20. Jones, K.G.; Brull, D.J.; Brown, L.C.; Sian, M.; Greenhalgh, R.M.; Humphries, S.E.; Powell, J.T. Circulation, 2001, 103, 2260. Ferroni, P.; Rosa, A.; Di Franco, M.; Palmirotta, R.; Guadagni, F.; Davì, G.; Bertazzoni, G.; Basili, S. Clin. Chim. Acta. 2007, 381, 151. Biasucci, L.M.; Liuzzo. G.; Fantuzzi, G.; Caligiuri, G.; Rebuzzi, A.G.; Ginnetti, F.; Dinarello, C.A.; Maseri, A. Circulation, 1999, 99, 2079. Luo, Y.; Jiang, D.; Wen, D.; Yang, J.; Li, L. Heart Vessels, 2004, 19, 257. Kinlay, S.; Schwartz, G.G.; Olsson, A.G.; Rifai, N.; Leslie, S.J.; Sasiela, W.J.; Szarek, M.; Libby, P.; Ganz, P. Circulation, 2003, 108, 1560. Pradhan, A. Nutr. Rev., 2007, 65, S152. Navab, M.; Gharavi, N.; Watson, A.D. Curr. Opin. Clin. Nutr. Metab. Care, 2008, 11, 459. Gustafson, B.; Hammarstedt, A.; Andersson, C.X.; Smith, U. Arterioscler. Thromb. Vasc. Biol., 2007, 27, 2276. Pradhan, A.D.; Manson, J.E.; Rifai, N.; Buring, J.E.; Ridker, P.M. JAMA, 2001, 286, 327. Lang, C.H.; Dobrescu, C.; Bagby, G.J. Endocrinology, 1992, 130, 43. Uysal, K.T.; Wiesbrock, S.M.; Marino, M.W.; Hotamisligil, G.S. Nature, 1997, 389, 610. Liang, H.; Yin, B.; Zhang, H.; Zhang, S.; Zeng, Q.; Wang, J.; Jiang, X.; Yuan, L.; Wang, C.Y.; Li, Z. Endocrinology, 2008, 149, 2943. Haffner, S.M.; Greenberg. A.S.; Weston, W.M.; Chen, H.; Williams, K.; Freed, M.I. Circulation, 2002, 106, 679. Deepa, R.; Velmurugan. K.; Arvind, K.; Pillarisetti, S.; Sientay, C.; Uday, S.; Mohan, V. Metabolism, 2006, 55, 1232. Ding, W.X.; Yin, X.M. J. Cell. Mol. Med., 2004, 8, 445. Bradham, C.A.; Plümpe, J.; Manns, M.P.; Brenner. D.A.; Trautwein, C. Am. J. Physiol., 1998, 275, G387. Bird, G.L.A.N.; Sheron, J.; Goka, G.J.; Alexander, R.S. Ann. Intern. Med., 1990, 112, 917. Gantner, F.M.; Leist, A.W.; Lohse, P.G.; Germann, A. Hepatology, 1995, 21, 190. Gonzalez-Amaro, R.C.; Garcia-Monzon, L.; Garcia-Buey, R.; Moreno-Otero, J. L.; Lonso, E.; Yague, J.P.; Pivel, M.; Lopez-Cabrera, E.; Fernandez-Ruiz, F. J. Exp. Med., 1994, 179, 841. Iimuro, Y.; Gallucci, R.M.; Luster, M.I.; Kono, H.; Thurman, R.G. Hepatology, 1997, 26, 1530. Küsters, S.; Tiegs, G.; Alexopoulou, L.; Pasarakis, M.; Douni, E.; Künstle, G.; Blüthmann, H.; Wendel, A.; Pfizenmaier, K.; Kollias, G.; Grell, M. Eur. J. Immunol., 1997, 27, 2870. Leist, M.; Gantner, F.; Jilg, S.; Wendel, A. J. Immunol., 1995, 154, 1307. Muto, Y.; Nouri-Aria, K.T.; Meager, A.; Alexander, G.J.; Eddleston, A.L.; Williams, R. Lancet, 1998, 8602, 72. Barton, C.C.; Barton, E.X.; Ganey, P.E.; Kunkel, S.L.; Roth, R.A. Hepatology, 2001, 33, 66. Künstle, G.; Hentze, H.; Germann, P.G.; Tiegs, G.; Meergans, T.; Wendel, A. Hepatology, 1999, 30, 1241. Biburger, M.; Tiegs, G. J. Immunol., 2005, 175, 1540.
96 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109]
Pillarisetti and Khanna
Stenvinkel, P. Kidney Int., 2005, 67, 1216. Avesani, C.M.; Carrero, J.J.; Axelsson, J.; Qureshi, A.R.; Lindholm, B.; Stenvinkel, P. Partners Crime Kidney Int., 2006, 70, S8. Baud, L.; Ardaillou, R. Miner. Electrolyte Metab., 1995, 21, 336. Mullin, J.M.; Snock, K.V. Cancer Res., 1990, 50, 2172. Marano, C.W.; Laughlin, K.V.; Russo, L.M.; Peralta Soler, A.; Mullin, J.M. J. Cell. Physiol., 1993, 157, 519. Liu, M.; Chien, C.C.; Burne-Taney, M.; Molls, R.R.; Racusen, L.C.; Colvin, R.B.; Rabb, H. J. Am. Soc. Nephrol., 2006, 177, 65. Faubel, S.; Lewis, E.C.; Reznikov, L.; Ljubanovic, D.; Hoke, T.S.; Somerset, H.; Oh, D.J.; Lu, L.; Klein, C.L.; Dinarello, C.A.; Edelstein, C.L. J. Pharmacol. Exp. Ther., 2007, 322, 8. Griffin, W.S. Am. J. Clin. Nutr., 2006, 83, 470S. Lucas, S.M.; Rothwell, N.J.; Gibson, R.M. Br. J. Pharmacol., 2006, 147, S232. Triantafillidis, J.K.; Papalois, A.E.; Parasi, A.; Anagnostakis, E.; Burnazos, S.; Gikas, A.; Merikas, E.G.; Douzinas, E.; Karagianni, M.; Sotiriou, H. World J. Gastroenterol., 2005, 11, 6843. Komatsu, M.; Kobayashi, D.; Saito, K.; Furuya, D.; Yagihashi, A.; Araake, H.; Tsuji, N.; Sakamaki, S.; Niitsu, Y.; Watanabe, N. Clin. Chem., 2001, 47, 1297. Wilhelm, S.M.; McKenney, K.A.; Rivait, K.N.; Kale-Pradhan, P.B. Clin. Ther., 2008, 30, 223. Holmdahl, R. Methods Mol. Med., 2007, 136, 185. Mohler, K.M.; Torrance, D.S.; Smith, C.A.; Goodwin, R.G.; Stremler, K.E.; Fung, V.P; Madani, H.; Widmer, M.B. J. Immunol., 1993, 151, 1548. McComb, J.; Gould, T.; Chlipala, E.; Sennelo, G.; Frazier, J.; Kieft, G.; Seely, J.; Edwards, C.K. 3rd.; Bendele, A. J. Rheumatol., 1999, 26, 1347. Feldmann, M. Nat. Rev. Immunol., 2002, 2, 364. Swierkot, J.; Szechiski, J. Pharmacol. Rep., 2006, 58, 473. Haroon, N.; Srivastava, R.; Misra, R.; Aggarwal, A. J. Rheumatol., 2008, 35, 975. Hanigan, M.H.; Devarajan, P. Cancer Ther., 2003, 1, 47. Dong, Z.; Atherton, S.S. Kidney Int., 2007, 72, 5. Ramesh, G.; Reeves, W.B. J. Clin. Invest., 2002, 110, 835. Kim, Y.K.; Choi, T.R.; Kwon, C.H.; Kim, J.H.; Woo, J.S.; Jung, J.S. Ren. Fail., 2003, 6, 909. Ramesh, G.; Reeves, W.B. Am. J. Physiol. Renal Physiol., 2003, 285, F610. Larson, A.M. Clin. Liver Dis., 2007, 11, 525. Jaeschke, H. Expert Opin. Drug Metab. Toxicol., 2005, 1, 389. Ishida, Y.; Kondo, T.; Tsuneyama, K.; Lu, P.; Takayasu, T.; Mukaida, N. J. Leukoc. Biol., 2004, 75, 59.
Frontiers in Drug Design & Discovery, 2009, 4, 97-121
97
Important Drug Interactions for Clinical Oncologists Hiroshi Ishiguro1,*, Ikuko Yano2 and Masakazu Toi3 1
Translational Research Center, Department of Clinical Trial Management / Outpatient Oncology Unit; 2Department of Pharmacy and 3Breast Surgery Department, Kyoto University Hospital, 54 Shogoinkawaharara-cho,Sakyo-ku, Kyoto-city, 606-8507, Japan Abstract: Drug interactions can cause severe side effects and lead to early termination of drug development, refusal of drug approval, prescribing restrictions or drug withdrawal from the market. Of drugs used to treat humans, cytotoxic anti-neoplastic drugs have a particularly strong action. Furthermore, they have a complex pharmacological profile, a narrow therapeutic window, and a steep dose-toxicity curve, and are associated with considerable inter- and intrapatient pharmacokinetic and pharmacodynamic differences. The recommended dose is usually close to the maximally tolerated dose in order to achieve the maximum therapeutic effect. Thus, some adverse effects are usually inevitable, so these drugs are approved for usage based on a clinical risk to benefit ratio. Therefore, drug interactions affecting the pharmacokinetics of anti-neoplastic drugs are of particular concern. Any physicians treating oncology patients must understand the pharmacokinetic behavior (absorption, distribution, metabolism, excretion, etc.) of a drug, as well as the factors affecting its pharmacokinetic behavior, for example the effects of concomitantly administered drugs, and hepatic and renal function. Medical oncologists must have expertise in achieving a good balance between safety and efficacy in medical treatment, with a proper knowledge of supportive care and an understanding of pharmacokinetics, pharmacodynamics and pharmacogenomics. We summarize the drug interactions that are important in day-to-day oncology practice. We cover pharmaceutical interactions as well as the interactions at the levels of absorption, distribution, metabolism and excretion. This review is the one of the most comprehensive to date in the field of clinical oncology, where the level of understanding of drug interactions can directly affect patient management.
INTRODUCTION Drug interactions can cause severe side effects and lead to the early termination of drug development, refusal of drug approval, severe prescribing restrictions or withdrawal of drugs from the market [1]. For example, in the 1980s, terfenadine was a widely used antihistamine, representing 60% of all antihistamine prescriptions written in the United States. After a case of fatal arrhythmia (Torsades de Point) in 1990, this drug was *Corresponding Author: Tel: +81-75-751-4770; Fax: +81-75-751-4772; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
98 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
removed from the market worldwide [2]. This is despite terfenadine cardiotoxicity having been only reported for overdoses. The problem arose because this drug is nearly entirely bio-transformed by the CYP3A enzyme system before entering the systemic circulation. When the CYP3A enzyme system is inhibited (e.g. by the simultaneous administration of erythromycin or ketoconazole) or overwhelmed (e.g. because of an overdose), or when its activity is reduced by disease (e.g. cirrhosis), the concentration of unmetabolized terfenadine entering the systemic circulation increases markedly, resulting in considerable prolongation of the QT interval [3]. Cytotoxic anti-neoplastic drugs have some of the strongest actions of drugs used clinically. They have a complex pharmacological profile, a narrow therapeutic window (Fig. 1), a steep dose-toxicity curve, and there are considerable inter- and intra-patient pharmacokinetic and pharmacodynamic differences [4]. In fact, some adverse effects are usually inevitable; therefore, usage of these drugs is based on assessment of the clinical risk to benefit ratio. For example, arsenic trioxide, which has been approved for the treatment of recurrent or refractory acute promyelocytic leukemia in Japan, was associated with a prolongation of QT of grade 3 or greater in more than 10% of patients in the Japanese registration trial [5]. In light of this finding, the drug was approved with the condition that a post-marketing study be performed for of all cases for 6 years after approval [5]. The recommended dose of cytotoxic anti-neoplastic drugs is usually close to the maximally tolerated dose in order to achieve maximum therapeutic effect. Therefore, drug interactions affecting the pharmacokinetics of anti-neoplastic drugs are of particular concern. Physicians engaged in treating oncology patients, especially with treatments involving anti-neoplastic drugs, must understand their pharmacokinetic behavior (absorption, distribution, metabolism and excretion, etc.) as well as the factors affecting this behavior (e.g. concomitantly administered drugs and hepatic and renal function). Medical oncologists must have expertise in achieving a good balance between safety and efficacy, with proper knowledge of supportive care, as well as an understanding of pharmacokinetics, pharmacodynamics and pharmacogenomics. In this review, we summarize the drug interactions that are important in day-to-day oncology practice. We have also included some non-antineoplastic drug interactions that may be relevant for clinical oncologists. TYPES OF DRUG-INTERACTIONS Pharmaceutical Interactions Some drugs require specific diluents in order to prevent precipitation, so possible incompatibility must be checked before drug preparation. Ideally, these drugs would be prepared by pharmacists that have been specifically trained in the preparation of antineoplastic drugs. Amphotericin B should be diluted in a non-electrolyte containing solution such as sterile water or 5% glucose solution, since it is likely to precipitate in sodium chloride solution [6]. Some penicillins, including carbenicillin, ticarcillin and piperacillin, have been shown to inactivate aminoglycosides in vitro [7]. This has been observed to a great extent with tobramycin and gentamycin, while amikacin has proven more stable against inactivation [7]. Concurrent use of these agents may pose a risk of reduced antibacterial
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 99
efficacy in vivo, particularly in the setting of profound renal impairment [7]. If combined penicillin/aminoglycoside therapy is desired in a patient, especially one with renal dysfunction, the clinician should consider separation of doses and routine monitoring of aminoglycoside level and clinical response [7]. A
Adverse Event
Response
Effect
Therapeutic window
Dose
Response
Effect
Therapeutic window
B
Adverse Event
Dose Fig. (1). Schematic illustration of the sizes of the therapeutic windows for general (A) and antineoplastic (B) drugs. Figure modified from reference [69]. The therapeutic windows are delineated by the vertical dashed lines, and the widths of the windows are indicated by the double-headed arrows. Variations in response are also illustrated, with the bold line indicating the average response.
100 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
Cisplatin is known to be unstable in aqueous solution. The primary mode of decomposition involves displacement of the chloride ligand, and so increasing the chloride ion concentration improves the stability of the drug in solution. In a solution with low sodium chloride concentration (e.g. a 5% glucose solution), cisplatin decomposes rapidly for the first 3 h, then gradually reaches an equilibrium at about 80% of the initial concentration (Fig. 2A) [8, 9]. However, in a 0.9% sodium chloride solution, the cisplatin concentration decreases by no more than 3% over a 24-h period (Fig. 2A) [8, 9]. Cisplatin may also become inactivated in an amino acid solution in as little as 4 h [10].
A
Cisplatin concentration (%)
Mesna, which is used to reduce the incidence of hemorrhagic cystitis arising from the use of ifosfamide or cyclophosphamide, may bind to cisplatin cations, thus inhibiting cisplatin DNA binding and leading to a reduced anti-tumor effect [11]. When mitomycin is dissolved in 5% glucose infusion fluid (pH 4–5), it is rapidly degraded into inactive mitosenes (Fig. 2B) [4]. Mitomycin changes color from bluish violet to reddish violet when pH and activity decrease [12]. 100 95
NS D5W
90 85 80 0
1
2
3
4
6
B
Mitomycin C concentration (%)
Hours
100
NS D5W
95
90
85 0
1
3
6
24
48
Hours Fig. (2). Stability of cisplatin (A) [8, 9] and mitomycin C (B) [12] in different solutions in vitro. NS, normal saline; D5W, 5% glucose solution.
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 101
Even small modifications of the dose and schedule of two drugs can have a profound effect on the plasma disposition of the drugs [13]. In one clinical study, when a 24-h infusion of paclitaxel preceded a 48-h infusion of doxorubicin, the end-of-infusion plasma doxorubicin concentration (Cmax) was 70% higher and the doxorubicin clearance was 32% lower on average compared with the reverse sequence (Fig. 3A,B), which resulted in a higher incidence of grade 2 and 3 stomatitis and hematological toxicity [14]. In another study, even when doxorubicin was given first, if the doxorubicin bolus was immediately followed by the paclitaxel infusion, such that the two drugs were both present at high concentrations, paclitaxel caused prolongation of the distribution and elimination phases of doxorubicin and its active metabolite, doxorubicinol (Fig. 3C, D) [13]. Doxorubicin clearance and area under the curve (AUC) are also significantly altered in the presence of Cremophor EL (BASF Corp.), an emulsifying agent and aqueous solubilizer
A N=8 p = 0.0021
2000 1000 0
60 40 20 0
Dox ⇒ Pac
Pac ⇒ Dox
Dox ⇒ Pac
C
4000
D
p < 0.05
N=8 p < 0.05
24 hours 2000 0
15 minutes
Pac ⇒ Dox
N=8 Not significant
40
CL (L/h/m2)
AUC (mg/L/h)
6000
N=8 p = 0.0062
80
CL (L/h/m2)
3000
AUC (μg/L/h)
B
20
24 hours 15 minutes
0
Doxorubicin
Fig. (3). AUC (A,C) and clearance (B,D) for doxorubicin in patients with breast cancer, according to administration sequence (A,B) [14], and the interval between administration of doxorubicin and paclitaxel (C,D) [13]. Doxorubicinol, a metabolite of doxorubicin, is also shown in (C). (A,B) Doses used were: doxorubicin 60 mg/m2 and paclitaxel 125 mg/m2. (C,D) Doses used were: doxorubicin 48 mg/m2 and paclitaxel 125 mg/m2. CL, clearance; Pac, paclitaxel; Dox, doxorubicin. Error bars represent standard deviation.
102 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
for hydrophobic agents such as paclitaxel. In the presence of Cremophor EL, doxorubicin AUC has been found to be increased by about 30% and the doxorubicinol AUC to be elevated by more than 100% (Fig. 4) [15]. These findings have been postulated to explain the increased incidence of doxorubicin cardiotoxicity that has been observed in some studies [16]. Similarly, a significantly larger docetaxel AUC value has been observed when docetaxel follows doxorubicin than when docetaxel is administered alone [17].
A
Cremophor EL not received Cremophor EL received
AUC (ng/mL・h)
3000 p = 0.02
2000
N = 11 p = 0.02
1000
0 Doxorubicin
Doxorubicinol
B
CL (mL/min/m2)
1000
Cremophor EL not received N = 11 p = 0.02
Cremophor EL received
800 600 400 200 0 Doxorubicin
Fig. (4). AUC (A) and clearance (B) of doxorubicin and its metabolite doxorubicinol in patients with advanced cancer according to whether they had also received Cremophor EL [15]. Doses used were: doxorubicin 60 mg/m2 and Cremophor EL 30mL/m2. CL, clearance. Error bars represent standard deviation.
Paclitaxel has also been shown to alter the pharmacokinetic profile of epirubicin. Sequence-dependent pharmacokinetic interactions between paclitaxel and epirubicin
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 103
have been described for breast cancer patients receiving either paclitaxel (175 mg/m2) followed by epirubicin (90 mg/m2) or the reverse sequence. The treatment sequence with paclitaxel first was associated with a higher level of hematological toxicity and a 36% increase in the AUC of epirubicin relative to the reverse sequence (Fig. 5A, B), which might also have contributed to some extent to the higher risk of epirubicin cardiotoxicity [18]. Overall, the combination of taxanes and anthracyclines is associated with consistent elevation of anthracycline plasma level and enhanced toxicity when the taxane is administrated concurrently with or just prior to the anthracycline. Therefore, the minimum recommended interval between administration of the two agents is 24 h, and it is recommended that the anthracycline be administered first [16]. In a phase 1 study in which the effects of sequence and dose for combined paclitaxel and cisplatin treatment were investigated, myelosuppression was more pronounced when paclitaxel was given after cisplatin compared with the reverse sequence [19]. This finding was explained by a 25% lower paclitaxel clearance (Fig. 5C) when cisplatin preceded paclitaxel than vice versa, possibly because of the effect of cisplatin on paclitaxelmetabolizing CYP enzymes [4, 19]. Absorption Although oral administration of chemotherapeutic agents has a number of advantages over other methods, drug bioavailability is limited and variable when delivered orally. The actions of drug transporters and CYP isoenzymes (e.g. CYP3A4 and CYP3A5) in the intestinal epithelium seem to be major obstacles to efficient drug uptake [4]. Quinolones inhibit DNA gyrase (topoisomerase II) in susceptible organisms, thereby inhibiting the relaxation of supercoiled DNA and promoting the breakage of DNA strands. When present, metal cations (aluminum, calcium, iron, magnesium, and zinc) bind to quinolones in the gastrointestinal tract and inhibit their absorption. Concurrent administration of quinolones with most antacids, oral electrolyte supplements, quinapril and sucralfate should be avoided where possible [20]. If these agents are necessary, quinolones should be taken at least 2 h before or after these agents [20]. For levofloxacin in particular, the peak/minimum inhibitory concentration (MIC) ratio is a better predictor of clinical and microbiological outcome than the AUC/MIC ratio [21]. Therefore, both dose splitting, which is recommended in the package insert of Cravit (levofloxacin) in Japan [20], and inhibition of absorption by concomitantly administered drugs are likely to reduce the anti-biological activities of quinolones. Treatment with allopurinol can lead to an increased level of mercaptopurine via inhibition of xanthine oxidase-catalyzed metabolic breakdown of mercaptopurine in the intestine and liver. Therefore, when allopurinol and mercaptopurine are used concomitantly, the dose of mercaptopurine should be reduced by 75% [22]. Absorption of gefitinib is affected by gastric pH, and it has been reported that the AUC for gefitinib is reduced by 50% when gastric pH is higher than 5 [23]. Patients with achlorydia (which occurs in many elderly patients), those that have undergone stomach resection, and those taking H2 blockers or proton pump inhibitors may obtain suboptimal benefit from gefitinib treatment [23]. In order to achieve maximum benefit, gefitinib is usually taken after meal, which promotes gastric acid secretion, in turn increasing
104 Frontiers in Drug Design & Discovery, 2009, Vol. 4
A
Ishiguro et al.
N = 15 p = 0.02
AUC (μg/L/h)
4000 3000 2000 1000 0 Epi ⇒ Pac
Pac ⇒ Epi N = 15 p = 0.02
B
CL (L/h/㎡)
60
30
0 Epi ⇒ Pac
Pac ⇒ Epi
C N = 15 P < 0.001
CL (mL/min/㎡)
600
400
200
0
Pac ⇒ CDDP
CDDP ⇒ Pac
Fig. (5). AUC (A) and clearance (B) of epirubicin when concomitantly administered with paclitaxel to patients with early-stage breast cancer, according to administration sequence [18]. Doses used were: epirubicin 90 mg/m2 and paclitaxel 175mg/m2. Clearance (C) of paclitaxel when concomitantly administered with cisplatin to patients with solid tumors, according to administration sequence [19]. Doses used were: paclitaxel 110-200 mg/m2 and cisplatin 50-75mg/m2. CL, clearance; Pac, paclitaxel; Epi, epirubicin; CDDP, cisplatin; Error bars represent standard deviation.
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 105
AUC and maximum plasma concentration (Cmax) by 37% and 32%, respectively, compared with an empty stomach [23]. For erlotinib, however, the manufacturer of the product marketed in Japan (Tarceva; Chugai) recommends administration on an empty stomach (at least 1 h before or 2 h after the ingestion of food) [24], even though this reduces drug absorption by approximately half relative to the same dose taken after a meal [24]. Thus, administration of erlotinib after a meal results in a doubling of the AUC [24]. Absorption of erlotinib is also affected by gastric pH and concomitant administration of proton pump inhibitors, which reduce the AUC by about half [24]. Protein Binding Anticancer drugs can bind to several blood components, including albumin, 1 acid glycoprotein, lipoproteins, immunoglobulin, and erythrocytes. When a drug is displaced from its plasma protein as a result of drug-drug interactions or disease-drug interactions, the concentration of biologically active unbound drug in the blood increases, potentially causing an increase in effect and/or toxicity. The net pharmacodynamic effect, however, is difficult to predict because drug displacement causing an increase in the free fraction not only makes the drug more available for its target, but also assists with metabolic and renal elimination [4]. Therefore, a change in plasma protein binding will usually only transiently influence the clinical exposure of a patient to a drug [25]. The cases described below were originally thought to be caused by a change in plasma protein binding, but other factors, such as inhibition of hepatic clearance (which will be discussed later), may actually be more significant. In a patient being treated with warfarin, after administration of paclitaxel, there was an increase in the international normalization ratio (INR): on day 2 of the first cycle, a warfarin INR value of 5.2 was observed, relative to a pre-chemotherapy INR of 3.0 [26]. A post-treatment rise was observed after all subsequent cycles. Given that 95-98% of paclitaxel is bound to plasma proteins, paclitaxel could conceivably displace warfarin from plasma protein binding sites. In the patient described above, that the warfarin INR increase occurred so soon after paclitaxel administration suggested that displacement of warfarin was the primary mechanism, but more recent findings indicate that this may not be the case [26]. The present authors have also experienced a similar phenomenon in a patient treated with warfarin and paclitaxel (unpublished data). Clearly, in patients treated with warfarin and paclitaxel, the warfarin INR must be monitored closely. There have been several reports of cases where warfarin effect has been potentiated by an etoposide-containing regimen [27]. Like paclitaxel, a very high proportion of etoposide (97%) binds to plasma proteins [27], leading to suggestions that it may bring about an increased warfarin INR in a similar manner to paclitaxel. In one patient being treated with warfarin, the INR value was found to be elevated within 24 h of receiving etoposide therapy [27]. In that patient, the INR value returned to baseline in 3 days [27]. For ifosfamide, an analogue of cyclophosphamide, 12-24% of molecules are bound to plasma proteins, but the proportion is higher for its metabolites [28]. Hall et al. described three patients who experienced serious disturbance of warfarin INR values while receiving ifosfamide/mesna-containing chemotherapy. In their cases, a previously stable warfarin INR became elevated to 5.5, 7.5 and 8.4, respectively [28]. Since warfarin has a very low hepatic extraction ratio and a long equilibration time, these phenomena are unlikely to be due to a change in plasma protein binding alone.
106 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
Other factors, such as inhibition of hepatic clearance, might also contribute to the effect on pharmacodynamics of warfarin [25]. Metabolism The hepatic CYP system is the major machinery for drug metabolism, and most drug-drug interactions arise because of their impact on this system [29, 30] (Table 1). For orally administered drugs, gut wall CYP3A is probably of major importance in drugdrug and drug-xenobiotic interactions [4]. Many anticancer drugs are cleared by the CYP3A4 system. There exists, therefore, the potential for drug-drug interactions between drugs that share the same CYP machinery, whether they are cytotoxic or noncytotoxic drugs. Table 1.
Major Routes of Elimination for Selected Anti-Neoplastic Agents
Class
Drug
Major Activating Enzyme
Major Inactivating Enzyme
Alkylating agent
Cyclophosphamide
CYP2B6
CYP3A4
Ifosfamide
CYP3A4
Antimetabolite
Busulfan
CYP3A4
5-FU
DPD
Capecitabine
Carboxyesterase Cytidine deaminase
DPD
Methotrexate
Antibiotic agent
Renal
Gemcitabine
Cytidine deaminase
Mercaptopurine
Xanthine oxidase
Doxorubicin
CYP2D6, CYP3A4
Epirubicin
Unknown
Bleomycin Anti-microtubule agent
Topoisomerase inhibitor
Renal
Vincristine
CYP3A4
Vinorelbine
CYP3A4
Irinotecan
Other Method of Excretion
Carboxyesterase
UGTA1A, CYP3A4
Paclitaxel
CYP2C8, CYP3A4
Docetaxel
CYP3A4
Etoposide
CYP3A4
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 107 (Table 1) contd....
Major Activating Enzyme
Major Inactivating Enzyme
Other Method of Excretion
Class
Drug
Platinum agent
Carboplatin
Renal
Cisplatin
Renal
Molecular targeted agents
Hormonal agent
Imatinib
CYP3A4
Gefitinib
CYP3A4
Tretinoin
CYP3A4
Tamoxifen
CYP2D6, CYP3A4
Toremifene
CYP3A4
Anastrozole
CYP3A4
Letrozole
CYP2A6, CYP3A4
Exemestane
CYP3A4
Progesterone
CYP3A4
Flutamide
CYP3A4
Increased Toxicity The AUC of midazolam, which is metabolized by CYP3A4, can be elevated two to 16-fold by moderate to strong inhibitors of CYP3A4, such as anti-fungal medications (fluconazole, itraconazole), macrolide antibiotics (erythromycin, clarithromycin), nondihydropyridine calcium channel blockers (diltiazem, verapamil) and even grapefruit juice [1]. The magnitude of inhibition depends on the drug, its dose, and the duration of administration. The inhibitory activity of a concomitantly administered drug is generally classified as strong (>5), moderate (~5) or weak (<2) based on the fold increase in midazolam plasma AUC [1] (Table 2). Although classified as “weak”, a two-fold increase in AUC for anti-neoplastic drugs is often very undesirable, since many of these agents have very narrow therapeutic windows. Metabolism of the vinca alkaloids generally appears to be mediated by the CYP3A subfamily. There have been several cases of severe neurotoxicity or even death possibly resulting from drug interactions between intravenously administered vinca alkaloids and CYP inhibitors such as the antifungal agent itraconazole [31-33]. Irinotecan is converted into its pharmacologically active metabolite, SN-38, by carboxyesterase. SN-38 is then inactivated by conversion to SN-38G by uridine diphosphate-glucuronosyltransferase (UGT) 1A1. A specific genetic polymorphism of UGT1A1 is known to be related to a higher level of irinotecan toxicity, and a diagnostic tool for this polymorphism is now available [34]. Irinotecan is also inactivated by CYP3A4, so
108 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 2.
Ishiguro et al.
CYP3A4 Inhibitors and Degree of Inhibition [1]
Strength of CYP3A4 Inhibition
Dose
CYP3A4 Substrate AUC (Fold Increase Relative to Baseline Value)
200 mg daily for 4 days
8-11
100 mg daily for 4 days
6
Clarithromycin
1000 mg daily for 7 days
7
Fluconazole
400 mg
3
Clarithromycin
500 mg daily for 5 days
3.6
Erythromycin
1500 mg daily for 5 days
3.8
Diltiazem
180 mg daily for 2 days
3.7
Verapamil
240 mg daily for 2 days
2.9
Grapefruit juice
~250 mL daily for 4 days
2.4
Grapefruit juice
200 mL
1.5
Azithromycin
400 mg daily for 3 days
1.3
Cimetidine
800 mg
1.3
Drug
Itraconazole Strong
Moderate
Weak
co-administration of a strong CYP3A4 inhibitor, ketoconazole, leads to significantly increased (109%, p = 0.04) levels of SN-38 [35]. Docetaxel is extensively metabolized by CYP3A4, and co-administration of ketoconazole has been found to decrease docetaxel clearance by 50% and increase AUC by more than two-fold (Fig. 6A, B) [36]. The AUC of docetaxel is reportedly a significant predictor of severe neutropenia, with a 50% decrease in docetaxel clearance corresponding to a 4.3-fold increase in the risk of developing grade 4 neutropenia and a 3.0-fold increase in the risk of developing febrile neutropenia [37]. Although a single dose of cimetidine (800 mg orally) does not significantly affect the AUC of midazolam (producing only a 1.3-fold increase) [1], the results of a study involving a small number of patients showed that 5 days of cimetidine treatment (400 mg orally twice per day) increased the AUC of epirubicin (a 50% increase, which was statistically not significant due to a small sample size) and its active metabolite, epirubicinol (41% increase) (Fig. 7A) [38]. These changes in exposure could not be explained by reduced cytochrome P-450 activity alone [38]. Cimetidine also reportedly affects the pharmacokinetics of 5-fluorouracil (5-FU): pretreatment with cimetidine for 4 weeks led to increased plasma concentration and AUC of 5-FU [39]. For oral 5-FU, the peak plasma concentration was increased by 74% and the AUC was increased by 72% [39]. For intravenous 5-FU, the AUC was increased by 27% and total body clearance was decreased by 28% (Fig. 7B, C) [39]. Since H2 blockers such as cimetidine are likely to be coprescribed or self-administered with anti-neoplastic drugs, oncologists should be aware of this potential drug interaction [38]. 5-FU clearance has been found to be significantly
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 109
reduced (26.9%) by prior administration of metronidazole, which leads to increased toxicity without enhanced therapeutic efficacy [40]. The mechanisms underlying these apparent interactions are unknown [40].
A
N=7 p = 0.018
AUC/dose (ng・h/mL・mg)
120 100 80 60 40 20 Doc alone
Doc+Keto
N=7 p = 0.018
B 50
CL (L/h)
40 30 20 10 Doc alone
Doc+Keto
Fig. (6). AUC/dose (A) and clearance (B) for docetaxel in patients with cancer treated with docetaxel [36] when administered alone or concomitantly with ketoconazole. Doses used were: docetaxel 100mg/m2 or 10mg/m2 with three 200mg dose of ketokonazole. CL, clearance; Doc, docetaxel; Keto, ketoconazole. Error bars represent standard deviation.
Calcium channel blockers such as verapamil inhibit both P-gp and CY3A. When paclitaxel was administered in combination with r-verapamil, there was a near doubling of the AUC and peak concentration of paclitaxel [41]. In another study, doxorubicin AUC appeared to be higher and plasma drug clearance appeared to be lower when concomitantly administered with verapamil, although these differences were not statistically significant due to a small sample size (Fig. 8A, B) [42]. Another calcium channel blocker, nifedipine, also affects the pharmacokinetics of vincristine, a CYP3A-metabolized drug. When patients receiving vincristine were treated with nifedipine, there was a decrease in vincristine clearance from the body (Fig. 8C, D) [43].
110 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
A
AUC (μg/L/min)
without cimetidine
Not significant
1600
with cimetidine
1200 p < 0.05
N=7
800
400
0 Epirubicin
Epirubicinol
B N = 15 p < 0.01
AUC (μg/mL/min)
2000 1500 1000 500 0 without cimetidine
with cimetidine
C
CL (mL/min)
1500 N = 15 p < 0.01
1000
500 without cimetidine
with cimetidine
Fig. (7). AUC for epirubicin and epirubicinol (A) [38] and AUC (B) and clearance (C) for 5-FU [39] when administered alone or concomitantly with cimetidine to patients with advanced cancer. Doses used were: (A) epirubicin 100 mg/m2 and cimetidine 400mg twice daily, (B,C) 5-FU 15 mg/kg and cimetidine 1000mg once daily. Error bars represent standard deviation.
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 111
A
N=5 Not significant
AUC (ng mL-1h)
4000 3000 2000 1000 0 without verapamil
with verapamil N=5 Not significant
B 80
CL (l/h)
60 40 20 0 without verapamil
with verapamil N = 26 p < 0.05
C AUC (μgxmin/mL)
15
10
5
0 without nifedipine
with nifedipine N = 26 p < 0.01
D CL (ml/min/m2)
1500
1000
500
0 without nifedipine
with nifedipine
Fig. (8). AUC (A,C) and clearance (B,D) for doxorubicin (A,B) or vincristine (C,D) when administered alone or concomitantly with verapamil (A,B) [42] or nifedipine (C,D) [43] to patients with advanced cancer. Dose used were: (A,B) doxorubicin 40 mg/m2 and verapamil 240-480mg daily, (C,D) vincristine 2mg and nifedipine 10mg three times a day. CL, clearance. Error bars represent standard deviation.
112 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
Decreased Benefit Corticosteroid and anti-convulsant (AC) medications that are commonly used for patients with brain tumors, such as phenytoin, carbamazepine and phenobarbital, can affect the cytochrome P-450 system and may cause increased clearance of antineoplastic drugs, which are also metabolized by the cytochrome P-450 system, thus producing a need for larger doses [44]. In a phase I study of paclitaxel in patients with recurrent malignant glioma, peak plasma concentration and clearances were similar for patients receiving 240 mg/m2 paclitaxel with no ACs and patients receiving 360 mg/m2 paclitaxel with ACs, which supports the theory that administration of ACs increases the clearance of paclitaxel [44]. Vincristine is metabolized by CYP3A4 to some extent, and in one study systemic clearance was found to be 63% higher, elimination half-life 35% shorter and total AUC 43% smaller in patients who were receiving carbamazepine or phenytoin than in those who were not (Fig. 9A, B) [45]. Etoposide clearance is also affected by ACs: AC therapy is reportedly associated with a 1.5-fold to two-fold increase in etoposide clearance, resulting in a 33% to 50% reduction in systemic exposure, which may result in ineffective therapy (Fig. 9C) [46]. In patients with malignant glioma who were being treated with irinotecan, all of whom were receiving chronic dexamethasone, and >90% of whom were receiving enzyme-inducing ACs, irinotecan, SN-38 and SN38G AUC values were approximately 40%, 25% and 25%, respectively, of the corresponding values in patients with metastatic colorectal cancer who were not receiving chronic dexamethasone and enzyme-inducing ACs (Fig. 10) [47]. It has been recommended that the dose of irinotecan be increased by 50% in patients who are being treated concomitantly with enzyme-inducing ACs, in order to achieve adequate exposure to irinotecan and SN-38 [29]. Pretreatment of patients with barbiturates has been found to stimulate docetaxel metabolism to a significant extent [48]. In any case, the American Academy of Neurology’s position is that ACs are not effective in preventing first seizure in patients with newly diagnosed brain tumor, and should not be used routinely [49]. St John’s wort has become one of the world’s most popular herbal preparations, particularly among cancer patients, because of its putative activity against mild to moderate depression. However, in cancer patients treated with irinotecan, concomitant administration of St John’s wort reduced the AUC of its active metabolite, SN-38, by 42%, which may have a deleterious impact on treatment outcome (Fig. 11) [50]. Cigarette smoke contains several constituents that are known to interact with drugmetabolizing enzymes. Among patients receiving irinotecan, irinotecan clearance was about 18% faster among smokers than among non-smokers, and systemic exposure to SN-38 was almost 40% lower in smokers [51]. In the same study, hematological toxicity was considerably less prevalent among smokers, which may reflect a less favorable therapeutic outcome [51]. CYP3A4 is the major enzyme responsible for metabolism of erlotinib, an epidermal growth factor receptor tyrosine kinase inhibitor [52]. Oral clearance of erlotinib has been shown to be 24% faster in smokers compared with nonsmokers [53], and this may affect overall survival in patients with non-small-cell lung cancer [51]. Tamoxifen is converted to the active metabolites 4-hydroxytamoxifen and 4-hydroxyN-desmethyltamoxifen (endoxifen), which is more than 100 times potent than tamoxifen with respect to its anti-estrogen effect [54, 55]. Plasma endoxifen steady-state concentrations are, on average, 6-10 times higher than corresponding 4-hydroxytamoxifen concen-
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 113
N = 15 p = 0.004
A AUC (ng・h/mL)
80
60
40
20 without AC
B
N = 15 p = 0.04
1200
CL (mL/min)
with AC
1000 800 600 400 200 0 without AC
with AC
C N = 14 p = 0.01
CL (mL/min/m2)
40 30 20 10 0 without AC
with AC
Fig. (9). AUC (A) and clearance (B) of vincristine [45] and clearance (C) of etoposide [46] when administered alone or concomitantly with various anticonvulsants to patients with advanced cancer. Dose used were: (A,B) vincristine 1.4mg/m2 (max 2mg), (C) etoposide 960-1500mg/m2. CL, clearance. Error bars represent standard deviation.
114 Frontiers in Drug Design & Discovery, 2009, Vol. 4
A
Ishiguro et al.
p < 0.0001
AUC (ng・h/mL)
30000
Glioma
(N = 32)
25000 20000
Non-CNS tumor (N = 163)
p < 0.0001
15000 p < 0.0001
10000 5000 0 CPT
SN-38 (x10)
SN-38G (x10)
p < 0.0001
B
20
CL
(L/h/m2)
40
0 Glioma (N = 32)
Non-CNS tumor (N = 163)
Fig. (10). AUC of irinotecan (CPT) and its metabolites (A) and clearance of CPT (B) in patients with glioma and non-central nervous system tumors [47]. Dose used were: irinotecan 125mg/m2 . CL, clearance; CNS, central nervous system. CL, clearance. Error bars represent standard deviation.
trations in women receiving 20 mg/day tamoxifen in the long term, and endoxifen has been shown to be the metabolite most responsible for tamoxifen activity [54, 55]. Plasma concentrations of endoxifen are lower in patients who are also taking paroxetine, a selective serotonin re-uptake inhibitor (SSRI) that is commonly prescribed for the nonhormonal treatment of hot flashes, than in patients who are not taking paroxetine (Fig. 12) [56]. Tamoxifen-treated patients with impaired CYP2D6 metabolism arising from the presence of one or two CYP2D6*4 alleles or the concomitant administration of a moderate or potent CYP2D6 inhibitor have a nearly two-fold higher risk of breast cancer recurrence, independent of standard prognostic factors [54].
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 115
A
AUC (μM・h)
80
without SJW with SJW
p = 0.14
60 p = 0.033
40
N=5
20 0 CPT
B
SN-38 (x10)
N=5 p = 0.11
80
CL (L/h)
60 40 20 0 without SJW
with SJW
Fig. (11). AUC (A) and clearance (B) of irinotecan (CPT) when administered alone or concomitantly with St John’s wort (SJW) in patients with cancer [50]. AUC data for SN-38 are also shown. Dose used were: irinotecan 350mg/m2, SJW 300mg three times a day. CL, clearance; SJW, St John’s wort. Error bars represent standard deviation.
Cyclophosphamide is a prodrug that requires bio-activation to become cytotoxic. The first step in this activation process is the cytochrome p450 isoenzyme-catalyzed hydroxylation of cyclophosphamide to 4-hydroxycyclophosphamide. ThioTEPA, an agent used in high-dose chemotherapy regimens, strongly inhibits the bio-activation of cyclophosphamide, resulting in AUC values that are decreased by 26%, presumably resulting in decreased efficacy [57]. Therefore, the sequence and scheduling of these two agents in high-dose chemotherapy regimens is likely to be extremely important [57]. Effect on Warfarin Pharmacokinetics Imatinib is a strong inhibitor of CYP2C9, which is the principal enzyme required for inactivation of warfarin [58]. In a phase 1 clinical trial of docetaxel/estramustin/imatinib
116 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
administered to prostate cancer patients who were also receiving warfarin prophylaxis, two out of thirteen patients experienced grade 3 prolongation of prothrombin time, which was attributed to the interaction between imatinib and warfarin [59]. 200
Endoxifen concentration (nM)
N = 34
N=6
N=3
150
100
50
0
None
Paroxetine
None
Wt/Wt
Wt/Wt
Null/Null
SSRI Genotype
Fig. (12). Plasma endoxifen concentrations in patients with breast cancer concurrently taking commonly used antidepressants, paroxetine and tamoxifen 20mg/day, according to CYP2D6 genotype [56]. Error bars represent standard deviation. SSRI, selective serotonin re-uptake inhibitor; Wt, wild type.
An adverse interaction between FU and warfarin has been reported, which may be due to FU impairing vitamin K absorption or interfering with warfarin absorption, vascular transport, degradation, or excretion, or FU interfering with metabolism in general [4]. Several studies have shown that a 1 mg/day (mini-dose) of warfarin reduces catheter-related thrombosis without causing an alteration in prothrombin time or activated partial thromboplastin time, or causing bleeding. However, Masci et al. reported that there was a warfarin INR elevation of more than 1.5 in 33% of patients receiving fluorouracil-based chemotherapy, with 19% having an INR value of more than 3.0 [60]. Eight percent of patients also experienced bleeding. These authors also observed an INR elevation in 28% of patients treated with a de Gramont regimen (FU and folinic acid), 26% treated with a FU, folinic acid and irinotecan (FOLFIRI) regimen and 49% treated with a FU, folinic acid and oxaliplatin (FOLFOX) regimen. There was a significantly higher incidence of INR elevation in the patients treated with FOLFOX regimen. Oxaliplatin is largely bound to plasma protein and especially to albumin (85%), therefore it was originally suggested that displacement of warfarin from plasma protein could explain the prolongation of INR in that study [60] but other factors are now thought to be more important.
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 117
There is a significant pharmacokinetic interaction between capecitabine and warfarin, resulting in exaggerated anticoagulant activity. During capecitabine treatment, the AUC for a single dose of warfarin increased by 57%, with a 51% longer elimination half-life and an INR that was increased 2.8-fold relative to warfarin without capecitabine [61]. Copur et al. described two patients who presented with bleeding most likely caused by an adverse interaction between capecitabine and warfarin after 6 weeks of concomitant therapy [62]. In both cases, there was a marked elevation in INR value (to >10), with subsequent gastrointestinal bleeding. Since the exact mechanism by which this interaction occurs is unknown, close monitoring of coagulation parameters is recommended for all patients receiving concomitant warfarin and capecitabine, with appropriate adjustment of warfarin dosage as necessary [62]. Excretion For approximately one-third of anticancer agents, 30% or more undergoes renal clearance as an active or toxic compound. For example, methotrexate and bleomycin are mostly excreted unaltered in the kidney. Thus, even a minor decrease in renal function in patients undergoing chemotherapy can have profound effect on renal clearance, leading to significant toxicity [29]. When the cumulative dose of cisplatin is greater than 300 mg/m2, the total plasma clearance of a renally excreted drug such as bleomycin is reportedly decreased by 50% [63]. Thus, close monitoring of renal function should be performed in patients receiving concurrent cisplatin, such as in the MVAC regimen for bladder cancer and the BEP regimen for germ cell tumors [29]. Since fatal pulmonary toxicity has been associated with delayed bleomycin elimination as a result of cisplatininduced renal failure, the dose of bleomycin should be reduced when creatinine clearance is less than 60 mL/min [64]. Methotrexate is not only filtered but is also actively secreted by the renal tubules [65]. Concurrent administration of certain non-steroidal anti-inflammatory drugs or penicillin may increase the serum concentration of methotrexate via competition in renal tubular secretion [65, 66]. Further, when urine becomes acidified, methotrexate may precipitate within the renal tubules, which delays its elimination, causing further toxicity. Therefore, medications that potentially acidify urine, such as furosemide and thiazide diuretics as well as high dose vitamin C, should be avoided in patients receiving methotrexate treatment [67, 68]. Patients being treated with high doses of methotrexate must be kept hydrated and with a urine pH above 7.0 by using acetazolamide [67, 68]. SUMMARY In the field of oncology, drug-drug interactions are a major concern in both day-today practice and drug development. The US Food and Drug Administration has issued guidelines detailing which in vitro and in vivo drug interaction studies should be conducted during drug development [1]. However, data arising from studies of this type are not available for many drugs that are already on the market, meaning that data must be garnered from case studies. Further compounding the problem is that data from clinical trials that are terminated due to unexpected toxicities are not always reported in full in the literature. Further, drug-drug interactions that result in a lower level of efficacy but no increase in toxicity have not been given much attention until recently. The interaction between tamoxifen and SSRIs such as paroxetine is one famous example.
118 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
For most anti-neoplastic drugs, especially cytotoxic drugs, the recommended dose is close to the maximum tolerated dose in order to achieve maximum clinical benefit. Although CYP3A4 inhibitory activity leading to a less than two-fold increase in AUC is classified as “weak” inhibition, for anti-neoplastic drugs even a minor increase in AUC (e.g. 30%) can result in severe dose-limiting toxicity. Conversely, in cancer patients a reduction in AUC because of drug interactions might result in a lower level of efficacy, potentially leading to disease recurrence or progression. This phenomenon is particularly important for adjuvant chemotherapy, when there is no evaluable disease that might be used to monitor drug efficacy. Therefore, knowledge of drug interactions is more important for clinical oncologists than for any other specialty. Medical oncologists must not only control patient compliance via education, but also take charge of controlling the many factors affecting the pharmacokinetic and pharmacodynamic variables controlling response to treatment (Fig. 13).
Fig. (13). Schematic diagram showing the factors affecting the pharmacokinetics and pharmacodynamics of drugs.
ABBREVIATIONS 5-FU
=
5-Fluorouracil
AUC
=
Area under the concentration-time curve
Cmax
=
Maximum plasma concentration
Important Drug Interactions
Frontiers in Drug Design & Discovery, 2009, Vol. 4 119
CYP
=
Cytochrome P450
INR
=
International normalization ratio
MIC
=
Minimum inhibitory concentration
UGT
=
Uridine diphosphate-glucuronosyltransferase
REFERENCES [1]
[2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18]
[19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34]
Bjornsson, T. D.; Callaghan, J. T.; Einolf, H. J.; Fischer, V.; Gan, L.; Grimm, S.; Kao, J.; King, S. P.; Miwa, G.; Ni, L.; Kumar, G.; McLeod, J.; Obach, S. R.; Roberts, S.; Roe, A.; Shah, A.; Snikeris, F.; Sullivan, J. T.; Tweedie, D.; Vega, J. M.; Walsh, J.; Wrighton, S. A. J. Clin. Pharmacol., 2003, 43, 443-69. Monahan, B. P.; Ferguson, C. L.; Killeavy, E. S.; Lloyd, B. K.; Troy, J.; Cantilena, L. R. Jr. JAMA, 1990, 264, 2788-90. Roden, D. M. N. Engl. J. Med., 2004, 350, 1013-22. Beijnen, J. H.; Schellens, J. H. Lancet Oncol., 2004, 5, 489-96. Arsenic trioxide (Trisenox) Package Insert (Japanese). In Nippon Shinyaku CO., LTD.: 2007. Amphotericin B (Fangizone) Package Insert (Japanese). In Bristol-Myers Squibb Company: 2007. Tobramycin (Tobracin) Package Insert (Japanese). In J-DOLPH Pharmaceutical CO., LTD.: 2008. Cheung, Y. W.; Cradock, J. C.; Vishnuvajjala, B. R.; Flora, K. P. Am. J. Hosp. Pharm., 1987, 44, 12430. Pujol Cubells, M.; Prat Aixela, J.; Girona Brumos, V.; Duran Pou, S.; Villaronga Flaque, M. Pharm. World Sci., 1993, 15, 34-6. Cisplatin (Randa) Package Insert (Japanese). In Nippon Kayaku CO., LTD.: 2007. Mesna (Uromitexan) Package Insert (Japanese). In Shionogi & CO., LTD.: 2005. Mitomycin C (Mitomycin) Package Insert (Japanese). In Kyowa Hakko Kogyo CO., LTD.: 2006. Gianni, L.; Vigano, L.; Locatelli, A.; Capri, G.; Giani, A.; Tarenzi, E.; Bonadonna, G. J. Clin. Oncol., 1997, 15, 1906-15. Holmes, F. A.; Madden, T.; Newman, R. A.; Valero, V.; Theriault, R. L.; Fraschini, G.; Walters, R. S.; Booser, D. J.; Buzdar, A. U.; Willey, J.; Hortobagyi, G. N. J. Clin. Oncol., 1996, 14, 2713-21. Millward, M. J.; Webster, L. K.; Rischin, D.; Stokes, K. H.; Toner, G. C.; Bishop, J. F.; Olver, I. N.; Linahan, B. M.; Linsenmeyer, M. E.; Woodcock, D. M. Clin. Cancer Res., 1998, 4, 2321-9. Baker, A. F.; Dorr, R. T. Cancer Treat. Rev., 2001, 27, 221-33. D'Incalci, M.; Schuller, J.; Colombo, T.; Zucchetti, M.; Riva, A. Semin. Oncol., 1998, 25(6 Suppl 13), 16-20. Venturini, M.; Lunardi, G.; Del Mastro, L.; Vannozzi, M. O.; Tolino, G.; Numico, G.; Viale, M.; Pastrone, I.; Angiolini, C.; Bertelli, G.; Straneo, M.; Rosso, R.; Esposito, M. J. Clin. Oncol., 2000, 18, 2116-25. Rowinsky, E. K.; Gilbert, M. R.; McGuire, W. P.; Noe, D. A.; Grochow, L. B.; Forastiere, A. A.; Ettinger, D. S.; Lubejko, B. G.; Clark, B.; Sartorius, S. E. J. Clin. Oncol., 1991, 9, 1692-703. Levofloxacin (Clavit) Package Insert (Japanese). In Daiichi Sankyo & CO., LTD.: 2008. Preston, S. L.; Drusano, G. L.; Berman, A. L.; Fowler, C. L.; Chow, A. T.; Dornseif, B.; Reichl, V.; Natarajan, J.; Corrado, M. JAMA, 1998, 279, 125-9. Mercaptopurin (Leukerin) Package Insert (Japanese). In Ohara Pharmaceutical CO., LTD.: 2007. Gefitinib (Iressa) Package Insert (Japanese). In AstraZeneca: 2006. Erlotinib (Tarceva) Package Insert (Japanese). In Chugai Pharmaceutical CO., LTD.: 2007. Benet, L. Z.; Hoener, B. A. Clin. Pharmacol. Ther., 2002, 71, 115-21. Thompson, M. E.; Highley, M. S. Ann. Oncol., 2003, 14, 500. Le, A. T.; Hasson, N. K.; Lum, B. L. Ann. Pharmacother., 1997, 31, 1006-8. Hall, G.; Lind, M. J.; Huang, M.; Moore, A.; Gane, A.; Roberts, J. T.; Cantwell, B. M. Postgrad. Med. J., 1990, 66, 860-1. Lam, M. S.; Ignoffo, R. J. J. Oncol. Pharma. Pract., 2003, 9, 45. Fujita, K. Curr. Drug Metab., 2006, 7, 23-37. Bosque, E. Ann. Intern. Med., 2001, 134, 427. Bohme, A.; Ganser, A.; Hoelzer, D. Ann. Hematol., 1995, 71, 311-2. Gillies, J.; Hung, K. A.; Fitzsimons, E.; Soutar, R. Clin. Lab. Haematol., 1998, 20, 123-4. United States Food and Drug Administration: Campostar label.
120 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [35] [36] [37]
[38] [39] [40] [41] [42] [43] [44]
[45] [46] [47]
[48] [49] [50] [51] [52] [53] [54]
[55] [56] [57] [58] [59] [60] [61]
Ishiguro et al.
Kehrer, D. F.; Mathijssen, R. H.; Verweij, J.; de Bruijn, P.; Sparreboom, A. J. Clin. Oncol., 2002, 20, 3122-9. Engels, F. K.; Ten Tije, A. J.; Baker, S. D.; Lee, C. K.; Loos, W. J.; Vulto, A. G.; Verweij, J.; Sparreboom, A. Clin. Pharmacol. Ther., 2004, 75, 448-54. Bruno, R.; Hille, D.; Riva, A.; Vivier, N.; ten Bokkel Huinnink, W. W.; van Oosterom, A. T.; Kaye, S. B.; Verweij, J.; Fossella, F. V.; Valero, V.; Rigas, J. R.; Seidman, A. D.; Chevallier, B.; Fumoleau, P.; Burris, H. A.; Ravdin, P. M.; Sheiner, L. B. J. Clin. Oncol., 1998, 16, 187-96. Murray, L. S.; Jodrell, D. I.; Morrison, J. G.; Cook, A.; Kerr, D. J.; Whiting, B.; Kaye, S. B.; Cassidy, J. Clin. Oncol., (R Coll. Radiol.), 1998, 10, 35-8. Harvey, V. J.; Slevin, M. L.; Dilloway, M. R.; Clark, P. I.; Johnston, A.; Lant, A. F. Br. J. Clin. Pharmacol., 1984, 18, 421-30. Bardakji, Z.; Jolivet, J.; Langelier, Y.; Besner, J. G.; Ayoub, J. Cancer Chemother. Pharmacol., 1986, 18, 140-4. Tolcher, A. W.; Cowan, K. H.; Solomon, D.; Ognibene, F.; Goldspiel, B.; Chang, R.; Noone, M. H.; Denicoff, A. M.; Barnes, C. S.; Gossard, M. R.; Fetsch, P. A.; Berg, S. L.; Balis, F. M.; Venzon, D. J.; O'Shaughnessy, J. A. J. Clin. Oncol., 1996, 14, 1173-84. Kerr, D. J.; Graham, J.; Cummings, J.; Morrison, J. G.; Thompson, G. G.; Brodie, M. J.; Kaye, S. B. Cancer Chemother. Pharmacol., 1986, 18, 239-42. Fedeli, L.; Colozza, M.; Boschetti, E.; Sabalich, I.; Aristei, C.; Guerciolini, R.; Del Favero, A.; Rossetti, R.; Tonato, M.; Rambotti, P. Cancer, 1989, 64, 1805-11. Chang, S. M.; Kuhn, J. G.; Rizzo, J.; Robins, H. I.; Schold, S. C., Jr.; Spence, A. M.; Berger, M. S.; Mehta, M. P.; Bozik, M. E.; Pollack, I.; Gilbert, M.; Fulton, D.; Rankin, C.; Malec, M.; Prados, M. D. J. Clin. Oncol., 1998, 16, 2188-94. Villikka, K.; Kivisto, K. T.; Maenpaa, H.; Joensuu, H.; Neuvonen, P. J. Clin. Pharmacol. Ther., 1999, 66, 589-93. Rodman, J. H.; Murry, D. J.; Madden, T.; Santana, V. M. J. Clin. Oncol., 1994, 12, 2390-7. Friedman, H. S.; Petros, W. P.; Friedman, A. H.; Schaaf, L. J.; Kerby, T.; Lawyer, J.; Parry, M.; Houghton, P. J.; Lovell, S.; Rasheed, K.; Cloughsey, T.; Stewart, E. S.; Colvin, O. M.; Provenzale, J. M.; McLendon, R. E.; Bigner, D. D.; Cokgor, I.; Haglund, M.; Rich, J.; Ashley, D.; Malczyn, J.; Elfring, G. L.; Miller, L. L. J. Clin. Oncol., 1999, 17, 1516-25. Royer, I.; Monsarrat, B.; Sonnier, M.; Wright, M.; Cresteil, T. Cancer Res., 1996, 56, 58-65. Glantz, M. J.; Cole, B. F.; Forsyth, P. A.; Recht, L. D.; Wen, P. Y.; Chamberlain, M. C.; Grossman, S. A.; Cairncross, J. G. Neurology, 2000, 54, 1886-93. Mathijssen, R. H.; Verweij, J.; de Bruijn, P.; Loos, W. J.; Sparreboom, A. J. Natl. Cancer Inst., 2002, 94, 1247-9. van der Bol, J. M.; Mathijssen, R. H.; Loos, W. J.; Friberg, L. E.; van Schaik, R. H.; de Jonge, M. J.; Planting, A. S.; Verweij, J.; Sparreboom, A.; de Jong, F. A. J. Clin. Oncol., 2007, 25, 2719-26. Ling, J.; Johnson, K. A.; Miao, Z.; Rakhit, A.; Pantze, M. P.; Hamilton, M.; Lum, B. L.; Prakash, C. Drug Metab. Dispos., 2006, 34, 420-6. Lu, J. F.; Eppler, S. M.; Wolf, J.; Hamilton, M.; Rakhit, A.; Bruno, R.; Lum, B. L. Clin. Pharmacol.. Ther., 2006, 80, 136-45. Goetz, M. P.; Knox, S. K.; Suman, V. J.; Rae, J. M.; Safgren, S. L.; Ames, M. M.; Visscher, D. W.; Reynolds, C.; Couch, F. J.; Lingle, W. L.; Weinshilboum, R. M.; Fritcher, E. G.; Nibbe, A. M.; Desta, Z.; Nguyen, A.; Flockhart, D. A.; Perez, E. A.; Ingle, J. N. Breast Cancer Res. Treat., 2007, 101, 11321. Lim, H. S.; Ju Lee, H.; Seok Lee, K.; Sook Lee, E.; Jang, I. J.; Ro, J. J. Clin. Oncol., 2007, 25, 383745. Jin, Y.; Desta, Z.; Stearns, V.; Ward, B.; Ho, H.; Lee, K. H.; Skaar, T.; Storniolo, A. M.; Li, L.; Araba, A.; Blanchard, R.; Nguyen, A.; Ullmer, L.; Hayden, J.; Lemler, S.; Weinshilboum, R. M.; Rae, J. M.; Hayes, D. F.; Flockhart, D. A. J. Natl. Cancer Inst., 2005, 97, 30-9. Huitema, A. D.; Kerbusch, T.; Tibben, M. M.; Rodenhuis, S.; Beijnen, J. H. Cancer Chemother. Pharmacol., 2000, 46, 119-27. Imatinib (Glivec) Package Insert (Japanese). In Novartis: 2008. Lin, A. M.; Rini, B. I.; Derynck, M. K.; Weinberg, V.; Park, M.; Ryan, C. J.; Rosenberg, J. E.; Bubley, G.; Small, E. J. Clin. Genitourin. Cancer, 2007, 5, 323-8. Masci, G.; Magagnoli, M.; Zucali, P. A.; Castagna, L.; Carnaghi, C.; Sarina, B.; Pedicini, V.; Fallini, M.; Santoro, A. J. Clin. Oncol., 2003, 21, 736-9. Camidge, R.; Reigner, B.; Cassidy, J.; Grange, S.; Abt, M.; Weidekamm, E.; Jodrell, D. J. Clin. Oncol., 2005, 23, 4719-25.
Important Drug Interactions [62] [63] [64] [65] [66] [67] [68] [69]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 121
Copur, M. S.; Ledakis, P.; Bolton, M.; Morse, A. K.; Werner, T.; Norvell, M.; Muhvic, J.; Chu, E. Clin. Colorectal Cancer, 2001, 1, 182-4. Yee, G. C.; Crom, W. R.; Champion, J. E.; Brodeur, G. M.; Evans, W. E. Cancer Treat. Rep., 1983, 67, 587-9. Kintzel, P. E.; Dorr, R. T. Cancer Treat. Rev., 1995, 21, 33-64. Liegler, D. G.; Henderson, E. S.; Hahn, M. A.; Oliverio, V. T. Clin. Pharmacol. Ther., 1969, 10, 84957. Ronchera, C. L.; Hernandez, T.; Peris, J. E.; Torres, F.; Granero, L.; Jimenez, N. V.; Pla, J. M. Ther. Drug Monit., 1993, 15, 375-9. Methotrexate (Methotrexate) Package Insert (Japanese). In Wyeth: 2006. Sketris, I. S.; Farmer, P. S.; Fraser, A. Cancer Treat. Rep., 1984, 68, 446-7. Minami, H. In Clinical Oncology Update - Essentials for Medical Oncologist (Japanese). Nankodo: Tokyo, Japan, 2006, pp. 240-6.
122
Frontiers in Drug Design & Discovery, 2009, 4, 122-134
Pharmacogenomic Considerations in Breast Cancer Management Hiroshi Ishiguro1,*, Ikuko Yano2 and Masakazu Toi3 1
Translational Research Center, Department of Clinical Trial Management / Outpatient Oncology Unit; 2Department of Pharmacy and 3Breast Surgery Department, Kyoto University Hospital, 54 Shogoinkawaharara-cho,Sakyo-ku, Kyoto-city, 606-8507, Japan Abstract: Many factors affect an individual’s response to a drug, and large inter-ethnic, intra-ethnic, and even intra-individual variations exist. These variations may affect both the therapeutic response to a drug, or the side effects that the patient experiences. Since anti-neoplastic drugs often have a very narrow therapeutic range, it is very desirable to be able to predict these variations in response or to ensure that these variations are as small as possible. It has recently become possible to predict some extreme responses, such as severe side effects, using pharmacogenomic approaches; for example, the uridine diphosphate-glucuronosyltransferase (UGT) 1A1 genetic polymorphism is a predictor of irinotecan toxicity. Further, adding pharmacokinetic and pharmacodynamic information may increase the accuracy of response prediction. Both severe toxicity and clinical benefit can be predicted using a combination of pharmacogenomic and pharmacokinetic information. For example, the clinical benefit obtained from adjuvant treatment with tamoxifen is reduced in patients who have either a particular Cytochrome P450 (CYP) 2D6 genetic polymorphism or who are taking paroxetine, a strong inhibitor of CYP2D6. It is possible to monitor pharmacodynamic parameters, such as serum estrogen levels, as a measure of the therapeutic effect of aromatase inhibitors. In this review, we summarize current knowledge in the field of pharmacogenomics as it relates to breast cancer, focusing particularly on clinical data.
INTRODUCTION Pharmacogenomics, the intersection of pharmacology and genomics, is an approach often used in drug development to assess the genetic determinants of drug response on a genome-wide basis. Pharmacogenetics, a subset of pharmacogenomics, is slightly more narrowly defined as the study of the genetic basis for variation in drug response [1]. Pharmacogenomics is an important field of research because the activity of drugmetabolizing enzymes often varies widely, even among healthy people, making drug metabolism highly variable between individuals (e.g. drug elimination rates can vary up to 40-fold [2]). Genetic factors and aging appear to account for most of these variations in healthy individuals [2]. Pharmacogenetic variations can have significant clinical con*Corresponding Author: Tel: +81-75-751-4770; Fax: +81-75-751-4772; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
Pharmacogenomic Considerations
Frontiers in Drug Design & Discovery, 2009, Vol. 4 123
sequences; for example, patients who metabolize certain drugs rapidly may require higher and/or more frequent doses to achieve therapeutic concentrations. In contrast, patients who metabolize certain drugs slowly may need lower and/or less frequent doses to avoid toxicity, particularly for drugs with a narrow margin of safety, such as antineoplastic drugs. However, most genetically determined differences in drug metabolism cannot currently be predicted before drug administration. Also, many environmental and developmental factors, such as concomitantly administered drugs, can interact with each other and with genetic factors to affect drug response [2]. DRUG SAFETY AND PHARMACOGENOMICS Anti-neoplastic drugs, especially cytotoxic anti-neoplastic drugs, have a very narrow therapeutic window, so pharmacogenomics has become an important research topic in this context. Irinotecan, a key agent used in the treatment of colorectal and lung cancer, is also often used for advanced breast cancer. Irinotecan is in fact a prodrug, which is converted to its active but toxic metabolite, SN-38, by carboxyesterase in vivo (Fig. 1) [3]. Irinotecan is also metabolized by CYP3A4 to the inactive metabolites 7-ethyl-10-[4N-(5-aminopentanoic acid)-1-piperidino] carbonyloxycampothecin (APC) and 7-ethyl10-(4-amino-1-piperidino] carbonyloxycampothecin (NPC). Over time, SN-38 is metabolized to an inactive form, SN-38G, by uridine diphosphate-glucuronosyltransferase (UGT) 1A1 glucuronidation. The area under the concentration-time curve (AUC) of SN38G is ~7-fold larger than for SN-38, suggesting extensive conversion of SN-38 into SN-38G in vivo, but high levels of inter-patient variability have also been found [3]. Recently, a series of studies have provided evidence that the UGT1A1*28 genetic polymorphism may have an important influence on irinotecan toxicity [4]. It has been shown that there is a significant reduction in SN-38 glucuronidation activity in individuals with the *28 variant, and that this haplotype is significantly associated with a reduced SN38G/SN-38 AUC ratio [5]. In 2005, the US Food and Drug Administration required that a warning to this effect be added to the irinotecan package insert [6], which now states: “Patients homozygous/heterozygous for the UGT1A1*28 allele: Patients homozygous for the UGT1A1*28 allele are at increased risk of neutropenia; initial one-level dose reduction should be considered for both single-agent and combination regimens. Heterozygous carriers of the UGT1A1*28 allele may also be at increased risk; however, most patients have tolerated normal starting doses.” Trastuzumab, a humanized anti-HER-2/neu immunoglobulin G (IgG) 1 monoclonal antibody (mAb), is a drug that has significant benefit in terms of breast cancer prognosis, but has also been associated with cardiotoxicity. Trastuzumab binds to the extracellular domain of HER2, which is encoded by the proto-oncogene HER2 (erbB-2, neu), and is a transmembrane protein with tyrosine kinase activity but no identified physiological ligand. The HER2 gene is amplified in 30% of cases of invasive breast cancer and HER2 gene amplification is correlated with poor prognosis. The most thoroughly investigated germinal HER2 polymorphism at a clinical level occurs at codon 655 (ATC/isoleucine to GTC/valine), which encodes a residue in the transmembrane domain of the HER2 protein. In a group of 61 patients with advanced breast cancer studied by Beauclair et al., although no cardiac toxicity was found in 36 patients with the Ile/Ile genotype, 24% (5/21) of patients with the Ile/Val genotype experienced a greater than 20% reduction of their left ventricular ejection fraction (LVEF). Only a few patients carried the Val/Val genotype, and none of them developed any symptoms of cardiotoxicity [7].
124 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
O APC HN
OH
CH2CH3 N N
O
O
Irinotecan O
CH2CH3 N
O
N
A4 P3 Y C
O N
O
N
O CH2CH3
N
OH O
CH2CH3
OH O
A4 P3 CY
Carboxyesterase
O
NH2
CH2CH3 OH
NPC
O
CH2CH3
N
N
O
O
N O
Carboxyesterase
SN-38
N O
OH O CH2CH3
UGT1A1
CH2CH3
O
N
O OH
OH O
CH2CH3 O
O
O N
OH
OH
N
OH
O SN-38G
CH2CH3
OH O
Fig. (1). Metabolic pathways of irinotecan. (Modified figure from reference [3]).
DRUG EFFICACY AND PHARMACOGENOMICS In the human genome, inter-individual differences occur approximately every 300 to 1000 nucleotides, with each individual harboring a an estimated total of 3.2 million single nucleotide polymorphisms [1]. There are thus likely to be many genetic polymorphisms within genes encoding pharmacokinetic- and pharmacodynamic-related proteins (e.g. metabolizing enzymes, transporters and receptors), and therefore variations in drug responses due to these polymorphisms are unlikely to be rare phenomena.
Pharmacogenomic Considerations
Frontiers in Drug Design & Discovery, 2009, Vol. 4 125
Cyclophosphamide is a drug that is commonly used to treat breast cancer. It is a prodrug that must be metabolized to 4-hydroxycyclophosphamide to become cytotoxic. After administration, 75-80% of the dose is bioactivated by several CYP enzymes, including (amongst others) CYP2B6, CYP3A4 and CYP3A5, of which CYP2B6 has the highest activity. 4-Hydroxycyclophosphamide and its ring-open tautomer aldophosphamide are detoxified by glutathione S-transferase (GST) and aldehyde dehydrogenase (Fig. 2) [8]. In cyclophosphamide-treated breast cancer patients with an unfavorable CYP3A/GST genotype, cyclophosphamide metabolism is affected, such that activation is slow and detoxification is rapid, leading to worse overall survival compared with patients with other genotypes (Fig. 3) [9, 10]. OO
ClH2CH2C N ClH2CH2C
N
OO
ClH2CH2C T GS
ClH2CH2C N
P
ClH2CH2C
HN SG
4-Glutathionylcyclophosphamide
OO
ClH2CH2C
CYP3A4?
P
N ClH2CH2C
HN OH
GS T
4-Hydroxycyclophosphamide
N ClH2CH2C
P NH2 O
O N
Decomposition
Aldophosphamide Cl
OH
P NH2
GS
GS T
Monochloromonoglutathionylphosphoramide mustard
O
ClH2CH2C N ClH2CH2C
O-
P
O
+
H2C
CH C
NH2
Phosphoramide mustard
HN
4-Ketocyclophosphamide
OO
ClH2CH2C
P
O
Equilibrium
OO
HN
Dechloromethylcyclophosphamide
CYP2B6 others
Cyclophosphamide
N
P
H
HN
ClH2CH2C
OO
ClH2CH2C
CYP3A4
P
H Acrolein
Fig. (2). Metabolic pathways of cyclophosphamide. (Modified figure from reference [7]).
126 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
Even though trastuzumab is usually only administered to patients with HER-2/neupositive breast cancer, only 25 to 30% of these patients will respond to this mAb. The mechanism by which it exerts an anti-tumor effect is not well understood, but antibodydependent cell-mediated cytotoxicity (ADCC) is thought to be involved [11]. In ADCC, the antibody binds to tumor cells and is then engaged by effecter cells via their receptors for IgG. FcR is an activating IgG fragment C receptor (FcR) that is expressed on monocytes/macrophages and natural killer cells. A polymorphism encoding valine (V) at amino acid 158 of FcR rather than phenylalanine (F) greatly increases the affinity of IgG1 to the Fc receptor, leading to a higher level of ADCC activity. The FcR158 V/V genotype is significantly correlated with a higher objective response rate and better progression-free survival in trastuzumab-treated breast cancer patients (Fig. 4) [11].
Overall survival (%)
100
Other CYP3A/GST genotype N = 86
p < 0.05
50 N=4 Unfavorable CYP3A/GST genotype
2
4
6
8
10
Years Fig. (3). Cyclophosphamide CYP3A/GST metabolic enzyme genotype versus overall survival for breast cancer patients treated with a cyclophosphamide-containing regimen. (Modified figure from references [9,10]).
TAMOXIFEN-METABOLIZING ENZYMES Tamoxifen is widely used as the standard treatment for recurrent or metastatic breast cancer, as well as in an adjuvant setting. Tamoxifen is a prodrug that requires CYP2D6catalyzed metabolic activation to 4-hydroxytamoxifen and 4-hydroxy-N-desmethyltamoxifen (endoxifen) for therapeutic pharmacological activity (Fig. 5). Compared with the parent drug, these two metabolites have 100-fold greater affinity to estrogen receptor and 30- to 100-fold greater potency in suppressing estrogen-dependent cell proliferation [12]. Steady-state plasma endoxifen concentrations have been found to be 5- to 10-fold higher than those of 4-hydroxytamoxifen [13]. Therefore, the relationships between endoxifen concentration and CYP2D6 activity, and endoxifen concentration and tamoxifen treatment efficacy and adverse events have been studied.
Pharmacogenomic Considerations
Frontiers in Drug Design & Discovery, 2009, Vol. 4 127
Progression-free survival (%)
100 V/V genotype N = 11
50
p = 0.0035
Other genotype
1
N = 43
2
3
4
5
Years Fig. (4). Progression-free survival of trastuzumab-treated breast cancer patients according to FcRIII genotype. (Modified figure from reference [11]).
N O
CH3 O
CH3
CYP2D6
H3C
N
H3C
CYP3A4
CYP3A4
4-hydroxy-tamoxifen
O
N H
CH3 O
CYP2D6 H3C
OH N-desmethyl-tamoxifen
CH3
OH
Tamoxifen
H3C
CH3
Endoxifen
Fig. (5). Metabolic pathways of tamoxifen. (Modified figure from reference [13]).
N H
CH3
128 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
CYP2D6 GENE POLYMORPHISM AND ENDOXIFEN CONCENTRATION The plasma concentration of endoxifen appears to be influenced by a patient’s CYP2D6 genotype. The CYP2D6*4 allele results in a splicing defect, the CYP2D6*3 and *6 alleles lead to translation frameshifts, and in the CYP2D6*5 allele the entire CYP2D6 gene is deleted. None of these alleles yields a functional CYP2D6 enzyme [14]. In one study, more than 20% of a German Caucasian population had one of these four polymorphisms, compared with about 7% of the Japanese population (Fig. 6) [15]. In another study, tamoxifen-treated breast cancer patients who were heterozygous or homozygous for the null genotype had mean plasma endoxifen concentrations that were 55% and 26%, respectively, of those who were homozygous for the wild-type CYP2D6 genotype (Fig. 7A) [14]. The CYP2D6*10 allele, which produces an unstable enzyme that is associated with decreased CYP2D6 activity, is a major variant in individuals of Japanese, Korean and Chinese ethnic background [16]. This allele is seen in 40 to 50% of these populations (Fig. 6) [15]. Although the steady-state plasma concentrations of endoxifen were comparable in wild type and CYP2D6*10 heterozygotes, the concentration in CYP2D6*10 homozygotes was about 40% that of wild type (Fig. 7B) [16]. Wild
Null
Low
Others
Japanese
Caucasian
0%
20%
40%
60%
80%
100%
Percentage of population Fig. (6). Differences between German Caucasian and Japanese populations in terms of CYP2D6 gene polymorphism [15].
CONCOMITANTLY ADMINISTERED DRUGS AND ENDOXIFEN CONCENTRATION In one study of tamoxifen-treated breast cancer patients, plasma endoxifen concentrations in wild-type CYP2D6 homozygotes were found to be highly variable. However, those who were using CYP2D6 inhibitors such as paroxetine, fluoxetine, sertraline, citalopram, amiodarone and metoclopramide had plasma endoxifen concentrations that were 58% lower than concentrations in those who were not using CYP2D6 inhibitors (Fig. 7C) [14]. The plasma endoxifen concentration in subjects taking paroxetine was
Pharmacogenomic Considerations
Frontiers in Drug Design & Discovery, 2009, Vol. 4 129
p = 0.002
Endoxifen concentration (nM)
100
A 80 60 40 20 0 Wild/Wild
Wild/Null
Null/Null
(N = 48)
(N = 29)
(N = 3)
Endoxifen concentration (ng/mL)
CYP2D6 genotype p < 0.001
25
B
20 15 10 5 0 Wild/Wild (N = 64)
Wild/Low
Low/Low
(N = 89)
(N = 49)
CYP2D6 genotype
Endoxifen concentration (nM)
p = 0.0025 100
C
80
p = 0.08
60 40 20 0 -
+ Wild/Wild
(N = 34)
(N = 13)
-
+ Wild/Null
(N = 17)
(N = 11)
-
CYP2D6 inhibitor
Null/Null
CYP2D6 genotype
(N = 3)
Fig. (7). Endoxifen concentrations in breast cancer patients treated with tamoxifen, according to CYP2D6 genotype. (A) Wild-type versus null genotype [14]. (B) Wild-type versus low genotype [16]. (C) Wild-type versus null genotype with or without CYP2D6 inhibitor [14].
130 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
similar to that for subjects with the CYP2D6 null genotype who were not taking a CYP2D6 inhibitor [14]. Thus, not only genetic factors, but also environmental factors, including concomitantly administered drugs, can have a significant effect on a drug’s pharmacokinetics. CYP2D6 ACTIVITY AND ADVERSE EVENTS The most common request from breast cancer survivors regarding additional treatment or treatment modification is relief from hot flashes [17]. Hot flashes are the subjective sensation of heat that is thought to be related to a narrowing of the thermoneutral zone. The onset of symptoms is generally thought to occur as a result of decreased estrogen or increased gonadotropin concentrations. However, other reported precipitators include psychological stress, hot weather, alcohol and caffeine, although the physiological mechanism underlying this phenomenon is not well understood [17]. In a study conducted by Goetz et al., moderate to severe [National Cancer Institute Common Toxicity Criteria (version 1) grade 2-3] hot flashes were experienced by 20% of tamoxifen-treated breast cancer patients with either wt/wt or *4/wt CYP2D6 genotypes, but no patients with a *4/*4 genotype experienced hot flashes (Fig. 8A) [13]. Similarly, in a recent presentation, Rae et al. described how the presence of active CYP2D6 alleles predicts a higher likelihood of tamoxifen discontinuation in breast cancer patients due to treatment-related side effects* (Fig. 8B). These findings raise the question of whether the alleviation of the hot flashes is related to a reduced concentration of active metabolites. ADVERSE EVENTS AND TREATMENT EFFICACY Among breast cancer patients who took tamoxifen for adjuvant endocrine treatment, those who reported hot flashes were found to be less likely to develop recurrent breast cancer than those who did not develop hot flashes (Fig. 9A). In that study, hot flashes were a stronger predictor of breast cancer-specific outcome than age, hormone receptor status, or even disease stage [17]. These findings suggest an association between side effects, efficacy, and tamoxifen metabolism. Cuzick et al. recently reported the results of the Arimidex, Tamoxifen, Alone or in Combination (ATAC) randomized, doubleblinded, multicenter trial, in which patients with early-stage breast cancer received treatment with anastrozole alone, tamoxifen alone or the two agents in combination*. The occurrence of hot flashes was an indicator of treatment efficacy for both tamoxifen and anastrozole, with similar differences in treatment efficacy between patients with and without hot flashes for both drugs (Fig. 9B). It is likely that the occurrence of hot flashes is an indirect measurement of estrogen suppression by anastrozole and tamoxifen. CYP2D6 ACTIVITY AND TREATMENT EFFICACY In the ATAC trial, using the annual hazard rate, the benefit of anastrozole over tamoxifen was seen throughout the follow-up period, and the early recurrence (less than 3 years) rate was significantly higher for the tamoxifen group (Fig. 10A) [18]. Among patients treated with adjuvant tamoxifen, an immediate broad peak in the hazard rate was seen in patients with decreased CYP2D6 activity (Fig. 10B) [19]. In contrast, the hazard rate in patients with extensive CYP2D6 activity was reduced and did not peak until _____________________ *2007 San Antonio Breast Cancer Symposium.
Pharmacogenomic Considerations
Frontiers in Drug Design & Discovery, 2009, Vol. 4 131
Incidence of hot flashes (%)
40
A p = 0.06
20
0 Null/Null (N = 13)
Wild/Null (N = 40)
Wild/Wild (N = 137)
CYP2D6 genotype
CYP2D6 genotype
B
High (N = 170)
Intermediate (N = 87)
Low (N = 10)
0
10
20
Treatment discontinuation (%)
Fig. (8). Relationship between CYP2D6 genotype and hot flashes [13] (A) and between CYP2D6 phenotype and treatment discontinuation (B) in breast cancer patients treated with tamoxifen*. *2007 San Antonio Breast Cancer Symposium.
nearly the fourth year after treatment. These findings suggest that the peak in the hazard rate for recurrence that was seen following the initiation of tamoxifen in the ATAC trial may be due to there being a subset of patients in which tamoxifen is not fully activated. Several other randomized trials have demonstrated that administration of an aromatase inhibitor following either 2-3 or 5 years of tamoxifen treatment (switching or extension) significantly prolongs disease-free survival and overall survival compared with tamoxifen treatment for 5 years. Further research is needed to determine whether prior assessment of CYP2D6 metabolism can be used to identify which patients are most suitable for these strategies without a risk of early recurrence [19]. SUMMARY There are many factors involved in drug response, and large variations in response can occur between patients with different ethnic backgrounds, between patients with the
132 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
Hot flashes
100 Disease free survival (%)
N = 674
A
p = 0.01 No hot flashes
50
N = 190
2
4
6
8
10
Years
100 Hot flashes Recurrence rate (%)
B
Tamoxifen Anastrozole
No hot flashes
(N = 1091)
(N = 963)
(N = 1614)
(N = 1738)
p < 0.05
50
2
4
6
8
10
Years
Fig. (9). Relationship between hot flashes and disease-free survival in tamoxifen-treated breast cancer patients [17] (A) and hot flashes and recurrence in breast cancer patients who participated in the ATAC trial* (B). Modified figure from reference [16] and a presentation by Cuzick J. et al.* *2007 San Antonio Breast Cancer Symposium.
same ethnic background, and even for single individual at different times. Pharmacogenomics is one approach that can be used to understand these variations. Ideally, for antineoplastic drugs, we would want to refer to pharmacodynamic or pharmacokinetic data such as the international normalized ratio for warfarin or the measures of drug concentration in the blood (pharmacokinetics) that exist for anti-convulsant and immunosuppressive drugs. These will be very useful in reducing variations in drug response in the field of oncology.
Pharmacogenomic Considerations
Frontiers in Drug Design & Discovery, 2009, Vol. 4 133
A
Smoothed hazard rate
Tamoxifen Anastrozole
Poor/Intermediate metabolizer
B
Extensive metabolizer
2
4
6
8
10
Years
Fig. (10). Hazard rates for recurrence in patients treated with tamoxifen or anastrozole [18] (A) and in breast cancer patients treated with tamoxifen according to CYP2D6 activity [19] (B). Modified figure from references [18,19].
ABBREVIATIONS ADCC
=
Antibody-dependent cell-mediated cytotoxicity
APC
=
7-Ethyl-10-[4-N-(5-aminopentanoic acid)-1-piperidino] carbonyloxycampothecin
ATAC trial =
Arimidex, Tamoxifen, Alone or in Combination trial
AUC
=
Area under the concentration-time curve: AUC
CYP
=
Cytochrome P450
FcR
=
Fragment C receptor
GST
=
Glutathione S-transferase
IgG
=
Immunoglobulin G
LVEF
=
Left ventricular ejection fraction
mAb
=
Monoclonal antibody
NPC
=
7-Ethyl-10-(4-amino-1-piperidino] carbonyloxycampothecinecine
UGT
=
Uridine diphosphate-glucuronosyltransferase
134 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Ishiguro et al.
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]
[15] [16] [17] [18] [19]
Relling, V.M.; Giacomini, MK. In Goodman & Gilman's The Pharmacological Basis of Therapeutics. eleventh ed.; McGRAW-HILL: New York, 2006, pp. 93-115. Cafiero, A.; Hussar, D.; Tarloff, B.J.; Vivian, M.E. In The Merck Manual of Diagnosis and Therapy. Eighteenth ed.; Merck Research Laboratories: Whitehouse Station, NJ, 2006, pp. 2513-44. Mathijssen, R. H.; van Alphen, R. J.; Verweij, J.; Loos, W. J.; Nooter, K.; Stoter, G.; Sparreboom, A. Clin. Cancer Res., 2001, 7, 2182-94. O'Dwyer, P. J.; Catalano, R. B. J. Clin. Oncol., 2006, 24, 4534-8. Sai, K.; Saeki, M.; Saito, Y.; Ozawa, S.; Katori, N.; Jinno, H.; Hasegawa, R.; Kaniwa, N.; Sawada, J.; Komamura, K.; Ueno, K.; Kamakura, S.; Kitakaze, M.; Kitamura, Y.; Kamatani, N.; Minami, H.; Ohtsu, A.; Shirao, K.; Yoshida, T.; Saijo, N. Clin. Pharmacol. Ther., 2004, 75(6), 501-15. United States Food and Drug Administration: Campostar label. Beauclair, S.; Formento, P.; Fischel, J. L.; Lescaut, W.; Largillier, R.; Chamorey, E.; Hofman, P.; Ferrero, J. M.; Pages, G.; Milano, G. Ann. Oncol., 2007, 18(8), 1335-41. Nakajima, M.; Komagata, S.; Fujiki, Y.; Kanada, Y.; Ebi, H.; Itoh, K.; Mukai, H.; Yokoi, T.; Minami, H. Pharmacogenet. Genomics, 2007, 17(6), 431-45. DeMichele, A.; Aplenc, R.; Botbyl, J.; Colligan, T.; Wray, L.; Klein-Cabral, M.; Foulkes, A.; Gimotty, P.; Glick, J.; Weber, B.; Stadtmauer, E.; Rebbeck, T. R. J. Clin. Oncol., 2005, 23(24), 5552-9. DeMichele, A.; Gimotty, P.; Botbyl, J.; Aplenc, R.; Colligon, T.; Foulkes, A. S.; Rebbeck, T. R. J. Clin. Oncol., 2007, 25(35), 5675-7. Musolino, A.; Naldi, N.; Bortesi, B.; Pezzuolo, D.; Capelletti, M.; Missale, G.; Laccabue, D.; Zerbini, A.; Camisa, R.; Bisagni, G.; Maria Neri, T.; Ardizzoni, A. J. Clin. Oncol., 2008, 26(11), 1789-96. Kiyotani, K.; Mushiroda, T.; Sasa, M.; Bando, Y.; Sumitomo, I.; Hosono, N.; Kubo, M.; Nakamura, Y.; Zembutsu, H. Cancer Sci., 2008, 99(5), 995-99. Goetz, M. P.; Rae, J. M.; Suman, V. J.; Safgren, S. L.; Ames, M. M.; Visscher, D. W.; Reynolds, C.; Couch, F. J.; Lingle, W. L.; Flockhart, D. A.; Desta, Z.; Perez, E. A.; Ingle, J. N. J. Clin. Oncol., 2005, 23(36), 9312-8. Jin, Y.; Desta, Z.; Stearns, V.; Ward, B.; Ho, H.; Lee, K. H.; Skaar, T.; Storniolo, A. M.; Li, L.; Araba, A.; Blanchard, R.; Nguyen, A.; Ullmer, L.; Hayden, J.; Lemler, S.; Weinshilboum, R. M.; Rae, J. M.; Hayes, D. F.; Flockhart, D.A. J. Natl. Cancer Inst., 2005, 97(1), 30-9. Fujita, K. Curr. Drug Metab., 2006, 7(1), 23-37. Lim, H.-S.; Ju, Lee, H.; Seok Lee, K.; Sook Lee, E.; Jang, I.-J.; Ro, J. J. Clin. Oncol., 2007, 25(25), 3837-45. Mortimer, J. E.; Flatt, S. W.; Parker, B. A.; Gold, E. B.; Wasserman, L.; Natarajan, L.; Pierce, J. P. Breast Cancer Res. Treat., 2008, 108(3), 421-426. Buzdar, A. U.; Cuzick, J. Clin. Cancer Res., 2006, 12(3 Pt 2), 1037s-1048s. Goetz, M. P.; Knox, S. K.; Suman, V. J.; Rae, J. M.; Safgren, S. L.; Ames, M. M.; Visscher, D. W.; Reynolds, C.; Couch, F. J.; Lingle, W. L.; Weinshilboum, R. M.; Fritcher, E. G.; Nibbe, A. M.; Desta, Z.; Nguyen, A.; Flockhart, D. A.; Perez, E. A.; Ingle, J. N. Breast Cancer Res. Treat., 2007, 101(1), 113-21.
Frontiers in Drug Design & Discovery, 2009, 4, 135-161
135
More Than Skin Deep: The Human Skin Tissue Equivalent as an Advanced Drug Discovery Tool Alexandra S. Bause, Sarah D. Lamore and Georg T. Wondrak* University of Arizona, College of Pharmacy, Arizona Cancer Center, Tucson, AZ, USA Abstract: Advanced human tissue equivalents are an emerging drug discovery tool that aims at minimizing costly candidate attrition in later stages of preclinical and clinical development. 3-D cellular tissue models amenable to screening in 96-well plate format and drug monitoring in real time provide more accurate compound screening in a physiologically relevant context, not achievable in traditional two dimensional, single cell type-based assays. Currently, equivalents that represent the healthy or pathological physiology of human tissues including skin, vasculature, lung, prostate, and various malignant tumors are used for rapid activity screening, toxicity profiling, and pharmacokinetic characterization of compound libraries. Human skin equivalents with intact stratum corneum reconstructed in vitro from matrix-embedded dermal fibroblasts and epidermal primary keratinocytes closely resemble the complex architecture and functional complexity of skin. Using human skin equivalents, delivery of novel topical agents for photoprotection has been studied, and toxicity and activity profiling of skin care products and FDA-approved drugs including sunscreens has been performed. Importantly, human skin equivalents are better predictors of drug activity than available animal models as mouse and guinea pig skin do not reflect the architecture of human skin in terms of physico-optical properties such as light reflectance and scattering, epidermal thickness, cellularity, and biochemical composition. Recently, cellular composition of advanced commercial skin reconstructs has been further optimized by incorporation of melanogenic melanocytes and immunomodulatory dendritic (Langerhans) cells in order to assess drug modulation of cutaneous pigmentation, inflammation, photo-immunosuppression, and photo-carcinogenesis. Continuous progress in skin equivalent engineering will ensure the expanding role of skin equivalents in disease model-based assays for rapid identification and development of novel cutaneous therapeutics.
1. INTRODUCTION. ENHANCING THE DRUG DISCOVERY PROCESS USING THREE-DIMENSIONAL RECONSTRUCTED TISSUE MODELS The paradigm of modern drug discovery is aptly represented by a complex funneling process that begins with initial identification and validation of a molecular target which
*Corresponding Author: Tel: 520-626-9017; Fax: 520-626-3797; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
136 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
is causatively involved in a disease process and therefore selected for future pharmacological intervention as shown in Fig. (1). Further activities should then ultimately lead to market introduction of a novel drug that provides superior therapeutic benefit. In a time of ever increasing biological understanding of the molecular pathology underlying human disease states, no dearth of promising drug targets exists. However, it seems that many academic and industrial drug development projects have proven fruitful by further increasing our biological understanding of the disease process (target discovery), but fail to deliver a novel therapeutic agent that could enter the marketing phase [1-4].
target target identification & validation assay development & screening
lead identification
lead
• potency • selectivity • druglikeness • toxicological profile
lead optimization
preclinical development costliness of candidate attrition
clinical trials
drug Fig. (1). The obligatory funneling process underlying modern drug discovery. The use of human skin reconstructs as a versatile drug discovery tool throughout all developmental stages from target identification to preclinical development of cutaneous therapeutics may significantly reduce costly candidate attrition at late stages.
The reasons underlying the discrepancy between success rates of pharmaceutical target versus drug discovery are manifold and may be connected to the very nature of pharmaceutical discovery. The process of iterative selection and optimization of drug candidates that must outperform earlier therapeutic agents is intrinsically associated with high candidate attrition rates. Apart from limitations associated with high throughput screening of compound libraries with insufficient molecular diversity and inadequate coverage of chemical space, an inadequate quality of cell-based screening systems has been invoked to rationalize the high failure rate correlated with generation of promising
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 137
preclinical candidates. Indeed, it has been argued that cell-based screening methodologies and even advanced animal models such as mouse xenograft tumor models that are dependent on single-cell type cell culture insufficiently represent the biological complexity of the molecular target in a meaningful tissue context. After target identification and validation, development of screening assays aims at the identification of promising lead compounds that then can undergo iterative lead optimization based on preclinical efficacy, toxicity, and pharmacokinetic profiling. These efforts will then lead to the identification of clinical candidates that can be advanced through the obligatory phases of clinical testing. Obviously, a major determinant of the success rate of this funneling process is the validity and predictive power of the assay systems that are used for efficacy and toxicology screening. Advanced human tissue equivalents that mimic the exact physiology of the human target tissue to a great extent are an emerging drug discovery tool that aims at minimizing costly candidate attrition in later stages of preclinical and clinical development. Three dimensional cellular tissue models amenable to screening in 96-well plate format and drug monitoring in real time provide more accurate compound screening in a physiologically relevant context, not achievable in traditional two dimensional, single cell type-based assays [5]. Currently, equivalents that represent the healthy or pathological physiology of human tissues including skin, vasculature, lung, prostate, and various malignant tumors are used for rapid activity screening, toxicity profiling, and pharmacokinetic characterization of compound libraries and selected lead compounds during preclinical development [6,7]. As discussed in this review, the use of human skin reconstructs is now firmly established as a versatile drug discovery tool throughout all developmental stages from target identification to preclinical development of cutaneous therapeutics. Due to technological accessibility, commercial availability, and biological validity of skin tissue reconstructs, skin pharmacology is now fundamentally impacted by this technology that rapidly emerges as the novel gold standard of preclinical skin models. Skin reconstructs of variable complexity have been generated ranging from human differentiated epidermal equivalents with stratum corneum reconstructed to full thickness reconstructs that include extracellular matrix-embedded dermal fibroblasts and other skin cells of particular interest including melanocytes and Langerhans cells [8,9]. All of these closely resemble the complex architecture and functional complexity of skin. Obviously, cellular and acellular composition of skin reconstructs is highly adjustable and can be customtailored in order to provide a relevant model of a specific pathological condition. For example, photoaged skin can be reconstructed by incorporation of photo-crosslinked collagen and senescent fibroblasts during dermal reconstruction, and psoriatic skin can be reconstructed by incorporation of patient derived keratinocytes cultured from psoriatic lesions [10]. Using human skin equivalents delivery of novel topical agents for skin photoprotection and anti-psoriatic intervention has been studied, and pharmacodynamic profiling of FDA-approved drugs including sunscreens has been performed [11]. Importantly, human skin equivalents are superior predictors of drug photoprotective activity compared to available animal models, since mouse and guinea pig skin do not reflect the architecture of human skin in terms of physico-optical properties such as light reflectance and scattering, epidermal thickness, cellularity, permeability, and biochemical composition. Moreover animal models of specific skin conditions are often completely unavailable or poorly represent the human disease. For example, transgenic melanoma mo-
138 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
dels in mice are of debatable relevance to the human disease, and drug candidates that display anti-melanoma activity in these transgenic mice may be completely inactive against human melanoma. Therefore, the commercial availability of human melanoma reconstructed skin that mimics the successive progressional stages of the disease may provide a stringent and powerful screening tool for the identification of promising chemopreventive and chemotherapeutic anti-melanoma agents as discussed below [12]. In addition to high costs associated with animal experimentation and poor representation of human skin physiology in animal models, ethical concerns and regulatory guidelines limit or completely ban the extensive use of animal models particularly in the earlier stages of the drug discovery process that would require screening in high or moderate throughput. In the European Union (EU), animal testing for cosmetic ingredients is completely banned by 2009, and ethical and legal guidelines in other countries equally create an urgent need for alternative in vitro methods that allow the cost effective and highly predictive toxicity and efficacy assessment of topical consumer products and pharmaceutical agents. In summary, the emerging role of human skin reconstructs as superior drug discovery tool is based on the following characteristics: I. adequate representation of human skin; II. high model-to-human predictability of drug action; (i.e., drug effects in the model are predictive of therapeutic efficacy in humans; III. adaptability to screening in high throughput; IV. commercial availability and cost competitiveness; V. inter-laboratory reproducibility and standardization; VI. adjustable degree of tissue complexity (e.g. simple epidermal reconstruct versus skin-type specific pigmented skin reconstructs); VII. validated tissue biomarkers that specify test compound action and can easily be measured (e.g. toxicity markers for corrosion and irritation studies). VIII. superiority over animal experimentation based on ethical, legal, economical, and operational considerations; IX. superior predictive power and biological validity over cell-based and animal-based screening methods Three major applications of reconstructed human skin can be distinguished: First, skin reconstructs serve as basic research tools for skin biology. Second, skin reconstructs serve as drug discovery tools that allow drug target identification and validation. Moreover, compound screening for efficacy, toxicity, and pharmacokinetic profiling can be performed. Thirdly, skin reconstructs are used therapeutically in the area of skin tissue replacement and reconstructive surgery, for example in burn victims. 2. THE BIOLOGICAL COMPLEXITY OF HUMAN SKIN Human skin is a complex organ composed of a variety of specialized cells, organized into three highly differentiated principal layers: the epidermis, the outermost layer, followed by the dermis (jointly referred to as cutis and separated by a basement membrane), and the deeper localized subcutis, mainly composed of adipose tissue [13]. The epidermis is a squamous epithelium derived from terminal differentiation of keratinocytes with associated hair follicles and glandular structures (sebaceous glands and sweat glands). Keratinocytes constitute approximately 80% of epidermal cells forming the multilayered differentiated epidermis comprising (-from inside out-) the stratum germinativum/basale, the stratum spinosum, the stratum granulosum, the stratum lucidum, and the stratum corneum. The avascular epidermis is characterized by self-renewal with undifferentiated basal cells proliferating and differentiating into a stratified multilayered tissue, where complete turnover of the interfollicular epidermis occurs every
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 139
four weeks. Multipotent stem cells within the epidermis produce daughter cells that differentiate along multiple lineages. Remarkably, hair follicle lineages, sebocytes and interfollicular epidermis originate from stem cells within the hair follicle bulge. The differentiated epidermis acts as an effective barrier against physical (UV radiation), chemical (xenobiotics), and biological (microbial) insults. The outermost acellular stratum corneum consists of dead corneocytes (cornified keratinocytes), the endproduct of epidermal terminal differentiation. Cornification involves changes in expression patterns of keratins, where basal layers express keratins 5 and 14, while the suprabasal keratinocytes produce keratins 1 and 10. Furthermore, cornification involves synthesis of involucrin, loricrin, and small proline rich proteins as structural constituents of the cornified envelope that is crosslinked through transglutaminase enzymatic activity. As another essential part of the human skin barrier, proteolysis of the structural protein filaggrin during terminal differentiation produces a specific mixture of amino acids (serine, glycine, arginine, ornithine, citrulline, alanine, histidine) and their derivatives (pyrrolidone-5-carboxylic acid, urocanic acid) called the natural moisturizing factor (NMF). Components of the NMF are crucially involved in skin hydration, pH control, antimicrobial defense, immunomodulation, and UV-photoprotection [14]. Keratinocytes produce a multitude of molecules involved in barrier formation (structural proteins including keratins, filaggrin, loricrin, and involucrin, and lipids such as ceramides and cholesterol), melanogenesis (-MSH), mitogenesis (CCL5), host defense (defensins), inflammatory signaling (IL, TNF-), angiogenesis (VEGF), and cellular differentiation and calcium homeostasis (vitamin D3 and calcitriol). In addition to keratinocytes, the pigment-producing melanocytes at the epidermal– dermal junction are interspersed among every 5–10 basal keratinocytes, forming the ‘epidermal–melanin unit’. It is thought that one melanocyte synthesizes and transports melanin-containing melanosomes to the surrounding keratinocytes for phagocytotic uptake and formation of supranuclear melanin caps, protecting keratinocyte DNA from the genotoxic effects of ultraviolet light [15]. Another important cellular constituent of human epidermis are Langerhans cells (LC), bone marrow derived dendritic cells, which circulate in the peripheral blood as progenitors, and then migrate into the suprabasal epidermis. These cells play a major role in the skin immune defense system. After exposure to exogenous antigens, LC migrate to the regional lymph nodes to present them to naive T cells. Recently, LC have emerged as crucial epidermal targets of UV-photodamage involved in induction of photo-immunosupression thought to contribute to solar photocarcinogenesis [16]. The dermis with its mesenchymal components, i.e. fibroblasts and blood vessels embedded in fibroblast-derived extracellular matrix containing collagen, elastin and glycosaminoglycans, provides mechanical and nutritional support and rigidity to the skin. At least three subpopulations of dermal fibroblasts have been identified which occupy unique niches in the dermis: papillary dermal fibroblasts (superficial dermis), reticular fibroblasts (deep dermis), and hair follicle-associated fibroblasts [17]. Importantly, keratinocytes and fibroblasts interact with each other to maintain skin integrity through paracrine cytokine loops. For example, fibroblasts modulate epidermal homeostasis and reepithelialization after epidermal damage with production of keratinocyte growth factor and TGF-1, and fibroblast–keratinocyte interactions are involved in the formation of the basement membrane [18].
140 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
In addition to these basic skin constituents other specialized cells including sebocytes, mast cells, sensory Merkel cells, and adipocytes contribute to the fascinating functional and structural complexity of human skin. Remarkably, simple skin models can be generated from single cellular constituents using keratinocytes (epidermal reconstruct) or fibroblasts (dermal reconstruct). More advanced skin reconstructs that better represent the functional complexity of skin are characterized by incorporation of multiple skin cell types including keratinocytes, melanocytes, Langerhans cells, and fibroblasts and their genetically engineered or malignant counterparts. 3. BASICS OF HUMAN SKIN RECONSTRUCTION Historically, the development of skin reconstructs as a drug discovery tool occurred in three stages based on fundamental advances in (I) skin cell culture technology, (II) the generation of complex skin tissue models for biological studies, and (III) the optimization and validation of skin models for toxicological and pharmacological studies. (I)
Cell culture. All tissue reconstruction is based on developing the fundamental technology of culturing cells in monolayers. For human skin keratinocytes, this was achieved through serial cultivation of strains of human epidermal keratinocytes with the formation of keratinizing colonies from single cells using serum containing culture medium and a feeder-layer of pre-irradiated quiescent fibroblasts [19]. The inclusion of serum became dispensable when calcium-regulated differentiation of normal human epidermal keratinocytes in chemically defined clonal culture and serum-free serial culture was reported [20], followed by cultivation of other human skin cell types including melanocytes and Langerhans cells [21,22].
(II)
Tissue reconstructs. Cellular monolayers obviously do not represent the functional complexity of differentiated skin in vivo and are therefore poor models for molecular studies of skin biology and pathophysiology. Moreover, drug discovery targeting complex alterations of skin structure and function (barrier impairment, immunological and inflammatory dysregulation, altered epidermal differentiation and melanogenesis, carcinogenesis, photodamage, and chronological aging) depends on model systems that adequately represent the target tissue. A fully differentiated epidermal reconstruct was first generated by Pruniéras et al. who described methods for cultivation of keratinocytes at the air–liquid interface [23], and many skin reconstructs of diverse complexity have been created as discussed below. Interestingly, the differences between gene expression profiles of reconstructed human epidermis and classical cultures of keratinocytes based on cDNA expression arrays comprising 505 genes related to cutaneous biology have been examined. A comparison of gene expression between keratinocytes grown as cultures on plastic dishes or in a three-dimensional reconstruct identified six genes with considerably higher expression in the reconstruct. All of these genes (keratin 1, corneodesmosin, filaggrin, loricrin, calmodulin-like skin protein, and caspase 14) are related to keratinocyte terminal differentiation as expected in a model that adequately represents human epidermis [5]. Importantly, a major driving force behind technological advances in full thickness skin reconstruction is the use of reconstructed skin for therapeutic applications as skin substitutes [24,25]. Currently, skin replacement based on epidermal, dermal, or full-thickness skin substitutes is in clinical use to treat different types of difficult-to-heal wounds including
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 141
large and deep burns, trauma-induced wounds, and ulcers resistant to conventional therapies. (III)
Skin reconstruction as a pharmacological tool. Moreover, skin reconstructs have become the focus of increased research activities since it has been demonstrated that they may excellently reproduce certain aspects of the pathology of various skin diseases including skin carcinogenesis and solar photodamage. For example, various studies demonstrated the usefulness of reconstituted three-dimensional human skin as a novel in vitro model of skin carcinogenesis using established carcinogens including benzo[a]pyrene, ultraviolet B (UVB) radiation, UVA, and psoralen-UVA [11,26]. Molecular consequences of carcinogen exposure (upregulation of c-fos and p53 expression, DNA adduct formation) were similar to those reported previously in human and murine skin after carcinogenic insult demonstrating that skin reconstructs can be used as a discovery tool for assessing genotoxicity and mechanism of action of mutagens/carcinogens in human skin [26]. Recently, melanoma skin reconstruction has been achieved representing the progressional stages of the disease and a commercial model is now available as discussed below [12,15]. Remarkably, it is now possible to reconstruct human skin containing the major cellular targets for skin photodamage: keratinocytes (UVB-induced nonmelanoma skin cancer), Langerhans cells (photo-immunosuppression), melanocytes (melanoma), and fibroblasts (UVA-induced photoaging). Inclusion of human skin color- and phototype-specific melanocytes allows the reconstruction of human skin with differential inducible and constitutive pigmentation and photosensitivity [27,28]. Further progress may be facilitated by inclusion of optimized biomaterials (e.g. epidermal reconstruction on fibrin scaffolds) and genetically modified cellular components.
A representative full thickness reconstruct that has been generated according to the following standard procedure closely resembles the anatomy of human skin as shown in Fig. (2): First, dermal equivalents are prepared by mixing human dermal fibroblasts and native bovine type I collagen in a Petri dish. After lattice contraction for three days at
A
B
Fig. (2). Human reconstructed skin closely resembles the architecture of skin in situ. Human skin reconstructed in vitro (B) closely resembles normal human skin (A) displaying a well-stratified epithelium covered by a dense stratum corneum on top of a fibroblast populated dermis. However, the absence of interpapillary ridges, functionally significant extensions of the epithelium into the connective tissue, and the somewhat less pronounced thickness of the stratum corneum are apparent in reconstructed skin. (reproduced with permission from reference [64], Elsevier Ireland Ltd.).
142 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
37°C, 5% CO2, adult human epidermal keratinocytes grown in primary culture are seeded on this support using stainless rings. The rings are then removed and the cultures are kept submerged for seven days, allowing the cells to form a monolayer. The medium used is composed of minimum essential Eagle’s medium supplemented with 10% fetal calf serum, 10 ng/mL epidermal growth factor, 8.4 ng/mL cholera toxin, 0.4 g/mL hydrocortisone. The cultures are then raised at the air–liquid interface on grids and kept one week to allow keratinocytes to stratify and differentiate completely during which time the reconstruct is fed by capillarity [11]. 4. SKIN RECONSTRUCTS AS A DRUG DISCOVERY TOOL The broad availability of reconstructs that correctly represent the complex biology of healthy human skin facilitated research that aimed at defining the toxicological effects of chemical and physical insult on human skin. Indeed, established biomarker profiles indicative of exposure-induced toxicity in human skin can easily be studied in reconstructs with high reproducibility. Reconstructed skin has therefore become an indispensable screening tool for toxicological profiling of test compounds intended for future topical application [29]. Moreover, screening of compounds that exert protective effects against environmentally relevant insults such as solar UV-radiation has rapidly become a major application of skin reconstructs [8]. Since pathological key mechanisms involved in complex diseases affecting human skin such as chronic inflammation, psoriasis and cancer are adequately reproduced in skin reconstructs, pharmaceutical screening of compound libraries with high throughput and predictive power for drug efficacy seems feasible. Usefulness of these models has been established for activity screening of topical retinoids and glucocorticoids as discussed below. In particular, reconstructed human skin has rapidly emerged as a superior experimental model for the identification of novel pharmacological and cosmeceutical agents designed to interfere with skin photodamage and chronological aging including sunscreens and topical antioxidants as discussed below. Various human epidermal reconstructs are now commercially available and have been used successfully as pharmaceutical screening tools. For example, EpiDerm™ (MatTek,Ashland, MA, USA; http://www.mattek.com) consists of normal, human epidermal keratinocytes that form a multilayered, highly differentiated model of the human epidermis on cell culture inserts. Recently, EpiDermFT, a full thickness skin model comprising normal human epidermal keratinocytes and dermal fibroblasts has been created enabling in vitro studies on fibroblast-keratinocyte cell interactions. Moreover, specialized models representing pigmented human skin (MelanoDermTM), human melanoma skin (MLNM-FT-A375), and oral epithelium (EpiOralTM and EpiGingivalTM) are available. Another common model, EPISKIN TM (SkinEthic, Nice, France; http://www. skinethic.com), consists of adult human keratinocytes cultured on a base of bovine collagen type I and III, coated with a thin layer of human collagen IV, and forming a fully differentiated epidermis with a functional horny layer. From the same manufacturer, another in vitro reconstructed human epidermis comprising normal human keratinocytes cultured in chemically defined medium on inert polycarbonate filters at the air-liquid interface can be obtained (RHE by SkinEthic). Again, more specialized models that represent pigmented epidermis (RHPE) and oral (HOE) and gingival epithelium (HGE) are available.
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 143
5. SKIN RECONSTRUCTS AS A TOXICITY SCREENING TOOL Skin toxicity as a result of compound exposure comprises a variety of adverse effects ranging from acute corrosive tissue damage to inflammatory irritation, and many parameters including speed of onset, persistence, and reversibility define the severity of these reactions [29]. Animal skin toxicity testing is compromised both for its limited predictive capacity for human toxicity, as well as for ethical and legal restrictions associated with animal experimentation. Reconstructed human epidermis allows topical application of test compounds and identification of agents that (I) corrode the skin (‘corrosion’), (II) irritate the skin or the eyes (‘irritation’), (III) sensitize the skin (‘sensitization’), or (IV) elicit toxic responses in combination with UV-light (‘phototoxicity’) as summarized in Fig. (3) [29-33].
NH O
O O
OH
S
O O
N
N
F Na+
HO O2N
NO2
F O
O N
OH
N
Cl CnH2n+1
n=8, 10, 12, 14, 16, 18 Corrosives
Irritants
N O
N
O O
Sensitizers
Cl
S Photosensitizers
IL-1α IL-8 TNFα PGE2 Hsp27 cell death
skin damage, pain, edema, erythema, itching Fig. (3). Biomarkers of skin toxicity validated in human skin reconstructs. For abbreviations see text. Skin toxicity inducing compounds by category (from top to bottom): acrylic acid, 2-tertbutylphenol (corrosion); SDS, benzalkonium chloride (irritation); 1-fluoro-2,4-dinitrobenzene, oxazolone (sensitization); ciprofloxacine, chlorpromazine (phototoxicity).
Current international regulations require assessment of skin corrosion for all chemicals placed on the market [34]. Predictive test methods must define the toxicity profile of consumer products, single ingredients (e.g. detergents), experimental drugs, cosmeceuticals, and other agents leading to classification as irritant/slightly irritant/non-irritant or
144 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
severely corrosive/ corrosive/ non-corrosive. In recent years, skin reconstructs have replaced more traditional animal-based test methods including the long established Draize test based on rabbit eye exposure as discussed in [29]. In the EU, the use of human skin reconstructs as predictive test models for corrosion testing of dangerous substances are now regulated through the OECD 431 guideline [35]. A common irritation protocol based on the use of human skin reconstructs has been established by the ECVAM (European Centre for the Validation of Alternative Methods) and is the subject of further validation studies [33,36,37]. 5.1. Corrosion Corrosive materials are identified by their ability to reduce cell viability below defined threshold levels at specified exposure periods [38]. Corrosive agents penetrate or erode the stratum corneum and cause cell death in the underlying cell layers. Skin damage from corrosion occurs within minutes after chemical exposure and results in irreversible tissue necrosis by physicochemical processes that occur independent of a biological damage response. An inflammatory response of the surrounding tissue is commonly a consequence of the skin corrosion process [38]. In an example test protocol and prediction model using reconstructed human epidermis, corrosion potential of test material is predicted from the mean tissue viabilities assessed after exposure times as short as 3 min [33]. Test materials are applied directly to the stratum corneum and exposure to the chemical is terminated by rinsing with phosphate buffered saline. Tissue viability is then assessed using the MTT assay with formazan solvent extraction followed by photometric analysis compared to untreated control tissue. A chemical is classified as ‘corrosive’ if the relative tissue viability is decreased below 50% after 3 min of exposure to a test material. In addition, test materials classified ‘non-corrosive’ after 3 min (viability > 50%) are classified ‘corrosive’ if the relative tissue viability is decreased below 15% after exposure for 1 h. Similar protocols have been implemented and used successfully according to ECVAM guidelines using various commercial and academic skin reconstruct models [33]. 5.2. Irritation During early development of novel cosmetic and pharmaceutical formulations designed for topical delivery, the potential for the induction of adverse reactions due to skin irritation must be assessed early on in order to exclude irritants from further development. In contrast to in vitro tests for skin corrosion, validation of in vitro tests for assessing the acute skin irritation potential of chemicals is ongoing, and consequently no generally accepted regulatory testing requirements based on the use of human skin reconstructs have been established. Skin irritation is an inflammatory damage response that occurs in the absence of cellular necrosis [29]. The principal clinical signs of cutaneous irritancy in humans are erythema, induration, and edema. Acute skin irritation occurs as a response to a single exposure to potent irritants, whereas cumulative skin irritation, the most common type of skin irritation, occurs only after repetitive exposures to mild irritants. Inflammatory events associated with irritation originate from the complex interaction between epidermal cells, dermal fibroblasts, endothelial cells and invading leukocytes that involve cytokines, lipid mediators, and other signaling molecules including reactive oxygen species as summarized in Fig. (3). Initiation of irritation can result from physi-
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 145
cochemical properties of the test agent that leads to lipid solving effects impairing the integrity of the stratum corneum lipid barrier and damaging skin cell membranes. Apart from their role as epidermal structural components, keratinocytes are key initiators of skin inflammatory and immunological reactions involved in skin irritation in response to a chemical or physical insult. Irritation by surfactants is a well-studied mechanism of chemical-induced skin irritation, with sodium dodecyl sulfate (SDS) serving as a reference substance for assessing irritant potentials in vivo [39]. Cytokines released from keratinocytes are crucial upstream mediators of skin irritation. The proinflammatory cytokine IL is considered a key inducer of the skin inflammatory cascade. This major biomarker of skin irritation, that is contained in the cytoplasm of keratinocytes is released during inflammatory signalling in response to irritation [40], and can easily be assessed by ELISA analysis. In human skin reconstructs, release of IL in response to surfactant exposure such as benzalkonium chloride is rapidly triggered and correlates with loss of cell membrane integrity as assessed by LDH release and loss of mitochondrial energy production as assessed by the MTT assay [30,41]. Induction of acute cutaneous oxidative stress is an important consequence of irritant exposure and production of cellular ROS in keratinocytes may contribute to inflammatory signalling [42]. For example, IL synthesis and release are induced upon exposure to the irritant tributyltin, which is thought to increase generation of reactive oxygen species through impairment of mitochondrial respiration. Importantly, IL is a potent inducer of cellular signaling through the NFB proinflammatory pathway. Regulation of epidermal homeostasis and inflammatory responses are controlled by NFB transcriptional activity upregulating expression of proinflammatory cytokines including IL-6 and IL-8. Indeed, release and cytoplasmatic concentration of IL in two full thickness skin equivalents after SDS induced irritation has been documented recently [43]. Studies comparing release of IL after surfactant exposure between full thickness and epidermal skin equivalents demonstrate that keratinocytes are the exclusive source of IL in skin reconstructs. In contrast, communication between keratinocytes and fibroblasts regulates an acute inflammatory response with IL-6 and IL-8 release following irritant exposure. Chemotactic IL-8 is released from tissue reconstructs in response to irritant exposure, and the ratio of released IL-8 to IL has been suggested as a biomarker that discriminates irritants from sensitizers [44]. Another pleiotropic inflammatory cytokine, TNF-, is stored in the epidermal mast cells and also produced by keratinocytes following irritant exposure inducing the expression of endothelial adhesion molecules. In a full thickness skin model TNF- expression levels have been proposed to serve as skin irritation markers based on mRNA detection. Apart from inflammatory cytokines, inflammatory lipid metabolites derived from arachidonic acid are synthesized and released in response to skin irritant exposure as discussed in [29]. Arachidonic acid is released from membrane pools by enzymatic activity of phospholipase A2, an important downstream target activated by IL-1. Further biotransformation of arachidonic acid by cyclooxygenase and PGE synthase induces formation of strongly inflammatory metabolites including prostaglandin E2 released in irritant exposed epidermal reconstructs as well as in skin in vivo. PGE2 is therefore an important marker of irritation that can be easily assessed by ELISA analysis of tissue culture medium. Moreover, in response to the irritant benzoyl peroxide the release of leukotriene B4 and 15-hydroxyeicosatetraenoic acid from arachidonate prelabelled skin equivalents has been demonstrated.
146 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
Discovery of novel markers indicative of skin irritation elicited by a test compound is ongoing with a focus on differential expression patterns of irritated and non-irritated skin equivalents using genomic and proteomic analysis as reviewed in [29]. For example, differential gene expression using cDNA array technology after treatment of epidermal reconstructs with SDS has been observed. Moreover, screening for differentially abundant proteins in SDS treated human skin by 2D-PAGE revealed seven potential new epidermal markers for skin irritation, among which heat shock protein 27 (Hsp27) was the most prominently up-regulated [45]. An in vitro human reconstructed epidermis model used for screening acute and chronic skin irritation potential was validated against in vivo data from skin tolerability studies in order to establish predictivity of skin reconstruct based models for acute and chronic skin irritation [46]. The cumulative irritation potential of topical agents (SDS, calcipotriol and trans-retinoic acid) was evaluated in a human clinical study. In parallel, using a reconstructed epidermis model, compoundinduced modulation of cell viability, the release and gene expression of IL-1 and IL-8, and morphological changes were evaluated during 3 days as endpoints representative of an inflammatory reaction. It was found that the topical agents that were non-irritating in the human study were non-cytotoxic and did not induce cytokine expression in the in vitro acute model. Moreover, all irritating controls exhibited specific cell viability and cytokine patterns, which were predictive of the in vivo human data. In this study, the human reconstructed epidermis model was a reliable preclinical tool for the prediction of the irritation potential of topical products. In another study, the predictive ability of reconstructed human epidermis equivalents for the assessment of skin irritation of cosmetics was assessed [30]. To this end, 22 formulations from product development test representing different cosmetic product classes were tested in vivo using the modified Frosch-Kligman soap chamber patch test with repetitive occlusive application and in vitro using a series of epidermis equivalents. In vivo, skin reactions including erythema, dryness, fissures, and transepidermal water loss were evaluated. In vitro, cell viability and the extracellular release of IL and LDH into the culture medium collected after topical application of the products were assessed. In general, in vivo and in vitro classifications as ‘irritant’ and ‘non-irritant’ of all test products was fully concordant suggesting the usefulness of the skin reconstructs for routine screening of developmental formulations before human in vivo dermatological evaluation. It is important to note that apart from cutaneous skin irritation testing performed in reconstructs, in vitro testing of intra-orally applied consumer products has been performed using a human oral epithelial reconstruct generated from primary normal human oral keratinocytes and fibroblasts [47]. In reconstructs exposed to low concentrations (0.015%) of SDS, increased epithelial thickness and proliferation based on immunohistochemical detection of the marker Ki-67 were observed, whereas reduced epithelial thickness and cell proliferation and massive induction of epithelial cell death was observed upon exposure to higher SDS concentrations (> 0.15%). 5.3. Sensitization Sensitization occurs as an immunological reaction in response to xenobiotics that form allergenic epitopes on epidermal proteins. Sensitization induces activation of immuno-competent dendritic cells (Langerhans cells) that migrate from the epidermis to the regional lymph nodes, where the processed antigen is presented to naive T-cells [48].
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 147
Skin sensitization assessment can be performed in vivo using complex procedures including including the local lymph node assay [44]. Recently, analysis of cell viability and IL and IL-8 expression and release in reconstructed human epidermis has been documented as a reliable predictor of in vivo skin irritation and/or sensitization. This method allowed discrimination between 1-chloro-2,4-dinitrobenzene, nickel sulfate, oxazolone, 2,4-dinitrofluorobenzene, and 2,4,6-trinitrobenzenesulfonic acid as established skin sensitizers from benzalkonium chloride, benzoic acid, and SDS as established skin irritants. The epidermal reconstruct used in this study recreates many of the structural and functional features of the epidermis, but is exclusively composed of keratinocytes. However, the absence of Langerhans cells and leukocytes reduces the complexity and biological relevance of this model. Recently, a Langerhans cell containing reconstructed human epidermis has been generated and successfully employed as screening tool to evaluate sensitization potential of test compounds [16,48]. Topical exposure of these reconstructs to known allergens and allergen-inducible cytokines resulted in an activated appearance of the Langerhans cells. Concomitantly, overexpression of IL-1 and the co-stimulatory Langerhans surface epitope CD86 mRNAs were detected in the reconstructed epidermis. 5.4. Phototoxicity During topical application of chemical compounds the formation of reactive photoexcited states may occur as a result of solar exposure of skin [49]. The specific photochemical reactivity of a compound is determined by the chemical structure of its lightabsorbing chromophore and may lead to light-driven production of cytotoxic singlet and triplet states and reactive oxygen species, including singlet oxygen and superoxide radical anions. Moreover, the photoexcited compound itself may react with tissue proteins leading to the formation of antigenic epitopes in skin that illicit an immune response upon repetitive exposure. Indeed, phototoxicity of cosmetic and pharmaceutical agents that induces photo-irritation, an inflammatory skin reaction in response to exposure to the combined action of solar photons and chemical agent, is a serious adverse reaction associated with many compounds that contain a photoactive chromophore. Early assessment of the phototoxic potential of novel compounds is therefore imperative, and three-dimensional skin models have been used as efficient and cost-effective phototoxicological screening tools [50-52]. Phototoxic standards (e.g. phenothiazines and fluoroquinolones) and test compounds are applied to the reconstruct and then irradiated with UV-light (-most often noncytotoxic doses of UVA and visible light-) to trigger a phototoxic reaction. Established markers of phototoxicity in reconstructed skin include leakage of LDH and loss of viability based on the MTT assay, and increased photoinflammatory signaling through IL and IL-8 release and expression of IL-8 mRNA. Reconstruct-based phototoxicity studies are often complemented by simple cell-based and chemical assays, e.g. photobinding to human serum albumin indicative of a photoallergic potential. 6. SKIN RECONSTRUCTS AS A DISCOVERY TOOL FOR CUTANEOUS PHARMACOLOGY Experimental evidence has shown that the cutaneous pharmacology of many test compounds can be studied in three-dimensional skin reconstructs.
148 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
6.1. Reconstruct-Based Pharmacokinetic Characterization of Test Compounds Kinetics of cutaneous absorption and residence time, permeation and penetration potentially leading to systemic availability is an important quality parameter of pharmaceutical and cosmetical test compounds and also impacts the skin toxicology of hazardous agents. Reconstructed human epidermis is an established tool for in vitro skin absorption studies of chemical compounds performed according to nationally-defined regulatory guidelines [31]. For example, according to OECD guideline 428, human skin preparations can be replaced by reconstructed human epidermis models based on experimental evidence that demonstrates equivalence with data obtained in human skin [53]. Earlier research has studied epidermal reconstructs in parallel with human and animal skin to determine percutaneous absorption and skin metabolism of glucocorticoids and glucocorticoid esters, estradiol, as well as absorption of flufenamic acid, chlorpheniramine and various other agents [31,54]. Recently, barrier function of reconstructed human epidermis was compared to human heat-separated epidermis using the Franz diffusion cell procedure. Skin permeation of caffeine and testosterone used as test compounds differed significantly between the two test systems suggesting methodological limitations for replacement of human skin for in vitro permeability experiments [55]. It has long been discussed that skin equivalents may differ from normal human skin with regard to penetration rate of substances through the stratum corneum [29,31,56,57]. More recent experimental evidence suggests the feasibility of technical improvements that will allow reconstruct-based screening tests for skin penetration with sufficient reliability and throughput [53]. 6.2. Reconstruct-Based Pharmacodynamic Studies Beneficial and adverse effects of topically applied glucocorticoid ointments have been demonstrated in a full-thickness skin model [58]. Anti-inflammatory efficacy was validated by assessing suppression of UVB-induced IL-6 and IL-8 upregulation. Glucocorticoid-induced skin atrophy with decrease of epidermal thickness and impaired dermal type I collagen synthesis, both classic adverse effects of topical glucocorticoid therapy, were closely reproduced. Reconstructed epidermis serves as a benchmark for assessment of retinoid performance [59]. For example, in reconstructed epidermis the topical retinoid tretitoin alters tissue morphology through inhibition of epidermal differentiation and modulates biomarker expression that reflects retinoid effects in normal human skin. Recently, the therapeutic benefit of topical application of acitretin, an aromatic synthetic retinoid, currently taken as an oral drug for the treatment of severe psoriasis in adults, was demonstrated in reconstructed human epidermis for local management of keratinization disorders [60]. Using a similar experimental methodology, development of topically applied cosmeceutical agents that provide health benefit to human skin can be facilitated by reconstruct technology as discussed below. 6.3. Reconstruct-Based Drug Discovery Targeting Skin Photodamage and Aging It is now well established that acute and chronic photodamage contributes to skin photoaging and photocarcinogenesis as reviewed in [61]. Photo-oxidative stress, originating from reactive oxygen species formed in sun-exposed skin, is a pathological key me-
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 149
chanism that underlies both skin extracellular matrix protein damage (involved in wrinkle formation) and skin cell photomutagenesis (involved in carcinogenesis). Importantly, the mechanisms by which solar UV-irradiation causes skin photodamage are wavelength dependent as reviewed in [49]. UVB (290-320 nm) is thought to cause direct structural damage to DNA in the form of epidermal cyclobutane pyrimidine dimers (CPD) and other photoproducts. However, most of the solar UV energy incident on human skin derives from the deeply penetrating UVA region (> 95%, 320-400 nm) that is not directly absorbed by DNA. UVA-induced photodamage of human skin that contributes to tumorigenic initiation and progression of non-melanoma and melanoma skin cancer is thought to originate from photo-oxidative stress [62,63]. The structural and functional alterations that result from skin photooxidative stress depend on a cascade of chemical events initiated by photoexcited states of endogenous skin chromophores that act as photosensitizers. These endogenous skin components with sensitizer activity (including porphyrins, melanin precursors, riboflavin, pyridoxine, enzymatic collagen crosslinks, advanced glycation endproducts, and lipofuscin-type pigments) form redoxreactive excited states that lead to formation of reactive oxygen (ROS) and carbonyl species (RCS) by light-driven redox cycling [49]. These reactive intermediates then induce skin protein, lipid and DNA damage, and trigger alterations of redox and inflammatory signaling leading to photoaging and photocarcinogenesis. Chronic photooxidative stress also leads to skin photoaging, characterized by accumulation of senescent dermal fibroblasts, extracellular matrix remodeling with collagen crosslinking, proteasedependent collagen breakdown, overexpression of dysfunctional elastin (solar elastosis), and chronic inflammatory signaling. 6.3.1. Skin Reconstructs for Mechanistic Studies on Photodamage Induced by Solar Radiation Skin reconstructs are excellent models for mechanistic studies on molecular pathways involved in photodamage. After exposure to solar UV many important biomarkers indicative of photodamage are equally expressed in normal and reconstructed human skin as summarized in Fig. (4) [64]. Indeed, exposure of reconstructed human skin to erythemogenic doses of UVB radiation induces changes characteristic of a moderate sunburn reaction in human skin [11,65]. Cyclobutane pyrimidine dimers (CPDs), UVB-fingerprint DNA lesions, can be detected immunohistochemically in all epidermal layers of a full thickness reconstruct using a CPD antibody followed by fluorescence labeling. At 24 h after exposure, sunburn cells, apoptotic keratinocytes that are classic markers of the epidermal sunburn response, can be visualized by H&E staining of the fixed reconstruct. Moreover, other important sunburn markers including accumulation of p53 are easily detected in UV-exposed skin reconstructs. At physiologically relevant doses of deeply penetrating UVA, specific changes in the dermal layer can be detected with induction of cellular oxidative stress using the fluorigenic redox probe DCFH-DA immediately after irradiation. Moreover, UVA-induction of fibroblast apoptosis can be detected within 6 h after irradiation by TUNEL staining of the reconstruct. 48 h after UVA, loss of fibroblast cellularity in the upper layer of the dermis with no changes in the epidermal layer can be detected. Moreover, UV-induction of MMP-1 production has been studied in reconstructed skin using ELISA analysis. Under UVA exposure conditions, MMP-1 production was directly induced in the dermal fibroblasts of full thickness reconstructs and removal of the epidermal layer immediately after UVA exposure did not alter this induction suggesting that this dermal UVA effect does not require the pre-
150 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
sence of keratinocytes. In contrast, UVB-induced MMP-1 production in dermal fibroblasts required the presence of the epidermis and was abolished when the epidermis was removed during the post-irradiation period. Using comparative studies on monolayered cultured keratinocytes and fibroblasts versus reconstructed skin that combines these two cell types, paracrine mechanisms based on epidermal release of IL-1 and IL-6 were demonstrated to cause UVB-induced MMP-1 upregulation in dermal fibroblasts [66,67]. This finding nicely exemplifies the superior experimental versatility of skin reconstructs for mechanistic studies on the cellular crosstalk involved in the molecular pathways of skin photodamage.
CPD
sunburn cells
p53
normal human skin
reconstructed human skin
Fig. (4). Biomarkers of UV-induced tissue damage in normal and reconstructed human skin. Immediately after UVB exposure (50 mJ/cm2) DNA lesions are detectable by immunohistochemical visualization of cyclobutane pyrimidine dimers (CPD). 24 h after irradiation, sunburn cells (arrows) can be observed using H&E staining. Moreover, 24 h after UVB, p53 protein accumulation becomes apparent (arrows). Remarkably, both systems display a similar pattern of UV-damage biomarkers. (reproduced with permission from reference [64], Elsevier Ireland Ltd.).
Based on their structural and functional similarities with human skin and the close concordance of photodamage biomarkers in reconstructs and human skin, the use of organotypic skin reconstructs has become a valuable discovery tool for the identification and development of pharmaceutical and cosmeceutical agents that target skin photodamage, aging, and dyspigmentation as summarized in Fig. (5) and Table (1). 6.3.2. Suncreens and Sunblockers Sunscreens and sunblockers are important photoprotective topical agents, and combinatorial use of modern broad-spectrum suncreens with complementary agents that exert skin photoprotection by other synergistic mechanisms of action has emerged as an important chemopreventive strategy targeting skin photoaging and photocarcinogenesis [61]. The U.S. FDA regulates sunscreen products as over-the-counter drugs, and appro-
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 151
val and marketing of novel sunscreen agents is therefore a rare event. The use of skin reconstructs in photoprotection studies examining novel therapeutic agents is now firmly established and promises to accelerate the preclinical drug discovery and development process of this important class of skin protectants. A study on the protective effect of the
solar photons [UVB, UVA, visible, infrared]
O O
photoprotective agents
N H
O
O O O
HO
P HO OH
OH
HO
DNA ROS p53 IL-8 PGE2 MMP-1 damage CPD 8-oxodG
sunburn cells Langerhans cells fibroblast
depletion
sunburn & photoaging chlorpromazine
Fig. (5). Biomarkers of UV-induced tissue damage validated in human skin reconstructs and antagonized by photoprotective agents. For abbreviations see text. Photoprotective small molecule compounds (from top to bottom): octylmethoxycinnamate (UVB-sunscreen), L-prolinemethylester (quencher of photoexcited states), 2-phospho-L-ascorbate (antioxidant), MexorylTMSX (broad spectrum UV-sunscreen). Table 1.
Drug Discovery Targeting Skin Photodamage, Aging, and Dyspigmentation Using Reconstructed Human Skin Agent
Mechanism of Action
References
MexorylTM SX
Sunscreen
Bernerd et al., 2000; [11]
L-Proline-methylester
QPES
Wondrak et al., 2005; [72]
Aminoguanidine
Glycation inhibitor
Pageon et al., 2008; [75]
-Carotene
Antioxidant
Hakozaki et al. 2008; [76]
Aloesin
Skin whitener
Wang et al., 2008; [80]
152 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
classic UVB sunscreen 2-ethylhexyl-p-methoxycinnamate and the UVA sunscreen MexorylTM SX on reconstructed human skin exposed to UVB or UVA Irradiation was able to discriminate molecular parameters in skin indicative of UVB (sunburn cell formation and CPD formation in nuclear DNA) versus UVA (fibroblast apoptosis) photoprotection [11]. Equally, measurements of the protective effect of the topically applied sunscreens Eusolex 8020, a UVA-blocking dibenzoylmethane derivative, and Eusolex 6300, a UVB-blocking benzylidene camphor-derivative was achieved using skin equivalents by measurement of residual cellular viability 24 h postirradiation using the 3-(4,5dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) test and assessment of the inflammation response by IL determination [68]. The use of cellularly more complex reconstructed human skin for the evaluation of photodamage and sunscreen efficacy has been described recently. Based on reconstructed epidermis, comprising keratinocytes, melanocytes, and Langerhans cells, UVinduced pigmentation, cellular parameters related to UV-induced immunosuppression (morphological alterations and reduction in numbers of epidermal Langerhans cells) and photoprotection achieved by topical application of a broad spectrum sunscreen were studied [8]. Similarly, assessment of the skin photoprotective capacities of an organomineral broad-spectrum sunblock containing Tinosorb M, OCM, ZnO and TiO2 was performed in ex vivo hairless rat skin and reconstructed human skin demonstrating prevention of UVA-specific damage characterized by dermal expression of the fibroblast marker vimentin [69]. In another study, the combinatorial protective effects of sunscreens and antioxidants (-tocopherol and ascorbic acid) on reconstructed epidermis with or without melanocytes 24 h after UVB, UVA, or UVA+B irradiation were compared using various markers including sunburn cell formation, CPD formation, immunohistochemical detection of total epidermal protein oxidation, and catalase and superoxide dismutase activities [70]. Moreover, using full thickness reconstructed skin as an adequate tool for the discrimination between UVB- and UVA-induced skin photodamage, it has been demonstrated that the photoprotection afforded by two sunscreen formulations having similar SPF values yet different UVA-screening efficacies is not equal with regard to prevention of dermal damage related to photoaging [71]. Based on the involvement of epidermal Langerhans cells in the mediation of UVinduced immunosuppression, an important factor in skin photocarcinogenesis, a Langerhans cell containing reconstructed human epidermis has been used to validate the efficacy of sunscreens to prevent UV-induced immunosuppression. Exposure of the reconstructed epidermis to solar simulated radiation depleted Langerhans cell numbers within 24 h of irradiation and induced dramatic morphological changes with loss of dendricity in the surviving cells, all of which were prevented by topical application of a broad spectrum UV filter before UV-exposure [16]. 6.3.3. Other Photoprotective and Anti-Aging Molecules Quenchers of photoexcited states (QPES compounds). Photooxidative stress is a key mechanism in UVA-induced skin photodamage involving photoexcited states of endogenous UVA chromophores such as porphyrins, melanin precursors, and cross-linkfluorophores of skin collagen as reviewed in [49]. Skin photodamage by endogenous UV-photosensitizers occurs by direct reaction with substrate molecules (type I photosensitization) or molecular oxygen (type II), leading to formation of reactive oxygen species. The causative role of photoexcited states in skin photodamage suggests that direct molecular antagonism of photosensitization reactions using physical quenchers of pho-
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 153
toexcited states (QPES compounds) offers a novel chemopreventive opportunity for topical skin photoprotection [72]. QPES compounds antagonize the harmful excited state chemistry of endogenous sensitizers by physical quenching, facilitating the harmless return of the sensitizer excited state to the electronic ground state by energy dissipation. Pilot studies using reconstructed full thickness human skin exposed to solar simulated light have provided proof-of concept efficacy of QPES compounds for topical skin photoprotection [72]. QPES compounds suitable for further preclinical development were identified using a series of chemical and cell-based screening methods. First, hit compounds were identified in a broad primary screening assay based on QPES suppression of photosensitized plasmid DNA cleavage. A secondary screen then confirmed hit compounds as QPES agents based on the nonsacrificial quenching of dye-sensitized singlet oxygen (1O2) formation by electron paramagnetic resonance detection of 2,2,6,6tetramethyl-piperidine-1-oxyl, a stable free radical indicative of 1O2 formation. These initial screens identified a pyrrolidine pharmacophore with pronounced QPES activity, and L-proline and other noncytotoxic proline derivatives containing this pharmacophore were then screened for efficacy in skin cell based assays of sensitized photodamage. Final confirmation of promising QPES lead compounds for further preclinical development was obtained in photoprotection studies using a human full thickness skin reconstructs, comprising a dermal equivalent from rat tail tendon type I collagen and human skin fibroblasts with a multilayered epidermis formed from immortalized human HaCaT keratinocytes. Various QPES compounds play an increasingly important role as combinatorial agents used together with regular sunscreens in order to enhance photostability of existing sunscreen formulation in order to provide additional skin UVA photoprotection by suppression of photodamage that originates from photoexcited states in skin [49]. Glycation inhibitors. Recent research has demonstrated that during chronological and actinic aging skin proteins accumulate significant chemical damage with formation of chromophore-epitopes that can act as protein crosslinks and UV-photosensitizers [73,74]. Non-enzymatic amino-carbonyl reactions (glycation) between reactive carbonyl species (RCS) and protein-bound amino-groups induce the accumulation of crosslinks and other posttranslational epitopes called advanced glycation endproducts (AGEs) on long-lived skin proteins such as dermal collagen, elastin, laminin, and fibronectin [61,75]. AGEs induce marked changes in extracellular matrix architecture and function including extensive protein crosslinking thought to contribute to loss of dermal elasticity and wrinkle formation during chronological and actinic aging of the human skin. Due to the causative involvement of glycation reactions in skin aging and photodamage, considerable research efforts aim at developing experimental therapeutics that act as topical inhibitors of glycation reactions capable of delaying or even reversing the detrimental consequences of skin extracellular matrix damage [61,73-75]. Recently, reconstructed full thickness human skin modified by glycation of the dermal equivalent was used in order to study glycation damage during skin aging [75]. Pre-glycation of the collagen used for dermal reconstruction induced significant structural changes throughout the full thickness reconstruct with increased levels of collagen IV and laminin in the basement membrane zone and expansion of 1-integrin to suprabasal layers of the epidermis, alterations that can also be observed in chronologically aged human skin and during wound healing. Using this chronologically-aged skin reconstruct model as a powerful screening tool, the potential of small molecule glycation inhibitors to suppress
154 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
glycation damage and glycation-induced dermal and epidermal alterations was demonstrated. Antioxidants. The causative involvement of reactive oxygen species (ROS) in UVinduced skin damage including inflammation, photoaging, and ultimately photocarcinogenesis is firmly established. Topical application of small molecule antioxidants is therefore a promising chemopreventive approach targeting skin photodamage as discussed in [49]. Improvement of dermal epidermal junction morphogenesis as well as enhanced synthesis of extracellular matrix components and basement membrane proteins by 2phospho-L-ascorbate supplementation has been demonstrated in a full thickness reconstruct [64]. Recently, a commercial human epidermal reconstruct was used to monitor UVB-induced formation of ROS in real time and to screen small molecule topical antioxidants for skin photoprotection [76]. Based on chemiluminescent visualization of spatial ROS distribution by topical application of a cypridina luciferin analog as a chemiluminescent probe using a high-performance low-light imaging luminograph system with a CCD camera the antioxidant and photoprotective effects of test agents such as ascorbate, -carotene, superoxide dismutase (SOD) were evaluated. As molecular endpoints, quantification of emitted chemiluminescence (CL) intensities, MTT assay and oxidative DNA damage using DNA 8-hydroxy-2-deoxyguanosine (8-oxodG) staining were performed. Thus, this model shows promise not only for visualizing the production of UVB-induced ROS in real-time but also for screening of topical anti-oxidants for photoprotective and anti-photo-oxidative efficacy. Similarly, reconstructed skin has been used successfully to demonstrate photoprotective and anti-photoaging effects of an antioxidant ethanolic extract of rosmarinus officinalis [77]. Importantly, a genetically modified human skin reconstruct has recently been generated in order to validate catalase as a key antioxidant enzyme in skin involved in the endogenous skin photoprotective and antioxidant network, providing an impressive example of using human skin reconstructs for drug target validation studies [78]. Epidermis was reconstructed with normal human keratinocytes overexpressing sustainably lentivirus-mediated catalase (CAT), copper/zinc superoxide dismutase (CuZnSOD) or manganese superoxide dismutase (MnSOD) enzymes. Upon exposure to UVB irradiation a significant decrease in sunburn cell formation, caspase-3 activation and p53 accumulation was observed in human reconstructed epidermis overexpressing CAT. Upon UVA exposure, UVA-induced hypertrophy and DNA oxidation (8-oxodG) were decreased by CAT overexpression. Remarkably, these effects were not achieved by overexpression of CuZnSOD or MnSOD. Thus, based on these studies using genetically engineered human skin reconstructs, vector-mediated CAT overexpression or therapeutic administration of small molecule catalase mimetics could be a novel photoprotective strategy in monogenic/polygenic photosensitive disorders characterized by ROS accumulation. Natural product anti-inflammatory agents. Due to the involvement of inflammatory signaling in the response of human skin to UV exposure, anti-inflammatory agents are promising ingredients of photoprotective topical formulations. Recently, the use of reconstructed human epidermis for screening of fragrance raw materials and essential oils with potential anti-inflammatory activity has been reported [79]. An initial screening with pig blood platelets was conducted on 900 perfumery raw materials, and active compounds which successfully reduced formation of PGE2, a key marker of keratinocyte irritancy and crucial mediator of inflammatory events in human skin, were then validated
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 155
in a human keratinocyte cell line. Application of anti-inflammatory fragrances in a hydrogel allowed reduction of UVB-induced PGE2 formation in reconstructed epidermis. Moreover, antiperspirant-induced formation of PGE2 was equally reduced to background levels in this epidermal screening model if these novel fragrances were incorporated into the antiperspirant formulation. It was concluded that the active ingredients identified by screening in reconstructed human skin may be part of a cosmetic formulation designed for optimal product mildness. Skin pigmentation modulators. Pharmacological modulation of skin pigmentation is an important area of current skin drug discovery. Epidermal reconstructs containing melanocytes that closely mimic the melanogenic potential of human skin have been developed and are used as powerful drug discovery tools in academic and pharmaceutical research circumventing the problems associated with species extrapolation, the use of laboratory animals, and time consuming clinical trials. The MelanoDermTM Skin Model (MatTek) is based on co-culture of human keratinocytes and melanocytes, which are alternatively comprised of Black, Caucasian, or Asian cells in order to accurately reflect the ethnic complexity of human skin pigmentation. The co-cultures, produced using serum free medium without artificial stimulators of melanogenesis, undergo spontaneous melanogenesis leading to tissues of varying levels of pigmentation. Topical application of test compounds acting as skin lighteners or self-tanning agents is possible, since the cultures are grown on cell culture inserts at the air-liquid interface. Screening of pharmacological modulators of skin pigmentation is then based on analytical methods to evaluate melanocyte dendricity and viability, pigment granule transfer to adjacent keratinocytes, bulk darkening of tissue, and total melanin content and synthesis rates. Topical application of inhibitors of melanogenesis will reduce melanin production and macroscopic darkening of the reconstructs, whereas stimulants of melanogenesis including melanocyte stimulating hormone will increase melanin content and macroscopic darkening over untreated controls. Numerous studies using melanocyte containing epidermal reconstructs for the identification of pigmentation altering agents have been published [27]. The skin lightening effects of aloesin and arbutin were tested in a human pigmented skin model demonstrating dose dependent reduction of tyrosinase activity and melanin content [80]. N-acetyl glucosamine (NAG) has previously been shown to reduce skin hyperpigmentation based on glucosamine inhibition of enzymatic glycosylation and conversion of inactive human pro-tyrosinase to the active tyrosinase. To identify additional mechanisms by which NAG might affect melanin production, an in vitro genomics experiment was conducted in SkinEthicTM skin equivalent cultures, which were topically dosed with NAG vs. a vehicle control [81]. Relative to vehicle, NAG reduced melanin production, and the expression of several pigmentation-relevant genes was significantly altered (downregulated or up-regulated) by NAG treatment . The effects of all-trans retinoic acid (RA) on melanogenesis and the mechanism of its action in topical treatment have recently been studied in the pigmented skin equivalent as well as in monolayer culture of melanocytes [82]. Interestingly, suppression of melanogenesis by RA was not observed in pigmented skin equivalents and monolayer culture of murine and human melanocytes, whereas the control substance hydroquinone showed strong inhibition of melanogenesis. The results suggested that the role of RA in combinatorial bleaching treatments appears to be based on promotion of keratinocytes proliferation and acceleration of epidermal turnover.
156 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
7. RECONSTRUCTED SKIN AS A TARGET DISCOVERY TOOL: RECONSTRUCTION OF DISEASED SKIN The commercial availability of advanced tissue reconstructs that adequately represent the complexity of functional human skin has significantly impacted the regulatory, cosmeceutical, and pharmaceutical domains of the drug discovery process. In addition to accelerating the rate at which screening of novel molecules as better consumer ingredients, cosmeceuticals, and drugs occurs, advanced skin reconstructs are currently revolutionizing the way that novel skin drug targets are identified and validated. Successful reconstruction of specialized skin models that adequately represent many aspects of the complex biology underlying important skin pathologies with unmet therapeutic needs including cancer (nonmelanoma and melanoma), immune dysregulation (psoriasis), and microbial infection (acne) has now opened the exciting opportunity to identify and validate novel molecular targets for drug discovery. 7.1. Skin Cancer As a striking example, reconstruction of human melanoma skin that represents the progressional stages of this devastating disease has now been achieved. Melanoma is a highly invasive and metastatic tumor that originates from skin melanocytes [83,84]. Alarming increases in incidence rates combined with the notorious resistance to chemotherapeutic intervention and the lack of adequate animal models that represent the human biology of this complex disease create an urgent need for better disease models that would allow rapid identification of promising preclinical candidates from compound libraries. Indeed, melanoma reconstructs that successfully reproduce early disease progression and invasion have been generated in the in commercial and academic environment [12,15]. These models would provide a tool for the identification of novel antimelanoma agents that discriminate between tumor cells and healthy tissue. Progression of cutaneous melanoma via radial growth phase (RGP) confined to the epidermis and vertical growth phase (VGP) with penetration of the basement membrane and invasion of the dermis can now be mimicked and pharmacologically manipulated using a full thickness melanoma skin model (MLNM-FT-A375, MatTek; http://www.mattek.com) as shown in Fig. (6). The MLNM-FT-A375 model consists of human malignant melanoma cells (A375), normal, human epidermal keratinocytes (with organized basal, spinous, granular, and cornified epidermal layers) and dermal fibroblasts, forming a welldifferentiated full thickness skin model with RGP, VGP, and metastatic progressional stages of melanoma dependent on the duration of tissue culture after reconstruction. The model closely parallels the progression of melanoma in vivo, thus providing a valuable tool to study the cell biology of melanoma and to screen and develop preventative and therapeutic treatments for the most serious cutaneous malignancy. It has already been stated that reconstituted three-dimensional human skin serves as a valid in vitro model for mechanistic studies on skin carcinogenesis induced by chemical carcinogens and UV exposure [26]. Recently, epidermal photocarcinogenesis has been carefully studied by reconstruction of DNA repair-deficient xeroderma pigmentosum skin in vitro that also serves as a model a model of skin hypersensitivity to UV light [85,86]. In xeroderma pigmentosum (XP), a genetic defect in nucleotide excision repair of ultraviolet (UV)-induced mutagenic lesions causes extreme photosensitivity and cancer proneness. Nucleotide excision repair is an enzyme-based mechanism of DNA repair that removes cyclobutane pyrimidine dimers and the pyrimidine-pyrimidone (6–4) pho-
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 157
toproducts, both signature DNA lesions caused in skin exposed to erythemogenic doses of solar ultraviolet light. After isolation and culture of several strains of XP-C keratinocytes and fibroblasts from human patients, a three-dimensional skin model in vitro comprising both XP epidermis and a XP dermal equivalent was generated. The XP model displayed repair deficiency characterized by long-lasting persistence of UV-induced DNA damage and p53 positive nuclei after UV-B exposure, closely resembling the pathological UV repair response observed in human XP skin. Genetic reintroduction of a functional XPC gene into keratinocytes and fibroblasts before reconstruction reversed various cellular consequences of UV-hypersensitivity. Reconstruction of XP skin in vitro may therefore serve as a superb model for the development of novel therapeutic strategies including gene therapy and may allow the identification of novel pharmacological agents that efficiently prevent or treat skin UV-hypersensitivity. In this context, topical delivery of DNA repair enzymes has shown great promise [87].
Fig. (6). Commercial melanoma tissue reconstruct (MLNM-FT-A375) representing progressional stages of the disease. Human Metastatic Melanoma Cells (A375) in full thickness melanoma skin model. Metastatic A375 cells develop radial growth phase (RGP) melanoma nodes at dermal/epidermal junction (day 11). With extended culture time, melanoma nodes adopt a vertical growth phase (VGP) morphology (day 18); subsequently, isolated clusters of cells invade the dermis (metastatic invasion) (day 29). Long arrows: melanoma cell clusters at the epidermal-dermal junction; short arrows: separated melanoma cell clusters infiltrating the dermis. [formalin fixed, paraffin embedded, H&E stained reconstructs; reproduced with permission, MatTek, Ashland, MA, USA; http://www.mattek.com).
7.2. Psoriasis Reconstruction of psoriatic skin, recently achieved using keratinocytes and fibroblasts derived from psoriatic patients, represents a valuable tool for drug discovery targeting this disease [10]. Psoriasis is a chronic inflammatory disease affecting the skin and joints that is associated with increased antigen presentation, T-helper (Th)-1 cytokine
158 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Bause et al.
production, and T cell activity. Consistent with inflammatory dysregulation of psoriatic skin, the proinflammatory genes TNF-, IFN-, and IL-8 were expressed at high levels in the psoriatic models, whereas minimal expression was detected in reconstructs built from normal skin keratinocytes and fibroblasts. Moreover, the chemokine receptor CXCR2 was overexpressed in the psoriatic model and localized to the granular layer of keratinocytes, an expression pattern closely resembling psoriatic skin in vivo. Remarkably, models derived from uninvolved psoriatic skin showed the same gene expression profile as those derived from involved skin along with an increased proliferation rate when compared to normal models suggesting an inherent predisposition of psoriatic individuals to develop the disease phenotype even in the absence of T cells. 7.3. Microbial Skin Infection Human skin equivalents have shown promise as discovery tools to study cutaneous microbial interactions, potentially useful for the identification of novel antimicrobial therapeutics. Ethical considerations associated with the intentional inoculation of pathogenic microorganisms onto human skin and biological insufficiencies of animal models create an urgent need for skin reconstructs with microbial inoculation, representing relevant pathogen-induced skin diseases including inflammatory acne, seborrhoeic dermatitis, and wound infection. Earlier studies have demonstrated feasibility of microbial inoculation of skin reconstructs for determining the adhesiveness of staphylococcus aureus and staphylococcus epidermidis to reconstructed epidermis [88]. Moreover, inoculation of epidermal reconstructs with the human skin commensals staphylococcus epidermidis, propionibacterium acnes and malassezia furfur as well as the transient pathogen staphylococcus aureus has been reported [89]. A novel, real-time growth monitoring method was also developed, using S. aureus containing a lux cassette with light output as analytical parameter representing bacterial colonization of the reconstruct. Successful colonization of reconstructed epidermis with constitutive and pathological microbial flora can therefore serve as a model to investigate interactions between resident and transient microbial communities on skin underlying the innate responses that protect skin from colonization by opportunistic pathogens. Microbial inoculation of reconstructed epidermis will serve as a unique discovery tool for screening of novel antimicrobial chemotherapeutics targeting infection of human skin by defined pathogens. The usefulness of the reconstruct-based skin disease models described above suggests that other pathologies that involve impaired skin barrier structure and function (atopic dermatitis and ichthyosis) [90], redox and inflammatory dysregulation (vitiligo), and early stages of carcinogenesis (actinic keratosis, dysplastic nevus) can be modeled adequately in vitro. These novel reconstructs would then offer exciting possibilities for future drug discovery targeting important pathologies with unmet therapeutic needs. Continuous progress in skin tissue engineering will be achieved based on gene-modified cellular components of increasing complexity and incorporation of stem cells and progenitor cell-derived populations [7,91,92]. This will ensure the expanding role of skin equivalents in disease model-based pharmaceutical assays for rapid identification and development of novel cutaneous therapeutics. ACKNOWLEDGEMENTS Supported in part by grants from the National Institutes of Health [R01CA122484, ES007091, Arizona Cancer Center Support Grant CA023074], and from the Arizona
More Than Skin Deep
Frontiers in Drug Design & Discovery, 2009, Vol. 4 159
Biomedical Research Commission (ABRC 0721). Support from the Arizona Science Foundation (SL) and German Academic Exchange Service (AB) is greatly acknowledged. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]
[13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35]
Kubinyi, H. Nat. Rev. Drug Discov., 2003, 2, 665-8. Schmid, E. F.; Smith, D. A. Drug Discov. Today, 2005, 10, 1031-9. Kamb, A.; Wee, S.; Lengauer, C. Nat. Rev. Drug Discov., 2007, 6, 115-20. Bender, A.; Bojanic, D.; Davies, J. W.; Crisman, T. J.; Mikhailov, D.; Scheiber, J.; Jenkins, J. L.; Deng, Z.; Hill, W. A.; Popov, M.; Jacoby, E.; Glick, M. Curr. Opin. Drug Discov. Devel., 2008, 11, 327-37. Mehul, B.; Asselineau, D.; Bernard, D.; Leclaire, J.; Regnier, M.; Schmidt, R.; Bernerd, F. Arch. Dermatol. Res., 2004, 296, 145-56. Ikada, Y. J. R. Soc. Interface, 2006, 3, 589-601. Garcia, M.; Escamez, M. J.; Carretero, M.; Mirones, I.; Martinez-Santamaria, L.; Navarro, M.; Jorcano, J. L.; Meana, A.; Del Rio, M.; Larcher, F. Mol. Carcinog., 2007, 46, 741-5. Duval, C.; Schmidt, R.; Regnier, M.; Facy, V.; Asselineau, D.; Bernerd, F. Exp. Dermatol., 2003, 12 Suppl 2, 64-70. Bechetoille, N.; Dezutter-Dambuyant, C.; Damour, O.; Andre, V.; Orly, I.; Perrier, E. Tissue Eng., 2007, 13, 2667-79. Barker, C. L.; McHale, M. T.; Gillies, A. K.; Waller, J.; Pearce, D. M.; Osborne, J.; Hutchinson, P. E.; Smith, G. M.; Pringle, J. H. J. Invest. Dermatol., 2004, 123, 892-901. Bernerd, F.; Vioux, C.; Asselineau, D. Photochem. Photobiol., 2000, 71, 314-20. Meier, F.; Nesbit, M.; Hsu, M. Y.; Martin, B.; Van Belle, P.; Elder, D. E.; Schaumburg-Lever, G.; Garbe, C.; Walz, T. M.; Donatien, P.; Crombleholme, T. M.; Herlyn, M. Am. J. Pathol., 2000, 156, 193-200. Fisher, G. J. Cutis, 2005, 75, 5-8; discussion 8-9. Elias, P. M.; Choi, E. H. Exp. Dermatol., 2005, 14, 719-26. Berking, C.; Herlyn, M. Histol. Histopathol., 2001, 16, 669-74. Facy, V.; Flouret, V.; Regnier, M.; Schmidt, R. Toxicol. In Vitro, 2005, 19, 787-95. Sorrell, J. M.; Caplan, A. I. J. Cell Sci., 2004, 117, 667-75. Marionnet, C.; Pierrard, C.; Vioux-Chagnoleau, C.; Sok, J.; Asselineau, D.; Bernerd, F. J. Invest. Dermatol., 2006, 126, 971-9. Rheinwald, J. G.; Green, H. Cell, 1975, 6, 331-43. Boyce, S. T.; Ham, R. G. J. Invest. Dermatol., 1983, 81, 33s-40s. Pittelkow, M. R.; Shipley, G. D. J. Cell Physiol., 1989, 140, 565-76. Caux, C.; Dezutter-Dambuyant, C.; Schmitt, D.; Banchereau, J. Nature, 1992, 360, 258-61. Prunieras, M.; Regnier, M.; Woodley, D. J. Invest. Dermatol., 1983, 81, 28s-33s. Boyce, S. T.; Goretsky, M. J.; Greenhalgh, D. G.; Kagan, R. J.; Rieman, M. T.; Warden, G. D. Ann. Surg., 1995, 222, 743-52. MacNeil, S. Nature, 2007, 445, 874-80. Zhao, J. F.; Zhang, Y. J.; Kubilus, J.; Jin, X. H.; Santella, R. M.; Athar, M.; Wang, Z. Y.; Bickers, D. R. Biochem. Biophys. Res. Commun., 1999, 254, 49-53. Cario-Andre, M.; Bessou, S.; Gontier, E.; Maresca, V.; Picardo, M.; Taieb, A. Cell. Mol. Biol. (Noisyle-grand), 1999, 45, 931-42. Yoshida, Y.; Hachiya, A.; Sriwiriyanont, P.; Ohuchi, A.; Kitahara, T.; Takema, Y.; Visscher, M. O.; Boissy, R. E. FASEB J., 2007, 21, 2829-39. Welss, T.; Basketter, D. A.; Schroder, K. R. Toxicol. In Vitro, 2004, 18, 231-43. Faller, C.; Bracher, M.; Dami, N.; Roguet, R. Toxicol. In Vitro, 2002, 16, 557-72. Netzlaff, F.; Lehr, C. M.; Wertz, P. W.; Schaefer, U. F. Eur. J. Pharm. Biopharm., 2005, 60, 167-78. Eppler, A. R.; Kraeling, M. E.; Wickett, R. R.; Bronaugh, R. L. Food Chem. Toxicol., 2007, 45, 210917. El Ghalbzouri, A.; Siamari, R.; Willemze, R.; Ponec, M. Toxicol. In Vitro, 2008, 22, 1311-20. Hoffmann, S.; Cole, T.; Hartung, T. Regulatory Toxicol. Pharm., 2005, 41, 159-66. OECD Guidelines for the Testing of Chemicals, Test No. 431, E-book, OECD Code: 979943101E1, OECD Publishing, 2004, Paris, France.
160 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53]
[54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71]
Bause et al.
Kandarova, H.; Liebsch, M.; Gerner, I.; Schmidt, E.; Genschow, E.; Traue, D.; Spielmann, H. Altern. Lab. Anim., 2005, 33, 351-67. Cotovio, J.; Grandidier, M. H.; Portes, P.; Roguet, R.; Rubinstenn, G. Altern. Lab. Anim., 2005, 33, 329-49. Robinson, M. K.; Cohen, C.; de Fraissinette Ade, B.; Ponec, M.; Whittle, E.; Fentem, J. H. Food Chem. Toxicol., 2002, 40, 573-92. Tupker, R. A.; Coenraads, P. J.; Fidler, V.; De Jong, M. C.; Van der Meer, J. B.; De Monchy, J. G. Br. J. Dermatol., 1995, 133, 358-64. Corsini, E.; Galli, C. L. Toxicol. Lett., 1998, 102-103, 277-82. Faller, C.; Bracher, M. Skin Pharmacol. Appl. Skin Physiol., 2002, 15 Suppl 1, 74-91. Rogers, J. V.; Gunasekar, P. G.; Garrett, C. M.; McDougal, J. N. J. Biochem. Mol. Toxicol., 2001, 15, 228-30. Ponec, M.; Kempenaar, J. Skin Pharmacol., 1995, 8, 49-59. Coquette, A.; Berna, N.; Vandenbosch, A.; Rosdy, M.; De Wever, B.; Poumay, Y. Toxicol. In Vitro, 2003, 17, 311-21. Boxman, I. L.; Hensbergen, P. J.; Van Der Schors, R. C.; Bruynzeel, D. P.; Tensen, C. P.; Ponec, M. Br. J. Dermatol., 2002, 146, 777-85. de Brugerolle de, F.; Picarles, V.; Chibout, S.; Kolopp, M.; Medina, J.; Burtin, P.; Ebelin, M. E.; Osborne, S.; Mayer, F. K.; Spake, A.; Rosdy, M.; De Wever, B.; Ettlin, R. A.; Cordier, A. Cell Biol. Toxicol., 1999, 15, 121-35. Neppelberg, E.; Costea, D. E.; Vintermyr, O. K.; Johannessen, A. C. Exp. Dermatol., 2007, 16, 574-9. Facy, V.; Flouret, V.; Regnier, M.; Schmidt, R. J. Invest. Dermatol., 2004, 122, 552-3. Wondrak, G. T.; Jacobson, M. K.; Jacobson, E. L. Photochem. Photobiol. Sci., 2006, 5, 215-37. Bernard, F. X.; Barrault, C.; Deguercy, A.; De Wever, B.; Rosdy, M. Cell Biol. Toxicol., 2000, 16, 391-400. Portes, P.; Pygmalion, M. J.; Popovic, E.; Cottin, M.; Mariani, M. Photodermatol. Photoimmunol. Photomed., 2002, 18, 96-102. Jones, P. A.; King, A. V.; Earl, L. K.; Lawrence, R. S. Toxicol. In Vitro, 2003, 17, 471-80. Schafer-Korting, M.; Bock, U.; Diembeck, W.; Dusing, H. J.; Gamer, A.; Haltner-Ukomadu, E.; Hoffmann, C.; Kaca, M.; Kamp, H.; Kersen, S.; Kietzmann, M.; Korting, H. C.; Krachter, H. U.; Lehr, C. M.; Liebsch, M.; Mehling, A.; Muller-Goymann, C.; Netzlaff, F.; Niedorf, F.; Rubbelke, M. K.; Schafer, U.; Schmidt, E.; Schreiber, S.; Spielmann, H.; Vuia, A.; Weimer, M. Altern. Lab. Anim., 2008, 36, 161-87. Schreiber, S.; Mahmoud, A.; Vuia, A.; Rubbelke, M. K.; Schmidt, E.; Schaller, M.; Kandarova, H.; Haberland, A.; Schafer, U. F.; Bock, U.; Korting, H. C.; Liebsch, M.; Schafer-Korting, M. Toxicol. In Vitro, 2005, 19, 813-22. Netzlaff, F.; Kaca, M.; Bock, U.; Haltner-Ukomadu, E.; Meiers, P.; Lehr, C. M.; Schaefer, U. F. Eur. J. Pharm. Biopharm., 2007, 66, 127-34. Asbill, C.; Kim, N.; El-Kattan, A.; Creek, K.; Wertz, P.; Michniak, B. Pharm. Res., 2000, 17, 1092-7. Schmook, F. P.; Meingassner, J. G.; Billich, A. Int. J. Pharm., 2001, 215, 51-6. Zoller, N. N.; Kippenberger, S.; Thaci, D.; Mewes, K.; Spiegel, M.; Sattler, A.; Schultz, M.; BereiterHahn, J.; Kaufmann, R.; Bernd, A. Toxicol. In Vitro, 2008, 22, 747-59. Bernard, F. X.; Pedretti, N.; Rosdy, M.; Deguercy, A. Exp. Dermatol., 2002, 11, 59-74. Hsia, E.; Johnston, M. J.; Houlden, R. J.; Chern, W. H.; Hofland, H. E. J. Invest. Dermatol., 2008, 128, 125-30. Wondrak, G. T. Curr. Opin. Investig. Drugs, 2007, 8, 390-400. Bowden, G. T. Nat. Rev. Cancer, 2004, 4, 23-35. Bachelor, M. A.; Bowden, G. T. Semin. Cancer Biol., 2004, 14, 131-8. Vioux-Chagnoleau, C.; Lejeune, F.; Sok, J.; Pierrard, C.; Marionnet, C.; Bernerd, F. J. Dermatol. Sci. 2006, 2, S1-S12. Bernerd, F.; Asselineau, D. J. Am. Acad. Dermatol., 2008, 58, S155-9. Fagot, D.; Asselineau, D.; Bernerd, F. Arch. Dermatol. Res., 2002, 293, 576-83. Fagot, D.; Asselineau, D.; Bernerd, F. Photochem. Photobiol., 2004, 79, 499-505. Augustin, C.; Collombel, C.; Damour, O. Photochem. Photobiol., 1997, 66, 853-9. Gelis, C.; Girard, S.; Mavon, A.; Delverdier, M.; Paillous, N.; Vicendo, P. Photodermatol. Photoimmunol. Photomed., 2003, 19, 242-53. Cario-Andre, M.; Briganti, S.; Picardo, M.; Nikaido, O.; Gall, Y.; Ginestar, J.; Taieb, A. J. Photochem. Photobiol., B 2002, 68, 79-87. Bernerd, F.; Vioux, C.; Lejeune, F.; Asselineau, D. Eur. J. Dermatol., 2003, 13, 242-9.
More Than Skin Deep [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 161
Wondrak, G. T.; Jacobson, M. K.; Jacobson, E. L. J. Pharmacol. Exp. Ther., 2005, 312, 482-91. Wondrak, G. T.; Cervantes-Laurean, D.; Roberts, M. J.; Qasem, J. G.; Kim, M.; Jacobson, E. L.; Jacobson, M. K. Biochem. Pharmacol., 2002, 63, 361-373. Pageon, H.; Asselineau, D. Ann. N. Y. Acad. Sci., 2005, 1043, 529-32. Pageon, H.; Techer, M. P.; Asselineau, D. Exp. Gerontol., 2008, 43, 584-588. Hakozaki, T.; Date, A.; Yoshii, T.; Toyokuni, S.; Yasui, H.; Sakurai, H. Arch. Dermatol. Res., 2008, 300, 51-56. Martin, R.; Pierrard, C.; Lejeune, F.; Hilaire, P.; Breton, L.; Bernerd, F. Eur. J. Dermatol.,2008, 18, 128-35. Rezvani, H. R.; Cario-Andre, M.; Pain, C.; Ged, C.; deVerneuil, H.; Taieb, A. Cancer Gene Ther., 2007, 14, 174-86. Natsch, A.; Wasescha, M. Int. J. Cosmet. Sci., 2007, 29, 369-76. Wang, Z.; Li, X.; Yang, Z.; He, X.; Tu, J.; Zhang, T. Int. J. Cosmet. Sci., 2008, 30, 121-30. Bissett, D. L.; Farmer, T.; McPhail, S.; Reichling, T.; Tiesman, J. P.; Juhlin, K. D.; Hurley, G. J.; Robinson, M. K. J. Cosmet. Dermatol., 2007, 6, 232-8. Yoshimura, K.; Tsukamoto, K.; Okazaki, M.; Virador, V. M.; Lei, T. C.; Suzuki, Y.; Uchida, G.; Kitano, Y.; Harii, K. J. Dermatol. Sci., 2001, 27 Suppl 1, S68-75. Wondrak, G. T.; Jacobson, M. K.; Jacobson, E. L. Curr. Cancer Ther. Rev., 2005, 1, 271-276. Wondrak, G. T.; Jacobson, M. K.; Jacobson, E. L. J. Pharmacol. Exp. Ther., 2006, 316, 805-814. Bernerd, F.; Asselineau, D.; Vioux, C.; Chevallier-Lagente, O.; Bouadjar, B.; Sarasin, A.; Magnaldo, T. Proc. Natl. Acad. Sci. USA, 2001, 98, 7817-22. Bernerd, F.; Asselineau, D.; Frechet, M.; Sarasin, A.; Magnaldo, T. Photochem. Photobiol., 2005, 81, 19-24. Yarosh, D.; Klein, J.; O'Connor, A.; Hawk, J.; Rafal, E.; Wolf, P. Lancet 2001, 357, 926-9. Lerebour, G.; Cupferman, S.; Bellon-Fontaine, M. N. J. Appl. Microbiol., 2004, 97, 7-16. Holland, D. B.; Bojar, R. A.; Jeremy, A. H.; Ingham, E.; Holland, K. T. FEMS Microbiol. Lett., 2008, 279, 110-5. Simonsen, L.; Fullerton, A. Skin Pharmacol. Physiol., 2007, 20, 230-6. Shepherd, B. R.; Enis, D. R.; Wang, F.; Suarez, Y.; Pober, J. S.; Schechner, J. S. FASEB J., 2006, 20, 1739-41. Andreadis, S. T. Adv. Biochem. Eng. Biotechnol., 2007, 103, 241-74.
162
Frontiers in Drug Design & Discovery, 2009, 4, 162-188
Across Skin Barrier: Known Methods, New Performances Krzysztof Cal* Department of Pharmaceutical Technology, Medical University of Gdansk, Hallera 107, 80-416 Gdansk, Poland Abstract: Skin is still the desirable route for the delivery of drug substances into the human body. Transdermal drug delivery offers many advantages over the conventional oral route of application, such as the elimination of hepatic first-pass effect, reduced side effects, constant concentration of a drug in the blood. Human intact skin is normally permeable for molecules with log P in the range of 1-3, smaller than 500 Da and present in a unionized form. Usually, obtained fluxes of drug substances are too low for the induction of systemic therapeutic effects. It is caused by the specific structure and composition of the outer layer of the skin – the stratum corneum. Closely packed, built as “brick” (corneocytes) and “mortar” (lipid bilayers), the stratum corneum is the most important limiter for transdermal drug delivery. Molecules that permeated the stratum corneum are easily taken up by capillary vessels present in the deeper skin layers. This chapter presents different methods used for skin permeation enhancement. Various drug forms and carriers, chemical permeation enhancers, electrically supported methods and devices, and the stratum corneum bypassing or removing methods are described, and the recent achievements in the field and possible practical use in market products are discussed. The special subchapters are dedicated to the skin disposition of one of the most often used penetration enhancers – terpenes, and the use of cyclodextrins in formulations applied onto the skin.
INTRODUCTION The application of exogenous substances onto the skin has been known for many thousands of years. The ancient Egyptians and Greeks applied olive oil, mixture of water and lead oxide in olive oil (astringent agent and occlusive barrier), and essential oils (penetration enhancers) onto the skin [1,2]. In the 16th century Fuchs [3] recommended soaking feet in water with Hyoscyamus niger for insomnia; and “witches” used the henbane containing liniments to obtain a “fly-like” feeling [3,4]. Over the years, the skin was regarded as an impermeable barrier [5]. The first conscious transdermal application can be dated in 1893, when the efficiency of salicylic acid in acute rheumatoid arthritis was demonstrated [1,6]. In the mid 20th century, through the tape-stripping technique
*Corresponding Author: Tel: +48 58 349 3183; Fax: +48 58 349 3190; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 163
and water permeability assessments, it was proven that the stratum corneum of epidermis is the major barrier in the skin permeation process [7,8]. It was confirmed by in vitro skin permeation studies with a wide range of drug substances [5,9,10]. The first, introduced on the market in 1954, transdermal preparation for the treatment of systemic disease was nitroglycerin in the form of an ointment [1,6]. Only thirty years later, FDA approved the first transdermal therapeutic system containing scopolamine [11]. Now, passive patches containing buprenorphine, clonidine, estradiol, ethinylestradiol + norelgestromin, fentanyl, flurbiprofen, indometacin, ketoprofen, levonorgestrel, lidocaine, methylphenidate, nicotine, norethisterone, oxybutynin, rivastigmine, rotigotine, selegiline and testosterone, and transdermal patches supported by iontophoresis are available in different countries [11-13]. This chapter describes the structure and composition of the skin, targets of skin applications of drugs, possible permeation pathways and different methods for skin permeation enhancement. The special subchapter is dedicated to the use of cyclodextrins in formulations applied onto the skin, because the mechanism of their action is complex and can consists of influences on drug molecules, vehicles and skin. THE BARRIER OF HEALTHY SKIN Skin is the easiest available and the largest human organ. It separates the organism from the external environment. The skin area in adults is 1.7-2 m2. In general, skin consists of epidermis with the stratum corneum and dermis. The dermis is about 0.5-2 mm thick. It is built from collagen and elastic fibers, and contains blood vessels, free nerve endings, and skin appendages (hair follicles, sweat and sebaceous glands) (Fig. 1). The main cells in the dermis are fibroblasts. Subcutaneous tissue (hypodermis) lies beneath the dermis and consists of fat lobules and muscles. The dermis is connected with the epidermis via the stratum basale. The epidermis consists of several layers of the skin cells – keratinocytes, reflecting different stages of differentiation. From the stratum basale the cells move up and undergo the keratinization process, and the cells die. The process of keratinocytes’ movement from the stratum basale to the surface of the skin, where cells are sloughed off, lasts from 2 to 4 weeks. In turn, the layer of dead cells corneocytes - is formed, the stratum corneum, that consists of 15-20 corneocytes layers. Corneocytes are composed mainly of keratin and natural moisturizing factor, and may contain from 10 to 80% of water [14-17]. Corneocytes are polygonal, elongated and flat (0.2-1.5 m thick and 34.0-46.0 m in diameter) [18]. Each corneocyte is encapsulated in an insoluble tough protein shell having a thickness of 10 nm, corneocyte envelope, which is covalently bound to an outer lipid envelope composed of a layer of ceramides [19]. Such enveloped corneocytes are embedded in a lipid matrix. This “brick” (corneocytes) and “mortar” (lipids) model of the stratum corneum was first presented by Michaels et al. [20]. The stratum corneum, according to the degree of hydration, is 10 to 50 m thick. The intercellular lipids of the stratum corneum have a unique composition compared to other epithelial lipid layers and consist of ceramides (40-50%), cholesterol (25%) and free fatty acids (10-20%). No phospholipids are present in the healthy stratum corneum [21,22]. Lipids are arranged in organized multilaminar structures, which basic unit is a bilayer. Hydrophilic polar groups (heads) of lipids are directed to each other and a monomolecular aqueous layer appears between them. Such a composition and structure
164 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
of the stratum corneum’ lipids is considered to be the most important for skin penetration and the permeation processes [23,24]. Drug form Dissolution Diffusion
Transepidermal pathway
Intracellular route
Shunt routes
Intercellular route Lipidic region
Corneocyte
Stratum corneum
Epidermis
Dermis
Capillary blood vessels
Sweat duct
Hair follicle
Fig. (1). Skin structure and possible permeation pathways.
Blood vessels, coming across the dermis and reaching up to the stratum basale, supply the skin with oxygen and nutrients, and remove toxins and metabolites. On the other hand, these vessels deliver substances that permeate the stratum corneum, directly to the systemic blood circulation [17]. TARGETS FOR THE SKIN’S APPLICATION OF DRUGS For many years, the skin was a target of the application of only dermatological drug products that were used in the local treatment of skin diseases. Currently, drug products applied onto the skin to treat muscles and joints inflammations or which purpose is to induce systemic effects are gaining significance [1]. This effect is possible thanks to the penetration of drug substances into not only the skin, but also permeation across the skin into the tissues localized beneath this organ or even into the circulation of blood. The surface of the skin is rarely the target for drug activity, except e.g. antiseptics and antimycotics, as far as dermatological use is concerned. Most of the drug substances, applied onto the skin for treatment purposes, must undergo diffusion to viable epidermis
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 165
(corticosteroids) or to the dermis (regulators of collagen synthesis). A drug substance, which penetrates into the epidermis or dermis, is partially or fully absorbed into systemic circulation. This might result in the occurrence of systemic side effects and, furthermore, even toxic effects. On the other hand, many drug forms are intended for transdermal delivery and are considered as a non-invasive and alternative for the conventional route of administration, e.g. in patients that cannot take drugs orally [1,12,14,17]. There are many unquestionable advantages of transdermal drug delivery [1,11,2528]. The drug substance avoids hepatic first pass effect and variances in the gastrointestinal tract that may influence on absorption. Transdermal systems enable achieving a steady-state drug concentration in the blood. This reduces the likelihood of peakassociated side effects, and ensures that drug concentration is above the minimal therapeutic concentration. Because transdermal patches are usually applied every three days, the dosing frequency is reduced. On the other hand, transdermal administration has some limitations. Transdermal delivered drug substances should be effective in the daily dose < 10 mg, and have appropriate physicochemical properties (log P in the range 1-3, molecular weight 100-500 Da, aqueous solubility > 100 g/ml, melting point < 200oC, unionized form). Sometimes, after the application of a drug form onto the skin, sensitization or irritation reactions appear, caused by the drug substance or excipients. Transdermal systems are relatively complex drug forms, expensive to develop and manufacture, and thus limited to relatively a few manufacturers. SKIN PERMEATION PATHWAYS Penetration is the movement of a molecule into a particular layer, while permeation is the passage through one layer into another layer or compartment. Absorption is usually the uptake of a substance into the blood vessels. The ability of exogenous substances to penetrate and to permeate the skin depends on the two following processes: the substance must be released in the dissolved form from the vehicle, and then it must overcome the stratum corneum barrier (Fig. 1). Both processes are closely related and depend on the physicochemical properties of permeant, a type of carrier and penetration pathway [17,29-31]. When the drug form is present on the skin, the drug substance can permeate the skin via the transfollicular pathway - through the skin appendages (Fig. 1). Substances diffuse even 1000 times faster through the skin appendages than via the other way. Unfortunately, the transfollicular pathway is of little significance, as skin appendages occupy less than 0.1% of the skin surface (except the hairy skin of the head), and secretions of glands being eliminated in the opposite direction impede the diffusion of substances. Thus, the transepidermal pathway - through the stratum corneum - is the main pathway of substance skin permeation. As viable skin layers, epidermis and dermis, might be considered to be a protein hydrogel, where the diffusion rate of permeating substance are comparable to the diffusion rate in the high viscous solution, the dead stratum corneum makes the factual barrier for substance permeation across the skin. Within this layer, molecules diffuse mainly via the tortuous lipid intercellular pathway. Corneocytes’ accurate arrangement and laminar composition of lipids surrounding the cells allow diffusing substances through the stratum corneum over 1000 times slower than through the viable skin layers. The lipophilic character of the stratum corneum favors permeation of nonpolar, unionized, small and lipophilic substances. Permeation of high lipophilic compounds may be limited by hydrophilic properties of deeper-situated viable skin layers.
166 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
Although it is possible for hydrophilic substances or ions to penetrate across the cells of the stratum corneum, these pathways have little part in the skin permeation process, and the permeant is subjected to a number of partitioning steps between hydrophilic corneocytes and lipophilic intercellular spaces [17,29]. It might be assumed that the stratum corneum is a homogenous lipid membrane, across which diffusion occurs according to the simply modified Fick’s first law (Eq. 1) [30]. This equation takes into consideration participation of a vehicle in the permeation process (K) and the process of diffusion across the stratum corneum.
J= where:
K D C L
(Eq. 1)
J-
steady-state flux - cumulative amount of permeating substance [g/cm2 h];
K-
partition coefficient of the substance between the stratum corneum and vehicle;
D-
diffusion coefficient of the substance in the stratum corneum [cm2/h];
C -
difference between applied concentration of substances and concentration below the stratum corneum (in vivo) or in the acceptor fluid (in vitro) [g/cm2];
L-
diffusion pathlength - thickness of the stratum corneum [cm].
Partition coefficient (K) of the substance between the stratum corneum and the vehicle is difficult to be determined, that is why the value of the partition coefficient of the substance between n-octanol and water is used for calculations. C can be replaced by applied concentration, because in normal conditions this concentration is much larger than the concentration in deeper skin layers. The real tortuous diffusion pathlength (L) is much longer (about 500 m) than the stratum corneum thickness (10-50 m); however, the stratum corneum thickness is mostly used, because it is measured more easily. STRATEGIES FOR ENHANCING SKIN PERMEATION As explained before, the stratum corneum is the principal barrier for skin absorption of drugs. The following groups of methods for increasing stratum corneum permeation are known and intensively investigated: -
methods related to the modification of drug molecules, choice of the appropriate vehicle and drug form, incorporation of drug into carrier;
-
methods related to the modification of the stratum corneum properties - hydration, chemical penetration enhancers;
-
electrically supported methods and devices - iontophoresis, electroporation, ultrasound;
-
methods related to the bypassing or removing of the stratum corneum - microneedles, jet injections, ablations.
DRUG MOLECULES, DRUG FORMS AND DRUG CARRIERS Prodrugs When the molecule has inappropriate physicochemical properties for permeation through the stratum corneum, usually lipophilicity, it can be optimized by altering its
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 167
chemical structure, which improves the stratum corneum/vehicle partitioning and diffusion characteristics [1]. Prodrugs are bioreversible derivatives of drug molecules that undergo an enzymatic and/or chemical transformation in vivo to release the active parent drug, which can then exert the desired pharmacological effect [32]. The skin contains appreciable non-specific esterase activity, and thus steroid esters provide greater topical anti-inflammatory activity than the parent steroids do, as the lipophilic moieties added to the steroids (e.g. the valerate in betamethasone-17-valerate) improve partitioning of the molecules into the stratum corneum compared with the parent steroid [17]. The most commonly applied prodrug strategy is to covalently bond an active substance with an inactive moiety by an ester bond [1,33]. The prodrugs were experimentally used to increase skin permeation of different groups of drug substances, such as non-steroidal antiinflammatory [34-36], beta-blockers [37], anti-cancers [38,39] and others [1]. Although about 5-7% of drugs approved can be classified as prodrugs, within preparations used onto the skin the most successful application of the prodrugs remains the steroids [1,32]. Finally, what is interesting, there are a lot of patent applications by Yu and Xu [40] concerning positively charged water-soluble prodrugs of acetaminophen, acetylsalicylic acid, diclofenac, diflunisal, ibuprofen, ketoprofen, prostaglandins and others, with a very high skin penetration rate. Ion Pairs Charged molecules do not readily penetrate lipid membrane, such as the stratum corneum [31]. The ion pairs approach utilizes oppositely charged species added to charged drugs, forming an ion pair in which the charge is temporarily neutralized. The pair of oppositely charged ions is held together by Coulomb attraction [41]. The ion pair partitions into the stratum corneum lipids, diffuses through this layer and dissociates in the viable epidermis into its charged compounds [1,31]. The good examples for ion pair are ibuprofen with triethylamine as a pairing agent [41] and equimolar mixtures of salicylic acid and various alkyl amines [42]. The recent studies on the ion pairs skin penetration enhancement method concern indapamide with organic acids [43] and aminolevulic acid with HCl/alcohol [44], but in general, the enhancement effect of ion pairs is rather modest. Thermodynamic Activity of Permeants and Supersaturation The greatest flux of a drug through a membrane occurs when the drug is at its maximum thermodynamic activity [1,31]. At saturation, equilibrium exists between the solid and liquid phase and activity equals 1. Therefore, all vehicles that contain a finely ground suspension exist as saturated solutions of the drug [1,31]. The most common method of creating a supersaturated system is the co-solvent method [1,45-49]. Saturated solutions of the co-solvent mixture are combined with a poor solvent to create a supersaturated system [35-37]. Supersaturated systems can also be obtained through the evaporation of a volatile co-solvent [50], by cooling a heated saturated solution down to skin temperature [51], or by moisture from the skin absorbing into a formulation, and acting as an antisolvent [52]. Although the drug flux from supersaturated systems increases over 10 times, such systems are ungrateful because of their instability represented by crystallization. There are many attempts to inhibit the drug’s crystallization, e.g. by using different polymers
168 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
as antinucleants [46,47,49,53-56], however to maintain a reasonable storage period of supersaturated systems is still a challenge. Eutectic Systems The molecule can diffuse across the membrane only when it is in a dissolved form. A eutectic system is a mixture of two or more compounds, which do not interact to form a new chemical entity, but its melting point is lower than the melting points of particular compounds and the solubility increases [1]. A linear correlation was seen when the log of steady-state flux is plotted against the melting point, indicating that the lower the melting point (greater solubility) the better the penetration [1,31]. The great achievements in such systems were to launch the drug products EMLA® (1:1 eutectic mixture of lidocaine and prilocaine) and S-Caine® (1:1 lidocaine and tetracaine) on the market [1,57]. Both mixtures have a melting point below room temperature and are incorporated into a cream form. The products are intended for use on intact skin to provide local analgesia, e.g. before injections and superficial dermatological procedures, such as dermal filler injections, pulsed dye laser therapy, facial laser resurfacing, and laser-assisted tattoos removal. The disadvantages of using EMLA® are a delay in analgesia and necessity of applying occlusive conditions. The vehicle of S-Caine® forms a pliable peel on the skin when exposed to air, that can be easily removed, meaning no occlusion is required [1]. The success of the EMLA® mixture stimulated the studies on eutectic systems with other drugs: testosterone [58], ibuprofen [59] and propranolol [60]. Liposomes Liposomes are rounded, stable vesicles composed of one or more concentric lipid bilayers [1,61]. Liposomes have two compartments: an aqueous central core and a lipophilic region within the lipid bilayer. Hydrophilic drugs can be incorporated into the inner aqueous compartment, while lipophilic drugs can be incorporated within the lipid bilayers. Conventional liposomes are composed of phospholipids, and usually with cholesterol to stabilize the vesicle. The most common phospholipid is phosphatidylcholine obtained from soybean or egg yolk. The conventional liposomes are most often prepared by the film hydration method [17,62], where the liposome compounds are dissolved in a volatile solvent, next the solvent is evaporated and a thin film of lipids deposits on the wall of the container. An aqueous solution of the drug to be incorporated is added at a temperature greater than the phase transition temperature of the lipids. As a result, multilamellar vesicles (MLV, < 10 m in diameter) are formed, containing several lipid bilayers surrounding an aqueous drop. MLV can be next sonicated or extruded to obtain large unilamellar vesicles (LUV, 1-5 m in diameter), or small unilamellar vesicles (SUV, 0.1-0.5 m in diameter). The conventional liposomes for transdermal application induced the unfounded enthusiasm. Simply, they are too large and too rigid to penetrate and permeate the stratum corneum. Liposomes are often adsorbed into the skin surface and fused with the outer layer of stratum corneum, and next release the free drug substance, which independently permeates the skin [31]. The liposomal lipids can also act as a skin penetration enhancer. The great problem with liposomes is their limited stability. A range of modifications in liposome compounding are proposed to decrease their rigidity and to increase stability, e.g. Oh et al. [63] describe polysorbate-based deformable liposomes for transdermal
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 169
delivery of retinol (see p. Transfersomes). There are many other attempts to formulate liposomes-like vehicles, such as niosomes, transfersomes, ehtosomes, and catezomes. Niosomes Niosomes (non-ionic surfactant vehicles) are formed by the self-assembly of nonionic surfactants in an aqueous dispersion; comparing to the liposomes they are flexible and more stable [1,2,31]. However, they often reduce fluxes of drugs in comparison to the conventional liposomes [64,65]. Transfersomes The transfersomes introduced by Cevc are highly deformable (or elastic) liposomes [1,17,31,66]. Transfersomes are composed of phospholipids, usually phosphatidylcholine, and surfactant (10-25%), such as sodium cholate, deoxycholate, Span 80, Tween 80 or dipotassium glycyrrhizinate [1,67]. Surfactant molecules act as an “edge activator” and give flexibility [17]. The formulation also contains a few percent of ethanol, and with the final aqueous lipid suspension, its total lipid concentration ranges between 4 and 10% [17]. Preparation methods are similar to those for conventional liposomes. The film hydration method is used most commonly [1,67]. Ultra deformable transfersomes are claimed to be able to squeeze through pores that are 10% of the vesicle diameter (about 20 nm diameter) [17,66]. Whereas the diffusion gradient is the driving force behind the topical delivery of drugs, the osmotic gradient across the skin is thought to be responsible for driving elastic vesicles [1]. The difference in water content varies from almost 100% at the epidermal/dermal junction to approximately 20% at the skin’s surface (depending on the environment) [1,17]. After applying transfersomes onto the skin’s surface, the formulation will dry, and the vesicles start to partially dehydrate, resulting in the vesicles becoming flattened or curved [1,17]. To maintain stability, the vesicle will penetrate deeper into the stratum corneum, where water content is higher [17]. Such a hydration theory is supported by the reduction in flux observed when the skin is occluded [1,68]. Transfersomes have been successfully used as topical and transdermal carriers for number of drug substances, including retinol [63], diclofenac [69], triamcinolone [70], dexamethasone [71], methotrexate [72], ketotifen [73], zidovudine [74], ethinylestradiol [75], often resulting in an increase of skin penetration/permeation. The transfersomal form of ketoprofen is in phase III of clinical trials [76-78]. Ethosomes Conventional liposomes and transfersomes can contain up to 10% of ethanol. Ethosomes are liposomes that contain 20-45% of ethanol, and as liposomes, they are composed of phospholipids [79]. Ethosomes are often prepared by first dissolving the lipids and drug in ethanol, then adding the aqueous component as a fine stream with thorough mixing [1,80]. High ethanol content results in ethosomes being much smaller than liposomes, and thus size reduction is unnecessary. Furthermore, ethanol enhances solubility of more lipophilic drugs [1]. The mechanism of ethosome action remains unclear [17]. Both the components of ethosomes, ethanol and phospholipids, can act as penetration enhancers [1,80]. However, ethosomes were much more effective penetration enhancers than hydro-ethanolic solu-
170 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
tions, ethanol or an ethanolic phospholipids solution [1,80]. It seems that ethanol initially disrupts the lipids within the stratum corneum, and next ethosomes, which are more flexible than liposomes, squeeze through the compromised stratum corneum [81]. Ethosomal formulations increased skin penetration of ketotifen [73], minoxidil [80], testosterone [80], acyclovir [81], trihexphenidyl hydrochloride [82], and cannabidiol [83]. The Supra-Vir cream containing acyclovir in ethosomes is available on some markets, and ethosomal erythromycin and clindamycin are clinically tested [84]. Catezomes Catezomes are a non-phospholipids vesicle with a cationic charge that makes them substantive to skin. They are used to deliver oil-soluble ingredients, like UV absorbers and ceramides or water-soluble drugs, like panthenol. Solid Lipid Nanoparticles By incorporating drugs into nanoparticles, such limitations as poor aqueous solubility or inadequate stability can be overcome [1]. Solid lipid nanoparticles can be obtained by high shear homogenization and ultrasonic dispersion, high pressure homogenization or emulsification, or vaporization methods [31,85,86]. Drug distribution within the nanoparticle is a function of particle composition (lipid, drug, and surfactant) and of the method of preparation [1]. A drug can be homogenously dispersed throughout the matrix of the nanoparticle, or it can be loaded into the core or the particle shell. The localization of the drug influences its release rate. A drug from the shell is usually released rapidly, whereas from the core – the process is more or less, prolonged [86]. Although there are many studies on the use of solid lipid nanoparticles in transdermal delivery of drugs, e.g. for isotretinoin [85], flurbiprofen [87], ketoprofen [88], naproxen [88], prednicarbate [89], retinol [90], the major chance for solid lipid nanoparticles propagations seems to be application in cosmetics, including sunscreens [91-93]. Microemulsions Microemulsions are defined as a system of water, oil and surfactants, which are transparent, a single optically isotropic and thermodynamic stable liquid [94]. Microemulsions can be considered as ideal liquid vehicles for drug delivery as they have most of the requirements for this: thermodynamic stability, ease of formulation, low viscosity, high solubilization capacity and small droplet size [95]. The major problem for microemulsions is that drugs can precipitate and crystallize as large crystals during storage or dilution. The dilution effects on the stability of the microemulsions were in most studies neglected and not considered [96]. In addition, the toxicity of the microemulsions compounds still imposes limitations in the use of this carrier [31]. Although there are many studies on skin application of drug-loaded microemulsions, and this carrier seems very promising, the obtainment of proper drug products based on microemulsions is still a challenge [94,96]. MODIFICATIONS OF STRATUM CORNEUM Hydration Hydration can be achieved by soaking the skin, applying a formulation with high water content and creating an occlusive condition [1,17]. Occlusion prevents natural
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 171
water loss from the skin, thus the water content of the stratum corneum equalizes with the underlying layers. Occlusion can be achieved by using dressings, hydrophobic ointments or patch-based formulations [17]. It is suggested that small amounts of water is present in the polar region of the stratum corneum lipids. The water insertion loosens lipid packing, increasing the mobility of the lipids chains [1,97], and larger amounts of water can exist as a separate phase in the intercellular spaces or can be taken up by the corneocytes [98]. Occlusion also causes an increase in temperature of the skin’s surface and dermal clearance in dermal capillaries. Unfortunately, long-term occlusion results in maceration of the skin and the high humidity of the skin favor the growth of microorganisms. Chemical Skin Penetration Enhancers Penetration enhancers are excipients that are added intentionally to the preparations for reversibly decrease the barrier of the stratum corneum against therapeutic substances [99,100]. A perfect penetration enhancer should be characterized by the following major properties [1,99,100]: -
it should be pharmacologically inactive;
-
it should be non-toxic, non-irritant, non-allergy inducing;
-
it should give a reproducible effect;
-
there should be a quick and complete reconstitution of the skin’s barrier after its removal;
-
it should be compatible with other compounds of formulation;
-
its action should be unidirectional - it should enhance drug absorption into the skin, but not promote the loss of endogenous substances;
-
it should be acceptable as far as its organoleptic properties are concerned - fragrancefree and colorless.
The most significant and characteristic feature of a safe penetration enhancer is its reversible effect, i.e. that a proper stratum corneum barrier is restored after preparation application. Elimination of a penetration enhancer from the stratum corneum may occur as a result of evaporation from the skin (e.g. ethanol or terpenes) or diffusion into deeper layers of the skin and, in consequence, into the blood circulation. The first elimination pathway, that is evaporation, is definitely favorable because it diminishes risk of penetration enhancer absorption into the blood. The mechanism of penetration enhancer action consists mainly in their polarity. These compounds, located in the stratum corneum, according to their physicochemical properties, may: -
disorder the fine arrangement of intercellular lipids within the stratum corneum;
-
increase fluidity or dissolve intercellular lipids;
-
change the hydration of lipids polar groups.
It is also a significant fact that penetration enhancers increase solubility of drug substances in the stratum corneum owing to their own solubility in the lipids of this layer, thus increasing the partition coefficient of penetrants between the stratum corneum and
172 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
the vehicle. The most widely used enhancers nowadays are alcohols (ethanol), glycols (propylene glycol), fatty acids (oleic acid), terpenes (menthol). Ethanol Ethanol is used in many transdermal formulations and is often the solvent of choice used in patches [17,99,100]. The mechanism of ethanol action at a low concentration (25%) on the stratum corneum consists in its interaction with lipids polar groups so that the fine arrangement of lipids becomes disturbed that, in turn, results in the increase of “fluidity” of these compounds. As far as higher concentrations (> 50%) are concerned, stronger physicochemical interactions may cause conformational changes of -keratin, an increase in the number of its free sulfhydryl groups and a partial extraction of lipids. It results in the formation of hydrophilic micropores enabling the absorption of polar molecules. Ethanol increases percutaneous absorption of, inter alia, 5–fluorouracil, hydrocortisone and estradiol [99-102], which was taken advantage of in transdermal therapeutic systems containing this drug substances. On the other hand, ethanol, at concentrations exceeding 60%, decreases the percutaneous absorption of many drugs probably because of dehydration of viable skin layers. Propylene Glycol It has been suggested that the propylene glycol effect on the lipids of the stratum corneum is comparable to the ethanol action [99,100]. Propylene glycol also causes changes in the structure of -keratin, probably because of the movement of bound water and solvation of this protein. The combination of propylene glycol with oleic acid in one system (action synergism) obtains the highest activity. It has also been proposed that the presence of water in the vehicle is necessary for the effect of propylene glycol to be exerted. Fatty Acids Fatty acids, that are used as penetration enhancers, are characterized by double bond in cis conformation (e.g. oleic acid), whereas unsaturated fatty acids in trans conformation have no influence on skin absorption of drug substances [99,100]. Fatty acids in cis conformation have a non-linear spatial structure and they incorporate within alkyl chains of the lipids of the stratum corneum, what in turn disturbs their organized structure leading to looseness and an increase in the fluidity of lipids. The molecules of oleic acid reveal high affinity to the lipids of the stratum corneum, however, their affinity to keratin and other components of this layer was not discovered. The probable mechanism of oleic acid action consists in the partition of the liquid and solid phase of lipids in the stratum corneum. The transport of polar and ionic compounds is also increased in the spots where partition takes place. Terpenes, Terpenoids and Essential Oils Terpenes/terpenoids are a large group of natural compounds deriving from active isoprene [103]. The number of carbon atoms in the terpene molecule is a multiple of 5; for example, C10 (monoterpenes), C15 (sesquiterpenes) and C20 (diterpenes). They are the main constituents of essential oils - the active substances of many plants. Terpenes have their-own very broad spectrum of biological activity, which in essential oils is an effect of synergistic or antagonistic activity of each compounds. Topical preparations containing terpenes exhibit antibacterial, rubefacient, mild-analgesic and anti-inflammatory
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 173
activity. Essential oils and/or terpenes are present in the most of cosmetics, household chemicals and drug products. Most of drug products are intended for topical treatment or, after spreading, for inhalation. They can be in the form of solutions, ointments, emulsions, balms, gels, lotions. Unfortunately, manufacturers of drug products and cosmetics very often do not take consider that terpenes are very strong skin penetration enhancers for different classes of substances. Interest in terpenes, terpenoids and essential oils, containing these substances, as penetration enhancers has been observed in recent years. These compounds enhance penetration of lipophilic, hydrophilic as well as ionized substances. Penetration enhancement is related to increase in substance solubility in the stratum corneum. Terpenes also increase stratum corneum/vehicle partition coefficient for lipophilic compounds and disorder the fine arrangement of intercellular lipids of the stratum corneum. Terpenes are particularly active to non-steroidal anti-inflammatory drugs [99,100,104]; and the synergism of terpenes and ethanol action is often used. It is interesting that in some carriers terpenes can be inactive for drug substance [105]. The penetration enhancement by terpenes has been recently reviewed in deep by Aqil et al. [106] and Satra et al. [107]. Other Chemical Skin Penetration Enhancers There are also other groups of penetration enhancers that are experimentally tested, such as sulfoxides, azone, pyrrolidones, fatty alcohols, surfactants, phospholipids, that have been preciously reviewed by Williams and Barry [99] and Thong et al. [100]. Most of penetration enhancers are present in various commercial cosmetics and drug products for completely different reasons. Propylene glycol is used as the vehicle forming components or, like alcohol, as solvents. Surfactants solubilize active substances within the vehicle or they enable to obtain proper form of a preparation. Although great number of studies, there is still not an ideal penetration enhancer. Skin Disposition of Terpenes Skin penetration and elimination kinetics can be one of the parameters determining the in use safety of penetration enhancers. The one of better examined in this field penetration enhancers are terpenes [108]. Although most of terpenes are Generally Regarded as Safe substances (GRAS), the side effects depending on their skin absorption during usage cannot be excluded [109-113]. Physicochemical Properties of Investigated Terpenes For the skin penetration studies both acyclic monoterpenes: (±)--citronellol, (±)linalool, linalyl acetate and cyclic monoterpenes: (-)--pinene, (-)--pinene, eucalyptol (1,8-cineole) and terpinen-4-ol were chosen. All investigated compounds appear in the liquid form at room temperature. Considering their structure, the investigated terpenes represent different chemical classes: (-)--pinene and (-)--pinene are hydrocarbons; (±)-citronellol, (±)-linalool and terpinen-4-ol are alcohols; linalyl acetate is an ester and eucalyptol is an oxide. Based on aqueous solubility and log P value, the penetrants investigated were grouped as follows [114]: -
acyclic with log P value 3: (±)--citronellol and (±)-linalool;
-
acyclic with log P > 4: linalyl acetate;
-
cyclic with log P 3: eucalyptol and terpinen-4-ol;
174 Frontiers in Drug Design & Discovery, 2009, Vol. 4
-
Krzysztof Cal
cyclic with log P > 4: (-)--pinene and (-)--pinene.
Ex Vivo Skin Penetration of Pure Terpenes Skin penetration of terpenes was determined by their application in pure form for 1, 2 and 4 h onto the full human skin placed in flow-through diffusion cell [115,116]. The system was maintained at a temperature of 37±0.5°C and the skin was occluded. At the end of experiment, the stratum corneum was separated using a tape-stripping technique. The collected stratum corneum samples and the remaining epidermis with dermis, were extracted with methanol and the extracts were analyzed by gas chromatography [115117]. The amounts of terpenes penetrating into the skin after 4 h were very large. It was in the range from 200 to 1800 μg/cm2 and depended on the type of applied terpene. The total skin penetration of pure terpenes increases in the following order: linalyl acetate < (-)--pinene < (-)--pinene = eucalyptol << (±)--citronellol < terpinen-4-ol < (±)linalool. Total skin cumulation was similar for acyclic and cyclic terpenes with log P > 4. Among terpenes with log P 3 total skin absorption was 2-3 times greater for compounds with acyclic structure than those with cyclic structure. Higher skin absorption for oxygenic derivatives (alcohols and oxides) than for ester- and hydrocarbon-type terpenes was observed [115,116]. Influence of Vehicle Type on Ex Vivo Skin Penetration of Terpenes The application of two terpenes: (±)-linalool and terpinen-4-ol, incorporated singly into commonly used dermatological vehicles (oily solution, hydrogel and oil-in-water emulsion) allows to determine the influence of the type of vehicle on the skin penetration process [118]. The vehicles used demonstrated different properties. An oily solution is a typical anhydrous vehicle, while hydrogel contained 98% water. Terpenes in oil were completely dissolved, whereas only partly dissolved and/or dispersed in the hydrogel. Emulsion had intermediate properties. In this form, the terpenes were most probably present in several forms: dissolved in the oily internal phase, emulsified with the surfactants and forming an internal oily phase, as well as dissolved in surfactant micelles. Increased absorption time from 1 h to 4 h resulted in increased amounts of terpenes in the skin layers [118]. The greatest total cumulation in the skin was observed when terpenes were applied in hydrogel. This results from the high partition coefficient of the lipophilic terpene molecule between lipophilic stratum corneum and the hydrophilic vehicle. Hydrogel and the oily solution favored penetration of terpene with higher aqueous solubility [114] - cumulation of terpinen-4-ol in the skin layers, particularly epidermis with dermis, was greater than culmination of (±)-linalool. Skin penetration of terpenes applied as an oily solution and an oil/water emulsion were comparable and lower than when applied as hydrogel, mostly because of the unprofitable partition coefficient of penetrants between the stratum corneum and the vehicle. For all the preparations investigated, the cumulation of terpenes in epidermis with dermis was comparable or greater than in the stratum corneum. Influence of Vehicle Type on In Vivo Stratum Corneum Absorption of Terpenes in Humans The in vivo stratum corneum penetration was determined for (±)-linalool and terpinen-4-ol incorporated into two vehicles - oily solution and hydrogel [119]. The prepa-
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 175
rations were applied onto the skin of ventral forearms in humans. After 1 h, the stratum corneum layer was tape-stripped. The highest stratum corneum penetration was observed when terpenes were applied as hydrogel, and the stratum corneum absorption of terpinen-4-ol was 2.5 times greater than (±)-linalool. The stratum corneum absorption of both terpenes after application in oily solution was similar and 1.8 times smaller for (±)linalool and 4.8 times for terpinen-4-ol than for hydrogel vehicle. Comparing in vivo and ex vivo data after 1 h application, higher in vivo than ex vivo stratum corneum absorption of (±)-linalool applied in oily solution was noted while the in vivo cumulation of this terpene applied in hydrogel and terpinen-4-ol in oily solution was smaller than ex vivo. The same in vivo and ex vivo stratum corneum absorption of terpinen-4-ol administrated in hydrogel was determined. It was also proven that (±)-linalool in vivo retention in the stratum corneum is constant and is about 10 g/cm2, even at least 2 h after removal of (±)-linalool containing preparations, while for terpinen-4-ol the stratum corneum retention is about 5 g/cm2 after this time. In conclusions, presented results indicate a distinct relationship between skin penetration of terpenes and their physicochemical properties and the type of vehicle used. Current ex vivo and in vivo experimental models are imperfect [120]; therefore, it is difficult to introduce unequivocal advice relating to the use of terpenes in practice. The fact that only a small amount of terpenes is necessary for saturation of the stratum corneum and that they are able to cumulate in large amounts in the skin layers indicate overcoming the stratum corneum barrier by terpenes and possibility of easy in vivo penetration into the blood circulation. Because a higher skin penetration of terpenes from a hydrophilic vehicle was demonstrated, such formulations can cause side effects more easily [121]. ELECTRICALLY SUPPORTED METHODS AND DEVICES Iontophoresis Iontophoresis uses electric current flowing between two electrodes (anode and cathode) located onto the skin (Fig. 2) [1,17,31,100]. Substance, that undergoes electrolytic dissociation and possessing electric charge is dissolved in proper solvent and applied to the skin beneath the electrode of the same charge. The second electrode is placed in other spot and closes the electric circuit. At the moment of induction of current flow (approximately 0.5 mA/cm2), charged particles of a drug are repulsed from the electrode close to which they were applied and they migrate to the electrode of the opposite charge. This process takes place through the whole skin thickness, including dermis, thanks to which it is possible to obtain increased absorption. It is significant that permeability of the skin increases already under the influence of electric current alone due to disturbance of the lipid arrangement in the stratum corneum. Iontophoresis enables not only percutaneous delivery of charged substances, but it also enhances percutaneous penetration of small neutral molecules via electroosmosis. The phenomenon of electroosmosis, that is migration of endogenic Na+ ions to the cathode, which is also accompanied, by migration of water molecules together with dissolved substance, is used at this point. The iontophoresis is widely used in physiotherapy for, e.g., the following substances: -
applied beneath the anode: histamine, procaine and lidocaine, calcium (as calcium chloride), zinc, epinephrine;
176 Frontiers in Drug Design & Discovery, 2009, Vol. 4
-
Krzysztof Cal
applied beneath the cathode: diclofenac, ketoprofene (as salts), iodine, hydrocortisone. Active Electrode
Power Supply Applied Formulation
Indifferent
Electrode
Stratum corneum
Epidermis
Dermis
Capillary blood vessels
Fig. (2). Schematic diagram of iontophoresis.
The main advantage of iontophoresis over other transdermal enhancement strategies is its ease of control [1], because electrical current is responsible for the increased delivery [122]. Thus, by manipulating current density and duration, the dose may be tailored to an individual patient’s needs. The major successes in iontophoresis are introduction on the market LidoSite® and Ionsys™ devices. LidoSite® is iontophoretic patch delivering lidocaine-epinephrine that provides fast, effective analgesia before blood draws, venipunctures, and other superficial dermatological procedures [123]. Ionsys™ contains fentanyl, and allows patients to self-administer the drug according to their personal requirements for pain relief (maximum 6 doses per hour) [1,124,125]. The other drugs recently studied in iontophoretic condition are ketoprofen [126], dexamethasone [127], botulinum toxin [128]. Electroporation Electric voltage was utilized in electroporation for percutaneous delivery of drug substances [1,17,31,100]. The phenomenon of electroporation consists in formation of water channels (pores) in intercellular lipid bilayers of the stratum corneum under the influence of short (micro- to milliseconds) electric impulses (voltage 10-1000 V) (Fig. 3). Occurrence of the channels due to the electric voltage applied also regards other biological membranes. It is believed that voltage applied also leads to melting of the lipids of the corneal layer, additionally increasing its permeability. It is worth mentioning that
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 177
macromolecular substances may be delivered by electroporation. It seems to be promising method; however miniaturization of devices is essential to facilitate routine use by patients.
Pulse Electrode
Power Supply
Applied Formulation
Stratum corneum
Epidermis
Dermis
Capillary blood vessels
Fig. (3). Schematic diagram of electroporation.
Ultrasound Ultrasound (sonophoresis or phonophoresis) uses energy of sound waves to increase percutaneous absorption of drugs (Fig. 4) [1,17,31,100,129]. Under the influence of ultrasounds, air bubbles appear in the intercellular spaces of the stratum corneum that disorganize lipid arrangement and enlarge intercellular spaces. Additionally, tissue temperature is raised by a few degrees in the spot of ultrasound activity, caused increase drug solubility and enhances dermal clearance by increasing the local blood flow. Sonophoresis is used in physiotherapy and sport medicine, to administer drugs in neuropathic pains, degenerative diseases and inflammations of joints and muscles. Ultrasound may enhance drug penetration into the stratum corneum by a thousand-fold [31]. Other Electrically Supported Methods There are some studies on skin permeation of diamagnetic substances under influence of magnetic field (magnetophoresis) [31]. The radio waves have been tested to increase skin permeability for hydrophilic drugs [130]. A pulse from a high-powered laser was used to create stress or compression waves. This photomechanical waves can be directed onto the skin to increase permeability [1,31,131]. Combined strategies for skin
178 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
permeation enhancement, such as: iontophoresis or electroporation with chemical enhancers, iontophoresis with electroporation, etc. were also studied [1,17]. Power Supply
Ultrasonic Transducer
Applied Formulation
Stratum corneum
Epidermis
Dermis
Capillary blood vessels
Fig. (4). Schematic diagram of sonophoresis.
BYPASSING OR REMOVING OF STRATUM CORNEUM It seems that bypassing or removing of the stratum corneum may allow delivering of practically any substance. Such bypassing methods as microneedles and jet injections are, in fact, similar to the dermal injections – the drug is directly applied to the viable skin layers. However, both methods are painless. Trials of safe removal of the stratum corneum with use of such methods as tape-stripping, laser ablation, suction ablation, thermal ablation, microscissioning and consecutive drug application onto the viable epidermis have been undertaken continuously. Although promising, stratum corneum removing has significant disadvantage, namely long-time local removal of the skin barrier function, what may result in unpredictable side effects. The bypassing or removing of stratum corneum does not constitute transdermal drug delivery because the permeant is not crossing intact skin [17]. Microneedles Successful trials to omit the stratum corneum barrier are patches containing microneedles of length of 100-200 m and thickness of 10-50 m, usually. The patch is placed onto the skin before application of drug form, and next the drug penetrates to viable skin layer via created microholes (Fig. 5). Microneedles do not reach the nerve endings located within the dermis. The microneedles may also be coated with drug or
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 179
filled with drug; they can be also soluble. Microneedles are mainly tested for delivering of insulin, vaccines, DNA [1,17,31,132,133]. Microneedles patch
Stratum corneum
Epidermis
Dermis Capillary blood vessels
Fig. (5). Schematic diagram of application of microneedles patch.
Jet Injections Jet injection is an alternative method for drug delivery into the skin with use of needle and syringe [1]. The drug is delivered as a high-pressure jet (> 100 m/s), with sufficient intensity to pierce the skin [134]. The two major types of jet injectors, liquid jet injector and powder jet injector can be differentiate [1]. As microneedles, jet injectors are studied to deliver insulin, vaccines, and hormones [1]. Tape-Stripping and other Ablation Methods Tape-stripping is a technique with use of an adhesive tape that remove the stratum corneum layer by layer. Fragments of an adhesive tape are sequentially adhered onto the same penetration area of the skin, and then stripped off [17]. Although inexpensive, tape-stripping may be unrepeatable and unacceptable for routine use in patients. The stratum corneum can be removed by laser ablation [1,17]. The suction ablation uses a vacuum to produce a small blister on the skin [1]. Next, using an epidermatome, the upper surface of the blister is removed. The thermal ablation utilizes a current at radiofrequency (100 kHz). Dermabrasion and microscissiong uses aluminium oxide particles to rub off the stratum corneum. CYCLODEXTRINS IN SKIN FORMULATIONS Physicochemical Properties of Cyclodextrins Cyclodextrins (CDs) are cyclic oligosaccharides composed of at least six D-(+)glucopyranose units linked by -(1-4) bonds (Fig. 6a) [135,136]. Natural CDs occur in
180 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
the form of white crystalline powder, and they form stable hydrates. CDs have quite rigid structures (stabilized by hydrogen bonds between C2 and C3 hydroxyl groups) lacking free rotation in -(1-4) bonds, thus they form torus-like molecules (truncated cone) (Fig. 6b) [136]. CD molecules have a hydrophilic outer surface (all hydroxyl groups in the ring are located in the exterior of torus) and a hydrophobic interior (there are skeletal carbons with hydrogen atoms and oxygen bridges inside the cavity). The nonbonding electron pairs of the oxygen bridges are directed toward the inside cavity, thereby generating high electron density [137,138]. There are three main natural CDs: -, -, and -CD composed of six, seven, and eight glucose units, respectively. They differ in ring size and physicochemical properties [135]. It is possible to achieve higher homologues, but because of their properties – large cavity dimension, high aqueous solubility, and weak complex formation – they cannot be of practical use. The CDs contain 18 (-CD), 21 (-CD), or 24 (-CD) hydroxyl groups that can be chemically modified. To improve some physicochemical properties of natural CDs, many types of derivatives have been developed: hydrophilic (methylated, hydroxyalkylated, and branched), hydrophobic (ethylated), ionic (sulphated and phosphated) [137-140]. Derivatization of parent crystal CDs usually leads to achieving amorphous mixtures of isomers; thus, their aqueous solubility is much higher [141].
a
HOH2 C O HO
O
OH
HOH2C
HO
HO H2 C
O HO
CH OH 2 O HO
O OH O
b
O
HO
OH OH O
O
HO
O
HO O
OH
O
CH2OH
OH HO O O CH2 OH
CH OH 2
Fig. (6). Structure (a) and torus-like shape (b) of -CD molecule [136].
The most important attribute of CDs is the ability to create inclusion complexes with a large number of molecules or their portions; however, not all molecules (drugs) can form stable complexes. There are some limitations, like very high aqueous-soluble substances, that generally cannot be included [142]. The size – geometric factor of the molecule is most important because it decides whether the molecule is able to form “stable” inclusion with -, -, or - CD. If the molecule had adequate properties, it interacts with CD inside cavity without forming covalent bonds; this interaction is “guest/host” type. CD inclusion complex is mainly formed via the substitution of included water by the appropriate “guest” molecule. Release of the enthalpy-rich water molecules from the cavity decreases the energy of the system. A decrease in the energy of the system is due to reduce the contact surface area between the solvent and solute as well as solvent (highly polar water) and imperfectly solvated (hydrophobic) CD cavity. Some other fac-
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 181
tors, such as hydrogen bonding, changes in surface tension, van der Waals’ interactions, and ring strain release, also can have some influence on the complex formation [143145]. Skin Toxicity Natural CDs are regarded as a non-irritant to skin [137]. -CD, applied under occlusion condition onto the skin surface in humans, does not induce irritation or allergenic reaction [137,138]. Skin compatibility with CDs, both natural and a wide range of derivatives, has been summarized recently by Piel et al. [146]. All tested CDs were well tolerated by the stratum corneum, with the exception of dimethylated derivatives where changes in corneoxenometry were observed, which can indicate disruption in the lipid bilayers. Overall, natural CDs and their hydrophilic derivatives are not able to permeate skin barrier in significant amounts; thus they are safe for topical applications [138,147149]. Only lipophilic RM--CD can interact with membranes more readily but in high concentrations [146,150]. Generally, all types of above-mentioned CDs can be used in skin formulations safely and without risk of irritation; even methylated CDs in low concentrations can be safely applied. Only for aqueous solution/suspension containing high concentration of CDs there is some probability that methylated CD will interact with the stratum corneum lipids (cholesterol, triglycerides) and temporarily affect the membrane integrity [150]. Application of Cyclodextrins in Skin Formulations Practical Aspects Practical use of the obtained complex with CDs is more complicated than forming the inclusion. As presented by Szejtli [151], drug formulations with CDs are usually not bioequivalent to their reference products. Even when only better stability is required, the absorption of the drug will be usually affected in the positive or negative way. CDs are able to modify dermal application both by increasing (supergenerics) [141,152-156] or by modifying delivery (retarded or prolonged release) [141,157-159]; hence, no reference products exist for performing a comparative study. Therefore, for registration purpose, all preclinical studies are necessary, which increase developmental cost for such products, and the reduction of dose does not solve this problem [151]. Advantages For liquid preparations intended for application onto the skin, the most biocompatible solvent is water. Montassier et al. [160,161] proposed aqueous-soluble tretinoin/CD complexes as an alternative for use 60% ethanol as a solvent. The solubility of drug was the same in both cases. Organic solvents, like ethanol, are corrosive to the skin and their volatility may cause recrystallisation of the drug substance during storage. CDs can be an alternative for them. In the skin formulations, improved solubility is usually associated with enhanced bioavailability, like for: sericoside [154], piroxicam [162], celeoxib [163], tretinoin [160,161], 4-biphenylylacetic acid [164], hydrocortisone [165], and bupranolol [166]. The increase in bioavailability is mainly due to higher concentration in the site of administration, caused rather by higher aqueous solubility, and thus improved availability onto the tissue surface than enhancement activity of CDs by itself. Schoch et al. [167]
182 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
demonstrated that modified CDs (HP--CD as well as oktakis--CD) significantly increased in vitro corneal permeability of diclofenac sodium in comparison to formulation containing Cremophor. The improvement of skin permeation is possible by increasing drug solubility, which improved availability onto the skin surface or by influence on the barrier function of stratum corneum (probably only for DM--CD) [168]. In the studies of Ventura et al. [168] on percutaneous absorption of celecoxib from 0.01% solution or suspension in presence of HP--CD and DM--CD, both CDs influenced the in vitro drug permeation through human skin by shortening the lag time from 2 h to 35 min for 5% DM--CD and to 1 h for the same concentration of HP--CD. The cumulative amount of celecoxib permeating through the skin after 24 h was up to 7–8 times greater when either HP--CD or DM--CD was present in the donor phase in contrast to the uncomplexed drug substance. Complexation of dexamethasone with -CD and HP--CD protects the drug substance against skin metabolism. Such studies performed by Lopez et al. [155] on the homogenised mouse skin during 2 h resulted in 30 and 65% degradation of dexamethasone for complexed and free drug, respectively. However, this stabilising effect is limited ex vivo and in vivo by the non-homogenized full skin because CDs are not able to penetrate into viable skin layers. Limitations One of the biggest hopes for CD application in topical formulations was using them as universal non-irritating penetration enhancer for transdermal application of drug substances. CDs are regarded by some authors as classic enhancers that are able to extract all the major lipid classes and proteins, and thus reduce skin barrier function [149,169], while others regarded their action as problematic and rather unproved [17]. Above hypotheses are based on the studies on the animal skin (hairless mice or rats) performed in 1990s by Legendre et al. [149], Vollmer et al. [169] and Bentley et al. [170]. Although Bentley et al. [170] indicated that HP--CD caused removal and possible disorganisation of the lipids in the stratum corneum and Legendre et al. [149] stated that HP--CD exhibited 2-fold higher activity in removing cholesterol from rat skin than RM--CD, it seems that only methylated CDs applied in high concentrations (10–20%) in aqueous solutions can have influence on the stratum corneum [146,166,169]. No effect of possible disruption was seen for -, HP-- and -CDs [170]. Pretreatment studies realised on the rat skin for bupranolol as drug substance showed no flux increases for 2 and 10% solution of HP--CD, while for RM--CD flux increased markedly for both concentrations with concentration dependency [166]. Shaker et al. [148] suggested that HP--CD and its inclusion complex with corticosterone do not effectively penetrate into or transport through the skin. In hairless mouse skin model, HP--CD did not change the barrier function of the stratum corneum, nor did it enhance transport of corticosterone. Similar results were obtained during studies with human skin [171-173]. Valjakka-Koskela et al. [171] reported that addition of 10% of HP--CD decreases levosimendan flux through human in vitro skin from the solution form. Preis et al. [172] reported that incorporation of hydrocortisone/HP--CD inclusion complex into gel formulation did not involves changes in the drug permeation through human ex vivo skin in comparison to preparation without the CDs. In the same studies incorporation of drug substance/-CD inclusion complex into formulation resulted in decrease release. Also in the studies by Simeoni et al. [173] with excised human skin, HP--CD had no
Across Skin Barrier
Frontiers in Drug Design & Discovery, 2009, Vol. 4 183
effect on the stratum corneum and epidermal concentration of sunscreen agent buthylmethoxydibenzoylmethane in comparison to free molecules applied as solution. Additionally, they found that sulfobutylether--CD markedly reduced epidermal absorption of such filter without reducing its stratum corneum penetration [173]. Some of CDs, like HP--CD, -CD or sulfobutylether--CD, may not only had no influence on barrier function but they even, in certain cases, may had protective properties against penetration of drug substances into deeper skin layers [173,174]. Ventura et al. [168] presented opposite statement. They concluded that both HP-CD and DM--CD enhanced drug flux through human stratum corneum and epidermis by means of an increase of dissolution rate of the drug as well as a direct action on the stratum corneum. The direct impact on the stratum corneum is overestimated because during 24 h experiment with 3% solution of CDs, HP--CD had no destructive effect and seems that was not able to act as penetration enhancer. Although DM--CD in above conditions caused separation of corneocytes layers and had significant influence on barrier function of the skin, its inclusion complex with the drug showed less injurious effect than DM--CD alone. Possible Applications of Cyclodextrins in Skin Formulations CDs can be used to improve lipophilic drug entrapment in the aqueous liposomal phase, and thus result in a new two-carrier system of drugs-in-CD-in-liposome formulations [175-178]. Maestrelli et al. [179] investigated such system for transdermal delivery of ketoprofen. They achieved improvement in drug entrapment for ketoprofen/HP--CD complex in equimolar ratio. Encapsulation efficacy increased with CD concentration; however, high concentration destabilised liposomal membrane. Liposomal formulations resulted in slower and prolonged drug permeation through membrane (about 40% drug permeated after 24 h) in comparison to drug solution (60% after 4 h). The CD interaction with liposome lipid membrane depends on the type of CD, complexed drug substance and lipids [176-178]. In addition, some other lipidic systems, such as microspheres, are combined with drug/CD complexes, especially for mucosal delivery [180-183]. As skin penetration enhancement is difficult by using only CDs and as the influence on drug flux depends not only on CDs and drug substance properties, but also on other formulation components, recently CDs have been applied with good efficacy as coenhancers or in combination with other methods – supersaturation, electroporation, or iontophoresis [156,184-188]. CONCLUSIONS As Morrow et al. [1] note, the most of transdermal drug formulations are based on the passive diffusion, often with use of penetration enhancers, of a low molecular weight, lipophilic, unionized drugs. Although the conception of chemical skin penetration enhancement is outstanding and still promising, thousands of studies do not result in an ideal penetration enhancer. What’s more, the activity of known penetration enhancers to the highly hydrophilic or macromolecular drugs is often disputable. On the other hand, among passive penetration enhancement strategies, a few have been successfully introduced in commercial products. Recently methods, such as iontophoresis, electroporation, microneedles, are less limiting by the physicochemical properties of permeants,
184 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Krzysztof Cal
and have been show to increase delivery of macromolecules through the skin. Iontophoretic transdermal systems are recently employed in the medical care. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39]
Morrow, D.I.J.; McCarron, P.A.; Woolfson, A.D.; Donnelly, R.F. Open Drug Deliv. J., 2007, 1, 36. Florence, A.T.; Salole, E.G. Topics in Pharmacy. Routes of Drug Administration, Wright Publishing Company: London, 1990. Fuchs, L. New Kreuterbuch, Basel, 1543. Przeglad Farmaceutyczny /Pharmaceutical Review/, 2008, 190, 46. /in Polish/ Scheuplein, R.J.; Blank, I.H. Physiol. Rev., 1971, 51, 702. Surber, C.; Smith, E.W. Dermatology, 2005, 210, 157. Blank, I.H. J. Invest. Dermatol., 1953, 21, 259. Wolf, J. Anat. Forsch., 1939, 46, 170. Scheuplein, R.J. J. Invest. Dermatol., 1967, 48, 79. Scheuplein, R.J.; Blank, I.H.; Brauner, G.J.; MacFarlane, D.J. J. Invest. Dermatol., 1969, 52, 63. Thomas, B.J.; Finnin, B.C. Drug Discov. Today, 2004, 9, 697. Benson, H.A.E. Curr. Drug Deliv., 2005, 2, 23. Rios, M. Pharm. Technol., 2007, 10. Wiedersberg, S.; Leopold, C.S.; Guy, R.H. Eur. J. Pharm. Biopharm., 2008, 68, 453. Burns, D.A.; Breathnach, S.M.; Cox, N.H.; Griffiths, C.E.M. Rook`s Textbook of Dermatology, Blackwell Science: Oxford, 2004. Anderson, R.L.; Cassidy, J.M. J. Invest. Dermatol., 1973, 61, 30. Williams, A.C. Transdermal and Topical Drug Delivery, Pharmaceutical Press: London, Chicago, 2003. Elias, P.M.; McNutt, N.S.; Friend, D.S. Anat. Rec., 1977, 189, 577. Madison, K.C. J. Invest. Dermatol., 2003, 121, 231. Michaels, A.S.; Chandrasekaran, S.K.; Shaw, J.W. AIChE J., 1975, 21, 985. Downing, D. J. Lipid Res., 1992, 33, 301. Elias, P.M. J. Control. Release, 1991, 15, 199. Harada, K.; Murakami, T.; Yata, N.; Yamamoto, S. J. Invest. Dermatol., 1992, 99, 278. Knutson, K.; Potts, R.O.; Guzek, D.B.; Golden, G.M.; McKie, J.E.; Lambert, L.D.; Higuchi, W.I. J. Control. Release, 1985, 2, 67. Brown, M.B.; Martin, G.P.; Jones, S.A.; Akomeah, F.K. Drug Deliv., 2006, 13, 175. Scheindlin, S. Mol. Interv., 2004, 4, 308. Berlin, C.M.; May-McCarver, D.G.; Notterman, D.A.; Ward, R.M.; Weismann, D.N.; Wilson, G.S.; Wilson, J.T. Pediatrics, 1997, 100, 143. Paparella, S. J. Emerg. Nurs., 2005, 31, 278. Bronaugh, R.L.; Maibach, H.I. Percutaneous absorption, Marcel Dekker: New York, 1999. Gabard, B.; Elsner, P.; Surber, C.; Treffel, P. Dermatopharmacology of topical preparations, Springer-Verlag: Berlin, 1999. Smith, E.W.; Maibach, H.I. Percutaneous penetration enhancers, CRC Press: Boca Raton, London, New York, 2006. Rautio, J.; Kumpulainen, H.; Heimbach, T.; Oliyai, R.; Oh, D.; Jarvinen, T.; Savolainen, J. Nature Rev. Drug Discov., 2008, 7, 255. Waller, D.G.; George, C.F. Brit. J. Clin. Pharmacol., 1989, 28, 497. Rautio, J.; Nevalainen, T.; Taipale, H.; Vepsalainen, J.; Gynther, J.; Laine, K.; Jarvinen, T. Eur. J. Pharm. Sci., 2000, 11, 157. Bonina, F.P.; Puglia, C.; Barbuzzi, T.; de Caprariis, P.; Palagiano, F.; Rimoli, M.G.; Saija, A. Eur. J. Pharm. Sci., 2001, 14, 123. Thomas, J.D.; Sloan, K.B. Int. J. Pharm., 2008, 346, 80. Ahmed, S.; Imai, T.; Otagiri, M. Pharm. Res., 1996, 13, 1524. Saab, A.N.; Sloan, K.B.; Beall, H.D.; Villaneuva, R. J. Pharm. Sci., 1990, 79, 1099. Beall, H.; Prankerd, R.; Sloan, K. Int. J. Pharm., 1994, 111, 223.
Across Skin Barrier [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 185
http://www.wipo.int/pctdb/en/ Sarveiya, V.; Templeton, J.F.; Benson, H.A. J. Pharm. Pharmacol., 2004, 56, 717. Megwa, S.A.; Cross, S.E.; Benson, H.A.; Roberts, M.S. J. Pharm. Pharmacol., 2000, 52, 919. Ren, C.; Fang, L.; Li, T.; Wang, M.; Zhao, L.; He, Z. Int. J. Pharm., 2008, 350, 43. Maisch, T.; Worlicek, C.; Babilas, P.; Landthaler, M.; Szeimies, R.M. Exp. Dermatol., doi:10.1111/j.1600-0625.2008.00705.x Poulsen, B.J.; Young, E.; Coquilla, V.; Katz, M. J. Pharm. Sci., 1968, 57, 928. Kondo, S.; Sugimoto, I. J. Pharmacobiodyn., 1987, 10, 587. Iervolino, M.; Cappello, B.; Raghavan, S.L.; Hadgraft, J. Int. J. Pharm., 2001, 212, 131. Iervolino, M.; Raghavan, S.L.; Hadgraft, J. Int. J. Pharm., 2000, 198, 229. Megrab, N.A.; Williams, A.C.; Barry, B.W. J. Control. Release, 1995, 36, 277. Moser, K.; Kriwet, K.; Kalia, Y.N.; Guy, R.H. J. Control. Release, 2001, 73, 245. Henmi, T.; Fujii, M.; Kikuchi, K.; Yamanobe, N.; Matsumoto, M. Chem. Pharm. Bull., 1994, 42, 651. Kemken, J.; Ziegler, A.; Muller, B.W. Pharm. Res., 1992, 9, 554. Cilurzo, F.; Minghetti, P.; Casiraghi, A.; Tosi, L.; Pagani, S.; Montanari, L. Eur. J. Pharm. Biopharm., 2005, 60, 61. Raghavan, S.L.; Trividic, A.; Davis, A.F.; Hadgraft, J. Int. J. Pharm., 2001, 212, 213. Kim, J.H.; Choi, H.K. Int. J. Pharm., 2002, 236, 81. Ma, X.; Taw, J.; Chiang, C. Int. J. Pharm., 1996, 142, 115. Nyqvist-Mayer, A.A.; Brodin, A.F.; Frank, S.G. J. Pharm. Sci., 1986, 75, 365. Kaplun-Frischoff, Y.; Touitou, E. J. Pharm. Sci., 1997, 86, 1394. Stott, P.W.; Williams, A.C.; Barry, B.W. J. Control. Release, 1998, 50, 297. Stott, P.W.; Williams, A.C.; Barry, B.W. Int. J. Pharm., 2001, 219, 161. Fourie, L.; Breytenbach, J.C.; Du Plessis, J.; Goosen, C.; Swart, H.; Hadgraft, J. Int. J. Pharm., 2004, 279, 59. Vemuri, S.; Rhodes, C.T. Pharm. Acta Helv., 1995, 70, 95. Oh, Y.K.; Kim, M.Y.; Shin, J.Y.; Kim, T.W.; Yun, M.O.; Yang, S.J.; Choi, S.S.; Jung, W.W.; Kim, J.A.; Choi, H.G. J. Pharm. Pharmacol., 2006, 58, 161. Agarwal, R.; Katare, O.P.; Vyas, S.P. Int. J. Pharm., 2001, 228, 43. Mura, S.; Pirot, F.; Manconi, M.; Falson, F.; Fadda, A.M. J. Drug Target., 2007, 15, 101. Cevc, G.; Blume, G. Biochim. Biophys. Acta, 1992, 1104, 226. Elsayed, M.M.; Abdallah, O.Y.; Naggar, V.F.; Khalafallah, N.M. Int. J. Pharm., 2007, 332, 1. El Maghraby, G.M.; Williams, A.C.; Barry, B.W. J. Pharm. Pharmacol., 2001, 53, 1311. Cevc, G.; Blume, G. Biochim. Biophys. Acta, 2001, 1514, 191. Cevc, G.; Blume, G. Biochim. Biophys. Acta, 2003, 1614, 156. Jain, S.; Jain, P.; Umamaheshwari, R.B.; Jain, N.K. Drug Dev. Ind. Pharm., 2003, 29, 1013. Trotta, M.; Peira, E.; Carlotti, M.E.; Gallarate, M. Int. J. Pharm., 2004, 270, 119. Elsayed, M.M.; Abdallah, O.Y.; Naggar, V.F; Khalafallah, N.M. Pharmazie, 2007, 62, 133. Jain, S.; Tiwary, A.K.; Jain, N.K. Curr. Drug Deliv., 2006, 3, 157. Garg, M.; Mishra, D.; Agashe, H.; Jain, N.K. J. Pharm. Pharmacol., 2006, 58, 459. Rother, M.; Lavins, B.J.; Kneer, W.; Lehnhardt, K.; Seidel, E.J.; Mazgareanu, S. Ann. Rheum. Dis., 2007, 66, 1178. http://www.idea-ag.de/web/en/product_opportunities/index.html Cevc, G.; Mazgareanu, S.; Rother, M. Int. J. Pharm., 2008, 360, 29. Touitou, E.; Alkabes, M.; Dayan, N.; Eliaz, M. Pharm. Res., 1997, 14, S305. Touitou, E.; Dayan, N.; Bergelson, L.; Godin, B.; Eliaz, M. J. Control. Release, 2000, 65, 403. Touitou, E.; Godin, B.; Weiss, C. Drug Dev. Res., 2000, 50, 406. Dayan, N.; Touitou, E. Biomaterials, 2000, 21, 1879. Lodzki, M.; Godin, B.; Rakou, L.; Mechoulam, R.; Gallily, R.; Touitou, E. J. Control. Release, 2003, 93, 377. Touitou, E.; Godin, B. Therapy, 2007, 4, 465. Liu, J.; Hu, W.; Chen, H.; Ni, Q.; Xu, H.; Yang, X. Int. J. Pharm., 2007, 328, 191. Muller, R.H.; Radtke, M.; Wissing, S.A. Adv. Drug Deliv. Rev., 2002, 54, S131. Han, F.; Li, S.; Yin, R.; Shi, X.; Jia, Q. Drug Dev. Ind. Pharm., 2008, 34, 453.
186 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136]
Krzysztof Cal
Puglia, C.; Blasi, P.; Rizza, L.; Schoubben, A.; Bonina, F.; Rossi, C.; Ricci, M. Int. J. Pharm., 2008, 357, 295. Maia, C.S.; Mehnert, W.; Schafer-Korting, M. Int. J. Pharm., 2000, 196, 165. Jenning, V.; Schafer-Korting, M.; Gohla, S. J. Control. Release, 2000, 66, 115. Souto, E.B.; Muller, R.H. Int. J. Cosmet. Sci., 2008, 30, 157. Wissing, S.A.; Muller, R.H. Int. J. Pharm., 2003, 254, 65. Wissing, S.; Muller, R.H. Int. J. Pharm., 2002, 242, 373. Kreilgaard, M. Adv. Drug Deliv. Rev., 2002, 54, S77. El Maghraby, G.M. Int. J. Pharm., 2008, 355, 285. Kogan, A.; Garti, N. Adv. Colloid Interface Sci., 2006, 16, 123, 369. Barry, B.W. J. Control. Release, 1987, 6, 85. Marjukka Suhonen, T.; Bouwstra, J.A.; Urtti, A. J. Control. Release, 1999, 59, 149. Williams, A.C.; Barry, B.W. Adv. Drug Deliv. Rev., 2004, 56, 603. Thong, H.Y.; Zhai, H.; Maibach, H.I. Skin Pharmacol. Physiol., 2007, 20, 272. Friend, D.; Catz, P.; Heller, J.; Reid, J.; Baker, R. J. Control. Release, 1988, 7, 243. Pershing, L.K.; Lambert, L.D.; Knutson, K. Pharm. Res., 1990, 7, 170. Harborne, J.B.; Baxter, H. Phytochemical Dictionary. A Handbook of Bioactive Compounds from Plants, Taylor and Francis: Philadelphia, 1999. Nokhodchi, A.; Sharabiani, K.; Rashidi, M.R.; Ghafourian, T. Int. J. Pharm., 2007, 335, 97. Cal, K. Drug Dev. Ind. Pharm., DOI: 10.1080/03639040801974121 Aqil, M.; Ahad, A.; Sultana, Y.; Ali, A. Drug Discov. Today, 2007, 12, 1061. Sapra, B.; Jain, S.; Tiwary, A.K. AAPS J., 2008, 10, 120. Cal, K. Pharm. Technol. Europe, 2007, 19, 41. Matura, M.; Skold, M.; Borje, A.; Andersen, K.E.; Bruze, M.; Frosch, P.; Goossens, A.; Johansen, J.D.; Svedman, C.; White, I.R.; Karlberg, A.T. Contact Dermatitis, 2005, 52, 320. Hammer, K.A.; Carson, C.F., Riley, T.V.; Nielsen, J.B. Food Chem. Toxicol., 2006, 44, 616. Hausen, B.M. Dermatitis, 2004, 15, 213. Rastogi, S.C.; Johansen, J.D.; Menne, T. Contact Dermatitis, 1996, 34, 423. Schempp, C.M.; Schopf, E.; Simon, J.C. Hautarzt, 2002, 53, 93. Cal, K. Yakugaku Zasshi, 2006, 126, 307. Cal, K.; Sznitowska, M. J. Control. Release, 2003, 93, 369. Cal, K.; Kupiec, K.; Sznitowska, M. J. Dermatol. Sci., 2006, 41, 137. Cal, K. Planta Med., 2006, 72, 311. Cal, K. Arch. Dermatol. Res., 2006, 297, 311. Cal, K.; Krzyzaniak, M. J. Dermatol. Sci., 2006, 42, 265. Cal, K. Eur. J. Pharm. Biopharm., 2008, 68, 838. Cal, K. Int. J. Essent. Oil Ther., 2007, 1, 1. Touitou, E. Expert Opin. Biol. Ther., 2002, 2, 723. http://www.vyteris.com/home/Our_Products/Lidosite.php Bain, K.T. J. Opioid Manag., 2006, 2, 314. Power, I. Brit. J. Anaesth., 2007, 98, 4. Okuno, M.; Takahashi, Y.; Isowa, K.; Machida, Y. Biol. Pharm. Bull., 2008, 31, 487. Gurney, B.; Wascher, D.; Eaton, L.; Benesh, E.; Lucak, J. J. Orthop. Sports Phys. Ther., 2008, 38, 238. Davarian, S.; Kalantari, K.K.; Rezasoltani, A.; Rahimi, A. Australasian J. Dermatol., 2008, 49, 2. Ogura, M.; Paliwal, S.; Mitragotri, S. Adv. Drug Deliv. Rev., 2008, 60, 1218. Sintova, A.C.; Krymberka, I.; Danielb, D.; Hannanb, T.; Sohnb, Z. J. Control. Release, 2003, 89, 311. Doukas, A.G.; Kollias, N. Adv. Drug Deliv. Rev., 2004, 56, 559. Pearton, M.; Allender, C.; Brain, K.; Anstey, A.; Gateley, C.; Wilke, N.; Morrissey, A.; Birchall, J. Pharm. Res., 2008, 25, 407. Lee, J.W.; Park, J.H.; Prausnitz, M.R. Biomaterials, 2008, 29, 2113. Shergold, O.A.; Fleck, N.A.; King, T.S. J. Biomech., 2006, 39, 2593. Cal, K.; Centkowska, K. Eur. J. Pharm. Biopharm., 2008, 68, 467. Loftsson, T.; Brewster, M.E. J. Pharm. Sci., 1996, 85, 1017.
Across Skin Barrier [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 187
Rowe, R.C.; Sheskey, P.J., Weller, P.J. Handbook of Pharmaceutical Excipients, Pharmaceutical Press: London, Chicago, 2003. Fromming, K.H.; Szejtli, J. Topics in Inclusion Science, Vol. 5: Cyclodextrins in Pharmacy, Kluwer Academic Publishers: Dortrecht, Boston, London, 1994. Rubinstein, M.H. Pharmaceutical Technology, Drug Stability, Ellis Horwood Ltd, John Wiley and Sons: New York, Chichester, Brisbane, Toronto, 1989. Del Valle, E.M.M. Process. Biochem., 2004, 39, 1033. Matsuda, H.; Arima, H. Adv. Drug Deliv. Rev., 1999, 36, 81. Martins, P.S.; Ochoa, R.; Pimenta, A.M.C.; Ferreira, L.A.M.; Melo, A.L.; Da Silva, J.B.B.; Sinisterra, R.D.; Demicheli, C.; Frezard, F. Int. J. Pharm., 2006, 325, 39. Loftsson, T.; Masson, M.; Brewster, M.E. J. Pharm. Sci., 2004, 93, 1091. Duan, M.S.; Zhao, N.; Ossurardottir, I.B.; Thorsteinsson, T.; Loftsson, T. Int. J. Pharm., 2005, 297, 213. Loftsson, T.; Matthiasson, K.; Masson, M. Int. J. Pharm., 2003, 262, 101. Piel, G.; Moutard, S.; Uhoda, E.; Pilard, F.; Pierard, G.E.; Perly, B.; Delattre, L.; Evrard, B. Eur. J. Pharm. Biopharm., 2004, 57, 479. Loftsson, T.; Masson, M. Int. J. Pharm., 2001, 225, 15. Shaker, D.S.; Ghanem, A.H.; Li, S.K.; Warner, K.S.; Hashem, F.M.; Higuchi, W.I. Int. J. Pharm., 2003, 253, 1. Legendre, J.Y.; Rault, I.; Petit, A.; Luijten, W.; Demuynck, I.; Horvath, S.; Ginot, Y.M.; Cuine, A. Eur. J. Pharm. Sci., 1995, 3, 311. Asai, K.; Morishita, M.; Katsuta, H.; Hosoda, S.; Shinomiya, K.; Noro, M.; Nagai, T.; Takayama, K. Int. J. Pharm., 2002, 246, 25. Szejtli, J. J. Incl. Phenom. Macrocycl. Chem., 2005, 52, 1. Godwin, D.A.; Wiley, C.J.; Felton, L.A. Eur. J. Pharm. Biopharm., 2006, 62, 85. Anadolu, R.Y.; Sen, T.; Tarimici, N.; Birol, A.; Erdem, C. J. Eur. Acad. Dermatol. Venerol., 2004, 18, 1. Rode, T.; Frauen, M.; Muller, B.W.; Dusing, H.J.; Schonrock, U.; Mundt, C.; Wenck, H. Eur. J. Pharm. Biopharm., 2003, 55, 191. Lopez, R.F.V.; Collet, J.H.; Vitoria, M.; Bentley, L.B. Int. J. Pharm., 2000, 200, 127. Sirdevei, S.; Rao Diwan, P.V. Eur. J. Pharm. Biopharm., 2002, 54, 151. Bilensoy, E.; Rouf, M.A.; Vural, I.; Sen, M.; Hincal, A.A. AAPS Pharm. Sci. Tech., 2006, 7, E38. Proniuk, S.; Liederer, B.M.; Dixon, S.E.; Rein, J.A.; Kallen, M.A.; Blanchard, J. J. Pharm. Sci., 2002, 91, 101. Scalia, S.; Tursilli, R.; Sala, N.; Iannuccelli, V. Int. J. Pharm., 2006, 320, 79. Montassier, P.; Duchene, D.; Poelman M.C. J. Incl. Phenom. Molec. Recogn. Chem., 1998, 31, 213. Montassier, P.; Duchene, D.; Poelman, M.C. Int. J. Pharm., 1997, 153, 199. Jug, M.; Becirevic-Lacan, M.; Kwokal, A.; Cetina-Cizmek, B. Acta Pharm., 2005, 55, 223. Kaur, I.P.; Smitha, R.; Aggarwal, D.; Kapil, M. Int. J. Pharm., 2002, 248, 1. Arima, H.; Miyaji, T.; Irie, T.; Hirayama, F.; Uekama, K. Chem. Pharm. Bull., 1996, 44, 582. Siguroardottir, A.M.; Loftsson, T. Int. J. Pharm., 1995, 126, 73. Babu, R.J.; Pandit, J.K. Int. J. Pharm., 2004, 271, 155. Schoch, C.; Bizec, J.C.; Kis, G. J. Incl. Phenom. Macrocycl. Chem., 2007, 57, 391. Ventura, C.A.; Tommasini, S.; Falcone, A.; Giannone, I., Paolino, D.; Sdrafkakis, V.; Mondello, M.R.; Puglisi, G. Int. J. Pharm., 2006, 314, 37. Vollmer, U.; Muller, B.W.; Peeters, J.; Mesens, J.; Wiffert, B.; Peters, T. J. Pharm. Pharmacol., 1994, 46, 19. Bentley, M.V.L.B.; Vianna, R.F.; Wilson, S.; Collett, J.H. J. Pharm. Pharmacol., 1997, 49, 397. Valjakka-Koskela, R.; Hirvonen, J.; Monkkonem, J.; Kiessvaara, J.; Antila, S.; Lehtonen, L.; Urtti, A. Eur. J. Pharm. Sci., 2000, 11, 343. Preis, A.; Mehnert, W.; Fromming, K.H. Pharmazie, 1995, 50, 121. Simeoni, S.; Scalia, S.; Benson, H.A.E. Int. J. Pharm., 2004, 280, 163. Simeoni, S.; Scalia, S.; Tursilli, R.; Benson, H. J. Incl. Phenom. Macrocycl. Chem., 2006, 54, 275. Skalko-Basnet, N.; Pavelic, Z.; Becirevic-Lacan, M. Drug Dev. Ind. Pharm., 2000, 26, 1279. Boulmedarat, L.; Piel, G.; Bochot, A.; Lesieur, S.; Delattre, L.; Fattal, E. Pharm. Res., 2005, 22, 962.
188 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188]
Krzysztof Cal
Hartel, S.; Diehl, H.A.; Ojeda, S.F. Anal. Biochem., 1998, 258, 277. Piel, G.; Piette, M.; Barillaro, V.; Castagne, D.; Evrard, B.; Delattre, L. Int. J. Pharm., 2006, 312, 75. Maestrelli, F.; Gonzalez-Rodriguez, M.L.; Rabasco, A.; Mura, P. Int. J. Pharm., 2006, 312, 53. Fundueanu, G.; Constantin, M.; Dalpiaz, A.; Bortolotti, F.; Cortesi, R.; Ascenzi, P.; Menegatti, E. Biomaterials, 2004, 25, 159. Mainardes, R.M.; Cocenza Urban, M.C.; Cinto, P.C.; Chaud, M.V.; Evangelista, R.C.; Daflon Gremiao, M.P. Curr. Drug Deliv., 2006, 3, 275. Jug, M.; Becirevic-Lacan, M.; Cetina-Cizmek B.; Horvat, M. Pharmazie, 2004, 59, 686. Filipovic-Gracic, J.; Voinovich, D.; Moneghini, M.; Becirevic-Lacan, M.; Magarotto, L.; Jaslenjak, I. J. Pharm. Sci., 2000, 9, 373. Murthy, S.N.; Zhao, Y.L.; Sen, A.; Hui, S.W. J. Control. Release, 2004, 99, 393. Dias, M.M.R.; Raghavan, S.L.; Pellett, M.A.; Hadgraft, J. Int. J. Pharm., 2003, 263, 173. Doliwa, A.; Santoyo, S.; Ygartua, P. Int. J. Pharm., 2001, 229, 37. Yamakawa, T.; Nishimura, S. J. Control. Release, 2003, 86, 101. Nandi, I.; Bateson, M.; Bari, M.; Joshi, H.N. AAPS Pharm. Sci. Tech., 2003, 4, E1.
Frontiers in Drug Design & Discovery, 2009, 4, 189-227
189
The Tape Stripping Method as a Valuable Tool for Evaluating Topical Applied Compounds J.J. Escobar-Chávez1,2,*, L.M. Melgoza-Contreras2, M. López-Cervantes1, D. Quintanar-Guerrero1 and A. Ganem-Quintanar1 1
División de Estudios de Posgrado (Tecnología Farmacéutica), Facultad de Estudios Superiores Cuautitlán-Universidad Nacional Autónoma de México, Cuautitlán Izcalli, Estado de México, México 54740 and 2Departamento de Sistemas Biológicos, Universidad Autónoma Metropolitana-Xochimilco, Calzada del Hueso 1100, Colonia Villa Quietud, México D.F. 04960, México Abstract: Quantification of drugs within the skin is essential for topical and transdermal delivery research. Over the last two decades, horizontal sectioning, consisting of tape stripping throughout the stratum corneum, has become one of the traditional investigative techniques. Tape stripping of human stratum corneum is widely used as a method for studying the kinetics and penetration depth of drugs. The Food and Drug Administration released a draft guidance proposing a Dermatopharmacokinetic method for evaluating bioavailability and/or bioequivalence of topical dermatological drug products. As specified in this document, the method measures topically applied drug levels in the outermost layer of the skin, the stratum corneum, as a function of time post-application and postremoval of the formulation, so as to generate a stratum corneum concentration versus time profile. The stratum corneum is collected by successive application and removal of adhesive tape providing a minimally invasive technique by which the drug’s concentration in the skin can be determined. The Dermatopharmacokinetic method assumes that: (i) in normal circumstances, the stratum corneum is the rate-determining barrier to percutaneous absorption, (ii) the stratum corneum concentration of drug is directly related to that which diffuses into the underlying viable epidermis, and (iii) Stratum corneum drug levels are more useful and relevant for assessing local, dermatological efficacy than plasma concentrations. This paper shows the applications of the tape stripping technique to evaluate drug penetration through the skin as well as stratum corneum composition and physiology, underlining its versatile application in the area of topical and transdermal drugs.
*Corresponding Author: Tel: (525) 56.23.20.65; Fax: (525) 58.93.86.75; E-mail:
[email protected]
Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
190 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
1. INTRODUCTION Tape stripping (TS) with adhesive tape is a widely accepted and used method to examine the localization and distribution of substances within the stratum corneum (SC) [1-7]. This is a minimally invasive technique to sequentially remove SC by the repeated application of appropriate adhesive tapes [8]. This technique can be used to investigate SC cohesion in vivo by quantifying the amount of SC removed [9]. Today, weighing with precision balances is the most frequently used method to determine the amount of SC removed on a tape strip. The method is also used to provide information about the kinetics of transdermal drug delivery, offering an apparently easy and quite non-invasive methodology for skin tissue sampling, and is the basis of the FDA’s so-called dermatopharmacokinetic (DPK) approach to the assessment of topical bioavailability and bioequivalence [10]. However, validation and optimization of the procedure have not come quickly and the proposed guidance document has been withdrawn for re-evaluation. More recent work has addressed at least some of the important limitations of the DPK approach [11-13] and has proposed modifications in order to incorporate it into an improved protocol. A number of excellent reviews that have been published contain detailed discussions concerning many aspects of the TS technique [14-16]. The present review shows an updated overview of the use of the TS technique in the pharmaceutical field, specifically in the area of topical and transdermal drug delivery. This focus is justified due to the magnitude of the experimental data available with the use of this technique. The use of the TS technique in experimental medicine and pharmaceutical sciences has a long history. The skin is the largest organ of the body [17-19], accounting for more than 10% of body mass, and the one that enables the body to interact more intimately with its environment. Essentially, the skin consists of four layers: The SC, that is the outer layer of the skin (non-viable epidermis), and forms the rate-controlling barrier for diffusion for almost all compounds. It is composed of dead flattened, keratin-rich cells, the corneocytes. These dense cells are surrounded by a complex mixture of intercellular lipids, namely, ceramides, free fatty acids, cholesterol, and cholesterol sulphate. Their most important feature is that they are structured as ordered bilayer arrays [20]. The predominant diffusional path for a molecule crossing the SC appears to be intercellular [21-23]. The other layers are: the remaining layers of the epidermis (viable epidermis), the dermis, and the subcutaneous tissues (Fig. 1). There are also several associated appendages: hair follicles, sweat ducts, apocrine glands and nails. In a general context, the skin’s functions may be classified as protective, homeostasis maintaining functions, or sensing [25]. The importance of the protective and homeostatic role of the skin is illustrated in one context by its barrier property. This allows survival in an environment of variable temperature and water content and presence of environmental dangers, such as chemicals, bacteria, allergens, fungi and radiation. In a second context, the skin is a major organ for maintaining the body’s homeostasis, especially in terms of its composition, heat regulation, blood pressure control and excretory roles [25]. Third, the skin is a major sensory organ in terms of sensing environmental influences, such as heat, pressure, pain, allergens and microorganism entry. Finally the skin is an organ in a continuous state of regeneration and repair. To perform each of these functions, the skin must be tough, robust, and flexible, with an effective communication between each of its intrinsic components.
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 191
Fig. (1). Layers of human skin.
Many agents are applied to the skin either deliberately or accidentally, with either beneficial or deleterious outcomes. The main interest in dermal absorption assessment is related to: a) Local effects in dermatology (e.g., corticosteroids for dermatitis); b) transport through the skin seeking a systemic effect (e.g., nicotine patches, hormonal drug patches, etc.); c) surface effects (e.g., sunscreens, cosmetics, and anti-infectives) [26]; d) targeting of deeper tissues (e.g., nonsteroidal anti-inflammatory agents) [7, 27-35]; and e) unwanted absorption (e.g., solvents in the workplace, pesticides or allergens) [36,37]. Fig. (2) summarizes the process of percutaneous absorption and the possible routes by which a substance penetrates through the skin. The skin became popular as a potential site for systemic drug delivery, on the one hand, because of the possibility of avoiding the problems of stomach emptying, pH effects, enzyme deactivation associated with gastrointestinal passage, and hepatic firstpass metabolism; and on the other hand, due to its capability to enable input control. 2. IN VIVO METHODS FOR PERCUTANEOUS ABSORPTION MEASUREMENT There is a persistent belief that skin viability has little importance in percutaneous absorption. This concept of skin as passive membrane has led to the domination of the study of percutaneous absorption by laws of mass action and physical diffusion. This concept has also led researchers to use skin excised from cadavers (human and animal)
192 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
Fig. (2). Processes of percutaneous absorption and transdermal delivery.
and the physically (e.g., by freezing or heat separation) and chemically isolated skin sheets or sections determining chemical diffusion across these treated tissues [38]. Fig. (3) shows a schematic representation of the penetration of a drug by passive diffusion throughout the SC.
Fig. (3). Schematic representation of the way by which a drug or chemical permeate throughout the SC by passive diffusion.
The need of studying percutaneous absorption has its reality in dermatotoxicity, by which compounds pose a threat to human health, and to dermatopharmacology, for
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 193
which drugs need to be delivered into and through the skin to treat diseases both locally and systemically. Almost all the molecules of interest and concern in dermatotoxicology and dermatopharmacology are lipophilic. The SC, the barrier to percutaneous absorption, is a lipid saturated tissue that is like a sink to topically applied lipophilic materials. The chemical and physical properties of the topical vehicle and the barrier/sink properties of the living SC determine the initial absorption of compounds into the skin. The vitality of the living skin will, in part, determine the metabolism, distribution and excretion of the compounds through the skin and the body [38]. Some of the in vivo methods used to evaluate percutaneous penetration/absorption are described in the following paragraphs. 2.1. Radioactivity in Excretions In vivo percutaneous absorption is usually determined by the indirect method of measuring radioactivity in excreta after topical application of a labeled compound. In human studies chemical plasma levels are extremely low after topical application, often below assay detection level, so it is necessary to use tracers methodology. The compound, usually labeled with 14C or tritium, is applied and the total amount of radioactivity excreted in urine plus feces is determined. The amount of radioactivity retained in the body or excreted by some route not assayed (CO2, sweat) is corrected by determining the amount of radioactivity excreted after parenteral administration. This final amount of radioactivity is then expressed as the percentage of applied dose that was absorbed [39-41]. The equation used to determine percutaneous absorption is: % = 100 (Total radioactivity after topical administration/Total radioactivity after parenteral administration) The limitation on determining percutaneous absorption from urinary or fecal radioactivity, or both, is that the methodology does not account for skin metabolism. 2.2. Radioactivity in Blood Plasma radioactivity can be measured and the percutaneous absorption determined by the ratio of the AUC from the plasma concentration versus time curves following topical and intravenous administration [42]. This method has given results similar to those obtained from urinary excretion [43], and the same limitations discussed from excreta also apply here. 2.3. Surface Recovery This method consists in determining the loss of material from the surface as it penetrates into the skin. Skin recovery from an ointment or solution application is difficult because total recovery of compound from the skin is never assured. With topical application of a transdermal delivery device, the total unit can be removed from the skin and the residual amount of drug in the device determined. It is assumed that the difference between applied dose and the residual dose is the amount of drug absorbed. 2.4. Surface Disappearance Related to the method above mentioned, it is possible to monitor the disappearance of 14C from the surface of skin using appropriate instrumentation. The limitation on this
194 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
methodology is that the disappearance is due both to movement of 14C-labeled chemical into the skin and to the quenching effect of the skin on the rays bouncing back to the instrument. The degree of quench of chemical in the various cell layers of the skin has not been defined. 2.5. Biological/Pharmacological Response Biological assay is substituted for a chemical assay, and absorption is estimated. An obvious disadvantage is that the biological responses are limited to compounds that elicit responses that can be measured easily and accurately. This method is more qualitative than quantitative [44]. 2.6. Tape Stripping Method The TS method determines the concentration of chemical in the SC at the end of a short application period and, by linear extrapolation, predicts the percutaneous absorption of that chemical for longer application periods. The chemical is applied to the skin of animal or humans, after a period of time the excess of formulation is wiped from the surface and the SC is removed by successive tape application and removal. The tapes are assayed for chemical content. The major advantages of this method are: 1) The elimitation of urinary or fecal excretions to determine absorption, and 2) The applicability of nonradiolabeled determination to percutaneous absorption, because in general, the TSs contain adequate chemical concentrations for nonlabeled assay methodology. This is an interesting methodology for which more research is needed to establish limitations [45-47]. 2.7. Absolute Topical Bioavailability The only way to determine the absolute bioavailability of a topically applied compound is to measure the compound by a specific assay in blood or urine after topical and intravenous administration. This is extremely difficult to do in plasma because concentrations after topical administration are often low. However, as advances in analytical methodology bring more sensitive assays, estimates of absolute topical bioavailability are becoming more available. 2.8. Real time In Vivo Bioavailability This method determines the bioavailability of organic solvents following dermal exposure. Breath analysis is used to obtain real-time measurements of volatile organics in expired air following exposure. Human volunteers and animal breathe fresh air via a new breath-inlet system that allows for continuous real time analysis of undiluted exhaled air [48]. 3. IN VITRO TECHNIQUES TO DETERMINE PERCUTANEOUS ABSORPTION In vitro percutaneous absorption methods have become widely used for measuring the absorption of compounds that come in contact with skin. Safety evaluations of toxic chemicals frequently rely on in vitro studies for human permeation data. Animal data
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 195
must be used cautiously for estimating human absorption due to differences in barrier properties of animal and human skin [49]. In vitro absorption studies can also be used to measure skin metabolism if viable skin is obtained for the study and if the viability is maintained in the diffusion cells [50]. The in vitro system allows for the isolation of skin so that the metabolism of the organ can be distinguished from systemic metabolism. In vitro protocols generally allow the use of either the flow-through or static diffusion cell (Fig. 4). Diffusion cells should be made of a material resistant to binding of test material, such a glass or Teflon.
Fig. (4). Schematic overview of the Franz diffusion cell for in vitro skin permeation studies.
196 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
4. TAPE STRIPPING TECHNIQUE OVERVIEW The simplest method for reducing the barrier imposed by the SC is to remove it. Theoretically, an adhesive tape removes a layer of corneocytes. In vivo, removal of the SC by TS is performed by the repeated application of adhesive tapes to the skin’s surface. In Fig. (5) we can observe a detailed procedure of the TS technique. It has been found that on the flexor surface of the forearm, about 30 tape strips are needed to strip off most of the horny layer [8]. Multiple strips remove a substantial skin barrier, as evidenced by 20 to 25-fold increases in transepidermal water loss (TEWL) [51]. Usually, the amount of SC removed by TS is not linearly proportional to the number of tapes
Fig. (5). The tape stripping technique procedure.
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 197
removed [8]. TS appears to be simple and easy to perform [52-53], however there are different parameters that can influence the quantity of SC removed by a piece of tape, and these include TS mode [51, 54], skin hydration, cohesion between cells (which increases with SC’s depth), the body site and inter-individual differences [9, 55]. The impact of these factors has been frequently investigated [8, 51-57]. After its description by Pinkus [8], TS has become a standard method in dermatological research [58]. This method can be used to obtain a more susceptible skin, e.g., prior to the application of an irritant [59] or an allergen [60-62]. Similarly, TS is performed to induce a defined disruption of the water barrier, e.g., to evaluate the effect of a subsequently applied skin care product in barrier restoration [63]. It may be also used to obtain cells for mycological culture [64,65] or to investigate SC quality [66]. In dermatopharmacology, the SC barrier function [63,67] and the bioavailability and bioequivalence of topical drugs [52,68-70] can be evaluated with the use of this technique [71,72]. Because of the limited systemic absorption of topical products, bioavailability and/or bioequivalence studies used end-point parameters or surrogate pharmacodynamic markers. However, if a drug does not penetrate or partition into de SC, a pharmacodynamic activity will not take place. The skin stripping methodology allows the determination of the uptake and elimination profile of topically applied drugs. TS appears to be simple and easy to perform. However, there are parameters which have to be defined, as they may change the outcome. Because various brands of tape differ in shape, surface area, composition and adhesive properties, the influence of the tape brand on the outcome seems apparent [51,54]. Other parameters which influence the procedure can be subsumed in the intrinsic properties of the SC [51]. Although these properties are often investigated, little is known about the anatomical sites (intrinsic factor) as well as the pressure with which the tape is applied on to the skin, the duration of pressure and the removal process (extrinsic factors) influencing SC removal. In the case of bioequivalence studies, topical bioavailability can be estimated from the drug concentration within the SC, which is expected to be related to the drug concentration at the target site (i.e., usually viable epidermis or dermis) since the SC is the rate limiting barrier for percutaneous absorption. Similarly to the determination of the drug concentration in blood and/or urine as surrogate for the concentration at the target tissue, the determination of the drug concentration in the SC may serve as a surrogate for the concentration in the viable (epi-)dermis [71]. A typical profile obtained from a skin permeation study with sodium naproxen is shown in Fig. (6). TS, which enables the removal of the SC layer by layer, is a useful DPK technique for the assessment of drug amounts in SC as a function of time [72]. 5. APPLICATIONS OF THE TAPE-STRIPPING TECHNIQUE Removal by TS of the outermost skin layer, the SC, has become a common practice in recent years [7,36,37,56,73,74]. The determination of the kinetics and penetration depth of different kind of permeants by tracing the concentration profiles in SC, has been facilitated by the use of the virtually non-invasive method of SC stripping with adhesive tape [1,2,5,56,57]. For this reason, TS also offers the possibility of evaluating bioequivalence of topical dermatological dosage forms [5]. DPK characterization of active drugs in human volunteers has been suggested to be able to replace comparative clinical trials as a means of documenting bioequivalence [72].
198 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
Fig. (6). Penetration profiles across human SC of Sodium naproxen formulated in PF-127 gels with different penetration enhancers [Azone®-Transcutol® mixture, Transcutol® applying an infinite dose, and a film with Azone®-Transcutol® mixture, Transcutol® ] (Mean± SD; n =6).
Moreover, in vitro methods are encouraged by regulatory agencies regarding the provision of percutaneous absorption data for drugs, pesticides and cosmetics [75]. All these points are emphasized in Table 1, which summarizes the research with the TS technique to determine the kinetics and penetration depth of permeants (drugs and toxic chemicals) [7, 26, 35, 37, 57, 76-110], in order to evaluate the factors that influence the physiology of the SC [55, 57, 118,119], to determine the composition of the SC [121], superficial infections in the skin [123, 124], and evaluate skin regeneration [125, 126], etc. 5.1. Kinetics and Penetration Depth of Drugs 5.1.1. Analgesic and Anti-Inflammatory Drugs Arima et al. [76] investigated the effect of hydroxypropyl-P-cyclodextrin (HP-P-CD) on the cutaneous penetration and activation of ethyl 4-biphenylyl acetate (EBA), a prodrug of the non-steroidal anti-inflammatory drug 4-biphenylylacetic acid (BPAA), from hydrophilic ointment, using hairless mouse skin in vitro. When the hydrophilic ointment containing a complex of EBA with HP-P-CD was applied to full-thickness skin, HP-PCD facilitated the penetration of EBA into the skin, the BPAA flux through the tapestripped skin was greater than that through full-thickness skin, while the activation of the prodrug in the skin was slowed by TS. Their results suggest that the enhancing effect of HP-P-CD on the cutaneous penetration of EBA would be largely attributed to an increase in the effective concentration of EBA in the ointment. Curdy et al. [35] administered piroxicam from a commercially available gel to human volunteers, both passively and under the application of an iontophoretic current. After treatment, the SC at the site of application was progressively tape-stripped and piroxicam transport into the membrane was assessed by UV-analysis of drug extracted from the tape-strips. Current application enhanced drug uptake into the SC, as indicated by both increased piroxicam concentrations in the horny layer and detectable concentrations at greater depths in the membrane. The total amount of drug recovered in the SC post-iontophoresis was significantly higher than that found following passive diffusion for each application time.
TS Method as a Valuable Tool
Table 1.
Frontiers in Drug Design & Discovery, 2009, Vol. 4 199
Research on the Tape Stripping Technique as a Method to Determine Skin Penetration of Different Kind of Permeants
Outcome
Author (Ref.) Year
Effect of Azone® and Transcutol® on skin permeation of sodium naproxen formulated in PF-127 gels.
The combination of Azone® and Transcutol® in PF-127 gels enhanced sodium naproxen penetration, with up to twofold enhancement ratios compared with the formulation containing Transcutol® only.
Escobar-Chávez et al. [7], 2005
Administration of piroxicam from a commercially available gel to human volunteers, both passively and under the application of an iontophoretic current.
The total amount of drug recovered in the SC post-iontophoresis by TS was significantly higher than that found following passive diffusion for each application time.
Curdy et al. [35], 2001
Effect of hydroxypropyl-P-cyclodextrin (HP-P-CD) on the cutaneous penetration and activation of ethyl 4-biphenylyl acetate (EBA), a prodrug of nonsteroidal anti-inflammatory drug 4biphenylylacetic acid (BPAA), from hydrophilic ointment, using hairless mouse skin in vitro.
The enhancing effect of HP-P-CD on the cutaneous penetration of EBA would be largely attributable to an increase in the effective concentration of EBA in the ointment.
Arima et al. [76], 1998
Production and characterization of monoleine (MO) dispersions as drug delivery systems for indomethacin.
Reflectance spectroscopy demonstrated that indomethacin incorporated into MO dispersions can be released in a prolonged fashion. TS experiments corroborated this finding.
Unilaminar films of Eudragit E-100 prepared from naproxen-loaded nanoparticles vs. conventional films.
In vivo penetration studies showed no statistical differences for the penetrated amount of naproxen across the SC and the depth of penetration for the two films. The films formulated from nanoparticle dispersions were shown to be effective for the transdermal administration of naproxen.
Ganem-Quintanar et al. [78], 2006
Investigation of pig ear skin as a surrogate for human skin in the assessment of topical drug bioavailability by sequential TS of the SC.
Pig ear skin ex vivo is promising as a tool for topical formulation evaluation and optimization.
Herkenne et al. [79], 2006
Research 1) Kinetics and penetration depth of drugs
Esposito et al. [77], 2005
200 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
(Table 1) contd....
Outcome
Author (Ref.) Year
Determination of DPK parameters describing the rate and extent of delivery into the skin of Ibuprofen in the ventral forearms of human volunteers.
Prediction and experimental tests agreed satisfactorily suggesting that objective and quantitative information, to characterize topical drug bioavailability, can be obtained from this approach.
Herkenne et al. [80], 2007
Examination of the diffusion of copper through human SC in vivo following application of the metal as powder on the volar forearm for periods of up to 72 h.
Copper will oxidize and may penetrate the stratum corneum after forming an ion pair with skin exudates. The rate of reaction seems to depend on contact time and oxygen availability. A marked inter-individual difference was observed in baseline values and the amounts of copper absorbed.
Hostnek et al. [81], 2006
Comparison of the bioavailability of ketoprofen in a photostabilised gel formulation without photoprotection using a new DPK TS model and an established ex vivo penetration method using human skin.
The comparison of the amount of ketoprofen in the skin after 45 min with the amount penetrated through the excised skin during 36 h, suggests a change in the thermodynamic activity of ketoprofen during exposure.
Lodén et al. [82], 2004
Penetration kinetics of SLs (sesquiterpene lactones) in Arnica montana preparations, by using a stripping method with adhesive tape and pig skin as a model.
Gel preparation showed a decrease in penetration rate, whereas the penetration rate of ointments remained constant over time. The total amount of SLs penetrated depends only on the kind of formulation and the SLs-content, but not on SLs composition or on the extraction agent used.
Wagner Steffen et al. [83], 2006
Penetration experiments investigating several incubation times with three different skin flaps, using the Saarbruecken penetration model and the lipophilic model drug flufenamic acid.
A direct linear correlation was found between the SC/water partition coefficients and the drug amounts penetrated into the SC for all time intervals tested.
Wagner Heike et al. [84], 2002
Effect of dose and application frequency on the penetration of triamcinolone acetonide (TACA) into human SC in vivo.
Considerable TACA amounts were retained within the SC, independently of dose and application frequency.
Pallenda et al. [85], 2006
Research 1) Kinetics and penetration depth of drugs
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 201 (Table 1) contd....
Outcome
Author (Ref.) Year
Sustained bactericidal activity of chlorhexidine base loaded poly(caprolactone) nanocapsules against Staphylococcus epidermidis inoculated onto porcine ear skin.
Topical application of chlorhexidine baseloaded positively charged nanocapsules in an aqueous gel achieved a sustained release of bactericide against Staphylococcus epidermidis for at least 8 h.
Lboutounne et al. [86], 2002
Design of an all-trans retinoic acid (RA) topical release system that modifies drug diffusion parameters in the vehicle and the skin, in order to reduce the systemic absorption and side-effects associated with the topical application of the drug to the skin.
RA encapsulation not only prolongs drug release, but also promotes drug retention in viable skin.
Fresno-Contreras et al. [87], 2005
Behaviour of a skin bioadhesive film containing lidocaine in vitro and in vivo.
In vivo experiments with TS indicated that the presence of water during film application is essential to achieve not only the proper adhesion, but also an effective accumulation.
Padula et al. [88], 2003
Keratolytic efficacy of topical preparations containing salicylic acid in humans by TS, and quantification of SC removal by protein analysis.
TS combined with protein analysis was sensitive in detecting the keratolytic effect of salicylic acid within hours of application.
Bashir et al. [89], 2005
Novel synthetic technique to synthesize the co-drug retinyl ascorbate (RA-AsA) ester from all-trans-retinyl chloride (RA) and l-ascorbic acid (AsA) suspended in ethanol at low temperature.
The data suggest the potential value of RA-AsA co-drug for treating damage to skin resulting from UV-induced production of free radicals.
Abdulmajed et al. [90], 2004
Glycerol replacement corrects each of the defects in aquaporine-3 (AQP3)-null mice.
The findings establish a scientific basis for the >200-yr-old empirical practice of including glycerol in cosmetic and medicinal skin formulations due to its influence on water retention and the mechanical and biosynthetic functions of the SC.
Hara et al. [91], 2003
Determination of the cutaneous bioavailability and bioequivalence of topically applied drugs in vivo. The procedure uses serial TS and TEWL measurements to quantify the thickness of the removed SC and to determine the intact membrane thickness.
Integration of the concentration profile over the entire SC thickness provides a measure of the cutaneous bioavailability and hence can be used to assess the bioequivalence of topically applied drugs.
Kalia et al. [92], 2001
Research 1) Kinetics and penetration depth of drugs
202 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
(Table 1) contd....
Outcome
Author (Ref.) Year
Contribution of SC barrier and microvascular perfusion in determining dermal tissue levels of hydrophilic drugs (aciclovir and penciclovir) in vivo.
There was no relationship between fibre depth and the amount of drug dialysed, which suggests free movement of antiviral drug on reaching the aqueous environment of the dermis.
Morgan et al. [93], 2003
Determination of the in-vitro dermal delivery of a new class of lipophilic, highly potent and uniquely selective anti-VZV nucleoside analogue compared with aciclovir.
Topical delivery of these compounds is highly promising as a new first line treatment for VZV infections.
Jarvis et al. [94], 2004
Effect of CpG oligodeoxynucleotide (CpG-ODN) on the immune response to an antigen applied to tape-stripped mouse skin by evaluating the production of cytokines and Ig isotypes.
Administration of CpG ODN through skin is a simple strategy for patients with diseases like atopic dermatitits, which is characterized by Th2dominated inflammation.
Inoue et al. [96,97], 2005, 2006
Administration of human immunodeficiency virus type-1 (HIV-1) DNA vaccine with cytokine-expressing plasmids to the skin of mice by a new topical application technique involving prior elimination of keratinocytes using TS.
Topical application is an efficient route for DNA vaccine administration and that the immune response may be induced by DNA plasmids taken in by DCs, Langerhans cells, or others such as antigen-presenting cells.
Liu et al. [98], 2002
Effect of sucrose esters (sucrose oleate and sucrose laureate in water or Transcutol®, TC) on the SC barrier properties in vivo. Impact of these molecules on the in vivo percutaneous penetration of 4-hydroxybenzonitrile (4-HB).
A combination of sucrose esters (oleate or laureate) and TC is able to temporarily alter the SC barrier properties, thereby promoting 4-HB penetration.
Ayala-Bravo et al. [99], 2003
Absorption of 4-cyanophenol (4CP) in humans using TS experiments to assess the conditions under which diffusion alters tape stripping results.
Chemical concentrations in TSs can be affected by diffusion during tape stripping, but with tTS < 0.2 tlag and an exposure time > 0.3 tlag , TS concentrations are not significantly affected by tTS.
Reddy et al. [100], 2002
Development of a sensitive method for the determination of polyethylene glycols with different molecular weights (MW) in the human SC obtained by TS.
The method showed to be suitable for studying permeability in normal and impaired skin with respect to MW in the range of 150–600 Da.
Jakasa et al. [101,102], 2004, 2007
Research 1) Kinetics and penetration depth of drugs
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 203 (Table 1) contd....
Outcome
Author (Ref.) Year
Dependence of permeability on molecular weight with different forms of barrier disruption.
Irrespectively of the form of barrier disruption, not only higher amounts, but also more varieties of chemicals (larger molecules) may penetrate into the skin in the presence of a compromised barrier, compared with normal skin.
Tsai et al. [103], 2003
Measurement of the status of skin surface hydration of uraemic patients with the corneometer and skin surface hydrometer, the functional capacity and the urea concentration in SC by TS technique, as well as the response of eccrine sweat gland to sudorific agent (0.05% pilocarpine HCL) in 18 age-matched haemodialysis patients and 10 healthy volunteers.
The functional abnormalities of eccrine sweat glands may be account for dry skin in uraemic patients at least in part, but there is no correlation between xerosis and pruritus.
Park et al. [104], 2001
Penetration of octyl methoxycinnamate (OMC) encapsulated in poly(caprolactone) nanoparticles, into and across porcine ear skin in vitro.
Nanoparticulate encapsulation of OMC increased its “availability” within the SC.
Alvarez-Román et al. [105], 2004
In vivo distribution profile of OMC contained in nanocapsules (NCs) through the SC. Comparison with a nanoemulsion (NE) and a conventional o/w emulsion (EM).
NE increased the extent of OMC penetration relative to the penetration achieved by NCs or EM.
Olvera-Martínez et al. [26], 2005
Quantification of four common sunscreen agents, namely 2-hydroxy-4 methoxybenzophenone, 2-ethylhexyl-pmethoxycinnamate, 2ethylhexylsalicylate (octylsalicylate) and salicylic acid 3,3,5trimethcyclohexyl ester in a range of biological matrices.
A preliminary clinical study demonstrates a significant penetration of all sunscreen agents into the skin, as well as of oxybenzone and its metabolites across the skin.
Sarveiya et al. [106], 2004
Amount of sunscreen present on the skin of people at the beach.
The best protected areas were the upper arm and décolleté, but even in these areas, most volunteers had applied only 10% of the COLIPA standard amount.
Lademann et al. [107], 2004
Research 1) Kinetics and penetration depth of drugs
204 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
(Table 1) contd....
Research
Outcome
Author (Ref.) Year
1) Kinetics and penetration depth of drugs Penetration of titanium dioxide (TiO2) and In vivo and in vitro penetration studies methylene bis-benzotriazoyl tetramethylshowed an absence of TiO2 penetration into butylphenol (MBBT), included in a viable skin layers through either transcorbroad-spectrum sunscreen formulation, neal or transfollicular pathways, and a into human skin in vivo, using the TS negligible transcutaneous absorption of method, and in vitro, using a compartMBBT. mental approach.
Mavon et al. [108], 2007
In vitro human skin permeation and distribution of geranyl nitrile (GN)
Systemic exposure resulting from the use of GN as a fragrance ingredient, under unoccluded conditions, would be low based on the currently reported use levels.
Brian et al. [109], 2007
Development of a method to investigate the effectiveness of reservoir closure by different formulations. Model penetrant: Patent Blue V.
Application of barrier creams cannot replace other protective measures and should be maximally used to inhibit low-grade irritants or in combination with other protectants, or in body areas where other protective measures are not applicable.
Teichmann et al. [110], 2006
Differences in the distribution and the Penetration of highly hydrophilic and localization of both dyes within the SC were lipophilic dyes into the skin using pure oil observed. These differences depend on the or water, comparing them with an o/w physicochemical properties of both the emulsion. vehicles and the dyes.
Jacobi et al. [111], 2006
2) Dermal absorption of toxic or irritant chemicals Naphthalene has a short retention time in the Development and testing of a simple, nonhuman SC and the tape stripping method, if invasive dermal sampling technique on used within 20 min of the initial exposure, human volunteers under laboratory condican be employed to measure the amount of tions to estimate acute dermal exposure to naphthalene in the SC due to a single expojet fuel (JP-8). sure to jet fuel.
Mattorano et al. [36], 2004
The amount of keratin removed with tape strips was not affected by an exposure of up Normalization of extracted concentrations to 25 min to JP-8, and there was a substanof naphthalene (as a marker for jet fuel tial decrease in the amount of keratin reexposure) from human volunteers, before moved with consecutive TSs from the same and after exposure to jet fuel (JP-8). site; thus, adjusting the amount of naphthaRemoval and quantification of keratin by lene to the amount of keratin measured in a SC-TS tape-strip sample should improve the interpretation of the amount of this analyte by using this sampling approach.
Chao et al. [37], 2004
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 205 (Table 1) contd....
Outcome
Author (Ref.) Year
Description of a physiologically based pharmacokinetic (PBPK) model developed to simulate the absorption of organophosphate pesticides, such as parathion, fenthion, and methyl parathion, through porcine skin with flow-through cells.
The study demonstrated the utility of PBPK models for studying dermal absorption, which can be useful as explanatory and predictive tools.
Van der Merwe et al. [112], 2006
Study of whether the sodium lauryl sulphate (SLS) penetration rate into the SC is related to an impairment of skin’s water barrier function and inflammation.
Variation in barrier impairment and inflammation of human skin depends on SLS penetration rate, which was mainly determined by SC thickness.
Jongh et al. [113], 2006
Modification and testing of a vacuuming sampler for removing particles from the skin.
Agreement between the vacuuming sampler and the TS technique.
Lundgren et al. [114], 2006
Development of a noninvasive sampling method for measuring dermal exposure to a multifunctional acrylate employing TS.
No significant difference was observed in recovery between TPGDA and UV-resin for the first tape stripping when calculated as a percentage of the theoretical amount. TS can be used to quantify dermal exposure to multifunctional acrylates
Nylander-French et al. [115], 2002
Research 2) Dermal absorption of toxic or irritant chemicals
3) Evaluation of factors that influence the physiology of the stratum corneum
Effect of gender on the physiology of the SC.
The skin of women was characterized by a significantly higher pH value (5.6±0.4) than that of men (4.3±0.4). Protein absorption was the only other parameter significantly dependent on gender.
Jacobi et al. [122], 2005
Influence of procedures inherent to each stripping protocol on changes in skin physiology.
Skin hydration was not influenced by the variables tested.
Löffler et al. [56], 2004
Efficacy of TS in removing complete cell layers from the superficial part of human SC.
Furrows in the skin can present difficulties when performing depth penetration studies. Although the largest part of the skin surface will be stripped properly, it has to be realized that small areas, represented by furrows, may still contain high concentrations of the substance applied.
Van der Molen et al. [57], 1997
206 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
(Table 1) contd....
Research
Outcome
Author (Ref.) Year
3) Evaluation of factors that influence the physiology of the stratum corneum Local changes in the ultrastructure of human skin after iontophoresis in human skin in vitro and in vivo.
No drastic changes in the ultrastructure of the SC were observed.
Fatouros et al. [123], 2006
Study of the differences in the SC lipid profile in healthy and diseased human skin relative to the SC lipid organization and to the skin barrier function in vivo.
Weerheim et al. [124], 2001
Evaluated, using attenuated total reflectance Fourier transform infrared spectroscopy, the SC bioavailability of terbinafine (TBF) following topical treatment with four different formulations, based on a vehicle consisting of 50% ethanol and 50% isopropyl myristate.
It was found that the formulation containing 5% oleic acid significantly enhanced the SC availability of TBF
Alberti et al. [125], 2001
Determined whether a structurally heterogeneous biomembrane, human SC, behaved as a homogeneous barrier to water transport.
The variation of TEWL as a function of SC removal behaved in a manner entirely consistent with a homogeneous barrier, thereby permitting the apparent SC diffusivity of water to be found.
Kalia et al. [126], 1996
4) Stratum corneum composition Establishment of a suitable analytical method for the determination of the local SC lipid composition. 5) Stratum corneum thickness
6) Determination of superficial infections and viruses New animal model for the purpose of studying superficial infections.
Evaluation of novel antimicrobial treatments for superficial infections caused by S. aureus and S. pyogenes.
Kugelberg et al. [129], 2005
Immunocompetent patients were tested for human papilloma virus (HPV) DNA in swab samples collected on top of skin tumors and in biopsies of the same tumors, obtained after stripping with tape to remove superficial layers.
HPV DNA is common in superficial layers of lesions, but is not necessarily present in tumors.
Forslund et al. [130], 2004
Tight junctions of regenerating epidermis may provide a functional barrier prior to regeneration of the corneal layer.
Malminen et al. [131], 2003
7) Skin regeneration Expression of tight junction components during the reepithelialization of suction blisters and the regeneration of the corneal layer after TS.
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 207 (Table 1) contd....
Outcome
Author (Ref.) Year
In vitro model for the developing skin of the premature neonate.
Porcine skin, in vitro, tape-stripped to a particular level, can provide a barrier corresponding to a specific degree of neonate maturation, and thus, can serve as a useful tool to explore whether transdermal drug delivery in this unique patient population may be beneficial.
Sekkat et al. [132], 2002
An in vivo human model was utilized to define the irritation potential of a topical agent after partial removal of the stratum corneum by cellophane TS
This model provides a method for the prediction, with exaggerated sensitivity, of chemical irritation and proclivity to enhance or retard water barrier repair.
Zhai et al. [133], 1998
Research 7) Skin regeneration
Escobar-Chávez et al. [7] determined the penetration of sodium naproxen, formulated in Pluronic F-127 gels containing Azone® and Transcutol® as penetration enhancers, through human skin in vivo by using the TS technique. It was found that the combination of Azone® and Transcutol® in PF-127 gels enhanced sodium naproxen penetration, with up to two-fold enhancement ratios compared with the formulation containing only Transcutol. These results were confirmed by TEWL and ATR-FTIR spectroscopy, suggesting a synergistic action for Azone® and Transcutol®. Esposito et al. [77] produced and characterized monoleine (MO) dispersions as drug delivery systems for indomethacin. An in vitro diffusion study was conducted using Franz cells associated to SC epidermal membrane on cubosome dispersions viscosized by carbomer. In vivo studies based on skin reflectance spectrophotometry and TS were performed to better investigate the performance of cubosome as an indomethacin delivery system. Indomethacin incorporated in viscosized MO dispersions exhibited a lower flux with respect to the analogous formulation containing the free drug in the aqueous phase and to the control formulation based on carbomer gel. Reflectance spectroscopy demonstrated that indomethacin incorporated into MO dispersions can be released in a prolonged fashion. TS experiments corroborated this finding. MO dispersions can be proposed as nanoparticulate systems able to control the percutaneous absorption of indomethacin Ganem-Quintanar et al. [78] used naproxen-loaded nanoparticles to prepare, in a one-step process, unilaminar films of Eudragit E-100. Nanoparticle films and conventional films were characterized in vitro by drug release studies through a cellulose membrane using Franz-type cells, and in vivo by penetration experiments with the TS technique. Concerning in vivo penetration studies, no statistical differences were found for the amount of naproxen penetrated across the SC and the depth of penetration for the two films.
208 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
Herkenne et al. [79] investigated pig ear skin as a surrogate for human skin in the assessment of topical drug bioavailability by sequential TS of the SC. Ex vivo experiments on isolated pig ears were compared with in vivo studies in human volunteers. Four formulations including ibuprofen in different propylene glycol (PG)-water mixtures (25:75, 50:50, 75:25 and 100:0), were compared. Derived DPK parameters characterizing the diffusion and partitioning of the drug in the SC ex vivo were consistent with those in vivo following a 30-minute application period. Furthermore, non-steady-state ex vivo results could be used to predict the in vivo concentration profile of the drug across the SC when a formulation was administered for 3 h (i.e., close to steady-state). Taken together, the results obtained suggest that pig ear skin ex vivo is promising as a tool for topical formulation evaluation and optimization. Continuing with their research, Herkene et al. [80] explored the potential of using SC-TS, post application of a topical drug formulation, to derive DPK parameters describing the rate and extent of delivery into the skin of Ibuprofen in the ventral forearms of human volunteers for periods ranging between 15 and 180 minutes. Subsequently, SC was tape-stripped, quantified gravimetrically, and extracted for drug analysis. Together with concomitant TEWL measurements, SC concentration–depth profiles of the drug were reproducibly determined and fitted mathematically. The SC-vehicle partition coefficient (K) and a first-order rate constant related to ibuprofen diffusivity in the membrane (D/L2, where L SC thickness) were derived from data-fitting and characterized the extent and rate of drug absorption across the skin. Integration of the concentration profiles yielded the total drug amount in the SC at the end of the application period. Using K and D/L2 obtained from the 30-minute exposure, it was possible to predict ibuprofen uptake as a function of time into the SC. Prediction and experiment agreed satisfactorily suggesting that objective and quantitative information, with which to characterize topical drug bioavailability, can be obtained from this approach. Hostnek et al. [81] sheded light on the long-standing controversy on whether wearing copper bangles benefits patients suffering from inflammatory conditions such as arthritis. Sequential TS was performed on healthy volunteers to examine the diffusion of copper through human SC in vivo, following application of the metal as powder on the volar forearm for periods of up to 72 h. Exposure sites were stripped 20 times, and the strips were analyzed for metal content by inductively coupled plasma-mass spectroscopy. The results indicate that, in contact with skin, copper will oxidize and may penetrate the SC after forming an ion pair with skin exudates. The rate of reaction seems to depend on contact time and oxygen availability. A marked inter-individual difference was observed in baseline values and amounts of copper absorbed. Lodén et al. [82] compared the bioavailability of ketoprofen in a photostabilised gel formulation without photoprotection using a new DPK tape stripping model and an established ex vivo penetration method using human skin. Analyses of the SC showed that during the first 45 minutes, about 12 μg/cm2 of ketoprofen were absorbed into the skin from the formulations. The area under the ketoprofen concentration–time curve (AUC0–6 h) for the photo-stabilised gel/transparent gel ratio was 73%. The rate of penetration of ketoprofen through isolated skin was approximately 0.2 μg/cm2h for both formulations. The ratio’s AUC0–36 h was 84%. Thus, the two methods did not disagree in terms of the relative efficacy of the two gels. The comparison of the amount of ketoprofen in the skin
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 209
after 45 min with the amount penetrated through the excised skin during 36 h, suggests a change in the thermodynamic activity of ketoprofen during exposure. A supersaturated formulation may have been formed initially due to evaporation of ethanol. Wagner Steffen et al. [83] studied the penetration kinetics of sesquiterpene lactones (SLs) in Arnica montana preparations; a stripping method with adhesive tape and pig skin as a model was used. For the determination of SLs in the stripped layers of the SC, a gas chromatography/mass spectrometry method was developed and validated. The penetration behavior of one gel preparation and two ointment preparations was investigated. The SLs of all preparations showed a comparable penetration and permeation through the SC, in the uppermost layer of the skin. Interestingly, the gel preparation showed a decreased penetration rate over 4 h, whereas the penetration rate of ointments remained constant over time. Moreover, they could demonstrate that the total amount of SLs penetrated depends only on the kind of formulation and the SLs-content in the formulation, but not on SLs composition or the extraction agent used. Wagner Heike et al. [84] carried out penetration experiments to investigate several incubation times with three different skin flaps using the Saarbruecken penetration model and the lipophilic model drug flufenamic acid. Drug distribution within SC was obtained by the TS technique, while the drug present in deep skin layers was determined by cryosectioning. In addition, for the lipophilic drug flufenamic acid, a direct linear correlation was found between SC/water partition coefficients and the drug amounts penetrated into the SC for all the time intervals tested. The authors concluded that SC/water-partition coefficients offer the possibility to predict drug amounts within the SC of different donor skin flaps, without a time-consuming determination of the lipid composition of the SC. 5.1.2. Corticosteroids The aim of Pellanda et al. [85] was to investigate the effect of i) dose and ii) application frequency on the penetration of triamcinolone acetonide (TACA) into human SC in vivo. The experiments were conducted on the forearms of 15 healthy volunteers, with i), single TACA doses (300 μg/cm2 and 100 μg/cm2), and ii) single (1 x 300 μg/cm2) and multiple (3 x 100 μg/cm2) TACA doses. SC samples were collected by TS after 0.5, 4 and 24 h (i) and after 4, 8 and 24 h (ii). In Experiment 1, TACA amounts within SC after application of 1 x 300 μg/cm2 compared to 1 x 100 μg/cm2 were only significantly different immediately after application, and were similar at 4 and 24 h. In ii), multiple applications of 3 x 100 μg/cm2 yielded higher TACA amounts compared to a single application of 1 x 300 μg/cm2 at 4 and 8 h. At 24 h, no difference was observed. In conclusion, by using this simple vehicle, considerable TACA amounts were retained within the SC, independently of dose and application frequency. 5.1.3. Disinfectants Lboutounne et al. [86] investigated the sustained bactericidal activity of chlorhexidine base loaded poly(-caprolactone) nanocapsules against Staphylococcus epidermidis inoculated onto porcine ear skin. The antimicrobial activity of these colloidal carriers was evaluated (i) in vitro against eight strains of bacteria, and (ii) ex vivo against Staphylococcus epidermidis inoculated for 12 h onto porcine ear skin surface treated for 3 min either with 0.6% chlorhexidine base loaded or unloaded nanocapsules suspended in hydrogel, or 1% chlorhexidine digluconate aqueous solution. Chlorhexidine absorption into
210 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
the SC was evaluated by the TS technique. The results showed that chlorhexidine nanocapsules in aqueous suspension with a 200–300 nm size and a positive charge exhibited similar minimum inhibitory concentrations against several bacteria, compared with chlorhexidine digluconate aqueous solution. Ex vivo, there was a significant reduction in the number of colony forming units from skin treated with chlorhexidine nanocapsules suspension for 3 min compared to chlorhexidine digluconate solution after an 8-h artificial contamination. Interestingly, nanocapsules were present in porcine hair follicles. Topical application of chlorhexidine base-loaded positively charged nanocapsules in an aqueous gel achieved a sustained release of bactericide against Staphylococcus epidermidis for at least 8 h. 5.1.4. Drugs for Keratinization Disorders Fresno-Contreras et al. [87] designed an all-trans retinoic acid (RA) topical release system that modifies drug diffusion parameters in the vehicle and the skin in order to reduce systemic absorption and side-effects associated with the topical application of the drug to the skin. RA, either in free form or encapsulated in SC lipid liposomes, was included in hydrogels prepared with Carbopol® UltrezTM 10 and hyaluronic acid. In vitro permeability experiments with [3H]-t-RA were carried out using a Franz-type diffusion cell in abdominal rat skin samples. Accumulation of the drug in the surface and skin layers was evaluated by both the TS technique and a dissection technique. The results show that RA encapsulation not only prolongs drug release, but also promotes drug retention in viable skin. At the same time, interaction between RA and hyaluronic acid has an obstructive effect on diffusion, which contributes to the formation of a reservoir. 5.1.5. Anesthetics Padula et al. [88] studied the behavior of a skin bioadhesive film containing lidocaine, in vitro and in vivo. Film characterization included in vitro and in vivo drug transport studies with and without iontophoresis. The release rate was compared with a lidocaine commercial gel. The permeation kinetics across the skin was not linear, but the patch acted as a matrix controlling drug delivery. Additionally, permeation rate increased with drug loading. The in vivo experiments with TS indicated that the presence of water during film application is essential to achieve not only the proper adhesion, but also an effective accumulation. The application of an electric current to the patch can further increase the amount of drug accumulated in the SC. 5.1.6. Keratolytics Bashir et al. [89] studied the keratolytic efficacy of topical preparations containing salicylic acid (SA) in humans by the TS technique, quantifying SC removal by protein analysis. In combination with TS, squamometry was used to evaluate the influence of SA on skin surface scaliness and desquamation. Furthermore, skin barrier perturbation and skin irritability were recorded and related to the dermatopharmacological effect of the preparations. In contrast to squamometry, TS combined with protein analysis was sensitive in detecting the keratolytic effect of SA within hours of application. Importantly, whereas the pH of the preparations had only a minimal influence on efficacy, local dermatotoxicity was significantly increased at an acidic pH. This indicates that the intent to increase the amount of free, non-dissociated SA is, in fact, counterproductive, as more acidic preparations resulted in skin irritation and barrier disruption.
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 211
5.1.7. Retinoids and Antioxidants Abdulmajed et al. [90] used a novel synthetic technique to synthesize the co-drug retinyl ascorbate (RA-AsA) ester from all-trans-retinyl chloride (RA) and l-ascorbic acid (AsA) suspended in ethanol at low temperature. The flux and permeation coefficient were determined using heat separated human skin membrane, and skin penetration was determined by TS using full-thickness human skin. All experiments were performed in parallel with retinyl palmitate and ascorbyl palmitate. Overall, the data suggest the potential value of RA-AsA co-drug for treating damage to skin resulting from UV-induced production of free radicals. 5.1.8. Aquaporine-3 Hara et al. [91] showed that glycerol replacement corrects each of the defects in aquaporine-3 (AQP3)-null mice. SC water content, measured by skin conductance and 3 H2O accumulation, was 3-fold lower in AQP3-null vs. wild-type mice, but was similar after topical or systemic administration of glycerol in amounts that normalized glycerol content in the SC. Orally administered glycerol fully corrected reduced skin elasticity in AQP3-null mice, as measured by the kinetics of skin displacement after suction, and the delayed barrier recovery, as measured by TEWL after TS. The analysis of [14C]glycerol kinetics indicated a reduced blood-to-SC transport of glycerol in AQP3-null mice, resulting in slowed lipid biosynthesis. These data provide functional evidence for a physiological role of glycerol transport by aquaglyceroporin, and indicate that glycerol is a major determinant of SC water retention and of mechanical and biosynthetic functions. Their findings establish a scientific basis for the >200 year old empirical practice of including glycerol in cosmetic and medicinal skin formulations. 5.1.9. Antimycotic Drugs Kalia et al. [92] presented a method to determine the cutaneous bioavailability and hence to evaluate the bioequivalence of topically applied drugs in vivo. The procedure uses serial TS and TEWL measurements to quantify the thickness of the removed SC and to determine the intact membrane thickness. Following TS, the drug is extracted from the tapes and assayed, by HPLC. This provides a drug concentration profile of terbinafine as a function of the normalized position within the SC. The data are fitted to a solution of Fick’s second law of diffusion in order to calculate characteristic membrane transport parameters. Integration of the concentration profile over the entire SC thickness, that is, the AUC, provides a measure of the cutaneous bioavailability and hence can be used to assess the bioequivalence of topically applied drugs. 5.1.10. Antiviral Drugs Morgan et al. [93] measured the contribution of SC barrier and microvascular perfusion in determining dermal tissue levels of two hydrophilic drugs (aciclovir and penciclovir) in vivo. Removal of the SC by TS resulted in a 1300-fold increase in penciclovir absorption and a 440-fold increase in aciclovir absorption, confirming that SC is the major barrier to hydrophilic drug absorption. 5.1.11. Anti-Varicella Zoster Virus Nucleoside Jarvis et al. [94] determined the in-vitro dermal delivery of a new class of lipophilic, highly potent and uniquely selective anti-Varicella Zoster virus nucleoside (VZV) ana-
212 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
logue compared with aciclovir. Three test compounds (Cf1698, Cf1743, and Cf1712) and acyclovir were formulated in propylene glycol/aqueous cream, and finite doses were applied to full-thickness pig ear skin for 48 hours in vertical Franz-type diffusion cells. Depth profiles were constructed following TS and membrane separation. All three test compounds reached the target basal epidermis in concentrations suggesting they would be highly efficacious in reducing viral load. Furthermore, the data showed that each of the test compounds would have a far superior performance than aciclovir. The dermatomal site of viral replication during secondary infection —the basal epidermis— was successfully targeted. 5.1.12. Vaccines The skin-associated lymphoid tissue, formed by powerful antigen-presenting cells (APCs), such as Langerhans cells (LCs), dermal dendritic cells (DCs), re-circulating T cells, and regional LNs, ensures the efficient presentation of antigen to immunocompetent cells and the induction of strong immune responses. LCs and dermal DCs commonly exist in the skin and are easy to target [95]. The TS technique has been used to study the effect of oligodeoxynucleotides on the immune response [96] and expression of immune receptors [97]. Inoue et al. [96] examined the effect of CpG-oligodeoxynucleotide (CpG-ODN) on the immune response to an antigen applied to tape-stripped mouse skin, by evaluating the production of cytokines and Ig isotypes. Confocal laser scanning microscopy revealed that the OVA (model antigen) and CpG-ODN easily penetrated the tape-stripped skin. Co-administration of CpG-ODN and OVA to the disrupted skin elicited an antigenspecific, Th1-predominant immune response, and enhanced the production of Th1-type cytokines, IL-12 and IFN-. On the other hand, the production of a Th2-type cytokine, IL-4, was drastically suppressed. In terms of antigen-specific antibody production, the IgG2a level, which is regulated by IFN-, was increased by CpG-ODN, but IgE production regulated by IL-4 was suppressed. Furthermore, the administration of CpG-ODN through the skin drastically attenuated the production of IgE in mice experiencing IgEtype immune response. Administration of CpG-ODN through the skin may shift the immune response from a Th2 to a Th1-like response. Continuing with their studies, Inoue et al. [97] also demonstrated that TS induces the expression of toll-like receptor (TLR)-9 in the skin, and enhances the Th1-type immune response triggered by CpG-ODN administered through the tape-stripped skin. TS induces the expression of TLR-9 and tumor necrosis factor (TNF)- in the skin, and CpGODN treatment through the tape-stripped skin enhances the migration of antigenpresenting cells to the draining lymph nodes. On the other hand, TLR-9 mRNA and TNF- mRNA were not observed in the skin when CpG-ODN was injected intradermally, or in Th1-type immune response. The transdermal application of CpG-ODN with an antigen through the tape-stripped skin is an effective way to induce a Th1-type immune response, and is also a simple, cost-effective and needle-free vaccination system. Liu et al. [98] administered human immunodeficiency virus type-1 (HIV-1) DNA vaccine with cytokine-expressing plasmids to the skin of mice by a new topical application technique involving prior elimination of keratinocytes using TS. Their results revealed that the topical application of HIV-1 DNA vaccine induced high levels of both humoral and cell-mediated immune activity against HIV-1 envelope antigen. Coadministration of the DNA vaccine with cytokine expression plasmids of IL-12 and gra-
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 213
nulocyte-macrophage colony-stimulating factor by this new method raised the levels of both the HIV-specific cytotoxic T lymphocyte (CTL) response and delayed-type hypersensitivity (DTH) and facilitated the induction of substantial immune responses by DNA vaccine. Skin biopsy sections showed significant increases of S-100 protein-positive dendritic cells (DCs). These results suggest that the topical application method is an efficient route of DNA vaccine administration and that the immune response may be induced by DNA plasmids taken in by DCs, Langerhans cells (LCs), or others such as antigen-presenting cells. This new topical application is likely to be of benefit in clinical use. 5.1.13. Other Kind of Permeants (Polyethylene Glycols, 4-Cyanophenol, Urea) Ayala-Bravo et al. [99] investigated the effect of sucrose esters (sucrose oleate and sucrose laureate in water or Transcutol®, TC) on the SC barrier properties in vivo, and examined the impact of these surfactant-like molecules on the in vivo percutaneous penetration of a model penetrant, 4-hydroxybenzonitrile (4-cyanophenol, 4CP). The effect of the enhancers on 4CP penetration was monitored in vivo using ATR-FTIR spectroscopy in conjunction with TS of the treated site. A combination of sucrose esters (oleate or laureate) and TC is able to temporarily alter the SC barrier properties, thereby promoting 4CP penetration. Results from TS experiments can be affected significantly by chemical diffusion into the SC during the time required to apply and remove all of the TSs, tTS (period of time required to completely remove the SC by TS). For this reason, Reddy et al. [100] studied dermal absorption of 4CP in humans using TS experiments to assess the conditions under which diffusion alters TS results. Mathematical models were developed to assess the effects of diffusion on parameter estimation. In an experiment with tTS > tlag (i.e., the lag time for a chemical to cross the SC), the permeability coefficient for 4CP, Psc,v, calculated including tTS, was consistent with the values from the literature. When diffusion during stripping was not included in the model, Psc,v, was 70% smaller. Calculations show that chemical concentrations in TSs can be affected by diffusion during TS, but with tTS < 0.2 tlag and an exposure time > 0.3 tlag, TS concentrations are not significantly affected by tTS. Jakasa et al. [101] developed a sensitive method for the determination of polyethylene glycols (PEGs) with different molecular weights (MW) in the human SC obtained by TS. The analysis is based on derivatization with pentafluoropropionic anhydride and gas chromatography–electron capture detection. The method showed to be suitable for studying permeability in normal and impaired skin with respect to MW in the range of 150–600 Da. In order to obtain more data to assess the barrier function of uninvolved skin in atopic dermatitis (AD) patients, Jakasa et al. [102] determined the percutaneous penetration of PEGs of various molecular sizes in vivo in AD patients and control subjects using TS of the SC. The apparent diffusion coefficient of PEGs through atopic skin was twice as high as through normal skin, and decreased with increasing MW in both groups. The partition coefficient in the skin of AD patients was half of that for normal skin, but as for normal skin, there was no MW dependence. Although atopic skin exhibited an altered barrier with respect to diffusion and partitioning, the permeability coefficients were nearly the same for atopic and normal skin. The results support the assumption of an altered skin barrier in AD patients, even if the skin is visibly unaffected by the disease.
214 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
Tsai et al. [103] further investigated the dependence of permeability on MW with different forms of barrier disruption. A series of PEGs with a MW ranging from nearly 300 to over 1000 Da were used to study the effects of TS and sodium dodecyl sulphate (SDS) treatment on MW permeability profiles of mouse skin in vitro. The total penetration of PEG oligomers across control skin and tape-stripped skin and SDS-treated to different degrees of barrier disruption progressively decreased with increasing MW. Penetration enhancement relative to control skin was more prominent with larger molecules. The MW cut-off for skin penetration increased with the degree of barrier disruption, irrespectively of the treatment applied, and was 986 Da (TS) and 766 Da (SDS treatment) at TEWL levels in the range of 10–20 g/m2 per h, compared with 414 Da for control skin. The results strongly suggest that, regardless of the form of barrier disruption applied, not only higher amounts, but also more varieties of chemicals (larger molecules), may penetrate into the skin in the presence of a compromised barrier compared with normal skin. Park et al. [104] measured the status of skin surface hydration of uraemic patients with the corneometer and skin surface hydrometer, the functional capacity and the urea concentration of SC by TS technique and the response of eccrine sweat gland to sudorific agent (0.05% pilocarpine HCL) in 18 age-matched haemodialysis patients and 10 healthy volunteers. They also performed the water sorption-desorption test to uraemic and control subjects after application of urea in various concentrations. Uraemic patient's skin showed decreased water content compared to control subjects. However, they found no correlation between dry skin and pruritus. Although the urea concentration determined by TS of the horny layer in uraemic patients was elevated compared to control subjects (28.2 g/cm2 vs 5.04 g/cm2), its moisturizing effect to relieve pruritus is questionable because its artificial application revealed no improvement of the functional capacity of horny layer in concentration 5 times higher than the physiological concentration. Uraemic patients showed decreased sweating response to sudorific agent. In conclusion, the functional abnormalities of eccrine sweat glands may be account for dry skin in uraemic patients at least in part, but there is no correlation between xerosis and pruritus. 5.1.14. UV Absorbers Alvárez-Román et al. [105] determined whether encapsulation of lipophilic compounds in polymeric nanoparticles is able to improve topical delivery to the skin. The penetration of octyl methoxycinnamate (OMC) encapsulated in poly(-caprolactone) nanoparticles, into and across porcine ear skin in vitro, was investigated using TS. Quantification of OMC in the skin using TS demonstrated that nanoparticulate encapsulation produced a 3.4-fold increase in the level of OMC within the SC. Nanoparticulate encapsulation of OMC increased its “availability” within the SC. Olvera-Martínez et al. [26] prepared polymeric nanocapsules (NCs) containing OMC, and their in vivo distribution profile through the SC was determined by the TS technique. The penetration degree of OMC formulated in NCs was compared with that obtained for a nanoemulsion (NE) and a conventional oil-in-water (o/w) emulsion (EM). In vivo percutaneous penetration, evaluated by the TS technique, demonstrated that NE increased the extent of OMC penetration relative to the penetration achieved by NCs or EM. Likewise, OMC accumulation in the skin was significantly greater with NE than with EM or NCs.
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 215
Sarveiya et al. [106] developed a reverse HPLC assay to quantify four common sunscreen agents, namely, 2-hydroxy-4-methoxybenzophenone, 2-ethylhexyl-p-methoxycinnamate, octylsalicylate and salicylic acid 3,3,5-trimethcyclohexyl ester in a range of biological matrices. This assay was further applied to study skin penetration and systemic absorption of sunscreen filters after topical application to human volunteers. The assay allows the analysis of sunscreen agents in biological fluids, including bovine serum albumin solution, plasma and urine, and in human epidermis by using the TS technique. The results from the preliminary clinical study demonstrate a significant penetration of all sunscreen agents into the skin. Lademann et al. [107] determined the amount of sunscreen present on the skin of people at the beach. The amounts of sunscreen applied to different body sites were quantitatively determined by TS. The actual amounts of sunscreen applied were compared with the COLIPA (European Cosmetic Toiletry and Perfumery Association) standard. Most volunteers had applied 10% or less of the COLIPA standard amount to all body sites assessed. Mavon et al. [108] assessed the penetration of titanium dioxide (TiO2) and methylene bis-benzotriazoyl tetramethylbutylphenol (MBBT), included in a broad-spectrum sunscreen formulation, into human skin in vivo, using the TS technique, and in vitro, using a compartmental approach. More than 90% of both sunscreens were recovered in the first 15 tape strippings. In addition, they have shown that the remaining 10% did not penetrate the viable tissue, but was localized in the furrows. Less than 0.1% of MBBT was detected in the receptor medium, and no TiO2 was detected in the follicles, the viable epidermis or the dermis. Thus, this in vivo and in vitro penetration study showed an absence of TiO2 penetration into viable skin layers through either transcorneal or transfollicular pathways, and a negligible transcutaneous absorption of MBBT. However, differences in distribution within the SC reinforced the need for a complementary approach, using a minimally invasive in vivo methodology and an in vitro compartmental analysis. This combination represents a well-adapted method for testing the safety of topically applied sunscreen formulations in real-life conditions. 5.1.15. Fragances In-vitro human skin permeation and distribution of geranyl nitrile (GN) were determined by Brian et al. [109] using epidermal membranes, following application in 70% ethanol, under non-occlusive conditions, at maximum in-use concentration (1%). Levels of GN in the epidermis (plus any remaining in SC after TS), filter paper membrane support, and receptor fluid were combined to produce a total absorbed dose value of 4.72±0.32%. The systemic exposure resulting from the use of GN as a fragrance ingredient, under unoccluded conditions, would be low based on the currently reported use levels. 5.1.16. Dyes Teichmann et al. [110] developed a method to investigate the effectiveness of reservoir closure by different formulations. Patent Blue V in water was used as a model penetrant. Its penetration, with and without barrier cream treatment, was analyzed by TS in combination with UV/VIS spectroscopic measurements. The investigations showed that the SC represents a reservoir for topically applied Patent Blue V in water. Furthermore, the barrier investigations showed that vaseline and bees wax form a 100% barrier on the
216 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
skin surface. The third barrier cream, containing waxes and surfactant, only partially showed a protective effect against the penetration of Patent Blue V in water. Strong inter-individual differences were observed for this barrier product. In conclusion, it was assumed that the application of barrier creams cannot replace other protective measures, and should be used to inhibit low-grade irritants or in combination with other protectants, or in body areas where other protective measures are not applicable. Jacobi et al. [111] studied the penetration of highly hydrophilic (Patent blue V) and lipophilic (curcumin) dyes into the skin using pure oil (o) or water (w), and comparing them with an o/w emulsion. The penetration and localization of both dyes were investigated in vivo using TS and microscopy techniques. Differences in the distribution and the localization of both dyes within the SC were observed. These differences depend on the physicochemical properties of both the vehicles and the dyes. The vehicle appears to affect, in particular, the penetration pathways. As we can observe, there is an ongoing search for the identification of testing methods to optimize topical dosage forms and to assess topical drug bioavailability. While in vitro screening continues to play an important role (and is relatively inexpensive and easy to use), regulatory approval of drug delivery systems to the skin, with few exceptions, requires clinical trials to be performed. For many drugs used topically, the problem remains unsolved, since an easily visualized pharmacodynamic response is not elicited. As a consequence, various alternative techniques have been considered, of which SC TS is being given the greatest attention [7,26,35,76-111]. While the former is technically more challenging, the potential reward is a drug concentration-time profile in a compartment presumed to be in close communication with the site of action of most dermatological drugs. The latter, in contrast, offers an apparently easy and quite non-invasive methodology for skin tissue sampling, and is the basis of the FDA’s so-called DPK. Validation and optimization of the procedure have not come quickly. The goal of the research described here is not only to contribute to further establishing the credibility of the tape stripping technique, but also to demonstrate that useful and relevant measurements can be made on a surrogate, ex vivo skin model. 5.2. Dermal Absorption of Toxic or Irritant Chemicals The rate and extent of dermal absorption are important in the analysis of risk from dermal exposure to toxic chemicals, and for the development of topically applied drugs, barriers, insect repellents, and cosmetics, and the TS technique has been widely used to determine the penetration of these kinds of substances [36, 37, 112-114]. Mattorano et al. [36] developed and tested a simple, non-invasive dermal sampling technique on 22 human volunteers to estimate acute dermal exposure to jet fuel (JP-8). Two sites on the ventral surface of each forearm were exposed to 25 μl of JP-8, and the SC was sequentially tape-stripped using an adhesive tape. The analysis of the first tape strips indicated that JP-8 was rapidly removed from the SC over the 20-min study period. On average, after 5 min of exposure, the first two tape strips removed 69.8% of the applied dose. The amount recovered with two tape strips decreased over time, to a recovery of 0.9% 20 min after exposure. By fitting a mixed-effect linear regression model to the TS data, the authors were able to accurately estimate the amount of JP-8 initially applied. This study indicates that naphthalene has a short retention time in the human SC
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 217
and that the TS technique, if used within 20 min of initial exposure, can be used to reliably measures the amount of naphthalene initially present in the SC due to a single exposure to jet fuel. Chao et al. [37] presented a TS method for the removal and quantification of keratin from the SC for normalization of extracted concentrations of naphthalene (as a marker for jet fuel exposure) from 12 human volunteers before and after exposure to jet fuel (JP8). Due to the potential for removal of variable amounts of squamous tissue from each tape strip sample, keratin was extracted and quantified using a modified Bradford method. Naphthalene was quantified in the sequential tape strips collected from the skin between 10 and 25 min after a single dose of JP-8 was initially applied. The penetration of jet fuel into the SC was demonstrated by the fact that the average mass of naphthalene recovered by a tape strip decreased with increasing exposure time and subsequent tape strips. The actual concentration of naphthalene (as a marker for JP- 8 exposure) per unit of keratin in a tape-strip sample can be determined by using this method, and may prove necessary when measuring occupational exposures under field conditions. Van der Merwe et al. [112] described a physiologically based pharmacokinetic model developed to simulate the absorption of organophosphate pesticides, such as parathion, fenthion, and methyl parathion through porcine skin with flow-through cells. Three parameters were optimized based on experimental dermal absorption data, including solvent evaporation rate, diffusivity, and a mass transfer factor. Diffusion cell studies were conducted to validate the model under a variety of conditions, including different dose ranges (6.3–106.9 μg/cm2 for parathion; 0.8–23.6 μg/cm2 for fenthion; 1.6–39.3 μg/cm2 for methyl parathion), different solvents (ethanol, 2-propanol and acetone), different solvent volumes (5–120 μL for ethanol; 20–80 μL for 2-propanol and acetone), occlusion versus open to atmosphere dosing, and corneocyte removal by TS. The study demonstrated the utility of PBPK models for studying dermal absorption. The similarity between the overall shapes of the experimental and model-predicted flux/time curves and the successful simulation of altered system conditions for this series of small, lipophilic compounds, indicated that the absorption processes described in the model successfully simulated important aspects of dermal absorption in flow-through cells. These data have a direct relevance in the assessment of topical organophosphate pesticides’ risk. Jongh et al. [113] studied whether sodium lauryl sulphate (SLS) penetration rate into the SC is related to an impairment of skin’s water barrier function and inflammation. The penetration of SLS into the SC was assessed using a non-invasive TS procedure in 20 volunteers after a 4-h exposure to 1% SLS. Additionally, the effect of a 24-h exposure to 1% SLS on the skin water barrier function was assessed by measuring TEWL. A multiple regression analysis showed that the baseline TEWL, SC thickness and SLS penetration parameters K (SC ⁄water partition coefficient) and D clearly influenced the increase in TEWL after the 24-h irritation test. They found that variation in barrier impairment and inflammation of human skin depends on SLS’ penetration rate, which was mainly determined by SC thickness. Lundgren et al. [114] modified and tested a vacuuming sampler for removing particles from the skin. The sampler was compared with two other skin and surface exposure sampling techniques. These were based on surrogate skin (a patch sampler-adhesive tape on an optical cover glass) and a TS removal procedure. All three samplers measure the mass of dust on the skin. Dust containing starch was deposited onto the skin in a whole-
218 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
body exposure chamber. Samples were taken from forearms and shoulders and analysed using optical microscopy. With the different sampling techniques, small differences in results were obtained. There was a good agreement between the vacuuming sampler and the TS technique. Nylander-French et al. [115] developed and tested a noninvasive sampling method for measuring dermal exposure to a multifunctional acrylate employing a TS technique of the SC. Samples were subsequently extracted and a gas chromatographic method was employed for quantitative analysis of tripropylene glycol diacrylate (TPGDA). This method was tested in 10 human volunteers exposed to an a priori determined amount of TPGDA or a UV-radiation curable acrylate coating containing TPGDA (UV-resin) at different sites on hands and arms. On the average, the first TS removed 94% of the theoretical quantity of deposited TPGDA and 89% of the theoretical quantity of deposited TPGDA in UV-resin 30 min after exposure. Quantities of TPGDA recovered from two consecutive TS´s accounted for all of the test agent, demonstrating both the efficiency of the method to measure dermal exposure and the potential to determine the rate of absorption with successive samples over time. In general, the amount removed by the first TS was greater for TPGDA than for UV-resin while the second TS removed approximately 6 and 21% of TPGDA and UV-resin, respectively. However, when the amounts removed with the first TS for TPGDA or UV-resin from the five different individual sites were compared, no significant differences were observed. No significant difference was observed in recovery between TPGDA and UVresin for the first tape stripping when calculated as a percentage of the theoretical amount. The results indicate that this TS technique can be used to quantify dermal exposure to multifunctional acrylates. 5.3. Evaluation of Factors that Influence the Physiology of the Stratum Corneum Many factors are known to influence the physiology of the SC. In this way, increasing age is related to decreased skin thickness [115] and a variation in skin lipids [116] and flora [117]. The anatomical region also influences lipid distribution [118, 119], microflora [120] and physical parameters such as TEWL [121]. However, there are already a few studies reporting conflicting results for the effect of gender [122] on skin physiology, as well as for the effect of anatomical site, pressure, pressure duration and tape removal rate in skin physiology [55], or the effect of skin transport technology, as iontophoresis, on human skin [123]. Jacobi et al. [122] studied the effect of gender on the physiology of the SC. The physiological parameters: TEWL, pH value, hydration and sebum content were determined on the flexor forearms of 6 female and 6 male volunteers. In addition, SC samples, removed by TS, were studied for amount, spectroscopic properties, protein content, and mass. The skin of women was characterized by a significantly higher pH value (5.6±0.4) than that of men (4.3±0.4). Protein absorption was the only other parameter significantly dependent on gender. Both effects might be caused by differences in human biology, such as hormonal status. Therefore, volunteers’ gender should be considered in dermatologic studies. Löffler et al. [56] investigated the influence of the procedures (anatomical site, pressure, pressure duration, tape removal rate) inherent to each stripping protocol on changes in skin physiology. A significant influence of all parameters on TEWL increase, as a function of tape strip number was observed. The fastest increase was demonstrated on the forehead, followed by the back and, lastly, the forearm. Rapid removal produced a
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 219
protracted increase compared with slow removal. Pressure for 10s induced a faster increase in TEWL than 2s pressure. Likewise, pressure at 330 g cm-2 induced an earlier increase than pressure at 165 g cm-2. Skin hydration was not influenced by the variables tested. Van der Molen, et al. [57] investigated the efficacy of TS in removing complete cell layers from the superficial part of the human SC. A histological section of skin that was tape-stripped 20 times, clearly showed non-stripped skin in the furrows, indicating persistent incomplete TS. Replicates of tape-stripped skin surface demonstrated that even after removing 40 tape strips, furrows were still present. They emphasize that the results from studies using the TS method have to be viewed from the perspective that cells on one tape strip of the SC may come from different layers, depending on the position of the tape strip in relation to furrow slope, and such results should be interpreted with considerable caution. Fatouros et al. [123] investigated the local changes in the ultrastructure of human skin after iontophoresis in human skin in vitro and in vivo. Human dermatomed skin was subjected to passive diffusion for 6 hours, followed by nine hours of iontophoresis at 0.5mA/cm2. In addition, iontophoresis patches were applied to healthy volunteers for 3.5h with 0.5h of passive delivery followed by 3h of iontophoresis at a current density of 0.25mA/cm2. Subsequently, a series of TSs were performed, and were visualized by freeze-fracture transmission electron microscopy. In vitro/in vivo studies suggest that iontophoresis results in the formation of intercellular water pools, and in a weakening of the desmosomal structure only in the upper part of the SC. However, no changes in lipid organization were observed in vitro and in vivo at the 0.5 and 0.25mA/cm2 current densities, respectively. Therefore, even at relatively high current densities, no drastic changes in the ultrastructure of the SC are observed. As far as structural changes in SC are concerned, iontophoresis is a safe method under the experimental conditions used. 5.4. Stratum Corneum Composition Weerheim et al. [124] established a suitable analytical method for the determination of local SC lipid composition. For this purpose, SC samples were collected by sequential stripping with Leukoplex tape in five healthy volunteers. Lipids were extracted with an ethyl acetate: methanol mixture (20:80) and separated by means of HPTLC. The results of this study revealed that the free fatty acid level is higher, and cholesterol and ceramide levels are lower in the uppermost SC layers. Levels remained unchanged in the underlying SC layers. In these layers, the ceramide level was about 60 wt % and free fatty acid and cholesterol levels were about 20 wt % each. Ceramides could be separated into seven different fractions, and the relative amounts of individual ceramide fractions did not significantly change with the SC depth. The method developed allowed to study the differences in the SC lipid profile in healthy and diseased human skin relative to the SC lipid organization and the skin barrier function in vivo. 5.5. Determination of Stratum Corneum Thickness Alberti et al. [125] evaluated, using attenuated total reflectance Fourier transform infrared spectroscopy, the SC bioavailability of terbinafine (TBF) following topical treatment with four different formulations, based on a vehicle consisting of 50% ethanol and 50% isopropyl myristate. Three of these formulations included a percutaneous penetration enhancer: either 5% oleic acid, 10% 2-pyrrolidone or 1% urea. The SC con-
220 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
centration profile of TBF was measured by repeated infrared spectroscopic measurements while sequentially stripping off the layers of this barrier membrane with adhesive tape. TEWL measurements were also performed, to permit facile estimation of SC thickness. The SC concentration profiles of TBF were fitted to the appropriate solution of Fick's second law of diffusion. This analysis enabled the efficacies of the different formulations tested to be compared to the non-enhancer control. While it was found that the formulation containing 5% oleic acid significantly enhanced the SC availability of TBF, the other formulations did not improve the apparent drug delivery. Kalia et al. [126] determined whether a structurally heterogeneous biomembrane, human SC, behaved as a homogeneous barrier to water transport. Impedance spectra (IS) of the skin and measurements of the rate of TEWL were recorded sequentially in vivo in human subjects as layers of the SC were progressively removed by the serial application of adhesive tape strips. The low-frequency impedance of skin was much more significantly affected by TS than the higher frequency values; removal of the outermost SC layer had the largest effect. In contrast, TEWL changed little as the outer SC layers were stripped off, but increased dramatically when 6-8 microns of the tissue had been removed. It follows that the two noninvasive techniques probe SC barrier integrity in somewhat different ways. After SC removal, recovery of barrier function, as assessed by increasing values of the low-frequency impedance, apparently proceeded faster than TEWL decreased to the pre-stripping control. The variation of TEWL as a function of SC removal behaved in a manner entirely consistent with a homogeneous barrier, thereby permitting the apparent SC diffusivity of water to be found. Skin impedance (low frequency) was correlated with the relative concentration of water within the SC, thus providing an in vivo probe for skin hydration. Finally, the SC permeability coefficient to water, as a function of SC thickness, was calculated and correlated with the corresponding values of skin admittance derived from IS. 5.6. Determination of Superficial Infections and Viruses Topical infections due to S. aureus and S. pyogenes are clinically relevant and cause a variety of serious symptoms, including toxic shock syndrome and skin lesions [127], that can progress to sepsis and systemic shock if they are left untreated [128]. These bacterial species are also the most common causes of impetigo in humans [128]. The TS technique has offered the possibility of studying superficial infections on the skin [129], as well as viruses in skin tumors [130]. Kugelberg et al. [129] presented a new animal model for the purpose of studying superficial infections. In this model, an infection is established by disruption of the skin barrier by partial removal of the epidermal layer by TS and subsequent application of the pathogens Staphylococcus aureus and Streptococcus pyogenes. The infection and the infection route were purely topical. Thus, the present model is considered more biologically relevant for the study of superficial skin infections in mice and humans. Established topical antibiotic treatments are shown to be effective. The procedures involved in the model are simple, a feature that increases throughput and reproducibility. This new model should be applicable to the evaluation of novel antimicrobial treatments for superficial infections caused by S. aureus and S. pyogenes. Forslund et al. [130] investigated 229 immunocompetent patients tested for human papilloma virus (HPV) DNA in swab samples collected on top of skin tumors and in biopsies of the same tumors, obtained after stripping with tape to remove superficial
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 221
layers. HPV DNA was detected on top of 69% of the lesions, and in 12% of the stripped biopsies. A difference was seen for the four types of tumors studied. Seborrheic keratosis had 79% HPV positivity on top of lesions versus 19% in biopsies; actinic keratosis had 83% HPV positivity on top of lesions versus 11% in biopsies; basal cell carcinoma had 63% on top of lesions versus 8% in biopsies; and squamous cell carcinoma had 58% on top of lesions versus 19% in biopsies. HPV DNA is common in superficial layers of lesions, but is not necessarily present in tumors. 5.7. Skin Regeneration Malminen et al. [131] investigated the expression of tight junction components during reepithelialization of suction blisters and regeneration of the corneal layer after TS. Suction blisters were induced in eight healthy volunteers, and skin biopsies were taken 4 or 6 days afterwards. The restoration of epidermal barrier function was evaluated by measuring water evaporation (WE) from the wound area. TS was performed on three volunteers to remove the corneal layer. Prior to the biopsies, WE from the blister wounds was markedly elevated compared with normal skin. In the epidermis surrounding the blister, occludin and ZO-1 (membrane-associated guanylate kinase homologue protein family) were expressed in the granular cell layer only. In the hyperproliferative zone adjacent to the border of the blister, the expression of ZO-1 was redistributed into several spinous cell layers, while occludin expression was restricted to the upper epidermis. In the leading edge of migrating keratinocytes, both proteins were expressed solely in the most superficial layer of keratinocytes. Double labelling for ZO-1 and involucrin showed expression of both proteins in the same layers of hyperproliferative keratinocytes, while the expression patterns were clearly different in migrating keratinocytes. Tight junctions of regenerating epidermis may provide a functional barrier prior to regeneration of the corneal layer. Sekkat et al. [132] developed an in vitro model for the developing skin of the premature neonate. Barriers of different levels of efficiency were produced by differential tapestripping of the SC from the skin of excised porcine ears, and were characterized by measurements of TEWL. In this way, it was possible to express the recorded TEWL as a function of percentage SC thickness (F), generating the following relationship: TEWL = 2.7+41.exp [-0.028.F]. These data were then compared to previously published in vivo measurements of TEWL obtained from a population of premature neonates at various post-conceptional ages (PCA). The former showed a remarkably parallel relationship to that found in vitro with the porcine skin model, namely TEWL = 3.3+41 exp [0.026.(PCA-160)]. Therefore, it can be suggested that the empirically adjusted PCA (i.e., PCA-160) has a close correlation with the developing thickness of the neonate’s SC. Consequently, porcine skin in vitro, tape-stripped to a particular level, can provide a barrier corresponding to a specific degree of neonate maturation, and thus, can serve as a useful tool to explore whether transdermal drug delivery in this unique patient population may be beneficial. Zhai et al. [133] used an in vivo human model to define the irritation potential of a topical agent after partial removal of the SC by cellophane TS. The tape was applied to and removed approximately 50 times (mean, 50.0 +/- 16.7) from each test site on the volar aspect of the forearm. TEWL was measured before and daily for 5 days. The TEWL values at baseline after stripping represented the point of maximal stripping barrier disruption. The barrier disruption and irritation potential were assessed with TEWL
222 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
measurements. The results showed that the model topical agent had no adverse effect on barrier repair, i.e. did not interfere with TEWL normalization. This model provides a method for the prediction, with exaggerated sensitivity, of chemical irritation and proclivity to enhance or retard water barrier repair. They believe that the model may predict the response of low irritation materials and may be more sensitive than patch testing on normal skin, particularly for products to be used on certain areas, e.g. the face, anus, etc., or even mucous membranes. CONCLUSIONS The quantification of drugs within the skin is essential for topical and transdermal delivery research. Over the last two decades, horizontal sectioning, consisting of TS, has been the traditional investigative technique. Many in vivo methods for measuring dermal absorption of chemicals are invasive (e.g., blood sampling) or slow (e.g., urine samples collected for extended periods). TS of the outermost skin layer, the SC, is a fast and relatively non-invasive technique for measuring the rate and extent of dermal absorption. The TS technique has the potential to meet the requirements for an efficient and reliable method to assess dermal exposures. Stripping with an adhesive has been widely accepted as a dermal sampling technique in dermatology. Tape stripping data have been used to estimate permeability coefficients and partition coefficients, SC mass, barrier function, drug reservoir from in vivo dermal exposures, and even to explain the SC physiology. TS has also been proposed as a method for evaluating the bioequivalence of topical dermatological dosage forms. DPK characterization of the penetration of active drugs in human volunteers has been suggested to be able to replace comparative clinical trials as means of documenting bioequivalence. It is suggested that DPK assessment of drug concentrations in the SC is comparable to blood/urine measurements of systemically administered drugs, where the concentration of a drug in the SC is expected to relate to its concentrations in viable tissue. Short-contact DPK experiments can be used to obtain diffusion and partitioning parameters that may subsequently be able to predict drug penetration into the SC following longer application periods. Although tape stripping is widely used to determine dermal absorption through the SC, several factors can influence the actual technique. Recent reviews on this topic provide updated and additional insights [134-136]. The investigation of variations in skin condition (dry versus moist skin, skin defects, etc.) to determine their potential impact on the sampling method is warranted. For these reasons, the tape stripping technique requires further development. ACKNOWLEDGMENTS José Juan Escobar-Chávez wishes to acknowledge the PROFIP/UNAM grant. The authors also thank the financial support from PAPIIT/UNAM (Reference IN213205). ABBREVIATIONS SC
=
Stratum corneum
DPK
=
Dermatopharmacokinetic
TS
=
Tape stripping
TS Method as a Valuable Tool
Frontiers in Drug Design & Discovery, 2009, Vol. 4 223
AUC
=
Area under curve
TEWL
=
Transepidermal water loss
HP-P-CD
=
Hydroxypropyl-P-cyclodextrin
EBA
=
Ethyl 4-biphenylyl acetate
BPAA
=
4-biphenylylacetic acid
MO
=
Monoleine
PG
=
Propylene glycol
SLs
=
Sesquiterpene lactones
TACA
=
Triamcinolone acetonide
TA
=
Retinoic acid
SA
=
Salicylic acid
RA-AsA
=
Retinyl ascorbate
RA
=
Retinyl chloride
AsA
=
Ascorbic acid
AQP3
=
Aquaporine-3
VZV
=
Varicella Zoster virus
APCs
=
Antigen-presenting cells
LCs
=
Langerhans cells
DCs
=
Dendritic cells
CpG-ODN
=
CpG-oligodeoxynucleotide
TLR-9
=
Toll-like receptor-9
TNF
=
Tumor necrosis factor
HIV-1
=
Human immunodeficiency virus type-1
CTL
=
cytotoxic -T- lymphocyte
TC
=
Transcutol
4CP
=
4-cyanophenol
TSs
=
Tape strippings
AD
=
Atopic dermatitis
SDS
=
Sodium dodecyl sulphate
OMC
=
Octyl methoxycinnamate
NCs
=
Nanocapsules
NE
=
Nanoemulsion
EM
=
Emulsion
224 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Escobar-Chávez et al.
COLIPA
=
European Cosmetic Toiletry and Perfumery Association
MBBT
=
Methylene bis-benzotriazoyl Tetramethylbutylphenol
GN
=
Geranyl nitrile
FDA
=
Food and Drug Administration
JP-8
=
Jet fuel-8
SLS
=
Sodium lauryl sulphate
TPGDA
=
Tripropylene glycol diacrylate
TBF
=
Terbinafine
IS
=
Impedance spectra
HPV
=
Human papilloma virus
WE
=
Water evaporation
PCA
=
Post-conceptional ages
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]
[11]
[12] [13]
[14] [15] [16] [17] [18] [19] [20] [21]
Bommannan, D.; Potts, R.O.; Guy, R.H. J. Invest. Dermatol., 1990, 95, 403-408. Higo, N.; Naik, A.; Bommannan, D.B.; Potts, R.O.; Guy, R.H. Pharm. Res., 1993, 10, 1500-1505. Lotte, C.; Wester, R.C.; Rougier, A.; Maibach, H.I. Arch. Dermatol. Res., 1993, 284, 56-459. Pershing, L.K.; Silver, B.S.; Krueger, G.G.; Shah, V.P.; Skelley, J.P. Pharm. Res., 1992, 9, 45-51. Rougier, A.; Dupuis, D.; Lotte, C.; Roguet, R.; Wester, R.C.; Maibach, H.I. Arch. Dermatol. Res., 1986, 278, 465-469. Tojo, K.; Lee, A.C. J. Invest. Dermatol., 1989, 92, 105-108. Escobar-Chávez, J.J.; Quintanar-Guerrero, D.; Ganem-Quintanar, A. Drug Develop. Ind. Pharm., 2005, 31, 447-454. Pinkus, H. J. Invest. Dermatol., 1951, 16, 383-386. King, C.S.; Barton, S.P.; Nicholls, S.; Marks, R. Br. J. Dermatol., 1979, 100, 165-172. Shah, V.P. Topical Dermatological Drug Product NDAs and ANDAs-In Vivo Bioavailability, Bioequivalence, In Vitro Release and Associated Studies, US Department of Health and Human Services, Rockville, 1998. Conner, D.P. Differences in DPK Methods. http://www.fda.gov/ ohrms/dockets/ac/01/slides/3804s2_ 05_conner/index.htm, Advisory Committee for Pharmaceutical Sciences Meeting, Center for Drug Evaluation and Research (CDER), Food and Drug Administration (FDA), Rockville, Maryland, November 29, 2001. Franz, T.J. Study #1, Avita Gel 0.025% vs Retin-A Gel 0.025%, Advisory committee for pharmaceutical sciences meeting, Center for Drug Evaluation and Research (CDER), Food and Drug Administration (FDA), Rockville, Maryland, November 29, 2001. Pershing, L.K. Bioequivalence assessment of three 0.025% tretinoin gel products: dermatopharmacokinetic vs clinical trial methods, advisory committee for pharmaceutical sciences meeting, Center for Drug Evaluation and Research (CDER), Food and Drug Administration (FDA), Rockville, Maryland, November 29, 2001. Touitou, E.; Meidan, V.M.; Horwitz, E. J. Control. Release, 1998, 56, 7-21. Moser, K.; Kriwet, K.; Naik, A.; Kalia, Y.N.; Guy, R.H. Eur. J. Pharm. Biopharm., 2001, 52, 103112. Pinkus, H. G. Ital. Dermatol. Minerva Dermatol., 1966, 107, 1115-26. Forslind, B.A. Acta Derm. Venereol., 1994, 74, 1-6. Potts, R.O.; Francoeur, M.L. J. Invest. Dermatol., 1991, 96, 495-499. Potts, R.O.; Guy, R.H. Pharm. Res., 1992, 9, 663-669. Ellias, P.M. J. Control. Release, 1991, 15, 199-208. Barry, B.W. Dermatological Formulations: Percutaneous Absorption. In: Swarbick J, Ed. Drugs and the Pharmaceutical Sciences. New York and Basel: Marcel Dekker, Inc. p. 202, 1983.
TS Method as a Valuable Tool [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58]
[59] [60] [61] [62] [63] [64] [65] [66] [67] [68]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 225
Hadgraft, J. Int. J. Pharm., 2001, 224, 1-18. Guy, R.H. Hadgraft, J. Transdermal drug delivery. New York: Marcel Dekker, Inc., p. 1-23, 2003. Walters, K.A.; Roberts, M.S. Dermatological and transdermal formulations. New York: Marcel Dekker, Inc., p. 1-39, 2002. Nevill, A.M. Am. Physiol. Soc., 1994, 2870, 2873. Olvera-Martínez, B.I.; Cazares-Delgadillo, J.; Calderilla-Fajardo, S.B.; Villalobos-García, R.; GanemQuintanar, A.; Quintanar-Guerrero, D. J. Pharm. Sci., 2005, 94, 552-1559. Escobar-Chávez, J.J.; López-Cervantes, M.; Naïk, A.; Kalia, Y.N.; Quintanar-Guerrero, D.; GanemQuintanar, A. J. Pharm. Pharm. Sci., 2006, 9, 339-358. Miyazaki, S.; Yokouchi, Ch.; Nakamura, T.; Hashiguchi, N.; Hou, W.-M.; Takada, M. Chem. Pharm. Bull., 1986, 34, 1801-1808. Chi, S.Ch.; Do, K.; Tan, H.K.; Chun, H.W. US 5, 527, 832, 1996. Fang, J.Y.; Leu, Y.L.; Wang, Y.Y.; Tsai, Y.H. Eur. J. Pharm. Sci., 2002, 15, 417-423. Shin, S.C.; Cho, C.W.; Oh, I.J. Int. J. Pharm., 2001, 222, 199-203. Liaw, J.; Lin, Y-Ch. J. Control. Release, 2000, 68, 273-282. Wang, Y.Y.; Hong, C.T.; Chiu, W.T.; Fang, J.Y. Int. J. Pharm., 2001, 224, 1-2. Kattan El, A. F.; Asbill, C. S.; Kim, N.; Michniak, B.B. Drug Deliv., 2000, 7, 147-153. Curdy, C.; Kalia, Y.N.; Naïk, A.; Guy, R.H. J. Control. Release, 2001, 76, 73-79. Mattorano, D.A.; Kupper, L.L.; Nylander-French, L.A. Ann. Occup. Hyg., 2004, 48, 139-146. Chao, Y.-Ch.; Nylander-French, L.A. Ann. Occup. Hyg., 2004, 48, 65-73. Wester, R.C.; Christoffel, J.; Hartway, T.; Poblete, N.; Maibach, H.I.; Forsell, J. Pharm. Res., 1998, 15, 82-84. Feldmann, R.J.; Maibach, H.I. J. Invest. Dermatol., 1969, 542, 89-94. Feldmann, R.J.; Maibach, H.I. J. Invest. Dermatol., 1970, 54, 399-404. Feldman, R.J.; Maibach, H.I. Toxicol. Appl. Pharmacol., 1974, 28, 126-132. Wester, R.C.; Noonan, P.K. J. Invest. Dermatol., 1978, 70, 92-94. Wester, R.C.; Maibach, H.I. Drug Metab. Rev., 1983, 14, 169-205. McKenzie, A.W.; Stoughton, R.B. Arch. Dermatol., 1962, 86, 608-610. Dupuis, D.; Rougier, A.; Rouguet, R.; Lotte, C. Br. J. Dermatol., 1986, 115, 233-238. Rougier, A.; Dupuis, D.; Lotte, C.; Rouguet, R.; Schaefer, H. J. Invest. Dermatol., 1983, 81, 275-278. Rougier, A.; Dupuis, D.; Lotte, C.; Rouguet, R.; Wester, R.C.; Maibach, H.I. Arch. Dermatol. Res., 1986, 278, 465-469. Poet, T.S.; Thrall, K.D.; Corley, R.A.; Hui, X.; Edwards, J.A.; Weitz, K.K.; Maibach, H.I.; Wester, R.C. Toxicol. Sci., 2000, 54, 42-51. Bronaugh, R.L.; Stewart, R.F.; Congdon, E.R. Toxicol. Appl. Pharmacol., 1982, 62, 481-488. Collier, S.W.; Sheikh, N.M.; Sakr, A.; Lichtin, J.L.; Stewart, R.F.; Bronaugh, R.L. Toxicol. Appl. Pharmacol., 1989, 99, 522-533. Tsai, J.-C.; Weiner, N.D.; Flynn, G.L.; Ferry, J. Int. J. Pharm., 1991, 72, 227-231. Sheth, N.V.; McKeough, M.B.; Spruance, S.L. J. Invest. Dermatol., 1987, 89, 598-602. Ohman, H.; Vahlquist, A. Acta Derm. Venereol., 1994, 74, 375-379. Bashir, S. J.; Chew, A. L.; Anigbogu, A. Skin Res. Technol., 2001, 7, 40-48. Martin, E.; Neelissen-Subnel, M.T.A.; De Haan, F.H.N.; Boddé, H.E. Skin Pharmacol., 1996, 9, 6977. Löffler, H.; Dreher, F.; Maibach, H.I. Br. J. Dermatol., 2004, 151, 746-752. Van der Molen, R.G.; Spies, F.; Van‘t Noordende, J.M.; Boelsma, E.; Mommaas, A.M.; Koerten, H.K. Arch. Dermatol. Res., 1997, 289, 514-518. Surber, C.; Schwarb, F.P.; Fmith, E.W. Tape stripping technique. In: Percutaneous Absorption – Drug – Cosmetics – Mechanisms – Methodology (Bronough H, Maibach HI, eds), 3rd ed. New York: Marcel Dekker, 395-409, 1999. Nangia, A.; Camel, E.; Berner, B. Pharm. Res., 1993, 10, 1756-1759. Van Voorst Vader, P.C.; Lier, J.G.; Woest, T.E. Acta Derm. Venereol., 1991, 71, 301-305. Surakka, J.; Johnsson, S.; Rosen, G. J. Environ. Monit., 1999, 1, 533-540. Kondo, H.; Ichikawa, Y.; Imokawa, G. Eur. J. Immunol., 1998, 28, 769-779. Fluhr, J.W.; Gloor, M.; Lehmann, L. Acta Derm. Venereol., 1999, 79, 418-21. Pechere, M.; Krischer, J.; Remondat, C. J. Dermatol., 1999, 26, 558-61. Pechere, M.; Remondat, C.; Bertrand, C. Dermatology, 1995, 191, 348-349. Ghadially, R.; Brown, B.E.; Sequeira-Martin, S.M. J. Clin. Invest., 1995, 95, 2281-90. Van der Valk, P.G.; Maibach, H.I. Clin. Exp. Dermatol., 1990, 15, 180-2. Hostynek, J.J.; Dreher, F.; Pelosi, A. Acta Derm. Venereol., 2001, 212, 5-10.
226 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [69] [70] [71] [72]
[73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110]
Escobar-Chávez et al.
Lademann, J.; Otberg, N.; Richter, H. Skin Pharmacol. Appl. Skin Physiol., 2001, 14, 17-22, Wilhelm, K.P.; Surber, C.; Maibach, H.I. J. Invest. Dermatol., 1991, 97, 927-32. Schafer, U. Topical Absorption of Dermatological Products. Robert L. Bronaugh and Howard I. Maibach (Editors), Marcel Dekker, New York, 1993. Shah, V.P.; Flynn, G.L.; Yacobi, A.; Maibach, H.I.; Bon, C.; Fleischer, N.M.; Franz, T.J.; Kaplan, S.A.J.K.; Lesko, L.J.; Marty, J.P.; Pershing, L.K.; Schaefer, H.; Sequeira, J.A.; Shrivastava, S.P.; Wilkin, J.; Williams, R.L. Pharm. Res., 1998, 15, 167-171. Surakka, J.; Lindh, T.; Rosen, G. Ann. Occup. Hyg., 2000, 44, 635-44. Nylander-French, L.A. Ann. Occup. Hyg., 2000, 44, 645-51. Howes, D.; Guy, R.; Hadgraft, J.; Heylings, J.; Hoeck, U.; Kemper, F.; Maibach, H.; Marty, J.P.; Merk, H.; Parra, J.; Rekkas, D.; Rondelli, I.; Schaefer, H.; Täuber, U.; Verbiese, N. ATLA, 1996, 24, 81-106. Arima, H.; Miyajib, T.; Irie, T.; Hirayama, F.; Uekamaa, K. Eur. J. Pharm. Sci., 1998, 6, 53-59. Esposito, E.; Cortesi, R.; Drechsler, M.; Paccamiccio, L.; Mariano, P.; Contado, C.; Stellin, E.; Menegatti, E.; Bonina, F.; Puglia, C. Pharm. Res., 2005, 22, 2163-2173. Ganem-Quintanar, A.; Silva-Alvarez, M.; Alvarez-Roman, R.; Casas-Alancaster, N.; CazaresDelgadillo, J.; Quintanar-Guerrero, D. J. Nanosci. Nanotechnol., 2006, 6, 3235-3241. Herkenne, C.; Naik, A.; Kalia, Y.N.; Hadgraft, J.; Guy, R.H. Pharm. Res., 2006, 23, 1850-1856. Herkenne, C.; Naik, A.; Kalia, Y.N.; Hadgraft, J.; Guy, R.H. J. Invest. Dermatol., 2007, 127, 887-894. Hostnek, J.J.; Dreher, F.; Maibach, H.I. Food Chem. Toxicol., 2006, 44, 539-1543. Lodén, M.; Akerstrom, U.; Lindahl, K.; Berne, B. Int. J. Pharm., 2004, 284, 23-30. Wagner, S.; Merfort, I. J. Pharm. Biomed. Anal., 2007, 43(1), 32-8. Wagner, H.; Kostka, K.-H.; Lehr, C.-M.; Schaefer, U.F. J. Pharm. Sci., 2002, 91, 1915-1921. Pellanda, C.; Ottiker, E.; Strub, C.; Figueiredo, V.; Rufli, T.; Imanidis, G.; Surber, C. Arch. Dermatol. Res., 2006, 298, 221-230. Lboutounne, H.; Chaulet, J.F.; Ploton, C.; Falson, F.; Pirot, F. J. Control. Release, 2002, 82, 319-334. Fresno-Contreras, M.J.; Jiménez-Soriano, M.M.; Ramírez-Diéguez, A. Int. J. Pharm., 2005, 297, 134145. Padula, C.; Colombo, G.; Nicoli, S.; Catellani, P.L.; Massimo, G.; Santi, P. J. Control. Release, 2003, 88, 277-285. Bashir, S.J.; Dreher, F.; Chew, A.L.; Zhai, H.; Levina, C.; Stern, R.; Maibach, H.I. Int. J. Pharm., 2005, 292, 187-194. Abdulmajed, K.; Heard, Ch. M. Int. J. Pharm., 2004, 280, 113-124. Hara, M.; Verkman, A.S. Proc. Natl. Acad. Sci. USA, 2003, 100, 7360-7365. Kalia, Y.N.; Alberti, I.; Naïk, A.; Guy, R.H. Skin Pharmacol. Appl. Skin Physiol., 2001, 14, 82-86. Morgan, C.J.; Renwick, A.G.; Friedmann, P.S. Br. J. Dermatol., 2003, 148, 434-443. Jarvis, C.A.; McGuigan, C.; Heard, C.M. Pharm. Res., 2004, 21, 914-919. Strid, J.; Hourihane, J.; Kimber, I.; Callard, R.; Strobel, S. Eur. J. Immunol., 2004, 34, 2100-9. Inoue, J.; Yotsumoto, S.; Sakamoto, T.; Tsuchiya, S.; Aramaki, Y. J. Control. Release, 2005, 108, 294- 305. Inoue, J. and Aramaki, Y. Vaccine, 2007, 22, 1007-13. Liu, L.-J.; Watabe, S.; Yang, J.; Hamajima, K.; Ishii, N.; Hagiwara, E.; Onardi, K.; Xin, K.-Q.; Okuda, K. Vaccine, 2002, 20, 42-48. Ayala-Bravo, H.A.; Quintanar-Guerrero, D.; Naik, A.; Kalia, Y.N.; Cornejo-Bravo, J.M.; GanemQuintanar, A. Pharm. Res., 2003, 20, 1267-1273. Reddy, M.B.; Stinchcomb, A.L.; Guy, R.H.; Bunge, A.L. Pharm. Res., 2002, 19, 292-298. Jakasa, I.; Calkoen, F.; Kezic, S. J. Chromatogr. B, 2004, 811, 77-182. Jakasa, I.; Verbek, M.M.; Esposito, M.; Bos, J.D.; Kezic, S. J. Invest. Dermatol., 2007, 127, 129-134. Tsai, J.-C.; Shen, L.-C.; Sheu, H.M.; Lu, C.-C. Arch. Dermatol. Res., 2003, 295, 169-174. Park, T.-H.; Park, C.-H.; Ha, S.-K.; Lee, S.-H.; Song, K.-S.; Lee, H.-Y.; Han, D.-S. Nephrol. Dial. Transplant., 1995, 10, 2269-2273. Alvarez-Román, R.; Naik, A.; Kalia, Y.N.; Guy, R.H.; Fessi, H. Pharm. Res., 2004, 21, 1818-1825. Sarveiya, V.; Risk, S.; Benson, H.A.E. J. Chromatogr. B, 2004, 803, 225-231. Lademann, J.; Schanzer, S.; Richter, H.; Pelchrzim, R.V.; Zastroe, L.; Golz, K.; Sterry, W. J. Cosmet. Dermatol., 2004, 3, 62-68. Mavon, A.; Miguel, C.; Lejeune, O.; Payre, B.; Moretto, P. Skin Pharmacol. Physiol., 2007, 20, 1020. Brian, K.R.; Green, D.M.; Lalko, J.; Api, A.M. Toxicol. In Vitro, 2007, 21, 133-138. Teichmann, A.; Jacobi, U.; Waibler, E.; Sterry, W.; Lademann, J. Contact Dermatitis, 2006, 54, 5-13.
TS Method as a Valuable Tool [111] [112] [113] [114] [115] [116] [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134]
[135] [136]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 227
Jacobi, U.; Tassopoulos, T.; Surber, C.; Lademann, J. Arch. Dermatol. Res., 2006, 297, 303-310. Van der Merwe, D.; Brooks, J.D.; Gehring, R.; Baynes, R.E.; Monteiro-Riviere, N.A.; Riviere, J.E. Toxicol. Sci., 2006, 89, 188-204. De Jongh, C.M.; Jakasa, I.; Verberk, M.M.; Kezic, S. Br. J. Dermatol., 2006, 154, 651-657. Lundgren, L.; Skare, L.; Lidén, C. Ann. Occup. Hyg., 2006, 50, 95-103. Nylander-French, L.-A. Ann. Occup. Hyg., 2000, 44, 645-651. Leveque, J.L.; Corcuff, P.; de Rigal, J.; Agache, P. Int. J. Dermatol., 1984, 23, 322-329. Denda, M.; Koyama, J.; Hori, J.; Horii, I.; Takahashi, M.; Hara, M.; Tagami, H. Arch. Dermatol. Res., 1993, 285, 415-417. Somerville, D.A. Br. J. Dermatol., 1980, 81, 248-258. Schurer, N.Y.; Plewig, G.; Elias, P.M. Dermatologica, 1991, 183, 77-94. Cua, A.B.; Wilhelm, K.P.; Maibach, H.I. Skin Pharmacol., 1995, 8, 246-251. Fluhr, J.W.; Dickel, H.; Kuss, O.; Weyher, I.; Diepgen, T.L.; Berardesca, E. Br. J. Dermatol., 2002, 146, 770-776. Jacobi, U.; Gautier, J.; Sterry, W.; Lademann, J. Dermatology, 2005, 211, 312-317. Fatouros, D.G.; Groeninka, H.W.M.; De Graaff, A.M.; Van Aelst, A.C.; Koertenc, H.K.; Bouwstra, J.A. Eur. J. Pharm. Sci., 2006, 29, 160-170. Weerheim, A.; Ponec, M. Arch. Dermatol. Res., 2001, 293, 191-199. Alberti, I.; Kalia, Y.N.; Naik, A.; Bonny, J. D.; Guy, R. H. J. Control. Release, 2001, 71, 319-327. Kalia, Y.N.; Pirot, F.; Guy, R. H. Biophys. J., 1996, 71, 2692-2700. Alouf, J.E.; Muller-Alouf, H. Int. J. Med . Microbiol., 2003, 292, 429-440. George, A.; Rubin, G. Br. J. Gen. Pract., 2003, 53, 480-487. Kugelberg, E.; Norström, T.; Petersen, T.K.; Duvold, T.; Andersson, D.I.; Hughes, D.E. Antimicrob. Agents Chem., 2005, 49, 3435-3441. Forslund, O.; Lindelöf, B.; Hradil, E.; Nordin-Stenquist, B.; Kirnbauer, R.; Slupetzky, K.; Dillner, J. J. Invest. Dermatol., 2004, 123, 388-394. Malmminen, M.; Koivukangas, V.; Peltonen, J.; Karvonen, S.-L.; Oikarinen, A.; Peltonen, S. Br. J. Dermatol., 2003, 149, 255-260. Sekkat, N.; Kalia, Y.N.; Guy, R.H. J. Pharm. Sci., 2002, 91, 2376-2381. Zhai, H.; Poblete, N.; Maibach, H.I. Int. J. Dermatol., 1998, 37, 386-389. Choi, M.J.; Zhai, H.; Kim, J.-H.; Maibach, H.I. Tape stripping method versus stratum corneum. In: Zhai, H.; Wilhelm, K.P. and Maibach, H.I. (eds.): Dermatotoxicology, 7th edition. CRC Press, Boca Raton, 327-337, 2008. Löffler, H.; Weimer, C.; Dreher, F.; Maibach, H.I. Parameters influencing stratum corneum removal by tape stripping. In: Zhai, H.; Wilhelm, K.P.; and Maibach, H.I. (eds.): Dermatotoxicology, 7th edition. CRC Press, Boca Raton, 339-342, 2008. Escobar-Chávez, J.J.; Merino-Sanjuán, V.; López-Cervantes, M.; Urbán-Morlán, Z.; Piñón-Segundo, E.; Quintanar-Guerrero, D.; Ganem-Quintanar, A. J. Pharm. Pharmaceut. Sci., 2008, 11, 104-130.
228
Frontiers in Drug Design & Discovery, 2009, 4, 228-246
Methylphenidate Extended-Release Capsules: A New Formulation for Attention-Deficit Hyperactivity Disorder Pilar García-García1, Francisco López-Muñoz1,*, Juan D. Molina2, Roland Fischer3 and Cecilio Alamo1 1
Department of Pharmacology, Faculty of Medicine, University of Alcalá, Madrid, Spain; 2Acute Inpatients Unit, Dr. Lafora Psychiatric Hospital, Madrid, Spain and 3 Medice chem.-phar., Fabrik Pütter GmbH & Co.KG, Iserlohn, Germany Abstract: In recent years Attention Deficit Hyperactivity Disorder (ADHD) has been the focus of growing interest, and different drugs have been introduced for its treatment. Thus, there is a range of medication for ADHD, but new formulations are necessary for more individualized therapy. The choice will depend upon the circumstances and detailed assessment. A new extendedrelease formulation of methylphenidate (Medikinet®) has increased the drugdelivery treatment options for ADHD. Medikinet® combines the advantages of immediate-release (IR) and extended-release (ER) formulations of methylphenidate, with rapid onset and prolonged duration of action (7-8 h), in a single dose intended for once-daily administration. The concentration-time profile is achieved through the particular formulation of Medikinet®, whose hardgelatine capsule contains 50% enteric-coated and 50% uncoated pellets, providing both a first, immediate release and a second, delayed release. The coated pellets only dissolve at a pH > 5.5 and release the active drug in a sustained way into the intestine. There is no difference in the bioavailability of the IR/ER product when administration follows a normal or high-calorie breakfast. Medikinet® also shows a bioavailability comparable to that of the b.i.d. 10 mg immediate release regime, as well as a high level of efficacy and good tolerability. In this review, we describe the pharmacokinetics of Medikinet® and compare its characteristics with those of other formulations used for treating ADHD.
INTRODUCTION Attention-deficit hyperactivity disorder (ADHD) is one of the neuropsychiatric conditions that has received most attention in the scientific literature in the last few years [1]. The majority of authors acknowledge that ADHD is the most widespread problem in developmental neurology and one of the commonest reasons for neuropaediatric consultations [2-4]. The prevalence of this disorder is generally accepted to be around 5-6% in *Corresponding Author: Tel: +34 91 724 8210; Fax: +34 91 724 8205; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 229
children of school age, and from 1–6% in the adult population [3,5-9]. ADHD is, therefore, a highly prevalent disorder across the developmental spectrum [9]. The core symptoms of ADHD include attention deficit, hyperactivity and impulsiveness, which should be reviewed in the context of each age group [10]. This disorder also has a wide range of consequences, the commonest being retarded academic progress, comprehension difficulties when reading, instability of relationships with friends and classmates, low selfesteem, and disorganization. Children suffering from this disorder are more likely, as juveniles, to be involved in traffic violations and car accidents and to have early and uncontrolled sexual relations, which can lead to sexually-transmitted diseases and unwanted paternity or maternity [4,11-13]. Furthermore, they generate a total medical cost and use of medical resources far in excess of those of their peers without ADHD. Pharmacological treatment of this disorder dates back more than 50 years, with the use of psychostimulants, the type of drug that has been subjected to most research in children with ADHD, and whose efficacy has been demonstrated in a large quantity of clinical trials [4,9,14-16]. Psychostimulants constitute the primary pharmacotherapy for children diagnosed with ADHD, and between 73% and 94% of children with ADHD respond to stimulants, such as methylphenidate, dexamphetamine and pemoline [17-19]. Methylphenidate is the best known and most widely used stimulant for treating ADHD in children, accounting for at least 70% to 90% of ADHD drug therapy, while dexamphetamine and pemoline are generally regarded as second-line therapies (and other therapeutic agents are still at the development stage). Data show that dexamphetamine is equally effective as methylphenidate, but the pharmacological class of amphetamines is used more reluctantly than methylphenidate. Indeed, in most European countries, dexamphetamine is not commercially available. Pemoline is available under restricted use, as it can cause an increase in liver enzymes and, very rarely, irreversible hepatotoxicity. Therefore, it is not considered as a drug of first choice [10,12,19]. In recent years, pharmacological research in this field, rather than focusing on the search for new therapeutic drugs, has launched itself into the development of new formulations of stimulants [20-23], aimed at achieving better control of children with ADHD through a single daily dose. However, there has also been research with nonstimulant drugs, such as atomoxetine, bupropion or modafinil. Even so, it is the new formulations with methylphenidate that are arousing the most therapeutic interest, given their advantages with regard to the management of ADHD, and consequently, to the improvement of its symptoms [4,10,18,24]. ATTENTION-DEFICIT HYPERACTIVITY DISORDER The combination of inattentive, impulsive and hyperactive behaviour is recognized as a clinically relevant disorder when these symptoms are serious, when mental development is inappropriate and when there is deterioration of social relations in the school and family contexts [10]. This behaviour disorder is currently known as attention deficit disorder (ADD) or attention-deficit hyperactivity disorder (ADHD), though over the course of the history of this pathology it has been referred to by other names, such as hyperkinesis, hyperkinetic syndrome or minimal brain dysfunction (MBD). The first clinical approximation to ADHD is attributed to the German doctor and man of letters Heinrich Hoffman, who in 1854 called his son “Fidgety Philip”. Later, in 1902, British pediatrician George Frederic
230 Frontiers in Drug Design & Discovery, 2009, Vol. 4
García-García et al.
Still referred to mental conditions related to abnormality of moral control (referring to the cognitive capacity to distinguish good from bad) in children of normal intelligence. At the beginning of the 20th century the principal hypothesis for explaining children’s hyperactivity was based on brain dysfunction. Thus, the initial conception involved minimal brain lesion, but by the late 1950s this had become minimal brain dysfunction. In 1960, Stella Chess introduced the idea of hyperactive child syndrome, considering it to be a developmental diagnosis with good prognosis. By the 1970s the triad of symptoms that currently define ADHD had been identified: attention deficit, hyperactivity and impulsiveness. In 1972, the conceptualization of this disorder took a radical turn thanks to the contribution of Virginia Douglas, of McGill University, who proposed that its chief symptom was attention deficit, rather than hyperactivity. Finally, the Diagnostic and Statistical Manual of the American Psychiatric Association (DSM-III, 1980) recorded attention deficit disorder with and without hyperactivity [25,26]. Currently, according to the International Classification of Diseases (ICD) [27] and the DSM [28], the terms mainly used are attention-deficit hyperactivity disorder (ADHD) and hyperkinetic disorder (HKD), which can be considered synonymous [29,30]. A review of the historical development of the ADHD concept reveals that this pathology is by no means of recent discovery, though today it is receiving more attention, its diagnosis is being improved and it is becoming better understood. Thus, both the DSM-IV and the ICD-10, used as diagnostic criteria by specialists in ADHD, break this disorder down into three core symptoms: attention deficit, hyperactivity and impulsiveness [20,31]. ADHD is, therefore, a persistent pattern of inattention and/or hyperactivity/impulsiveness which is more frequent and severe than that normally observed in subjects of a similar developmental level. Some symptoms must necessarily appear before age 7, but it is only from this age onwards that the condition is usually diagnosed. Symptoms must occur in two different contexts: that of school or work and that of family. Within the different subtypes of ADHD identified, the commonest is the combined form (60%), which encompasses the inattentive (30%) and hyperactive/impulsive (10%) forms. With regard to gender, ADHD is more prevalent in boys, the boy:girl ratio being 3:1, though the literature includes ratios of up to 9:1 [32-33], and by the subtypes mentioned above the ratio is 4:1 for the hyperactive subtype and 2:1 for that of inattention [25]. According to the DSM-IV, the estimated prevalence of ADHD is 3% to 5%. However, extensive epidemiological studies have demonstrated that the prevalence of symptoms of ADHD in children is 9% to 20%, depending on the methodology used to diagnose the disorder [27,30,32,34]. As ADHD is often associated with comorbid disorders, it is important to seek evidence of other conditions, such as stress disorders and adjustment disorders. A recent comprehensive review of literature reported the rates of comorbidity for clinically referred children with ADHD at 30-50 % for conduct disorders, 15-75% for mood disorders, approximately 25% for anxiety disorders and between 10% and 90% for learning disorders [30,34-36]. There is growing consensus that ADHD continues through adolescence and often also into adulthood, although in slightly altered form. It is estimated that about 70% of children will take their ADHD into adolescence, and 10% into adulthood. Its high prevalence in childhood, combined with the follow-up results, suggests that approximately 2%
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 231
of adults may suffer from ADHD. This would make ADHD a relatively common adult disorder that may be under-identified in adult psychiatric clinics [34,37]. To diagnose ADHD, clinicians should elicit the history of specific symptoms from those who know the child best – usually the parents and teachers. Rating scales with specific ADHD symptoms have been developed, and provide a systematic approach for documenting clinical history. Additionally, various psychometric instruments can be used to rate behaviour and performance and to measure improvement of ADHD. Direct observation in clinical settings may often not confirm reports from parents or teachers. If confirmation of subjective reports is necessary, then observation in the natural settings of home or school is recommended [8,36]. Psychological tests, especially with tasks of attention/concentration and learning, are supportive methods for diagnosis [30,38], but it is still difficult to separate cause and consequence. Attempts to clarify the pathophysiology of ADHD have been made from various perspectives, but the etiology of ADHD remains unknown [32,34,39-41]. However, there would appear to exist a considerable genetic basis that makes these children vulnerable to other factors, such as obstetric and postnatal complications, as well as other environmental factors (smoking during pregnancy, toxic agents such as lead, etc.). Thus, the frequency of ADHD and of other psychiatric disorders is greater in first-order relatives of patients with this illness, though what is inherited may be not the clinical manifestations but rather a particular neurobiological vulnerability [8,25], as suggested by studies of families, comparisons of twins, and studies of adopted children, which provide some support for possible genetic transmission. Genetic influences of around 60% heredity for some forms of ADHD have been reported. Genes influencing the dopaminergic systems may play a role in genetic transmission, and it has been suggested that the dopamine transporter gene and the D4 dopamine receptor gene are associated with ADHD [32,34,41]. Current research supports a neurobiological cause of ADHD. Neurochemically, ADHD seems to be linked to an imbalance between the dopaminergic, noradrenergic and serotonergic neurotransmitter systems, characterized by decreased dopamine (DA) and increased noradrenaline (NA) [42]. The neuroanatomical location of deficits seems to be in the cortical and subcortical areas, particularly the frontal lobe and the prefrontal lobe. Positron emission tomography (PET) scans reveal decreased cerebral metabolism in the premotor and prefrontal superior cortex, as well as diminished blood flow to the corpus striatum in ADHD children [43]. Both NA and DA participate in numerous cognitive processes (attention, alertness, executive function, etc.) that appear to function defectively in ADHD. The maintenance of high levels of vigilance and attention depend on the catecholamine modulation of the prefrontal, cingulate and parietal cortices, as well as of the striate nucleus, the hippocampus and the thalamus. In all these structures there is substantial participation of catecholaminergic neurons. Although the physiopathology of ADHD has not been fully clarified, there appears to be a dysregulation of DA and NA levels that would lead to an alteration in the frontallymediated executive functions, and the adrenergic imbalance could mean that the psychological functions necessary for appropriate responses to environmental stimuli become altered. The prefrontal cortex depends on adequate dopaminergic and noradrenergic innervation for maintaining optimum functioning, as pointed out previously. Thus, dopaminergic hypofunction is the pathogenic core of ADHD, in which there is an increase in
232 Frontiers in Drug Design & Discovery, 2009, Vol. 4
García-García et al.
the dopamine transporter, suggesting a dysfunctional regulation of dopamine or of its receptors, which will lead in turn to a smaller quantity of dopamine in the synaptic cleft. The consequence of catecholaminergic dysregulation is that processing of external and internal stimuli is affected, so that impulsive behaviours would indicate a failure in the system for delayed gratification, while excessive motor activity would indicate a failure in the systems of inhibition, which involve frontal and limbic structures [39,40]. THERAPEUTIC MANAGEMENT OF ADHD Two treatment modes can be distinguished for ADHD; on the one hand, the pharmacological mode, which tends to be applied in conjunction with psychological treatment, and on the other, the non-pharmacological mode, which would include diverse therapies such as educational, cognitive-behavioural and other psychological and psychiatric approaches [8,17,25,36,44]. The majority of authors recommend multimodal treatment, which combines pharmacotherapy and psychological treatment. It should be borne in mind that without appropriate treatment for each patient, ADHD can lead, as we have pointed out, to a range of circumstances – academic failure, social stigmatization, and so on. Moreover, the disorder can cause a deterioration of occupational and social functioning and can have a negative effect on emotional stability [12,17]. The treatment regimes mentioned emerged from the MTA study (Multimodal Treatment Study of Children with Attention-Deficit/Hyperactivity Disorder) carried out in the United States, one treatment arm of which was called “Community Care”, in which the treatment of ADHD was not stipulated by the study protocol. It can be assumed that treatment selection in this study arm reflects the customary treatment decisions for ADHD in the USA. Medication was given to 67% of these children. Of the 97 children with medication in this arm, 84 (87%) were treated with methylphenidate, while only 6 (6%) were treated with amphetamines. There are considerable cultural differences in the use of stimulants. In Europe, where their prescription has been restricted by custom and by law, clinical guidelines recommend initially a rigorous regime of multiple psychosocial interventions, such as behaviour modification, cognitive therapy, family therapy and teacher consultation; in North America, where the prescription of stimulants has been generally accepted for decades, clinical guidelines recommend an initial pharmacological therapeutic regime [12,17,19,44]. The principal conclusions of the MTA study suggest that a careful medication regime (mostly methylphenidate) is superior to behavioural treatment alone. Medication therapy combined with behavioural treatment appears to be the most beneficial treatment for ADHD, especially in children with additional anxiety disorders [12,17,45]. Thus, the drug most widely used in these patients is methylphenidate, whose efficacy is endorsed by numerous studies and indeed by its consistent use over five decades [25,45,46]. At the same time, it is important to stress that the treatment of ADHD patients is complex and multidisciplinary, being conditioned by a large number of factors, some intrinsic, such as the individual characteristics of each patient, sex, age, etc., and some extrinsic, such as family and social environment. In sum, therapy should be based on four therapeutic pillars: the family, pedagogical, psychological and pharmacological approaches. The general goal of the treatment should be to improve the child’s cognitive, behavioural and social functioning and increase his/her self-esteem, with the minimum
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 233
of side effects. Of all therapeutic approaches, the combined (or multimodal) treatment is that which, in general, appears to offer the best results at the present time. METHYLPHENIDATE The various guidelines for the treatment of ADHD concur on the fact that stimulants of the central nervous system are extremely effective in the improvement of ADHD symptoms. Prominent among such stimulants is methylphenidate (MPH), as already mentioned. Indeed, the pharmacology of MPH has become a focus of major interest in the wake of the development of different controlled-release formulations with particular drug-delivery and pharmacokinetic profiles [10,16,18,19,47-51]. The mechanism by which MPH produces psychostimulant effects appears to depend prominently upon the facilitation of catecholaminergic neurotransmission. MPH is thought to act as an indirect sympathomimetic substance, but predominantly via dopamine (DA) transmission, and less via the norepinephrine (NE) or serotonin (5-HT) systems. Recognizing that MPH binds with high affinity to the DA transporter or uptake channel, it has been proposed that such binding blocks the synaptic clearance of impulse-released DA, leading to prolonged post-synaptic neurochemical mediation [44,52]. Methylphenidate hydrochloride (methyl 2-phenyl-2-(2-piperidyl)acetate hydrochloride) is a basic ester of phenylacetic acid (Fig. 1). It is formulated as freely soluble hydrochloride salt. The molecular structure of MPH contains a basic phenylethylamine moiety which is common to psychostimulant agents such as amphetamines, and is thought to be responsible for its amphetamine-like action profile. The presence of two chiral centres in the structure of MPH allows four possible stereoisomers, but all current MPH products, as well as the product-marketing authorization applied for, contain the drug in the racemic form, a 50:50 mixture of the threo-R,R (+)- and threo-S,S (-)isomers. The threo-R,R (+)-stereoisomer appears to be almost exclusively responsible for the catecholaminergic effects of racemic MPH [53].
* N H
*
CH3O
* HCl O
Fig. (1). Molecular structure of methylphenidate hydrochloride.
Pharmacokinetics of Methylphenidate The pharmacokinetic properties of MPH hydrochloride, shown in Table 1, are thoroughly described in the literature [10,25,44,54-57]. As known from other psychoactive drugs, there is a considerable range of inter-individual variation in MPH serum levels. In addition, serum levels of MPH show intra-individual variations – that is, a given subject may have different serum levels on different days. Therefore, plasma levels have not proven useful in standard clinical practice. Data regarding the correlation between serum levels of MPH and clinical response are contradictory [25,55,58].
234 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 1.
García-García et al.
Pharmacokinetic Parameters of the Different Formulations of Methylphenidate
Parameters
Concerta® 1 Capsule of 18 mg
Rubifen® 1 Capsule of 10 mg*
Medikinet ® 1 Capsule of 20 mg
% immediate release
22
100
50
% extended release
78
0
50
Cmax (ng/mL)
3.7
9
6.4
6.8
1-2
2.75
AUC0-inf (ng.h.ml )
41.8
Interindividual variability
48.9
T1/2 (h)
3.5
2
3.2
Duration of effect (h)
12
3-4 h
7-8
Tmax (h) -1
*One dose of Concerta is equal to three doses of Rubifen; Cmax: Maximum concentration; Tmax; time necessary for attaining maximum concentration; AUC: area under the curve; t1/2: half-life of elimination.
Methylphenidate is a basic drug, and as such is poorly bound (approximately 15%) to plasma protein. Low plasma binding makes it highly amenable to crossing the bloodbrain barrier. Typical therapeutic doses of MPH provide a Tmax of 1.5 - 2.5 hours, reaching a Cmax of 6 - 15 ng/ml with a T1/2 of 2 - 3.5 hours. Pharmacokinetic parameters for children and adults appear to be comparable [10,59]. Animal studies (rats) have shown that MPH accumulates in easily perfused tissues, favouring kidney > lung > brain > heart > liver. Brain concentration of MPH is about eight times that of serum. Methylphenidate is rapidly metabolized and excreted in urine, small amounts also appearing in the faeces [10,60,61]. The major metabolic route is the hydrolysis of the ester function to the corresponding carboxylic acid, commonly known as ritalinic acid (2-phenyl-2-piperidyl acetic acid), which accounts for approximately 80% of the dose. Minor metabolites include p-hydroxy, oxo, and conjugated derivatives of MPH. A small portion of MPH is metabolized by the hepatic microsomal oxidase system. Less than 1% appears in the urine as unchanged MPH [10,61]. METHYLPHENIDATE FORMULATIONS There are three pharmaceutical formulations of MPH: immediate release or shortacting formulation, sustained release or intermediate-acting methylphenidate, and extended release or long-acting methylphenidate [9,10,23,57]. Immediate-release methylphenidate (Ritalin®, Rubifen®), whose pharmacokinetic properties are shown in Table 1, releases 100% of the methylphenidate in the capsule on administration. This formulation has a therapeutic effect of 2 to 4 hours, which means that in many cases two or three administrations are necessary to achieve a sustained effect [62]. Moreover, the abrupt decrease of plasmatic concentrations of MPH can produce a rebound effect. Because of the relatively short half-life of two to (at most) four hours, in many patients a single early-morning dose of rapid-release methylphenidate
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 235
results in the ADHD symptoms increasing again during the latter part of the school morning; this also creates considerable difficulties in the family context at homework time [63]. Furthermore, rebound effects are not infrequently seen in school as the effect subsides. Many affected children, therefore, require a second pre-midday dose while still at school. D-methylphenidate (Focalin®) is a racemic variant of methylphenidate marketed in the United States. This is a development of the classic formulation, from which the relatively inactive L isomer has been removed. Thus, this product contains only the D isomer, so that with half the dose the same effect is obtained as with the conventional formulation [10,64-66]. With the aim of solving the problem of administering multiple doses of MPH, modified release systems were designed. Both in the United States and in Europe there are different modified systems of release of MPH and other drugs used in the treatment of ADHD, which means that the products have different characteristics (Table 1). Among the specific formulations of methylphenidate are the SODAS technology, the CD system, the OROS technology, the transdermic release (skin patch) system and, finally, capsules containing different types of pellets. The present review will deal in depth with the pharmacological characteristics of this last-named formulation. Methylphenidate administration with the SODAS (Spheroidal Oral Drug Absorption System) technology, whose representative is Ritalin LA®, involves the use of a capsule made up of a shell containing half the MPH dose in the form of immediate-release MPH, while the other half has an enteric protection layer that permits release of the active agent after 4 hours. In reality, the LA formulation mimetizes, in a single application, the immediate-release administration of MPH twice, separated by 4 hours. The main advantage of this formulation is convenience of administration [10,67,68]. Methylphenidate CD, whose commercial name is Metadate CD®, is characterized by the combination, in a single capsule, of 30% of immediate-release MPH and 70% of slow-release MPH. In this way a prolonged effect, sustained over 9 hours, is achieved. From the pharmaceutical perspective it uses a technique similar to the SODAS technique [10,67-69]. Methylphenidate with OROS (Concerta®) technology consists in an osmotic-release capsule. Each capsule includes a shell of immediate-release MPH and three compartments, two with MPH and another with an osmotic polymer/polymeric agent, coated with a semi-permeable membrane. After oral administration, the coating of the capsule provides immediate release of 22% of the dose. From that point on, the osmotic compartment becomes hydrated due to the passing of intestinal juices through the semipermeable membrane, and increases in volume, acting as a plug. The OROS formulation provides two-phase kinetics with two peaks of concentration, corresponding to the two periods of MPH release, with total exposure to the drug equivalent to 3 doses of immediate-release MPH. Fluctuations of plasma concentration of the drug are fewer than in the case of repeated administration of immediate-release stimulants, thus eliminating the daily variation of pharmacological effects associated with the older formulations. MPH with the OROS release system was designed to replace the three-administration regime, morning, midday and evening, since with its use the second peak of plasma concentration occurs later than with other sustained-release systems. This means that the therapeutic effect lasts 12 hours, which can result in insomnia or lack of appetite in the
236 Frontiers in Drug Design & Discovery, 2009, Vol. 4
García-García et al.
evening, with the consequent negative effect on children and secondary effects on the family. In such cases it would be advisable to use a shorter release system [10,19,48, 68,70-72]. In April 2006 the Food and Drugs Administration (FDA) approved the first skin patch for MPH release (Daytrana®), though it has yet to be authorized in Europe. The patch has been patented with the so-called “DOT Matrix” technology. This dot-matrix transdermic technology uses a semi-solid suspension of microscopic cells of the drug concentrated uniformly and dispersed by a silicon glue. The steep diffusion gradient between each cell of the drug and the skin means that the drug adequately penetrates the skin. With this release system the methylphenidate passes directly into the bloodstream, providing constant levels throughout the day. The patch must remain on the skin for 9 hours, even during the child’s everyday activities, including swimming, exercise and bathing. Clinical trials have shown that the patch, fixed to the skin for 9 hours, has an effect lasting 12 hours [73-76]. METHYLPHENIDATE EXTENDED-RELEASE CAPSULES: OVERVIEW OF BIOPHARMACEUTICS The aim of the development process of the newly-developed modified-release formulation of MPH hydrochloride (Medikinet® retard modified-release capsules) (Fig. 2) was to optimize broaden the range of existing MPH modified-release formulations, as Metadate CD®, which was approved by the FDA [69,77]. Methylphenidate extended-release capsules consist of two fractions of active substance in a 1:1 ratio. The hard-gelatine Medikinet® capsules contain two types of pellets in equal proportion: 50% immediaterelease at the gastric level (white pellets) and 50% extended-release at the intestinal level, which have a gastro-resistant coating that permits the delayed release of MPH (blue pellets) (Fig. 3).
Fig. (2). Capsule of Medikinet®.
The retard pellet has two layers with different characteristics: an outer releasedelaying layer (enteric coat) and an inner methylphenidate layer. The enteric coating of the outer layer comprises (co-)polymers of (meth)acrylic acid and (meth)acrylate con-
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 237
taining carboxyl groups, thereby causing sustained release of the psychostimulant in vivo. The addition of an alkaline agent to this formulation results in a partial neutralization of the carboxyl groups of the polymer forming the enteric coating, and thus in the formation of small channels in the enteric coating which allow a slight diffusion of the MPH through the coating even at a pH of under 5.5, which is typical in the stomach of human patients. This formulation results in optimum plasma concentration in vivo.
e
Fig. (3). Pellets of Medikinet®.
The in vivo release profile of the Medikinet® capsules, as shown in Fig. (4), was designed so that the first portion would be released in the stomach’s acid medium immediately, within 30 minutes of ingestion, and the modified-release portion would be released only when pH values were higher than 5.5, this second release therefore taking place in the intestine (see Fig. 4). Studies have shown that MPH is rapidly and almost completely absorbed from the gastrointestinal tract. Medikinet® features a modified pharmacokinetic profile with prolonged Tmax and T1/2 values, offering more stable daily profiles of the active drug level, and combines the advantages of a rapid influx rate with the development of a relatively prolonged plateau phase [78]. Influence of Food The influence of food on rate and extent of absorption is a controversial issue. Various studies have shown that food does not affect the pharmacokinetics of MPH [79], though others claim that food modifies absorption rate, accelerating it. However, a more recent study showed that food produces a significant increase in the duration of absorption, but does not affect the final rate [80]. When the medication is taken in the morning after breakfast, the non-delayed release portion of the capsule is rapidly dissolved and an initial maximum concentration is achieved after a mean of 2 hours. Methylphenidate is then released from the sustainedrelease part of the capsule, and contributes to creating a plateau phase during which the concentration does not fall below 75% of the maximum achieved concentrations [81].
238 Frontiers in Drug Design & Discovery, 2009, Vol. 4
García-García et al.
25
Amount released [mg]
20
15 pH 1.2
pH 6.8
10
5
0 0
20
40
60
80
100
120
140
160
Time [min]
Fig. (4). Release profile of methylphenidate from Medikinet retard® capsules.
After single daily doses of 20 mg, 30 mg and 40 mg of Medikinet®, the maximum concentrations and the areas under the concentration-time curves are proportional to the administered dose. Because of the short half-life of MPH, the modified-release principle also does not result in accumulation of the active drug when the prescribed dosing interval is maintained. In a balanced, randomized cross-over design, the bioavailability of MPH from one 10 mg Medikinet® capsule (single dose) was compared to 10 mg MPH water solution in 12 healthy volunteers, all recruited out of the 24 from the preceding study. Medikinet® is bioequivalent to the reference solution in terms of extent of absorption, but it shows a lower peak concentration at a later time point compared to the reference. Cmax is outside common bioequivalence limits, and Tmax is delayed by 1.92 hours. The plasma concentration time curve suggests that when the drug is taken with food, retardation occurs as the Cmax summit ends in a “plateau-like” phase. Nevertheless the influence of food on the kinetic behaviour of the substance was not very pronounced, but rather demonstrated a “standard” profile. Apart from this, the results match the theoretical expectations for a hybrid formulation such as Medikinet retard® with respect to Cmax and Tmax, where bioequivalence to the “immediate release” solution would not be expected. Within the assessment of this study an additional exploratory evaluation was carried out, comparing the results of the same 12 volunteers’ participation in the two studies from the 20 mg Medikinet retard® sequence (fasting), with the 10 mg Medikinet retard® sequence (highcalorie fed), normalized to 20 mg assuming linear kinetics. In order to further elucidate the influence of food intake it was attempted to explore whether the kind of food taken as breakfast does influence the bioavailability, since children (at least in Europe) most likely do not eat a high-calorie, high-fat breakfast. A “normal” child’s breakfast was now introduced in one of the sequences. In this balanced,
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 239
concentration [ng/ml]
randomized, crossover, open, four-way, single-dose design the bioavailability of MPH hydrochloride from Medikinet® modified-release 10 mg and 20 mg capsules after singledose application and 10 mg Ritalin® immediate-release after b.i.d. application following a normal breakfast were compared to single-dose application of Medikinet® capsule following a high-calorie breakfast. Both Medikinet retard® formulations (10 mg and 20 mg capsules) were now dose proportional with regard to the excipients and identical to the product-marketing authorization applied for. The study was carried out with 16 healthy male and female volunteers, all 16 of whom completed the study and were considered in the statistical analysis. For better comparability, Ritalin® was also administered after a normal breakfast, and not on an empty stomach as recommended by the designer of the original study. The results (Fig. 5) demonstrate that there is no difference in bioavailability between the Ritalin® IR in a dosage of 10 mg b.i.d. and the 20 mg Medikinet® formulation given after a normal breakfast. There is bioequivalence with regard to rate and extent of absorption. There is no difference in bioavailability of the 20 mg Medikinet® formulation given after a normal and after a high-calorie breakfast. Once again there is bioequivalence with regard to rate and extent of absorption. Nevertheless, as far as the individual curves are concerned, it becomes apparent that the typical double-peak pattern is less pronounced after the high-calorie breakfast. Dose linearity is given between Medikinet® 10 mg and Medikinet® 20 mg, both administered after a normal breakfast. 8
test 1 (Ritalin® 10mg, normal BF, tau=4h)
arithmetic means (n=16)
test 2 (Medikinet® Retard 10mg, normal BF) test 3 (Medikinet® Retard 20mg, normal BF) test 4 (Medikinet® Retard 20mg, high calory BF) 6
4
2
0 0
6
12
18
24 time [h]
Fig. (5). Plots of the mean whole blood levels of methylphenidate hydrochloride over the 24-hour sampling period (BF: breakfast).
As food intake with a normal breakfast does not alter stomach acidity to above a pH of 4, the enteric-coated pellet portion of Medikinet retard® will only release the active ingredient in the duodenum, creating the prerequisites for the retardation effect. In contrast, if children show severe anacidity of the stomach, reaching ph values > 5.5, the retardation will not come into effect, and such children should not be treated with Medikinet®.
240 Frontiers in Drug Design & Discovery, 2009, Vol. 4
García-García et al.
Dosage and Frequency of Use Medikinet® modified release is a hard-gelatine capsule with different dosage options: 10, 20, 30 or 40 mg. Administration of this drug once a day is generally sufficient for achieving the therapeutic goal. Research shows that the dose for children should not exceed 60 mg per day. Thus, when the dose set by the clinician is sufficient, onset of the effect occurs in the first hour after administration of the drug, the effect being maintained until early afternoon. As mentioned earlier, duration of action of this drug is approximately 8 hours. In some cases it may be necessary to administer an additional dose of immediate-release MPH to prolong the effects in the afternoon; however, this decision should be made on the basis of the patient’s particular symptoms. In order to maintain the appropriate pharmacokinetics of Medikinet®, the drug should be administered with or after breakfast, and never on an empty stomach. Safety of the Capsule Shell One of the main problems of therapeutic compliance in children with ADHD is the difficulty for parents to administer the medication, especially in children who have trouble swallowing capsules. Although data on this matter are scarce, some authors refer to swallowing problems in 26% of patients [82,83]. In this regard, Medikinet® brings considerable advantages, especially for those patients with difficulties for swallowing the capsules, such as small children; the advantages extend to cases in which the capsule tears open or is bitten into, or in which a patient takes the capsule contents without the shell, since the capsule, as well as being swallowed whole, can be opened and its content ingested without modification of the drug’s pharmacokinetic properties. In this regard, Fischer et al. [79] carried out a study to check the bioequivalence of whole-capsule ingestion versus its content only. The resulting curve profile showed that the sustainedrelease characteristics are preserved, and that there is no immediate release of the total dose of methylphenidate (dose dumping). It can be assumed that biting the capsule, due to which some of the pellets can enter the oral cavity, does not result in a changed plasma concentration curve compared to the normal and correct practice of swallowing the entire capsule whole. Given the small size of the Medikinet® pellets it is advisable that when the drug is taken without the capsule its contents are mixed into a spoonful of semi-solid food, such as apple purée, jam or yoghurt, so as to avoid loss of pellets and the consequent decrease in efficacy [79,81]. Methylphenidate Extended-Release Capsules Versus Methylphenidate (b.i.d.) Döpfner et al. [78] carried out a study to explore whether the bioavailability of MPH extended-release capsules was equivalent to that of immediate-release MPH administered b.i.d. This multi-centre, randomized, double-blind, controlled clinical trial was carried out in children who already responded to MPH, with the aim of assessing the efficacy and tolerability of Medikinet®. A crossed design was used, comparing it with both immediate-release MPH (2 x 10mg), administered every four hours, and with placebo. The study took place in successive treatment phases of four-day observation, total observation time being limited to 2 weeks.
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 241
For the clinical assessment the SKAMP questionnaire (Swanson, Kotkin, Afler, MFlynn, and Pelham Scale) was used, and parents/caregivers assessed the children’s behaviour taking into account ADHD symptoms and rule-compliance, as well as recording the number of arithmetic problems correctly solved. Likewise, individual ADHD symptoms and aggressive behaviour, throughout the morning and afternoon activities, were assessed by means of the German questionnaires for the assessment of hyperkinetic disorders (FBB-HKS; Fremdbeurteilungsbogen für Hyperkinetische Störungen) and for social behaviour disorders (FBB-SSV; Fremdbeurteilungsbogen für Störungen des Sozialverhaltens), in accordance with Corrigan’s Agitated Behaviour Scale (ABS). In addition, the child’s activity was measured throughout the entire period (“level of agitation”) by means of an actimeter. The study included 82 schoolchildren aged 8-14, divided into groups of 5-10 patients each. The children were subjected to relatively high demands, with five concentration tests over the course of the day, but they also received throughout the study behavioural therapy and training in social adaptation, and were attended by doctors, psychologists and education experts. Dosage was identical in the two MPH treatment groups, and did not exceed 1 mg/kg of body weight. Administration of the drugs was designed in such a way that Medikinet® was taken at 9 o’clock in the morning and immediate-release MPH and placebo at 9 in the morning and 13.00 hours, using the double-simulation technique. Daily administration of a capsule of Medikinet® considerably reduced symptoms of inattention and hyperactivity in the 79 patients assessed, and the superiority of Medikinet® versus placebo (p<0.001) was demonstrated, as was the equivalence of Medikinet® and immediate-release MPH, according to the SKAMP questionnaire scores. This assessment confirmed the superiority of Medikinet® in the improvement of attention and in rule-compliance, not only during the first phase of the study in the morning, but at all 5 time points: 9.00, 11.00, 12.30, 15.00 and 16.15 hrs. As regards overall scores (FBB-HKS, FBB-SSV and ABS), they indeed revealed net differences in favour of Medikinet®. Compared to placebo, Medikinet® considerably reduced symptoms of hyperactivity, aggressiveness and social behaviour disorders (p<0.001), these results being identical to those obtained with immediate-release MPH. The hyperkinetic disorder questionnaire (FBB-HKS) was applied in the morning (prior to 13.00 hrs.) and in the afternoon (prior to 16.45 hrs.) by two parents/caregivers in each case. The mean obtained indicated, in both the morning and afternoon assessments, that no difference at all could be demonstrated between the two MPH groups. METHYLPHENIDATE EXTENDED-RELEASE CAPSULES IN THE TREATMENT OF ADHD Table 2 shows some of the most relevant studies carried out with extended-release methylphenidate, which we shall now briefly describe. Döpfner et al. [84] carried out a multi-centre, randomized, double-blind, placebocontrolled clinical trial with 85 patients aged between 6 and 16. Patients were given placebo (n=42) or Medikinet® (n=43) randomly. The principal variable of the study was assessment of ADHD symptoms, by means of the FBB-HKS scale. Safety was rated through assessment of adverse effects (SERS-D scale) and physical examination of patients. The results of this study showed a clear positive effect of the drug on ADHD symptoms (p<0.001). It was found that 72% of patients responded to treatment with
242 Frontiers in Drug Design & Discovery, 2009, Vol. 4
García-García et al.
Medikinet®, versus 26% who responded to the placebo. This study, in which intensity of the effect of Medikinet® was rated by teachers, concluded that Medikinet® displays good clinical efficacy. Teachers also reported that the effects persist or tend to persist until 13.00 hrs. (in a school timetable of 8.00-14.00 hrs.) in 65.1% of patients treated with Medikinet®, versus 19% of patients treated with placebo. Table 2.
Clinical Studies on Methylphenidate Extended-Release Capsules
Treatment Author, Year
Treatment Regime
ResultS
Control Regime
Principal Variable
Absolute Difference of Risks/Relative Risk
Complications/ Adverse Reactions
Effect size d=1.0 (Teachers) Effect size d=0.4 (Parents)
Not assessed
Sinzig et al., 2007 [86]
20-60 mg/day 5 weeks
Placebo
FBB-HKS, FBB-SSV
Döpfner et al., 2004 [78]
10-40 mg/day
Placebo MFD-imm.release
SKAMP, PERMP
p<0.01 vs Pla. NS vs MFD-IR
Not assessed
Sinzig et al., 2004 [85]
20-60 mg/día 4 weeks
Placebo
FBB-HKS
Effect size d=1.2
Not assessed
Döpfner et al., 2003 [84]
20-60 mg/day 4 weeks
Placebo
FBB-HKS
Effect size d=1.0
Tics, decreased appetite, insomnia, hyper-excitability
For their part, Sinzig et al. [85] carried out a multi-centre study in 2004, with doubleblind, randomized and placebo-controlled design, for which they recruited 102 patients aged between 6 and 16 years. ADHD diagnosis was made according to the diagnostic criteria of the DSM-IV. Likewise, specific questionnaires (FBB-HKS) were used for parents and teachers with a view to rating the effectiveness of Medikinet® in the afternoon. The parents concluded, at 4 weeks, that the medication had an intense therapeutic effect. Moreover, it was observed that, as against placebo, the symptoms decreased significantly with Medikinet®. Identical results were obtained in the assessment carried out by the teachers. The parents rated the effectiveness of Medikinet® as good or very good in 60% of cases, versus 5% for placebo. In the light of the results obtained, the authors conclude that Medikinet® reduces both hyperactivity and ADHD.
Methylphenidate Extended-Release Capsules
Frontiers in Drug Design & Discovery, 2009, Vol. 4 243
More recently, Sinzig et al. [86] published a double-blind, randomized study lasting 5 weeks, in which they assessed the response to extended-release MPH in 85 children and adolescents aged 6-16 years, diagnosed with ADHD, and who presented an oppositional-defiant and conduct disorder according to the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV). The authors reported that a total of 64.9% of the children showed oppositional defiant disorder/conduct disorder (ODD/CD) symptoms. A statistically significant effect was found in the group treated with MPH (verumgroup). Furthermore, according to Cohen’s criteria, high effects were found for aggressive symptoms in school, but not in the afternoon. There were also lower effect sizes for more severe aggressive symptoms. Likewise, a characteristic correlation was found between ODD/CD symptoms and the ADHD hyperactivity/impulsivity subscale, compared to the inattention subscale. Thus, Sinzig et al. [86] conclude that long-acting MPH is effective in the treatment of oppositional-defiant and aggressive behaviour, especially as regards milder symptoms. Also, the expected correlation between impulsivity and aggressiveness was confirmed. CONCLUSIONS The important role of psychostimulants in general and of MPH in particular in the control of ADHD symptoms is well known. Recent years have seen, moreover, an increase in the number of publications aimed at improving our understanding of both the pathology and its treatment. Likewise, new pharmaceutical formulations have emerged to swell and improve the therapeutic arsenal for ADHD. Among these is Medikinet®, a new formulation of extended-release MPH, consisting in capsules incorporating two types of pellet; white immediate-release pellets, which, thanks to their rapid absorption, permit immediate control of symptoms, and blue retarded-release pellets, which permit a total duration of action of 8 hours – an ideal duration for children who must adhere to a school timetable. Furthermore, the gradual release of MPH from the blue pellets avoids fluctuations in plasma concentration of the active agent that can lead to a rebound effect. An additional advantage of this new formulation is the possibility of opening the capsule and mixing the contents in a spoonful of semi-solid food (purée, yogurt, jam, etc.), thus facilitating administration of the drug in those patients with difficulty for swallowing. Apart from these advantages, Medikinet® has shown its clinical efficacy in the improvement of core symptoms of ADHD, as well as in patients with comorbid disorders, such as oppositional defiant disorder/conduct disorder (ODD/CD). REFERENCES [1] [2] [3] [4] [5] [6] [7] [8]
Fischman, A.J.; Madras, B.K. Biol. Psychiatry, 2005, 57, 1374-1376. Biederman, J. Biol. Psychiatry, 2005, 57, 1215-1220. Polanczyk, G.; Silva, M.; Lessa, B.; Biedermanm, J.; Rodhe, L.A. Am. J. Psychiatry, 2007, 164, 942948. López-Muñoz, F.; Alamo, C.; Quintero-Gutiérrez, F.J.; García-García, P. Eur. Child Adolesc. Psychiatry, 2008, in press. Faraone, S.V.; Sergean, J.A.; Gillberg, C.; Biederman, J. World Psychiatry, 2003, 2, 104-113. Kessler, R.C. Prevalence of ADHD in the US: Results from the NCS-R, American Psychiatric Association: Washington DC, 2004. Kessler, R.C.; Berglund, P.; Demler, O.; Jin, R.; Merikangas, K.R.; Walters, E.E. Arch. Gen. Psychiatry, 2005, 62, 593-602. Spencer, T.J.; Biederman, J.; Mick, E. J. Pediatr. Psychol., 2007, 32, 631-642.
244 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [9] [10] [11] [12]
[13] [14] [15] [16] [17] [18]
[19]
[20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44]
García-García et al.
Kolar, D.; Keller, A.; Golfinopoulos, M.; Cumyn, L.; Syer, C.; Hechtman, L. Neuropsychiatr. Dis. Treat., 2008, 4, 107-121. Patrick, K.S.; González, M.A.; Straughn, A.B.; Markowitz, J.S. Exp. Opin. Drug Deliv., 2005, 2, 121143. Popper, C.; West, S. In DSM-IV: Tratado de psiquiatría, 3rd ed.; Hales, R.E.; Yudofsky, S.C.; Talbott, J.A. Eds.; Masson: Barcelona, 2000, pp. 825-856. Valdizan, J.R. Consenso multidisciplinar en TDAH: infancia, adolescencia y adultos, 2005. Available at: http://www.aapi.org.ar/Aapi/ArchivosTDAH/TDAH%20%20documento%20presentado%20en%20M as%20Casadevall.pdf. Rappley, M. New Engl. J. Med., 2005, 352, 165-173. Conners, C.K. J. Attent. Disord., 2002, 6, S17-S30. Faraone, S.V.; Spencer,T.; Aleardi, M.; Pagano, C.; Biederman, J.J. Clin. Psychopharmacol., 2004, 24, 24-28. Barkley, R.A. In Attention-Deficit Hyperactivity Disorder. A Handbook for Diagnosis and Treatment, 3rd ed.; Barkley, R.A. Ed.; Guilford Press: New York, 2006, pp. 3-75. MTA Cooperative Group. Arch. Gen. Psychiatry, 1999, 56, 1073-1086. Taylor, E.; Döpfner, M.; Sergeant, J.; Asherson, P.; Banaschewski, T.; Buitelaar, J.; Coghill, D.; Danckaerts, M.; Rothenberger, A; Sonuga-Barke, E.; Steinhausen, H.C.; Zuddas, A. Eur. Child Adolesc. Psychiatry, 2004, 13, S7-S30. Banaschewski, T.; Coghill, D.; Santosh, P.; Zuddas, A.; Asherson, P.; Buitelaar, J.; Danckaerts, M.; Döpfner, M.; Faraone, S.V.; Rothenberger, A.; Sergeant, J.; Steinhausen, H.C.; Sonuga-Barke, E.J.S.; Taylor, E. Eur. Child Adolesc. Psychiatry, 2006, 15, 476-495. Connor, D.F.; Steingard, R.J. CNS Drug, 2004, 18, 1001-1030. Madaan, V.; Kinnan, S.; Daughtonm J.; Kratochvilm C.J. Exp. Rev. Neurother., 2006, 6, 1375-1385. Biederman, J. J. Clin. Psychiatry, 2006, 67, S3-S6. Keith, S. Prog. Neuro-Psychopharmacol. Biol. Psychiatry, 2006, 30, 996-1008. Prince, J.B. Child Adolesc. Psychiatry Clin. North Am., 2006, 15, 13-50. Quintero-Gutiérrez, F.J.; Correas, J.; Quintero-Lumbreras, F.J. Déficit de atención e hiperactividad a lo largo de la vida, Editorial Ergon: Madrid, 2006. Eisenberg, L.J. Child Adolesc. Psychopharmacol., 2007, 17, 279-283. World Health Organization. The international classification of diseases, 10th ed., World Health Organization: Geneva, 1993. American Psychiatric Association. Diagnostic and statistical manual of mental disorders, 4th ed., American Psychiatric Association: Washington DC, 1994. Greenhill, L.L. Psychiatr. Ann., 1989, 19, 604-613. Swanson, J.M.; Sergant, J.A.; Taylor, E.; Sonuga-Barke, E.J.S.; Jensen, P.S.; Cantwell, D.P. Lancet, 1998, 351, 429-433. Lee, S.I.; Schachar, R.J.; Chen, S.X.; Ornstein, T.J.; Charach, A.; Barr, C.; Ickowicz, A. J. Child. Psychol. Psychiatr., 2008, 49, 70-78. Sandberg, S. Br. J. Psychiatry, 1996, 169, 10-17. Garitty, C.M. Can. Pharm. J., 1996, 129, 36-42. Cosgrove, P.V.F. Prim. Care Psychiatry, 1997, 3, 101-113. Jaideep, T.; Reddy, Y.C.J.; Srinath, S. Bipolar Dis., 2006, 8, 182-187. Cumyn, L.; Kolar, D.; Keller, A.; Hechtman, L. Exp. Rev. Neurother., 2007, 7, 1375-1390. Biederman, J. J. Clin. Psychiatry, 1998, 59, 4-16. Safer, D.J. CNS Drugs, 1997, 7, 264-272. Ballard, S.; Bolan, M.; Burton, M.; Snyder, S.; Pasterczyk-Seabolt, C.; Martin, D. Adolescence, 1997, 32, 855-862. Kieling, C.; Goncalves, R.R.F.; Tannock, R.; Castellanos, F.X. Child Adolesc. Psychiatr. Clin. North Am., 2008, 17, 285-307. Mick, E.; Faraone, S.V. Child Adolesc. Psychiatr. Clin. North Am., 2008, 17, 261-284. Arnsten, A.F. J. Clin. Psychiatry, 2006, 67(Suppl 8), S7-S12. Fassbender, C.; Schweitzer, J.B. Clin. Psychol. Rev., 2006, 26, 445-465. Biederman, J.; Spencer, T.J. Child Adolesc. Psychiatr. Clin. North Am., 2008, 17, 439-458.
Methylphenidate Extended-Release Capsules [45] [46] [47] [48] [49]
[50] [51] [52] [53] [54] [55]
[56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72]
[73] [74] [75] [76] [77] [78]
[79] [80] [81]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 245
Jensen, P.S.; Arnold, L.E. Pediatrics, 2004, 113, 754-761. Smoot, L.C.; Boothby, L.A.; Gillett, R.C. Int. J. Clin. Pract., 2007, 61, 1730-1738. Scottish Intercollegiate Guidelines Network. Attention Deficit and Hyperactivity Disorders in Children and Young People. A National Clinical Guideline, SIGN: Edinburgh, 2001. Wilens, T.; Spencer, T.; Biederman, J.A. J. Attent. Dis., 2002, 5, 189-202. Kutcher, S.; Aman, M.; Brooks, S.J.; Buitelaar, J.; Van Dalen, E.; Fegert, J.; Findling, R.L.; Fisman, S.; Grrennhill, L.L.; Huss, M.; Kusumakar, V.; Pine, D.; Taylor, E.; Tyano, S. Eur. Neuropsychopharmacol., 2004, 14, 11-28. Sonuga-Barke, E.; Van Lier, P.; Swanson, J.; Coghill, D.; Wigal, S.; Vandenberghe, M.; Hatch, S. Eur. Child Adolesc. Psychiatry, 2008, 17, 245-254. Harpin,V.A. Arch. Dis. Child Educ. Pract., 2008, 93, 58-65. Kostrzewa, R.M.; Kostrzewa, J.P.; Kostrzewa, R.A.; Nowak, P.; Brus, R. J. Neural Transm., 2008, 115, 287-298. Challman, T.D.; Lipsky, J.J. Mayo Clin. Proc., 2000, 75, 711-721. Wolraich, M.L.; Doffing, M. CNS Drug, 2004, 18, 243-250. Wigal, S.B.; Gupta, S.; Greenhill, L.; Posner, K.; Lerner, M.; Steinhoff, K.; Wigal, T.; Kapelinski, A.; Martinez, J.; Modi, N.B.; Stehli, A.; Swanson, J. J. Child Adolesc. Psychopharmacol., 2007, 17, 153164. Tuerck, D.; Appel-Dingemanse, S.; Maboudian, M.; Pommier, F.; Wang, Y.; Sedek,G. J. Clin. Pharmacol., 2007, 47, 64-69. Pelz, R.; Banaschewski, T.; Becker, K. Klin. Padiatr., 2008, 220, 93-100. Liu, F.; Muniz, R.; Minami, H.; Silva, R.R. Psychiatr. Q., 2005, 76, 259-269. Swanson, J.; Kinsbourne, M.; Roberts, W.; Zucker, K. Pediatrics, 1978, 61, 21-29. Holmes, V. Int. J. Psychiatr. Med., 1995, 25, 1-19. Patrick, K.S.; Markowitz, J.S. Hum. Psychopharmacol., 1997, 12, 527-546. Schachter, H.M.; King, J.; Langford, S.; Moher, D. Can. Med. Assoc. J., 2001, 27, 1475-1488. Schulte-Markwort, M.; Warnke, A. Methylphenidat, Thieme: Stuttgart, 2004. Wigal, S.; Swanson, J.M.; Feifel, D.; Sangal, R.B.; Elia, J.; Casat, C.D.; Zeldis, J.B.; Conners, C.K. J. Am. Acad. Child Adolesc. Psychiatry, 2004, 43, 1406-1414. McGough, J.J.; Pataki, C.S.; Suddath, R. Exp. Rev. Neurother., 2005, 5, 437-441. Robinson, D.M.; Keating, G.M. Drugs, 2006, 66, 661-668. Biederman, J.; Quinn, D,; Weiss, M.; Markabi, S.; Weidenman, M.; Edson, K.; Karlsson, G.; Pohlmann, H.; Wigal, S. Paediatr. Drugs, 2003, 5, 833-841. Artigas-Pallares, J. Rev. Neurol., 2004, 38, S117-S123. Anderson, V.R.; Keating G.M. Pediatr. Drugs, 2006, 8, 319-333. Keating, G.M.; McClellan, K.; Jarvis, B. CNS Drugs, 2001, 15, 495-500. Mulas, F.; Tellez de Meneses, M.; Hernandez-Muela, S.; Mattos, L.; Pitarch, I. Rev. Neurol., 2005, 40, S49-S55. Wigal, S.B.; Gupta, S.; Greenhill, L.; Posner, K.; Lerner, M.; Steinhoff, K.; Wigal, T.; Kapelinski, A.; Martinez, J.; Modi, N.B.; Stehli, A.; Swanson, J.; Wilens, T.; McBurnett, K.; Stein, M.; Lerner, M.; Spencer, T.; Wolraich, M. J. Am. Acad. Child Adolesc. Psychiatry, 2005, 44, 1015-1023. Abramowitz, M. Med. Lett., 2005, 48, 49-51. López, F.A. J. Dev. Behav. Pediatr., 2006, 27, 410-416. Anderson, V.R.; Scott, L.J. Drugs, 2006, 66, 1117-1126. Wilens, T.E.; Boellner, S.W.; López, F.A.; Turnbow, J.M.; Wigal, S.B; Childress, A.C.; Abikoff, H.B.; Manos, M.J. J. Am. Acad. Child Adolesc. Psychiatry, 2008, 47, 700-708. Anonymous. Med. Lett., 2001, 43, 83-84. Döpfner, M.; Gerber, W.D.; Banaschewski, T.; Breuer, D.; Freisleder, F.J.; Gerber-von Müller, G.; Günter, M.; Hässler, F.; Ose, C.; Rothenberger, A.; Schmeck, K.; Sinzig, J.; Stadler, C.; Uebel, H.; Lehmkuhl, G. Eur. Child Adolesc. Psychiatry, 2004, 13, 93-101. Fischer, R.; Schütz, H.; Grossman, M.; Leis, H.J.; Armmer R. Int. J. Clin. Pharmacol. Ther., 2006, 44, 135-141. Lerner, M.; Wigal, T. Pediatr. Ann., 2008, 37, 37-45. SPC Medikinet. MEDICE Arzneimittel Pütter GmbH & Co. KG, Fach information: Berlin, 2005.
246 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [82] [83] [84]
[85] [86]
García-García et al.
Ghuman, J.K.; Cataldo, M.D.; Beck, M.H.; Slifer, K.J. J. Child. Adolesc. Psychopharmacol., 2005, 14, 601-611. Beck, M.H.; Cataldo, M.; Slifer, K.J.; Pulbrook, V.; Guhman, J.K. Clin. Pediatr., 2005, 44, 515-526. Döpfner, M.; Banaschewski, T.; Schmidt, J.; Uebel, H.; Schmeck, K.; Gerber, W.D.; Günter, M.; Knölker, U.; Gehrke, M.; Häbler, F.; Möhler, E.; Brünger, M.; Ose, C.; Fischer, R.; Poustka, F.; Lehmkuhl, G.; Rothenberger, A. Nervenheilkunde, 2003, 22, 85-92. Sinzig, J.K.; Döpfner, M.; Pück, J.; Banaschewski, T.; Stephani, U.; Lehmkuhl, G.; Rothenberger, A.; Arbeitsgruppe methylphenidat. Z. Kinder-Jugendpsychiatr., 2004, 32, 225-233. Sinzig, J.K.; Döpfner, M.; Lehmkuhl, G.; German Methylphenidate Study Group. J. Child Adolesc. Psychopharmacol., 2007, 17, 628-632.
Frontiers in Drug Design & Discovery, 2009, 4, 247-272
247
Osmotic-Controlled Release Oral Delivery System (OROS Technology) in Chronic Pain Management Flaminia Coluzzi* and Consalvo Mattia I.C.O.T. – Polo Pontino, Dept. Anaesthesiology, Intensive Care Medicine and Pain Therapy, University of Rome “La Sapienza”, Rome, Italy Abstract: Chronic pain requires long-term therapy and controlled-release opioids are widely used for its management. One of the major causes of suboptimal therapy outcomes is poor adherence to prescribed therapy. Better adherence to treatment can be obtained by reducing dosing frequency. Oncedaily administrations were shown to be effective and well accepted by patients. The osmotic-controlled release oral delivery system (OROS®) is an advanced drug delivery technology that uses osmotic pressure as the driving force to deliver pharmacotherapy in many therapeutic areas. Over the past decades, different OROS osmotic dosage forms have been studied to optimize extendedrelease oral administration by controlling the rate of drug release, providing different delivery profiles. OROS hydromorphone has been recently introduced in clinical practice for treating patients with chronic cancer and noncancer pain. OROS hydromorphone ensures a constant, monophasic delivery of hydromorphone over a 24-hour period. The pharmacokinetic profile of OROS hydromorphone is minimally affected by food and alcohol. Doseconversion studies showed that patients with chronic pain could be switched easily from previous opioid therapy to OROS hydromorphone, without loss of pain control. These studies support the clinical utility of the 5:1 ratio used for the conversion of oral morphine to oral OROS hydromorphone. Oncedaily OROS hydromorphone was shown to be effective in patients with chronic low back pain, and to provide similar pain relief to that of ER oxycodone in patients with moderate to severe osteoarthritis pain. OROS products can result in an improved safety profile, more stable drug concentrations, uniform drug effects, and reduced dosing frequency.
INTRODUCTION Despite major improvements in pain control over the last 15 years, chronic pain continues to be a significant global public health concern. Chronic pain requires long-term therapy. Opioids are widely used for both chronic cancer and non-cancer pain management. When possible, drugs should be given by mouth and doses repeated at regular intervals so that pain is prevented from returning. Controlled-release formulations
*Corresponding Author: Tel: +39 339 244 10 90; Fax: +39 06 8086839; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
248 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
represent the gold standard for chronic pain management, in order to keep a stable plasmatic drug concentration and offer a stable level of analgesia [1-2]. One of the major causes of sub-optimal therapy outcomes is poor adherence to prescribed treatment regimen. Adherence to a medication regimen is defined as the extent to which patients take medications as prescribed by their health care providers [3]. The ability of a patient to follow a treatment regimen is dependent on a variety of factors: social and economic factors, the health care system, health care professionals, the nature of the disease, type of treatment, and personal circumstances. In chronic medical conditions the level of adherence to treatment has been the subject of intense investigations. In patients with chronic malignant pain, one factor that seems to contribute to ineffective cancer pain management is poor adherence to the analgesic regimen [4]. The adherence rates for opioid analgesics prescribed on an aroundthe-clock (ATC) basis ranged from 84.5% to 90.8%, whereas for opioids prescribed on an as-needed (PRN) basis resulted from 22.2% to 26.6%. Similarly, a poor adherence to analgesic treatment has been observed in patients with osteoarthritis. Patients were generally adherent to their other medications; however they were reluctant to take painkillers, and when they did, they generally took them at a lower dose or frequency than prescribed [5]. Pain medications seem to be perceived differently than other medications. Patients appeared invested in taking analgesics only when they felt it was absolutely necessary. This could be one of the reasons for the observed difficulty to achieve adequate pain control in chronic conditions. It has been shown that treatment adherence is positively influenced by a good patient-doctor relationship and support within a community. In contrast, complicated treatment regimens and high frequency of dose often confound the patient and reduce the compliance to prescribed therapies. Better adherence to prescribed therapy can be obtained by reducing dosing frequency. Reducing the number of daily doses through controlled-release formulations has been shown to provide the patient with better symptom control in a number of diseases. A meta-analysis of 76 studies covering different indications, revealed that adherence to medication is inversely proportional to frequency of dose. Dose-compliance was 51% when medication was prescribed four times daily, 65% for three times daily, 69% for twice daily and increased up to 79% for once daily. The improvements in dosecompliance between four times daily and once-daily, and between three times daily and once-daily were statistically significant (p < 0.001) [6]. The consequences of poor compliance to long-term therapies are poor health outcome and increased health care costs. Improving adherence to medications also enhances the safety and tolerability of a treatment [7]. Many oral drug delivery systems have been developed with the intention of improving patient compliance to treatment in chronic pain management. Once-daily administrations have been shown to be effective and well accepted by patients. A number of opioids are available as once-daily oral formulations, such as Tramadol, Morphine, and Hydromorphone. However, some technologies have been shown to be more effective than the others in keeping an adequate plasmatic drug concentration and offering a stable analgesia [8].
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 249
The osmotic-controlled release oral delivery system (OROS osmotic technology, Alza Pharmaceuticals, Palo Alto, California) is an advanced drug delivery technology, specifically developed for controlled-release formulations that could improve patient adherence to chronic therapies. Over the past decade, the OROS osmotic technology has been used in more than a dozen products marketed around the world, for treatment of chronic diseases, such as hypertension, diabetes, benign prostatic hyperplasia, overactive bladder, attention-deficit hyperactive disorder, schizophrenia, and pain [9]. OROS Hydromorphone is a novel controlled-release formulation, marketed in some European countries for severe chronic pain management. This method of drug delivery is intended to minimize peak-through plasma concentration fluctuations associated with traditional Immediate-Release (IR) Hydromorphone. OSMOTIC-CONTROLLED RELEASE ORAL DELIVERY SYSTEM: OROS TECHNOLOGIES OROS Technology and Design Besides drug discovery research to find new therapies, the pharmaceutical industry over the past decades focused on the key role of drug delivery systems in providing optimized products for existing drugs. A controlled drug delivery system requires simultaneous consideration of several factors, such as drug properties, biocompatibility, ability to targeting, nature of delivery vehicle, mechanism of drug release, duration of delivery, and route of administration [10]. The oral route remains the most acceptable mode of administration especially for long-term therapies to manage chronic pathologies. The difficulty is to design products that are resistant to breakdown by the fluids in the gastrointestinal (GI) tract. Extendedrelease formulations require the drug to have reasonable absorption through the length of the GI tract. The OROS osmotic technology is one of the most complex, and relies on the principle of osmosis as the driving force for controlled drug release. This technology can aid in the transformation of the standard pharmaceutical tablet into an advanced drug delivery system containing one or more drugs. Different OROS osmotic dosage forms have been studied to optimize extended-release oral administration by controlling the rate of drug release, providing different delivery profiles (Figs. 1 and 2). The first elementary osmotic pump (EOP) was introduced in the 1970s. The system consisted of a tablet with a single osmotic core containing a water-soluble drug enclosed in a semi-permeable membrane laser-drilled with a delivery orifice. The membrane is permeable to water, but impermeable to ions and to the drug itself. Following ingestion, as water is absorbed from the gastrointestinal tract, the drug is dissolved in a solution and is delivered through the delivery orifice at the same rate as the water entering the table. Delivery is thus controlled primarily by the dosage form, rather than environmental factors such as pH or motility. This system delivers drugs at zero-order rates and is only suitable for water-soluble drugs [11]. The osmotic technology has evolved over the last 30 years and it is currently available in four different rate controlled formulations, namely Elementary Osmotic Pump,
250 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
a) Elementary Osmotic Pump Osmotic Delivery Orifice
Semipermeable Membrane
Water
Osmotic Core containing Drug
b) OROS® Push-Pull System Osmotic Delivery Orifice
Semipermeable Membrane
g Drug e Core
Water
Osmotic Push Compartment
c) OROS® Multi-Layer Push-Pull System Osmotic Delivery Orifice Semipermeable Membrane
g1 Drug e Core g2 Drug e Core
W Water
Osmotic Push Compartment
d) L-OROS® System Osmotic Delivery Orifice Semipermeable Membrane Drug g e Core Water W
Inner layer Osmotic Push Compartment
Soft Gelatine
Fig. (1). OROS technologies for once-daily pharmacotherapy.
OROS Push-Pull System, OROS Multi-Layer Push-Pull System, and Liquid OROS System. A new two-layer osmotic push-pull table was designed to deliver also poorly soluble drug moieties. This more sophisticated system, named OROS Push-Pull System, consists of a bilayer core surrounded by a semipermeable tablet shell membrane that is per-
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 251
vious to water but not the drug. The drug layer contains the active drug and excipients, and represents the 60-80% of the tablet weight. The push layer contains a hydrophilic expanding compartment (osmotic agents and water-soluble polymers) and constitutes 20-40% of the table. Upon ingestion of the OROS Push-Pull System, when the tablet is exposed to water in the gastrointestinal tract, both the layers become osmotically active. A drug solution or suspension is formed in the drug layer and is delivered through the orifice, as with the elementary osmotic pump. However, an additional driving force comes from the water-soluble polymeric expanding compartment of the push layer, which maintains the drug delivery at a target rate. This system can also be used for water-insoluble drugs and is currently the most frequently used OROS technology. The depleted empty shell does not dissolve but it is excreted intact in the faeces [9]. Drugs commercialized using the OROS Push-Pull System composed of a bilayer tablet core include Glipizide, Nifedipine, and Hydromorphone, for hyperglicemia, angina/hypertension, and pain control, respectively.
Zero-order
Double profile (2 drugs)
Pulse + zero order
Ascending profile
Multiple pulses
Delayed zero-order
Modified from Bass DM [14]
Fig. (2). Examples of OROS osmotic drug delivery technology delivery profiles.
The OROS Multi-Layer Push-Pull System is an advanced design technology, which increases the flexibility of the system, by using a longitudinally compressed tablet multilayer formulation. Multiple drug layers can be formulated either with different drugs to administer combination therapy or with different drug concentrations to modulate the release rate profile. At the beginning, most of the drug is released from the first compartment, which is adjacent to the delivery orifice, followed by the drug released from the second compartment at a different rate. A moderate level of mixing between the two layers can be programmed to achieve a smooth ascending release profile. The amount of mixing between the multiple drug layers can be controlled, by varying the relative viscosity and hydrophilicity of components. Moreover, a layer without a drug can be included between the two drug layers to deliver a drug intermittently [9]. Methylphenidate,
252 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
a drug used to treat attention-deficit disorder, is commercialized using the OROS Multi-Layer Push-Pull System, composed of a trilayer tablet core. A liquid OROS (L-OROS ) system has been also designed for the delivery of emulsions or microparticulate suspensions of highly insoluble drugs or polypeptides. The L-OROS system provides controlled delivery of liquid drug formulations in a capsule form. This liquid OROS softcap design involves coating a gelatin capsule (which contains the liquid formulation) with multiple layers, including an inner barrier membrane, then an expandable osmotic push layer and then a rate-controlling semi-permeable membrane. Like other OROS systems, a hole is drilled through the rate-controlling membrane to permit the release of drug and then the system is dried at 30°C overnight. When the system is administrated, water permeates through the rate-controlling membrane and activates the osmotic engine. As the osmotic engine expands, hydrostatic pressure inside the system builds up, thereby forcing the liquid formulation to be pumped out of the system. When the liquid is squeezed out, the gelatin capsule shell becomes flattened. The inner barrier membrane, located between the gelatin capule and the osmotic push layer, functions to minimize the hydration of the gelatin shell during release. If the hydrated gelatin shell would lose its integrity, it would mix with the liquid fill, leading to an erratic release profile. The osmotic push layer is the driving force of this system and comprises a high molecular weight hydrophilic polymer and an osmotic agent, mixed in a pseudo-suspension of sodium chloride and ethanol. The semi-permeable membrane is composed of a water-insoluble but permeable polymer, and a water flux enhancer. Products utilizing this system are currently under development. Acetaminophen soft gelatin capsule have being studied [12]. Recent advances in pharmaceutical research led to development of osmotic timecontrolled drug delivery systems to meet the emerging chronotherapeutic requirements. The idea of targeting drug release to the specific time of day when there is maximal clinical manifestation of the disease could have obvious advantages in the treatment of asthma, heart disease, duodenal ulcer or arthritis. The difficulty lies in designing products that are resistant to breakdown by the fluids in the GI tract and that release drug only at the required time. Verapamil has been marketed with OROS technology in a controlled-onset extended release (COER) formulation. The drug is released overnight, four to five hours after the table is ingested, to provide optimal blood pressure control in the early hours of the day, with the aim of addressing the peak morning risk of cardiac events. An additional layer between the active drug core and the semipermeable membrane guarantees this OROS Delayed Push-Pull System. As water from the GI tract enters the tablet, the osmotically active layer expands and pushes on the drug core. The additional layer enables release of drug to be delayed [9]. Novel elementary osmotic tablets have also been developed for time-controlled release system using the cores of drug-resin complexes and are named drug-resin complex osmotic pump tablets (DRCOPT). This system led to a zero-order drug release profile after an initial lag of time. Propranolol has been used as the model drug in the treatment of hypertension [13]. An alternative osmotic drug delivery platform, named EnSoTrol (Shire Laboratories Inc., Rockville, MD, US), has been developed for poorly soluble compounds. The table consist of a single layer core, which is surrounded by a semi-permeable membrane, with a laser drilled delivery orifice. The table core includes the drug, osmotic agent, and a
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 253
solubility enhancer that helps dissolved drug to be delivered in the GI lumen. The EnSoTrol technology has been tested with water-insoluble nifedipine and glizipide. Safety of OROS Formulations The OROS technologies have been included for the last two decades in prescription medications for urology, central nervous system, and cardiovascular indications, as well as over-the-counter nasal/sinus congestion medications. These nondeformable tablets or capsules that incorporate OROS technologies are excreted unmodified in stools, suggesting an increased risk of gastrointestinal events compared to other oral formulations. A retrospective study examined gastrointestinal safety data for OROS formulations, following two decades of use [14]. An estimated cumulative distribution of these products in United States through June 2000 is 12.85 billion tablets. The overall incidence of clinically significant GI irritation and obstruction was estimated 13 cases per billion tablets distributed, ore one case per 76 million tablets. Most of cases (78%) were recorded in patients taking OROS formulation of nifedipine. Six of these patients died for duodenal ulcer haemorrhage (two cases), peptic ulcer haemorrhage (one case), gastric ulcer (one case), and gastrointestinal events of unclear origin (two cases). 87 cases that involved one or more symptoms of gastric or lower gastrointestinal obstruction (bezoar, ileus, faecal impaction, stenosis, impaired gastric emptying, volvulus of the bowel, and intestinal obstruction) were recorded. Despite OROS formulations were suspected to have a greater risk of injury and obstruction than other oral dosage forms, these data showed that gastrointestinal events are infrequent and typically associated with known effects of the drug substance rather than the dosage form. OROS HYDROMORPHONE: PUSH-PULL DELIVERY SYSTEM IN PAIN MANAGEMENT Pharmacology of Hydromorphone Hydromorphone HCl is a semi-synthetic opioid that has been used to treat acute and chronic pain for almost 80 years. It is considered an effective alternative to morphine in the treatment of moderate-to-severe pain for its efficacy and safety profile [1, 15-16]. Hydromorphone is structurally very similar to morphine, from which it differs by the presence of a 6-keto group and the hydrogenation of the double bond at the 7-8 position of the molecule. Hydromorphone exerts its analgesic effects primarily through muopioid receptors in the central nervous system and to a lesser degree on delta receptors, without effects on kappa, sigma, or epsilon receptors. It has a potency of 5 to 7 times that of morphine [17]. Hydromorphone is currently available in several dosage forms: immediate-release tablets, controlled-release capsules, oral liquid, rectal suppositories, powder, cough syrup, and solution for intravenous, subcutaneous, and intramuscular injection. The pharmacokinetic profile of orally administrated immediate-release hydromorphone has been well characterized. Following oral administration, hydromorphone is rapidly absorbed in the upper small intestine, achieving peak plasma concentration within approximately 1 hour. The onset of action is about 30 min with duration of about 4 hours, due to the half-life of approximately 2 to 3 hours [18].
254 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
Hydromorphone is mainly metabolized in the liver and a variety of water-soluble metabolites are excreted in the urine. 62% of the oral dose is eliminated by the hepatic first-pass metabolism. Although chemically similar to morphine, hydromorphone does not have an analgesically active 6-glucuronide metabolite. Morphine-6-glucuronide has analgesic activity and accumulates in presence of renal failure, leading to increased risk of opioid-related respiratory depression. Conversely, hydromorphone is extensively metabolized to hydromorphone-3-glucuronide, which is about 2.5 times as potent as morphine-3-glucuronide as a neuroexcitant. With chronic dosing of oral hydromorphone, blood levels of hydromorphone-3-glucuronide may increase of about 30 times compared to the blood levels of the parent drug. This ratio may rise to 100-fold in patients with renal impairment [17]. A link between metabolite accumulation and excitatory side effects has not been well established in humans. Clinical studies on administration of high doses of hydromorphone in patients with renal failure showed an increased incidence of nausea [19] and delirium [20] that are likely to be related to hydromorphone metabolites. However, recent investigations on the use of hydromorphone in palliative care patients with renal impairment support the use of hydromorphone even in end-stage renal failure, where it can be used effectively and safely, and it is better tolerated than morphine [21]. The major compounds recovered in the urine include hydromorphone-3-glucuronide (H3G) conjugate (~40%), unchanged drug (~6%), dihydroisomorphine (< 0.1%), and dihydromorphine (< 0.1%) [22]. Hydromorphone can be given through various routes of administration as it has the potential to be absorbed from almost all types of mucosa. Parenteral administration of hydromorphone is the most frequent alternative to oral treatment. Hydromorphone solutions are available for intravenous, subcutaneous, and intramuscular injection. After intravenous dosing, the onset of action is approximately 5 minutes. The lipid solubility of hydromorphone is higher than that of morphine, leading to a faster onset of action, but it is slower than a highly lipid soluble opioid such as fentanyl. The oral to parenteral equianalgesic ratio is about 5:1. When prepared in highly concentrated solution (up to 100 mg/mL) hydromorphone can be administrated as a subcutaneous infusion, which has 78% of the bioavailability of intravenous dosing and is reported to be a safe and effective alternative. The absorption of hydromorphone by the intramuscular route is erratic and dose not offer advantages to the patient. All these three parenteral routes require regularly repeated doses, due to the short half-life of hydromorphone. Therefore, the parenteral route is considered time-consuming, painful, and difficult to maintain in the home setting. An alternative to the oral administration is the rectal route, which has the advantage of bypassing first-pass hepatic metabolism. However, following administration of hydromorphone suppositories, low bioavailability and large inter-individual variation have been observed. After transdermal delivery, hydromorphone remains undetected during the first few hours but when steady-state concentrations are achieved, they are maintained for 24 hours without any depot formation in the skin. The sublingual and buccal routes are unlikely to be preferred routes of administration for hydromorphone, as they require drugs with high lipophilicity and poorly ionized at the pH of saliva. Hydromorphone does not possess the required attributes. Similarly, the inhalation route is considered an inappropriate route of administration for hydromorphone, and the intranasal route provide a bioavailability of < 60% [23]. Hydromorphone can also be administrated via the epidural route, with duration of action after single shot of about 7 to 19 hours. The epidural to parenteral equianalgesic
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 255
ratio is about 1:2 [17]. Hydromorphone has the potential for being administrated by the intraspinal route. When delivered by this route, hydromorphone is detected in the blood one minute after administration, with a peak after one hour [23]. Extended-Release Hydromorphone Formulations According to its pharmacokinetic profile, the conventional immediate-release tablets of hydromorphone HCl must be administered every 4–6 h to provide continuous pain relief, due to its short elimination half-life. Fluctuating plasma levels associated with intermittent administration of immediaterelease analgesic drugs can result in inadequate analgesia. Administration of opioids on an around-the-clock basis seems to maintain the drug levels within the therapeutic range. Moreover, frequent dosing may be inconvenient for many patients. Compliance with taking a medication is inversely related to the dosing frequency [6]. Currently, three opioid analgesics are available in extended-release formulations for oral administration: morphine, oxycodone, and hydromorphone, most of which require dosing at least twice daily. Hydromorphone is available as extended-release formulation for twice-daily administration. It has been marketed under the trade name Hydromorphon Contin in Canada and as Palladone SR (Mundipharma International Limited, Cambridge, UK) in Britain and Germany. The pharmacokinetics, clinical efficacy and safety of this 12-hour modified-release hydromorphone have been well characterized in patients with chronic pain [24]. A novel once-daily extended-release hydromorphone formulation, named Palladone XLTM (Purdue Pharma, Stamford, CT), has been approved in Canada in 2005. It has been found to be effective in patients with chronic pain from malignant and nonmalignant causes. The analgesic effect, side effects, and safety profile were comparable to IR hydromorphone, but significantly reduced plasma fluctuations were observed for the once-daily product. Following a single oral administration, a rapid peak blood level can be detected within 2 hours and a second slow peak at about 18-24 hours. Steady state was achieved in three days. This modified-release capsule could be administrated by sprinkling on food or through a feeding tube if necessary in patients unable to swallow [17]. Despite the favourable profile resulting from several pre-marketing studies, in 2005, the United States Food and Drug Administration asked Purdue Pharma to withdraw this formulation over concerns related to dose dumping when ingested with alcohol [25, 26]. A New Technology for OROS Hydromorphone A new sustained-release formulation, named OROS hydromorphone, has been recently developed for achieving stable plasma concentrations using a once-daily dosing regimen. This novel controlled-release formulation utilizes the patented OROS PushPull System. Hydromorphone is the first opioid analgesic to be included in this unique OROS Push-Pull delivery technology that uses a bilayer core to provide constant drug release. One pull or drug layer contains hydromorphone HCl and hydrophilic, osmotically active polymers, along with standard tablet excipients. The second push layer contains a high molecular weight, osmotically active expansion polymer, an adjunct osmotic agent (sodium chloride), a colorant (ferric oxide or iron oxide) for layer discrimination
256 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
during laser drilling, and other standard excipients. The layers are bonded together by tablet compression to form a single round, bilayer core. Following ingestion, in the gastrointestinal tract, water diffuses across the semipermeable membrane surrounding the bilayer tablet core and a gel-like suspension is formed in the drug layer. Water also expands the osmotic push compartment, which in turn forces the drug suspension (hydrated hydromorphone) through the laser-drilled delivery orifice at a near constant rate into the lumen of the gastrointestinal tract for absorption. The empty shell is excreted intact in the stools. Four dosages are currently available in most of European countries (8 mg, 16 mg, 32 mg, and 64 mg), whereas this formulation is not marketed in United States yet. OROS HYDROMORPHONE: PHARMACODYNAMIC AND PHARMACOKINETIC PROFILE Pharmacodynamics of orally administrated OROS hydromorphone has been studied in twelve healthy volunteers [27]. Two study sessions have been conducted by administering a single dose of IR hydromorphone (8 mg) and OROS hydromorphone (single dose of 8, 16, and 32 mg). Analgesic efficacy has been evaluated by using an electrical experimental pain paradigm and has been assessed for up to 30 hours after administration. Peak plasma concentration was significantly higher after administration of IR hydromorphone than after any dose of OROS hydromorphone (4.74 ± 1.76 ng/ml vs 0.77 ± 0.33 ng/ml). The OROS hydromorphone plasma concentration peaked significantly later compared to the IR hydromorphone (12.0 h [12.0-18.0] vs 0.8 h [0.8-1.0]), without differences among the different doses of hydromorphone. Similarly, different doses of OROS hydromorphone did not result in significantly different intersubject variability in plasma concentration. However, the OROS hydromorphone plasma concentration was maintained significantly longer at greater than 50% of peak concentration (24.2 ± 10.3 h vs 1.1 ± 0.7 h) compared to conventional tablets. The system ensured the release of the drug over a period of 24 hours. Similarly, the analgesic effects of OROS hydromorphone, measured as analgesic tolerance, peaked significantly later than IR hydromorphone (9.0 h [9.0-12.0] vs 1.5 h [1.0-2.0]), but were maintained significantly longer at greater than 50% of peak analgesic effect (13.3 ± 6.3 h vs 3.6 ± 1.7 h). No statistically significant differences were observed in time to peak pain tolerance and duration of analgesic effects among different doses of OROS hydromorphone. Following administration of different doses of OROS hydromorphone, a significant linear relation existed between hydromorphone plasma concentration and pain tolerance, whereas there was no significant relation between hydromorphone plasma concentration and the pain threshold. The OROS hydromorphone formulation showed pronounced sustained-release characteristics with plasmatic concentrations sustained for approximately 24 h and analgesic efficacy to experimental pain beyond 24 h of its administration. Pharmacokinetics of intravenous, oral IR hydromorphone, and OROS hydromorphone has been compared in twelve subjects in a randomized, crossover-design study [28]. Results suggested continued release of hydromorphone from the OROS delivery
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 257
system over 24 hours, with a maximum absorption rate detected at approximately 10 h and decreasing progressively thereafter. The mean value of the maximum rate of release for the 8 mg OROS hydromorphone formulations was 1.2 mg/h. The bioavailability (AUC Area Under the Curve) per milligram oral dose was the same for 8, 16, and 32 mg OROS hydromorphone doses, showing a proportional drug release and absorption profile for the three different doses. The mean bioavailability for the 8 mg OROS hydromorphone was significantly larger than that for the IR formulation (0.24 vs 0.19; p<0.05). This may be partially explained by the release and absorption of hydromorphone from the OROS formulation in the colon and rectum, thus avoiding the amount of the first-pass effect. IR formulations indeed are adsorbed mainly in the upper small intestine from the duodenum and enter the mesenteric vessels to be transported to the liver, where they undergo extensive metabolism before entering the rest of systemic circulation. Approximately 62% of the oral dose is eliminated by the liver on first pass, leading to a high variability in oral bioavailability ranging from 1:2 to 1:8 [29]. Moreover, similarly to morphine, the presence of enterohepatic cycling of hydromorphone could explain the sustained absorption for 24 hours from the OROS formulation. Unchanged drug is excreted in the bile and reabsorbed across the intestinal wall into the systemic circulation. Although hydromorphone is chemically similar to morphine, hydromorphone is extensively metabolized to hydromorphone-3-glucuronide and dihydroisomorphine, but it does not have an analgesically active 6-glucuronide metabolite [17]. The degree of fluctuation of plasma concentrations, as calculated by mathematic simulator, is expected to be greater with the IR hydromorphone (130%) than the OROS hydromorphone (39%) [28]. This may be clinically translated to more side effects for the IR formulation compared to the sustained-release formulation, which is likely to provide a more stable analgesia and much less peak-to-through variations. Data from the simulation suggest the peak plasma concentration would be 95% of the steady state peak by the third dose of OROS hydromorphone. Thus, dose adjustments with OROS hydromorphone should be done at three days interval. These pharmacokinetic parameters after administration of OROS hydromorphone have been confirmed in a recent open-label, crossover study [30]. The aim of the study was to investigate the dose proportionality of hydromorphone that utilizes the OROS Push-Pull System. Thirty-two healthy volunteers have been randomized to receive a single dose of 8, 16, 32, and 64 mg of OROS hydromorphone. The opioid antagonist Naltrexone was administrated, concomitantly with each dose level of OROS hydromorphone, by three or four doses, to minimize the opioid related adverse events in these opioid-naïve subjects. Naltrexone acts competitively at mu, kappa, and delta opioid receptors in the central nervous system. Following a single oral dose of OROS hydromorphone, median time to peak plasma concentration (Tmax: 12.0-16.0 hours) and median terminal half-life (T1/2: 10.6-11.0 hours) were found to be independent of dose. Similarly the bioavailability (AUC0- Area Under the Curve for zero to infinity) per milligram oral dose was the same for 8, 16, 32, and 64 mg OROS hydromorphone doses (AUC0- 2.44-2.79 ng*h/mL/mg). The bioavailability and the peak plasma concentration (Cmax) increased linearly and in a manner proportional to the dose of OROS hydromorphone. Regression analyses with OROS hydromorphone showed that the relationship
258 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
was linear (p 0.05) and that the intercept did not differ significantly from zero (p 0.05). Plasma concentration was sustained at or near maximum levels up to approximately 30 hours after oral administration. In conclusion, OROS hydromorphone ensures a constant, monophasic delivery of hydromorphone over a 24-hour period. The rate of release is consistent across a range of doses (8 mg, 16 mg, 32 mg, and 64 mg), with similar rates of absorption into the systemic circulation for all doses. In contrast to IR hydromorphone that is rapidly released, resulting in a peak plasma concentration within one hour of administration, OROS hydromorphone is released more slowly, reaching a stable plasmatic concentration within 6 to 8 hours and keeping this level until approximately 24 hours post-dose (Fig. 3). When steady-state conditions are achieved, OROS hydromorphone plasmatic concentrations are maintained above the minimum IR hydromorphone plasma concentrations for each 24-hour dosing interval [31]. 5 IR Hydromorphone 8mg
4,5
Hydromorphone (ng/ml)
OROS® Hydromorphone 8 mg 4
OROS® Hydromorphone 16 mg
3,5
OROS® Hydromorphone 32 mg
3 2,5 2 1,5 1 0,5 0 0
4
8
12
16
20
24
Time (hours) Modified from Gupta S [31]
Fig. (3). Hydromorphone plasma concentrations after single dose administration of IR Hydromorphone and OROS® Hydromorphone.
Effects of Food and Alcohol on the Pharmacokinetic Profile of OROS Hydromorphone Food can interfere with the bioavailability of a drug via several mechanisms, including drug binding, altering gastric pH, modifying hepatic blood flow, altering drug solubility, and promoting disintegration of the drug formulation. Pharmacokinetics of IR hydromorphone are affected by food [32]. A 24% increase in the overall bioavailability has been observed, with a 25% decrease in Cmax, when administrated in fed conditions compared to the fasted state.
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 259
The effects of food on the integrity of the drug formulation are particularly important for sustained-release formulations that contain larger amounts of drug compared to conventional formulations. The effects of food and alcohol on the pharmacokinetic profile of OROS hydromorphone have been recently investigated in two studies [33,34] and are summarized in Table 1. Table 1.
Effects of Food and Alcohol on the Pharmacokinetic Profile of OROS Hydromorphone Cmax (ng/mL) mean (SD)
Tmax (h) mean (range)
T (h) mean (SD)
AUC 0- (ng*h/mL) mean (SD)
Sathyan G et al. [33] Fasted (N=25)
Fed (N=27)
Fasting + Naltrexone (N=26)
1.11
16
14.7
38.8
(0.20)
(6-36)
(6.1)
(9.6)
1.35
12
12.5
36.1
(0.36)
(6-20)
(5.2)
(10.0)
1.63
12
12.0
37.2
(0.57)
(6-24)
(3.9)
(9.9)
Sathyan G et al. [34] Fasted + 0% alcohol (N=20) Fasted + 40% alcohol (N=17) Fed + 0% alcohol (N=20) Fed + 40% alcohol (N=17)
1.37
16
12.4
40.6
(0.32)
(6-27)
(5.1)
(11.0)
1.89
12
11.1
42.2
(0.85)
(6-24)
(3.0)
(13.2)
1.42
16
11.6
37.1
(0.50)
(6-27)
(5.1)
(8.6)
1.56
16
10.8
34.8
(0.56)
(6-27)
(4.8)
(11.9)
In an open-label, randomized, crossover trial, thirty healthy subjects have been randomized to receive a single dose of OROS hydromorphone 16 mg under fasting conditions, immediately after a high-fat breakfast (fed conditions), and under fasting conditions with opioid antagonism with Naltrexone 50 mg [33]. Results showed a minimal effect of food on the pharmacokinetic profile of OROS hydromorphone (Fig. 4). A slightly lower mean Tmax value (12.0 vs 16.0 h) and a 20% increase in mean Cmax value (1.352 vs 1.107 ng/mL) have been observed in fed conditions. These effects could be attributed to delayed gastric emptying with food, resulting in the longer duration of drug
260 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
Hydromorphone (ng/ml)
release in the small intestines, where the drug absorption is more efficient. However, the overall hydromorphone exposure was not affected by the consumption of the high fat meal. The bioavailability of hydromorphone, expressed as the mean area under the concentration-time curve for zero to time t (AUC0-t), was 31.12 ng*h/mL in the fasting state and 30.20 ng*h/mL in the fed state. Thus, food seems to affect only the first few hours of dosing, by interacting with the gastrointestinal motility of the upper tract. 1,6
OROS® Hydromorphone 16 mg Fasted
1,4
OROS® Hydromorphone 16 mg Fed
1,2
OROS® Hydromorphone 16 mg Fasted + Naltrexone 50 mg
1 0,8 0,6 0,4 0,2 0 0
4
8
12
16
20
24
30
36
42
48
Time (hours) Modified from Sathyan G [33]
Fig. (4). Mean plasma concentrations of OROS® Hydromorphone 16 mg under fasted, fed conditions, and fasted condition plus naltrexone.
Co-administration of OROS hydromorphone with Naltrexone under fasting conditions seems to increase the rate, but not the extent, of drug absorption, with a 39% increase in Cmax (from 1.107 ng/mL to 1.635 ng/mL) and a 4.5 hours reduction in Tmax compared to fasting conditions. The overall hydromorphone exposure, in terms of AUC0-t and AUC0- was not affected by a 50 mg dose of Naltrexone. An open-label crossover study was conducted to investigate the pharmacokinetic profile of OROS hydromorphone in the presence of alcohol [34]. Alcohol is a commonly abused substance and many people consume alcohol regularly at meals. Alcohol can interfere with drug absorption by affecting the integrity of the formulation and resulting in an increase of the release rate (dose dumping) and potentially life-threatening absorption of opioid. Other controlled-release formulations, such as Palladone XL and oxymorphone hydrochloride, showed alcohol-interaction that affected the absorption rate of the drug. When subjects received 12 mg of Palladone XL with 240 mL of 40% alcohol, the mean peak plasmatic hydromorphone concentration increased by almost 6 times compared with subjects who received the drug with water, and for one subject the in-
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 261
crease was 16-fold. Alcohol disrupted the controlled-release mechanisms of these formulations, with potentially fatal consequences [25]. Similarly, when ER oxymorphone hydrochloride was administered with 40% alcohol in healthy fasted subjects, the mean bioavailability increased by 13% compared with subjects who received the drug with water. In this study, 48 healthy volunteers were randomised to receive four single doses of OROS hydromorphone 16 mg with 240 mL solutions (orange juice) of 0%, 4%, 20% or 40% alcohol. Opioid antagonism was provided by Naltrexone. Two groups of 24 patients were studied under fasted (no food for at least 12 hours before dosing) and fed (after a standardized breakfast) status, as food may interfere with the absorption of alcohol. No immediate rapid dose dumping of drug occurred when OROS hydromorphone was administrated in presence of alcohol. At the first measurement, 2 h after dosing, plasma hydromorphone concentrations were close to the lower limit of quantification. Thereafter, plasma hydromorphone concentrations were slightly higher after dosing with all alcohol treatments. No differences were observed in the fasted and fed subject groups. Mean Cmax values increased after subjects received the drug with alcohol, at all concentrations, without a clear alcohol dose-response relationship. The increase was slightly lower in the fed state (1.1-fold) compared to the fasted state (1.3-fold). The overall increase in Cmax was substantially lower than that seen with other extendedrelease formulations. By comparison, when patients received Palladone XL with alcohol, the mean peak hydromorphone concentration increased by almost 6-times compared with patients who received the drug with water [25]. The median Tmax values (12 - 16 h) were similar to those observed in previous studies, after oral administration of OROS hydromorphone in absence of alcohol [30,33], and without statistically significant difference in all treatments. These results confirmed that the controlled-release properties of the OROS formulation are maintained in the presence of alcohol, which minimally affected the pharmacokinetic profile of OROS hydromorphone. While other opioid formulations ensure controlled-release properties by the dissolution characteristics of the capsule, the OROS technology is minimally affected by the gastrointestinal environment and relatively insensitive to solvent characteristics. OROS HYDROMORPHONE: CLINICAL DATA Hydromorphone has been widely studied for acute and chronic pain. Hydromorphone resulted a potent analgesic, with a dose-related clinical effect, which can be considered a good alternative to morphine for the management of moderate-to-severe chronic pain [35]. The analgesic efficacy and safety of OROS hydromorphone was established in five published clinical trials involving a total of 1145 patients with chronic pain who were treated with OROS hydromorphone. Three trials were open-label, multicenter doseconversion studies, and two clinical trials were conducted in patients with chronic nonmalignant pain (low back pain and osteoarthritis). Dose-Conversion Studies with OROS Hydromorphone Three dose-conversion studies with OROS hydromorphone are currently available in literature (Table 2). These are open-label, multicentre trials conducted in patients who
262 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
were receiving oral or transdermal opioid therapies to manage their chronic malignant and non-malignant pain. Table 2.
Type of study (Author, year)
Dose-Conversion Studies with OROS Hydromorphone in Chronic Pain Management
Diagnosis (N. patients)
Open-label Chronic Ma(Palangio M lignant and 2002) [36] non-malignant pain (445) Open-label Chronic nonMulticenter malignant pain (Wallace M (366) 2007) [37] Open-label Multicenter (Wallace M 2008) [38]
Chronic Malignant pain (148)
Design
Phase 1 opioid dose stabilization Phase 2 titration
Efficacya
pain intensity pain relief
Phase 3 maintenance pain interference ratings Phase 1 opioid dose stabilization Phase 2 titration
pain intensity pain interference ratings
Phase 3 maintenance
Withdrawalb n. (% of N.)
Phase 1: 41 (9.2%) Phase 2-3: 131 (29.4%)
Phase 1: 30 (8.2%) Phase 2-3: 114 (31.1%)
pain on average intensity
Phase 1: 21 (14.1%)
pain interferPhase 3 maintenance ence ratings
Phase 2-3: 42 (28.4%)
Phase 1 opioid dose stabilization Phase 2 titration
OROS Hydromorphone mean daily dose at end of phase 3
63.4 ± 129.2 mg
70.1 ± 146.1 mg
56.7 ± 65.1 mg
a
Only statistically significant (p < 0.05) results are reported in “efficacy” Withdrawal is expressed as number of patients (n) and percentage on the total number (N) of patients who were enrolled in the phase 1 of the study.
b
In 2002, Palangio et al. published an open-label trial to evaluate the clinical outcomes associated with conversion from prior opioid therapy to OROS hydromorphone, in the treatment of chronic malignant and non-malignant pain [36]. The study was divided in three phases. Phase 1 was a prior opioid dose stabilization phase ( 3 days). Patient was considered stabilized when for a minimum of 3 consecutive days the total daily dose of baseline opioid medication (oral opioid or transdermal fentanyl) was stable and the number of opioid rescue medications per day was 3. The duration of phase 1 ranged from 3 to 40 days. Phase 2 was the conversion and titration phase to OROS hydromorphone (3-21 days). Each patient’s 24-hour opioid dose was converted to a single daily administration of OROS hydromorphone. The following dose-conversion ratios were used: 5:1 oral morphine sulfate equivalents (mg) to oral OROS hydromorphone (mg) or to 8 mg daily OROS hydromorphone for each 25 mcg/h of transdermal fentanyl patch. IR hydromorphone was provided as rescue dose for breakthrough pain. Dose adjustments were considered after two days of therapy to ensure that steady-state blood levels were achieved, using increments of 25% to 100% of the current total daily dose. Phase 3 was a maintenance phase at a stable dose of OROS hydromorphone (14 days).
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 263
A total of 445 patients were enrolled. Of these, only 404 received the study medication and only 273 (67.6%) completed the study and had a termination visit. The mean daily dose of OROS hydromorphone for patient at the end of the phase 3 was 63.4 ± 129.2 mg, whereas the mean daily dose of IR hydromorphone was 11.5 ± 36.4 mg (1.7 ± 1.3 doses per day). Conversion from prior opioid therapy and titration to a stable dose of OROS hydromorphone was readily performed. Dose stabilization was achieved by 73.8% of patients, most of which in less than two titration steps, in a mean of 12.1 ± 5.7 days. Analgesia was maintained during the transition to OROS hydromorphone. Pain intensity ratings at the Brief Pain Inventory (BPI) were significantly lower (p < 0.001) when comparing pretreatment to endpoint. Similarly mean pain relief ratings increased significantly from pretreatment to endpoint (56.2 ± 23.6% vs 61.1 ± 24.4%; p < 0.001), along with mean pain interference of function (general activity, mood, walking ability, normal work, relationships with others, sleep, and enjoyment of life) ratings that decreased significantly (p < 0.0001) for all pair-wise comparisons. Another open-label, multicentre study was conducted by Wallace et al. to assess the dose conversion to once-daily OROS hydromorphone from previous stable opioid agonist therapy [37]. This study was conducted only in patients with chronic non-malignant pain. Similarly to the previous trial [36], the study was divided in three phases: stabilization on the previous opioid therapy, switch to OROS hydromorphone and titration to adequate analgesia (3-16 days), and thereafter maintenance period (14 days). Conversion was obtained using a 5:1 ratio of morphine equivalent to OROS hydromorphone. For patients receiving transdermal fentanyl, each 25 mcg/hour of fentanyl were converted in 8 mg/day OROS hydromorphone. 366 patients were enrolled in the study; of these 336 received the study medication and 222 (66%) completed the study. The mean daily dose of prior opioid therapy at the end of stabilization period was 154.5 ± 172.6 mg morphine equivalents, and this dosage was converted to a starting daily dose of 30.1 ± 37.9 mg OROS hydromorphone. At the end of titration, the daily dose of OROS hydromorphone increased to a mean of 56.6 ± 63.3 mg, and at the end of the maintenance period to 70.1 ± 146.1 mg. These results are similar to those obtained in the previous dose-conversion study [36]. The number of rescue medication doses decreased during OROS hydromorphone therapy, but the mean daily dose of rescue drug does not. Dose stabilization of OROS hydromorphone was achieved by over 94% of treated patients. Mean pain intensity ratings at the Brief Pain Inventory (BPI) decreased significantly (p < 0.001) with OROS hydromorphone. Pain relief also improved during OROS hydromorphone treatment. All BPI pain interference scores significantly (p < 0.001) decreased at last assessment compared to baseline. Both patient and investigator assessment of the study medication improved along the treatment. At the end of the maintenance phase 26% of patients and 27% of investigators rated the effectiveness of medication as “very good” or “excellent”, compared to 8% and 9% at baseline, respectively. Results of this study confirmed the findings of the previous analysis and showed that patients with chronic malignant and non-malignant pain could be switched easily from previous opioid therapy to OROS hydromorphone, without loss of pain control. In 2008, Wallace et al. published a dose-conversion study on OROS hydromorphone in patients with chronic cancer pain [38]. In this multicenter, open-label trial,
264 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
patients were eligible if they were receiving at least 45 mg of morphine (or its equivalents) daily to control their pain. Similarly to the previous trials [36, 37], the study period was divided into three phases: stabilization, conversion (3-21 days) and maintenance (14 days). Dose conversion was obtained with a 5:1 ratio of morphine equivalents to hydromorphone. A total of 148 patients were enrolled in the prior opioid dose stabilization phase (phase 1). Of the 127 patients who underwent conversion to OROS hydromorphone (phase 2), 85 (67%) completed the maintenance phase (phase 3). Dose stabilization with OROS hydromorphone was achieved in 119 of 127 (94%) patients who received the study medication. The majority of patients did not require more than one or two titration steps. The mean daily dose of OROS hydromorphone was 44.6 ± 50.9 mg at the beginning of conversion (phase 2). This dose increased to 61.6 ± 73.4 mg at the end of the dose stabilization with OROS hydromorphone (phase 2) and decreased to 56.7 ± 65.1 mg at the end of maintenance phase (phase 3). The mean number of daily rescue medications and their mean daily dose increased during the titration phase (phase 2), but decreased during the maintenance phase (phase 3). The mean level of pain relief remained stable through the study period. The only statistically significant difference (p < 0.001) on BPI ratings was “pain on average” that decreased from 4.2 to 3.4 when comparing pretreatment (previous opioid stabilization phase) and endpoint (end of maintenance phase with OROS hydromorphone). Comparison of BPI interference ratings, at the end of phase 3 compared to pretreatment, showed a significant difference (p < 0.05) for all categories assessed, such as general activity, mood, walking ability, normal work, relation with others, sleep, and enjoyment of life. The study medication was rated as “very good” or “excellent” by 27% of patients and 25% of investigators. The result of this study confirmed that patients with chronic cancer pain could easily undergo conversion from previous opioid therapy to OROS hydromorphone, as previously shown in the study of Palangio et al. [36] In conclusion, all these three dose-conversion studies [36-38] support the clinical utility of the 5:1 ratio used for the conversion of oral morphine to oral OROS hydromorphone, because the majority of patients obtained adequate and stable analgesia with little titration from the starting dose, with no loss of efficacy or increase in side effects (Fig. 5). Similar results were obtained when previous opioid therapies were converted with a 8:1 ratio to a different formulation of oral once-daily extended-release hydromorphone in patients with persistent moderate to severe cancer-related or non-cancer-related pain [39]. These studies [36-38] showed that patients with chronic malignant or non-malignant pain could be switched easily from stable opioid agonist therapy to OROS hydromorphone, without the need for an intermediate IR opioid phase. OROS hydromorphone was shown to be effective in improving pain relief and in decreasing the degree to which pain interfere with life. Mean daily dose of OROS hydromorphone at the end of the maintenance period was similar in all these trials. Limitations of these studies include the open label design and the lack of a control group. However these results could represent the basis for further clinical investigations, by prospective, randomized and controlled trials.
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 265
The number of patients who achieved dose stabilization on the total number of patients treated with the studied drug is reported in brakets.
Fig. (5). Percentage of patients requiring titration steps to achieve stabilization of the OROS® Hydromorphone dose after converting from standard opioid therapy.
OROS Hydromorphone in Chronic Low Back Pain A pilot non-randomized, non-comparative, open-label, repeated-dose study was conducted in 207 patients with chronic moderate-to-severe low back pain, for efficacy and safety evaluation of once-daily OROS hydromorphone (Table 3) [40]. Phase 1 was a prior opioid dose stabilization phase (2 to 7 days). Opioids were titrated to levels of adequate analgesia and stable doses (i.e. requiring 3 doses of rescue medication per day for a minimum of 2 consecutive days). During phase 2, patients underwent conversion to once-daily OROS hydromorphone, using a 5:1 ratio of morphine equivalents to hydromorphone. The doses of hydromorphone were titrated every two days to achieve a stable dose (i.e. requiring 3 doses of rescue medication per day for a minimum of 3 consecutive days). Patients who required doses higher than 96mg/day were discontinued from the study. Phase 3 was a maintenance period of 28 days at the stable dose of OROS hydromorphone. 209 patients were enrolled and 207 patients proceeded to the phase 2. 131 patients (63.3%) completed all three periods of the study. The mean daily dose of OROS hydromorphone at the end of the titration period was 40.0 ± 26.43 mg. The mean number of days required to achieve a stable dose of OROS hydromorphone was 7 (1 to 14). The average daily dose of rescue medication did not differ in the phase 2 and phase 3 of the study (12.7 vs 12.4 mg). Pain was assessed using the BPI. OROS hydromorphone provided significant pain relief compared with baseline at all post-baseline visits (p < 0.0001) and endpoint (p < 0.005). Moreover, a significant reduction in the levels of pain interference with social and physical activity of daily living was observed within one week of starting therapy and was maintained through all the post-baseline visits (p < 0.0001). All the items 9 A-G of the BPI (i.e. general activity, mood, walking
266 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
ability, sleep, normal work, relationships with others, and enjoyment of life) were significantly improved at all post-baseline visits. Significant increments in quality of life, evaluated using the Medical Outcomes Survey (MOS) Short-Form 36 (SF-36), were observed at endpoint compared with baseline (p<0.01) At endpoint, approximately 66% of investigators and 63% of patients rated the treatment with OROS hydromorphone as good, very good or excellent. Finally, OROS hydromorphone significantly improved sleep adequacy, quantity, and problems in studied patients, as assessed by the MOS Sleep Questionnaire (p < 0.0001). Table 3.
Clinical Trials on OROS Hydromorphone in Chronic Pain Management
Type of study (Author, year)
Diagnosis (N. patients)
Design
Duration
Nonrandomized Open Label (Wallace M 2007) [40]
Chronic Low Back Pain (207)
Phase 1 opioid dose stabilization
7 weeks
Randomized Open Label (Hale M 2007) [41]
Osteoarthritis (138)
Efficacy
pain intensity (BPI) pain interference ratings (BPI)
Phase 2 titration
quality of life (SF-36)
Phase 3 maintenance
sleep quantity (MOS)
Opioid:
sleep disturbance
OROS hydromorphone
patient and investigators’ assessment of treatment as very good or excellent
Phase 1 opioid dose stabilization
6 weeks
Phase 2 titration
= improvement in pain relief = reduction of pain intensity
Opioid:
= patient and investigators’ assessment of treatment as very good or excellent
OROS hydromorphone vs
= improvement in physical status (WOMAC)
ER oxycodone
Significant better improvement (p < 0.05) of sleep problems (MOS) in patients treated with OROS hydromorphone
Phase 3 maintenance
OROS Hydromorphone in Chronic Osteoarthritis Pain An open-label, randomized, multicentre trial compared the analgesic efficacy and tolerability of once-daily OROS hydromorphone and twice-daily extended-release oxycodone in 140 patients with chronic, moderate-to-severe, osteoarthritis (OA) pain (Table 3) [41]. To be eligible for enrolment, all patients were required to meet the American College of Rheumatology (ACR) clinical criteria for OA of the knee or hip for 3 months and to be unable to manage their pain with non-opioid analgesics or with asneeded use of an opioid analgesic. Patients receiving daily opioids were excluded. Phase
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 267
1 was designed for randomization, dose-titration, and stabilization (2 weeks). 138 patients received study medications. OROS hydromorphone (n=71) was started at 8 mg/day, whereas ER oxycodone (n=67) was started at 10 mg twice-daily (BID). Each opioid dose was titrated every 2 days to achieve adequate analgesia. Patients who achieved a stable level of analgesia with 64 mg/day of OROS hydromorphone or 160 mg/day of ER oxycodone entered into the phase 2 of maintenance that was conducted for 28 days (4 weeks). Overall drug use was similar in both groups. Mean daily dose of OROS hydromorphone and ER oxycodone, at the end point, was respectively 15.8 ± 10.53 mg/day and 24.0 ± 11.71 mg/day. The results showed equivalent level of pain relief for both opioids. The time to the third day of moderate to complete pain relief was similar in both treatment groups (6.2 days vs 5.5 days for OROS hydromorphone and ER oxycodone, respectively). The mean change of pain intensity score from baseline to endpoint was also similar in both groups (-0.6 vs -0.4 points, on a 4-point ordinal scale, for OROS hydromorphone and ER oxycodone, respectively). No statistically significant differences were observed in mean changes of all scores measured with the Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index. Similar improvement in treatment satisfaction for both patients and investigators were observed with OROS hydromorphone and ER oxycodone. Approximately 68% of patients in both treatment groups rated their treatment as good, very good, or excellent at study endpoint. Similarly, approximately 70% of investigators rated both treatments as good, very good, or excellent at study endpoint. Finally results for MOS Sleep Problems Index I indicated significantly less sleep disruption and daytime somnolence in the OROS hydromorphone group compared with ER oxycodone (p < 0.012), whereas changes on the MOS Sleep Problems Index II were comparable in both groups. Safety and Tolerability of OROS Hydromorphone The safety and tolerability profile of OROS hydromorphone has been well established in 5 clinical trials [36-38; 40-41], involving a total of 1145 patients with chronic malignant and non-malignant pain. As expected, the adverse events most commonly reported by patients receiving OROS hydromorphone involved the gastrointestinal and central nervous system, such as nausea, constipation, somnolence, dizziness, and vomiting (Table 4). The majority of events were mild to moderate in severity and easily managed. In the open-label, randomized, multicentre trial comparing the analgesic efficacy and tolerability of once-daily OROS hydromorphone and twice-daily extended-release oxycodone in patients with chronic, moderate-to-severe, osteoarthritis pain [41], no statistically significant differences were observed among two groups with regard to the incidence, quality, and severity of adverse events (Fig. 6). CONCLUSIONS In chronic pain management, extended-release formulations are preferred to achieve a stable plasmatic analgesic concentration and keep adequate level of analgesia. Patients’ evaluation of treatment showed that once-daily administration is well accepted and tolerated. OROS hydromorphone has been introduced in clinical practice for the management of chronic cancer and non-cancer pain. The OROS push-pull system is used to
268 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 4.
Coluzzi and Mattia
Most Common Adverse Eventsa Occurring in Patients with Chronic Pain Treated with OROS Hydromorphone (N = 1145)
Event
Any event
Nausea
Constipation
Somnolence
Headache
Dizziness
Vomiting
Insomnia
Fatigue
Asthenia
Palangio M
Wallace M
Wallace M
Wallace M
Hale M
[36]
[37]
[38]
[40]
[41]
(N=404)
(N=336)
(N=127)
(N=207)
(N=71)
316 (78.2%)
264 (78.6%)
105
162
56
(82.7%)
(78.3%)
(78.9%)
86
79
30
41
25
(21.3%)
(23.5%)
(23.6%)
(19.8%)
(35.2%)
70
60
23
43
21
(17.3%)
(17.8%)
(18.1%)
(20.7%)
(29.6%)
60
49
16
29
18
(14.9%)
(14.6%)
(12.6%)
(14.0%)
(25.4%)
69
61
14
29
4
(17.1%)
(18.2%)
(11.0%)
(14.0%)
(5.6%)
59
54
14
20
10
(14.6%)
(16.1%)
(11.0%)
(9.7%)
(14.1%)
50
51
20
20
12
(12.4%)
(15.2%)
(15.7%)
(9.7%)
(16.9%)
-
20 (5.9%)
8
19
(6.3%)
(9.2%)
35
18 (5.4%)
-
(8.7%) Peripheral oedema
Pruritus
-
-
9
18
(7.1%)
(8.7%)
10
14
12
(5.1%)
(11.0%)
(5.8%)
-
Sweating
-
-
-
17
(5.6%)
(13.4%)
22 (6.5%)
-
227 (19.8%) 172 (15.0%) 167 (14.6%) 157 (13.7%) 153 (13.4%) 47
45 (3.9%)
-
12
-
45
-
43 (3.7%)
-
(5.8%)
19
261 (22.8%)
(3.9%)
17
28
903 (78.8%)
(4.1%)
(7.9%)
(8.3%) Diarrhoea
-
-
Overall Incidence (%)
40 (3.5%)
-
36 (3.1%)
-
-
22 (1.9%)
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 269 (Table 4) contd....
Event
Palangio M
Wallace M
Wallace M
Wallace M
Hale M
[36]
[37]
[38]
[40]
[41]
(N=404)
(N=336)
(N=127)
(N=207)
(N=71)
-
-
-
Dry mouth
-
Back pain
-
20 (5.9%) -
11
-
-
(5.3%) Confusion
-
9
-
-
-
-
20 (1.7%) 11 (0.9%)
-
(7.1%) Dyspnea
Overall Incidence (%)
9 (0.8%)
8
-
(6.3%)
-
8 (0.7%)
Most common were defined adverse events occurring at a frequency of 5% in each study, except for the trial of Hale et al. [41] (frequency 10%). a
!
"!
#!
$!
%!
Modified from Hale M et al. [35]
Fig. (6). Adverse events occurring in 10% of patients in either the OROS hydromorphone or extended-release (ER) oxycodone group.
deliver hydromorphone in a continuous, monophasic manner over 24 hours, providing constant analgesia. The pharmacokinetic characteristics are dose proportional and are not affected by environmental factors, such as GI motility and pH. Food and alcohol mini-
270 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Coluzzi and Mattia
mally affected the pharmacokinetics of once-daily OROS hydromorphone. This formulation can safely be used to switch patients from previous opioid therapy, without need of transition to IR formulations and without loss of pain control. Dose-conversion studies support the clinical utility of the 5:1 ratio used for the conversion of oral morphine to oral OROS hydromorphone. In patients with chronic low back pain, once-daily OROS hydromorphone significantly reduced pain intensity and pain interference scores on BPI. Improvements were also recorded in sleep interference and quality of life. In patients with moderate to severe osteoarthritis pain, once-daily OROS hydromorphone provided similar pain relief and improvement in physical conditions to that of twice-daily ER oxycodone. In both these studies, the percentage of patients and investigators who considered the treatment very good or excellent significantly increased at endpoint compared to baseline. These results support the use of OROS technology for the treatment of chronic pain, as for other chronic diseases. In the management of chronic pain, OROS technology has the advantages of reduced dosing frequency, more stable drug concentrations, and uniform drug effects, which can result in better adherence to treatment, improved safety profile and optimal therapy outcomes. ABBREVIATIONS ACR
=
American College of Rheumatology
ACT
=
Around-the-clock
AUC
=
Area Under the concentration-time Curve (bioavailability)
AUC0-
=
Area Under the concentration-time Curve for zero to infinity
AUC0-t
=
Area Under the concentration-time Curve for zero to time t
BID
=
Twice-daily
BPI
=
Brief Pain Inventory
Cmax
=
Peak plasma concentration
COER
=
Controlled-Onset Extended Release
DRCOPT
=
Drug-Resin Complex Osmotic Pump Tablets
EOP
=
Elementary Osmotic Pump
ER
=
Extended-release
GI
=
Gastrointestinal
IR
=
Immediate-release
L-OROS
=
Liquid-OROS System
MOS
=
Medical Outcomes Survey
OA
=
Osteoarthritis
OROS
=
Osmotic-controlled Release Oral delivery System
PRN
=
as-needed
SF-36
=
Medical Outcomes Survey (MOS) Short-Form 36
OROS Technology
Frontiers in Drug Design & Discovery, 2009, Vol. 4 271
T1/2
=
Terminal Half-life
Tmax
=
Time to peak plasma concentration
vs
=
versus
WOMAC
=
Western Ontario and McMaster Universities Osteoarthritis Index
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37]
Hanks, G.W.; de Conno, F.; Cherny, N.; Hanna, M.; Kalso, E.; McQuay, H.J. Br. J. Cancer, 2001, 84, 587-93. Kalso, E.; Edwards, J.E.; Moore, R.A.; McQuay, H.J. Pain, 2004, 112, 372-80. Osterberg, L.; Blaschke, T. N. Engl. J. Med., 2005, 353, 487-97. Miaskowski, C.; Dodd, M.J.; West, C.; Paul, S.M.; Tripathy, D.; Koo, P. J. Clin. Oncol., 2001, 19, 4275-79. Sale, J.E.M.; Gignac, M.; Hawker, G. Arthritis Rheum., 2006, 55, 272-8. Claxton, A.J.; Cramer, J.; Pierce, C. Clin. Ther., 2001, 23, 1296-310. Sabate, E. Adherence to long-term terapie: evidence for action, Geneva, World Health Organization, 2003. Coluzzi, F.; Mattia, C. Ther. Clin. Risk. Manag., 2007, 3, 819-29. Conley, R.; Gupta, S.K.; Sathyan, G. Curr. Med. Res. Opin., 2006, 22, 1879-92. Paolino, D.; Fresta, M.; Sinha, P.; Ferrari, M. Drug delivery systems, Webster, J.G.; Encyclopedia of medical devices and instrumentation, 2nd ed, John Wiley & Sons, Inc. 2006. Theeuwes, F. J. Pharm. Sci., 1975, 64, 1987-91. Dong, L.; Shafi, K.; Wong, P.; Wan, J. Drug Deliv. Technol., 2002, 2, 1-5. Wang, C.; Chen, F.; Heng, P.W.S., Li, J.; Xiang, L.; Ye, G. Chem. Pharm. Bull. 2008, 56, 457-63. Bass, D.M.; Prevo, M.; Waxman, D.S. Drug Saf., 2002, 25, 1021-33 Jacox, A.; Carr, D.B.; Payne, R. N. Engl. J. Med., 1994, 330, 651-5. Quigley, C.; Wiffen, P. J. Pain Symptom Manage., 2003, 25, 169-78. Murray, A.; Hagen, N.A. J. Pain Symptom Manage., 2005, 29, 57-66. Sarhill, N.; Walsh, D.; Nelson, K.A. Support. Care Cancer, 2001, 9, 84-96. Babul, N.; Darke, A.C.; Hagen, N. J. Pain Symptom Manage., 1995, 10, 184-6. Fainsinger, R.; Schoeller, T.; Boiskin, M.; Bruera, E. J. Palliat. Care, 1993, 9, 53-5. Lee, M.A.; Leng, M.E.F.; Tiernan, E.J.J. Palliat. Med., 2001, 15, 26-34. Zheng, M.; Mcerlane, K.M.; Ong, M.C. Xenobiotica, 2002, 32, 427-39. Kumar, M.A.; Lin, S. J. Pharm. Pharm. Sci., 2007, 10, 504-18. Hagen, N.A.; Babul, N. Cancer, 1997, 79, 1428-37. FDA asks Purdue Pharma to withdraw Palladone. FDA Consum., 2005, 39, 7. Murray, S.; Wooltorton, E. J. Ayub Med. Coll. Abbottabad, 2005, 173, 756. Angst, M.S.; Drover, D.R.; Lotsch, J.; Ramaswamy, B.; Naidu, S.; Wada, D.R.; Stanski, D.R. Anesthesiology, 2001, 94, 63-73. Drover, D.R.; Angst, M.S.; Valle, M.; Ramaswamy, B.; Naidu, S.; Stanski, D.R.; Verotta, D. Anesthesiology, 2002, 97, 827-836. Vallner, J.J.; Stewart, J.T.; Kotzan, J.A. J. Clin. Pharmacol., 1981, 21, 152-6. Sathyan, G.; Xu, E.; Thipphawong, J.; Gupta, S.K. BMC Clin. Pharmacol., 2007, 7, 3. Gupta, A.; Sathyan, G. J. Pain Symptom Manage., 2007, 33, 19-24. Durnin, C.; Hind, I.D.; Channon, E.; Yates, D.B.; Cross, M. Proc. West. Pharmacol. Soc., 2001, 44, 75-6. Sathyan, G.; Xu, E.; Thipphawong, J.; Gupta, S.K. BMC Clin. Pharmacol. 2007, 7, 2. Sathyan, G.; Sivakumar, K.; Thipphawong, J. Curr. Med. Res. Opin., 2008, 24, 297-305. Quigley, C. Cochrane Libr., 2005, 2, 1-31. Palangio, M.; Northfelt, D.W.; Portenoy, R.K. J. Pain Symptom Manage., 2002, 23, 355-68. Wallace, M.; Rauck, R.L.; Moulin, D.; Thipphawong, J.; Khanna, S.; Tudor, I.C. Int. J. Clin. Pract., 2007, 61, 1671-76.
272 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [38] [39] [40] [41]
Coluzzi and Mattia
Wallace, M.; Rauck, R.L.; Moulin, D.; Thipphawong, J.; Khanna, S.; Tudor, I.C. J. Int. Med. Res., 2008, 36, 343-52. Weinstein, S.M.; Shi, M.; Buckley, B.J.; Kwarcinski, M.A. Clin. Ther., 2006, 28, 86-98. Wallace, M.; Skowronski, R.; Khanna, S.; Tudor, J.; Thipphawong, J. Curr. Med. Res. Opin., 2007, 23, 1-9. Hale, M.; Tudor, I.C.; Khanna, S.; Thipphawong, J. Clin. Ther., 2007, 29, 874-88.
Frontiers in Drug Design & Discovery, 2009, 4, 273-290
273
Drug Delivery Systems Prepared by Membrane Emulsification C. Charcosset* and H. Fessi Laboratoire d’Automatique et de Génie des Procédés, Université de Lyon, UMR CNRS 5007, ESCPE-Lyon, 43 Bd du 11 Novembre 1918, 69 622 Villeurbanne Cedex, France Abstract: Colloidal delivery systems such as oil-in-water emulsions, liposomes, microparticles and nanoparticles, show great potential as a means of delivering a drug to its site of action efficiently, thereby minimizing any unwanted toxic effects. In this point of view, membrane emulsification has received increasing attention over the last 10 years. In the membrane emulsification process, a liquid phase is pressed through the membrane pores to form droplets at the permeate side of a membrane; the droplets are then carried away by a continuous phase flowing across the membrane surface. The purpose of the present paper is to provide a state of the art on membrane processes to prepare drug delivery systems. The following are included: principles of membrane emulsification, colloidal delivery systems, reduction of adverse affects, active principles, encapsulation efficiency, and drug release properties.
INTRODUCTION Colloidal delivery systems such as oil-in-water emulsions, liposomes, microparticles and nanoparticles, show great potential as a means of delivering a drug to its site of action efficiently, thereby minimizing any unwanted toxic effects. For treatments that require repeated administration, via ingestion or injection, and for compounds such as proteins with very short half life, the possibility of a single administration followed by a slow and controllable release is an improvement on the usual forms of drug delivery. The controlled release of drugs to the specific site of action at the therapeutically optimal rate and dose regimen has been a major goal in designing such devices. Liposomes have been used as potential drug carriers instead of conventional dosage forms because of their unique advantages which include ability to protect drugs from degradation, target the drug to the site of action and reduce the toxicity or side effects [1]. However, developmental work on liposomes has been limited due to inherent problems such as low encapsulation efficiency, rapid leakage of water-soluble drug in the presence of blood components and poor storage stability. Polymeric nanoparticles offer some specific advantages over liposomes. For instance, nanoparticles help to increase the stability of drugs/proteins and possess useful controlled release properties.
*Corresponding Author: Tel: +00 33 4 72 43 18 67; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
274 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Charcosset and Fessi
Polymeric nanoparticles and microparticles are classified based on their sizes. The microparticles range in diameter from 1 to 250 μm, while the size of nanoparticles ranges between 10 and 1000 nm [2, 3]. Nanospheres have a matrix type structure with drugs adsorbed at their surface, entrapped in the particle or dissolved in it. Nanocapsules have a polymeric shell and an inner liquid core, the drugs being dissolved in the inner core, or adsorbed at their surface. Nanoparticles have been investigated for the entrapment of a wide variety of drugs, for applications ranging from ophthalmic delivery to carriers in chemotherapy [4, 5]. They are characterized in terms of morphology (i.e. by transmission electron microscopy and scanning electron microscopy), size and size distribution (i.e. by photon correlation spectroscopy), Zeta potential and density (i.e. by isopycnic centrifugation) [5]. Important properties of nanoparticles are their drug entrapment efficacy (amount of drug loaded in the nanoparticles expressed as the percentage of the total amount drug added in the process), and their drug release behaviour (in vitro and in vivo) [5]. In recent years, biodegradable polymeric nanoparticles have attracted considerable attention as potential drug delivery devices in view of their applications, their ability to target particular organs/tissues, as carriers of DNA in gene therapy, and in their ability to deliver proteins, peptides and genes through a peroral route of administration. Several methods for the preparation of nanoparticles are available, involving either a dispersion of preformed polymers or a polymerization of dispersed monomers [4]. Nanospheres prepared by dispersion of preformed polymers involve the use of purified natural molecules or preformed synthetic polymers, and for nanospheres prepared by polymerization reaction the following polymers may be used: polyacrylamide, poly (alkyl methacrylates), poly(alkyl cyanoacrylates) and polyglutaraldehyde. The nanoprecipitation method developed by (Fessi et al.) [6] is an easy and reproducible method involving dispersion of preformed polymers. It is based on the interfacial deposition of a polymer following displacement of a semi-polar solvent miscible with water from a lipophilic solution [7]. The organic phase (solvent, polymer, eventually oil, and drug) is added dropwise under moderate stirring into the aqueous phase (water, and surfactant). An other method to prepare nanoparticles is the interfacial polymerization technique in which two monomers, one oil-soluble and the other water-soluble, are employed and a polymer is formed on the droplet surface [8]. The organic phase (solvent, monomer, eventually oil, and drug) is added into the aqueous phase (water, co-monomer and surfactant). The purpose of the present paper is to provide a state of the art on membrane processes to prepare drug delivery systems. The following are included: principles of membrane emulsification, colloidal delivery systems, reduction of adverse affects, active principles, encapsulation efficiency, and drug release properties. MEMBRANE EMULSIFICATION PRINCIPLES Membrane emulsification has received increasing attention over the last 15 years [914]. The membrane emulsification process is shown in Fig. (1). The dispersed phase is pressed through the pores of a microporous membrane, while the continuous phase flows along the membrane surface. Droplets grow at pore openings until they detach when having reached a certain size. Surfactant molecules in the continuous phase stabilize the newly formed interface, to prevent droplet coalescence im-
Drug Delivery Systems
Frontiers in Drug Design & Discovery, 2009, Vol. 4 275
mediately after formation. The distinguishing feature is that the resulting droplet size is controlled primarily by the choice of the membrane and not by the generation of turbulent droplet break-up. The apparent shear stress is lower than in classical emulsification systems, because small droplets are directly formed by permeation of the dispersed phase through the micropores, instead of disruption of large droplets in zones of high energy density. Besides the possibility of using shear-sensitive ingredients, emulsions with narrow droplet size distributions can be produced. Furthermore, membrane emulsification processes allow the production of emulsions at lower energy input (104-106 J/m3) compared to conventional mechanical methods (106-108 J/m3) [14].
Continuous phase Tangential flow
Droplets/or particles
Membrane Dispersed phase Permeation under applied pressure Fig. (1). Schematic diagram of the membrane emulsification process.
Previous reviews on membrane emulsification [i.e. 11, 15-18] focused on membrane emulsification principles, influence of process parameters, comparison with other methods, and applications. Nakashima et al. [12] provided a review recalling that membrane emulsification was introduced by these authors at the annual Meeting of the Society of Chemical Engineers, Japan, in 1988. The fundamentals of membrane emulsification are presented and the applications: food emulsions, synthesis of monodispersed microspheres and drug delivery systems are described. Gijsbertsen et al. [17] presented a state of the art on membrane emulsification, as well as an analysis of an industrial scale production of culinary cream, for which a microsieve membrane with a low porosity was found the best suitable. Vladisavljevi and Williams [18] provided a very complete review on manufacturing emulsions and particulate products using membranes, ranging from the production of simple o/w and w/o emulsions to multiple emulsions of different types, s/o/w dispersions, coherent solids (silica particles, solid lipid microspheres, solder metal powder), and structured solids (solid lipid microcarriers, gel microbeads, polymeric microspheres, core-shell microcapsules and hollow polymeric microparticles). EXPERIMENTAL DEVICES A schematic picture of a typical membrane emulsification set-up is shown in Fig. (2). The system incorporates a tubular microfiltration membrane, a pump, a feed vessel, and a pressurized (N2) oil container. The dispersed phase is pumped under gas pressure through the pores of the membrane into the continuous phase which circulates through the membrane device. The membrane should not be wetted with the dispersed phase.
276 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Charcosset and Fessi
Therefore, at the beginning of the experiment, the membrane is wetted with the continuous phase, i.e. a hydrophilic membrane for o/w emulsions is wetted with the water phase and a hydrophobic membranes for w/o emulsions is wetted with the oil phase. At the end of the experiment, the membrane is cleaned, using an appropriate solution, until the pure water flux is restored to its original value.
Pressurized vessel Dispersed phase
M Stirrer
M
Membrane module
N2 bottle
Continuous phase
Pump
Fig. (2). Typical experimental set-up for the membrane emulsification process. M: manometer.
The membrane emulsification process may also be carried out in batch mode without tangential flow of the continuous phase [i.e. 19], or in a stirred cell configuration [i.e. 20]. These configurations are particularly suited for the preparation of small amounts of emulsions or microcapsules loaded with high values chemicals. A rotating membrane device was tested to increase the performances of the membrane emulsification process, especially to increase the flux through the membrane [i.e. 21, 22]. Premix membrane emulsification is an other configuration of membrane emulsification. A pre-emulsion with a large droplet size is passed through the porous membrane into the continuous phase, instead of directly passing the oil or water. The droplets of the pre-emulsion are disrupted into fine droplets during their permeation through the membrane. For similar mean pore sizes, the mean droplet size resulting from premix membrane emulsification is smaller than in direct membrane emulsification, which is often an advantage [14, 23]. This technique is called repeated or multi-stage premix membrane emulsification [23, 24]. INFLUENCE OF PROCESS PARAMETERS Membranes Material The most commonly used membranes for the preparation of emulsions are the Shirasu porous glass (SPG) membranes (Ise Chemical Co., Japan), because of their nar-
Drug Delivery Systems
Frontiers in Drug Design & Discovery, 2009, Vol. 4 277
row pores size distribution and tubular shape [25]. The SPG membrane is synthesized from CaO-Al2O3-B2O3-SiO2 type glass which is made from “Shirasu”, a Japanese volcanic ash. The SPG membrane has uniform cylindrical interconnected micropores, a wide spectrum of available mean pore sizes (0.05 - 30 μm), and a high porosity (50-60 %). In addition to the SPG membranes, o/w emulsions were successfully prepared using silicon and silicon nitride microsieves membranes (Aquamarijn Microfiltration BV, The Netherlands) [26-29]. These are made by photolithographic treatment of a silicon wafer and subsequent etching, or electrochemical metal deposition on a skeleton in an electrolysis bath, respectively. These membranes have interesting properties, such as a smooth and flat surface, a very low membrane resistance and narrow pores size distribution. Different pore geometries (circular, square, slit shaped), pore size, pore edges and membrane porosities are available. Polycarbonate track-etch membranes (Millipore, Inc.) having a very narrow pores size distribution were also tested for the preparation of particles [30]. Other commercial microfiltration membranes are attractive because of their availability in very large surface area, and their high flux through the membrane pores: ceramic aluminium oxide (-Al2O3) membranes (Membraflow, Germany) [31], -alumina and zirconia coated membranes (SCT, France) [32], and polytetrafluoroethylene (PTFE) membranes (Advantec Tokyo Ltd., Japan [33, 34] and Goretex Co. Ltd., Japan [35]). W/o emulsions were successfully prepared using microporous polypropylene hollow fibers (Microdyn module, Wuppertal, Germany) [36], macroporous silica glass membranes [37], polytetrafluoroethylene (PTFE) membranes [33, 38], polyamide hollow fibers membrane [39, 40], and home made silica-based monolithic membrane [41]. Membrane Pores Size Several authors have shown that the average droplet diameter, d d , increases with the average membrane pore diameter,
d p , by a linear relationship, for given operating con-
ditions:
d d = cd p
(1)
where c is a constant. For SPG membranes, values of c range typically from 2-10. This range was explained by differences in operating conditions, and by the type of SPG membrane used [42]. For membranes other than SPG, the values reported for c are higher, typically 3-50. Monodispersed emulsions can be produced if the membrane pore-size distribution is sufficiently narrow. Using SPG membranes between 0.4 μm and 6.6 μm mean pore size, Vladisavljevi and Schubert [13] found that the span of the droplet size distribution was lower than that reported for ceramic membranes. Omi et al. [43] using SPG membranes stated that fairly uniform droplets were obtained with a coefficient of variation (CV) around 10 %, due to the uniformity of the membrane pore size. Transmembrane Pressure The membrane emulsification method involves using a transmembrane pressure to force the dispersed phase to permeate through the membrane into the continuous phase.
278 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Charcosset and Fessi
The transmembrane pressure Ptm is defined as the difference between the pressure of the dispersed phase, Pd, and the mean pressure of the continuous phase,
Ptm = Pd
( Pc ,in + Pc ,out ) 2
(2)
where Pc,in and Pc,out are the pressure of the flowing continuous phase at the inlet and at the outlet of the membrane module, respectively. The applied transmembrane pressure required to make the discontinuous phase (i.e. oil) flow can be estimated from the capillary pressure, assuming that the pores are ideal cylinders:
Pc =
4 cos dp
(3)
where Pc is the critical pressure, the o/w interfacial tension, the contact angle of the oil droplet against the membrane surface well wetted with the continuous phase and d p the average pore diameter. The actual transmembrane pressure required to make the discontinuous phase flow may be greater than predicted by eq. 2, due to tortuosities in the pores, irregular pore openings at the membrane surface and the significant effects of surface wettability [44]. The dispersed phase flux Jd is related to the transmembrane pressure according to Darcy’s law:
Jd =
K Ptm μL
(4)
where K is the membrane permeability, L the membrane thickness, and μ the dispersed phase flux viscosity. In cases where the membrane may be assumed to have n uniform cylindrical pores of radius r, the permeability K is given by the Hagen-Poiseuille equation:
K=
nr 2 8
(5)
The emulsification result is expressed in terms of the dispersed phase flux Jd, through the membrane calculated as [31]:
Jd =
Md d A
(6)
where Md is the mass flowrate of the dispersed phase, A the membrane surface area and d the dispersed phase density. The definition of the dispersed phase flux allows the comparison of results from different types or sizes of membrane. The dispersed phase flux is an essential parameter of the economy of the membrane emulsification process. Increasing transmembrane pressure increases the flux of dispersed phase through the membrane, according to Darcy’s law. At high fluxes, the average droplet size and the size distribution tend to increase because of increased droplet coalescence at the membrane surface. Therefore, an increase in flux may be at the ex-
Drug Delivery Systems
Frontiers in Drug Design & Discovery, 2009, Vol. 4 279
pense of droplet size distribution. The effect of transmembrane pressure is dependent on operating conditions, as crossflow velocity and type of surfactant [31, 45-47]. Crossflow Velocity Droplets formed at the membrane surface detach under the influence of the flowing continuous phase. The characteristic parameter of the flowing continuous phase is the cross flow velocity or the wall shear stress. It is shown that the droplet size becomes smaller as the wall shear stress increases and that the influence is greater for small wall shear stresses [32, 44, 46, 48]. The effect of the wall shear stress on reducing droplet size is dependent on the membrane pore size, being more effective for smaller membrane pore sizes [31]. Surfactants The influence of the type of surfactant in the membrane emulsification process has been studied by several authors [31, 37, 45, 47-50]. Surfactants played two main roles in the formation of an emulsion. Firstly, they lowered the interfacial tension between oil and water. This facilitated droplet distribution and in case of membranes lowers the minimum emulsification pressure. Secondly, surfactants stabilize the droplets against coalescence and/or aggregation. Schröder et al. [45] and Schröder and Schubert [31] showed that the type of surfactant used greatly influenced the droplet size. Droplet diameters obtained with Tween 20 were about twice the size of the droplets stabilized with SDS, in agreement with the ratio of equilibrium interfacial tensions. These authors suggested that the interfacial tension force was one of the key forces governing droplet formation during the membrane emulsification process. Van der Graaf et al. [51] carried out droplet formation experiments with a microengineered membrane by measuring the droplet diameter and droplet formation time as a function of the surfactant concentration in the continuous phase. Their experiments confirmed that the interfacial tension influenced the process of droplet formation: higher surfactant concentrations lead to smaller droplets and shorter droplet formation times. Viscosity The viscosity of the dispersed phase has also an important effect on the membrane emulsification process performance. According to Darcy’s law, the dispersed flux is inversely proportional to the dispersed phase viscosity. If the dispersed phase viscosity is high, then the dispersed flux will be low, and as a consequence the droplet diameter will be large compared to the mean pore diameter. Kukizaki and Goto [52] showed that in a system composed of decane containing liquid paraffin and SDS solution containing polyethylene glycol, the resulting droplet diameter increased with increasing waterphase viscosity, while droplet diameter decreased with increasing oil-phase viscosity. Droplet diameter decreased as the ratio of oil-phase to water-phase viscosity increased. However, droplet diameter did not change in the case of a constant viscosity ratio. PREPARATION OF DRUG DELIVERY SYSTEMS A number of emulsions and particles have been prepared so far by the membrane emulsification technique, including simple o/w and w/o emulsions to multiple emulsions of different types, s/o/w dispersions, coherent solids (silica particles, solid lipid micro-
280 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Charcosset and Fessi
spheres, solder metal powder), and structured solids (solid lipid microcarriers, gel microbeads, polymeric microspheres, core-shell microcapsules and hollow polymeric microparticles) [18]. The present review article focuses on the preparation of drug delivery systems with this method, such as emulsions, multiple emulsions, solid-in-oil-in-water dispersions, solid lipid particles, albumin microspheres, alginate microspheres, chitosan microspheres, polymeric nano/microspheres, and polymeric nano/microcapsules. Drug loaded particles prepared by the membrane emulsification technique method are summarized in Table 1. Emulsions Micro/sub-micron emulsions have received a lot of attention in the last years for their use as vehicles for drug administration [53]. These systems are finely dispersed emulsions with droplet sizes in the sub-micron range, consisting of a continuous aqueous phase, a dispersed oily phase, and surfactants that stabilize the interface and form the oily drops. The main advantages of these systems are the high apparent solubility and the bioavailability potential for lipophilic and hydrophilic substance. An o/w submicron-emulsion containing 5 % ectoin was prepared by the membrane emulsification technique [53]. In recent dermatological studies, ectoin was demonstrated to protect skin cells and cell biopolymers from external stress factors such as UV radiation, surfactants contact and heat. The authors showed that in comparison to conventional emulsification methods, membrane emulsification may provide more control of the drop size and size distribution. A drop size of 300-320 nm with a narrow size distribution was achieved using a tubular Al2O3 ceramic membrane (pore size of 200 nm). Based upon the results of ectoin permeability into porcine ear skin, the authors show that the application of ectoin into the skin depends on the drop size of the submicronemulsion. It was found that the formulation prepared with the membrane system was more efficient in absorbing ectoin into the skin than the formulation prepared by homogenizing. Multiple Emulsions Production of multiple emulsion involves the preliminary emulsification of two phases (e.g., w/o, o/w or e/o), followed by secondary emulsification into a third phase leading to a three phase mixture, such as w/o/w, o/w/o or e/o/w. The primary emulsion is usually prepared under intense homogenization conditions to obtain very fine droplets [18]. The secondary emulsification step is carried out under less severe conditions in order to avoid rupture of the liquid membrane between the internal disperse phase and continuous phase of the multiple emulsion. If the second step is carried out using a conventional emulsification device, the external droplets are in most cases highly polydisperse or the entrapment efficiency is very small. Membrane or microchannel emulsification enables to obtain narrow size distribution of the external droplets, and yet to maintain a high encapsulation yield of the internal droplets. The first application proposed was an w/o/w emulsions for the treatment of liver cancer by arterial injection chemotherapy [54-58]. The aqueous anticancer drug solution (epirubicin or carboplation) was mixed with the oil (iodinated poppy-seed oil or lipiodol). This mixture was then sonicated to form submicron sized w/o emulsion, and permeated through a SPG membrane into a glucose solution forming w/o/w emulsions. The
Drug Delivery Systems
Frontiers in Drug Design & Discovery, 2009, Vol. 4 281
efficacy of the w/o/w emulsions was proved clinically, for patients bearing hepatocellular carcinoma nodules recurrent after hepatectomy. Table 1.
Drug Loaded Particles Prepared by Membrane Emulsification
Drug and Applications
Systems
Ref.
epirubicin or carboplation treatment of liver cancer
w/o/w emulsions
[54-58]
diclofenac sodium long-term pain control
s/o suspension
[59]
ectoin topical administration for neurodermitic skin
o/w submicron emulsion
[53]
adriamycin, daunomycin anti-cancer drugs
PLGA microspheres
[60]
insulin
S/O/W emulsion
[61]
insulin
chitosan microsphere
[62-64]
rifampicin
PLGA micropsheres
[65]
hemoglobin
alginate micropsheres
[24]
spironolactone paediatric use
nanocapsules
[66]
progesterone
PLA and PLGA microspheres
[67]
riboflavin
albumin microspheres
[68]
lidocaine-hydrochloride sodium salicylate 4-acetaminophen
Alginate microspheres
[69]
blue dextran
PLGA microspheres
[70]
vitamin E
SLN
[71, 72]
vitamin E
nanocapsules
[73]
insulin
PLA and PLA/PLGA microcapsules
[74, 75]
vitamin B12
solid lipid microcapsule
[76]
lidocaine-hydrochloride sodium salicylate 4-acetaminophen
PCL microcapsules
[77]
bovine serum albumin
chitosan microspheres
[78]
Model Drugs
Solid-in-Oil-in-Water Dispersions A S/O/W dispersion for oral administration of insulin was prepared by passing a preliminary emulsified S/O/W dispersion several times through a SPG membrane [61]. The S/O suspension containing surfactant-coated insulin dispersed in soybean oil was first mixed with aqueous solution containing a hydrophilic surfactant, sodium cholate and Dglucose to prepare a coarse S/O/W dispersion using a conventional rotor stator homoge-
282 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Charcosset and Fessi
nizer. The coarse S/O/W dispersion was then homogenized by passing through the SPG membrane to prepare a monodispersed emulsion. The surfactant-coated insulin was prepared by mixing an aqueous solution of insulin with a hexane solution containing a lipophilic surfactant, followed by freeze-drying. The S/O/W dispersion showed the hypoglycemic activity for a long period after oral administration to rats, owing to the conversion of insulin into a lipophilic complex and uniform droplets of S/O suspension. With the same technique, (Piao et al.) [59] prepared a S/O suspension for oral administration of diclofenac sodium (DFNa), a non-steroidal anti-inflammatory drug widely used in long-term pain control. DFNa was suspended in medium chain triglyceride by forming a complex with an edible lipophilic surfactant. Two types of suspensions were prepared through the membrane emulsification technique for two different membrane pore sizes. The preparations were evaluated according to the degree of gastric damage following multiple oral administration in rats. It was shown that gastric ulcerogenecity of DFNa was reduced by the surfactant-drug complexes, at doses up to 12 mg/kg, whereas severe gastric damage was observed upon oral administration of the aqueous solution at doses of 6 mg/kg. Solid Lipid Particles Solid lipid particles are realised by exchanging the liquid lipid (oil) of the emulsions by a solid lipid, which means lipids are solid at room temperature but also at body temperature [79-81]. The use of solid lipids instead of liquid oils is a very attractive idea to achieve controlled drug release, because drug mobility in a solid is considerably lower compared with liquid oil. Solid lipid nanoparticles are prepared using high pressure homogenization techniques [79-80]. Two general approaches, the hot and the cold homogenization techniques, can be used. Solid lipid nanoparticles were prepared using a membrane process [71, 72]. The lipid phase was pressed, at a temperature above the melting point of the lipid, through the membrane pores allowing the formation of small droplets. The aqueous phase circulated inside the membrane module, and swept away the droplets forming at the pore outlets. SLN were formed by the following cooling of the preparation to room temperature. The influence of process parameters (aqueous phase and lipid phase temperatures, aqueous phase cross-flow velocity and lipid phase pressure, membrane pore size) on both the SLN size and the lipid phase flux were investigated. Vitamin E-loaded SLN were prepared, and their stability was demonstrated. With the same technique, solid lipid microcapsules (SLMCs) were prepared by membrane emulsification using tripalmitin (a high melting point triglyceride) as the lipid phase [76]. W/o/w emulsions encapsulating vitamin B12 as a model drug were prepared by a two-step membrane emulsification technique using SPG membranes. The resultant w/o/w emulsions were immediately cooled to solidify the oil phase, and were then filtered. Uniformly sized SLMCs with high encapsulation yields of vitamin B12 around 95 % were obtained, with a mean particle diameter between 3 and 32 μm. No leakage of vitamin B12 from the uniformly sized SLMCs was observed at body temperature over a period of 10 days, when the SLMCs were redispersed into normal saline. Albumin Microspheres Albumin microspheres have found many applications in the medical diagnosis and treatment in recent years. Albumin micropheres possess advantages over other types of
Drug Delivery Systems
Frontiers in Drug Design & Discovery, 2009, Vol. 4 283
polymeric microspheres, because they are non-antigenic and metabolizable and can accommodate a variety of drugs in a relatively non-specific fashion [82]. The distribution of particles in the body is dependent on their size and their surface characteristics. Monodisperse albumin microspheres were prepared by both chemical and thermal hardening methods via membrane emulsification [68, 82]. A monodisperse w/o emulsion was first prepared by passing the albumin solution through a hydrophobic SPG membrane into kerosene. After heating the emulsion, monodisperse albumin microspheres were obtained and the heat-denaturing process itself did not affect their monodispersity once uniform droplets had been prepared. The shape and size of the albumin microspheres strongly depended on the concentration of the albumin solution and the heatdenaturing temperature. Loaded riboflavin released quite rapidly from the albumin microspheres. Increasing both albumin concentration and denaturing temperature caused the drug release to be sustained. According to the authors [68], a rigid and dense matrix might form within the albumin microspheres. Alginate Microspheres Alginate microspheres with biocompatibility may be prepared under mild conditions, even physiological conditions, so they are suitable for drug delivery devices. Especially in recent years, there has been increasing interest in the study of the use of alginate microspheres as drug-delivery systems of proteins and polypeptides [83]. Monodisperse Ca-alginate microspheres were prepared using the membrane emulsification method [24, 69]. Monodisperse microspheres were obtained with a mean size of 4 μm. Lidocaine.HCl (cationic), sodium salicylate (anionic) and 4-acetamidophenol (non-ionic) were selected as ionic model drugs and included in the alginate microspheres. Lidocaine.HCl (cationic) release was more retarded than that of the anionic drug. The authors explained this fact by the electrostatic attraction between the negative charge of the ionized carboxyl group in the alginate chain and the positive charge of the cationic drug. In acidic release medium, a slow release was observed due to the low swelling characteristic and the increased viscosity of alginate, regardless of ionic type of drug. Chitosan Microspheres Chitosan microspheres have important application in controlled release of protein and peptide drugs, because they show excellent mucoadhesive and permeation enhancing effect across the biological surfaces. In the conventional preparation methods of chitosan microspheres, the w/o emulsion is usually prepared by mechanical stirring method, and the droplets are solidified by glutaraldehyde. There exist limitations such as broad size distribution, de-activity of bio-drug and difficulty in drug release because protein and peptide drugs have the same amino group as chitosan. A method to prepare chitosan microsphere was established by combining the membrane emulsification technique and a step-wise crosslinking method [62-64]. The chitosan/acetic acid aqueous solution was pressed through the pores of a SPG membrane into a paraffin/petroleum ether mixture containing PO-500 emulsifier, to form a w/o emulsion with uniform droplet size. The uniform droplets were solidified by a two-step crosslinking method. A tripolyphosphate (TPP) solution was dropped gradually in the emulsion, TPP diffused into the droplet to crosslink chitosan by an ionic linkage, generating a microgel. As a result, chitosan microspheres were obtained with a coefficient of
284 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Charcosset and Fessi
variation less than 11%, an encapsulation efficiency of 80%, a high chemical stability of insulin (>95%), a low burst release and steady release behavior. On the other hand, pH-sensitive quaternized chitosan microspheres were prepared by combining SPG membrane emulsification technique and a thermal-gelation method [78]. In this preparation process, the mixture of quaternized chitosan solution and -glycerophosphate (--GP) was used as water phase and dispersed in oil phase to form uniform w/o emulsion. The droplets then solidified into microspheres at 37°C by thermal-gelation. The coefficient of variation of the obtained microspheres diameters was below 15 %. Bovine serum albumin (BSA) as a model drug was encapsulated in microspheres, and it was released rapidly in acid solution and slowly in neutral medium. Polymeric Nano/Microspheres Micro- and nanospheres offer a high potential to be used as carriers for drug therapy as they can provide a slow release of drug and thus reduce the frequency of drug administration and the toxicity levels. A vast range of biopolymers and biopolymers combinations has been considered [70]. Once the biopolymer enters the body, the environmental conditions cause it to degrade in a predictable manner into monomers that are already present in the body. This degradation can be controlled through the polymer composition and the characteristics of the administration, to gradually release the encapsulated drugs. Poly(lactic-co-glycolic acid) (PLGA) is a highly biocompatible and biodegradable synthetic polymer, which is hydrolytically degraded into non-toxic oligomer and finally to lactic acid and glycolic acid [65, 84]. The most general and simple preparation technique of PLGA microspheres is a solvent evaporation from an o/w emulsion. An organic solvent (mainly dichloromethane, chloroform, or acetonitrile, etc.) dissolving drugs and PLGA is added under mechanical stirring into an aqueous phase containing a dispersion stabilizer (generally polyvinyl alcohol) to obtain an o/w emulsion. The microspheres are then obtained by evaporation of the organic solvent. However, with mechanical stirring, relatively wide size distributions are measured and or polydisperse microspheres are prepared due to the fusion or coagulation of emulsion droplets under mechanically agitation of emulsion droplets. To overcome the instability of the emulsions and to obtain monodispersed PLGA microspheres, the membrane emulsification technique was applied. The first attempt to prepare poly(D,L-lactide) (PLA) and poly(lactic-co-glycolic acid) (PLGA) microspheres by the membrane emulsification technique was reported by (Shiga et al.) in 1996 [67]. An emulsion was prepared with an organic solvent (methylene chloride) as the dispersed phase and water containing sodium lauryl sulphate as the continuous phase. The emulsion thus obtained was gently stirred with a magnetic stirrer until the solvent completely evaporated (solvent evaporation method), or was mixed with the same volume of cold methanol to precipitate PLA or PLGA (solvent extraction method). However, the authors concluded that in the present system of emulsion, the surfactant used was limited to ionic ones and the amount of polymers available for the formation of microspheres was inevitably too small in concentration to entrap sufficient amounts of drug (progesterone). Later, several preparations of PLA and PLGA microspheres by the membrane emulsion technique were reported. Monodispersed rifampicin (RFP)-loaded PLGA micropsheres were prepared using the membrane emulsion technique with SPG mem-
Drug Delivery Systems
Frontiers in Drug Design & Discovery, 2009, Vol. 4 285
branes, followed by solvent evaporation [65]. Rifampicin is a drug used for the treatment of tuberculosis. RFP/PLGA microspheres with average diameters of 1.3, 2.2, 5.2 and 9 μm were obtained. They were relatively monodisperse with a variation coefficient between 7 and 16 %. The loading efficiency of RFP was in the range between 50 and 67 % independent of the microsphere size. The release ratio of RFP was measured in pH 7.4 PBS at 37°C. From RFP/PLGA microspheres with average diameters of 1.3 and 2.2 μm, almost 60 % of RFP loaded in the microspheres was released in the initial day and the release was terminated within 10 days. From RFP/PLGA microspheres with average diameters of 5.2 and 9 μm, the release of RFP was observed even 20 days after release started. PLGA microspheres containing adriamycin and daunomycin (anti-cancer drugs) were also produced using the membrane emulsification technique [60]. PLGA was dissolved in methylene chloride and adriamycin or daunomycin was dissolved in an aqueous phase. The resulting mixture, after emulsification by high speed homogenization is added through a membrane into an aqueous solution to make a w/o/w emulsion. The formed emulsion was then gently stirred at room temperature until complete evaporation of the solvent. The particles were finally collected by centrifugation, washed with distilled water and lyophilized. Particles in the range of 50 μm were thus obtained. With the same technique, microparticles of PLGA containing a water-soluble model drug (blue dextran) were prepared [70]. The dispersed phase (PLGA dissolved in dichloromethane (DMC)) was injected through the membrane pores into an aqueous phase, followed by solvent removal. Particles in the range of 40 to 140 μm and encapsulation efficiencies of up to 100 % were obtained for concentrations of 15 % PLGA dissolved in the DCM and injected through a 40 μm membrane (Micropore Technologies Ltd., UK). Different PLGA concentrations, particle size and osmotic pressures were considered in order to find their effect on encapsulation efficiency. Nanoparticles were also obtained using a nanoprecipitation reaction [73, 85]. The organic phase containing a solvent, a polymer and a drug, was used as the dispersed phase, and the aqueous phase containing a surfactant as the continuous phase. Nanoparticles were also obtained using interfacial polymerization [85]. The organic phase contained a solvent, a monomer, and a drug and the aqueous phase a co-monomer and a surfactant. Examples of formulation are given in Table 2. Formulations 1 and 3 involve the nanoprecipitation reaction, and Formulation 2 the interfacial polymerization reaction. Polymeric Nano/Microcapsules Microencapsulation is defined as the embedding of at least one ingredient (active agent or core) into at least one other (shell material). The two main capsule structures are: embedded particles (or core) in the shell of the capsule (matrix microcapsules), and a continuous shell surrounding the core (core-shell microcapsules). Core-shell microcapsules are preferred if the active agent has to be released slowly over a long time. Nanocapsules have been obtained by adding an oil to the dispersed phase of the formulations used for the nanoparticles preparation (Formulation 3 for vitamin E loaded nanocapsules). Moreover, nanocapsule-loaded spironolactone (a diuretic for children) were prepared [66]. The optimized formulations in a becher and by the membrane technique lead to the preparation of spironolactone-loaded nanocapsules with a mean size of 320 nm and 400 nm respectively, a high encapsulation efficiency (96.21% and 90.56%
286 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 2.
Charcosset and Fessi
Examples of Formulations Used for the Nanoparticle Preparations [Charcosset and Fessi 73, 85] Formulation 1 Organic phase:
Aqueous phase:
Acetone: 0.6 l
Water: 1.2 l
Polycaprolactone: 15 g
Tween 20: 2.04 g
Formulation 2 Organic phase:
Aqueous phase:
Acetone: 0.6 l
Water: 1.2 l
Span 80: 1.2 g
Tween 20: 2.04 g
Hexyl laurate: 6 g
Diethylene triamine: 19.65 g
Sebacoyl chloride: 3.3 g Formulation 3 Organic phase:
Aqueous phase:
Acetone: 0.6 l
Water: 1.2 l
Polycaprolactone: 15 g
Tween 20: 2.04 g
Vitamin E: 15 g
respectively), both stable for 6 months (Fig. 3). The release of spironolactone from nanocapsules was rapid and complete in a simulated gastric fluid. The optimized formulations lead to a high drug-concentration in the liquid preparation (1.5 mg/ml) allowing minimizing the preparation volume administered for children medication. Also, recourse to spironolactone nanoencapsulation should enhance its oral bioavailability and probably its efficiency. Biodegradable poly(lactide) (PLA) microcapsules with various sizes were prepared by combining the membrane emulsification and w1/o/w 2 double emulsion-solvent method [74, 75]. A water phase was used as the continuous phase, and a mixture solvent of dichloromethane (DCM) and toluene dissolving PLA and Arlacel 83 was used as the dispersed phase. The two solutions were emulsified by a homogenizer to form a w1/o primary emulsion. The primary emulsion was permeated through the pores of a SPG membrane into the continuous phase to form the w1/o/w2 double emulsion. Recombinant human insulin (rhI), as a model protein, was encapsulated into the microcapsules with difference sizes. Its encapsulation efficiency and cumulative release were investigated. The authors underline that the advantage of preparing drug-loaded microcapsules by membrane emulsification technique is that the size of microcapsules can be controlled and thus the drug release profile can be adjusted by changing the size of the microcapsules. Moreover, higher encapsulation efficiency was found when compared with the conventional mechanical stirring method.
Drug Delivery Systems
Frontiers in Drug Design & Discovery, 2009, Vol. 4 287
Fig. (3). Transmission electron microscopy of spironolactone-loaded NC obtained by the membrane emulsification technique [66].
Monodispersed microcapsules of a biodegradable polymer, polycaprolactone (PCL) were also prepared by controlling various conditions of membrane emulsification [77]. PCL was dissolved in a mixture of dichloromethane and PVA, and was used as an oil phase. The aqueous phase containing sodium dodecyl sulphate and the model drug (lidocaine-hydrochloride, sodium salicylate or 4-acetaminophen) was used as continuous phase. Controlling membrane emulsification parameters, uniform PCL microcapsules with about 5 μm of the mean size were prepared. CONCLUSION Colloidal delivery systems have a great potential as a means of delivering a drug to its site of action, thereby minimizing any unwanted toxic effects. Colloid delivery systems include emulsions, liposomes, microparticles and nanoparticles. Membrane emulsification was introduced over fifteen years ago, as a new emulsification technique based on microporous membranes. This new process is increasingly reported for the preparation of drug delivery systems. The main advantages of this method are the control of the particle size by an appropriate choice of the pores size and the process parameters, and the potentiality of scaling-up by increasing the membrane surface area. This last advantage might be essential when the preparations have to be produced at the industrial scale. A number of active principles have been included in such systems, including anti-cancer drugs, pain control agents and insulin. Some of the presented studies describe the protocol of preparation and the characterization of the obtained drug release systems. Other works include in vivo studies. The purpose of this review article is to provide a state of the art of the membrane emulsification method to prepare drug release systems. In the future, other drug systems should be tested, and more complete in vivo and in vitro studies should be carried out. Hopefully, these new studies will lead to industrial preparations and commercialisations of drug delivery systems prepared by membrane emulsification.
288 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Charcosset and Fessi
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8]
[9] [10] [11] [12] [13] [14]
[15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34]
[35] [36] [37] [38] [39]
Soppimath, K.S.; Aminabhavi, T.M.; Kulkarni, A.R.; Rudzinski, W.E. J. Control Release, 2001, 70, 1-20. Douglas, S.J.; Davies, S.S.; Illum L. Crit. Rev. Ther. Drug Carrier Syst., 1987, 3, 233-261. Kreuter, J. Nanoparticles, Colloidal Drug Delivery Systems Kreuter, J. Ed., Marcel Dekker, New York, 1994, pp. 219-342. Allémann, E.; Gurny, R.; Doelker E. Eur. J. Pharm. Biopharm., 1993, 39, 173-191. Couvreur, P.; Barratt, G.; Fattal, E.; Legrand, P.; Vauthier, C. Crit. Rev. Ther. Drug Carrier Syst., 2002, 19, 99-134. Fessi, H.; Puisieux, F.; Devissaguet, J.P.; Ammoury, N.; Benita, S. Int. J. Pharm., 1989, 55, R1-R4. Govender, T.; Stolnik, S.; Garnett, M.C.; Illum, L.; Davis H. S. J. Control. Release, 1999, 57, 171185. Hincal, A.A.; Kas, H.S. Microencapsulation Technology: Interfacial polymerization method, in Wise, D. Ed., Handbook of Pharmaceutical Controlled Release Technology; Marcel Dekker, New York, 2000, pp. 271-285. Peng, S.J.; Williams, R.A. Chem. Eng. Res. Des., 1998, 76, 894-901. Peng, S.J.; Williams, R.A. Part. Syst. Charact., 1998, 15, 21-25. Joscelyne, S.M.; Trägårdh, G. J. Memb. Sci., 2000, 169, 107-117. Nakashima, T.; Shimizu, M.; Kukizaki, M. Adv. Drug Deliv. Rev., 2000, 45, 47-56. Vladisavljevi, G.T.; Schubert, H. Desalination, 2002, 144, 167-172. Altenbach-Rehm, J.; Suzuki, K.; Schubert, H. Production of O/W-emulsions with narrow droplet size distribution by repeated premix membrane emulsification. 3ième Congrès Mondial de l’Emulsion, 2427 September 2002, Lyon, France. Charcosset, C.; Limayem, I.; Fessi, H. J. Chem. Biochem. Technol., 2004, 79, 209-218. Charcosset, C.; Fessi, H. Rev. Chem. Eng., 2005, 21, 1-32. Gijsbersten-Abrahamse, A.J.; van der Padt, A.; Boom, R.M. J. Memb. Sci., 2004, 230, 149-159. Vladisavljevi, G.T.; Williams, R.A. Adv. Colloid Interface Sci., 2005, 113, 1-20. Kukizaki, M.; Goto, M. J. Memb. Sci., 2007, 299,190-199. Stillwell, M.T.; Holdich, R.G.; Kosvintsev, S.R.; Gasparini, G.; Cumming, I.W. Ind. Eng. Chem. Res., 2007, 46, 965-972. Aryanti, N.; Williams, R.A.; Hou, R.; Vladisavljevic, G.T. Desalination, 2006, 200, 572-574. Schadler, V.; Windhab, E.J. Desalination, 2006, 189, 130-135. Vladisavljevi, G.T.; Shimizu, M.; Nakashima, T. J. Memb. Sci., 2004, 244, 97-106. Ribeiro, A.J.; Silva, C.; Ferreira, D.; Veiga, F. Eur. J. Pharm. Sci., 2005, 25, 31-40. Nakashima, T.; Shimizu, M.; Kukizaki, M. Key Eng. Mat., 1991, 61-62, 513-516. Zhu, J.; Barrow, D. J. Memb. Sci., 2005, 261, 136-144. Brans, G.; van der Sman, R.G.M.; Schröen, C.G.P.H.; van der Padt, A.; Boom, R.M. J. Memb. Sci., 2006, 278, 239-250. Brans, G.; Kromkamp, J.; Pek, N.; Gielen, J.; Heck, J.; van Rijn, C.J.M.; van der Sman, R.G.M.; Schröen, C.G.P.H.; Boom, R.M. J. Memb. Sci., 2006, 278, 344-348. Geerken, M.J.; Lammertink, R.G.H.; Wessling, M. Colloids Surf. A., 2007, 292, 224-235. Tangirala, R.; Revanur, R.; Russell, T.P.; Emrick, T. Langmuir, 2007, 23, 965-969. Schröder, V.; Schubert, H. Spec. Publ. –R.Soc. Chem., 1999, 227, 70-80. Joscelyne, S.M.; Trägårdh, G. J. Food Eng., 1999, 39, 59-64. Suzuki, K.; Fujiki, I.; Hagura, Y. Food Sci. Technol., 1998, 4, 164-167. Kanichi, S.; Yuko, O.; Yoshio, H. Properties of solid fat O/W emulsions prepared by membrane emulsification method combined with pre-emulsification. 3ième Congrès Mondial de l’Emulsion, 24-27 September 2002, Lyon, France. Yamazaki, N.; Yuyama, H.; Nagai, M.; Ma, G.-H.; Omi, S. J. Dispersion Sci. Technol., 2002, 23, 279-292. Sotoyama, K.; Asano, Y.; Ihara, K.; Takahashi, K.; Doi, K. J. Food Sci., 1999, 64, 211-215. Fuchigami, T.; Toki, M.; Nakanishi, K. J. Sol-Gel Sci. Technol., 2000, 19, 337-341. Yamazaki, N.; Naganuma, K.; Nagai, M.; Ma, G.-H.; Omi, S. J. Dispersion Sci. Technol., 2003, 24, 249-257. Giorno, L.; Li, N.; Drioli, E. J. Memb. Sci., 2003, 217, 173-180.
Drug Delivery Systems [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50]
[51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 289
Giorno, L.; Mazzei, R.; Oriolo, M.; De Luca, G.; Davoli, M.; Drioli, E. J. Colloid Interface Sci., 2005, 287, 612-623. Hosoya, K.; Bendo, M.; Tanaka, N.; Watabe, Y.; Ikegami, T.; Minakuchi, H.; Nakanishi, K. Macromol. Mater. Eng., 2005, 290, 753-758. Omi, S. Colloids Surf. A., 1996, 109, 97-107. Omi, S.; Ma, G.-H.; Nagai, M. Macromol. Symp., 2000, 151, 319-330 Williams, R.A.; Peng, S.J.; Wheeler, D.A.; Morley, N.C.; Taylor, D.; Whalley, M.; Houldsworth, D.W. Chem. Eng. Res. Des., 1998, 76, 902-910. Schröder, V.; Behrend, O.; Schubert, H. J. Colloid Interface Sci., 1998, 202, 334-340. Katoh, R.; Asano, Y.; Furuya, A.; Sotoyama, K.; Tomita, M. J. Memb. Sci., 1996, 113, 131-135. Yuyama, H.; Watanabe, T.; Ma, G.-H.; Nagai, M.; Omi, S; Colloids Surf. A., 2000, 168, 159-174. Kobayashi, I.; Yasuno, M.; Iwamoto, S.; Shono, A.; Satoh, K.; Nakajima, M. Colloids Surf. A., 2002, 207, 185-196. Kandori, K.; Kishi, K.; Ishikawa, T. Colloids Surf., 1991, 55, 73-78. Muschiolik, G.; Dräger, S.; Scherke, I.; Rawel, H.M.; Stang, M. Protein-stabilized emulsions prepared by the micro-porous glass method, in Food Colloids: Proteins, lipids and polysaccharides, Dickinson Ed., Royal Society of Chemistry, Cambridge, 1997, p. 393-400. van der Graaf, S.; Schroën, C.G.P.H.; van der Sman, R.G.M.; Boom R.M. J. Colloid Interface Sci., 2004, 277, 456-463. Kukizaki, M.; Goto, M. Membrane, 2006, 31, 215-220. Ebrahimi, M., Lavi, G., Schmidts, T., Runkel, F.; Czermak, P. Desalination, 2008, 224, 40-45. Higashi, S.; Shimizu, M.; Setoguchi, T. Drug Deliv. Systems, 1993, 8, 59-61. Higashi, S.; Shimizu, M.; Nakashima, T.; Iwata, K.; Uchiyama, F.; Tateno, S.; Tamura, S.; Setogushi, T. Cancer, 1995, 75, 1245-1254. Higashi, S.; Maeda, Y.; Kai, M.; Kitamura, T.; Tsubouchi, H.; Tamura, S.; Setoguchi, T. HepatoGastroenterology, 1996, 43, 1427-1430. Higashi, S.; Tabata, N.; Kondo, KH.; Maeda, Y.; Shimizu, M.; Nakashima, T.; Setoguchi, T. J. Pharmacol. Exp. Ther., 1999, 289, 816-819. Higashi, S.; Setoguchi, T. Adv. Drug Deliv. Rev., 2000, 45, 57-64. Piao, H.; Hirata, A.; Yokoyama, H.; Fujii, T.; Shimizu, I.; Ito, S.; Kamiya, N.; Goto, M. Pharm. Devel. Technol., 2007, 12, 321-325. Costa, M.S.; Cardoso, M.M. Desalination, 2006, 200, 498-500. Toorisaka, E.; Ono, H.; Arimori, K.; Kamiya, N.; Goto M. Int. J. Pharm., 2003, 252, 271-274. Wang, L.Y.; Gu, Y.H.; Zhou, Q.Z.; Ma, G.H.; Wan, Y.H.; Su, Z.G. Colloids Surf. B., 2006, 50,126135. Wang, L.Y.; Gu, Y.H.; Su, Z.G.; Ma, G.H. Int. J. Pharm., 2006 , 311, 187-195. Wang, L.-Y.; Ma, G.-H.; Su, Z.-G. J. Control. Release, 2005, 106, 62-75. Ito, F.; Makino, K. Colloids Surf. B., 2004, 39, 17-21. Limayem, I.; Charcosset, C.; Sfar, S.; Fessi, H. Int. J. Pharm., 2006, 325, 124-131. Shiga, K.; Muramatsu, N.; Kondo, T. J. Pharm. Pharmacol., 1996, 48, 891-895. Muramatsu, N.; Nakauchi, K. J. Microencapsulation, 1998, 15, 15-723. You, J.O.; Park S. B.; Park H.Y.; Haam S., Chung C.H., Kim, W.S. J. Microencapsulation, 2001, 18, 521-532. Gasparini G., Kosvintsev S.R.; Stillwell, M.T.; Holdich, R.G. Colloids Surf. B., 2008, 61, 199-207. Charcosset, C.; El-Harati, A.A.; Fessi, H. J. Control. Release, 2005, 108, 112-120. El-Harati, A.A.; Charcosset, C.; Fessi, H. Pharm. Dev. Technol., 2006, 11, 153-157. Charcosset, C.; Fessi, H. Drug Dev. Ind. Pharm., 2005, 31, 987-992. Liu, R.; Ma, G.H.; Wan, Y.H.; Su, Z.G. Colloids Surf. B., 2005, 45, 144-153. Liu, R.; Huang, S.S.; Wan, Y.H.; Ma, G.H.; Su, Z.G. Colloids Surf. B., 2006, 51, 30-38. Kukizaki, M.; Goto, M. Colloids Surf. A., 2007, 293, 87-94. Yun, T.H.; Kim, K.S.; Cho, S.H.; Youm, K.H. Memb. J., 2007, 17, 67-69. Wu, J.; Wei, W.; Wang, L.Y.; Su, Z.G.; Ma G. H. Colloids Surf. B., 2008, 63, 164-175. Müller, R.H.; Dingler, A.; Schneppe, T.; Gohla. Handbook of Pharmaceutical Controlled Release Technology, Wise, D. Ed., Marcel Dekker, New York, 2000, pp.359-376. Müller, R.H.; Radtke, M.; Wissing, S.A. Adv. Drug Deliv. Rev., 2002, 54, S131-S155. Mehnert, W.; Mäder, K. Adv. Drug Deliv. Rev., 2001, 47, 165-196.
290 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [82] [83] [84] [85]
Charcosset and Fessi
El-Mahdy, M.; Ibrahim, E.S.; Safwat, S.; El-Sayed, A.; Ohshima, H.; Makino, K.; Muramatsu, N.; Kondo, T. J. Microencapsulation, 1998, 15, 661-673. Liu, X.D.; Bao, D.C.; Xue, W.M.; Xiong, Y.; Yu, W.T.; Yu, X.J.; Ma, X.J.; Yuan, Q. J. Appl. Polym. Sci., 2003, 87, 848-852. Mundargi, R.C.; Babu, V.R.; Rangaswamy, V.; Patel, P.; Aminabhavi, T.M. J. Control. Release, 2008, 125, 193-209. Charcosset, C.; Fessi, H. J. Memb. Sci., 2005, 266, 115-120.
Frontiers in Drug Design & Discovery, 2009, 4, 291-332
291
In Silico ADME Approaches Nuria E. Campillo* and Juan A. Páez* Instituto de Química Médica (CSIC), Juan de la Cierva, nº 3, 28006-Madrid, Spain Abstract: During the past decade, the pharmaceutical industry has invested considerably in technologies that have the potential to increase throughput in discovery projects. With the increase in the numbers of molecules synthesized in a typical drug discovery program, as well as the large amount of information utilized in the selection of a drug candidate, it is very useful and necessary ADME and pharmacokinetic information during the discovery process. Over the past decade, many in vitro, and even in vivo, ADME/PK screening have been developed and routinely deployed to generate this information in support of drug discovery efforts. In the past few years, alternative methods as in silico methods have been published. The great challenge for in silico methods is generation of models that correlate more closely with in vivo systems. This review attempts to summarize the in silico approaches taking into account the (i) ways of evaluating a molecule for ADME properties (type of descriptor) and (ii) the employed method for prediction of ADME properties. Progress in this area will be discussed, focussing on oral absorption, blood-brain barrier and metabolism, and some aspects on excretion will be also briefly commented.
Key Word: Oral absorption, blood-brain barrier, metabolism, excretion, ADME, in silico, QSPR, CYPs. 1. INTRODUCTION The success of a drug-independently of its pharmacological activity- depends on the difficulty or limitations that the drug finds during its journey through the body until it reaches the biological target. These difficulties are dependent on ADME (Absorption, Distribution, Metabolism and Elimination or Excretion) properties. The ideal oral drug will not have problems and it will be rapidly and completely absorbed from the alimentary canal, it will find its way directly and specifically to its site of action and so on. But the reality is quite different to this ideal situation, drugs rarely have the correct combination of characteristics to exhibit the ideal pharmacokinetic profile in order to have a perfect journey, allowing the drug development program to be focused on a selected group of potential lead candidates [1,2]. During the last decade preclinical ADME screening strategies have emerged as a tool for the early identification of weak drug candidates and the development of correct ADME profile together with the early prediction of these properties have taken an *Corresponding Authors: E-mail:
[email protected] and
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
292 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
utmost importance in the development of a drug. This fact is due to different aspects such as unsuitable drugs/cost drug development, undesirable side-effects, the new REACH regulation [3] and the development of alternative methods to the animal experimentation. In the following lines, we will comment the importance of these aspects on drug discovery and development. According to W. Bains, chief scientific officer of Amedis Pharmaceuticals (Royston, UK), the 40%-60% of developing drugs failed during the clinical trials because of ADME/Tox deficiencies [4]. In addition to the failure of the drugs due to the poor ADME properties, it must be taken into account the undesirable side-effect once the drugs are in the market. Hodgon et al. [2] presented an interesting example of the undesirable side-effects that happened with a calcium channel blocker, mibefradil (Posicor) from Roche (Nutley, NJ) [5]. Posicor was used to control hypertension and angina, however, this drug is also a potent inhibitor of liver metabolism, and leads to the accumulation in dangerously high concentrations of other drugs commonly prescribed to people with high blood pressure. It was the potential for such drug-drug interactions that caused Posicor to be withdrawn from the market in June 1998, only a year after its launching. The cost per new drug development along with the idea of “getting it right, quickly” [2] are aspects that the companies have to take into account in the development of a new drug. The cost per new drug- varies depending of the authors- could be from $500 million to more than $2,000 million [6,7]. An illustrative example was suffered by GlaxoWellcome and its Sumatriptan (migraine drug) which was launched into market in 1991. This drug was the pioneering medicine in migraine, but it had pharmacokinetic problems, as only 14% of dose of Sumatriptan was bioavailable. The drug had a short half-life and it was metabolized by monoamine oxidase (MAO), so there was risk of drug-drug interactions in people taking MAO inhibitors. Then, GlaxoWellcome started the development of a second generation of migraine drug getting a compound that showed right ADME properties and efficacy. All perfect except because during the optimization time other pharmaceutical companies launched their second-generations of migraine drug into market. REACH is the new European Community Regulation on chemicals and their safe use (EC 1907/2006). It deals with the Registration, Evaluation, Authorisation and Restriction of Chemical substances [3]. The new law entered into force on 1 June 2007. The aim of REACH is to improve the protection of human health and the environment through the better and earlier identification of the intrinsic properties of chemical substances. According to REACH regulation, toxicity and safety data for each chemical (more than 30,000) will need to be provided for registration. It is estimated that the testing of these existing chemicals will result in the use of at least 4 million of animals, therefore it is necessary the development and validation of alternative methods to be fully available in the near future [8]. In this sense, efforts to avoid the indiscriminate use of animals in scientific research have risen. The guiding principles that any researcher should follow is known as Three Rs. The three Rs are: Replace the use of animals with alternative techniques, Reduce the number of animal used to a minimum, gaining information from fewer animals or more information from the same number of animals, and Refine the way experiments are carried out, to make sure animals suffers as little as possible [9].
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
293
Therefore, early ADME studies would also minimize time-, cost-, and labor-intensiveness of screening and testing by looking at only the promising compounds. In Silico Methods On one hand, in silico prediction methods are gaining popularity in drug discovery processes as they are inexpensive and less time-consuming. On the other hand, reducing the use of animal experimentation represents a main goal of the European Union since the last decade, stimulating the development of new alternative methods. Thus, in silico methods can be envisaged as powerful alternative techniques to animal testing. Herein, we describe the current development in theoretical models to predict ADME properties. The physicochemical properties used, methods, and the relevance of predictive models in the evaluation of Human Intestinal Absorption (HIA), Blood-Brain Barrier (BBB) permeation and metabolism are discussed and finally, we briefly comment some aspects on excretion. 2. DESCRIPTORS Some of the most used molecular descriptors in the development of ADME models will be considered in this section. Polar Surface Area (PSA) The polar surface area (PSA) represents a very useful property for predicting different ADME properties as absorption. It is usually defined as those parts of the Van der Waals or solvent-accessible surface of a molecule that are associated with hydrogen bond-accepting and hydrogen bond-donating capability. Three types of PSA have been used in studying ADME properties: (i) Dynamic PSA (PSAd) [10], (ii) Static PSA (PSA) [11] and (iii) Two-dimensional or topological PSA (TPSA) [12]. An important amount of PSA models have been developed during the last years [13-15]. The first PSA model was developed by Palm et al. [10]. They demonstrated a sigmoidal relationship between PSAd and fraction absortion (FA) in human for 20 structurally diverse compounds. The major disadvantage of PSAd is that it is computationally expensive, which makes PSAd inappropriate for computational screening of large virtual libraries. Clark et al. [11] compared the results of PSA and PSAd and found that PSA is not very sensitive to the different conformation of small organic molecules. Using PSA in combination with modified Lipinsky rules they found that the FA is potentially low if PSA is greater than 140Å and at least two of the following criteria are met: MW greater than 550g mole-1; more than 12 hydrogen-bond acceptors, more than five hydrogen-bond donors and logP above 6.2. Other PSA approach is based on the 2D topology information. Ertl et al. [12] developed a method to generate topological PSA (TPSA) based on 3D PSA values for 43 fragments. They found that the correlation between PSA and TPSA is very high. Others authors as Clark [16] or Egan [17] employed PSA with logP as an additional descriptor. Lipophilicity Parameters Lipophilicity is a critical compound property that affects the suitability of compounds as potential drug candidates. The octanol/water partition coefficient is used to measure lipophilicity and it is expressed by the logP. It measures the distribution between a solute dissolved in an aqueous buffer (aqueous phase) and n-octanol as lipid
294 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
mimic (organic phase) [18-20]. It has been suggested, for example, that compounds with logP values between 0 and 3-4 are the most suitable candidates for passive transcellular absorption across intestinal epithelia [21]. As lipophilicity is increased (logP> 3-4), solubility and hence absorption progressively decline. In contrast, more hydrophilic compounds (logP< 0) are likely to traverse the epithelium more slowly via paracellular channels. Organic compounds that at different pH have different ionisable state show different logP values. The existence of more than one species yielding an average partition coefficient (e.g., ionisable compounds) leads to the apparent partition or distribution coefficient D, which is pH-dependent. Comparing the overall metabolism of numerous drugs clearly reveals a relation with lipophilicity. Several in vivo metabolic studies revealed a dependence of biotransformation on lipophilicity, suggesting a predominant role for transport in partitioning processes [22]. Abraham Descriptors Abraham and coworkers [23] developed a set of parameters to model solvatation and H-bonding properties of organic molecules. These descriptors are based on the physically meaningful theoretical cavity model of solute-solvent interactions and can be applied for prediction of a variety of physicochemical and pharmacokinetic properties. The analysis method uses the general linear free energy relationship LFER: SP = c + eE + sS + aA + bB + vV
Eq. 1
The dependent variable SP, or solute property, in Equation 1 refers to a property of a series of solutes in a given system. The independent variables in Equation 1 are solute descriptors. E is the solute excess molar refractivity, S is the solute polarity/polarizability, A and B are the overall or sum hydrogen bond acidity and basicity, and V is the McGowan characteristic volume. Volsurf Descriptors Volsurf descriptors developed by Cruciani and coworkers [24], take into account the interaction of the drug with the biological membranes by means of surface properties such as shape, Van der Waals forces, electrostatics, hydrogen bonding and hydrophobicity. They used the GRID forcefield to calculate energetically favourable interaction sites around the molecules producing 3D molecular interaction fields (MIFs) [24-26]. MolSurf Descriptors MolSurf descriptors [14,25,27] are derived from quantum mechanical calculations. They describe various electrostatic properties such as hydrogen-bonding strengths and polarizability, as well as Lewis base and acid strengths. It is possible to derive a statistically good and predictive model for intestinal absorption using the MolSurf parameters. Although PSA and the MolSurf parameters are highly correlated, it is believed that the MolSurfs descriptors give a more comprehensive characterization of the molecule and they are also easier to be interpreted than PSAd. Electrotopological State Index (E-State) The electrotopological state index (E-state) has been developed from chemical graph theory and uses the chemical graph for generation of atom-level structure indices. The
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
295
index is based on the electronic effect of each atom on the other atoms in the molecules as modified by molecular topology [28,29]. CODES Descriptors CODES© was created by Stud [30] and developed by our workgroup [31,32]. It is based on neural computing and enables the easy generation of the required numerical descriptors of the structures involved in a QSPR study. The main advantages of CODES are that it can avoid both the risky choice of physicochemical descriptors and problems associated to conformation. The problem associated with the structural diversity of many QSPR has been solved by CODES using the dimensions reduction methodology [31,32]. The implementation of this methodology allows any set of molecules (non-congeneric families) to be considered for study. This program codifies each molecule into a set of numerical parameters taking into account exclusively the information of its chemical structure using a not supervised neural network. The stereochemistry is also considered during the codification process and the R or S configuration of a stereogenic center is expressed by a corrective non-linear function. As we have mentioned at the beginning of this section, these descriptors are some of the parameters that it is possible to find in the literature [see reviews 15,33,34]. 3. IN SILICO APPROACHES The different computational approaches to predict ADME can be divided into three categories: the classical statistical methods, machine learning methods and structure bases (3D) molecular modelling [35]. The most important methods involved are briefly introduced in this section. 3.1. Statistical Methods Multiple linear regression (MLR). MLR [36,37] is the most used linear correlation method, which can model the relationship between two or more explanatory variables (X) and a response variable (Y) by fitting a linear equation to the observed data. Some important aspects are: 1.
As a general rule, the samples (N) should be larger than 2m (m is the number of descriptors used in correlation).
2.
Each variable should have equal chance to influence the outcome of the analysis. This can be to achieve by scaling the variables in a correct way.
3.
MLR assumes each variable to be exact and relevant.
4.
Variable must be independent. Strong collinear variables must be eliminated from the analysis.
Partial least square (PLS). PLS [38,39] is based on linear transformation from a large number of original descriptors into a new variable space based on small number latent variables (linear combinations of the original variables). Some differences regarding to MLR are: 1.
The descriptors are not treated as exact and relevant.
2.
Strong correlations between relevant variables are not a problem in PLS and all variables are used in the analysis.
296 Frontiers in Drug Design & Discovery, 2009, Vol. 4
3.
Campillo and Páez
The number of initial descriptors may exceed the number of compounds in the analysis.
3.2. Machine Learning Algorithm Machine learning methods utilize nonlinear learning methods to develop statistical models capable of predicting more diverse range of structures and physicochemical properties than those described by the available QSAR and QSPR models [40,41]. Artificial neural networks (ANNs). ANNs [42-44] represent, just opposite to PLS and MLR, a nonlinear statistical analysis technique. NNs are a class of machine leaning methods inspired by the way of biological nervous systems, such as the brain, of processing information. The two of the most prominent members in drug discovery are the fee-forward ANN with error back propagation (BPN) and the Kohonen self-organizing maps (SOMs) [45,46]. The main difference between both networks consists of the training process. In supervised training, both the inputs (X) (i.e., molecular descriptor) and the output (Y) (target property) are provided. The network then processes the inputs and compares its resulting outputs against the desired outputs. Errors are then propagated back through the system, causing the system to adjust the weights which control the network. In unsupervised training, the network is provided with inputs but not with desired outputs. The network itself must then decide what features it will use to group the input data. This is often referred to as self-organization. A net of artificial neurons is formed by several layers (input, hidden and out layers) with different number of neurons. The network will be trained to reproduce a target property for a set of molecules from input features (i.e., molecular descriptors). The number of hidden nodes in a NN must be identified through an optimization procedure with careful monitoring of the predictive behaviour of the final model. Some of the problems of a network can be overfitting by means of generation of chance correlations between the input variables and the output and over-training [45,46]. The risk of overfitting can be reduced by adjusting the number of neurons or by penalizing too complex networks during the training. Overtraining of NN can be avoided using a sufficiently large test set that is necessary to supervise the training of the ANNs models. Another way to solve these problems is using Bayesian regularized artificial neural networks (BRANNs). Instead of a single set of optimized network parameters, all possible parameter combinations are explored, and then predictions are computed as a weighted average over all possible networks. SOMs or Kohonen maps are trained in an unsupervised manner, i.e., the target property is not employed during the training process. This kind of network is applied to cluster and classify data sets. Genetic algorithms (GAs). Genetic algorithms [47-49] are a class of flexible and robust search and heuristic optimization techniques based on the biological principle of evolution through natural selection. A genetic algorithm simultaneously operates on a group of candidate solutions with the predicted property (Y) and a set of descriptors (X) that represent ‘chromosome’ for the population. The individuals are scored according to the fitness score. Good solutions are preferentially chosen from this group, their internal information is recombined and a small degree of variation is applied in a process of
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
297
algorithmic selection, cross-over and mutation. Cycling through this process will typically result in progressively better groups of solutions to the optimization problem. In principle, GAs can be combined with any correlation or classification approaches, such as MLR, PLS or ANNs [50]. Support vector machines (SVMs). SVM [51-53] are based on Vapnik´s structural risk minimization principle (SRM) from computational learning theory. SVM build several hyperplanes that separates the different classes of the dataset (Fig. 1). Among all hyperplanes (H1, H2…) SVM find the unique hyperplane having the maximum margin (m) that separates the two classes (this can be extended to multiclass problems), thus it is possible to minimize the expected error. In many cases SVM has been found to be consistently superior to other supervised learning methods and less prone to overfitting [54,55].
A
A
A
H1
A A
m A
A
H2
A A
B B B
B B
B
B
B B Fig. (1). SMV. Maximum separation between hyperplanes. Adapted by Ivanciuc [55].
Decision tree or rule based system. Decision trees (DT) create an iterative branching topology in which the branch taken at each intersection is determined by a rule related to a descriptor of the molecule. Finally, each terminating leaf of the tree is assigned to a class [56]. Some of the advantages of this method are that it is easy to understand and is an efficient training algorithm. Naïve bayesian classifier (NBC). The Bayesian approach for neural networks was pioneered in Buntine et al. [57] and MacKay et al. [58], and reviewed in Bishop [59], MacKay [60] and more recently in Lampinen [61]. Bayesian statistics is a method of categorization or classification base on the learn-by example. The Naïve Bayesian (NB) model is trained to distinguish interesting training set members such as highly active compounds of a data set of compounds. In contrast to other machine learning techniques, the Naïve Bayesian model is based on frequencies and probabilities of features set occurrences. Gaussian processes (GP). Gaussian Processes [62-64] is based on a Bayesian probabilistic approach. The method is suitable for modelling nonlinear relationships, does not require subjective determination of the model parameters, works for a large number of
298 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
descriptors, and is inherently resistant to overtraining. This method overcomes many of the problems of existing QSAR modelling techniques: 1.
Most importantly, it does not require the subjective a priori determination of parameters such as variable importance or network architectures.
2.
It is suitable for modelling nonlinear relationships.
3.
The method has a built-in tool to prevent overtraining and does not require crossvalidation.
4.
This technique has an inherent ability to select important descriptors.
The Gaussian Processes technique has been proven to compare well with and often exceed artificial neural networks (ANNs) in performance and has been shown to be equivalent to an ANN with a single hidden layer containing an infinite number of nodes A potential disadvantage of the Gaussian Processes technique is that it generates “black box” models, which are difficult to interpret. 3.3. Structure Bases (3D) Molecular Modelling These approaches are mostly applied to predicting metabolism [65-69]. Pharmacophore Model One possibility to derive a model for the active site of an enzyme is the creation of a pharmacophore model. With this technique, information on the active site is derived indirectly from the shape, electronic properties and conformation of substrates, inhibitors, or metabolic products. The construction of the pharmacophore model would not be possible with out taking that all substrates will be oriented in a similar way (both electronically and sterically) in the active site of the enzyme. Based on this assumptions it is possible to derive a template from all structures. For reviews of pharmacophore technique see Van Drie [70], and Hoffmann [71]. Enzyme-Ligand Model From docking studies of a specific ligand o a set of ligands into the appropriate enzymes structure their metabolism or inhibitory properties can be directly predicted. The aim of molecular docking is to evaluate the feasible binding geometry of a ligand with a target whose 3D structure is known [72,73]. To perform these studies the first step is the determination of the 3D structure of the target macromolecule, by experimental techniques as X-ray crystallography and NMR spectroscopy or by computational methods such as homology modelling [74]. The modelling process by homology, shown in Fig. (2), consists in the construction of a 3D model of the protein (target) using the structural information derivative of a homologue protein with structure 3D known (template). There are several methods to model a sequence of amino acids [75] (i) the “cut and paste” methods [76] comparative protein modelling by satisfaction of homology restrains [77,78]. Both methods are based on the assumptions that protein homologues (with >30% of identity) show similar structure. The most important requirement to built a homology model are (i) to have at least one crystal structure of a homologue protein to
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
299
the target and (ii) to perform a correct alignment between the target sequence and the template or templates with 3D known.
Homologues search
Target-template alignment
Model building
No
Analysis: Model Validation Yes Refine Model
Final Model Fig. (2). Flow chart for protein homology modelling.
4. ADME PROPERTIES Oral Absorption Oral administration is the most convenient and common way for patients to receive medication, therefore an important aspect in the process of drug development is to obtain a drug with a good oral bioavailability. Absorption can be defined as the movement of drugs across the outer mucosal membrane of the gastrointestinal (GI) tract, while bioavailability is defined as the capability of drugs to reach the general circulation or site of pharmacological actions. Oral drug absorption is a dynamic complex process comprising of several competitive components. Once the molecule is dissolved in the aqueous contents of the gastrointestinal tract, it will be transferred across different barriers from the intestinal lumen into the blood. These processes depend on a large number of factors, some of which are related to the properties of the drug itself, such as solubility and permeability, others to the formulation, such as the dissolution and release rates and to the physiology of the GI tract, such as gastric emptying and intestinal transit. The principal way for the drug permeability through the barrier is passive diffusion that is driven by a concentration gradient. However, some molecules, such as amino acid and glucose can be actively transported by specific transporters. The sets of drug-
300 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
transporter proteins such as P-gp, BCRP and MRP2 are efflux proteins in the apical membrane, MRP3 is an influx transporter on the basolateral side, and PEPT1 and ASBT are influx transporters on the apical side of human intestinal mucosal cells [33,79]. Therefore, intestinal drug permeability, together with aqueous solubility [33, 80, 81] are the most important factors that influence on drug absorption. In recent years, diverse in vitro and in situ permeability models have been developed to help predict oral drug absorption and bioavailability (Table 1). These include Caco-2 [82], Madin-Darby canine kidney (MDCK) [83,84] and 2/4/A1 cell culture models [85] , as well as the parallel artificial membrane permeability assay (PAMPA), immobilized artificial membrane (IAM) and physicochemical models [86-88]. Among the in vitro methods, the most popular cell-based model for intestinal permeability is the Caco-2 cell system. Some studies have shown that human oral drug absorption and Caco-2 permeability coefficient have a good sigmoidal relationship [89], suggesting that the human absorption can be well predicted by this in vitro model. However, it has several limitations, such as long preparation time, very slow absorption times compared to human intestine, large interlaboratory differences in quantitative results and its inability to predict quantitatively the level of active drug transport in vivo. Several in silico models have been developed to predict caco-2 in vitro data. In general, all models have a training set and mainly test set with a few number of compounds using some parameters included in section 3 [14,80,81,86,90]. Nevertheless, it is worth to mention the model of Hou et al. [91] that was built with a data set of 77 compounds and a test set of 23 compounds, using four descriptors: the experimental distribution coefficient (logD), the high-charged polar surface areas based on Gasteiger partial charges (HCPSA), the radius of gyration (rgyr) and the fraction of rotable bonds. As a less expensive alternative to Caco-2 methods, Kansy et al. [92] developed the parallel artificial membrane permeability assays (PAMPA) [93]. PAMPA is a non-cell based lipid membrane filter designed for the evaluation of passive transcellular permeability. Recently, a biomimetic artificial membrane permeation assay (BAMPA) has been introduced as an improved version of PAMPA. It utilizes a similar lipid composition to intestinal brush border membrane leading to significantly increased predictability of oral absorption [94,95]. Thus, passive oral drug absorption in humans can be predicted with in vitro permeability test as an alternative methodology to in silico models [90,93]. During the last years, starting with the articles of Lipinski and his “rule of 5”, proposed in 1997 [96,97], a considerable amount of research has been performed in order to develop in silico models for intestinal absorption in humans as well as other transport properties. There are numerous approaches based on computational models for bioavailability and human intestinal absorption, although the models proposed with a greater data set and test set correspond to the prediction of intestinal absorption. The intestinal absorption is measured by fraction absorption, which is defined by the total mass absorbed divided by the given dose of the drug. This simplified equation shows that the absorption is mainly dependent on the concentration at the intestinal wall, the permeability of the drug, and the given dose. Numerous models of HIA based on different descriptors have been gathered on different reviews [14,81,98-100]. As a consequence, there are predictive models that corre-
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
301
late oral absorption with fragment descriptors [101], hydrophobicity (logP) [17,102107], hydrogen bonding descriptors [102-104,108], topological indices [105,109-112], polar surface area [10,11,17,108,113,114] or quantum chemical parameters. With regards to the computational techniques, the most commonly mathematical models used in the generation of oral absorption models are Multiple Linear Regression (MLR) [10, 115] and Partial Least Square (PLS) [105, 116-119]. Furthermore, theoretical models for the prediction of oral absorption have been recently generated using Genetic Algorithms [14], Support Vector Machine (SVM) [120] and Artificial Neural Networks [110, 112]. The small size of the training set of most of these models limits the general applicability for prediction of other molecules. We will briefly comment the most interesting models on the basis of the techniques used for the building of HIA model. For example, Palm and coworkers [10] found that an excellent sigmoidal relationship could be established between humans fractional absorption values (%FA) and dynamic polar surface area (PSAd) (R2=0.94) for a set of 20 drugs. The dynamic polar surface area is a statistical average in which the surface area of each conformation is weighted by its probability to exist. Dynamic surface properties of each compound were calculated considering all low energy conformations within 2.5 kcal/mol of the global minimum based on Monte Carlo conformational search and energy minimizations. A different model was published by Zhao et al. [121] based on the Abraham descriptors to correlate the HIA data of 169 drugs. The model was obtained with a training set of 38 drugs and displays good correlation and external prediction ability for 131 compounds of the test set. The analysis of the model showed that there are mainly three descriptors that contribute to the absorption: the summation of solute hydrogen bond acidity, the summation of solute hydrogen and the volume term, being the first two the most important. Other model published to predict human intestinal absorption was developed by Klopman et al. [122] in which the parameters were structural descriptors identified by the CASE program, together with the number of hydrogen bond donors. The model developed with a data set of 417 molecules was able to correlate the percentage of absorbed drug of the training set with a square coefficient correlation (R2) of 0.79 and a standard deviation (SD) of 12.3%. The SD for an external test set for 50 drugs was 12.3%. Peres et al. [123] developed a HIA model from a data set of 82 compounds by using a topological substructural molecular design approach (TOPS-MODE). The drugs were divided into three classes according to reported cutoff values for HIA. "Poor" absorption was defined as HIA ≤ 30%, "high" absorption as HIA ≥ 80%, whereas "moderate" absorption was defined between these two values (30% < HIA < 79%). The best model developed with 82 compounds yielded a correct classification percentage of 89%. The external test set was carried out with 127 compounds and showed a correct classification percentage of 93%. Jones et al. [124] proposed a model for prediction of HIA of neutral molecules based upon surface charges of the molecule calculated by density functional theory (DFT). The model was built with a training set of 38 compounds and a data set of 107 drugs. The statistical values were similar to Zhao et al. method [121]. Finally, a paper appeared in the literature describing the application of support vector machine to HIA prediction [120]. A training set of 480 compounds and test set of 98
302 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
molecules was used for developing a new model. In the construction of classification model a set of 10 molecular descriptors were considered. The final model based on seven descriptors showed the importance of topological polar surface area (TPSA) and Table 1.
In Vitro, In Situ and In Vivo Systems for Permeability
System
Method
Comments
Advantages
Limitations
In vitro
Octanol-water partition coefficient
Lipophilic character (logP)
Simple procedure
Time-consuming Large amounts of solute
Liposomes
Lipid bilayer vesicles used as models for biological lipid bilayer membranes
Correlates better with human drug absorption than n-octanol-water partition coefficient
Time-consuming Non suitable for HTS
Immobilised artificial membrane (IAM)
Chromatographic model of lipidic membranes for studying passive absorption.
Automatisable system
Columns variability Instability interactions of the silica support
Suitable for HTS
Lacks active transport and also paracellular pathways
Useful for studying transcellular route
Extrapolation in vivo is difficult
PAMPA is a phospholArtificial Membrane ipid-based parallel Permeability Assay artificial membrane. (PAMPA) Permeability for passive absorption Membrane Vesicles.
Its preparation involves tissue homogenation Brush Border Memand differential sedibrane Vesicles (BBMV) and Baso- mentation, fractionation, and differential precipilateral membrane tation. vesicles (BLMV)
Most well established Less correlation with cell model that exHuman adenocarcinoma active carrier-mediated presses relevant carrierCell cultures. Caco 2 cell line expressing P-gp transport. Intermediated systems. laboratory variability. Suitable for HTS.
Isolate tissues
intestinal tissue
In situ
Single-pass perfusion animals
Rat small intestine
In vivo
Animals
Rat, dog
Complexity and characteristic closer to the in vivo situation
Less suitable for HTS
No suitable for HTS, Passive transcellular less accurate estimation permeability and active of performance of carrier-mediated system hydrophilic compounds
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
303
predicted apparent octanol-water distribution coefficient at pH 6.5 (logD 6.5). The best SVM classifier gave satisfactory predictions for the training set (97.8% for the poorabsorption class and 94.5% for the good-absorption class). Moreover, 100% of the poorabsorption class and 97.8% of the good-absorption class in the external test set could be correctly classified. Blood-Brain Barrier The blood-brain barrier (BBB), a highly regulated membranous barrier of brain capillaries, consists of an intricate network of tight junctions (TJs) that segregate the central nervous system (CNS) from systemic blood circulation whose purpose is to maintain homeostasis of the CNS by separating the brain from the systemic blood circulation. Therefore it is the most important biological barrier separating brain tissue from the peripheral circulation. This membranic structure acts primarily to protect the brain from chemicals in the blood, blocking the movement of all molecules except those that cross cell membranes by means of lipid solubility and those that are allowed in by specific transport systems. The extent to which drug molecules move from the blood into the brain is governed by two physiologically and anatomically related systems, BBB and the blood-cerebral spinal fluid (CSF) barrier, which forms two pathways through which drug compounds partition between plasma and brain tissue [125]. The ability of a drug to penetrate the blood–brain barrier is of fundamental importance in drug design. Blood-brain (BB) distribution of a molecule is a key characteristic for assessing the suitability of the molecule as a drug for the central nervous system. In the case of effective CNS acting drugs, the knowledge of their penetration through BBB is critical to screen potential therapeutic agent, while in the case of drugs with peripheral activity is necessary these knowledge to minimize CNS related side effects. The relative affinity for the blood or brain tissue can be expressed in terms of the blood-brain partition coefficient, usually expressed as logBB defined by the logarithm of the brain/blood concentration ratios log(Cbrain/Cblood) where Cbrain and Cblood are the equilibrium concentrations of the drug in the brain and the blood, respectively. The prediction of this property is important since the experimental determination (in vivo and in vitro) of BBB penetration is difficult and costly. In all animal models, the fraction of the drug transported is measured either by direct assay in the brain tissue or in blood samples studying the disappearance of the drug. There are several methodologies available (Table 1) for this purpose such as brain–blood partitioning, brain perfusion, the indicator dilution technique, brain uptake index, the capillary depletion technique, and intracerebral microdialysis [126]. Hence, there is a great demand for rapid and efficient methods capable of evaluating the penetration of drugs across the BBB. Within this context, the development of computational methods for predicting blood–brain partitioning and the creation and the development of software for prediction ADME properties constitute a challenge task of great scientific and economical value. In accordance with this relevance, a lot of in silico models for blood-brain barrier have been published in the last decade and many of these have been reviewed previously [14,99,127-131]. The more recent and significant prediction models, together with the methods used for their development and the predictive ability are gathered in Table 2. The data has been organized in chronological order. The statistical methods most often employed for developing logBB models are linear multivariate methods such as multiple linear regression (MLR) or partial least squares
304 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
(PLS), although other nonlinear methods have also been employed. The most commonly used nonlinear method in logBB modelling are neural networks (NN), genetic algorithm (GA) and support vector machine (SVM). There are a great number of models which correlate logBB from linear multivariate methods such as multiple linear regression (MLR) with very diverse parameters such as lipophilicity, H-bonding capability, solute size, and the ionization state. The initial models published were based mainly in descriptors that describe lipophilicity, H-bonding capability, and surface properties. A typical example is the model generated by Young et al. [132] in which a suitable correlation of logBB is established with octanol-water partition coefficient (logP) and the difference between logP in octanol/water and logP in cyclohexane/water systems (∆logP). In 1992, Van de Waterbeemd and Kansy [133] established a correlation of logBB with polar surface area (PSA) together with the molar volume (Vm) and non-polar surface area (NPSA). In 1996, Kelder et al. [134] found a good correlation with the dinamic polar Van der Waals surface area (dPSA). In 1999, Clark introduced a simple model from the polar surface area (PSA) and the clogP [16]. Considering that for PSA calculations it is necessary to generate a reasonable 3D molecular geometry and the calculation of the surface itself, a new approach for the calculation of the PSA was presented by Ertl et al. [118], based on the summation of tabulated surface contributions of polar fragments named topological PSA (TPSA), where the 3D structure does not need to be considered. So et al. introduced a new molecular descriptor named lipoaffinity (LA) to account for the effect of molecular hydrophobicity on BBB penetration [135]. Other singular approach was the proposed by Abraham et al. [136, 137]. They introduced the named Abraham descriptors that express the solvation model and the hydrogen bond donor/acceptor properties of the solute (see section 2). Thus, the logBB is defined according to descriptors such as the excess molar refraction, the polarizability/dipolarity, hydrogen bonding acidity and McGowan volume. Lombardo et al. [138] proposed a different approach that describes a solvation model to establish a correlation of logBB with the free energy of salvation (calculated using AMSOL methodology). A modified approach has been used later for Kesserü et al. using the generalised Born/surface continuum solvation model [139]. Through a rather different approach, Rose et al. [140] developed a model to predict logBB based on electrotopological state index. A modification using a similar strategy has been proposed by Cabrera et al. [141] in which a linear regression model was developed to predict the blood-brain partitioning coefficient from a data set of 119 compounds by using a topological substructural molecular design approach (TOPS-MODE). Pan et al. [142] described a new method that used 4D-molecular similarity measures to built BBB penetration model from cluster analysis applied to data set of 150 chemically diverse compounds. Some of the most recent and representative work in the area of BBB prediction will be summarised in the following lines on the basis of the techniques used for the establishment of the logBB models. With regard to MLR methods, it is worth to highlight the procedure described by Abraham et al. [143]. Their investigation was based on the combination of the different type of BBB data in the construction of a model. In particular, they checked that the in vivo rat data (blood, plasma, or serum) is possible to combine. However, they found that it is not able to combine in vivo and in vitro data on volatile
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
305
organic and inorganic compounds, because there is a systematic difference between the two sets of data. To obtain the BBB model they employed the general linear free energy relationship (LFER) (see section 2) using the Abraham descriptors together with two indicator variables Ic (indicator for carboxylic acids Ic = 1) and Iv (Iv=1 for the in vitro data and Iv=0 for the in vivo data) to indicate the different type of logBB data (in vivo and in vitro). The final equation yielded a R2 of 0.75 and a SD of 0.33 log units. The data set and the test set used by Abraham are largest applied to date. The resulting model built with 164 molecules and 7 descriptors showed a R2 of 0.73 and a SD of 0.34. This model was used to predict an external test set of 138 compounds with a SD of 0.25 log units. A new strategy has been proposed by Katrizky et al. [144] where the experimental blood-brain partition coefficient (logBB) for a diverse set of 113 drug molecules were correlated with computed structural descriptors using CODESSA-PRO and ISIDA programs. The best model corresponded to the linear correlation CODESSA-PRO of fivedescriptor (ClogP, Kier flexibility index, number of double bonds, H-donors CPSA and maximum partial charge for all atom types). This model yielded a correlation coefficient of R2 of 0.78 and a SD of 0.123. The models were successfully validated using the CNS active data of an external test set of 19 drug molecules with correlation coefficient R2 of 0.766 and a SD of 0.032 for the CODESSA model. Wichmann et al. [145] developed a model based on a molecular set of 103 neutral molecules using the set of 5 COSMO-RS sigma-moments as descriptors obtained from quantum chemical calculations using the continuum solvation model COSMO. The best model with only three descriptors yielded a R2 of 0.71 and a RMS error of 0.40 log units, although no external test set was used to validate the prediction model ability. Konovalov et al. have described two models [146,147], first of them was developed form a data set of 328 blood-brain distribution (logBB) values, based on the linear freeenergy relationship (LFER) descriptors. The obtained results indicated that the LFERbased k-NN-MLR model was the most accurate predictive logBB model [146]. In the second, a new variable selection wrapper method named the Monte Carlo variable selection (MCVS) method was developed. Thus, starting from 1500 molecular descriptors, the authors found that only TPSA(NO) was relevant for the BBB [147]. These methods have been implemented in the QSAR-BENCH program, which is freely available from www.dmitrykonovalov.org for academic use. Recently Bultink et al. [148] have published a new model for the prediction of logBB, based on a molecular set of 82 diverse structures. The descriptors are derived from quantum chemical ab initio calculations, and classical 2D descriptors. The quantum chemistry methods were used to determine the geometry and properties of the molecules corresponding to the minimum energy conformation. A model with eight-parameters could reproduce the logBB data, based on the information of 2D classical descriptors as the number of Cl atoms (nCl), the number of the fusion of 6-and 7-membered rings (nR11) in the molecules, clogP (theoretically determined value for the octanol–water partition coefficient), TPSA(NH) descriptor (topological polar surface area, where each N-containing fragment in the molecule contributes to the total value) and descriptors based on theoretical quantum chemical calculated at the B3LYP/631G* level as the maximum Mulliken charge-derived descriptor on the carbon atom and on the fluorine, the maximal separation in Mulliken charges on the hydrogens in the molecule and the
306 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
dipole moment (µ). The best model built with 70 compounds yielded a R2 of 0.80 and a SD of 0.36. Regarding the prediction of the test set (only 12 compounds) displayed a R2 of 0.76. A recent article of Obrezanova [149] has been published about an automatic model generation process for building blood-brain barrier penetration models using Gaussian Processes, a machine learning modelling method. The user is only required to provide initial information: a set of molecules and a set of associated property values to be modelled. From this starting point, the process applies a number of different mathematical techniques to generate a series of models before selecting the best one. The work set (151 structures) was split randomly into training (108 structures) and internal evaluation (43 structures) sets. A set of 157 SMARTS-based 2D descriptors including calculated logP, descriptors relating to molecular size and shape plus those related to hydrogen bonding was calculated and the number of descriptors was then reduced to 64 using the unsupervised forward selection (UFS) program. The Gaussian Processes technique with nested sampling (GP-Nest) produced the best model for this data set (R2= 0.79, RMSE= 0.32). The predictions for the internal validation set (23 compounds) displayed a R2 value of 0.72 and RMSE of 0.38 log units and the external test set (22 compounds) yielded a R2 value of 0.66 and RMSE of 0.49 log units. The performance of this automatic model generation process was comparable to manual model building. Other statistical methods as partial least squares (PLS) have been applied in the prediction of logBB. Norinder et al. [116] used MolSurf parameterisation to calculate various properties related to the molecular valence region applying PLS to develop a QSAR model. Luco [150] has presented a model using topological and constitutional descriptors using PLS statistics. The analysis of PLS revealed that the first three components had influenced on BBB model being related with the polarity, the molecular size and the molecular shape, respectively. The PLS technique was also used by Crivori et al. [151] to correlate transformed 3D molecule fields descriptors and experimental permeation values. Another study of Norinder has been published [105] using theoretical molecular descriptors related to electrotopological state indices. Additional parameters related to size and lipophilicity (calculated molar refraction (CMR) and octanol-water partition coefficient (CLOGP)) were also used in the statistical modelling. A different approach was introduced by Cruciani et al. [152] based on molecular fields computed by GRID to calculate the descriptors set. These VolSurf descriptors are referred to molecular size and shape hydrophobic and hydrophilic regions, hydrogen bonding, amphiphilic moments and critical packing. Zhao et al. [153] have been publish a new model for predictions of BBB penetration from 19 simple molecular descriptors calculated from Algorithm Builder and fragmentation schemes. Several BBB models based on hydrogen-bonding properties, such as Abraham descriptors, polar surface area (PSA), and number of hydrogen bonding donors and acceptors have been built using binomial-PLS analysing, using a 1093 compound in the training set and tested on a 500 compound set. The finals model developed with 1–5 simple descriptors presented test set accuracies for classifications on the range 96.5– 99.8% for BBB+ molecules and 65.3–79.6% for BBB- molecules. The results showed that the overall classification accuracy for the training set is over 90%, and overall prediction accuracy for the test set is over 95%.
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
307
The use of SVM has gained importance in the last year. A paper has been reported about the application of support vector machine (SVM) to logBB prediction with results favourable when is compared with neural network model [154]. A data set consisting of 179 CNS active molecules and 145 CNS inactive molecules were studied with two different machine-learning algorithms to predict the BBB permeability of different classes of molecules. The first algorithm is based on a multilayer perceptron neural network and the second algorithm uses a support vector machine. The parameters considered include molecular weight, lipophilicity, hydrogen bonding, and other variables that govern the ability of a molecule to diffuse through a membrane. The results displayed that the SVM predict up to 96% of the molecules correctly, averaging 81.5% over 30 test sets, which comprised of equal numbers of CNS positive and negative molecules. This is quite favourable when compared with the neural network's average performance of 75.7% with the same 30 test sets Li et al. [155] analysed the results yielding by different methods including logistic regression, linear discriminate analysis, k nearest neighbour, C4.5 decision tree, probabilistic neural network, and support vector machine. Molecular descriptors were selected by using a feature selection method, recursive feature elimination (RFE). A data set of 415 compounds including 276 BBB+ and 139 BBB- was used for building a new model. Regarding to the descriptors, the authors selected 35 descriptors from a data set of 199 molecular descriptors calculated from the feature selection method (RFE). The molecular parameters include 2 descriptors of molecular connectivity and shape class, 18 descriptors on the class of electrotopological state, 7 descriptors of quantum chemical properties class, and 8 descriptors of geometrical properties class. Of the six statistical learning methods tested, SVM appears to give slightly higher prediction accuracy than other methods with 88.6% of accuracy for BBB+ and 75.0% for BBB-. During the last years the use of artificial neural networks (ANNs) to develop logBB model has increased, reflecting in the number of publication. Fu [156] has proposed an ANN model to predict logBB of drugs from their molecular structural parameters obtained from quantum chemical calculations. These molecular structural parameters were the molecular volume (V), the sum of the absolute values of the net atomic charges of oxygen and nitrogen atoms which are hydrogen-bond acceptors (QO, N), and the sum of the net atomic charges of hydrogen atoms attached to oxygen or nitrogen atoms (QH). The final model developed with a training set of 56 compounds displayed a root mean squared errors (RMSE) of 0.24. The external validation was performed only using 5 compounds and yielded a RMSE of 0.26. A Bayesian neural network was used to establish the relationship between a 85compound logBB data set and a set of computed molecular property descriptors including counts of hydrogen-bond acceptors and donors, hydrophobes, rotable bonds, logP, MW and PSA [157]. The model with the best statistics (R2= 0.81, SD = 0.37) was obtained using four nodes in the hidden layer. The test set of 21-compound shown a q2 (cross-validated R2) value of 0.65 (SD = 0.54). The most important descriptor was logP, closely followed by the count of rotable bonds and PSA where the hydrogen-bond donors had a greater influence than hydrogen-bond acceptors on brain permeation. A new kind of nonlinear method, general regression neural network (GRNN), has been explored for blood-brain barrier penetration [158] using DRAGON Web (Version 3.0), a total of 1497 1D, 2D, and 3D molecular descriptors were computed. The BBB
308 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
penetration model was building with 159 compounds, with a training set of 129 compounds and a test set of 30 compounds. A seven-descriptor subset was selected by the descriptor selection algorithm as the optimum set for GRNN model of BBB penetration. The prediction capability of GRNN-developed models was compared to those developed using MLR and a nonlinear multilayer feed-forward neural network (MLFN) method. The model with the best statistics for the test set was obtained using general regression neural network (GRNN) (R2= 0.701, MSE= 0.13). Garg [159] has developed a ANN model using molecular structural parameters to predict logBB. Seven calculated descriptors including P-gp Substrate Probability, Molecular Weight, No. of H-Bond Acceptors and Donors, No. of Rotatable Bonds, Topological Polar Surface Area and ClogP were used for model building. The neural network has been constructed with a training set of 132 compounds, after removing 9 outliers. The model (R2= 0.82, s = 0.30) was obtained using 4-layered 7-5-2-1 architecture. The test set of 50 compound shown a R2 value of 0.80 (SD = 0.32). Hemmateenejad et al. [160] developed a back-propagation three-layered artificial neural network using an error-learning algorithm to process the nonlinear relationship between the quantum chemical calculated descriptors and logBB data. Different electronic descriptors were calculated for each molecule from the optimum 3D geometry obtained by ab initio calculations at the level of RHF/STO-3G and logP as a measure of hydrophobicity and different topological indices were also calculated. Thus, a diverse data set of 123 chemicals was chosen in this study and genetic algorithm (GA) was used as a feature selection method to select the most relevant set of descriptors as the input of the network. Modelling of the logBB data by the only quantum descriptors produced a 5:4:1 ANN structure with RMS error of validation and crossvalidation equal to 0.224 and 0.227, respectively. When the logP and the principal components of the topological indices to electronic descriptors were included better nonlinear model (RMS(V) and RMS(CV) equals to 0.097 and 0.099, respectively) was obtained. An external test set was used to predict the logBB of 23 molecules that did not have contribution in the steps of model development. The best model produced RMS error of prediction 0.14, and could predict about 98% of variances in the logBB data. Recently, other neural network model of logBB prediction has been reported [31]. A set of 108 compounds of wide structural diversity was chosen in this study. Thus the molecules are described from a not supervised neural network using a new methodology, the CODES program. This program codifies each molecule into a set of numerical parameters taking into account exclusively its 2D chemical structure. In order to compress the dynamic matrix data to a set of three numeric codes for each molecule reduction of dimension (RD) using TSAR© program [30] which applies Monte Carlo algorithm, was carried out by a back-propagation neural network with architecture (AxR)-c-y-c-(AxR), where (AxR) represents CODES matrix, c is the number of neurons in codification layer and y is the number of hidden neurons. The original data set of 108 compounds was divided into two different series (training set and external prediction set). Using a supervised artificial neural network with different training was developed a set of models of 3-layered 3-b-1 and 4-b-1 architectures, where the input layer (3 and 4) are the encoding variables for each structure (hidden neurons values in reduction of dimension process) and the output value is the experimental logBB value for each compound. The best predictive model was obtained using a 3-layered 3-5-1 architecture with 30 compounds. The model presented a R2 value of 0.79 (SD = 0.43) yielding 83% of accuracy in the training
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
309
set. Regarding the capability of prediction of the model, the prediction of the 78 compounds of the external test averages 73% of accuracy in the prediction of logBB. This model presents one of the biggest test set which makes it a reliable predictive model. About descriptor calculation, we used an easy way to code the molecules from the graphical structure without properties calculation or molecular optimization. Table 2.
Selected Models for the Prediction of Blood-Brain Partitioning External Test Set
Training Set Descriptors
Ref.
Method
n
m
SE
R2
n
Prediction
Abraham, LFER
MLR
164
7
0.35
0.71
138
SE = 0.25
Abraham et al.[143]
18 classical
NN
132
7-5-2-1
0.30
0.81
50
R2 = 0.80, SE = 0.32
Garg et al. [159]
21 Quantum Chemical, 6 Chemical (LogP etc.), 257 topological indexes
ANN
123
5:4:1
RMSE = 0.22
-
23
RMS = 0.14
Hemmateen ejad et al. [160]
0.35 0.22
0.78
19
R2 = 0.77, SE = 0.18
0.87
19
457 Calculated
MLR
113 112
CODESAPro:5 ISIDA:5
19 Classical
PLS
1093
5
-
Quantum Chemical COSMO-RS
MLR
103
3
RMSE = 0.40
0.71
Abraham, LFER
MLR
291
7
0.30
E-DRAGON (1500)
MLR
289
MCVS: 1 (TPSANO)
157 SMARTS based 2D
GP-nest
108
Quantum Chemical and 2D Classical
MLR
CODES
NN
95 % + 500 82 % -
R2 = 0.83, SE = 0.16
Katritzky et al. [144]
98% (+) 80% (-)
Zhao et al. [153]
-
-
Wichmann et al. [145]
0.75
-
-
Konovalov et al. [146]
0.43
0.49
-
-
Konovalov et al. [147]
UFS: 64
RMSE = 0.32
0.79
22
70
8
0.36
0.80
12
q2ext = 0.75; R2 = 0.76
Bultinck et al. [148]
30
3-5-1
0.43
0.79
78
73%
Campillo et al. [31]
R2 = 0.66; Obrezanova RMSE = 0.49 et al. [149]
310 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
SE: standard error. m: number descriptors used. q2ext: external explained variance. n: number of molecules. GP-nest: Gaussian Processes Technique with nested sampling variable selection. MCVS: Monte Carlo Variable Selection. UFS: unsupervised forward selection. RMSE: root mean square error.
The majority of in silico models use logBB as the index of BBB permeability. However, this parameter has been severely criticised, because the brain concentration used is the sum of the bound and free drug concentrations [161] being possible that, due to extensive binding to brain tissue, a compound with a high logBB value may in fact has a lower free concentration in the brain than a compound with a lower logBB value. Therefore, it has been propose a new parameter named permeability–surface area PS as alternative permeation model. logPS is measured using a short-duration vascular perfusion method from which a permeability surface area product is calculated being a measure of the rate of transfer of the compound from the blood to the brain and therefore logPS is a true permeability measure. Some works published in this field will comment very briefly in the following lines. Few studies using logPS have been published, and the proposed models have been established with a reduced number of experimental data. The models of Levin et al. [162], Abraham et al. [163,164], Liu et al. [165] have been established with data set between 18 and 30 compounds. We have to mention the model of Lanevsky [166] that has been published recently. This model is given by a system of non linear equation from a data set of 125 compounds and using five descriptors: logP, Hbond donors, H-bond aceptors, Vx and ion fractions (pKa function). The best model presents a R2=0.78 and SD=0.12. The ISIDA model gave R2=0.82 and RMSE=0.48 to remove 1 outlier. The models were successfully validated using an external test set of 53 molecules yielding a RMSE=0.49. Metabolism The metabolic degradation of drugs is one of the key determinants of drug clearance and accounts most frequently for different pharmacokinetic profiles [167]. Compounds begin to be process as soon as they enter to the body and the initial compound is converted to new compounds called metabolites. Metabolites may also be pharmacologically active, sometimes more so than the parent drug. Traditionally, a distinction is made between phase I (functional reactions) and phase II metabolism (conjugative reactions). In phase I metabolism, a molecule is functionalized through oxidation, reduction or hydrolysis. In phase II metabolism, the functionalized molecule is further transformed through of a group of reactions known as conjugation reactions, as glucuronidation and sulfation. The most important enzymes that are involved in these phase are the cytochrome P450 (CYP) a heme monooxygenase, and UDP-glucuronosyltranferse (UGT). CYPs and UGTs are responsible for the elimination of more than 90% of drugs cleared by the liver [168-173]. Cytochrome P450 is the principal drug metabolizing system of enzymes for phase I metabolism. Of the 57 human CYPs, only a few of them have been shown to play a significant role in hepatic drug metabolism. In particular the CYP3A4, CYP2D6, CYP1A2, CYP2C9, CYP2C19 and CYP2E1 isoenzymes are responsible for more than 90% of the metabolism of all pharmaceuticals in current clinical use. Some of these isoforms (CYP2D6 and CYP2C9) display polymorphisms which can result in the poor metabolism of drugs. Inhibition of P450 enzymes is undesirable because of the risk of severe side effects due to drug-drug interactions. CYP3A4 and CYP2D6 are responsible for the metabolism of a huge variety of drugs. Inhibition of these enzymes by one drug might lead to a decreased clearance of another drug when two or more drugs are co-
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
311
administered. Such unexpected drug-drug interactions can potentially have fatal consequences for the patients, and new chemical entities (NCEs) should be investigated for CYP inhibition as early as possible in drug research [174]. An overview of the different types of substrates for CYPs together with their inducers and inhibitors are shown in Table 3. Table 3.
Mammalian Cytochromes P450 (Adapted by Lewis [175])
CYP
Substrate Classes
Typical Substrates
Inhibitors
Inducers
1A2
Planar poly(hetero)-aromatic amines and amides
Caffeine
Furafylline
TCDD, PAHs
2A
Small to medium mwt ketones
Coumarin
Metyrapone
Poorly induced
2B
Non-planar lipophilic molecules
Phenobarbital
Secobarbital
Phenobarbital
2C
Non-Planar molecules with hydrogen bond potential
Tolbutamide
Sulphaphenazole
Poorly induced
2D
Generally polar molecules with basic protonable nitrogen
Debrisoquine
Quinidine
Non-inducible
2E
Small mwt compounds of diverse structure
p-Nitrophenol
Disulfiram
Ethanol
3A
Large mwt compunds of diverse structure
Erithromycin
Gestodene
Sinthetic steroids
TCDD: 2,3,7,8-tetrachlorodibenzo-p-dioxin, PAHs: Planar polyaromatic hydrocarbons.
CYP3A4 is the most abundant hepatic CYP isoform responsible for the metabolism of almost 50% of known drugs. Inhibition of CYP3A4 by co-administered drugs has been shown to cause adverse clinical drug-drug interactions [176]. Therefore the identification of inhibitors of CYP 3A4 is very beneficial in drug development in order to minimize the risk associate with clinically relevant drug-drug interactions. CYP2D6 is estimated to be involved in the metabolism of approximately 30% of the drugs currently on the market [177]. It is absent in 5-9% of the Caucasian population, resulting in diminished metabolism of numerous drugs. When CYP2D6 is inhibited by one compound, the metabolism of another compound will decrease which the subsequent and unexpected drug-drug interactions. This is due to accumulation of the latter compound as it is not being metabolized. Therefore, inhibition of CYP2D6 is an unwanted feature in a drug candidate [178]. Phase II metabolism has been less pursued compared with that of phase I enzymes. UGT (UDP-dependent glucuronosyl treanferase) is the principal drug metabolizing system of enzymes for phase II metabolism. Glucoronidation of small lipophilic molecules by UGTs is probably the most important phase II process for the clearance of drugs. To date, 18 UGT proteins have been identified which can be classified into two families, UGT1 and UGT2. The most important UGTs are UGT1A1, UGT1A4, UGT1A6, UGT1A7 UGT1A9, and UGT2B4. Predictive models to evaluate the UGT metabolism are less advanced relative to CYP [41].
312 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
It is important to mention that the metabolism in animals might be different form that in humans [179]. Several in vitro test systems (hepatocytes, microsomes or recombinant proteins) are available to determine metabolism parameters as stability and inhibition (Table 4) [180,181]. Table 4.
In Vitro Systems for Metabolism and Excretion
Method
Comments
Advantages
Limitations
Microsomes
Small vesicles from fragmented hepatocyte. Contain Phase I metabolism
Relatively inexpensive technique Contains important ratelimiting enzimes
Contains only phase I. Requiere strictly specific substrates and inhibitors or antibodies for individual drug metabolising enzyme
Hepatocyte
Liver
Contain the whole complement of drugmetabolisin enzyme (DME).
Relatively difficult to obtain. Require specific techniques and well-established procedures. The levels of many DMEs
Liver Slices
Liver tissue. Contain Phase I and II metabolism.
Contain the whole DME.
Require specific techniques and well-established procedures
DNA microarrays
Rat Liver. Phase I and II metabolism
Can be utilised with HTS
Problems in extrapolation
Chemicals with poor drug metabolism properties in humans are unfavorable for medicinal use, typically presenting problems with bioavailability, half life, interindividual variability, and drug-drug interactions. Therefore it is very important to screen chemicals for these properties as early as possible in the development process. The approaches to predict metabolism can be divided into QSAR and 3D modelling. The last one includes techniques such as development of pharmacophore and modelling of proteins (see section 3). The metabolism in silico models can be classified regarding the approach prediction: 1) take into account the used prediction method or 2) according to the complex enzymedrug. The first classification yields two categories: (i) Mathematical methods (statistical techniques and learning machine methods) and (ii) 3D-Modelling (Table 5). Regarding Table 5.
Overview of Models Prediction
Approach QSAR
Structure based model
Property Linear methods
Non lineal methods
MLR, PLS
ANNs, SOM, DT, SVM, kNN, NBC
ligand-based
protein-based
Inhibition or substrate CYPs protein-ligand interaction based
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
Pharmacophore 3D-QSAR Pharmacophore Model, QM/MM
Homology modelling, Crystallographic Protein Models
Docking DM
313
Ki, Km, CSP, mutagenicity, IC50, HA
CSP: catalytic site prediction, HA: Heteroactivacion activity.
the second classification, the models prediction can be classified into three approaches; (i) those that are ligand-based, (ii) those that are protein-based and (iii) those that are protein-ligand interaction based [67]. Regarding to the metabolic endpoints, the metabolism of P450 substrates can be analyzed using kinetics parameters as IC50, Ki (to study drug-drug interaction) and from their Michaelis-Menten kinetics [182] on the basis of intrinsic clearance; this is determined by the ratio of catalytic rate, kcat, or maximal velocity, Vmax, to apparent Michaelis constant (Km) as follows: Clint = kcat/Km
Eq. 2
The clearance concept will discus in the next section. Several reviews [15,41,99,173,177] have been published about the different approaches and models to predict metabolism activity. An overview of the approaches and models is given in Table 6 and 7. Table 6 displays the different models based on QSAR approaches meanwhile Table 7 gathers the structure based models. The majority of prediction models are base on Phase I metabolism, CYPs. Phase II metabolism (UGT) in silico methods are less development relative to CYP. For a revision of UGT in silico methods see Smith et al. [183], Chohan et al. [173], Radominska et al. [184] and Sorich et al. [185]. Table 6.
In Silico Models of Metabolic
Property/ Enzyme
Training Set
Test Set
MOE, atom and bond features
236
60
BRANN, PLS, MLR, RP
Topological, geometrical, electronic despcriptor
109
249
-
96%
[187]
CYP3A4, CYP2C9 CYP2D6 inhibition and substrates
SVM, PLS, MLR, kNN
Dragon, electropological state indices, Abraham descripotr
602
100
-
94-97%
[188]
CYP2D6 inhibition
DT
2D structural descriptor
100
51
75%
80%
[189]
CYP2D6 inhibition
BPN, NBC
E-state keys, Barnard fingerprints, functional class fingerprints, AlogP, MW, Hbond donors and acceptors
1810
600
76%
99%
[190]
CYP3A4 inhibition
BPN
Unity fingerprints
218
72+9
-
90.3
[191]
Method
Descriptors
CYP1A2 inhibition
NPA-SVM
CYP1A2 inhibition
Training Results
Test Results
Ref.
[186]
314 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
CYP3A4 inhibition
SVM, kNN, ridge regression
MOE2D, Ghose-Crippen, electronic, topological, count and structural descriptor
410
85
-
67.1
[192]
CYP3A4 inhibition
DT, NBC, SVM, kNN
Barnard fingerprints, MACSS fingerprints
4000
470
-
81.9
[193]
Property/ Enzyme
Method
Descriptor
Training Set
Test Set
Training Results
Test Results
Ref.
CYP2D6 inhibition/ CYP3A4 inhibition
SVM, PLS,DA
2D structural, MOE2D, Volsurf, QM descriptors
712/807
475/538
-
72.0/75.0 (71..0)
[194, 195]
CYPs inhibition
SOM(Kohone n)
Lipophilicity, charge distribution, topological features, steric and surface parameters (60 descriptor)
491
33 reversible competitive 15 competitive inhibitors
CYP3A4: 91% (high Km) 97% (lowKm)
94% as low Km 87% as low Km
[196]
Route of CYPmediated metabolism
DT
Physico-chemical descriptors
96
51
94%
CYP3A4 inhibition
BPN, PLS-DA
Cats, Ghose-Crippen, topological electronic, structural descriptor
311
50/10
-
90/90
[198]
CYP3A4 inhibition
PLS
Size, shape, lipophilicity, counts, surface areas, MOE2D, Volsurf
511
379
-
66%
[199]
CYP3A4, 2D6 inhibition
k-NN
Occurrence and frequency of ring fragments and functional groups
1037/ 865
345/288
87% /83%
84% /82%
[200]
CYP1A2,2C9, 2C19,2D6,2E, 3A4 inhibition
DT
Topological, electrotopolicial, hydrophobiciy, electronic, hydrogen bond, molecular inonization
161
One-tenth of original data
12% misclassification
15.7% misclassification
[201]
CYP2C9 inhibition
LWRP, NERP, Gravity method, SUBDUE method
Global structure descriptor
276
50/11
-
CYP2C9, 2D6, 2A4 inhibition
SVM, DT
Shape, polarizability, charge
600
100
-
84-90%
[203]
CYP2C9, 2D6, 2A4 regioselectivity (metabolic site)
DT
2D and 3D descriptor
92-316
9-19
72-77% of molecules where top two atoms contains
67-80% of molecules where top two atoms contains
[204]
(Table 6) contd....
a
[197]
Consensus 3 [202] methods: 90%
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4 oxidation site
315
oxidation site
a
LWRP: Line Walking Recursive Partioning; NERP: Normal Equation Recursive Partitioning; Gravity Method is a “nearestneighbour” type of algorithm; SUBDUE Method: The data is represented as a graph and the predictor for a class is represented as a set ob subgraphs.
Table 7.
Overview of CYP Pharmacophore and 3D-QSAR Models [66]
Method
Traininga
Testb
Predicted Property
Key Features of Pharmacophore
Ref.
CYP1A2 Combine and GRID/GOLPE + homology model
12
na
Inhibitor model includes H-bonding and hydrophoMutageni- bic binding sites (on the protein). Several residues in [215] city the homology model are found to interact with the pharmacophore model.
Substrates and inhibitors have a negative electrostatic potential close to a lactone moiety, steric effects around a methoxy group on methoxasalen, [216] and positive electrostatic potential para to a methoxy group.
CYP2A5
CoMFA
16
na
IC50
PLS MS-WHIM
16
na
Ki
Potent inhibitors have a positive molecular electrostatic potential and a H-bond acceptor.
[217]
23
5 (ss)
IC50
Potent CYP2A6 inhibitors do not include a lactone moiety.
[218]
16
na (ss)
Km, CSP
Substrate model includes at least three hydrophobic regions 3.1, 4.6, and 5.3 Å from a H-bond acceptor.
[219]
Substrate model includes two hydrophobic regions and one H-bond acceptor. Substrate catalytic site is 4.0 and 3.4 Å from two hydrophobic regions and 4.6 [220[ Å from a H-bond acceptor in model A and is 4.9, 4.1, and 7.8 Å, in model B.
CYP2A6 CoMFA and GRID/GOLPE CYP2B6 Catalyst and PLS MSWHIM
Catalyst + homology model
16
5 (4)
Km, CSP
na
na
CSP
Combined protein-based inverse substrate model includes two hydrophobic and two electropositive binding sites (on the protein).
[221]
7
na
CSP
Substrate model with H-bond acceptor and a hydrophobic and a cationic region.
[222]
manual superposition
8
na
CSP
Substrate model protein H-bond donor is 7 Å from substrate catalytic site.
[223]
manual superposition
20
na
CSP
Substrate model includes anionic site is 7.8 Å from
[224,
CYP2C8/9/18/19 GRID/CPCA + homology model + docking CYP2C8 Catalyst + homology model + docking CYP2C9
316 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez catalytic site, between hydrophobic regions.
225]
[226, 227]
Ref.
27
14 (13)
Ki, CSP
Inhibitor model includes two cationic binding sites, along with an aromatic binding region and a steric region (on the protein). Substrates possess a partial negative charge at 10 Å and an anionic site at 6 Å from the catalytic site.
Traininga
Testb
Predicted Property
Key Features of Pharmacophore
Catalyst and PLSWHIM
9 29
14 (10) 14 (12)
Ki Ki
GRID/GOLPE + homology model + docking
21
8 (8)
Ki
GRID/GOLPE/ALMO ND + homology model + docking
42 21
39 (29) 12 (11)
Ki Ki
Alignment-independent inhibitor model includes two H-bond donors at 6 Å from each other, a hydrophobic region at 13 Å from a H-bond donor, and two H- [230] bond acceptors at 12 and 15 Å from two H-bond donors (on the protein).
manual superposition + homology model
27
4 (4)
CSP
Substrate model includes a H-bond acceptor site and an aromatic region and suggests Arg108 for electrostatic and Phe476 for aromatic interactions.
[231]
Catalyst
36
na (ss) (22-39)
HA
Heteroactivator model includes two hydrophobic regions at 11.6-11.8 Å from a H-bond acceptor and at 10.3-10.9 Å from an aromatic ring and a H-bond acceptor 4.8 Å from an aromatic ring.
[232]
GRID/GOLPE/ALMO ND + homology model
43
na
CSP
Alignment-independent protein-based inverse model represents substrates as a set of fingerprints for each [233] H-atom present.
GRID/GOLPE/ALMO ND + homology model + docking
22
12 (11)
Ki
Alignment- and conformer-independent inhibitor model includes two H-bond donors and three hydrophobic binding sites (on the protein).
[234]
manual superposition
15
na
CSP
Substrates possess a basic nitrogen atom at 5 or 7 Å from site of catalysis and coplanar aromatic rings.
[235237]
manual superposition
156
na
CSP
Carboxylate group in protein is responsible for the 5 or 7 Å distance between basic nitrogen and site of catalysis.
[238]
manual superposition + homology model
10
na
CSP
Asp301 is identified as residue binding to basic nitrogen atom in ligands.
[239]
manual superposition + homology model + molecular orbital calc
40 + 14
7 (6)
CSP
Substrates have a basic nitrogen atom at 5, 7, or 10 Å from site of catalysis. Separate models are presented for O-demethylation and N-dealkylation.
[240, 241]
CoMFA + homology model
(Table 7) contd….
Method
CYP2C9 Inhibitor models include at least one hydrophobic and one H-bond acceptor at 3-5.8 Å from each other. [228] H-bond acceptor and H-bond donor/acceptor are 3.45.7 Å apart. Inhibitor model is in agreement with site-directed mutagenesis data indicating important roles of Leu102, Val113, Phe114, and Leu362.
[229 ]
CYP2D6
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
317
Glu216, Asp301, and Phe481 are identified as interaction sites.
manual superposition
6
Inhibitors contain a tertiary nitrogen atom (protonated at physiological pH), a flat hydrophobic region, and two regions in which functional groups with lone pairs are allowed.
na
[242]
(Table 7) contd….
Method
Traininga
Testb
Predicted Property
20 31
15 (9) 15 (10)
Ki Ki
Key Features of Pharmacophore
Ref.
CYP2D6 Catalyst
Inhibitor models include a H-bond acceptor and a H[243] bond donor and two to three hydrophobic regions. Substrate model includes one cationic site at 5.5 Å from a H-bond acceptor and two hydrophobic regions respectively at 8.7 and 8.3 Å from the Hbond acceptor and at 19 o "counterclockwise" from the cationic site and 28 o from the first hydrophobic region. Asp301 and Phe481 are involved in electrostatic and aromatic interactions, respectively.
Catalyst and PLS MSWHIM + homology model
24 52
28 (ss) na
Km
CoMFA
24
15 (ss)
Km
12
na
Clintr
Substrate model includes a long hydrophobic access channel.
19
na
CSP
Substrates possess a H-bond acceptor 5.5-7.8 Å from catalytic site and 3 Å from the heme-associated [247] oxygen. Inhibitor model includes three hydrophobic regions at 5.2-8.8 Å from a H-bond acceptor, three hydrophobic regions at 4.2-7.1 Å from a H-bond acceptor, and 5.2 Å from another H-bond acceptor or one hydrophobic regions at 8.1-16.3 Å from the two furthest of three H-bond acceptors.
[244]
Substrates possess a positive electrostatic site 5, 7, or [245] 10 Å away from the site of catatalysis.
CYP2E1 CoMFA
[246]
CYP3A4
manual superposition
Catalyst
14 32
8 (7) 8 (7)
Ki Ki
[248]
Catalyst
38
12 (12)
Km, CSP
Substrate model includes two H-bond acceptors, one [249] H-bond donor, and one hydrophobic region.
16
na
IC50
Inhibitor model includes two aromatic rings, one Hbond acceptor, and one or three hydrophobic regions [250] for CYP3A4 or CYP3A5 and 3A7, respectively.
8
na
IC50
Inhibitor model includes one or more ligand-protein H-bonds and an orthogonal angle with the heme
CYP3A4/A5/A7
Catalyst
CYP17 (17α) manual superposition + GRID + homology
[251]
318 Frontiers in Drug Design & Discovery, 2009, Vol. 4 model
Campillo and Páez plane.
7
na
IC50
Inhibitor model includes three hydrophobic regions and one or two H-bond acceptors and is used to extract known inhibitors/substrates from a large database.
[252]
Traininga
Testb
Predicted Property
Key Features of Pharmacophore
Ref.
manual superposition
11 + 13
ss
IC50
Inhibitors include one H-bond acceptor and hydrophobic groups
[253]
CoMFA and GOLPE
50
ss
IC50
CoMFA
49
8 (7)
IC50
Multiple binding orientations are for different classes of nonsteroidal inhibitors
[255]
CoMFA
40
15
IC90
Imidazoles are more active than triazole inhibitors
[256]
Apex-3D and CoMFA
18
4 (4)
IC
N3 and N4 atoms in azoles, ethereal oxygen, and aromatic ring are essential for azole inhibitors
[257]
CoMFA and Catalyst
58
na (ss)
IC50
Inhibitor models include three hydrophobic regions and one H-bond acceptor
[258]
Catalyst/HipHop
(Table 7) contd….
Method
CYP19 (arom)
Steroidal inhibitors include two hydrophobic pockets in the C6 region, a large pocket in the α-face, and a [254] smaller pocket at the β-6 position
CYP51 (14α)
a Size of data set of ligands with experimentally determined parameters with respect to (binding) affinity (Ki, IC50, and Km values), heteroactivation activity (HA), or catalytic site prediction (CSP) used to construct the model. Multiple models in the same study are on separate lines, or a training set can be added to a previous model indicated by "+".b Size of data set of ligands with experimentally determined parameters with respect to (binding) affinity (Ki, IC50, and Km values), heteroactivation activity (HA), or catalytic site prediction (CSP) used to test the model. The number of successes in the test set is given in parentheses; "ss" indicates that the results were statistically significant but no success rate was given. When no test set is used, this is referred to as "na" (not applied).
QSAR Approaches Different in silico models for the classification of CYP3D6, CYP3A4 and CYP2C9 inhibition have been published. For a revision of the different models see Kriegl et al. [41]. Table 6 shows some of the most representative prediction models. In general the CYP2D6 and CYP3A4 are the most studies system. The CYP2D6 models are based on 100-1810 diverse compounds and the test set accuracies varying from 80-99%. Regarding CYP3A4, the models are based on 218-4000 diverse compounds, and the accuracies cover a field from 66-94% (Table 6). Yap et al. [205] published a comparative study between different statistical learning methods for predicting CYP3A4 inhibition. Table 8 displays the classification accuracies of the various classification models such as LR, LDA, PLS, DT, k-NN, SVM and PNN. The inhibitor accuracies predictions are in the range of 25-80% while non-inhibitor accuracies prediction are in a better range of 49-99% (Table 8). Regarding to the statistical
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
319
methods goodness, the SVN methods yield the better overall accuracy and using this method it is possible to invert the negative results about the inhibitor prediction obtained with the other methods. De Groot and co-workers [12] demonstrated that is more convenient the use of multiple predictive methods. They used a combination of two neural networks and one Bayesian model to increase the predictive capacity for 2D6 inhibition. The using of two comTable 8.
Results from the Comparative Study Inhibitors
Non Inhibitors
Overall
Method TP
FN
Pp (%)
TN
FP
Pn (%)
Accuracy (%)
SVM
16
4
80.0
70
4
94.6
91.5
LR
8
12
40.0
65
9
87.8
77.7
LDA
5
15
25.0
36
38
48.6
43.6
PLS
8
12
40.0
67
7
90.5
79.8
DT
9
11
45.0
65
9
87.8
78.7
k-NN
6
14
30.0
73
1
98.6
84.0
PNN
12
8
69.0
65
9
87.8
81.9
TP: True Positive; FN: False Negative; Pp: Positive Accuracy; Pn: Positive Accuracy.
combined models increases the correct classification of the model. Other metabolism models are based on the regioselectivity [204,206-210]. CYPs enzyme catalyze a number of chemical changes that all probably involve the transfer of the oxygen radical from the heme iron of the enzyme to the molecule as one of the steps. To know where a molecule would be preferentially oxidized would help to block the metabolism and make the drug candidates more stable in vivo [206]. The most of regioselectivity methods use·3D knowledge [206-209]. Two QSAR-based regioselectivity models for CYP3A4, 2D6 and 2C9 have been published [204,210]. They estimate the AM1 dehydrogentation energy (the energy necessary to remove a hydrogen radical from a particular atom) based on the local chemical environment of the atoms (see Table 6). They found that the most predictive model uses three substructure descriptors and two physical property descriptors and it is able to explain about 70% of the regioselectivity. The authors admit that up to now the available models include their models have not all critical and necessary information to explain more accuracy the regioselectivity. Structure Based Models Two different approaches can be used to derive the structure based models, based on a set of known ligands or inhibitors or based on the interaction enzyme-ligand. The first approach can be divided into two broad categories, namely, quantum mechanical (QM) and pharmacophores models. The second one based on the complex enzyme-ligand in-
320 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
cludes the known 3D of the enzyme by X-Ray or by homology modelling and a set of known ligands of a specific enzyme. Ligand Based Models A pharmacophore was first defined by P. Ehrlich in 1909 as a molecular framework that support the minimal and essential features responsible of biological activity of a drug [211]. Later, P. Gund introduced the receptor term updating Elrlich´s definition by "a set of structural features in a molecule that is recognized at a receptor site and is responsible for that molecule's biological activity" [212]. The IUPAC definition of a pharmacophore is "an ensemble of steric and electronic features that is necessary to ensure the optimal supramolecular interactions with a specific biological target and to trigger (or block) its biological response" [213]. A pharmacophore does not represent a real molecule or a real association of functional groups, but a purely abstract concept that accounts for the common molecular interaction capacities of a group of compounds towards their target structure. Pharmacophore models are constructed by superimposing ligands onto each other by matching chemically similar groups and derive the minimal common features of all ligands without knowledge of the active site. This approach can be used to perform catalytic site prediction and qualitative predictions with respect to CYP activity and binding affinity (Ks, Ki, IC50, Km) [66]. The main assumptions for this approach are that all ligands share the same binding site and binding mode in the active site and that the minimum energy conformation of the ligands corresponds to the biologically active conformation. This approach that could be namely that classical pharmacophore model does not take into account the interaction with the active site. An update version of this approach is 3D-QSAR that uses information derived of the interaction with the active site (3D-QSAR), this approach use spatial atomic descriptors as molecular interaction fields (MIFs) [214], functional groups, electronic properties and shape to describe a set of structure into a 3D space that mimic the active site. With this information it is generated a matrix that is analyzed using multivariate analysis like principal component analysis (PCA) to extract and condense the information contained in the matrix to explain the chemical properties of the structure set. [66]. The constructed 3DQSAR model can be used to perform quantitative predictions of properties for molecules (Kd, Ki, IC50, Km) that were no used to build the QSAR model and to predict sites of catalysis of CYP substrates. The basic pharmacophore of the CYP2D6 substrates consist of basic nitrogen that is protonated at physiological pH and site of metabolism at 5 or 7 Å distances from the nitrogen. Regarding to CYP3A4, the pharmacophore model trained with 38 compounds contained two hydrogen bond acceptors, one hydrogen bond donor and one hydrophobic region [177]. An excellent overview of the different models obtained with pharamacophore and 3D-QSAR approaches was published by de Graff et al. (Table 7, reproduce with the permission of the authors) [66]. The other approach into ligand-based models is Quantum Mechanical (QM) calculations. In CYP enzymes, QM calculations can yield very useful information on activation barriers for substrates or heme-oxygen species during the catalytic cycle and on relative energies between species and relevant geometries. Quantum Mechanical approaches have been used to predict hydrogen abstraction potentials and likely sites of metabolism of drug molecules [259,260]. AM1, Fukui functions and DFT calculations could identify potential sites of metabolism [99].
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
321
Complex Based Models The 3D-structure of a protein-ligand complex can provide valuable insight into its function. Knowledge of 3D structure of the enzyme together molecular docking methods have been successfully used in the study of structure function of P450s [66]. Molecular docking studies are an important method to predict protein-ligand interactions. Docking methods predict energetically favourable conformations and orientations of ligands in the binding site of a protein. Several docking algorithms and scoring functions have been described in the past few years, the most commonly used for CYPS are GOLD, FlexX, DOCK, AutoDock, and the scoring functions form C-Score (Tripos Inc., St. Louis). More deeply review about docking techniques by Muegge et al. [72,261]. In order to perform computational protein-ligand docking experiments, a 3D structure of the target protein at atomic resolution must be available. The most reliable sources are crystals and solution structures provided by the Protein Data Bank (PDB) [262]. As mentioned above, ideally experimental technique such as X-ray crystallographic or NMR are used to determine the 3D structure of proteins. Unfortunately, it is not possible to determine all proteins using these experimental techniques by different reason as difficulty in crystallize process, insufficiently soluble or too large for NMR studies [67]. In such case homology modelling is the choice method for determine the 3D structure of proteins. An overview of the solved structures is summarized in Table 9. The information has been obtained from SCOP data base [263]. These enzymes belong to the class “All α proteins” and show a multihelical fold. The most of them have been crystallized in complex with ligand. In several cases, different protein conformations have been found in crystal structures for the same isoform. Until beginning of the 2000, structural models of human CYPs were based on the known bacterial P450s however this protein is distantly related with CYPs. The determination of the crystal structure of different rabbit [264269] and human CYPs [270-274] (see Table 9) in free form or in complex with inhibitors have improved the reliability of comparative models for human CYPs. Homology modelling is based on the assumption that the proteins with similar sequences might have analogous 3D structures. The modelling process by homology consists in several steps (Fig. 2) essential for obtaining a correct sequence alignment of the target sequences with the homologous (template) used as basic structure [77,78]. From the best alignment, 3D models can be constructed using different methods as MODELLER [77], SWISS-MODEL [275]. For reviews of different homology models and docking studies see Laak et al. [76], Maréchal et al. [67], Yamashita et al. [68] and de Graff et al. [66]. The general protocol followed by different authors [276-283] is schematized in Fig. (3). If the 3D structure of the enzyme is not known them it is modelled in order to perform docking studies with different aims.
322 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
Prediction of IC50. (Yu et al.) Catalytic differences (Prasad et al.) Target Sequence
Multiple alignment with templates
Validation and Refinement
Model Building
Docking procedure Prediction of regioselectivity (Lafite et al., Oh et al.)
Homology modelling Active site architecture and identification of key residues (Li et al.)
Catalytic site prediction and virtual screning ( Graff et al., Locuson et al.)
Fig. (3). Flow chart for docking studies. Table 9.
Overview of all Currently Available CYP Crystal Structures (Data from SCOP Data Base)
CYP Isoform
Species
No. of pdb
101-CAM
Pseudomonas putida
61
102-BM-3
Bacillus megaterium
23
55A1-NOR
Fungus (Fusarium oxysporum)
18
107A1-ERYF
Saccharopolyspora erythraea
8
EpoK
Sorangium cellulosum
3
108-TERp
Pseudomonas sp.
1
51(14α )
Mycobacterium tuberculosis
7
152A1
Bacillus subtilis
1
OxyB
Amycolatopsis orientalis
3
implicated in an oxidative phenol coupling reaction during vancomycin biosynthesis
OxyC
Amycolatopsis orientalis
1
implicated in an oxidative C-C coupling reaction during vancomycin biosynthesis
121-Mt2
Mycobacterium tuberculosis
14
154a1 monooxygenase
Streptomyces coelicolor
1
154c1 monooxygenase
Streptomyces coelicolor
1
154c1 monooxygenase
Streptomyces coelicolor
1
175a1
Thermus thermophilus
2
thermostable P450
119
Archaeon Sulfolobus solfataricus
5
thermophilic P450
Archaeon Sulfolobus tokodaii
1
Rabbit (Oryctolagus cuniculus)
3
2C5
Comments
functionalizes macrolide ring systems functionalizes macrolide ring systems
PDB codes:1dt6, 1n6b, 1nr6
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
323
2B4
Rabbit (Oryctolagus cuniculus)
3
PDB codes: 2bdm, 1suo, 2q6n
2C8
Human (Homo sapiens)
5
PDB codes: 1pq2, 2nni, 2nnj, 2nnn, 2rn0
2C9
Human (Homo sapiens)
3
PDB codes: 1r9o, 1og2, 1oq5
3A4
Human (Homo sapiens)
6
PDB codes:1tqn, 2j0d, 2v0m, 1w0f, 1w0f, 1w0e
158A2
Streptomyces coelicolor
1
Excretion The kidney and the liver play a major role in drug elimination by excretion and metabolism respectively [284]. The excretion process can either be passive or active via energy dependent processes. Prediction methods to describe the excretion kinetics quantitatively are at an early development stage. The ability to predict the biliary and intestinal clearance (CLbile and CLi) of drugs in therefore valuable in the design and selection processes of candidate drugs [285]. The clearance is a measurement of the renal excretion ability. Although clearance may also involve other organs than the kidney, it is almost synonymous with renal clearance or renal plasma clearance. Clearance based on measurements of the disappearance of parent compound. Clearance refers to the volume of in vitro medium or blood from which the substance is cleared, and so it is inversely proportional to area under the concentration-time curve [286]. 5. PERSPECTIVES AND CONCLUSIONS Compounds display different kinetic behaviour in function to their physicochemical and structural features and therefore these molecules reach higher or lower concentrations in certain organs and tissues. Absorption of compounds, distribution between blood and tissues and the passage of special barriers are the special relevance. Computational methodology is widely used to development of new lead compound providing valuable information for drug discovery. Thus, in silico methods can play a major role in decreasing time to market, reducing animal experiments and cost associated with discovery of new drugs. In this context, different strategies for ADME prediction have been developed using experimental data obtained from in vitro or in vivo experiments. In this work some aspects of the most crucial parts of the kinetic behaviour have been revised. The most relevant in silico methods for the prediction of HIA, BBB and metabolism have received extra attention. In general, the current generation of models exhibit similar statistical quality, even when it is derived using quite different descriptors and computational techniques. In relation blood-brain barrier, there are numerous methods that can fit and predict log BB values to around 0.3 and 0.4 log units, respectively. These values although are elevated are within the experimental error. A broad range of protocols for determining the blood–brain uptake is available. Thus, there are multiple types of administration, time points of measuring, test animals and techniques for plasma separation and brain homogenizing. To develop a predictive
324 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Campillo and Páez
model for BBB, all molecules in the set should ideally be characterized experimentally using exactly the same procedure. Therefore, the developed model should ideally be based on experimental data generated by a standardized test protocol. It may thus be expected that in the coming years, a new generation of BBB permeation models will emerge that is based on logPS, rather than logBB data. The models based on HIA have been developed from the same type of descriptors used in the study of other properties. However, the response parameters used generally show large variability depending on the methodology used to determine the fraction absorbed in humans and the inter-individual variability, and hence the accuracy of the model obtained is heavily influenced and therefore these models should be interpreted with caution. Although the biological complexity of in vivo absorption processes complicates the application of in vitro models, the development of techniques with greater capacity of prediction is necessary. In relation to descriptors, in principle, parameters based on 3D-descriptors are superior to lower-dimensional descriptors because they capture important information about physicochemical properties such as internal hydrogen bonds. The 3D descriptors may also be easier to interpret than some of the other mentioned variables. However, the choice of the correct 3D conformation may, cause problems in some cases. 1D and 2D descriptors are generally much faster to compute than the corresponding 3D-based descriptors. Moreover, the possible problems associated with generating a reasonable 3D conformation for the investigated structure are eliminated. The most interesting approaches are those using variables based on 2D-structures, since avoid the optimization of tridimensional structures and the study of the conformational domain. Regarding to metabolism, the aim of this paper has been collect the different modelling approaches to understand metabolism process focus on the knowledge and prediction of CYPS. The in silico approaches development during the last years in CYPs modelling cover a broad range of techniques as QSAR approaches, pharmacophore modelling of the ligands, enzyme modelling and docking studies to predict the ligand binding and metabolite formation. This variation of techniques is reflected on the different of reliability of the results obtained. It is quite sure that to obtain reliable results is necessary the combination of various approaches [12,205]. Some models often explore a linear relationship between the pharmacokinetic property of interest and the structural and physicochemical properties of the studied compounds, which are not applicable to those agents with nonlinear relationships. Hence, statistical methods capable of modelling nonlinear relationships need to be developed. In principle, good predictive models for pharmacokinetics properties depend crucially on selecting the correct descriptors and a sufficiently large set of experimental data. A model that is externally predictive should also be robust, although a robust model is not necessarily predictive. A high value of the “leave many out” cross-validated correlation coefficient can be regarded as necessary, but insufficient condition for a model displays a high predictive ability. The prediction capability of a regression model estimated by comparing the predicted and observed values of a sufficiently large and
In Silico ADME Approaches
Frontiers in Drug Design & Discovery, 2009, Vol. 4
325
representative external test set of compounds that were not used in the model development it is essential for the development of ADME models. Of course, the in silico prediction models are not simply of academic interest, they should play a very important role in development of new drug. Although the computational (in silico) approach for the prediction of ADME is very attractive, its accuracy is still not satisfactory. Thus, high throughput experimental and automatic computational methods for the determination of ADME properties continue being of great importance in drug development.
ABBREVIATIONS ADME
=
Absorption, Distribution, Metabolism and Elimination or Excretion
PK
=
Pharmacokinetic
QSAR
=
Quantitative Structure Activity Relationships
QSPR
=
Quantitative Structure Property Relationships
REACH
=
Registration, Evaluation, Authorisation and Restriction of Chemical Substances
CNS
=
Central Nervous System
OA
=
Oral Absorption
BBB
=
Blood-Brain Barrier
HIA
=
Human Intestinal absorption
FA
=
Fraction absorption
GI
=
Gastrointestinal
2D
=
Two-dimensional
3D
=
Tree-dimensional
MLR
=
Multiple Linear Regression
PLS
=
Partial Least Square
ANN
=
Artificial Neural Network
SVM
=
Support Vector Machine
DT
=
Decision Tress
k-NN
=
k-Nearest neighbors
GP
=
Gaussian Process
GA
=
Genetic Algorithm
2
R
=
Square coefficient correlation
SD
=
Standard Deviation
CYP
=
Cytochromes
326 Frontiers in Drug Design & Discovery, 2009, Vol. 4
UGT
=
Campillo and Páez
UDP-glucoronosyltransferases
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41]
Van de Waterbeemd, H.; Gifford, E. Nature, 2003, 2, 192-204. Hogdon J. Nature, 2001, 19, 722-26. REACH: http://ec.europa.eu/environment/chemicals/reach/reach_intro.htm Bains, W.; Gilbert, R.; Sviridenko, L.; Gascon, J.M.; Scoffin, R.; Birchall, K.; Harvey, I.; Caldwell, J. Curr. Opin. Drug Discov. Devel., 2002, 5, 44–51. SoRelle, R. Circulation, 1998, 98, 831-2. DiMasi, J. A.; Hansen, R.W.; Grabowski, H. G. J. Health Econ., 2003, 22, 151-85. Christopher, P.; Brantner, V. V. Health Aff., 2006, 25, 420-8. Clemedson, C.; Kolman, A.; Forsby, A. Altern. Lab. Anim., 2007, 35, 33-8. Mohan, C.C.; Gandhi, T.; Carg, D.; Shinde R. Mini Rev. Med. Chem., 2007, 7, 499-507. Palm, K.; Stenberg, P.; Luthman, K.; Artursson, P. Pharm. Res., 1997, 14, 568-71. Clark D.E. J. Pharm. Sci., 1999, 88, 807-14. Ert, P.; Rohde, B.; Selzer, P. J. Med. Chem., 2000, 43, 3714-7. Van de Waterbeemd, H. In Pharmacokinetic Optimization in Drug Research, Testa, B.; Van de Waterbeemd, H.; Folkers, G.; Guy, R. Eds. Wiley-VCH, 2001. Hou, T.; Wang, J.; Zhang, W.; Wang, W.; Xu X. Curr. Med. Chem., 2006, 13, 2653-67. Ruiz-Garcia, A.; Bermejo, M; Moss, A.; Casabo, V.G. J. Pharm. Sci., 2008, 97, 654-90. Clark, D.E. J. Pharm. Sci., 1999, 88, 815-21. Egan, W.J.; Merz, K.M.; Baldwin, J.J. J. Med. Chem., 2000, 43, 3867-77. Merino, V.; Freixas, J.; del Val Bermejo, M.; Carrigues, T.M.; Moreno, J.; Pla-delfina, J.M. J. Pharm. Sci., 1995, 84, 777-82. Bermejo, M.; Avdeff, A.; Ruiz, A.; Nalda, R.; Ruell, J. A.; Tsinman, O.; Gonzalex, I.; Fernandez, C.; Sanchez, G.; Garrigues, T.M.; Merino, V. Eur. J. Pharm. Sci., 2004, 21, 429-41. Van de Waterbeemd, H.; Smith, D.A.; Beaumont, K.; Walder, D.K. J. Med. Chem., 2001, 44, 1313-33. Smith, D.A. Eur. J. Drug Metab. Pharmacokinet., 1994, 3, 193-9. Smith, D.A.; Brown, K.; Neale, M.G. Drug Metab. Rev., 1985, 16, 365. Abraham, M.H.; Chadha, H.S.; Martins, F; Mitchell, R.C.; Bradbury, M.W.; Gratton, J.A. Pestic. Sci., 1999, 55, 78. Cruciani, C.; Crivori, P.; Carrupt, P.A.; Testa, B. J. Mol. Struct. Theochem., 2000, 503, 17-30. Cruciani, G.; Meniconi, M.; Carosati, E.; Zamora, I.; Mannhold. In Drug bioavailability, Van de Waterbeemd, H.; Lennernas, H.; Artursson, P. (Eds.). Wiley-VCH, 2003. Cruciani, G.; Clementi, S.; Crivori, P.; Carrupt, P-A.; Testa, B. In Pharmacokinetic Optimization in Drug Research, Testa, B.; Van de Waterbeemd, H.; Folkers, G.; Guy, R. Eds. Wiley-VCH, 2001. Sjöberg, P. In Computer-Assisted Lead Finding and Optimization, van de Waterbeemd, H., Testa, B., Folkers, G., Eds; Verlag Helvetica Chimica Acta: Basel, MolSurf version 2.0, Qemist AB, Hertig Carls allé 29, S-691 41 Karlskoga, Sweden. 1997, 81-92. Kier, L.B; Hall, L.H. Pharm. Res., 1990, 7, 801-7. Kier L. B. “Chemometric Methods in Molecular Design (Methods and Principles in Medicinal Chemistry)”, by Waterbeemd H. V.(Ed) .VCH Verlagsgesellschaft mbh, Weinheim, 1995, 39-44. M. Stud CODES©; v1.0 (revision 3) ed. Guerra, A.; Páez, J.A.; Campillo, N.E. QSAR Comb. Sci., 2008, 27, 586-94. Dorronsoro, I.; Chana, A.; Abasolo, M.I.; Castro, A.; Gil, C.; Stud, M.; Martinez, A. QSAR Comb. Sci., 2004, 23, 89-98. Van de Waterbeemd, H.; Lennernas, H.; Artursson, P. Drug bioavailability, Estimation of Solubility, Permeability, Absorption and Bioavailability. V.18. Wiley-VCH, 2003. Testa, B.; Van de Waterbeemd, H.; Folkers, G.; Guy, R. Pharmacokinetic Optimization in Drug Research, Wiley-VCH, 2001. Wishart, D.S. Drug RD, 2007, 8, 349-62. Draper, N.R.; Smith, H. Applied Regression Analysis, Wiley, New York, 1998. Freund R.J.; Wilson, W.J. Regression Analysis. Academic Press, USA, 1998. Linderg, W.; Persson, J.A.; Wold, S. Anal. Chem., 1983, 55, 643. Geladi, P.; Kowalshi, B.R. Anal. Chim. Acta, 1986, 185, 1. Li, H.; Yap, C.W.; Ung, C.Y.; Xue, Y.; Li, Z.R.; Han, L.Y.; Lin, H.H.; Chen, Y.Z. J. Pharm. Sci., 2007, 96, 2838-60. Fox, T.; Kriegl, J.M. Curr. Top. Med. Chem., 2006, 6, 1579-91.
In Silico ADME Approaches [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84]
Frontiers in Drug Design & Discovery, 2009, Vol. 4
327
Devillers. J. Neural Networks in QSAR and Drug Design. Academic Press, USA, 1997. Livingstone, D.J.; Manallack, D.T.; Tetko I.V. J. Comput. Aided Mol. Des., 1997, 11, 135-42. Livingstone, D.J; Manallack, D.T. J. Med. Chem., 1993, 26, 1295-7. Saarinen, J.; Kohonen, T. Perception, 1985, 14, 711-9. Kohonen, T. Self-Organizing Maps, 3rd ed., Spinger, Berlin, 2001. Holland, J.H. Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology Control and Artificial Intelligence. Ann Arbor, MI, USA: University of Michigan Press. 1975. Devillers. J. Genetic Algorithms in Molecular Modelling. Academic Press, USA, 1996. Rogers, D.R.; Hopfinger, A.J. J.Chem. Inf. Comput. Sci., 1994, 34, 854. Bayram, E.; Santago, P.; Harris, R.; Xiao, Y.D.; Clauset A.J.; Schmitt, J. D. J. Comput. Aided Mol. Des., 2004, 18 483-93. Vapnik, V.; Chapelle, O. Neural Comput., 2000, 12, 2013-36. Cortes, C.; Vapnik, V. Mach. Learn., 1995, 20, 273-97. Byvatov, E.; Scheneider, G. Appl. Bioinformatics, 2003, 2, 67-77. Yang, R. Brief. Bioinform., 2004, 5, 328-38. Ivanciuc. O. In Reviews in Computacional Chemistry. Lipkowitz, K.B.; Cundari, T.R (Eds). WileyVCH, Weinheim, 2007, 291-400. Quinlan, J. R, Mach. Learn., 1986, 1, 81-106. Buntine, W.L.; Weigend, A.S. Complex Syst., 1991, 5, 603-43. MacKay, D.J.C. Neural Comput., 1992, 4, 448-72. Bishop, C.M. Neural networks for pattern recognition, Oxford University Press, 1995. MacKay, D.J.C. Network Comp. Neural, 1995, 6, 469-505. Lampinen, J.; Vehtari, A. Neural Networks, 2001, 14, 257-74. Williams, C.; Rasmussen C. in Proc. of the Conf. on Neural Information Processing Systems (NIPS). MIT Press, 1995, pp. 514-20. Seeger M. Int. J. Neural Syst., 2004, 14, 69-106. Obrezanova, O.; Csánzy, G.; Gola, J.M.R.; Segall, M.D. J. Chem. Inf. Model., 2007, 47, 1847-57. Ekins, S.; de Groot, M.J.; Jones, J.P. Drug Metab. Dispos., 2001, 29, 936-44. de Graaf, C.; Vermeulen, N.P.E.; Feenstra, K.A. J. Med. Chem., 2005, 48, 2725-55. Maréchal, J.D.; Sutcliffe, M.J. Curr. Top. Med. Chem., 2006, 6, 1619-26. Yammashita, F.; Hashida, M. Drug Metab. Pharmacokinet., 2004, 19, 327-38. de Groot M. Drug Discov. Today, 2006, 11, 601-6. Van Drie, J.H. Computational Medicinal Chemistry for Drug Discovery, Bultinck, P.; de Winter, H.; Langenaeker, W.; Tollenaere, J.P (Eds.). Marcel Dekker, Inc. New York. 2004, 437-460. Hoffmann, R.D.; Meddeb, S.; Langer, T. Computational Medicinal Chemistry for Drug Discovery, Bultinck, P.; de Winter, H.; Langenaeker, W.; Tollenaere, J.P (Eds.). Marcel Dekker, Inc. New York. 2004, 461-482. Muegge, I.; Enyedy I. In Computational Medicinal Chemistry for Drug Discovery. Bultinck, P.; de Winter, H.; Langenaeker, W.; Tollenaere, J.P (Eds.). Marcel Dekker, Inc. New York. 2004, 405-436, Krumrine, J.; Raubacher, F.; Brooijmans, N.; Kuntz, I. Structural Bioinformatics, Bourne, P.E.; Weissig, H. Wiley-Liss, USA, 2003, 443-476. Bourne, P.E. Weissig Structural Bioinformatics, Bourne, P.E.; Weissig, H. Wiley-Liss, USA, 2003. Wallner, B.; Elofsson, A. Protein Sci., 2005, 14, 1315-27. ter Laak, A.M.; Vermeulen, N.P.E. Pharmacokinetic Optimization in Drug Research, Testa, B.; Van de Waterbeemd, H.; Folkers, G.; Guy, R. Eds. Wiley-VCH, 2001, 551-588. Sali, A.; Blundell, T. L. J. Mol. Biol., 1993, 234, 779-815. Burke, D.; Deane, C.; Nagarajaram, H.; Campillo, N.; Martin-Martinez, M.; Mendes, J.; Molina, F. Perry, J.; Reddy, B.; Soares, C.; Steward, R.; Williams, M.; Carrondo, M.; Blundell, T.; Mizuguchi, K. Proteins, 1999, 3, 55-60. Zhang, D.; Zhu, M.; Humpherys, W.G. Drug Metabolism in Drug Design and Development, WileyVCH, 2007. Johnson, S. R.; Zheng, W. AAPS J., 2006, 8, 27-40. Norinder, U.; Bergstrom, C. A. Chem. Med. Chem., 2006, 1, 920-37. Hidalgo, I. J.; Raub, T. J.; Borchardt, R. T. Gastroenterology, 1989, 96, 736-49. Irvine, J. D.; Takahashi, L.; Lockhart, K.; Cheong, J.; Tolan, J. W.; Selick, H. E.; Grove, J. R. J. Pharm. Sci., 1999, 88, 28-33. Cho, M. J.; Thompson, D. P.; Cramer, C. T.; Vidmar, T. J.; Scieszka, J. F. Pharm. Res., 1989, 6, 71-7.
328 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133]
Campillo and Páez
Matsson, P.; Bergstrom, C. A.; Nagahara, N.; Tavelin, S.; Norinder, U.; Artursson, P. J. Med. Chem., 2005, 48, 604-13. Fagerholm, U. J. Pharm. Pharmacol., 2007, 59, 905-16. Avdeef, A. Expert Opin. Drug Metab. Toxicol., 2005, 1, 325-42. Volpe, D. A. J. Pharm. Sci., 2008, 97, 712-25. Artursson, P.; Karlsson, J. Biochem. Biophys. Res. Commun., 1991, 175, 880-5. Linnankoski, J.; Ranta, V. P.; Yliperttula, M.; Urtti, A. Eur. J. Pharm. Sci., 2008, 25, 417-25. Hou, T. J.; Zhang, W.; Xia, K.; Qiao, X. B.; Xu, X. J. J. Chem. Inf. Comput. Sci., 2004, 44, 1585-600. Kansy, M.; Senner, F.; Gubernator, K. J. Med. Chem., 1998, 41, 1007-10. Verma, R. P.; Hansch, C.; Selassie, C. D. J. Comput. Aided Mol. Des., 2007, 21, 3-22. Sugano, K.; Hamada, H.; Machida, M.; Ushio, H. J. Biomol. Screen., 2001, 6, 189-96. Proulx, P. Biochim. Biophys. Acta, 1991, 1071, 255-71. Lipinski, C. A., Lombardo, F., Dominy B. W., Feeney P. J. Adv. Drug Deliv. Rev., 1997, 23, 3-25. Lipinski, C. A. J. Pharmacol. Toxicol. Methods, 2000, 44, 235-49. Ekins, S.; Mestres, J.; Testa, B. Br. J. Pharmacol., 2007, 152, 21-37. Egan, W. J. Ann. Rep. Med. Chem., 2007, 42, 449-67. Dokoumetzidis, A.; Kalantzi, L.; Fotaki, N. Expert Opin. Drug Metab. Toxicol., 2007, 3, 491-505. Andrews, C. W.; Bennett, L.; Yu, L. X. Pharm. Res., 2000, 17, 639-44. Sugawara, M.; Takekuma, Y.; Yamada, H.; Kobayashi, M.; Iseki, K.; Miyazaki, K. J. Pharm. Sci., 1998, 87, 960-6. Ren, S.; Lien, E. J. Prog. Drug Res., 2000, 54, 1-23. Oprea, T. I.; Gottfries, J. J. Mol. Graph. Model., 1999, 17, 261-74, 329. Norinder, U.; Osterberg, T. J. Pharm. Sci., 2001, 90, 1076-85 Yoshida, F.; Topliss, J. G. J. Med. Chem., 2000, 43, 2575-85. Stenberg, P.; Luthman, K.; Ellens, H.; Lee, C. P.; Smith, P. L.; Lago, A.; Elliott, J. D.; Artursson, P. Pharm. Res., 1999, 16, 1520-6. van de Waterbeemd, H.; Camenisch, G.; Folkers, G.; Raevsky, O. A. QSAR Comb. Sci., 1996, 15, 48090 Yamashita, F.; Wanchana, S.; Hashida, M. J. Pharm. Sci., 2002, 91, 2230-9. Wessel, M. D.; Jurs, P. C.; Tolan, J. W.; Muskal, S. M. J. Chem. Inf. Comput. Sci., 1998, 38, 726-35. Votano, J. R.; Parham, M.; Hall, L. H.; Kier, L. B. Mol. Divers., 2004, 8, 379-91. Niwa, T. J. Chem. Inf. Comput. Sci., 2003, 43, 113-9. Palm, K.; Luthman, K.; Ungell, A. L.; Strandlund, G.; Beigi, F.; Lundahl, P.; Artursson, P. J. Med. Chem., 1998, 41, 5382-92. Palm, K.; Luthman, K.; Ungell, A. L.; Strandlund, G.; Artursson, P. J. Pharm. Sci., 1996, 85, 32-9. Ertl, P.; Rohde, B.; Selzer, P. J. Med. Chem., 2000, 43, 3714-7. Norinder, U.; Osterberg, T.; Artursson, P. Pharm. Res., 1997, 14, 1786-91. Norinder, U.; Sjoberg, P.; Osterberg, T. J. Pharm. Sci., 1998, 87, 952-9. Fujiwara, S.; Yamashita, F.; Hashida, M. Int. J. Pharm., 2002, 237, 95-105. Osterberg, T.; Norinder, U. J. Chem. Inf. Comput. Sci., 2000, 40, 1408-11. Hou, T.; Wang, J.; Li, Y. J. Chem. Inf. Model., 2007, 47, 2408-15. Zhao, Y. H.; Le, J.; Abraham, M. H.; Hersey, A.; Eddershaw, P. J.; Luscombe, C. N.; Butina, D.; Beck, G.; Sherborne, B.; Cooper, I.; Platts, J. A. J. Pharm. Sci., 2001, 90, 749-84. Klopman, G.; Stefan, L. R.; Saiakhov, R. D. Eur. J. Pharm. Sci., 2002, 17, 253-63. Perez, M. A.; Sanz, M. B.; Torres, L. R.; Avalos, R. G.; Gonzalez, M. P.Diaz, H. G. Eur. J. Med. Chem., 2004, 39, 905-16. Jones, R.; Connolly, P. C.; Klamt, A.; Diedenhofen, M. J. Chem. Inf. Model., 2005, 45, 1337-42. Kramer, S.D.; Abbot, N.J.; Begley, D. J. In Pharmacokinetic Optimizacion in Drug Research. Testa, B. Eds. Wiley-VCH, 127-154. Bickel, U. NeuroRx, 2005, 2, 15-26. Norinder, U. ; Haeberlein, M. Adv. Drug Deliv. Rev., 2002, 54, 291-313. Lombardo, F.; Gifford, E.Shalaeva, M. Y. Mini Rev. Med. Chem., 2003, 3, 861-75. Ecker, G. F.Noe, C. R. Curr. Med. Chem., 2004, 11, 1617-28. Clark, D. E. Drug Discov. Today, 2003, 8, 927-33. Clark, D. E. Ann. Rep. Med. Chem., 2005, 40, 413-25. Young, R. C.; Mitchell, R. C.; Brown, T. H.; Ganellin, C. R.; Griffiths, R.; Jones, M.; Rana, K. K.; Saunders, D.; Smith, I. R.; Sore, N. E. J. Med. Chem., 1988, 31, 656-71. van de Waterbeemd, H.; Kamsky, M. Chimia, 1992, 46, 299-303.
In Silico ADME Approaches [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180]
Frontiers in Drug Design & Discovery, 2009, Vol. 4
329
Kelder, J.; Grootenhuis, P. D.; Bayada, D. M.; Delbressine, L. P.Ploemen, J. P. Pharm. Res., 1999, 16, 1514-9. Liu, R.; Sun, H.So, S. S. J. Chem. Inf. Comput. Sci., 2001, 41, 1623-32. Abraham, M. H.; Chadha, H. S.Mitchell, R. C. J. Pharm. Sci., 1994, 83, 1257-68. Abraham, M. H.; Takacs-Novak, K.Mitchell, R. C. J. Pharm. Sci., 1997, 86, 310-5. Lombardo, F.; Blake, J. F.Curatolo, W. J. J. Med. Chem., 1996, 39, 4750-5. Keseru, G. M.Molnar, L. J. Chem. Inf. Comput. Sci., 2001, 41, 120-8. Rose, K.; Hall, L. H.; Kier, L. B. J. Chem. Inf. Comput. Sci., 2002, 42, 651-66. Cabrera, M. A.; Bermejo, M.; Perez, M.Ramos, R. J. Pharm. Sci., 2004, 93, 1701-17. Pan, D.; Iyer, M.; Liu, J.; Li, Y.Hopfinger, A. J. J. Chem. Inf. Comput. Sci., 2004, 44, 2083-98. Abraham, M. H.; Ibrahim, A.; Zhao, Y.Acree, W. E., Jr. J. Pharm. Sci., 2006, 95, 2091-100. Katritzky, A. R.; Kuanar, M.; Slavov, S.; Dobchev, D. A.; Fara, D. C.; Karelson, M.; Acree, W. E., Jr.; Solovev, V. P.Varnek, A. Bioorg. Med. Chem., 2006, 14, 4888-917. Wichmann, K.; Diedenhofen, M.Klamt, A. J. Chem. Inf. Model., 2007, 47, 228-33. Konovalov, D. A.; Coomans, D.; Deconinck, E.Heyden, Y. V. J. Chem. Inf. Model., 2007, 47, 164856. Konovalov, D. A.; Sim, N.; Deconinck, E.; Heyden, Y. V.Coomans, D. J. Chem. Inf. Model., 2008, 48, 370-83. Van Damme, S.; Langenaeker, W.; Bultinck, P. J. Mol. Graph. Model., 2008, 26, 1223-36. Obrezanova, O.; Gola, J. M.; Champness, E. J.Segall, M. D. J. Comput. Aided Mol. Des., 2008, 22, 431-40. Luco, J. M. J. Chem. Inf. Comput. Sci., 1999, 39, 396-404. Crivori, P.; Cruciani, G.; Carrupt, P. A.Testa, B. J. Med. Chem., 2000, 43, 2204-16. Cruciani, G.; Pastor, M.; Guba, W. Eur. J. Pharm. Sci., 2000, 11, 29-39. Zhao, Y. H.; Abraham, M. H.; Ibrahim, A.; Fish, P. V.; Cole, S.; Lewis, M. L.; de Groot, M. J.; Reynolds, D. P. J. Chem. Inf. Model., 2007, 47, 170-5. Doniger, S.; Hofmann, T.; Yeh, J. J. Comput. Biol., 2002, 9, 849-64. Li, H.; Yap, C. W.; Ung, C. Y.; Xue, Y.; Cao, Z. W.; Chen, Y. Z. J. Chem. Inf. Model., 2005, 45, 1376-84. Fu, X. C.; Wang, G. P.; Liang, W. Q.; Yu, Q. S. Pharmazie, 2004, 59, 126-30. Winkler, D. A.; Burden, F. R. J. Mol. Graph. Model, 2004, 22, 499-505. Yap, C. W.; Chen, Y. Z. J. Pharm. Sci., 2005, 94, 153-68. Garg, P.; Verma, J. J. Chem. Inf. Model, 2006, 46, 289-97. Hemmateenejad, B.; Miri, R.; Safarpour, M. A; Mehdipour, A. R. J. Comput. Chem., 2006, 27, 112535. Pardridge, W. M. Drug Discov. Today, 2004, 9, 392-3. Levin, V. A. J. Med. Chem., 1980, 23, 682-4. Gratton, J. A.; Abraham, M. H.; Bradbury, M. W.; Chadha, H. S. J. Pharm., 1997, 49, 1211-6. Abraham, M. H. Eur. J. Med. Chem., 2004, 39, 235-40. Liu, X.; Tu, M.; Kelly, R. S.; Chen, C.; Smith, B. J. Drug Metab. Dispos., 2004, 32, 132-9. Lanevskij, K.; Japertas, P.; Didziapetris, R.Petrauskas, A. J. Pharm. Sci., 2008. Kriegl, J. M.; Eriksson, L.; Arnhold, T.; Beck, B.; Johansson, E.; Fox T. Eur. J. Pharm. Sci., 2005, 24, 451-463. Krishna, D.R.; Klotz, U. Clin. Pharmacokinet., 1994, 26, 144-60. de Wildt, S.N.; Kearns, G.L.; Leeder, J.S.; van den Anker, J.N. Clin. Pharmacokinet., 1999, 36, 43952. Rendic, S. Drug Metab. Rev., 2002, 34, 83-448. Daly, A.K. Fundam. Clin. Pharmacol., 2003, 17, 27-41. Smith, P.A.; Sorich, M.J.; Low, L.S.C.; McKinnon, R.A.; Miners, J.O. J. Mol. Graph. Model., 2004, 22, 507-17. Chohan, K.K.; Paine, S.W.; Waters, N.J. Curr. Top. Med. Chem., 2006, 6, 1569-78. Jensen, B.F.; Vind, C.; Padkjær, S.B, Brockhoff, P.B.; Refsgaard, H.H.F. J. Med. Chem., 2007, 50, 501-11. Lewis, D.F.T. J. Chem. Technol. Biotechnol., 2001, 76, 237-44. Wolbrette, D.L. Curr. Cardiol. Rep., 2004, 6, 379-84. Arimoto, R. Curr. Topics. Med. Chem., 2006, 6, 1609-18. O'Brien S.E.; de Groot, M. J. J. Med. Chem., 2005, 48, 1287 -91. Van de Waterbeemd H.; Gifford E. Nature, 2003, 2, 192-2004. Riley, R J.; Grime, K. Drug Discov. Today, 2004, 1, 365-72.
330 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [193] [194] [195] [196] [197] [198] [199] [200] [201] [202] [203] [204] [205] [206] [207] [208] [209] [210] [211] [212] [213] [214] [215] [216] [217] [218] [219] [220] [221] [222] [223]
Campillo and Páez
Lipscomb, J.C.; Poet, T.S. Pharmacol. Ther., 2008, 118, 82-108. Lewis, D.F.V. Inflammopharmacology, 2003, 11, 43-73. Smith, P.A.; Sorich, M.J.; Los, L.S.C.; McKinnon, R.A.; Miners, J.O. J. Mol. Graph. Model., 2004, 22, 507-17. Radominska-Pandya, A.; Ouzzine, M.; Fournel-Gigleux, S.; Magdalou, J. Meths. Enzymol, 2005, 400, 116-47. Sorich, M.J.; Miners, J.O.; Mckinnon, R.A.; Smith, P.A. Mol. Pharm., 2004, 65, 301-8. Kless, A.; Eitrich, T. Lect. Notes Comput. Sci., 2004, 3303, 191-205. Chohan, K. K.; Paine, S. W.; Mistry, J.; Barton, P.; Davis, A. M. A J. Med. Chem., 2005, 48, 5154-61. Yap, C. W.; Chen, Y. Z. J. Chem. Inf. Model., 2005, 45, 982-92. Susnow, R. G.; Dixon, S. L. J.Chem. Inf. Comput. Sci., 2003, 43, 1308-15. O'Brien, S. E.; de Groot, M. J. J. Med. Chem., 2005, 48, 1287-91. Molnár, L.; Keserü, G. M. Bioorg. Med. Chem. Lett., 2002, 12, 419-21. Merkwirth, C.; Mauser, H.; Schulz-Gasch, T.; Roche, O.; Stahl, M.; Lengauer, T. J. Chem. Inf. Comput. Sci., 2004, 44, 1971-8. Arimoto, R.; Prasad, M. A.; Gifford, E. M. J. Biomol. Screen., 2005, 10, 197-205. Kriegl, J. M.; Arnhold, T.; Beck, B.; Fox, T. QSAR Comb. Sci., 2005, 24, 491-502. Kriegl, J. M.; Arnhold, T.; Beck, B.; Fox, T. J. Comput. Aided Mol. Des., 2005, 19, 189-201. Balakin, K. V.; Ekins, S.; Bugrim, A.; Ivanenkov, Y. A.; Korolev, D.; Nikolsky, Y. V.; Skorenko, A. V.; Ivashchenko, A. A.; Savchuk, N. P.; Nikolskaya, T. Drug Metab. Dispos., 2004, 32, 1183-9. Manga, N.; Duffy, J. C.; Rowe, P. H.; Cronin, M. T. SAR QSAR Environ. Res., 2005, 16, 43-61. Zuegge, J.; Fechner, U.; Roche, O.; Parrott, N. J.; Engkvist, O.; Schneider, G. Quant. Struct. Act. Relat., 2002, 21, 249-56. Kriegl, J.M.; Ericsson, L.; Arnhold, T.; Beck, B.; Johansson, E.; Fox.T. Eur. J. Pharm. Sci., 2005, 24, 451-63. Jensen, B. F., Vind, C.; Padkjaer B.S.; Brockhoff, P.B.; Refsgaard, H.F. J. Med. Chem., 2007, 50, 50111. Yamashita, F.; Hara, H.; Ito, T.; Hashida, M. J.Chem. Inf. Model, 2008, 48, 364-9. Hudelson, M.G.; Ketkar, N.S.; Hodler, L.B.; Carlson, T.J. Peng, C.C.; Waldher, B.J.; Jones, J.P. J. Med. Chem., 2008, 51, 638-54. Hudelson, M.G.; Jones, J.P. J. Med. Chem., 2006, 49, 4367-73. Sheridan, R.P.; Korzekwa, k.R.; Torres, R.A; Walder M.J. J. Med. Chem., 2007, 50, 3173-84. Yap, C.W.; Xue, Y.; Li, Z.R.; Chen, Y.Z. Curr. Med. Chem., 2006, 6, 1593-607. de Graaf, C.; Oostenbrink, C.; Keizers, P. H. J.; van der Wijst, T.; Jongejan, A.; Vermeulen, N. P. E. J. Med. Chem., 2006, 49, 2417-30. Zhou, D.; Afzelius, L.; Grimm, S. W.; Andersson, T. B.; Zauhar, R. J.; Zamora, I. Drug Metab. Disp., 2006, 34, 976-83. Cruciani, G.; Carosati, E.; DeBoeck, B.; Ethirajulu, K.; Mackie, C.; Howe, T.; Vianello, R. J. Med. Chem., 2005, 48, 6970-9. Singh, S. B.; Shen, L. Q.; Walker, M. J.; Sheridan, R. P. J. Med. Chem., 2003, 46, 1330-6. Jones, J. P.; Mysinger, M.; Korzekwa, K. R. Drug Metab. Dispos., 2002, 30, 7-12. Ehrlich, P. Ber. dtsch. Chem. Ges., 1909, 42, 17-47. Gund, P. Prog. Mol. Subcell. Biol., 1977, 5, 117-43. Guérin GA., Pratuangdejkul J., Monica Alemany M., Launay J and Manivet P. Drug Discov. Today, 2006, 11, 991-8. Cruciani, G. Molecular Interaction Fields. Application in Drug Discovery and ADME Predictions, V27. Wiley-VCH, 2006. Lozano, J. J.; Pastor, M.; Cruciani, G.; Gaedt, K.; Centeno, N. B.; Gago, F.; Sanz, F. J. Comput. Aided Mol. Des., 2000, 14, 341-53. Poso, A.; Juvonen, R.; Gynther, J. Quant. Struct. Act. Relat., 1995, 14, 507-11. Bravi, G.; Wikel, J. H. Quant. Struct. Act. Relat., 2000, 19, 29-38. Poso, A.; Gynther, J.; Juvonen, R. J. Comput. Aided Mol. Des., 2001, 15, 195-02. Ekins, S.; Bravi, G.; Ring, B. J.; Gillespie, T. A.; Gillespie, J. S.; Vandenbranden, M.; Wrighton, S.A.; Wikel, J.S. J. Pharmacol. Exp. Ther., 1999, 288, 21-9. Wang, Q.; Halpert, J. R. Drug Metab. Dispos., 2002, 30, 86-95. Ridderstrom, M.; Zamora, I.; Fjellstrom, O.; Andersson, T. B. J. Med. Chem., 2001, 44, 4072-81. Tanaka, T.; Kamiguchi, N.; Okuda, T.; Yamamoto, Y. Chem. Pharm. Bull. (Tokyo), 2004, 52, 836-41. Jones, B. C.; Hawksworth, G.; Horne, V. A.; Newlands, A.; Morsman, J.; Tute, M.S.; Smith, D.A. Drug Metab. Dispos. 1996, 24, 260-6.
In Silico ADME Approaches [224] [225] [226] [227] [228] [229] [230] [231] [232] [233] [234] [235] [236] [237] [238] [239] [240] [241] [242] [243] [244] [245] [246] [247] [248] [249] [250] [251] [252] [253] [254] [255] [256] [257] [258] [259] [260] [261] [262]
Frontiers in Drug Design & Discovery, 2009, Vol. 4
331
Mancy, A.; Dijols, S.; Poli, S.; Guengerich, P.; Mansuy, D. Biochemistry, 1996, 35, 16205-12. Mancy, A.; Broto, P.; Dijols, S.; Dansette, P. M.; Mansuy, D. Biochemistry, 1995, 34, 10365-75. Jones, J. P.; He, M. X.; Trager, W. F.; Rettie, A. E. Drug Metab. Dispos., 1996, 24, 1-6. Rao, S.; Aoyama, R.; Schrag, M.; Trager, W. F.; Rettie, A.; Jone, J.P. J. Med. Chem., 2000, 43, 278996. Ekins, S.; Bravi, G.; Binkley, S.; Gillespie, J. S.; Ring, B. J.; Wikel, J.H.; Wrighton, S.A. Drug Metab. Dispos., 2000, 28, 994-1002. Afzelius, L.; Zamora, I.; Ridderstrom, M.; Andersson, T. B.; Karlen, A.; Masimirembwa, C. M. Mol. Pharmacol., 2001, 59, 909-19. Afzelius, L.; Masimirembwa, C. M.; Karlen, A.; Andersson, T. B.; Zamora, I. J. Comput. Aided Mol. Des., 2002, 16, 443-58. de Groot, M. J.; Alex, A. A.; Jones, B. C. J. Med. Chem., 2002, 45, 1983-93. Egnell, A. C.; Eriksson, C.; Albertson, N.; Houston, B.; Boyer, S. J. Pharmacol. Exp. Ther., 2003, 307, 878-87. Zamora, I.; Afzelius, L.; Cruciani, G. J. Med. Chem., 2003, 46, 2313-24. Afzelius, L.; Zamora, I.; Masimirembwa, C. M.; Karlen, A.; Andersson, T. B.; Mecucci, S.; Baroni, M.; Cruciani. G. J. Med. Chem., 2004, 47, 907-14. Wolff, T.; Distelrath, L. M.; Worthington, M. T.; Groopman, J. D.; Hammons, G. J.; Kadlubar, F.F.; Prough, R.A.; Martin, M.V.; Guengerich, F.P Cancer Res., 1985, 45, 2116-22. Meyer, U.A.; Gut, j.; Kronbach, T.; Skoda, C.; Meier, U.T. Xenobiotica, 1986, 16, 449-64. Islam, S. A.; Wolf, C. R.; Lennard, M. S.; Sternberg, M. J. Carcinogenesis, 1991, 12, 2211-9. Koymans, L.; Vermeulen, N. P.; Van Acker, S. A.; Te Koppele, J. M.; Heykants, J. J.; Lavrijsen, K.; Meuldermans, W.; Donne-Op den Kelder, G.M. Chem. Res. Toxicol., 1992, 5, 211-9. de Groot, M. J.; Bijloo, G. J.; Martens, B. J.; van Acker, F. A.; Vermeulen, N. P. Chem. Res. Toxicol., 1997, 10, 41-8. de Groot, M. J.; Ackland, M. J.; Horne, V. A.; Alex, A. A.; Jones, B. C. J. Med. Chem., 1999, 42, 4062-70. de Groot, M. J.; Ackland, M. J.; Horne, V. A.; Alex, A. A.; Jones, B. C. J. Med. Chem., 1999, 42, 1515-24. Strobl, G. R.; von Kruedener, S.; Stockigt, J.; Guengerich, F. P.; Wolff, T. J. Med. Chem., 1993, 36, 1136-45. Ekins, S.; Bravi, G.; Binkley, S.; Gillespie, J. S.; Ring, B. J. Pharmacogenetics, 1999, 9, 477-89. Snyder, R.; Sangar, R.; Wang, J. B.; Ekins, S. Quant. Struct. Act. Relat., 2002, 21, 357-68. Haji-Momenian, S.; Rieger, J. M.; Macdonald, T. L.; Brown, M. L. Bioorg. Med. Chem., 2003, 11, 5545-54. Waller, C. L.; Evans, M. V.; McKinney, J. D. Drug Metab. Dispos., 1996, 24, 203-10. Lewis, D. F.; Eddershaw, P. J.; Goldfarb, P. S.; Tarbit, M. H. Xenobiotica, 1996, 26, 1067-86. Ekins, S.; Bravi, G.; Binkley, S.; Gillespie, J. S.; Ring, B. J. J. Pharmacol. Exp. Ther., 1999, 290, 42938. Ekins, S.; Bravi, G.; Wikel, J. H.; Wrighton, S. A. J. Pharmacol. Exp. Ther., 1999, 291, 424-33. Ekins, S.; Stresser, D. M.; Williams, J. A. Trends Pharmacol. Sci., 2003, 24, 161-6. Schappach, A.; Holtje, H. D. Pharmazie, 2001, 56, 835-42. Clement, O. O.; Freeman, C. M.; Hartmann, R. W.; Handratta, V. D.; Vasatis, T. S.; Brodie, A.M.; Njar, V.C. J. Med. Chem., 2003, 46, 2345-51. Nagy, P. I.; Tokarski, J.; Hopfinger, A. J. J. Chem. Inf. Comput. Sci., 1994, 34, 1190-7. Oprea, T. I.; Garcia, A. E. J. Comput. Aided Mol. Des., 1996, 10, 186-200. Recanatini, M.; Cavalli, A. Bioorg. Med. Chem., 1998, 6, 377-88. Tafi, A.; Anastassopoulou, J.; Theophanides, T.; Botta, M.; Corelli, F.; Massa, S.; Artico, M.; Costi, R.; Di Santo, R.; Ragno, R. J. Med. Chem., 1996, 39, 1227-35. Talele, T. T.; Kulkarni, S. S.; Kulkarni, V. M. J. Chem. Inf. Comput. Sci., 1999, 39, 958-66. Bargar, T. M.; Secor, J.; Markley, L. D.; Shaw, B. A.; Erickson, J. A. Pestic. Sci., 1999, 55, 1059-69. Fishelovitch, D.; Hazan, C.; Hirao, H.; Wolfson, H.J. Nussinov, R.; Shaik, S. J. Phys. Chem., B. 2007, 111, 13822-32. Fishelovitch, D.; Hazan, C.; Shaik, S.; Wolfson, H.J.; Nussinov, R. J. Am. Chem. Soc., 2007, 129, 1602-11. Muegge I, Rarey M. In Reviews in Computational Chemistry, Boyd, D.B.; Lipkowitz, K.B (Eds). Wiley-VCH New York, 2001, 1-60. Berman, H.M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bath, T.N; Wissig, H.; Shindyalov, I.N.; Bourne, P.E. Nucleic Acids Res., 2000, 28, 235-42.
332 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [263] [264] [265] [266] [267] [268] [269] [270] [271] [272] [273] [274] [275] [276] [277] [278] [279] [280] [281] [282] [283] [284] [285] [286]
Campillo and Páez
Murzin A. G., Brenner S. E., Hubbard T., Chothia C. J. Mol. Biol., 1995, 247, 536-40. William, P.A.; Cosme, J.; Sridhar, V.; Johnson, E.F.; McRee, D.E. J. Inorg. Biochem., 2000, 81, 18390. Wester, M.R.; Johnson, E.F.; Marques-Soares, C.; Dansette, P.M.; Mansuy, D.; Stout, C.D. Biochemistry, 2003, 4, 6370-9. Wester, M.R.; Johnson, E.F.; Marques-Soares, C.; Dijols, S.; Dansette, P.M.; Mansuy, D.; Scout, C.D. Biochemistry, 2003, 42, 9335-45. Zhao, Y.;White, M.A., Muralidhara, B.K., Sun, L., Halpert, J.R., Stout, C.D. J. Biol. Chem., 2006, 281, 5973-81. Scott, E.E.; White, M.A.; He, Y.A.; Johnson, E.F.; Stout, C.D.; Halpert, J.R. J. Biol. Chem., 2004, 279, 27294-301. Zhao, Y.; Sun, L.; Muralidhara, B.K.; Kumar, S.; White, M.A.; Stout, C.D.; Halpert, J.R. Biochemistry, 2007, 46, 11559-67. Schoch, G.A.; Yano, J.K.; Wester, M.R.; Griffin, K.J. ; Stout, C.D. ; Johnson, E.F. J. Biol. Chem., 2004, 279, 9497-503. Wester, M.R.; Yano, J.K.; Schoch, G.A.; Yang, C.; Griffin, K.J.; Stout, C.D.; Johnson, E.F. J. Biol. Chem., 2004, 279, 35630-7. Yano, J.K.; Wester, M.R.; Schoch, G.A.; Griffin, K.J.; Stout, C.D.; Johnson, E.F. J. Biol. Chem., 2004 279, 38091-4. Ekroos, M.; Sjogren, T. Proc. Natl. Acad. Sci. USA, 2006, 103 13682-7. Williams, P.A.; Cosme, J.; Vinkovic, D.M.; Ward, A.; Angove, H.C.; Day, P.J.; Vonrhein, C.; Tickle, I.J.; Jhoti, H. Science, 2004, 305, 683-6. Schwede T.; Kopp, J.; Guex N.; Peitsch M.C. Nucleic Acids Res., 2003, 31, 3381-5. Li, W.; Tang, Y.; Liu, H.; Cheng, J.; Zhu, W.; Jiang H. Proteins, 2008, 71, 938-49. Lafite, P.; André, F.; Zeldin, D.C.; Dansette, P.M.; Mansuy D. Biochemistry, 2007, 46, 10237-47. Pradad J.C.; Goldstone J.V.; Camacho C.J.; Vajda S.; Stegeman J. J. Biochem., 2007, 46, 2640-54. Yu, J.; Paine, M.; Marécha J.D.; Kemp, C.A.; Ward, C.J.; Brown, S.; Sutcliffe, M.J.; Roberts, G.C.; Rankin, E.M.; Wolf, C.R. Drug Metab. Disp., 2006, 34, 1386-92. de Graaf, C.; Oostenbrink, C.; Keizers, P H. J.; van der Wijst, T., Jongejan, A.; Vermeulen, N.P.E. J. Med. Chem., 2006, 49, 2417-30. Ekroos, M.; Sjögren, T. Proc. Natl. Acad. Sci. USA, 2006, 103, 13682-7. Oh, W.S.; Kim, D.N.; Jung, J.; Cho, K.H.; No, K.T. J. Chem. Inf. Model, 2008, 48, 591-601. Charles, W; Locuson, P. M.; Gannett, R.A.; Tracy, T. S. J. Med. Chem., 2007, 50, 1158-65. Theil, F.P.; Guentert, T.W.; Haddad, S.; Poulin, P. Toxicol. Lett., 2003, 138, 29-49. Fagerholm, U. J. Pharm. Pharmacol., 2008, 60, 535-42. Lipscomb, J. C.; Poet T.S. Pharmacol. Therapeut., 2008, 118, 82-103.
Frontiers in Drug Design & Discovery, 2009, 4, 333-350
333
Advances in ADMET Predictions and Modeling: Rapid Drug Discovery Efforts in 21st Centuries Mahmud Tareq Hassan Khan* Department of Pharmacology, Institute of Medical Biology, Faculty of Medicine, University of Tromsø, 9037 Tromsø, Norway Abstract: A large number of compounds in the development stage got failure due to the unfavourable ADMET (absorption, distribution, metabolism, excretion and toxicity) profiles. The utilities of ADMET properties are becoming progressively more imperative in the drug discovery processes, assortment, development and promotion processes. In recent years several review papers have been published about the possibilities of the prediction or the ADMET properties using different structural features of the molecules, i.e., molecular descriptors, and utilizing multiple approaches. One of the most important approaches is QSAR modelling of the data derived from their activity profiles and their different structural features (i.e., quantitative molecular descriptors). More and more efforts are put to the field of ADMET predictions. This chapter will critically assess some of the most important and recently reported topics for the effective in silico predictions of the ADMET properties of the potential drug candidates based on QSAR modelling approaches.
Key Words: ADMET, molecular descriptor, QSAR modeling, logP, logD, artificial neural network, PSA, multiple linear regressions, machine learning, support vector machine, inductive logic programming. INTRODUCTION Over the past century, drug discovery has involved the alliance of chemistry and biology, specifically pharmacology [1]. Chemical compounds were synthesized or extracted from natural sources and exposed to different in vitro as well as well in vivo assays to identify potentially active candidates [2]. The selection of potential drug candidate is now widely viewed as an important and relatively new, yet largely unsolved, bottleneck in the drug discovery and development process. In order to achieve an efficient selection process, high quality, rapid, predictive and correlative absorption, distribution, metabolism, excretion and toxicity (ADMET) models are required in order for them to be confidently used to support critical financial decisions [2].
*Corresponding Author: Tel: +47-776-46755; Fax: +47-776-45310; E-mail:
[email protected];
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
334 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Mahmud Tareq Hassan Khan
In past few years, combinatorial chemistry and high-throughput screening (HTS) have radically augmented the number of compounds for which early data on ADMET are necessary, which has in turn driven the development of a variety of medium and high-throughput in vitro ADMET screens [3, 4]. Considerations about ADMET are particular concern with issues such as high plasma protein binding requiring extra studies for FDA approval [5]. During drug development, approximately 40-60% [6] of the compounds are rejected due to unfavorable properties of ADMET. Increasing efforts are therefore put into the in silico prediction of ADMET properties [7-9], which offers an economical and high-throughput way of evaluating these properties prior to synthesis and biological testing. Such an approach may reduce the need of experimental work considerably. High oral bioavailability is a very desirable condition for a drug candidate. In silico ADMET predictions may give the possibility of rejecting a molecule with low oral bioavailability early in the development process [9]. Oral bioavailability prediction was pioneered by the rule of five (ROF) proposed by Lipinski and coworkers (Lipinski 'Rule-of-five') [9, 10]. The ROF defined different descriptors for identifying compounds with possible poor absorption and permeability [10]. The descriptors for the Lipinski Rule-of-five (ROF) are [10, 11]: (1) Molecular weight >500, (2) Calculated log P > 5, (3) Number of hydrogen bond donors (OH and NH groups) > 5, and (4) Number of hydrogen-bond acceptors (N and O atoms) > 10. Structural features of the molecules (like aqueous solubility, molecular weight, polarizability, etc.) can be exploited to envisage information about the ADMET which could be the significant strategies in the early stage of the drug development [11, 12]. However, information about the metabolism is not easy to adopt from the structural features of a compound. Hansch and co-workers (in 2004) reported that the permeability of organic compounds into Caco-2 cells, as a model for intestinal lining, were useful for modelling the absorption. However, they also suggested that the most general and useful parameter for predicting absorption was the octanol/water partition coefficient (logP) [12]. A similar conclusion was also given by others [13]. In addition to the logP values, it was suggested that molecular weight (MW) and the aqueous solubility profiles were important parameters for estimating absorption [13]. The use of data-based modeling approaches are effective for many drug ADMET features, such as passive membrane permeation, where their molecular mechanism is barely delineated. Therefore, quantitative structureactivity relationship (QSAR) approaches have been applied to generate relationships between the ADMET parameters and the structure and properties of the molecules [11, 14]. In addition to the basic ADMET parameters, features like solubility, bioavailability, oral absorption, and side effects are important for modeling of drug molecules or drug candidates. The modelling process must generate models that are predictive for all these features and provide information that guides the synthesis of new entities with aspire of increasing advantageous properties, whilst declining the undesirable ones. Therefore, the focus of the modeling must be on the structural attributes that are appreciably allied to the definite properties [15].
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 335
In the present chapter the ADMET predictions utilizing QSAR modeling approaches based on the molecular descriptors and chemical features of druggable molecules are discussed. MOLECULAR DESCRIPTORS (MDs) Molecular descriptors (MDs) are the numerical values that characterize properties of molecules and obtained by well-specified algorithms (like, 2D fingerprints). These MDs could be empirical or calculated. These are the final results of logical and mathematical procedures which can transform chemical information encoded within symbolic representation of molecules into useful numbers [16]. In other words, MDs are the mathematical values that describe the structure or shape of molecules, valuable for predicting the activity and properties of molecules in complex experiments. These descriptors allow to find the structure/response correlations; help to perform similarity searching; and to perform substructure searching [16]. The descriptors are able to be used as contrivance for data mining in chemical registry files, like in the Chemical Abstracting Services (CAS). The CAS system contains information about molecular formula and weight, ring systems, hydrogen bond donors (HBD) and acceptors (HBA), logP, logD, pKa, solubilities, rotatable bonds, PSA, TPSA (topological PSA) [17], HCTPSA (high-charged topological polar surface area) [17], polarizability, etc, which are the most commonly used MDs in general practice. These features describe intrinsic molecular characteristics as well as how molecules can interact with their surrounding environments, and can also be used to predict the relevance of the molecules [18]. The MDs can vary complexity of encoded information and in computational time. Table 1 showing some molecular descriptors and their codes very often used in studies for the profiling of drug candidates [5, 19]. Table 1.
A Selection of Molecular Descriptors (MDs) Used in Compound Profiling [5, 19]
Descriptors (codes)
Descriptors (codes)
Calculated logD pH 7.4 (logD)
Molecular flexibility (FLEX)
Calculated logP (logP)
Calculated molar refractivity (CMR)
Hydrogen bond acceptors (HBA)
Molecular weight (MW)
Hydrogen bond donors (HBD)
Polar surface area (PSA)
Negatively ionizable group (NEG)
Rotatable bonds (bonds)
Positively ionizable group (POS)
Heavy atom count (heavy)
In Todeschini and Consonni’s Handbook of Molecular Descriptors [16], several hundreds of descriptors are described with an explanation of their mathematical derivation. A list of books describing MDs and their chemical-mathematical derivations is inc-
336 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Mahmud Tareq Hassan Khan
luded at the end of this chapter. More information is available about the molecular descriptors at www.moleculardescriptors.eu [11]. ABSORPTION Oral administration is the most convenient rout for patients to receive drugs. When the drug is administered orally, it must be absorbed across the epithelium of the small intestine. The intestinal absorption of a drug molecule includes two stages – (a) dissolution, and (b) membrane transfer. [6] Once the molecule is dissolved in the gastrointestinal contents, which mainly is in aqueous phase, it must be transferred to the blood circulation through the barrier of the intestinal tract. In the blood circulation the molecules are facing several in vivo barriers, quite similar to the barrier of the intestinal tract [6]. Other barriers that the molecule needs to overcome include the blood-brain barrier (BBB) and the stratum corneum of the skin. The relative affinity of a drug to the blood or brain tissue can be expressed in terms of blood-brain partition coefficient (logBB) [20]. All these in vivo barriers are similar and are mainly composed of a lipid-bilayer, which results from the special arrangements of different types of lipids, like phospholipids, glycolipids, cholesterol, etc., within the water medium [6, 21]. A large arrays of proteins, like selective ion channels (Na+, K+, Ca2+, Cl-, etc.), membrane receptor and transport proteins are embedded within the membrane lipid-bilayers [21]. The “Lipinski rule of 5” [10] is a good determinant for estimating the “druglikeness” of molecules, as discussed in earlier section, but the rule is too general for being valid for accurate ADME predictions. However, this thumb rule, ROF, can be used to distinguish well-absorbed molecules from poorly-absorbed molecules [6]. For precise and efficient prediction of intestinal absorption, several successful in vitro methods have been developed [6]. The most prominent cell-based model for intestinal permeability is the Caco-2 cell system [22-24]. Extensive studies have shown that oral drug absorption and the permeability coefficient derived from Caco-2 model have a significant sigmoidal relationship [24], indicating that intestinal absorption can be efficiently predicted by this in vitro model. Predictions of the ADMET properties are usually performed by two approaches modeling the data, i.e., QSAR modeling of the drug molecules, based on quantitative data from different descriptors which are being correlated with the biological activity of the set of ligands; [25-28] and molecular modeling, i.e., molecular mechanics (MM), pharmacophore modeling, docking, and quantum mechanics (QM), calculations usually used to explore the interaction between the potential drug molecules and the proteins related to ADMET processes, like cytochrome P450 subtypes [6, 25-28]. The modeling of parameters derived from different physicochemical features of the molecules as descriptors for their activity are mainly done using different QSAR and QSPR (quantitative structure-property relationships) approaches. Based on different appropriate descriptors, QSAR and QSPR models exploiting from simple multiple regression analysis (MLR) to the most modern and complex multivariate analysis or machinelearning methods are generated [6]. Some of the most modern and multivariate approaches included are – partial least square (PLS) [29], linear discriminant analysis (LDA), artificial neural networks (ANN) [29], genetic algorithms (GAs) [29], support vector machines (SVMs), inductive logic programming (ILP) [30-32], k-nearest neighbor
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 337
(kNN) method [29, 33-36], Bayesian modeling [37-39], self-organizing map (SOM) [37], and several others. The QSAR and QSPR modeling are most effective when a large number of molecules are included, the chemical diversity between the ligand is low, and the biological data included are consistent [11]. AQUEOUS SOLUBILITY (logS) The aqueous solubility (logS) of organic molecules should also be considered in drug design, since this parameter usually has a significant impact on many ADME-concerned properties of drugs, including uptake, distribution, transport, and bioavailability [8, 40]. It was observed that the synthesis of combinatorial libraries seemed to give compounds with more suitable lipophilicity (logP) and logS values, and thereby more ‘drug-like’ molecules than classical synthetic approaches. Therefore, for orally administrated drugs, in silico strategies are used to select sub- and fragment libraries with the relevant physicochemical properties, such as logP and logs [11, 40]. A large number of methods have been suggested in earlier reports for the calculation and prediction of the solubility (the logS) [41-61]. These methods generally consist of MLR or artificial neural networks (ANN) using various molecular descriptors, [40] and can be roughly divided into three categories – (a) experiment-related methods, (b) descriptor-based methods, and (c) group contribution methods [40]. SKIN PERMEABILITY Human skin is a highly complex organ made of multi-layers of composite materials including the subcutaneous tissue, the dermis and the epidermis [62-65]. A main function of the human skin is to regulate the entry of foreign substances into the body [66]. The barrier function of skin has attracted great scientific interest because of the relevance to a wide range of applications including transdermal delivery of drugs, [67] enhancement of sensorial and functional benefits of skin care products [68] and risk assessment of hazardous exposure to chemicals [69] In recent years large number of mathematical and QSAR models have been proposed for predicting skin permeability, mostly empirical and very few are deterministic. Early empirical models use simple lipophilicity parameters. The current tendency is to use more intricate molecular descriptors. There has been much debate on which models best predict skin permeability [66]. In a very recent scientific report Lian et al. (2008) [66] evaluated various mathematical models using a comprehensive investigational dataset of skin permeability for 124 molecules extracted from different resources. Of the seven models compared, the deterministic model of Mitragotri gives the best prediction. The simple QSPR model of Potts and Guy gives the second best prediction. The two models have many features in common. Both assume the lipid matrix as the pathway of transdermal permeation. Both use octanol–water partition coefficient and molecular size. Even the mathematical formulae are similar. All other empirical QSPR models that use more complicated descriptors fail to provide satisfactory prediction. The descriptors in the more complicated QSPR models are empirically related to skin permeation. Mathematically it is an ill-defined approach to use many co-linearly related parameters rather than fewer independent parameters in MLR [66]. Authors proposed some of the following models for the prediction of the skin permeability [66]:
338 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Mahmud Tareq Hassan Khan
logKp = 0.71logKow – 0.0061MW – 6.3 (n = 93, r2 = 0.67) where the skin permeability is in cm/sec, Kow is the octanol–water partition coefficient and MW is the molecular weight [66]
log K p = 5.241 + 0.437 R 2 0.41 2H 1.631 2H 3.286 2H + 2.012Vx (n = 47, r2 = 0.9567)
where the skin permeability is in cm/sec, R2 is an excess molar refraction, larity-polarizability,
H 2
the overall hydrogen bond acidity,
2H the dipo-
H 2
the overall
hydrogen bond basicity and Vx is the McGowan characteristic molecular volume [66]. PREDICTION OF BLOOD-BRAIN BARRIER DIFFUSION (logBB) A good example that exemplifies the great utility of a predictive computational model in drug discovery is a model for predicting blood-brain barrier (BBB) penetration [17]. Predicting the permeation through the BBB have been extensively studied, and is one of the most important ADME properties considered. The blood-brain partition coefficient (logBB) at steady state is a parameter commonly used to express the extent of a drug passing the BBB. Penetration of the BBB by new chemical entities is an important parameter in the drug discovery and development process [20]. The BBB is a complex cellular system consisting of endothelial cells of the brain capillaries, and takes part in maintaining the homeostasis of the central nervous system (CNS) by extrication the brain from the systemic blood circulation [70]. In the search of new CNS acting drugs, the ideal drug candidates must be able to penetrate BBB effectively, while peripherally acting drugs must have limited ability to cross BBB to avoid adverse CNS effects [11, 20]. Experimentally, the relative affinity of a drug to the blood or brain can be expressed by the terms of the logBB [20], logBB = log (Cbrain/Cblood) where Cbrain and Cblood are the equilibrium concentrations of the drug in the brain and the blood, respectively. Table 2 showing some examples molecules with their experimental blood-brain partition coefficient (logBB) [15]. This table points toward several structural aspects which are varying, such as number and types of rings, polarity, H-bond donors and/or acceptors, elements, bond orders, etc. [15]. Experimental determination of BBB penetration is time consuming, expensive and requires a sufficient quantity of the pure compounds, often in radio-labeled form, [20] and constitutes a major bottleneck for high-throughput screening of large molecular libraries [71-73]. This is one of the reasons for that a large number of in silico models for logBB estimation is reported in the literature and reviewed several times [72, 74-76]. A reliable and easily applicable computational model for predicting BBB permeation of drug candidates can help in an early identification of compounds with poor BBB penetration profile, prior even to chemical synthesis and will have a significant impact on drug discovery and development [4, 20, 77]. Numerous computational techniques have been
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 339
employed for calculating and predicting the BBB-penetrating (BBB+) and non-penetrating (BBB-) properties of compounds, with general precisions of 75-97% [72]. Table 2.
Examples of Experimental Blood-Brain Partition Coefficient (Logbb) Values of Some Molecules (Adopted from Hall, 2004) [15] Molecules
logBB
H N N
1.2
Cl N
H N
0.11
N H Cl
O -0.1
N N
H N
H N N
S CN
N
NH
-1.42
Several reports have predicted the BBB permeation using molecular properties such as lipophilicity (logP), topological indices, PSA, quantum chemical (QM) descriptors, and many others. Some of the reported models suffer from the use of computationally intensive calculations, which is making them not amenable for virtual screening of large libraries of compounds [17, 78-91]. The most common statistical methods applied for these studies are multiple linear regression (MLR) [17, 78-80, 82, 83, 86-89], principal component analysis (PCA) [84] and partial least squares regression (PLS) [11, 18]. PLASMA PROTEIN BINDING It is usually implicit that only free drug can cross membranes and bind to the intended molecular target(s), [92] and it is consequently significant to calculate the fraction of drug bound to plasma proteins [4]. The binding of a drug to plasma protein influences the pharmacodynamic behavior of the drug molecules significantly.Prediction of the corresponding ADME properties are important in the early stages of drug design [15]. Hall and co-workers proposed several models for the predicting the percent protein bin-
340 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Mahmud Tareq Hassan Khan
ding (%PPB) using E-state and descriptors [93, 94]. They selected a group of penicillin derivatives as training set for developing a QSAR model for the %PPB in human serum. They following predictive model was proposed[93, 94]: PPB = 14.6(±1.6)*(R2 2v) + 1.35(±0.25)*(CST(sF Cl)) + 4.44(±1.01)*(R2 ST(ssCH2)) – 21.1(±1.9)*(R2 C HS T(sNH2 NH)) + 2.25(±0.45)*ST(arom) + 13.2(±4.6) r2 = 0.8, s = 12.1, F = 60, n = 74, q2press = 0.76, spress = 13.4 The training set of penicillins was tested as a basis for designing new molecules that hopefully could become commercial drugs. The developed QSAR model was used to predict the putative %PPB of the designed penicillins, and the results indicated that the QSAR model could be used reliably in the drug-design process [93, 94] Drugs can bind to a variety of particles in the blood, including RBC, WBC and platelets, in addition to proteins, like albumin (in particular acidic drugs), 1-acid glycoproteins (e.g., basic drugs), lipoproteins (e.g., neutral and basic drugs), erythrocytes and , , -globulins [4, 92]. PREDICTION OF METABOLISM Quite a lot of facets of metabolism and elimination are pertinent to drug discovery, including the rate and extent of metabolism (turnover), the enzymes involved and the products formed, each of which can give rise to different concerns [4]. The degree and velocity of metabolism affect the clearance, while the contribution of meticulous enzymes might lead to issues related to the polymorphic nature of some of these enzymes and to drug–drug interactions [4]. QSAR and molecular modelling approaches for predicting metabolism could have an increasingly important role as a possible alternative to in vitro metabolism studies. In silico approaches to predicting metabolism can be divided into QSAR and 3D-QSAR [27] studies, protein and pharmacophore models [28, 95, 96] and predictive databases [4]. The most rationally rewarding molecular modelling studies are those based on the crystal structure of the metabolizing enzymes (different subtypes of cytochrome P450, e.g., CYP2B4 [97], see Fig. (1)). For the prediction of the metabolism different approaches, like pharmacophore modelling, 2D-quantitative structure metabolism relationships (2D-QSMR), 3D-QSMR, and non-linear pattern recognition techniques such as ANN and SVMs for modelling metabolism by UDP-glucoronosyltransferase (UGT) have been reported [98]. The cytochrome P450 enzymes (CYP’s), and the UGT’s are quantitatively (more than 90%) the most important functionalization (Phase I) and conjugation (Phase II) enzymes involved in the metabolism of xenobiotics [98]. During the recent years several reviews have been published discussing numbers of proposed QSAR approaches and models for the prediction of metabolism [3, 99-121]. Computational approaches of modelling CYP-catalyzed biotransformation and metabolism have advanced in parallel with the increasing availability of CYP isoform and the knowledge about substrate and inhibitor selectivity. The development of such models has also been aided by access to homology models of human CYP’s. But the improve-
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 341
ment of the models for predicting metabolism and for the characterizing structural features of substrates for UGT isoforms is less advanced relative to CYP [98].
Fig. (1). The three-dimensional structure (PDB code 1suo, www.pdb.org) of mammalian cytochrome P450 2B4 complexed with 4-(4-chlorophenyl) imidazole (zoomed in box, showing heme and the ligand) [97]. The figure was generated using PyMol (www.pymol.org) and rendered with RayTrace. Red dots are water molecules inside the crystal.
TOXICITY PREDICTION In recent years the pharmaceutical, agrochemical and personal product companies are producing large numbers of new, effective products at the same time as reducing significant amount of time and costs during the development processes. Taking advantages of the combinatorial chemistry and high-throughput screening (HTS), the numbers of new candidate molecules coming out of the discovery line has been increasing significantly, which ultimately produced vast requirements for faster screening of the toxicological properties of those molecules. It is not unpredictable, the in silico approaches for the prediction of toxicity offer attractive solutions to this problem because of their ability to screen large numbers of molecules sometimes even before the synthesis – virtually [122].
342 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Mahmud Tareq Hassan Khan
There are a large number of computer softwares, like DEREK (Deductive Estimation of Risk from Existing Knowledge, originally devised at Schering Agrochemicals and currently being marketed by LHASA Ltd.), OncoLogic (developed and marketed by LogiChem Inc.), HazardExpert (produced by CompuDrug), COMPACT (ComputerOptimized Molecular Parametric Analysis of Chemical Toxicity), CASE (Computer Automated Structure Evaluation; developed by MultiCASE and marketed by Charles River), TOPKAT (TOxicity Prediction by Komputer Assisted Technology, originally developed by Health Designs and now developed and marketed by Accelrys Ltd.), and more recent software applications such as QSARIS and TOXSYS developed and marketed by Scivision, etc., are available for the prediction of toxicity for large molecular datasets. Most of them are commercial, but good news is some of them are also available free for academic users in limited size. Some compounds released into the environment by human activities can modulate endogenous hormone activities and have been termed endocrine-disrupting compounds (EDCs). It has been hypothesized that such compounds may elicit a variety of adverse effects in both humans and wildlife including promotion of hormone dependent cancers, reproductive tract disorders, and a reduction in reproductive fitness [123] There are a number of receptor-mediated hormonal responses to toxicity. This phenomenon is referred to as receptor-mediated toxicity. These responses include xenobiotic effects on thyroid hormone receptor, epidermal growth factor (EGF) receptor, [124,125] Aryl hydrocarbon receptor (AhR) [126-128] as well as effects mediated by the estrogen receptor (ER) and the androgen receptor (AR) [129-133]. During 2005 Lill et al., demonstrated that they were able to both recognize the toxic compounds substantially different from those used in the training set as well as to classify harmless compounds clearly as being non-toxic [129]. The problem of predicting the toxicity of chemicals is, however, quite different from that of modeling medicinal properties. In contrast to lead optimization studies, toxicity testing is motivated by social concerns for human health and safety. Therefore, the available data pool for modeling chemical toxicity tends to be both structurally and mechanistically diverse [134, 135]. The primary goal of chemical toxicity prediction is to distinguish between toxicologically active and inactive compounds. Typically, multiple mechanisms can lead to the same toxicological endpoint and therefore predictive models need to be able to distinguish multiple regions of activity amidst a mass of inactive chemical structures [122]. MODEL VALIDATION Model validation is an indispensable assignment while trying to building up a statistically legitimate, extrapolative and prognostic model [136, 137]. Cronin and Schultz, in 2003, declared following reasonably valuable “essentials” in relation to developing QSARs in toxicology[138]: 1. Well-defined and quantifiable target; 2. Chemical and biologically diverse data set divided into representative training set and separate and consistent test set; 3. Structural description that is constant with the modelled target;
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 343
4. Exploitation of suitable statistical approaches; 5. Sturdy mechanistic basis, where possible. Additional potential “essentials” incorporate the utilities of jack-knifing, i.e. removing objects/compounds that do not fit the model (outliers), or cross-validation in conjunction with variable selection during model development [137] The correct identification of outliers is a long and well-discussed issue that has been debated for many years [139, 140]. CONCLUSION Proper approaches that can be relied upon to predict accurate performance in humans is not existed, and many important decisions have been made using tools whose capabilities could not be verified until candidates went to clinical trial, leading to the high failure rates historically observed. However, with the sequencing of the human genome, advances in proteomics, the anticipation of the identification of a vastly greater number of potential targets for drug discovery, and the potential of pharmacogenomics to require individualized evaluation of drug kinetics as well as drug effects, there is an urgent need for rapid and accurately computed pharmacokinetic properties [2]. Contrast to in vitro or in vivo experimental methods, the in silico approaches have the advantage that they do not necessitate compound synthesis. They can therefore be applied to “virtual” compounds permitting the rapid exclusion of likely failures at the “drawing-board” stage [77, 141]. Evidently, in vivo studies supply the most physiologically significant test system and assimilate processes ongoing in individual tissues. Indeed, the ability to translate data to in vivo from in vitro models in preclinical species may validate such extrapolation or projection from human in vitro test systems to man [142]. Additional improvement is considered necessary for a molecule be shown to be either a hERG [143] or P450 inhibitor [144] due to the possibility of cardiac arrhythmia or drug-drug interactions, respectively [5]. Although the early forecast of ADMET properties from the structure of the drug candidates is an imperative ambition to reach, it can significantly reduce the cost of the drug development process. However, the ADMET is composed of extremely complex features arising from several complex physiological processes [6, 12]. For dealing with the vast biological complexities it is very crucial with large sets of available high quality and reproducible experimental data, both in vitro and in vivo. Larger dataset can reduce the error of the prediction models. The diversity of the chemical structures is also very important. From highly diverse compound sets, it is more difficult to predict and propose good quality QSAR models. In a review Van de Waterbeemd and Gifford stated that, “Good predictive models for ADME parameters depend crucially on selecting the right mathematical approach, the right molecular descriptors for the particular ADMET endpoint, and a sufficiently large set of experimental data relating to this endpoint. In particular, more needs to be learnt about how the size of the training set influences the choice of the most capable model” [4]. Progressively, there is a requirement for ‘global’ or generalizable ADMET models [4, 98, 145]. The qualitative and quantitative interspecies differences in the regulation, expression and functional activity of key ADMET processes puzzle extrapolation from animals to
344 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Mahmud Tareq Hassan Khan
man. Nevertheless, complementary screens using animal and human material may assist the interpretation of safety assessment findings and help project the risk for early human studies [142]. Some Important Definitions of the Terms, Used in QSAR Modeling (Taken from Khan and Sylte, 2007 [11]) In the following some of the definitions and most important terms which are often used in QSAR modellings are given. The definitions of the terms are taken from Todeschini and Consonni’s Handbook of Molecular Descriptors [16]. Coefficient of determination, R2. The squared multiple correlation coefficient is the percent of total variance of the response explained by a regression model. This can be calculated from the model sum of squares MSS or from the residual sum of squares RSS: n
( yi yˆi )
2
R2 =
MSS RSS = 1 = 1 i =1 n TSS TSS
( yi yi )
2
i =1
where TSS is the total sum of squares around the mean. A value of one indicates perfect fit, i.e., a model with zero error terms. Residual mean square, RMS or s2. The estimated s2 of the error variance 2, defined as:
s
2
=
RSS dfE
where RSS is the residual sum of squares and dfE is the error degrees of freedom. Adjusted R2, adj. R2 or R2adj. A parameter adjusted for the degrees of freedom, so that it can be used for comparing models with different numbers of predictor variables:
R
2 adj
= 1
RSS / dfE = 1 1 TSS / dfT
2 n 1
R . n p
where RSS and TSS is the residual and total, respectively, sum of the squares; and R2 is the coefficient of determination. Predictive residual sum of the squares, PRESS. The sum of squared difference between the observed and estimated response by validation techniques: n
PRESS =
y yˆ i =1
where
yˆ
i
i /i
denotes the response of the i th object estimated by using a model obtained i /i
without using the i th object. Cross-validated R2, R2cv (or Q2). The explained variance in prediction:
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 345 n
2
R Q cv
2
= 1
y yˆ
PRESS = 1 i =1 n TSS
i
i /i
y y i =1
i
2
2
where PRESS is the predictive residual sum of the squares and TSS is the total sum of squares. Standard deviation error of prediction, SDEP. This is also called standard error in prediction, SEP. n
SDEP SEP =
y yˆ i =1
i
n
i /i
2
=
PRESS = PSE n
where PSE is the predictive squared error, which is the average of the predictive error sum of squares. Suggested Further Readings (taken from Khan and Sylte, 2007 [11]) Balaban, A.T. (Ed.), Chemical Applications of Graph Theory. Polytechnic Univ., Bucharest, Romania, 1976. Balaban, A.T. (Ed.), From Chemical Topology to Three-Dimensional Geometry. Polytechnic Univ., Bucharest, Romania Plenum Press, 1997. Bonchev, D. Information Theoretic Indices for Characterization of Chemical Structures. Virginia Commonwealth University, Research Studies Press, Richmond, USA, 1983. Devillers, J. and Balaban, A.T. (Eds.), Topological Indices and Related Descriptors in QSAR and QSPR, Taylor & Francis, 2000. Diudea, M.V. (Ed.), QSPR/QSAR Studies by Molecular Descriptors. University BabesBolya, Cluj-Napoca, Romania, Nova Science, 2001. Diudea, M.V., Florescu, M.S. and Khadikar, P.V. Molecular Topology and its Applications. EfiCon Press, 2006. Diudea, M.V., Gutman, I. and Lorentz, J. Molecular Topology, Nova Science, 2001. Karelson, M. Molecular Descriptors in QSAR/QSPR. Department of Chemistry, University of Tartu, Estonia, Wiley-Interscience, 2000. Kier, L. and Hall, L. Molecular Structure Description. USA, Academic Press, 1999. Kier, L. Molecular Connectivity in Structure-Activity Analysis. Virginia Commonwealth University, Research Studies Press, Richmond, USA, 1986. King, R.B. and Rouvray, D.H. (Eds.), Graph Theory and Topology in Chemistry. University of Georgia, Athens, Georgia, USA, Elsevier, 1987. King, R.B. and Rouvray, D.H. (Eds.), Topology in Chemistry: Discrete Mathematics of Molecules. University of Georgia, Athens, Georgia, USA, Albion/Horwood Pub., 2002.
346 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Mahmud Tareq Hassan Khan
Li, X. and Gutman, I. Mathematical Aspects of Randic-Type Molecular Structure Descriptors. University of Kragujevac, Kragujevac, Serbia, 2006. Mezey, P.G. Shape in Chemistry: An Introduction to Molecular Shape and Topology, University of Saskatchewan, Canada, John Wiley & Sons, 1993. Rouvray, D.H. and Bonchev, D. (Eds.), Chemical Graph Theory: Introduction and Fundamentals. Taylor & Francis, 1991. Rouvray, D.H. and Bonchev, D. (Eds.), Complexity: Introduction and Fundamentals, USA, CRC Press, 2003. Todeschini, R.; Consonni, V., Handbook of Molecular Descriptors. Methods and Principles in Medicinal Chemistry. Mannhold, R., Kubinyi, H., and Timmerman, H. (Eds.), Vol. 11, 2000. Trinajstic, N. Chemical Graph Theory. Rugjer Boskovic Institute, Zagreb, Croatia, CRC Press, 1992. ABBREVIATIONS ADMET
=
Absorption, distribution, metabolism, excretion and toxicity
AhR
=
Aryl hydrocarbon receptor
ANN
=
Artificial neural network
AR
=
Androgen receptor
BBB
=
Blood-brain barrier
CAS
=
Chemical abstracting services
CNS
=
Central nervous system
CYP
=
Cytochrome P450
dfE
=
Error degrees of freedom
EDC
=
Endocrine-disrupting compound
EGF
=
Epidermal growth factor
ER
=
Estrogen receptor
GA
=
Genetic algorithms
HCTPSA
=
High-charged topological polar surface area
HTS
=
High-throughput screening
ILP
=
Inductive logic programming
kNN
=
k-Nearest neighbor
LDA
=
Linear discriminant analysis
MD
=
Molecular descriptor
MLR
=
Multiple linear regressions
MSS
=
Model sum of squares
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 347
MW
=
Molecular weight
NCE
=
New chemical entities
PCA
=
Principal component analysis
PLS
=
Partial least squares
PRESS
=
Predictive residual sum of the squares
PSA
=
Polar surface area
PSE
=
Predictive squared error
QSAR
=
Quantitative structure-activity relationship
QSMR
=
Quantitative structure-metabolism relationships
QSPR
=
Quantitative structure-property (or permeability) relationships
ROF
=
Rule of five
RSS
=
Residual sum of squares
SEP
=
Standard error in prediction
SVP
=
Support vector machines
TPSA
=
Topological polar surface area
TSS
=
Total sum of squares
UGT
=
UDP-glucoronosyltransferase
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24]
Drews, J. Science, 2000, 287, 1960-4. Grass, G.M.; Sinko, P.J. Adv. Drug Deliv. Rev.,2002, 54, 433-51. Lombardo, F.; Gifford, E.; Shalaeva, M.Y. Mini Rev. Med. Chem., 2003, 3, 861-75. van de Waterbeemd, H.; Gifford, E. Nat. Rev. Drug Discov., 2003, 2, 192-204. Gleeson, M.P. J. Med. Chem., 2008, 51, 817-34. Hou, T.; Wang, J.; Zhang, W.; Wang, W.; Xu, X. Curr. Med. Chem. 2006, 13, 2653-67. Kennedy, T. Drug Discov. Today, 1997, 2, 436-44. Hou, T.; Xu, X. Curr. Pharm. Des., 2004, 10, 1011-33. Hou, T.; Wang, J.; Zhang, W.; Xu, X. J. Chem. Inf. Model., 2007, 47, 208-18. Lipinski, C.A.; Lombardo, F.; Dominy, B.W.; Feeney, P.J. Adv. Drug Deliv. Rev., 1997, 23, 3-25. Khan, M.T.; Sylte, I. Curr. Drug Discov. Technol., 2007, 4, 141-9. Hansch, C.; Leo, A.; Mekapati, S.B.; Kurup, A. Bioorg. Med. Chem., 2004, 12, 3391-400. Lobell, M.; Sivarajah, V. Mol. Divers., 2003, 7, 69-87. Hashida, M. Yakugaku Zasshi, 2005, 125, 853-61. Hall, L.H. Chem. Biodivers., 2004, 1, 183-201. Todeschini, R.; Consonni, V. Handbook of Molecular Descriptors. Methods and Principles in Medicinal Chemistry. Mannhold, R., Kubinyi, H., Timmerman, H., Eds. 2000, 11, 368-371. Hou, T.J.; Xu, X. J. J. Chem. Inf. Comput. Sci., 2003, 43, 2137-52. Anonymous, Chemical Abstracting Services. (www.cas.org/infopro/). Gleeson, M.P. J. Med. Chem., 2007, 50, 101-12. Narayanan, R.; Gunturi, S.B. Bioorg. Med. Chem., 2005, 13, 3017-28. Mitic, L.L.; Anderson, J.M. Annu. Rev. Physiol., 1998, 60, 121-42. Artursson, P.; Ungell, A.L.; Lofroth, J.E. Pharm. Res., 1993, 10, 1123-9. Artursson, P. Crit. Rev. Ther. Drug Carrier Syst., 1991, 8, 305-30. Artursson, P.; Karlsson, J. Biochem. Biophys. Res. Commun., 1991, 175, 880-5.
348 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70]
Mahmud Tareq Hassan Khan
Ekins, S.; Mankowski, D.C.; Hoover, D.J.; Lawton, M.P.; Treadway, J.L.; Harwood, H.J., Jr. Drug Metab. Dispos., 2007, 35, 493-500. Ekins, S.; Andreyev, S.; Ryabov, A.; Kirillov, E.; Rakhmatulin, E.A.; Bugrim, A.; Nikolskaya, T. Opin. Drug Metab. Toxicol., 2005, 1, 303-24. Ekins, S.; Bravi, G.; Binkley, S.; Gillespie, J.S.; Ring, B.J.; Wikel, J.H.; Wrighton, S.A. Drug Metab. Dispos., 2000, 28, 994-1002. Ekins, S.; de Groot, M. J.; Jones, J.P. Drug Metab. Dispos., 2001, 29, 936-44. Hoffman, B.; Cho, S.J.; Zheng, W.; Wyrick, S.; Nichols, D.E.; Mailman, R.B.; Tropsha, A. J. Med. Chem., 1999, 42, 3217-26. Hirst, J.D.; King, R.D.; Sternberg, M.J. J. Comput. Aided Mol. Des., 1994, 8, 421-32. Hirst, J.D.; King, R.D.; Sternberg, M.J. J. Comput. Aided Mol. Des., 1994, 8, 405-20. King, R.D.; Muggleton, S.; Lewis, R.A.; Sternberg, M.J. Proc. Natl. Acad. Sci. USA, 1992, 89, 113226. Jensen, B.F.; Vind, C.; Padkjaer, S.B.; Brockhoff, P.B.; Refsgaard, H.H. J. Med. Chem., 2007, 50, 501-11. Zheng, W.; Tropsha, A. J. Chem. Inf. Comput. Sci., 2000, 40, 185-94. Yuan, H.; Wang, Y.; Cheng, Y. J. Chem. Inf. Model., 2007, 47, 159-69. Ajmani, S.; Jadhav, K.; Kulkarni, S.A. J. Chem. Inf. Model., 2006, 46, 24-31. Hoshi, K.; Kawakami, J.; Kumagai, M.; Kasahara, S.; Nishimura, N.; Nakamura, H.; Sato, K. Chem. Pharm. Bull. (Tokyo), 2005, 53, 1570-4. Fernandez, M.; Caballero, J. J. Mol. Graph. Model., 2006, 25, 410-22. Fernandez, M.; Caballero, J.; Tundidor-Camba, A. Bioorg. Med. Chem., 2006, 14, 4137-50. Hou, T.J.; Xia, K.; Zhang, W.; Xu, X.J. J. Chem. Inf. Comput. Sci., 2004, 44, 266-75. Yalkowsky, S.H. Chemosphere, 1993, 26, 1239-61. Jain, N.; Yalkowsky, S.H. J. Pharm. Sci., 2000, 90, 234-52. Ran, Y.; Jain, N.; Yalkowsky, S.H. J. Chem. Inf. Comput. Sci., 2001, 41, 1208-17. Ran, Y.; Yalkowsky, S. H. J. Chem. Inf. Comput. Sci., 2001, 41, 354-7. Huuskonen, J.; Salo, M.; Taskinen, J. J. Chem. Inf. Comput. Sci., 1998, 38, 450-6. Huuskonen, J. J. Chem. Inf. Comput. Sci., 2000, 40, 773-7. McElroy, N.R.; Jurs, P.C. J. Chem. Inf. Comput. Sci. 2001, 41, 1237-47. McFarland, J.W.; Avdeef, A.; Berger, C.M.; Raevsky, O.A. J. Chem. Inf. Comput. Sci., 2001, 41, 1355-9. Liu, R.; So, S.S. J. Chem. Inf. Comput. Sci., 2001, 41, 1633-9. Tetko, I.V.; Tanchuk, V.Y.; Kasheva, T.N.; Villa, A.E. J. Chem. Inf. Comput. Sci., 2001, 41(6), 148893. Tetko, I.V.; Tanchuk, V.Y.; Kasheva, T.N.; Villa, A.E. J. Chem. Inf. Comput. Sci., 2001, 41, 246-52. Engkvist, O.; Wrede, P. J. Chem. Inf. Comput. Sci., 2002, 42, 1247-9. Yan, A.; Gasteiger, J. J. Chem. Inf. Comput. Sci., 2003, 43, 429-34. Butina, D.; Gola, J.M. J. Chem. Inf. Comput. Sci., 2003, 43, 837-41. Wegner, J. K.; Zell, A. J. Chem. Inf. Comput. Sci., 2003, 43, 1077-84. Nirmalakhandan, N.N.P.; Speece, R.E. Environ. Sci. Technol., 1989, 23, 708-13. Suzuki, T. J. Comput. Aided Mol. Des., 1991, 5, 149-66. Kuhne, R.; Ebert, R.U.; Kleint, F.; Schmidt, G.; Schuurmann, G. Chemosphere, 1995, 30, 2061-77. Lee, Y.; Myrdal, P.B.; Yalkowsky, S.H. Chemosphere, 1996, 33, 2129-44. Klopman, G.; Wang, S.; Balthasar, D.M. J. Chem. Inf. Comput. Sci., 1992, 32, 474-82. Klopman, G.; Zhu, H. J. Chem. Inf. Comput. Sci., 2001, 41, 439-45. Bouwstra, J.A.; Honeywell-Nguyen, P.L. Adv. Drug Deliv. Rev., 2002, 54(Suppl 1), S41-55. Honeywell-Nguyen, P.L.; de Graaff, A.M.; Groenink, H.W.; Bouwstra, J.A. Biochim. Biophys. Acta, 2002, 1573, 130-40. Honeywell-Nguyen, P.L.; Frederik, P.M.; Bomans, P.H.; Junginger, H.E.; Bouwstra, J.A. Pharm. Res., 2002, 19, 991-7. Bouwstra, J.A.; Salomons de Vries, M.A.; van den Bergh, B.A. I.; Gooris, G.S. Int. J. Pharm., 1996, 144(1), 81-9. Lian, G.; Chen, L.; Han, L. J. Pharm. Sci., 2008, 97, 584-98. Prausnitz, M.R.; Mitragotri, S.; Langer, R. Nat. Rev. Drug Discov., 2004, 3, 115-24. Thiele, J.J.; Hsieh, S.N.; Ekanayake-Mudiyanselage, S. Dermatol. Surg., 2005, 31, 805-13. Estrada, E.; Uriarte, E.; Gutierrez, Y.; Gonzalez, H. SAR QSAR Environ. Res., 2003, 14, 145-63. de Vries, H.E.; Kuiper, J.; de Boer, A.G.; Van Berkel, T.J.; Breimer, D.D. Pharmacol. Rev., 1997, 49, 143-55.
Advances in ADMET Predictions and Modelling Frontiers in Drug Design & Discovery, 2009, Vol. 4 349 [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117]
Subramanian, G.; Kitchen, D.B. J. Comput. Aided Mol. Des., 2003, 17, 643-64. Zhao, Y.H.; Abraham, M.H.; Ibrahim, A.; Fish, P. V.; Cole, S.; Lewis, M.L.; Groot, M.J.; Reynolds, D.P. J. Chem. Inf. Model., 2007, 47, 170-5. Reichel, A.; Begley, D.J. Pharm. Res., 1998, 15, 1270-4. Norinder, U.; Haeberlain, M. Adv. Drug Deliv. Rev., 2002, 54, 291-313. Lobell, M.; Molnar, L.; Keseru, G.M. J. Pharm. Sci., 2003, 92, 360-70. Goodwin, J.T.; Clark, D.E. J. Pharmacol. Exp. Ther., 2005, 315, 477-83. Clark, D.E.; Grootenhuis, P.D. Curr. Opin. Drug Discov. Devel., 2002, 5, 382-90. Young, R.C.; Mitchell, R.C.; Brown, T.H.; Ganellin, C.R.; Griffiths, R.; Jones, M.; Rana, K.K.; Saunders, D.; Smith, I.R.; Sore, N.E. J. Med. Chem., 1988, 31, 656-71. Lombardo, F.; Blake, J.F.; Curatolo, W.J. J. Med. Chem.,1996, 39, 4750-5. Abraham, M.H.; Takacs-Novak, K.; Mitchell, R.C. J. Pharm. Sci., 1997, 86, 310-5. Norinder, U.; Sjoberg, P.; Osterberg, T. J. Pharm. Sci., 1998, 87, 952-9. Clark, D.E. J. Pharm. Sci., 1999, 88, 815-21. Kelder, J.; Grootenhuis, P.D.; Bayada, D.M.; Delbressine, L.P.; Ploemen, J.P. Pharm. Res., 1999, 16, 1514-9. Crivori, P.; Cruciani, G.; Carrupt, P.A.; Testa, B. J. Med. Chem., 2000, 43, 2204-16. Ertl, P.; Rohde, B.; Selzer, P. J. Med. Chem., 2000, 43, 3714-7. Keseru, G. M.; Molnar, L. J. Chem. Inf. Comput. Sci., 2001, 41, 120-8. Liu, R.; Sun, H.; So, S.S. J. Chem. Inf. Comput. Sci., 2001, 41, 1623-32. Rose, K.; Hall, L. H.; Kier, L.B. J. Chem. Inf. Comput. Sci., 2002, 42, 651-66. Hou, T.; Xu, X. J. Mol. Model., 2002, 8, 337-49. Hutter, M.C. J. Comput. Aided Mol. Des., 2003, 17, 415-33. Abraham, M.H. Eur. J. Med. Chem., 2004, 39, 235-40. Smith, D.A.; Van de Waterbeemd, H.; Walker, D.K. Pharmacokinetics and Metabolism in Drug Design, Wiley–VCH: Weinheim, Germany, 2001. Hall, L.M.; Hall, L.H.; Kier, L.B. J. Chem. Inf. Comput. Sci., 2003, 43, 2120-8. Hall, L.M.; Hall, L.H.; Kier, L.B. J. Comput. Aided Mol. Des., 2003, 17, 103-18. de Groot, M.J.; Ackland, M.J.; Horne, V.A.; Alex, A.A.; Jones, B.C. J. Med. Chem., 1999, 42, 406270. de Groot, M.J.; Ackland, M.J.; Horne, V.A.; Alex, A.A.; Jones, B.C. J. Med. Chem., 1999, 42, 151524. Scott, E.E.; White, M.A.; He, Y.A.; Johnson, E.F.; Stout, C.D.; Halpert, J.R. J. Biol. Chem., 2004, 279, 27294-301. Smith, P.A.; Sorich, M.J.; Low, L.S.; McKinnon, R.A.; Miners, J.O. J. Mol. Graph. Model., 2004, 22, 507-17. Sakurai, A.; Kurata, A.; Onishi, Y.; Hirano, H.; Ishikawa, T. Expert Opin. Drug Saf., 2007, 6, 71-86. Chohan, K.K.; Paine, S.W.; Waters, N.J. Curr. Top. Med. Chem., 2006, 6, 1569-78. Srinivas, E.; Murthy, J.N.; Rao, A.R.; Sastry, G.N. Curr. Drug Metab., 2006, 7, 205-17. Yang, G.F.; Huang, X. Curr. Pharm. Des., 2006, 12, 4601-11. Arimoto, R. Curr. Top. Med. Chem., 2006, 6, 1609-18. Allen, D.D.; Geldenhuys, W.J. Life Sci., 2006, 78, 1029-33. Cronin, M.T.; Walker, J.D.; Jaworska, J.S.; Comber, M.H.; Watts, C. D.; Worth, A.P. Environ. Health Perspect., 2003, 111, 1376-90. Ekins, S.; Wrighton, S.A. J. Pharmacol. Toxicol. Methods, 2001, 45, 65-9. Gao, H.; Hansch, C. Drug Metab. Rev., 1996, 28, 513-26. Hansch, C.; Mekapati, S.B.; Kurup, A.; Verma, R.P. Drug Metab. Rev., 2004, 36, 105-56. Kulkarni, S.A.; Zhu, J.; Blechinger, S. Xenobiotica, 2005, 35, 955-73. Lewis, D.F. Scientific WorldJournal, 2003, 3, 536-45. Lewis, D.F.; Modi, S.; Dickins, M. Drug Metab. Drug Interact., 2001, 18, 221-42. Li,Y.; Harte, W.E. Curr. Pharm. Des., 2002, 8, 99-110. Locuson, C.W.; Wahlstrom, J.L. Drug Metab. Dispos., 2005, 33, 873-8. Mekenyan, O.G.; Dimitrov, S.D.; Pavlov, T.S.; Veith, G.D. Curr. Pharm. Des., 2004, 10, 1273-93. Rettie, A.E.; Jones, J.P. Annu. Rev. Pharmacol. Toxicol., 2005, 45, 477-94. Siraki, A.G.; Chevaldina, T.; Moridani, M.Y.; O'Brien, P.J. Curr. Opin. Drug Discov. Devel., 2004, 7, 118-25. Soffers, A.E.; Boersma, M.G.; Vaes, W.H.; Vervoort, J.; Tyrakowska, B.; Hermens, J.L.; Rietjens, I. M. Toxicol. In Vitro, 2001, 15, 539-51.
350 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128]
[129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145]
Mahmud Tareq Hassan Khan
van de Waterbeemd, H., High-throughput and in silico techniques in drug metabolism and pharmacokinetics. Curr. Opin. Drug Discov. Devel., 2002, 5, (1), 33-43. Verma, R.P.; Hansch, C. Bioorg. Med. Chem., 2005, 13, 4597-621. Wang, R.B.; Kuo, C.L.; Lien, L.L.; Lien, E.J. J. Clin. Pharm. Ther., 2003, 28, 203-28. Yamashita, F.; Hashida, M. Drug Metab. Pharmacokinet., 2004, 19, 327-38. Greene, N. Adv. Drug Deliv. Rev., 2002, 54, 417-31. Zacharewski, T. Environ. Sci. Technol., 1997, 31, 613-23. Madhukar, B.V.; Brewster, D.W.; Matsumura, F. Proc. Natl. Acad. Sci. USA, 1984, 81, 7407-11. Matsumura, F.; Brewster, D.W.; Madhukar, B.V.; Bombick, D.W. Arch. Environ. Contam. Toxicol., 1984, 13, 509-15. Gustaffson, J.A. Toxicol. Lett., 1995, 135, 465-70. Okey, A.B.; Riddick, D.S.; Harper, P.A. Toxicol. Lett., 1994, 70, 1-22. Safe, S.; Krishnan, K. Cellular and molecular biology of Aryl hydrocarbon (Ah) receptor-mediated gene expression - In Archives of Toxicology, Degen, D. H., Seiler, J.P., Bentley, P., Eds. Springer: Berlin, 1995; Vol. (Suppl 17), pp. 116-24. Lill, M. A.; Dobler, M.; Vedani, A. SAR QSAR Environ. Res., 2005, 16, 149-69. Vedani, A.; Dobler, M.; Dollinger, H.; Hasselbach, K.M.; Birke, F.; Lill, M.A. J. Med. Chem., 2005, 48, 1515-27. Vedani, A.; Dobler, M.; Lill, M.A. ALTEX, 2005, 22, 123-34. Vedani, A.; Dobler, M.; Lill, M.A. Toxicol. Appl. Pharmacol., 2005, 207, 398-407. Vedani, A.; Dobler, M.; Lill, M.A. J. Med. Chem., 2005, 48, 3700-3. Benigni, R.; Richard, A.M. Methods, 1998, 14, 264-76. Benigni, R.; Richard, A. Methods Enzymol., 1998, 14, 264-76. Matsson, P.; Bergstrom, C.A.; Nagahara, N.; Tavelin, S.; Norinder, U.; Artursson, P. J. Med. Chem., 2005, 48, 604-13. Norinder, U. SAR QSAR Environ. Res., 2005, 16, 1-11. Cronin, M.T.D.; Schultz, T.W. J. Mol. Struct., 2003, 622, 39-51. Massart, D.L.; Vandeginste, B.G.M.; Buydens, L.M.C.; De Jong, S.; Lewi, P.J.; Smeyers-Verbeke, J. Handbook of Chemometrics and Qualimetrics: Part A. Elsevier: Amsterdam, 1997. Martens, H.; Naes, T., Multivariate Calibration. Wiley: Chichester, 1989. Clark, D. E. Adv. Drug Deliv. Rev., 2002, 54, 253-4. Riley, R.J.; Kenna, J.G. Curr. Opin. Drug Discov. Dev., 2004, 7, 86-99. Viskin, S. Lancet, 1999, 354, 1625-33. Rettie, A.E.; Korzekwa, K.R.; Kunze, K.L.; Lawrence, R.F.; Eddy, A.C.; Aoyama, T.; Gelboin, H.V.; Gonzalez, F. J.; Trager, W. F. Chem. Res. Toxicol., 1992, 5, 54-9. Buchwald, P.; Bodor, N. Drugs Future, 2002, 27, 577-88.
Frontiers in Drug Design & Discovery, 2009, 4, 351-377
351
Computational Intelligence Methods for ADMET Prediction David Hecht1,* and Gary B. Fogel2 Southwestern College, 900 Otay Lakes Rd., Chula Vista, CA 91910, USA and Natural Selection, Inc., 9330 Scranton Rd., San Diego, CA 92121, USA Abstract: Quantitative structure-property relationship (QSPR) models have proven to be an effective approach for increasing the efficiency of small molecule drug discovery and development processes. Despite their importance to drug discovery, difficulties remain in the appropriate selection and weighting of descriptors, determination of appropriate descriptor combinations, and optimization strategies that can increase the value of QSPR models. Here we review the utility of some of the more popular applications of computational intelligence to QSPR modeling including: artificial neural networks, fuzzy logic, and evolutionary computing.
Key Words: Computational intelligence, evolutionary algorithms, artificial neural networks, fuzzy logic, machine learning, support vector machines, QSPR, ADME-tox, high-throughput screening, virtual screening. 1. INTRODUCTION The discovery and development of a new drug often takes 12-15 years to bring to market at a cost of more than $1.3B [1-3]. For every 5,000-10,000 compounds screened, only 250 enter preclinical testing. Of these, only 5 will survive to enter clinical testing with only 1 approved drug by the U.S. Food and Drug Administration (FDA) after an average of 15 years of total research and development [2, 4]. Only 2 out of every 10 approved and marketed drugs recover their research and development costs [2]. It has been estimated that more than 75% of the high cost of drug discovery and development is actually spent on compounds that fail later in the more costly portions of the drug development process (e.g., during clinical development) [5]. In fact, for the very expensive Phase III clinical trials, only half of compounds tested end up being approved [6]. During the 1980s and 1990s, roughly 40% of the failures during clinical trials were attributed to the absorption, distribution, metabolism, and excretion (ADME) properties of the clinical candidates [7, 8]. A more detailed analysis of Phase I failures indicated that during this period 33% failed for lack of efficacy, 9% for market reasons, 18% for toxicity and adverse events, and 40% for poor pharmacokinetic and ADME properties [9, 10]. *Corresponding Author: Tel: (619) 421-6700; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
352 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
In light of these data, in the late 1990s pharmaceutical and biotechnology companies realized the importance of optimizing absorption, distribution, metabolism, excretion as well as toxicological (ADMET) properties in early phases of the drug discovery and development process [9,11]. Traditionally, compounds were first identified through screening and then optimized for potency and specificity for molecular targets with in vitro enzyme/binding assays or in vivo cell-based assays. Optimization of ADMET properties was reserved for pre-clinical and clinical development. A typical clinical development process is shown below in Fig. (1).
Fig. (1). A “typical” clinical development workflow from the initial new drug (IND) filing through the three phases of clinical trials culminating in the new drug application (NDA) and market launch. At each stage of this process drug candidates are eliminated as the costs increase exponentially.
Unfortunately the very same physico-chemical properties that were optimized for potency in new leads discovery and optimization often resulted in poor ADMET profiles. For example, lead compounds that were optimized for high molecular weight and
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 353
increased lipophilicity tended to have high potency but poor absorption. These observations lead to the development of “filters” that could be used to select for more “druglike” characteristics of small molecules. Perhaps the most famous of these is Lipinski’s “Rule of Five” [12]. Caution has to be exercised when using these filters blindly as there are examples in the literature of very successful drugs that fail one or more filter criteria [13]. In order to avoid spending hundreds of millions of dollars for compounds that fail in late development, the paradigm quickly switched to elimination of high-risk compounds in early phases. In silico modeling in addition to high-throughput in vitro screening was very quickly and widely adopted [14-16]. These included various property predictions as well as quantitative structure-property relationship (QSAR) and quantitative structureproperty relationship (QSPR) models [5, 17-20]. As a testament to the successful widespread adoption of pre-clinical ADME screening and testing, the clinical failure rate due to ADME has dropped to 10-14% in 2008 [7,9,10]. An analysis of recent Phase I failures indicated that 36% failed for lack of efficacy; 7% for “other,” 43% for toxicity and adverse events, and only 14% for poor pharmacokinetic and ADME properties. Currently toxicity and lack of efficacy are the main causes for failure [10]. Even more troublesome is the observation that greater than 90% of recent market withdrawals have been due to toxicity causing adverse events and side effects in patients [7, 9, 10]. QSAR and QSPR models have proven to be an effective approach for handling the massive quantities of structural and biological data generated with combinatorial libraries and HTS in lead discovery, lead optimization, and drug development [21]. QSAR/QSPR models are essentially a function relating parameters/descriptors/features based on physicochemical properties of small molecule compounds to a biological response. These descriptors are quite easy to calculate for small molecules. However, only a fraction of the descriptors are truly useful for predicting activity or other properties. In addition, some descriptors that are not very useful on their own, may be very informative when in combination with other descriptors. Modeling approaches that can relate the appropriate selection and weighting of descriptors in automated, improved, and efficient ways is a very active area of research. One rather new development has been the interest in applying existing tools and techniques from the field of computational intelligence [22]. Computational intelligence (CI) is a broad field of computer science that makes use of nature-inspired modeling paradigms for optimization and pattern recognition. These approaches, such as artificial neural networks (ANNs), fuzzy logic, evolutionary computation (EC), can be used in addition to other machine learning methods to automatically select, analyze and interpret relevant data and information [23-27]. Here we review the utility of some of the more popular applications of computational intelligence to QSPR modeling. This paper is organized as follows. In the next section ADME and toxicological properties are briefly summarized. The subsequent section briefly describes current QSAR and QSPR technologies. An additional section introduces general concepts of computational intelligence and the section following presents examples of their application to QSPR modeling. The final section discusses some of the future challenges for ADMEtox modeling.
354 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
2. ADMET Pharmacokinetics is the study of what happens to drugs as they are administered and pass through the body. Oral administration is the preferred method for most drug development programs as it is convenient and helps to ensure compliance. An orally-administered drug dissolves in the stomach and/or intestine and passes through the intestinal wall. In many cases, drugs first enter the liver where they are processed by enzymes, the most important class of which are the cytochrome P450s. Metabolized drugs can often become quite toxic or have other negative effects on the body. Unprocessed drugs then exit via the portal vein. The percentage of the initial dose that reaches the circulatory system is called the “bioavailability” of the drug. The “half-life” or t of a drug is the time it takes for the blood plasma concentration of the drug to reach of its initial value. Serum albumin and other proteins often bind the drugs and prevent them from reaching their intended targets. A fraction of the unbound drug is then distributed to the various organs and tissues where it binds to its target and has its effect. Ultimately, the remainder is excreted and passes out of the body. Pharmacodynamics is the study of the effects a drug has on the body and on metabolism. These effects can be positive or negative. Positive effects include the desired effects of the drug. Negative effects can include genotoxicity. This section presents a brief introduction to adsorption, distribution, metabolism, excretion and toxicology. 2.1. Absorption In order for orally administered drugs to reach their intended molecular targets, they first need to dissolve in the stomach and/or small intestine and then pass through the epithelial cell layer in the intestine in order to get into the circulatory system. Drugs targeted for the central nervous system (CNS) need to pass through an additional barrier, the blood-brain barrier. Yet another barrier of interest is the skin – for topical administration and absorption. Drugs generally move across epithelial barriers, such as the one in the small intestine, via the paracellular pathway (between cells) or via the transcellular pathway (through cells). Generally, only small molecular weight compounds (<200 MW) are able to pass between epithelial cells via the paracellular pathway. Most drugs use transcellular pathways which often involve both active and passive transport processes [28]. Drugs, like other small molecules, can pass through cells and membranes via passive diffusion or via active transport processes. Passive diffusion processes generally involve one of two mechanisms: 1) the drug enters the cell membrane and diffuses from the apical to basolateral side of the cell within the cell membrane, or 2) the drug diffuses across the apical cell membrane, passes through the cytoplasm and exits through the basolateral cell membrane on the opposite side of the cell. Active transport processes include carrier-mediated and vesicular-transport mechanisms [29]. 2.1.1. In Vitro Models of Absorption Because of the importance of bioavailability, in vitro models of intestinal absorption have been developed and are used routinely in a “high-throughput” mode [30, 31]. These models use artificial membranes as well as cultured cell monolayers. The Caco-2 assay is perhaps the most common example [32]. Caco-2 cells are human intestinal epithelial
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 355
cells derived from a carcinoma and have many of the morphological and functional properties of the “normal” intestinal epithelial layer. In the assay, these cells are grown as a monolayer in a partitioned well. Compounds to be tested are placed on top of the monolayer. After 2-3 hours the amount of compound that is present in the bottom compartment is measured. These assays have been validated to have a sigmoidal relationship with human oral drug absorption [33]. Their accuracy is ~80-85% which is in a range useful for screening purposes [5]. Madin-Darby canine kidney (MDCK) cells are another commonly used cell line for studying membrane permeability [34]. The blood-brain barrier (BBB) separates the blood from the CNS including the brain. This barrier has its origins in the tight junctions between cells that prevent paracellular transport [30]. In general, hydrophobic molecules as well as relatively small polar molecules can cross this barrier via transcellular pathways. Larger polar molecules and hydrophilic molecules cannot pass this layer. As would be expected, important descriptors from in silico models of BBB penetration include: polar surface area (PSA), molecular size and shape, as well as hydrogen bonding [35]. 2.1.2. In Silico Models of Absorption QSPR models of absorption have primarily focused on passive transcellular transport [5, 17, 36]. As these model the movement of compounds across lipid bilayers, perhaps the most important property in these models is the “lipophilicity” or the degree to which a molecule prefers to be in a non-aqueous, non-polar environment. The octanol-water partition coefficient (logP) is often used a measure of lipophilicity [37]. In fact, the very first QSAR models used logP as a descriptor [38]. LogD is the octanol-water partition coefficient for a given pH. LogD is useful, as an orally administered drug will experience a wide range of pH values from <2 in the stomach to ~7.4 in blood plasma. PSA is another very commonly used descriptor in absorption models. PSA is the Van der Waals or solvent accessible surface area of a compound containing oxygen, nitrogen, hydrogen, or other polar atoms such as sulfur or phosphorous. Likewise, the nonpolar surface area (NPSA) is also a commonly used descriptor. Additional descriptors have also proven to be useful in absorption models. Some of these include: molecular weight; molar refractivity (MR); number of H-bond donors and acceptors; number of fused rings; number of rotatable bonds; basicity; polarizability; solubility; non-polar surface areas (NPSA) among others [5, 12, 39, 40]. In 1997, Lipinski and co-workers analyzed descriptors generated from 2245 compounds from the World Drug Index and came up with the Rule-of-Five [12]. The Ruleof-Five is commonly used as a filter on compound libraries in lead discovery in order to remove compounds that have: a)
molecular weight > 500
b)
calculated logP > 5
c)
number of hydrogen-bond donors (-OH and -NH groups) > 5
d)
number of hydrogen-bond acceptors (N and O atoms) > 10
Similar analyses have been performed that include additional descriptors such as: molar refractivity [39], counts of the number of rings, rotatable bonds, as well as hydrogen bond donors and acceptors [41, 42].
356 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
Recent analyses comparing drug candidates in development with marketed drugs concluded that larger, more lipophilic compounds tended to be identified from highthroughput screening of compound libraries in lead discovery [42, 43]. As compounds pass through the different stages of pre-clinical and clinical development, the mean molecular weight of drug candidates tends to converge to that of marketed drugs. Likewise, many promising lipophilic compounds are discontinued as development proceeds. Additionally it was determined that drugs developed for oral administration tended to be lighter and had fewer rotatable bonds and hydrogen bond acceptors and donors than drugs developed for other indications [43]. These observations explain a significant portion of the inefficiency and high-attrition rates of current drug discovery and development approaches. A major limitation of using predictive rules like the Rule-of-Five is that many of the “inactive” or “non-druggable” compounds in commercially available screening libraries also obey these rules. For example 68.7% of the compounds in the Available Chemical Directory (ACD) screening database (containing >2.4 million compounds) do not violate the Rule-of-Five [44]. In other words, simple ADME filters are not enough to eliminate “non-druglike” molecules from screening libraries. This often results in precious resources being spent on optimization and development of drug candidates that ultimately fail. The later the failure in the drug discovery process, the greater the cost. This has given rise to the mantra: “fail fast, fail early.” In order to address this deficiency, more sophisticated in silico models of absorption, often modeling activities in in vitro assays such as Caco-2 or MCDK and BBB, have been developed [44]. Many of these models use computational intelligence methodologies such as ANNs, EC, and fuzzy logic and will be discussed in more detail in Section 5. Several commercial software packages for the prediction of drug absorption include properties such as: aqueous solubility and partition coefficients; Caco-2 cell permeability; BBB permeability; MDCK cell and skin permeability; and cell absorption. Some representative programs (and companies) include: ADMET Predictor (Simulations-Plus, www.simulations-plus.com); Discovery Studio, TOPKAT and Accord (Accelrys, www. accelrys.com); ChemSilico (ChemSilico, www.chemsilico.com); KnowItAll (Bio-Rad, www.biorad.com); ADME boxes (Pharma-Algorithms, www.pharma-algorithms.com); Pre-ADMET (www.preadmet.bmdrc.org); QikProp (Schrödinger, www.schrodinger. com); VolSurf (Molecular Discovery, www.moldiscovery.com); VolSurf and Sybyl (Tripos, www.tripos.com); and various MDL databases and tools (Symyx, www.mdli. com). 2.2. Distribution Prediction of tissue distribution of a drug is a very important consideration in drug development. Descriptors such as logP, molecular weight, as well as acidity have been proven to be useful in modeling distribution [45]. Additional terms are usually added to account for plasma-protein binding, tissue composition, blood composition, as well as blood flow to the tissues [46-49] Modeling of plasma protein binding is also very important as bound drugs are often prevented from crossing cell membranes and getting to their intended targets [50]. On the other hand, drugs that bind to proteins tend to have a longer t [51]. Plasma proteins
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 357
that bind drugs include: albumin (for acidic drugs); 1-acid glycoproteins (for basic drugs); lipoproteins (for neutral and basic drugs) as well as erythrocytes and ,,globulins [20, 51]. Several commercially available software packages exist for prediction of multicompartment pharmacokinetic models, drug release, dissolution, and transport, elimination half-life, and plasma protein binding. Some representative examples include: GastroPlus (Simulations-Plus, www.simulations-plus.com); Pre-ADMET (www.preadmet.bmdrc. org); and KnowItAll (Bio-Rad, www.biorad.com). 2.3. Metabolism Predicting the potential interactions and metabolic pathways of a drug is extremely difficult. There is great interest in developing models for biotransformation (e.g., toxic metabolites & intermediates), enzyme and/or receptor binding and inhibition (e.g., cytochrome P450, hERG potassium channels), and synergistic/antagonistic drug-drug interactions [52, 53]. Biotransformation studies are important to identify what enzymes are metabolizing a drug, what metabolites are produced and if so, how they are cleared [54, 55]. If these metabolites are reactive, they can cause toxicity or other adverse events. Rule-based expert systems have been developed based on chemical similarities and decision trees [55]. These include: MDLI Metabolite Database (Symyx, www.mdli.com); Meteor (Lhasa, www.lhasalimited.org); MetaDrug (GeneGo, www.genego.com); MexAlert and MetabolExpert (CompuDrug, www.compudrug.com); and MetaSite (Molecular Discovery, www.moldiscovery.com). Potential cytochrome P450 interactions (as well as interactions with other metabolically important enzymes) have traditionally been studied using QSAR models [56, 57]. Because of the potential for arrhythmia and cardiac failure, there is also currently interest in developing QSAR models for potential interactions with the hERG potassium channel [58, 59]. Unfortunately there have been relatively few models of metabolic stability and its effect on t, or on potential drug-drug interactions [54]. Because of the complexity of modeling metabolism and metabolic pathways, there is currently great interest in applying computational intelligence methodologies. This will be discussed in more detail in Section 5 below. 2.4. Excretion Currently very little effort has been directed towards in silico models of excretion processes [36]. While most drugs are excreted to via the kidneys or the bile to some extent, for the most part they are eliminated via other routes (e.g., they are metabolized). 2.5 Toxicology As toxicity is currently the major reason for drug candidate failure in clinical trials, there is currently considerable interest in developing predictive in silico models. These models generally fall into one of two following categories: expert systems (based on rules generated from human experts as well as the scientific literature) and QSAR models – in particular for cytochrome P450’s and hERG receptors (as discussed in Section 2.3) [60-62].
358 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
Expert systems have been used to predict toxicological endpoints that include: rodent carcinogenicity; Ames mutagenicity; developmental toxicity potential; skin and eye irritation; acute oral toxicity LD50; acute inhalation toxicity LC50; acute toxicity LD50; acute toxicity EC50; maximum tolerated dose (MTD); chronic lowest observable adverse effect level (LOAEL); and skin sensitization [63]. Several representative examples of predictive toxicological software include: Actelion Property Explorer (Actelion, www.actelion.com); ADMET-Predictor (SimulationsPlus, www.simulations-plus.com); ChemSilico (ChemSilico, www.chemsilico.com); DEREK (Lhasa, www.lhasalimited.org); Hazard Expert (CompuDrug, www.compudrug. com); KnowItAll (Bio-Rad, www.biorad.com); LAZAR (www.predictive-toxicology. org/lazar/); MCASE, CASE, MTOX (Multicase, www.multicase.com); OncoLogic (www.epa.gov/oppt/sf/); Pre-ADMET (www.preadmet.bmdrc.org); TOPKAT (Accelrys, www.accelrys.com); ToxBoxes (Pharma-Algorithms, www.pharma-algorithms.com); and ToxScope (Lead Scope, www.leadscope.com). Recent developments of microarray technologies have completely transformed the fields of toxicogenomics and pharmacogenomics [64-71]. Not only are microarray experiments used identifying biomarkers and validating drug targets, they are also used to study the metabolic and potential toxicological effects of compounds in a highthroughput mode. The amount of data generated from these experiments is astronomical and CI approaches are routinely employed in these analyses [64-71]. 3. QSAR & QSPR QSAR models are in essence a mathematical function that relates features and descriptors generated from small molecule structures to some experimental determined activity or property. The first QSAR models introduced in 1969 were rather simple, identified the relationship between the water-octanol partition coefficient and biological activity [38]: log(1/C) = k1logP – k2(logP)2 + k3s + k4
(1)
where C is the concentration of the compound that gives a biological response, P is the water-octanol partition coefficient, and k1, k2, k3, and k4 are constants. QSPR models are used often to model and predict ADMET properties. QSAR and QSPR are very similar in that much of the same computational approaches are used in their development and optimization. The major differences arise from the activities/properties they are designed to predict. For QSAR models, relevant biological responses most often include: the concentration needed to inhibit 50% of activity (IC50); the dose required to reduce activity by 50% in cell based or animal studies (ED50); the inhibition constant, Ki; as well as the bonding constant Kd. As mentioned above, QSPR models are generated to predict physico-chemical properties and as well as biological activities relevant to ADMET. These often include the dose required to kill 50% of the cells or animals tested (LD50), solubility, lipohilicity and partition coefficients, absorption through intestinal walls, measures of cell membrane permeabilities, as well as BBB penetration. Some of the more widely used commercially available software packages for performing QSAR and QSPR include Cerius2 and Cata-
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 359
lyst (Accelrys), MOE (Chemical Computing Group), OpenEye, MDL, and Sybyl (Tripos). 3.1. QSAR and QSPR: 1D & 2D Models QSAR and QSPR models are based on molecular descriptors or features. While there are literally thousands of descriptors available, they generally fall into one of four major classes: 1) counts of features; 2) physico-chemical properties; 3) topological indices and atom connectivities; and 4) calculated intramolecular energies. The first class includes descriptors such as hydrogen bond acceptors, hydrogen bond donors, aromatic ring systems, carbonyl groups, basic nitrogens, and carboxyl groups. Descriptors based on predicted physico-chemical properties include dipole moments, volumes, polarizabilities, water-octanol partition coefficients, solubilities, molecular weights, melting points, boiling points, heat of sublimations, and molar refractivities. Topological indices and atom connectivities are based on the two- and three-dimensional structures of compounds. These include branching indices, kappa shape indices, electrotopological state indices, atom-pairs, topological torsions, as well as surface areas both polar and non-polar. Finally, there are many descriptors based on calculated intramolecular energies using both quantum mechanical as well as empirical methodologies. Numerous software packages are available for generating molecular descriptors. Some of the more popular and well known ones include: Sybyl (Tripos), Catalyst and Cerius2 (Accelrys), MOE (Chemical Computing Group), OpenEye, and Dragon. QikProp (Schrödinger) is primarily focused on generating descriptors and predicted activities relevant to ADMET. Because of the large number of descriptors available, the rate limiting step in the development of QSAR and QSPR models is often their identification, appropriate reduction, and weighting. A variety of techniques are typically employed for this purpose including multiple linear regression (MLR), partial least squares regression (PLS), and principle component analysis (PCA). MLR is perhaps the most widely used method for modeling linear correlations between descriptors and activities. For best results, the number of samples should be > 2n, where n is the number of descriptors. It is also important that descriptors used are not significantly correlated in order to avoid redundancies. PLS is useful for cases where the number of samples is small with respect to the number of descriptors. Unfortunately this is very often the case in drug discovery and development where data points are often very expensive and difficult to obtain. In PLS there is a linear transformation of the original descriptors into a new space composed of a smaller number of orthogonal variables. PCA is useful for transforming a large number of correlated descriptors into a far fewer number of orthogonal descriptors or principal components. The first principal component accounts for as much of the variability as possible, with each subsequent principal component accounting for additional variability. Development of improved and more efficient strategies is a very active area of research, and some of the more popular techniques include computational intelligence methodologies such as ANNs and evolutionary algorithms (EAs) [72-74]. In particular, ANNs have proven useful for selection of features that are nonlinearly correlated to small molecule activities [75-83]. The “genetic function approximation” (GFA) is another variation of evolutionary computing in which populations of QSAR models are generated and optimized [84].
360 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
This approach has become very popular and is included in Accelrys’ Cerius2 software package for QSAR model generation. 3.2. Pharmacophore Modeling, CoMFA and CoMSIA QSAR and QSPR models using pharmacophore modeling have proven to be very useful in drug discovery and development [85-87]. A pharmacophore is essentially the three dimensional substructure of an active compound or structure class that is both necessary and sufficient for bioactivity. The first step in generating a pharmacophore requires generation of a 3D structural alignment of a set of active compounds. Common structural and chemical features in the aligned structures are then identified and the distances and angles between the features are calculated. These features often include: hydrogen bond donors and acceptors, charged or polar groups, as well as aromatic groups. These models are extremely computationally efficient and large numbers of compounds (literally millions) can be screened against these models. Scoring is based on how well they fit the model. One very popular variation on pharmacophore modeling is comparative molecular field analysis (CoMFA) [85, 87]. As in pharmacophore modeling, a 3D structural alignment is performed on a set of training compounds. However, for a CoMFA model, the structural alignment is performed in a lattice of grid of points to which a molecular force field is applied [88]. Interaction energies are calculated for the molecule at each point of the lattice. These energies typically have steric, electrostatic and hydrophobic terms. Because of the large numbers of descriptors, PCS and/or PLS are typically used to reduce the number of descriptors during model development. Comparative molecular similarity indices analysis (CoMSIA), is very similar to CoMFA but is instead based on similarity [89]. 4. COMPUTATIONAL INTELLIGENCE AND MACHINE LEARNING The field of computational intelligence has many tools and techniques for building predictive models for processes that are extremely complex and where our understanding of the fundamentals is limited [76, 90, 91]. There are very few problems more complex than that of modeling biological responses in response to administration of a drug. It should not be surprising, then, that many of these tools and methodologies have been successfully applied to QSAR & QSPR modeling [22, 59, 92]. In QSAR & QSPR models, these computational intelligence approaches are used to predict experimental activities based on descriptors or features requiring a method of supervised learning. Perhaps one of the most useful applications has been that of feature selection. As mentioned previously, there are literally thousands of descriptors currently available. This section presents a brief introduction to ANNs, fuzzy logic, EC, as well as other machine learning approaches. 4.1. Artificial Neural Networks Artificial neural networks are transfer functions modeled loosely after the neural architecture of the human brain that accept some number of input features and yield some output decision. ANNs (or more commonly referred to as simply “neural networks”) are patterned after the neuronal structure of the brain as a tool for pattern recognition [9395]. Supervised learning of ANNs occurs using a training set of examples in which the neural net learns the relevant mapping of inputs to output decisions.
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 361
A typical ANN architecture consists of an input layer, one or more hidden layers, and an output layer. An example is shown in Fig. (2). Linear neural net models do not have a hidden layer: input nodes are directly connected to the output node(s). Non-linear models have at least one hidden layer with connections to both the input layer and the output layer. The number of connections between the nodes of each layer and their relative weightings will vary from model to model. For QSAR and QSPR, inputs to the ANNs are molecular descriptors and the output is a decision concerning the predicted activity or other property [77]. As was the case for use of MLR, PLS, and PCA, feature selection needs to be performed in order to select which ones to include or exclude from the model as input. Each input or feature then needs to be weighted with respect to maximizing predictive accuracy on the output decision over the training examples. The relative weights of each input are often unknown.
Output Layer
Input Layer
Hidden Layer(s)
Fig. (2). An artificial neural network architecture using five input nodes, one hidden layer with four nodes, and two output nodes. This architecture is a feed-forward multi-layer perceptron. Other architectures are possible making use of recurrence, a variable number of connections, variable number of nodes, nodes per layer, layers, processing elements internal to each node.
Optimization of the relative weights and/or the architecture of the ANN (e.g., the connections between layers) can be performed in order to minimize the mean squared error (MSE) between the predicted output and actual values over the training set. For example:
MSE =
1 N
N
(P k =1
k
Ok ) 2
(2)
where P was the predicted activity, O was the observed activity, and N was the number of patterns in the training set. Backpropagation is one of the most commonly used of the training algorithms for weight adjustment. A validation set of held-out examples (not used for training) is used to test the best model. The model can be re-designed if neces-
362 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
sary. In some cases a second held-out testing set of data is used to assess final predictive accuracy. 4.2. Fuzzy Systems Fuzzy systems which are based on fuzzy set theory [96, 97], attempt to build models that capture uncertainties and imprecision not easily quantified by other methods. Fuzzy algorithms have proven useful for clustering or classification in bioinformatics [98-102] where they are used to handle uncertainties in rule-based representations. For prediction of “drug-likeness” a fuzzy model representation might take the form: IF the activity score is ACTIVE and compliance to the Rule-of-Five is MOSTLY TRUE THEN the decision of drug-likeness is TRUE For prediction of toxicity a fuzzy model representation might take the form: IF the structure is SIMILAR TO a known cytrochrome P450 inhibitor and the predicted metabolite score is ACTIVE THEN the decision of toxicity is TRUE Fuzzy systems seem ideal for modeling toxicity and metabolism where the inputs used to generate the model do not cleanly separate into discrete values or are subjective. Whereas other methods would force the inputs or continuous variables into partitions on user defined discrete intervals, a fuzzy system can be designed to represent membership in vaguely defined partitions. This is useful when the discrete interval boundaries are largely subjective and/or difficult to determine empirically. There are many subdisciplines of fuzzy logic theory that have been developed to handle linguistic variables, and many of these are appropriate for use in biological problems such as prediction of toxicity or of metabolism. 4.3. Evolutionary Computation Evolutionary algorithms are designed to mimic natural evolution as a populationbased optimization process. An typical example is provided in Fig. (3). EAs use random variation and selection as a means for discovering solutions to complex problems. A typical evolutionary computation process starts with an initial set (population) of solutions. These are randomly altered (e.g., mutated and/or recombined) to generate the individuals comprising the current population which are subsequently evaluated using a fitness function (defined by the user). Based on their scores, a subset of individual solutions in the population are chosen to be parents for the next generation. The cycle then continues with random alteration, scoring with the fitness function and then selection. This continues until a halting criterion has been met, such as a specific number of generations or exceeding the available time. Methods of evolutionary computation include evolutionary programming [103], evolution strategies [104], genetic algorithms [105107], genetic programming [108], particle swarm optimization [109], ant-colony optimization [110], differential evolution [111, 112], and others. Each approach has its own advantages and disadvantages relative to specific problems. The “No Free Lunch” theorem indicates that no single optimization approach will work best over all problems [113].
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 363
4.4 Evolved Artificial Neural Networks and Evolved Fuzzy Systems One very powerful application of evolutionary computing has been the optimization of the connections (and weightings) between the input layer, the hidden layer(s), and the output layers in neural networks [114, 115]. The evolutionary algorithm creates populations of ANNs and scores each ANN based on mean squared error between the predicted and actual outputs. Likewise, evolutionary algorithms can be used to optimize any fuzzy classifiers or fuzzy inputs that are used in ANNs (e.g., fuzzy neural nets) or in fuzzy systems [116]. Evolutionary computing can also be used to evolve the selection of features to be used in a neural net model simultaneous with the optimization of that model’s architecture [117-120].
Population Initialization
Random Variation
Fitness Scoring
Parent Solutions
Process Termination
Fig. (3). A flow diagram of a standard evolutionary algorithm. The loop of variation, scoring, and, generation of parent solutions for the next “generation” of evolution continues until a termination criterion is satisfied.
4.5. Other Common Machine Learning Approaches Support vector machines (SVM) have recently been used for prediction of compound activities [121, 122]. Support vector machines represent the input descriptors/features as vectors that are projected onto higher-dimensional space. An optimal hyperplane is then constructed separating the actives and inactives. The hyperplane is used to predict the activity of new compounds that are tested [123-125]. Other techniques employed for modeling of ADMET properties include clustering and decision trees with recursive partitioning [22, 44, 59, 126]. K-means clustering is one of the oldest and most widely used clustering methods. Data are grouped by similarities in their features/descriptors. Decision trees consist of nodes where each node is connected to all the outcomes of a decision based on a single attribute. Recursive partitioning is often used to examine every attribute of the data and rank them with regards to their ability to partition the rest of the data. In general, the tree is first grown to its full size by evaluating each and every attribute and generating nodes for each outcome. The tree is then pruned back based on its predictive performance. 5. COMPUTATIONAL INTELLIGENCE AND ADMET MODELING In this section, current applications of computational intelligence to predictive ADMET and QSPR models are reviewed. These models have focused on a relatively small (but very important) subset of ADMET properties and activities reflecting the needs of drug development programs to increase the survivability of drug candidates. 5.1. Absorption Absorption is critical to the development of orally available pharmaceuticals. Models for aqueous solubility, intestinal absorption, Caco-2 permeability as well as BBB penet-
364 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
ration are well established and are routinely applied in lead discovery as a screen for “non-druglike” compounds. As these models predict properties and activities related to solute-solvent interactions including hydrophobic interactions, they often employ topological and surface property based descriptors such as PSA. Aqueous solubility is perhaps one of the most commonly modeled ADMET properties. Solubility refers to the maximum amount of compound that can dissolve in a given quantity of water. These models are commonly used as experimental determinations of solubility are very costly in terms of time, money and perhaps most importantly, the amount of compound used. In general, milligrams of compound are required for solubility determination. Table 1 provides representative examples of computational intelligence based aqueous solubility models. While MLR was a common approach used method early on [130-135], in recent years there has been an increased use of ANNs and EAs [120, 144152, 155-163]. When EAs are combined with other techniques such as ANNs, they are most often used for feature selection [162]. However, EAs can also be used effectively in order to evolve the ANNs themselves [120]. Table 1.
Models of Aqueous Solubility
Reference #
Method
Descriptors
127-139
MLR
calculated molecular descriptors
140, 141
MLR
topological and molecular descriptors
142
MLR
surface and calculated molecular descriptors
143
PLS
infrared spectral data
144-147
ANN
calculated molecular descriptors
148-156
ANN
topological and molecular descriptors
157-160
EA
calculated molecular descriptors
161
EA
topological and molecular descriptors
120, 162
EA & ANN
calculated molecular descriptors
163
ANN & Fuzzy Logic
topological descriptors
164
SVM
calculated molecular descriptors
165
EA, ANN, SVM
calculated molecular descriptors
Experimental determinations of intestinal absorption are generally very low throughput, extremely time consuming, and require costly animal models [166]. Because of these considerations, in silico models are very commonly used – particularly early on in drug discovery and development. Table 2 provides representative examples of human intestinal absorption models. While MLR and PLS were the techniques most commonly
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 365
used in the development of these models, there are a number of models using ANNs and EA as well as decision trees and recursive partitioning [183]. In two models, EAs were used for descriptor selection and then ANNs were generated [181, 182]. While the types of descriptors used in these models varied considerably, topological and surface properties were often used as well as H-bonding terms and logP values. Table 2.
Models of Intestinal Absorption
Reference #
Method
Descriptors
167-169
MLR
calculated molecular descriptors
170
MLR
H-bond descriptors and other calculated molecular descriptors
171
MLR
molecular groups
172
MLR
PSA
173
PLS
MolSurf
174
PLS
H-Bond descriptors and logP
175
PLS
calculated molecular descriptors
176
PLS
atom types
177
ANN
molecular hashkeys
178
ANN
PSA, logP, and topological descriptors
152
ANN
topological descriptors
158, 179, 180
EA
calculated molecular descriptors
181, 182
EA & ANN
calculated molecular descriptors
183
Recursive Partitioning
calculated molecular descriptors
184
Decision Trees
calculated molecular descriptors
164, 185-187
SVM
calculated molecular descriptors
Although Caco-2 permeability studies are less costly and easier to run than other intestinal absorption models, there remains great interest in using in silico filters – especially when screening large libraries [9, 11]. As was the case for the intestinal absorption models, both MLR, PLS, and ANNs were the techniques most commonly used. Again, topological and surface property based descriptors proved to be the most useful for these models. As experimental models of BBB penetration tend to be relatively expensive and low throughput [166]. It is imperative that drugs targeted for CNS indications are able to pass through this barrier. For these reasons, in silico BBB permeability filters are applied early in drug discovery and development. Table 4 lists BBB models constructed using
366 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
MLR, PLS, ANNs as well as EAs and SVMs. Of particular interest are models that combine computational intelligence techniques [229, 231]. Table 3.
Models of Caco-2 Permeability
Reference #
Method
Descriptors
188, 189
MLR
PSA and MW
190, 191
MLR
topological and surface descriptors
192, 193
MLR
calculated molecular descriptors
194
MLR
H-bond and molecular descriptors
175, 195, 196
PLS
calculated molecular descriptors
197
PLS
MolSurf descriptors
198
PLS
VolSurf descriptors
174
PLS
logP and H-bonding descriptors
181
ANN
topological descriptors
199-201
ANN
calculated molecular descriptors
100
ANN
calculated molecular descriptors
101
ANN
calculated molecular descriptors
202
EA
topological descriptors
158
EA
calculated molecular descriptors
203
SVM
calculated molecular descriptors
Table 4.
Models of BBB Permeability
Reference #
Method
Descriptors
204-207
MLR
PSA, logP & molecular descriptors
208-214
MLR
calculated molecular descriptors
215
MLR
topological descriptors
216
MLR
Solvation energy
217
MLR
PSA, H-bond descriptors, logP
174
PLS
H-bonding descriptors and logP
218
PLS
MolSurf descriptors
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 367 (Table 4) contd....
Reference #
Method
Descriptors
219
PLS
surface and molecular descriptors
175, 196, 220, 221
PLS
calculated molecular descriptors
222
PCA
VolSurf descriptors
223
ANN
topological descriptors
58, 82, 224, 225
ANN
calculated molecular descriptors
226
EA
calculated molecular descriptors
227
EA, CoMFA, CoMSIA
calculated molecular descriptors
164, 228
SVM
calculated molecular descriptors
229
SVM & ANN
calculated molecular descriptors
230
Decision Tree
calculated molecular descriptors
231
ANN, SVM, Clustering, & Decision Tree
calculated molecular descriptors
5.2. Distribution, Clearance, and Metabolism In addition to optimizing compounds for the ability to be absorbed, it is also very important to optimize compounds for their distribution to different organs and tissues, their clearance from the body, as well as their metabolic stabilities. As was the case for absorption, in silico models are used routinely to screen compounds for these purposes. Models of cytochrome P450 activity are extremely important for evaluating the potential for metabolism and reactive intermediate formation before compounds transit through the portal vein and into general circulation. As these models are designed to predict enzyme activity they often include 3D-QSAR techniques such as pharmacophore modeling [245, 246] as well as CoMFA [252, 253]. Examples of modeling techniques such as ANNs, EAs, SVMs, MLR, and PLS are also presented in Table 5. Table 5.
Models of Predicted Cytochrome P450 Activity
Reference #
Method
Descriptors
232, 233
MLR
surface descriptors, logP
234
PLS
logP
158
EA
topological descriptors
235, 236
ANN
calculated molecular descriptors
164, 237-242
SVM
calculated molecular descriptors
368 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hecht and Fogel
(Table 5) contd....
Reference #
Method
Descriptors
243
ANN, SVM
calculated molecular descriptors
244
3D QSAR
structural fragments
245, 246
3D QSAR
pharmacophore
56, 247-251
3D QSAR
calculated molecular descriptors
252, 253
3D QSAR
CoMFA
254
Clustering
calculated molecular descriptors
255-257
Clustering
microarray gene expression data
As mentioned previously, once drugs enter the blood stream, they often bind to blood proteins such as albumin and may be prevented from reaching their targets. Again, in silico models to predict albumin binding have also proven to be very useful drug development [9,11]. Some representative examples of these models are presented in Table 6. A variety of techniques have been used including ant colony optimization, a type of evolutionary algorithm [264]. Table 6.
Models of Human Serum Albumin Binding
Reference #
Method
Descriptors
258
MLR
calculated molecular descriptors
259-260
MLR
logP, topological descriptors, PSA
261
PLS
calculated molecular descriptors
58, 262, 263
ANN
calculated molecular descriptors
180
EA
calculated molecular descriptors
264
EA (Ant Colony Optimization)
calculated molecular descriptors
164
SVM
calculated molecular descriptors
265
Expert System
pharmacophores
Although there are fewer examples, models of clearance, t and metabolic stability are also very important. Table 7 lists several based on EAs and SVMs. Examples of models based on fuzzy logic or clustering are also provided.
Computational Intelligence Methods
Table 7.
Frontiers in Drug Design & Discovery, 2009, Vol. 4 369
Models of Clearance & Metabolic Stability
Reference #
Method
Property
266
Fuzzy Systems
clearance time
180
EA
urinary excretion
267
EA
volume of distribution
268
Clustering
metabolic stability
269
SVM
metabolic stability
164
SVM
t and volume of distribution
5.3. Toxicity Currently toxicity is the major cause of drug candidate failure during development and clinical trials and is responsible for >90% of the drugs pulled off the market [7, 9, 10]. The ramifications of toxicity are enormous not only in terms of costs, but in actual lives. One of the most common examples is that of hERG receptor modeling. In some cases, drugs will bind to the hERG receptor and cause arrhythmia and hear failure. It is therefore important to screen out compounds with the potential for this adverse effect as early as possible. The examples presented in Table 8 utilize EAs and SVMs. Table 8.
Models of Predicted hERG Receptor Binding
Reference #
Method
Descriptors
270
EA
calculated molecular descriptors
271
EA & CoMFA
calculated molecular descriptors
58, 272-274
SVM
calculated molecular descriptors
275
SVM and Clustering
calculated molecular descriptors
Other examples of in silico toxicology models are presented in Table 9. While most of these models have been generated using molecular descriptors, there are a couple based on gene expression profiles from microarray data. 6. GENERAL TRENDS AND FUTURE DIRECTIONS Table 10 presents an analysis of the literature cited in Tables 1-9 grouping them by year and methodology. From this analysis, a couple of general overall trends for the field seem to emerge, although the data presented here is only a sampling of the available literature.
370 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 9.
Hecht and Fogel
Models of Toxicity
Reference #
Method
Descriptors
Toxicity
152
ANN
topological descriptors
Ames genotxocity
276
EA & ANN
calculated molecular descriptors
drug transfer to breast milk
277
ANN, SVM, Clustering, Decision Trees
calculated molecular descriptors
overall predicted toxicity
278
SVM & ANN
calculated molecular descriptors
overall predicted toxicity
70
Clustering
gene expression profiles
overall predicted toxicity
279
Clustering
gene expression profiles
hepatotoxicity
280
Clustering
structures and data in LAZAR
Ames genotxocity
281
Clustering
Structures and molecular descriptors
overall predicted toxicity
Prior to 2000, MLR appeared to be the preferred methodology for generating ADMET models. This changed in the early 2000s, when computational intelligence-based approaches became more popular - particularly with the use of ANNs and EAs. Since 2005, SVM and clustering approaches (including decision tree analyses) have also been used with increasing regularity. The increased use of clustering and decision tree analyses in recent years reflects the great interest in developing models of metabolism and toxicity using gene expression data coming from microarray data. Because of the astronomical quantity of data produced, it is to be anticipated that computational intelligence methodologies will continue to play a major role. Table 10. Number of References Cited Grouped by Methodology and Year
Method
1980s-1999
2000-2004
2005 -2008
ANN
6
21
7
EA
2
9
9
MLR
18
28
2
PLS
7
8
0
SVM
0
3
16
Clustering & Decision Trees
0
2
9
Computational Intelligence Methods
Frontiers in Drug Design & Discovery, 2009, Vol. 4 371
Again, this reflects the shift in recent years from development of newer and improved models relevant to absorbance to models of predicted metabolic activities and toxicity. Perhaps one of the largest hurdles to overcome is the amount of proprietary metabolic and toxicity data maintained in the databases of pharmaceutical companies. A promising recent development has been microarray data that has been made public (from NCBI as well as other sources). Another very important area of opportunity is the development of improved models of distribution and clearance. There are currently relatively few examples of applications of computational intelligence and this is an area of likely future application. 7. CONCLUSIONS As discussed previously, there is great interest in developing new and improved ADMET models in order to improve the efficiency and productivity of drug discovery and development. Because of the great complexities, scarce and “noisy data,” as well as overwhelming numbers of parameters involved, researchers have borrowed heavily from the field of computational intelligence and machine learning. In this paper we have reviewed applications of computational intelligence methods to the development of predictive ADMET models. There is a great opportunity for the development of novel approaches and methodologies that will increase the likelihood of survival of drug candidates through the development process. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25]
DiMasi, J.A.; Grabowski, H.G. Manage. Decis. Econ., 2007, 28, 469-479. PhRMA Industry Profile 2008 Report, www.phrma.org/publications Center for Drug Evaluation and Research: www.fda.gov/cder/rdmt Tufts Center for the Study of Drug Development, Backgrounder: How New Drugs Move through the Development and Approval Process (November 2001). Darvas, F.; Keser, G.; Papp, Á.; Dormán, G.; Ürge, L.; Krajcsi, P. Curr. Top. Med. Chem., 2002, 2, 1287-1304. Von Eschenbach, A.C. “Statement Before the Senate Agriculture, Rural Development, Food and Drug Administration and Related Agencies Appropriations Subcommittee, ” U.S. Food and Drug Administration, June 2007, www.fda.gov/ola/2007/criticalpath060107.html (accessed 27 November 2007). Kennedy T. Drug Discov. Today, 1997, 2, 436-44. Prentis, R.A.; Lis, Y.; Walker, S.R. Br. J. Clin. Pharmacol., 1988, 25, 387-96. Tsaioun, K. Drug Discov., 2007, 20, 21. Schuster D.; Laggner, C.; Langer, T. Curr. Pharm. Des., 2005, 11, 3545-59. Weiss, A.J. www.dddmag.com, 2002, 27. Lipinski, C. A.; Christopher, A. L. Adv. Drug Deliv. Rev., 1997, 23, 3-25. Kubinyi, H. Nat. Rev., 2003, 2, 665-668. Lahoz, A.; Gombau, L.; Donato, M.T.; Castell, J.V.; Gómez-Lechón, M.J. Mini Rev. Med. Chem., 2006, 6, 1053-1062. Wunberg, T.; Hendrix, M.; Hillisch, A.; Lobell, M.; Meier, H.; Schmeck, C.; Wild, H.; Hinzen, B. Drug Discov. Today, 2006, 11, 175-180. Kumar, R.A.; Clark, D.S. Curr. Opin. Chem. Biol., 2006, 10, 162-168. Yamashita, F.; Hashida, M. Drug Metab. Pharmacokin., 2004, 19, 327-338. Lüpfort, C.; Reichel, A. Chem. Biodivers., 2005, 2, 1462-1486. Lombardo, F.; Gifford, E.; Shalavea, M.Y. Mini Rev. Med. Chem., 2003, 3, 861-875. van de Waterbeemd, H.; Gifford, E. Nat. Rev. Drug Discov., 2003, 2, 193-204. Desai, P.V.; Coutinho, E.C. Asian Chem. Lett., 2001, 5, 77-86. Duch, W.; Swaminathan, K.; Meller, J. Curr. Pharm. Des., 2007, 13, 1497-1508. Engelbrecht, A.P. Computational intelligence: An introduction, New York: J. Wiley; 2003. Konar, A. Computational intelligence; principles, techniques and applications, Berlin: Springer 2005. Duda R.O.; Hart, P.E.; Stork, D.G. Pattern classification, New York: J. Wiley, 2nd edt 2001.
372 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72]
Hecht and Fogel
Webb, A. Statistical pattern recognition, New York: J. Wiley 2002. Hastie, T.; Tibshirani, R.; Friedman, J. The elements of statistical learning, Springer 2001. Kramer, S. D. Pharm. Sci. Technol. Today, 1999, 2, 373-380. Stenberg, P.; Bergstrom, C. A. S.; Luthman, K.; Artursson, P. Clin. Pharmacokinet., 2002, 41, 877. Sanders, N. R.; Habgood, M. D.; Dziegielwska, K. M. Clin. Exp. Pharmacol. Physiol., 1999, 26, 1119. Stenberg, P.; Norinder, U.; Luthman, K.; Artursson, P. J. Med. Chem., 2001, 44, 1927-1937. Yazdanian, M; Glynn, S.L.; Wright, J.L.; Hawi, A. Pharm. Res., 1998, 15, 1490-1494. Artursson, P.; Karlsson, J. Biochem. Biophs. Res. Commun., 1991, 175, 880. Irvine, J.D.; Takahashi, L.; Lockhart, K.; Cheong, J.; Tolan, J.W.; Selick, H.E.; Grove, J.R. J. Pharm. Sci., 1999, 88, 28-33. Basak, S. C.; Gute, B. D.; Drewes, L.R. Pharm. Res., 1996, 13, 775-778. Boobis, A.; Gundert-Remy, U.; Kremers, P.; Macheras, P.; Pelkonen, O. Eur. J. Pharm. Sci., 2002, 17, 183-193. Tetko, I.V.; Tanchuk, V.Y.; Villa, A.E. J. Chem. Inf. Comput. Sci., 2001, 41, 1407-1421. Hansch, C. Acc. Chem. Res., 1969, 2, 232-239. Ghose, A. K.; Vishwanadhan, V. N.; Wendoloshki, J. J. J. Comb. Chem., 1999, 1, 55-68. Muthas, D.; Sabnis, Y.A.; Lundborg, M.; Karlén, A. J. Mol. Graph. Model., 2008, 26, 1237-1251. Oprea, T. I. J. Comput. Aided Mol. Des., 2000, 14, 251. Wenlock, M. C.; Austin, R. P.; Barton, P.; Davis, A. M.; Leeson, P.D. J. Med. Chem., 2003, 46, 1250. Vieth, M.; Siegel, M. G.; Higgs, R. E.; Watson, I. A.; Robertson, D. H.; Savin, K. A.; Durst, G. L.; Hipskind, P. A. J. Med. Chem., 2004, 47, 224. Hou, T.; Wang, J.; Zhang, W.; Wang, W.; Xu, X. Curr. Med. Chem., 2006, 13, 2653-2667. Poulin, P.; Schoenlein, K.; Theil, F.P J. Pharm. Sci., 2001, 90, 436-447. Lombardo, F.; Obach, R.S.; Shalaeva, M.Y.; Gao, F. J. Med. Chem., 2002, 45, 2867-2876. Poulin, P.; Theil, F.P. J. Pharm. Sci., 2000, 89, 16-35. Poulin, P.; Theil, F.P. J. Pharm. Sci., 2002, 91, 129-156. Poulin, P.; Theil, F.P. J. Pharm. Sci., 2002, 91, 1358-1370. Smith, D. A.; Van de Waterbeemd, H.; Walker, D. K. Pharmacokinetics and Metabolism in Drug Design, Wiley–VCH, Weinheim, Germany, 2001. Lutsevich, A.N. Pharm. Chem. J., 1990, 24, 593-599. Ekins, S.; Waler, C.L.; Swaan, P.W.; Cruciani, G.; Wrighton, S.A.; Wikel, J.H. J. Pharmacol. Toxicol. Methods, 2000, 44, 251-272. Ekins, S.; Wrighton, S.A. J. Pharmacol. Toxicol. Methods, 2001, 45, 65-69. Baranczewski, P.; Staczak, A.; Sundberg , K.; Svensson, R.; Wallin, A; Jansson, J.; Garberg, P.; Postlind , H. Pharmacol. Rep., 2006, 58, 453-72 Wishart, D.S. Drugs R D , 2007, 8, 349-362. Ekins, S.; Bravi, G.; Blinkley, S.; Gillespie, J. S.; Ring, B.J.; Wikel, J. H.; Wrighton, S. A. J. Pharm. Exp. Ther., 1999, 290, 429-438. Ekins, S.; Bravi, G.; Blinkley, S.; Gillespie, J. S.; Ring, B.J.; Wikel, J. H.; Wrighton, S. A. Pharmacogenetics, 1999, 9, 477-489. Yap C.W.; Chen, Y.Z. J. Pharm. Sci., 2005, 94, 153-168. Li, H.; Yap, C.W.; Ung, C.Y.; Xue, Y.; Li, Z.R.; Han, L.Y.; Lin, H.H.; Chen, Y.Z. J. Pharm. Sci., 2007, 96, 2838-2860. Richard, A. M.; Benigni R. SAR QSAR Environ. Res., 2002, 13, 1-19. Greene, N. Adv. Drug Deliv. Rev., 2002, 54, 417-431. Durham, S. K.; Pearl, G. M. Drug Discov., 2001, 4, 110-115. Mohan, C.G.; Gandhi, T.; Garg, D.; Shinde, R. Mini Rev. Med. Chem., 2007, 7, 499-507. Gomase, V.S.; Tagore, S. Curr. Drug Metab., 2008, 9, 250-254. Gomase, V.S.; Tagore, S.; Kale, K.V. Curr. Drug Metab., 2008, 9, 221-31. Collings, F.B.; Vaidya, V.S. Toxicology, 2008, 245, 167-74. Mendrick, D.L. Toxicology, 2008, 245, 175-81. Ganter, B.; Zidek, N.; Hewitt, P.R.; Müller, D.; Vladimirova, A. Pharmacogenomics, 2008, 9, 35-54. Ekins, S. J. Pharmacol. Toxicol. Methods, 2006, 53, 38 - 66. Maggioli, J.; Hoover, A.; Weng, L. J. Pharmacol. Toxicol. Methods, 2006, 53, 31 -37. Wishart, D.S. Drug Metab. Rev., 2005, 37, 279-310. Liu, S.S.; Liu, H.L.; Yin, C.S.; Wang, L.S. J. Chem. Inf. Comput. Sci., 2003, 43, 964-969.
Computational Intelligence Methods [73]
[74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 373
Embrechts, M.J.; Ozdemir, M.; Lockwood, L.; Breneman, C.; Bennett, K.; Devogelaere, D.; Rijckaert, M. In Evolutionary Computation in Bioinformatics, Fogel, G. and Corne, D. Eds., Morgan Kauffman, San Francisco (2002), 317-339. Nicolotti, O.; Gillet, V.J.; Fleming, P.J.; Green, D.V.S. J. Med. Chem., 2003, 45, 5069-5080. Fogel, G.B. Brief. Bioinformatics, 2008, 9, 307-316. Devillers, J. Eds. Neural Networks in QSAR and Drug Design, an essential reference source for those on the frontiers of this field. Academic Press 1996. Chalk, A.J.; Beck, B.; Clark, T. J. Chem. Inf. Comput. Sci., 2001, 41, 457-462. Lui, B.; Nadramija, D.; Baic, I.; Tranajsti, N. J. Chem. Inf. Comput. Sci., 2003, 43, 1094-1102. Mattioni, B.E.; Jurs, P.C. J. Mol. Graph. Model., 2003, 21, 391-419. Weekes, D.; Fogel, G.B. Biosystems, 2003, 72, 149-158. Winkler, D.A.; Burden, F.R. J. Mol. Graph. Model., 2004, 22, 499-505. Lewis, R.A. J. Med. Chem., 2005, 48, 1638-1648. Karakoc, E.; Sahinalp, S.C.; Cherkasov, A. J. Chem. Inf. Model., 2006, 46, 2167-2182. Yap, C.W.; Li, H.; Ji, Z.L.; Chen, Y.Z. Mini Rev. Med. Chem., 2007, 7, 1097-107. Kellogg, G.E.; Semus, S.F. 3D QSAR in modern drug design, In Hillisch, A and Hilgenfeld R. (Eds.): Modern Methods of Drug Discovery. Birkhauser Verlag, Switzerland, (2003) 223-241. Selassie, C.D. History of quantitative structure-activity relationships, Burger’s medicinal chemistry and drug discovery, 6th Edt 2003, 1, 1-48. Akamatsu , M. Curr. Top. Med. Chem., 2002, 2, 1381-94. Cramer III, R. D.; Patterson, D. E.; Bunce, J. D. J. Am. Chem. Soc., 1988, 110, 5959-5967. Klebe, G.; Abraham, U.; Mietzner, T. J. Med. Chem., 1994, 37, 4130-4146. Mjolsness, E.; DeCoste, D. Science, 2001, 293, 2051-5. Fogel, G.B.; Corne, D.W.; Pan, Y. (eds). Computational Intelligence in Bioinformatics, Hoboken, N.J. Wiley, 2008. Kitchen, D.B.; Stahura, F.L.; Bajorath, J. Mini Rev. Med. Chem., 2004, 4, 1029-39. McCulloch, W.S.; Pitts W. Bull. Math. Biophys., 1943, 5, 115-33. Rosenblatt, F. Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms, Washington, DC: Spartan Books, 1962. Haykin, S. Neural Networks: A Comprehensive Foundation, Upper Saddle River, NJ: Prentice Hall, 1998. Zadeh, L.A. Information Control, 1965, 8, 338-53. Zadeh, L.A. Information Control, 1968, 12, 94-102. Torres, A.; Nieto, J.J. J. Biomed. Biotechnol., 2006, 2, 91908. Mordeson, J.N.; Malik, D.S.; Cheng, S.-C. Physica, 2000. Szczepaniak, P.S.; Lisoba, P.J.G.; Kacprzyk J. Physica, 2000. Dong , X.; Bondugula, R.; Popescu, M. 2006 IEEE Int. Conf. Fuzzy Syst., 2006, 817-24. Ruspini, E.H.; Bonissone, P.P.; Pedrycz, W. (eds). Handbook of Fuzzy Computation, Bristol, UK: Oxford University Press, 1998. Fogel, L.J.; Owens, A.; Walsh , M.J. Artificial Intelligence Through Simulated Evolution, New York, NY: Wiley, 1966. Rechenberg , I. Evolutionsstrategie: Optimerung technischer Systeme nach Prinzipien der biologischen Evolution, Stuttgart, Germany: Fromman-Holzboog, 1973. Bremmerman H.J. Optimization through evolution and recombination, In: Yovits M.C.; Jacobi G.T.; Goldstein G.D. (eds). Self-Organizing Systems. Washington DC: Spartan Press, 1962. Holland, J.H. Adaptation in Natural and Artificial Systems, Ann Arbor, MI: University of Michigan Press, 1975. Michalewicz, Z. Genetic Algorithms + Data Structures = Evolution Programs, 3rd edn. Berlin, Germany: Springer, 1996. Koza, J. Genetic Programming: On the Programming of Computers by Means of Natural Selection, Cambridge, MA: MIT Press, 1992. Eberhart, R.C.; Shi , Y.; Kennedy, J. Swarm Intelligence, San Francisco, CA: Morgan Kaufmann, 2001. Dorigo, M; Gambardella, L.M. IEEE Trans. Evol. Comput., 1997, 1, 53-66. Storn, R.; Price, K. Technical Report TR-95-012, ICSI, March 1995. Storn , R. IEEE Trans. Evol. Comput., 1999, 3, 22-34. Wolpert, D.H. and Macready W.G. IEEE Trans. Evol. Comp., 1997, 1, 67-82. Fogel D.B.; Fogel L.J.; Porto V.W. Biol. Cybern., 1990, 63, 487-93. Yao X. Proc. IEEE, 1999, 87, 1423-47.
374 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [116] [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] [164] [165]
Hecht and Fogel
Fogel, G.B.; Cheung, M. IEEE Congress on Evolutionary Computation, Edinburgh, UK, 2005, 274281. MA, C.Y.C.; Wong, S.W.M.; Hecht, D.; Fogel, G.B. IEEE Congress on Evolutionary Computation, Vancouver, Canada, 2006, 9284. Hecht, D.; Fogel, G.B. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2007, 4, 476. Hecht, D.; Cheung, M.; Fogel, G.B. Biosystems, 2008, 92, 10-15. Cheung, M.; Johnson, S.; Hecht, D.; Fogel, G.B. IEEE Congress on Evolutionary Computation, Hong Kong, 2008. Han, L.Y. ; Ma, X.H. ; Lin, H.H.; Jia, J.; Zhu, F.; Xuec, Y.; Li, Z.R.; Cao, Z.W. ; Ji, Z.L.; Chen, Y.Z. J. Mol. Graph. Model., 2008, 26, 1276-1286. Martin, O. and Schomburg, D. Proteins, 2008, 70, 1367-1378. Vapnik, V. Statistical Learning Theory, Wiley: New York, 1998. Plewczynski, D.; Tkacz, A.;Godzik, A.;Rychlewski, L. Cell. Mol. Biol. Lett., 2005, 10, 73-89. Chang, C. C.; Lin, C. J. Neural Comput., 2001, 13, 2119-2147. Tan, P.-N.; Steinbach, M.; Kumar, V. Introduction to Data Mining, Addison Wesley, Pearson International Edition, 2006. Gao, H.; Shanmugasundaram, V.; Lee, P. Pharm. Res., 2002, 19, 497-503. Klamt, A.; Eckert, F.; Hornig, M.; Meck, M.E.; Burger T. J. Comput. Chem., 2002, 23, 275-281. Jorgensen, W.L.; Duffy, E.M. Bioorg. Med. Chem. Lett., 2000, 10, 1155-1158. Medir, M.; Giralt, F. AIChE J., 1982, 28, 341. Nirmalakhandan, N. N.; Speece, R. E. Environ. Sci. Technol., 1988, 22, 328. Nirmalakhandan, N. N.; Speece, R. E. Environ. Sci. Technol., 1989, 23, 708. Patil, G. S. Chemosphere, 1991, 22, 723. Makino, M. Environ. Int., 1998, 24, 653. Katritzky, A. R. J. Chem. Inf. Comput. Sci., 1998, 38, 720. Yin, C.; Liu, X.; Guo, W.; Lin, T.; Wang, X.; Wang L. Water Res., 2002, 36, 2975-2982. Delgado, E. J. Fluid Phase Equilib., 2002, 199, 101. Jorgensen, W. L.; Duffy, E. M. Adv. Drug Deliv. Rev., 2002, 54, 355. Abraham, M. H.; Joelle, J. J. Pharm. Sci., 1999, 88, 868. Chen, X. Q.; Cho, S. J.; Li, Y.; Venkatesh, S. J. Pharm. Sci., 2002, 91, 1838-1852. Huibers, P. D. T.; Katritzky, A. R. J. Chem. Inf. Comput. Sci., 1998, 38, 283. Bergstrom, C. A. S.; Norinder, U.; Luthman, K.; Artursson, P. Pharm. Res., 2002, 19, 182. Collette, T. W. Vib. Spectrosc., 1997, 15, 113. Manallack, D.T; Tehan, B.G.; Gancia, E.; Hudson, B.D.; Ford, M.G.; Livingstone, D.J.; Whitley, D.C.; Pitt, W.R. J. Chem. Inf. Comput. Sci., 2003, 43, 674-679. Yan, A.; Gasteiger, J. J. Chem. Inf. Comput. Sci., 2003, 43, 429-434. Tetko, I.V.; Tanchuk, V.Y.; Kasheva, T.N.; Villa, A.E.P. J. Chem. Inf. Comput. Sci., 2001, 41, 14881493. Liu, R.; So, S.-S. J. Chem. Inf. Comput. Sci., 2001, 41, 1633-1639. Engkvist, O.; Wrede, P. J. Chem. Inf. Comput. Sci., 2002, 42, 1247-1249. Huuskonen, J. J. Chem. Inf. Comput. Sci., 2000, 40, 773-777. Bruneau, P. J. Chem. Inf. Comput. Sci., 2001, 41, 1605-1616. Huuskonen, J.; Salo, M.; Taskinen, J. J. Chem. Inf. Comput. Sci., 1998, 38, 450-456. Votano, J.R.; Parham, M.; Hall, L.H.; Kier, L.B. Mol. Divers., 2004, 8, 379-391. Sutter, J. M.; Jurs, P. C. J. Chem. Inf. Comput. Sci., 1996, 36, 100. Mitchell, B. E.; Jurs, P. C. J. Chem. Inf. Comput. Sci., 1998, 38, 489. Huuskonen, J.; Rantanen, J.; Livingstone, D. Eur. J. Med. Chem., 2000, 35, 1081. McElroy, N. R.; Jurs, P. C. J. Chem. Inf. Comput. Sci., 2001, 41, 1237. Nicolotti, O.; Carotti, A. J. Chem. Inf. Model., 2006, 46, 264-276. Yamashita, F.; Fujiwara, S.-I.; Wanchana, S.; Hashida, M. J. Drug Target., 2006, 14, 496-504. Cheng, A.; Merz, K.M. Jr. J. Med. Chem., 2003, 46, 3572-358. Wegner, J.K.; Zell, A. J. Chem. Inf. Comput. Sci., 2003, 43, 1077-1084. Wanchana, S.; Yamashita, F.; Hashida, M. Pharmazie, 2002, 57, 127-129. Votano, J.R.; Parham, M.; Hall, L.H.; Kier, L.B.; Hall, L.M. Chem. Biodivers., 2004, 1, 1829-1841. Yaffe, D.; Cohen, Y.; Espinosa, G.; Arenas, A.; Giralt, F. J. Chem. Inf. Comput. Sci., 2001, 41, 1177. Balakin, K.V.; Ivanenkov, Y.A.; Savchuk, N.P.; Ivashchenko, A.A; Ekins, S. Curr. Drug Discov. Technol., 2005, 2, 99-113. Palmer, D.S.; O'Boyle, N.M.; Glen, R.C.; Mitchell, J.B. J. Chem. Inf. Model., 2007, 47, 150-158.
Computational Intelligence Methods [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [194] [195] [196] [197] [198] [199] [200] [201] [202] [203] [204] [205] [206] [207] [208] [209] [210]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 375
Harrison, A.P.; Erlwanger, K.H.; Elbrønd, V.S.; Anderson, N.K.; Unmack, N.K. J. Pharmacol. Toxicol. Methods, 2004, 49, 187-199. Zhao, Y. H.; Le, J.; Abraham, M. H.; Hersey, A.; Eddershaw, P. J.; Luscombe, C. N.; Boutina, D.; Beck, G.; Sherborne, B.; Cooper, I.; Platts, J. A. J. Pharm. Sci., 2001, 90, 749. Abraham, M. H.; Zhao, Y. H.; Le, J.; Hersey, A.; Luscombe, C. N.; Reynolds, D. P.; Beck, G.; Sherborne, B.; Cooper, I. Eur. J. Med. Chem., 2002, 37, 595. Jones, R.; Connolly, P. C.; Klamt, A.; Diedenhofen, M. J. Chem. Inf. Model., 2005, 45, 1337. Raevsky, O. A.; Fetisov, V. I.; Trepalina, E. P.; McFarland, J. W.; Schaper, K. J. Quant. Struct. Act. Rel., 2000, 19, 366. Klopman, G.; Stefan, L. R.; Saiakhov, R. D. Eur. J. Pharm. Sci., 2002, 17, 253. Clark, D. E. J. Pharm. Sci., 1999, 88, 807. Norinder, U.; Osterberg, T.; Artursson, P. Eur. J. Pharm. Sci., 1999, 8, 49. Osterberg, T.; Norinder, U. J. Chem. Inf. Comp. Sci., 2000, 40, 1408. Norinder, U.; Osterberg, T. J. Pharm. Sci., 2001, 90, 1076-1085. Sun, H. M. J. Chem. Inf. Comp. Sci., 2004, 44, 748. Ghuloum, A. M.; Sage, C. R.; Jain, A. N. J. Med. Chem., 1999, 42, 1739. Niwa, T. J. Chem. Inf. Comp. Sci., 2003, 43, 113-119. Wegner, J. K.; Frohlich, H.; Zell, A. J. Chem. Inf. Comp. Sci., 2004, 44, 931. Wang, J.; Krudy, G.; Xie, X.-Q.; Wu, C.; Holland, G. J. Chem. Inf. Model., 2006, 46, 2674-2683. Wessel, M. D.; Jurs, P. C.; Tolan, J. W.; Muskal, S. M. J. Chem. Inf. Comp. Sci., 1998, 38, 726-735. Agatonovic-Kustrin, S.; Beresford, R.; Yusof, A. P. M. J. Pharmaceut. Biomed. Anal., 2001, 25, 227. Zmuidinavicius, D.; Didziapetris, R.; Japertas, P.; Avdeef, A.; Petrauskas, A. J. Pharm. Sci., 2003, 92, 621. Deconinck, E.; Hancock, T.; Coomans, D.; Massart, D. L.; Vander Heyden, Y. J. Pharmaceut. Biomed. Anal., 2005, 39, 91. Xue, Y.; Li, Z. R.; Yap, C. W.; Sun, L. Z.; Chen, X.; Chen, Y. Z. J. Chem. Inf. Comput. Sci., 2004, 44, 1630-1638. Liu, H. X.; Hu, R. J.; Zhang, R. S.; Yao, X. J.; Liu, M. C.; Hu, Z. D.; Fan, B. T. J. Comput. Aided Mol. Des., 2005, 19, 33. Hou, T.; Wang, J.; Li, Y. J. Chem. Inf. Model., 2007, 47, 2408-2415. Palm, K.; Luthman, K.; Ungell, A. L.; Strandlund, G.; Artursson, P. J. Pharm. Sci., 1996, 85, 32. van de Waterbeemd, H.; Camenisch, G.; Folkers, G.; Raevsky, O. A. Quant. Struct. Act. Rel., 1996, 15, 480-490. Krarup, L. H.; Christensen, I. T.; Hovgaard, L.; Frokjaer, S. Pharmaceut. Res., 1998, 15, 972. Ponce, Y. M.; Perez, M. A. C.; Zaldivar, V. R.; Ofori, E.; Montero, L. A. Int. J. Mol. Sci., 2003, 4, 512. Kulkarni, A.; Han, Y.; Hopfinger, A. J. J. Chem. Inf. Comp. Sci., 2002, 42, 331. [193] Hou, T. J.; Zhang, W.; Xia, K.; Qiao, X. B.; Xu, X. J. J. Chem. Inf. Comp. Sci., 2004, 44, 1585. Ren, S.; Lien, E.J. Prog. Drug Res., 2000, 54, 1-23. Nordqvist, A.; Nilsson, J.; Lindmark, T.; Eriksson, A.; Garberg, P.; Kihlen, M. Qsar Comb. Sci., 2004, 23, 303 Segarra, V.; Lopez, M.; Ryder, H.; Palacios, J. M. Quant. Struct. Act. Rel., 1999, 18, 474. Norinder, U.; Osterberg, T.; Artursson, P. Pharmaceut. Res., 1997, 14, 1786. Cruciani, C.; Crivori, P.; Carrupt, P. A.; Testa, B. J. Mol. Struct., 2000, 503, 17. Fujiwara, S.; Yamashita, F.; Hashida, M. Int. J. Pharm., 2002, 237, 95-105. Di Fenza, A.; Alagona, G.; Ghio, C.; Leonardi, R.; Giolitti, A.; Madami, A. J. Comput. Aided Mol. Des., 2007, 21, 207-221. Deim, Z. Drug Dev. Ind. Pharm., 2005, 31, 935-942. Yamashita, F.; Wanchana, S.; Hashida, M. J. Pharm. Sci., 2002, 91, 2230-2239. Guangli, M.; Yiyu, C. J. Pharm. Pharm. Sci ., 2006, 9, 210-21. Young, R. C.; Mitchell, R. C.; Brown, T. H.; Ganellin, C. R.; Griffiths, R.; Jones, M.; Rana, K. K.; Saunders, D.; Smith, I. R.; Sore, N. E.; Wilks, T. J. J. Med. Chem., 1988, 31, 656. Van de Waterbeemd, H.; Kansy, M. Chimia, 1992, 46, 299. Clark, D. E. J. Pharm. Sci., 1999, 88, 815. Iyer, M.; Mishra, R.; Han, Y.; Hopfinger, A.J. Pharm. Res., 2002, 19, 1611-1621. Abraham, M. H.; Chadha, H. S.; Mitchell, R. C. J. Pharm. Sci., 1994, 83, 1257. Platts, J. A.; Abraham, M. H.; Zhao, Y. H.; Hersey, A.; Ijaz, L.; Butina, D. Eur. J. Med. Chem., 2001, 36, 719. Kaznessis, Y. N.; Snow, M. E.; Blankley, C. J. J. Comput. Aided Mol. Des., 2001, 15, 697.
376 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [211] [212] [213] [214] [215] [216] [217] [218] [219] [220] [221] [222] [223] [224] [225] [226] [227] [228] [229] [230] [231] [232] [233] [234] [235] [236] [237] [238] [239] [240] [241] [242] [243] [244] [245] [246] [247] [248] [249] [250] [251] [252] [253] [254] [255] [256]
Hecht and Fogel
Rose, K.; Hall, L. H.; Kier, L. B. J. Chem. Inf. Comp. Sci., 2002, 42, 651. Hutter, M. C. J. Comput. Aided Mol. Des., 2003, 17, 415. Hou, T. J.; Xu, X. J. J. Chem. Inf. Comp. Sci., 2003, 43, 2137. Narayanan, R.; Gunturi, S. B. Bioorg. Med. Chem., 2005, 13, 3017. Cabrera, M. A.; Bermejo, M.; Perez, M.; Ramos, R. J. Pharm. Sci., 2004, 93, 1701. Lombardo, F.; Blake, J. F.; Curatolo, W. J. J. Med. Chem., 1996, 39, 4750. Feher, M.; Sourial, E.; Schmidt, J.M. Int. J. Pharmaceutics., 2000, 201, 239-247. Norinder, U.; Sjoberg, P.; Osterberg, T. J. Pharm. Sci., 1998, 87, 952. Stanton, D. T.; Mattioni, B. E.; Knittel, J. J.; Jurs, P. C. J. Chem. Inf. Comp. Sci., 2004, 44, 1010. Luco, J. M. J. Chem. Inf. Comp. Sci., 1999, 39, 396. Subramanian, G.; Kitchen, D. B. J. Comput. Aided Mol. Des., 2003, 17, 643. Crivori, P.; Cruciani, G.; Carrupt, P. A.; Testa, B. J. Med. Chem., 2000, 43, 2204-2216. Dorronsoro, I.; Chana, A.; Abasolo, I.; Castro, A.; Gil, C.; Stud, M.; Martinez, A. Qsar Comb. Sci., 2004, 23, 89. Ajay; Bemis, G. W.; Murcko, M. A. J. Med. Chem., 1999, 42, 4942-4951. Liu, R. F.; Sun, H. M.; So, S. S. J. Chem. Inf. Comput. Sci., 2001, 41, 1623. Hou, T. J.; Xu, X. J. J. Mol. Model., 2002, 8, 337. Lessigiarska, I.; Pajeva , I.; Cronin, M.T.; Worth, A.P. SAR QSAR Environ. Res., 2005, 16, 79-91. Kortagere, S.; Chekmarev, D.; Welsh, W.J.; Ekins S. Pharm. Res., 2008, 25, 1836-1845. Doniger, S.; Hofmann, T.; Yeh, J. J. Comput. Biol., 2002, 9, 849-864. Zhao, Y.H.; Abraham, M.H.; Ibrahim, A;, Fish, P.V.; Cole, S.; Lewis, M.L.; de Groot, M.J.; Reynolds, D.P. J. Chem. Inf. Model., 2007, 47, 170-175. Li, H.; Yap, C. W.; Ung, C. Y.; Xue, Y.; Cao, Z. W.; Chen, Y. Z. J. Chem. Inf. Model., 2005, 45, 1376. Lewis, D.F.V.; Dickins, M. Toxicology, 2002, 170, 45-53. Lewis, D.F.V.; Modi, S.; Dickins, M. Drug Metab. Rev., 2002, 34, 69-82. Caldwell, J.; Gardner, I.; Swales, N. Toxicol. Pathol., 1995, 23, 102-114. Moon, T.; Chi, M.H.; Kim, D.H.; Yoon, C.N.; Choi, Y.S. QSAR, 2000, 19, 257-263. Bazeley, P.S.; Prithivi, S.; Struble, C.A.; Povinelli, R.J.; Sem, D.S. J. Chem. Inf. Model., 2006, 46, 2698-2708. Ekins, S.; Ring, B.J.; Grace, J.; McRobie-Belle, D.J.; Wrighton, S.A. J. Pharmacol. Toxicol. Methods, 2000, 44, 313-324. Zheng, C.J.; Han, L.Y.; Yap, C.W.; Ji, Z.L.; Cao, Z.W.; Chen, Y.Z. Pharmacol. Rev., 2006, 58, 259279. Kriegl, J. M.; Arnhold, T.; Beck, B.; Fox, T. QSAR Comb. Sci., 2005, 24, 491-502. Terfloth, L.; Bienfait, B.; Gasteiger, J. J. Chem. Inf. Model., 2007, 47, 1688-1701. Kriegl, J.M.; Arnhold, T.; Beck, B.; Fox, T. J. Comput. Aided Mol. Des., 2005, 19, 189-201. Arimoto, R.; Prasad, M.A.; Gifford, E.M. J. Biomol. Screen., 2005, 10, 197-205. White, R.E. Annu. Rev. Pharmacol. Toxicol., 2000, 40, 133-157. Boyer, S.; Zamora, I. J. Comp. Aided Mol. Des., 2002, 16, 403-413. de Groot, M.J.; Alex, A.A.; Jones, B.C. J. Med. Chem., 2002, 45, 1983-1993. de Groot, M.J.; Ackland; Horne, V.A.; Alex, A.A.; Jones, B.C. J. Med. Chem., 1999, 42, 1515-1524. Wang, Q.; Halpert, J.R. Drug Metab. Disp., 2002, 30, 86-95. Ekins, S.; Bravi, G.; Ring, B.J.; Gillespie, T.A.; Gillespie, J.S.; Vandenbranden, M.; Wrighton, S.A.; Wikel, J.H. J. Pharm. Exp. Ther., 1999, 288, 21-29. Ekins, S.; Bravi, G.; Wikel, J.H.; Wrighton, S.A. J. Pharm. Exp. Ther., 1999, 291, 424-433. Afzelius, L.; Masimirembwa, C.M.; Karlen, A.; Andersson, T.B.; Zamora, I. J. Comput. Aided Mol. Des., 2002, 16, 443-458. Ekins, S.; Bravi, G.; Binkley, S.; Gillespie, S.; Ring, B.J.; Wikel, J.H.; Wrighton, S.A. Drug Metab. Disp., 2000, 28, 994-1002. Poso, A.; Gynther, J.; Juvonen, R. J. Comput. Aided Mol. Des., 2001, 15, 195-202. Rao, S.; Aoyama, R.; Schrag, M.; Trager, W.F.; Rettie, A.; Jones, J.P. J. Med. Chem., 2000, 43, 27892796. Jensen, B.F.; Vind, C.; Padkjær, S.B.; Brockhoff, P.B.; Refsgaard, H.H.F. J. Med. Chem., 2007, 50, 501-511. Coe, K.J.; Nelson, S.D.; Ulrich, R.G.; He, Y.; Dai, X.; Cheng, O.; Caguyong, M.; Roberts, C.J.; Slatter, J.G. Drug Metab. Dispos., 2006, 34, 1266-1275. Slatter JG, Cheng O, Cornwell PD, de Souza A, Rockett J, Rushmore T, Hartley D, Evers R, He Y, Dai X, Hu R, Caguyong M, Roberts CJ, Castle J, Ulrich RG. Xenobiotica, 2006, 36, 902-937.
Computational Intelligence Methods [257] [258] [259] [260] [261] [262] [263] [264] [265] [266] [267] [268] [269] [270] [271] [272] [273] [274] [275] [276] [277] [278] [279] [280] [281]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 377
Slatter, J.G.; Templeton, I.E.; Castle, J.C.; Kulkarni, A.; Rushmore, T.H.; Richards, K.; He, Y.; Dai, X.; Cheng, O.J.; Caguyong, M.; Ulrich, R.G. Xenobiotica, 2006, 36, 938-962. Katritzky, A.R.; Karelson, M.; Lobanov, V. Pure Appl. Chem., 1997, 69, 245-248. Trotter, M.W.B.; Holden, S.B. QSAR Comb. Sci., 2003, 22, 533-548. Colmenarejo, G.; Alvarez-Pedraglio, A.; Lavandera, J.-L. J. Med. Chem., 2001, 44, 4370-4378. Kratochwil, N.; Huber, W.; Muller, F.; Kansy, M.; Gerber, P.R. Biochem. Pharm., 2002, 64, 13551374. Manallack, D.T.; Livingstone, D.J. Eur. J. Med. Chem., 1999, 34, 195-208. Yao, X.; Liu, H.; Zhang, R.; Liu, M.; Hu, Z.; Panaye, A.; Doucet, J.P.; Fan, B. Mol. Pharm., 2004, 2, 348-356. Gunturi, S.B.; Narayanan, R.; Khandelwal, A. Bioorg. Med. Chem., 2006, 14, 4118-4129. Saiakhov, R.D.; Stefan, L.R.; Klopman, G. Persp. Drug Disc. Des., 2000, 19, 133-135. Nestorov, I.; Gueorguieva, I.; Jones, H.M.; Houston, B.; Rowland, M. Drug Metab. Dispos., 2002, 30, 276-82. Ghafourian, T.; Barzegar-Jalali, M.; Dastmalchi, S.; Khavari-Khorasani, T.; Hakimiha, N.; Nokhodchi, A. Int. J. Pharm., 2006, 319, 82-97. Shen, M.; Xiao, Y.; Golbraikh, A.; Gombar, V.K.; Tropsha, A. J. Med. Chem., 2003, 46, 3013-3020. Sakiyama, Y.; Yuki, H.; Moriya, T.; Hattori, K.; Suzuki, M.; Shimada, K.; Honma, T. J. Mol. Graph. Model., 2008, 26, 907-915. Yoshida, K.; Niwa, T. J. Chem. Inf. Model., 2006, 46, 1371-1378. Klein, C.D.P.; Hopfinger, A.J. Pharm. Res., 1998, 15, 303-311. Jia, L.; Sun, H. Bioorg. Med. Chem., 2008, 16, 6252-6260. Li, Q.; Jørgensen, F.S.; Oprea, T.; Brunak, S.; Taboureau, O. Mol. Pharm., 2008, 5, 117-127. Leong, M.K. Chem. Res. Toxicol., 2007, 20, 217-226. Chekmarev, D.S.; Kholodovych, V.; Balakin, K.V.; Ivanenkov, Y.; Ekins, S.; Welsh, W.J. Chem. Res. Toxicol., 2008, 21, 1304-1314. Agatonovic-Kustrin, S.; Ling, L.H.; Tham, S.Y.; Alany, R.G. J. Pharm. Biomed. Anal., 2002, 29, 103119. Judson, R.; Elloumi, F.; Stzer, R.W.; Zhen, L.; Shah, I. BMC Bioinformatics, 2008, 9, 241-257. Zhao, C.Y.; Zhang, H.X.; Zhang, X.Y.; Liu, M.C.; Hu, Z.D.; Fan, B.T. Toxicology, 2006, 217, 105119. Young, M.B.; DiSilvestro, M.R.; Sendera, T.J.; Freund, J.; Kriete, A.; Magnuson, S.R. Pharmacogenomics J., 2003, 3, 41-52. Mazzatorta , P.; Tran, L.-A.; Schilter, B.; Grigorov, M. J. Chem. Inf. Model., 2007, 47, 34-38. Yuan, H.; Wang, Y.; Cheng, Y. J. Chem. Inf. Model, 2007, 47, 159-169.
378
Frontiers in Drug Design & Discovery, 2009, 4, 378-428
Data Modeling and Chemical Interpretation of ADME Properties Using Regression and Rule Mining Techniques Kiyoshi Hasegawa1 and Kimito Funatsu2,* 1
Kamakura Research Laboratories, Chugai Pharmaceutical Co., LTD, 200 Kajiwara, Kamakura, Kanagawa, 247-8530, Japan and 2Department of Chemical System Engineering, The University of Tokyo, Hongo 7-3-1, Bunkyo-ku, Tokyo, 113-8656, Japan Abstract: In pharmaceutical industry, in addition to high potency, good absorption, distribution, metabolism and excretion (ADME) profiles of compounds are needed for drug development. Data modeling of ADME model is a crucial step for efficient drug design. However, we have to avoid so called a situation, the black box, which might be difficult for chemical interpretation. The right balance between precision and interpretation is required for practical drug design. We review the related articles to focus several ADME modeling techniques. As regression, multiple linear regressions (MLR), partial least squares (PLS), artificial neural networks (ANN), support vector machines (SVM) are picked up and their algorithms and the representative applications are introduced. We pay attention to rule mining methods for chemical interpretation. As rule mining, rough set theory (RST) is shown as an example. Visualization is a classical but never neglected technique for easily understanding the overall behaviors of huge compounds. We especially spend more pages about kohonen neural networks (KNN) and decision trees (DT) as the representative methods. Furthermore, web application for chemists is another important aspect for practical drug design. Recent trend about this topic is shown in two industry cases. As conclusion, we will show future direction concerning in silico ADME prediction.
1. INTRODUCTION In a process of potential leads to candidates suitable for clinical trials, numerous hurdles must be overcome. In addition to adequate potency and safety, candidate compound must have optimal pharmacokinetic and metabolic properties. These properties are called as absorption, distribution, metabolism and excretion (ADME) properties in general terms [1]. Behind tight relationship between inadequate ADME properties and
*Corresponding Author: Tel: +81(03)-5841-7751; Fax: +81(03)-5841-7771; E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 379
failure of candidate compounds, optimizing ADME properties is a critical step in drug discovery [2]. Traditionally, pharmaceutical company has been embedded in a serial process whereby potency test is carried out first, followed by ADME tests that are essential to reveal the profile of candidates. In this scenario, if an ADME problem were to arise, it was late in the development stage and not much could be done to remedy the problem. This scenario has dramatically changed under the pressure of cost reduction. Many pharmaceutical companies have recently employed parallel processing, so that potency test and surrogate ADME tests are carried out in close proximity [3, 4]. As a result, high-throughput instruments for ADME profiles have been invested in pharmaceutical companies [5]. Early ADME studies can provide the necessary information to learn more about the fundamental mechanisms and specific properties that are used to select better compounds [6]. For example, in vitro metabolism studies in liver microsome can provide information as to which compounds in the chemical series are more metabolically stable to cytochrome P450 (CYP) enzymes. Absorption models can provide information to help rank order compounds to their ability to permeate the cell membrane and can be used as a predictor of intestinal absorption. Inhibition studies can provide valuable information on the potential for inhibition of CYP enzymes. Those kinds of information, if obtained early, can help guide chemists to modify the chemical series as means to design better candidate compounds. These studies can also help limit the number of compounds moving forward to the more time-consuming ADME studies and provide specific guidance to help better select those compounds moving forward towards drug development [4]. The advent of in vitro methods for measuring specific ADME properties have led to an increase in the availability of data on a wide range of compounds, making it possible to investigate the rules by which the chemical structure of a molecule determine its ADME properties, to build predictive models of ADME properties. The most well known is Rule-of-five [7]. Compounds are most likely to have poor absorption when molecular weight > 500, calculated octanol-water partition coefficient Clogp > 5, number of H-bond donors > 5 and number of H-bond acceptors > 10. Computation of these properties is now used as a simple but efficient ADME filter. The Rule-of-five should be seen as a qualitative absorption/permeability predictor, rather than a quantitative predictor. While calculated simple filters may be sufficient in library design, more sophisticated ADME models are required in lead optimization [8]. In silico modeling of ADME properties can be broadly divided into three categories: molecular modeling [9], physiologically based pharmacokinetic (PBPK) modeling [10] and data modeling [11, 12]. Molecular modeling approaches include quantum and classical mechanical methods, homology modeling and pharmacophore models and can be used where the underlying molecular mechanism of a property is fully understood. PBPK modeling integrates several factors responsible for ADME processes in one model and attempts to simulate the pharmacokinetics (PK) of compound in the whole organism. Data modeling is applied when the molecular mechanism of an ADME property is not clear or cannot be efficiently modeled at the molecular level and largely uses several statistical approaches. In this review, we will concentrate on data modeling techniques [13]. Other two approaches are beyond of this review and readers who are interested in should refer to the following literatures [14, 15].
380 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
This article will review the progress made from 2000 years in the development of predictive ADME models and the current state-of-the-art [16-18]. We will discuss recent trends in four fundamental aspects (regression, rule mining, visualization and web application). Other important topics (partition of training/test sets, chemical descriptors, descriptor selection and domain of applicability), when building any predictive models, are not explicitly described here. Concerning these topics, the following review should be referred to [19]. As regression, multiple linear regressions (MLR), partial least squares (PLS), artificial neural networks (ANN), support vector machines (SVM) are picked up and their algorithms and the representative applications are introduced. We pay attention to rule mining methods for chemical interpretation. As rule mining, RST (rough set theory) is shown as an example. Visualization is a classical but never neglected technique for easily understanding the overall behaviors of huge compounds. We especially spend more pages about kohonen neural networks (KNN) and decision trees (DT) as the representative methods. Furthermore, web application for chemists is another important aspect for practical drug design. Recent trend about this topic is shown in two industry cases. As conclusion, we will show future direction concerning in silico ADME prediction. In the next section, we will introduce each method with short explanation of the algorithm followed by the representative examples. 2. REGRESSION 2.1. MLR Multiple linear regressions (MLR), called Hansch-Fujita approach, are the most widely used linear correlation method [20, 21]. Its basic assumption is that ADME property is expressed as the additive combinations of each chemical descriptor. K
y = bj x j + C
(1)
j =1
where y is ADME property. xj is descriptor variable. bj is the regression coefficient. C is the constant value. K is the number of descriptors. Least squares are used to estimate the regression coefficient bj. The great advantage of MLR is that a causal model is obtained and the physical meaning is obvious due to the simple description. However, the following severe conditions must be satisfied to apply MLR: The number of compounds should be at least five times greater than that of the descriptors. The descriptors should be independently distributed [21]. Though these statistical limitations, many MLR equations have been proposed as ADME models [22, 23]. Applications with MLR method are cited in literatures [24-35]. Hasegawa et al. have performed a quantitative structure-pharmacokinetic relationship (QSPKR) study of antifungal N-myristoyltransferase (Nmt) inhibitors [36]. For predicting rat elimination half-life (t1/2) values, they have constructed a comprehensive multivariate statistical analysis based on various chemical descriptors. The t1/2 values of 105 inhibitors were obtained by cassette dosing experiments in a high-throughput manner. The 30 physicochemical descriptors were generated for an entire 3D structure as well as for an individual fragment (Table 1). The correlation between the t1/2 values and the chemical descriptors was examined by a stepwise MLR based on the F values. They have obtained a significant MLR model with just only three variables.
Data Modeling and Chemical Interpretation
Table 1.
Frontiers in Drug Design & Discovery, 2009, Vol. 4 381
Chemical Structures and Observed Three PK Parameters in the First Series. Cited from Table 1 in Ref. [36] N O
N
O
O
R
1-60
No.
-R
In (t1/2)
In (Cl)
In (Vss)
-0.386
2.104
1.946
-0.274
2.208
2.079
0.501
2.241
2.901
0.663
2.219
2.986
1188
-0.105
1.099
2126
0.916
3.040
2.454
0.470
3.016
2.518
0.262
2.827
CH3
1 O O
O
2
CH3
O
CH3 CF3
3 O O
CF2CF2H
4 O O
5
O
CF3 CF3
O CF3
6 O
CF3 O CF2CF3
7 O O
CF2CF2CF3
8 O O
382 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu (Table 1) contd....
N O
N
R O 61-83
No.
-R
In (t1/2)
In (Cl)
In (Vss)
0.4332
1.131
1.841
0.621
1.194
1.960
1.008
1.099
2.282
0.476
1.723
2.407
0.647
1.825
2.639
0.673
1.569
2.416
0.779
1.194
2.067
-0.994
1548
0.833
O
27
O
F
28
F
O
29 O
CF3 F
30
F
F
31
F F F
32
F
F F
33
F F F
34
N
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 383
ln(t1/2) = 0.011*PNSA(157.04) + 0.198*AlogP(63.09) – 0.081 *Rotlbonds(6.57) – 2.330 (n = 105, R = 0.843, Q = 0.827)
(2)
In Eq. 2, the numbers in parentheses are the F value of each variable, n is the number of compounds, R is the correlation coefficient and Q is the cross-validated version of R derived from the leave-one-out (LOO) procedure. The most important factor influencing the model is PNSA according to the F value. PNSA means the sum of the solventaccessible surface areas of negatively charged atoms such as oxygen, nitrogen, and halogen atoms. The model means that PNSA should be kept to a maximum value by introducing negatively charged groups to the terminal part of a chain from the C-2 position. The high lipophilicity (AlogP) of the compound is an advantage for long t1/2. Fewer rotational bonds (Rotlbonds) are also beneficial for the compound to remain in the plasma. The authors have stressed that the combination of high-throughput cassette dosing experiments and the multivariate statistical approach is helpful in designing new Nmt inhibitors with a long t1/2 value. Colmenarejo et al. have proposed the MLR model for predicting binding affinities to human serum albumin (HSA) [37]. They have determined the binding affinities to HSA of 95 diverse drugs by high-performance affinity chromatography. All the compounds but one (Captopril) in the database were used to generate the MLR model. The dataset was split into two subsets: a training one (84 molecules) and an external validation one (10 molecules). Genetic algorithm (GA) was used to exhaustively search and select multivariate equations, starting from 53 chemical descriptors. The resulting best model was logK’hsa = -0.607873 + 0.06784*(HbondDon – 3)2 – 9*10-6*(JursTPSA) – 0.028261*(EHOMO + 7.4076)2 + 0.005697*(AM1dip2) + 0.182595*(ClogP) + 2.33529*(6ring) (n = 84, R2 = 0.83, Q2 = 0.79)
(3)
Both the goodness of the fit and the predictive power of the model from the cross validation (CV) was achieved. When applying the logK’hsa of the external validation set, significant correlation was observed (Rpred2 = 0.82). Fig. (1) displays the descriptor usage versus the number of generations in the GA evolution. It can be seen that good convergence actually is achieved. (Especially, ClogP, HbondDon, 6ring and AM1dip) From the equation, an increase of hydrophobicity (ClogP) within a series of compounds is expected to result in an increased HSA binding. 6ring is a sixth-order, ring type Kier and Hall topological index. The equation indicates that molecules with non-substituted six-member rings are expected to bind more tightly to HSA. JursTPSA is the sum of solvent-accessible surface area of atoms with the absolute value of partial charges greater than or equal to 0.2. The equation indicates the logK’hsa is inversely proportional to this descriptor. Therefore, binding is favored for molecules with large non-polar surfaces. Other additional factors, like the number of hydrogen bond donors (HbondDon), HOMO energy (EHOMO) and AM1 dipole moment (AM1dip) can be important in determining HSA binding to some extent. The binding to HSA turns out to be determined by a combination of hydrophobic forces together with some modulating shape factors. Yoshida et al. have studied to construct a predictive model for human oral bioavailability (BA) of 232 structurally diverse drugs [38]. The oral BA was assigned one of four categories and analyzed in relation to chemical descriptors by the ORMUCS (ordered multi-categorical classification) method. ORMUCS is a modified form of discriminant MLR analysis using a simplex algorithm. Two lipophilicity measures, expressed as the distribution coefficient at pH 6.5 (logD6.5), and the difference between the fractions of the
384 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
usage vs. # of crossovers 1.2 Variable x103 #
ClogP use Hbond donor use K i e r C h i 6 use Jurs-TPSA use AM-1 dipole use HOMO MOPAC use Jurs-RNCG use AlogP use Rotlbonds use Jurs-TASA use
1.0
O f t i m e s v a r i a b l e
0.8
0.6
0.4
i s u s e d
0.2
0.0 0.0
0.5
1.0
1.5
# of crossovers
2.0
2.5
3.0 x10 5
Fig. (1). Variable usage vs genetic algorithm generations for the second global model. Selection of model descriptors during evolution can be seen, as well as the achieved convergence. Cited from Fig. (6) in Ref. [37].
neutral form at two given pH values (logD = logD6.5 - logD 7.4) were found to be significant factors influencing BA. The addition of 15 structural descriptors relating primarily to well-known metabolic processes and the squared term of logD6.5 yielded a satisfactory equation. The list of 18 descriptors is presented in Table 2. The final model with 18 descriptors has a correct classification rate of 71% and a Spearman rank correlation coefficient (Rs) of 0.851. In LOO, an average of 67% of drugs are correctly classified with Rs of 0.812. The predictive power of the model was evaluated using a separate test set of 40 compounds, of which 60% are correctly classified. The important role of lipophilicity terms in the final model is shown by their contribution index (CI) values. From the final equation, the optimum logD6.5 is -0.3 and a progressive negative impact on BA are seen as values move away from this level. The logD descriptor might adjust the logD of a
Data Modeling and Chemical Interpretation
Table 2.
Frontiers in Drug Design & Discovery, 2009, Vol. 4 385
The QSAR Model. Cited from Table 5 in Ref. [38] S(X) = wisi
No.
Descriptors Si
Weight wi
CIa
nb
1
(log D6.5)
-0.27
0.05
232
-0.046
0.25
232
0.370
0.23
232
2
2
(log D6.5)
3
log D (log D6.5 – log D 7.4) c
4
phenolic OH (excluding di-ortho-subst)
-1.032
0.45
22
5
SO2NH 2
-1.014
0.17
7
6
alcoholic OH (excluding tert-OH)d
-0.177
0.09
59
7
hydrolysis: esters, lactones, -lactams, alkyl carbamates
-1.074
0.37
24
-0.599
0.26
33
-0.235
0.12
47
-0.201
0.09
13
8 9
e
f
aromatic p-hydroxylation g
ArCH2-R (excluding di-ortho subst Ar) h
10
allylic oxidation (C-C=C)
11
tert-alicyclic amine (no ring heteroatoms)i
-0.340
0.10
24
12
XCCNR (R = Me, Et; X = N,O, Ar, C=C)j
-0.410
0.15
28
13
readily oxidized moieties: thiols, dihydropyridines
-1.137
0.24
11
-0.493
0.10
15
k
14
ketones
15
NO2 on a benzene ring (excluding ortho subst)
-0.148
0.03
7
16
ArNH2, ArNHNH2 , ArCONHNH2, ArC(=NH) NH2 as pKa value1
-0.034
0.04
16
17
HOCCNH tert-alkyl, HOCCN < (cyclic rings)
0.210
0.05
16
18
benzodiazepine (with no additional fused rings)
0.231
0.05
10
constant
4.358
n- 232 (four classes); boundaries, 2.0, 3.0, 4.0; recognition, nmis = 67 (8), Rs = 0.851 (p < 0.0001); leave-one-out, nmis = 76 (10), Rs = 0.812 (p < 0.0001) a Contribution index. The product of the weighting coefficient and standard deviation for each descriptor. b n = total number of each of the descriptors used in the analysis. c Except with o-CO2H, o-CONH2 and o-CH2OH substituents, which can undergo intramolecular hydrogen bonding with the phenolic OH group. d For steroids this descriptor is 0 for 11--OH substituents (steric hindrance) and 2 for 17--OH substituents (unless tert) due to high susceptibility to first-pass metabolism. e Weighting is 0.5, where the carbon to the carbonyl is tertiary or where the carbonyl can undergo intramolecular hydrogen bonding with a nearby group. f Applies where there is an open para position with respect to the activating groups OR, N(R)R1, NHC(=O) R (R, R1 = H, alkyl, aryl, aralkyl) with no ring substituents beyond one ortho to the activating group. The activating group and ortho substituent may part of a fused ring. g R = H, CH2X, where X = C or H and is not attached to a polar atom. h Excluding steroidal ring A dienones and allylic substructures of types C=C-C-X and C=C-C-C-X, where X is a polar atom. i Excluding ring systems with a bridged N atom such as quinine and quinidine. j Weighted by 2 for O=CCNR; see the . k Excluding , unsaturated dienones, diaryl-ketones, and ketones with a heteroatom attached at the -position. Corrected (-0.5) for branching on the aliphatic side and , -unsaturation. i If a molecule contains two or more amino groups the most basic group with no ortho substitution is selected.
386 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
compound caused by the change in pH following intestinal absorption. The presence of phenolic OH and SO2NH2 groups have marked and highly significant effects in reducing oral BA. Hydrolytic cleavage of susceptible functions such as esters, lactones, -lactams, and alkyl carbamates is a strong and highly significant contributor in reducing BA. Hydroxylation of an activated aromatic ring has the largest effect in reducing BA, with aryl methyl and allylic groups contributing to a lesser extent. The model has the advantage of transparency for chemical interpretation. This means which factors may affect BA and the extent of that effect, thus providing the basis for designing improved compounds. 2.2. PLS As mentioned in the section of MLR, MLR has severe statistical limitations when applying. In order to overcome theses limitations, partial least squares (PLS) have been invented and developed in chemometrics [39]. PLS finds some latent variables on which to perform regression. These latent variables are chosen to simultaneously satisfy two conditions: (a) that they are highly correlated with ADME property and (b) that they model as much of variations among descriptor variables as possible. In PLS, the descriptor variables X and ADME property y are modeled by the latent variable t. A
X = t h p hT + E
(4)
h =1
A
y = t h qh + f
(5)
h =1
where ph and qh are the loadings for X and y blocks in the h-th component, respectively. E is a matrix of X residuals, and f is a vector of y residuals. A is the optimum number of components determined by the LOO procedure. The latent variable t is a linear combination of the descriptor variables X and the weight vector w.
t = Xw
(6)
If Eq. 6 is substituted into Eq. 5, then the MLR-like model that can be directly comparable to the Hansch equation is obtained.
y = XW ( PW ) 1 q = Xb
(7)
where W, P and q are the weight matrix, loading matrix and coefficient vector, respectively. b is the regression coefficient of the MLR-like model equation. Applications with PLS method are cited in literatures [40-51]. Gleeson et al. have constructed the model for human volume of distribution (Vss) of 199 marketed drugs [52]. They have conducted in parallel with model building for rat Vss based on a data set of 2086 in-house compounds. The two data sets were randomly partitioned into training and test sets, 75%/25% for the human data and 80%/20% for rat. The 123 descriptors, which broadly describe lipophilicity, size, topological, geometrical, and electronic features of molecules, were calculated. The results of the human Vss model in fit and prediction using PLS were 0.641 of R2, 0.597 of Q2 and 0.587 of Rpred2, respectively. The first component in the human model describes 51% of the variance and the second 13%. Component 1 (t1) essentially describes the lipophilicity/molecular
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 387
weigh dependency of Vss, and the second component (t2) describes the charge state or charge distribution. Fig. (2) shows the overall normalized/scaled coefficients. Basic moieties/large positive charge and lipophilicity based descriptors have positive coefficients. In contrast, acidic/negative charge based descriptors have negative coefficients. The results of the rat Vss model in fit and prediction using PLS were 0.519 of R2, 0.506 of Q2 and 0.463 of Rpred2, respectively. t1 (45% of the explained variance) of the PLS model relates to lipophilicity/molecular weigh, t2 (5% of the explained variance) to charge state/charge distribution, and t3 (2% of variance) to descriptors related to aromatic molecular features. Fig. (3) shows the overall normalized/scaled coefficients. Basic/positive charge and lipophilicity based descriptors have a positive coefficient. In contrast, acidic/negative charge based descriptors have negative coefficients. Comparing Figs. (2) and (3), the two trends of coefficients are relatively similar. So, the authors have attempted to predict the rat data using the human model and vice versa. The human model predicted the rat data with root mean squares error (RMSE) of 0.38. The rat model predicted the human data with RMSE of 0.50. To understand this difference, PCA was performed against the combined data set. Fig. (4) shows the PCA score plot. From the plot, it can be seen that while the majority of human compounds have rat nearneighbors, the converse is not true. The chemical space of human data is larger than that of rat data. This may explain why the human model predicts the rat results but not vice versa. This is the first reported design and application of entirely in silico models for the prediction of an in vivo PK parameter. The authors have pointed out that the predictive
Variable Coetficient
0.20
0.10
0.00
-0.10
Pos_ioniz
MaxRing2
Neg_ioniz
Lipinski
HYBOT_max_donor
HMO_reson _energy
VDW_POL_AREA
Aver_neg _charge_G_H
NPat
POSCH
MWNPat
NEGCHARGED
MM_ZAP_PCR1
MM_VDW_EP_P_VAR
MM_MAXPOS
MM_SOLVNRG
MM_HDSA
MM_HASA
MM_FHASA
ACDLogP _v70
ACDLogD65_v70
Amine2
Amine3
-0.20
Variable ID
Fig. (2). Coefficients derived from the descriptor human Vss PLS model. (r2 = 0.64, q2 = 0.60, comp = 2). Red descriptors are acid/negative charge descriptors/indicators, blue descriptors are base/positive charge descriptors/indicators, black are lipophilicity/size based descriptors, gray are size/aromaticity based descriptors, and yellow are others. Cited from Fig. (7) in Ref. [52].
388 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
0.20
Variable Coetficient
0.10
0.00
Posioniz
MineV1
Negioniz
MaxRing3
Max_neg_charge_Gast
KierChi4p
KierChi5p
Charge_range_G_H
HMO_HOMO_energy
Aver_neg_charge_G_H
Aver_neg_charge_Gast
POSCH
POSCHARGED
NEGCH
NEGCHARGED
MWNPat
M2M
CLOGP
HAROM
Amine3
-0.20
CHARGED
-0.10
Variable ID
Fig. (3). Coefficients derived from the descriptor rat Vss PLS model. (r2 = 0.52, q2 = 0.51, comp = 3). Red descriptors are acid/negative charge descriptors/indicators, blue descriptors are base/positive charge descriptors/indicators, black are lipophilicity/size based descriptors, gray are size/aromaticity based descriptors, and yellow are others. Cited from Fig. (8) in Ref. [52]. 12 10
Component Two (t2)
8 6 4 2 0 -2 -4
Human Rat
-6
Component One (t1)
Fig. (4). PCA model showing the relationship between the human and rat data sets using 23 key physicochemical descriptors. The first two components are shown, describing 61% of total variance in the data set: component 1 (35%) and component 2 (26%). The human data set differs significantly from the rat based data set on component 1. Cited from Fig. (12) in Ref. [52].
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 389
ADME models together with suitable high-quality data sets may prove useful in the early stages of drug discovery prior to resource-intensive chemical synthesis and data acquiring.
1.35 0.00 -1.35 -2.69 -4.04 -5.38
BBB BBB
-8.07
-6.73
PLS score (Latent Variables 2)
2.69
4.04
5.38
Crivori et al. have developed a quantitative model for blood brain barrier (BBB) permeation [53]. All compounds were classified into two categories: either brain penetrating (BBB+) or no brain penetrating (BBB-). The chemical descriptors were the originally developed VolSurf descriptors that are calculated from 3D interaction energy with chemical probes. They have searched for a relation between the 3D structure and the BBB permeation of the dataset consisting of 229 compounds and 72 VolSurf descriptors. PLS discriminant analysis was carried out, assigning to the BBB+ compounds a score +1, and to the BBB- compounds a score -1. Two significant latent variables emerged from the PLS model. The PLS t1-t2 score plot of the resulting model is shown in Fig. (5). The model can distinguish well between the BBB+ and BBB- compounds. The model correctly predicted more than 90% of the BBB permeation data. Since the prediction error of the discriminant PLS was 0.6 units, a confidence interval was built in the t1t2 space between the BBB+ and BBB- regions, as shown in Fig. (5). In this interval, BBB prediction can be borderline and doubtful. The coefficient plot of the model (Fig. (6)) reports the contribution of all VolSurf descriptors. Hydrophilic region descriptors
-12.74
-10.19
-7.64
-5.10
-2.55
0.00
2.55
5.10
7.64
10.19
12.74
PLS scores (Latent Variable 1) Fig. (5). Discriminant PLS t1-t2 score plot for the global model. The model offers a good discrimination between the BBB+ and BBB- compounds, since it assigned a correct BBB profile to > 90% of the compounds. A confidence interval is built in the t1-t2 space, where BBB prediction can be borderline and doubtful. , BBB+ compounds; , BBB- compounds. Cited from Fig. (4) in Ref. [53].
390 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
0.04 BBB
0.03
0.01
Hydrophobic regions
Integy Moments
0.02
Critical Packing
Size & Shape
0 -0.01 -0.02 Amphiphilic Moments
-0.03
H-Bonds
-0.04
Hydrophilic regions
Capacity Factors
-0.05
Fig. (6). PLS coefficient plot for the global model (training and test sets combined) for the correlation of VolSurf descriptors with blood-brain barrier permeation. Shading refers to the different energy levels used. Cited from Fig. (5) in Ref. [53].
refer to polar water-accessible surface areas, indicating the BBB permeation decreases when the polar surface increases. Capacity factor descriptors refer to polar interactions per surface unit. While diffuse polar regions are tolerable for BBB permeation, dense and localized polar regions are markedly detrimental. An increase in H-bonding capacity is known to be detrimental for permeation. The contribution of the integy moment demonstrates that, besides the number of H-bonds, their 3D distribution also influences BBB permeation. The descriptors of hydrophobic interactions are directly correlated with BBB permeation, but their role appears smaller than that of the polar descriptors. The size and shape descriptors have no marked impact on BBB permeation. In contrast, critical packing and the hydrophilic-lipophilic balance are important descriptors. The authors have pointed out the advantages of VolSurf descriptors. VolSurf descriptors are independent of the alignment of molecules and relatively independent of conformational sampling and averaging. VolSurf descriptors are also fast to compute and easy to interpret. Singh et al. have developed a rapid semi-quantitative model for evaluating the relative susceptibilities of different sites on 50 drug molecules by CYP3A4 [54]. The model is based on the energy necessary to remove hydrogen radical from each site, plus the surface area exposure of the hydrogen atom. The energy can be estimated by AM1 semiempirical molecular orbital calculation. However, AM1 calculations take too long to be a practical usage. Therefore, they have developed a statistical trend vector model, which is used to estimate the AM1 hydrogen abstraction energy. That is, in their study, y is hydrogen abstraction energy and X is trend vector matrix in Eq. 7. Trend vectors can
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 391
capture the local topological environment of the hydrogen. The trend vectors are of the form H-AT1, H-AT1-AT2, H-AT1-AT2-AT3, H-AT1-AT2-AT3-AT4, where AT1 is the type of atom to which the hydrogen is directly bonded, AT2 , the atom two bonds away, etc. After PLS analysis, the significant model with 14 components were obtained. The R2 for 3984 hydrogens was 0.98, and the standard error of the prediction was 2.06 kcal/mol. A plot of the estimated lowest energy sites (circles) and metabolic site energies (asterisks) versus surface area is shown in Fig. (7). It is evident from Fig. (7) that none of the hydrogens with surface area exposure < 8 Å2 are susceptible to metabolism (no overlap of circles and asterisks). If we consider all hydrogens with hydrogen abstraction energies < 27 kcal/mol and surface area exposure > 8 Å2, these hydrogens include a major metabolic site in the CYP3A4 substrates 78% of the time. The authors have concluded that this simple thumb can suggest likely sites of metabolism for compounds that could potentially be CYP3A4 substrates. 50 CYP3A4 Substrates
Solvent accessible surface area (A * *2)
25.0
Trend vector hydrogen abstraction energy vs. surface area
20.0
15.0
10.0
5.0 Low est energies Metabolic energies Surface area line at BA
0.0 0.0
10.0
20.0
30.0
40.0
50.0
60.0
TV predicted hydrogen abstraction energy (Kcal/mol)
Fig. (7). Plot of trend vector predicted hydrogen abstraction energy versus the surface area. Cited from Fig. (5) in Ref. [54].
Kaneko et al. have proposed ICA (independent component analysis) approach and applied it to a quantitative structure-property relationship (QSPR) analysis of aqueous solubility [55]. ICA is a method that extracts mutually independent components from descriptor variables. The public aqueous data set (logS) was divided into a training set of 878 molecules and a test set of 412 molecules. After ICA, 169 independent components were extracted from 173 chemical descriptors. The relationship between the 169 components and logS was modeled by using MLR with the least-squares manner. The R2, Q2,
392 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
and Rpred2 values were 0.937, 0.868, and 0.894, respectively. Fig. (8) shows the regression coefficient values for each independent component. There are three components whose absolute value is larger than 0.2, and their numbers are 52, 101, and 105. Fig. (9) shows the weight values for these components (w52, w101 and w105). Weight values give information on contribution of each chemical descriptor to each independent component. Table 3 shows chemical descriptors corresponding to each weight whose absolute values are larger than 1.0. The value of w52 corresponding to SSS(12-1_dCH-) is positive, and that of bSy corresponding to s52 is negative. SSS(12-1_dCH-) contributes to logS negatively, because a positive value times a negative value equals a negative value. Molecule is insoluble in water if the number of aromatic bonds in the molecule is large, because substructure =CH- of SSS(12-1_=CH-) is included mostly in aromatic rings. Other factor, WTPT3 also plays significantly role in water solubility. WTPT3 is the descriptor that represents like reciprocal size of molecules. This descriptor contributes to logS positively from two signs of bSy and w101. Molecule is soluble in water if the size is small. Table 3 shows that logP also contributes to logS significantly. The authors have concluded that more simple and interpretable model could be constructed by selecting important independent components. 0.3
value of b
0.2 0.1 0 -0.1 -0.2 -0.3 0
20
40
60
80
100
120
140
160
independent component Fig. (8). Value of b to each independent component. Cited from Fig. (7) in Ref. [55].
2.3. ANN Artificial neural networks (ANN) are computer-modeled system containing a number of nodes that are connected into net-like structure. ANN can handle the non-linear relationships where the standard linear MLR and PLS are useless. The network consists of input, hidden and output layer nodes which are schematically drawn as circles and connected by bonds with connection weights as shown in Fig. (10) [56, 57]. Each input layer node obtains descriptor variables X. Similarly, the output layer node produces ADME property y. Each of the input and hidden layers has an additional bias node for accommodating non-zero offset in the data modeling. The bias node obtains a signal with intensity 1.0 and distributes the signal to the next layer. If descriptor variables are given in the input layer, the input nodes only serve as distributors of input signals to the hidden layer. The net input sa is calculated by Eq. 8 in the a-th hidden node and it is transformed by the sigmoidal functions f(sa) by Eq. 9 to give the output ta of the hidden node.
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 393
1.2 1
W52
0.8 0.6 0.4 0.2 0 -0.2 0
20
40
60
80
100
120
140
160
180
120
140
160
180
120.0
140.0
160.0
180.0
W101
descriptor 9 8 7 6 5 4 3 2 1 0 -1 -2 -3 -4 -5 -6 0
20
40
60
80
100
W105
descriptor 4 3 2 1 0 -1 -2 -3 -4 -5 -6 0
20
40
60
80
100.0
descriptor
Fig. (9). Value of w to each independent component on that the value of b is larger than 0.2. Cited from Fig. (8) in Ref. [55].
D
s a = Wap X p + a
(8)
t a = f ( s a ) = (1 + exp( s a ) / T )) 1
(9)
p =1
394 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 3.
Hasegawa and Funatsu
Descriptors to Each w Whose Absolute Values are Larger than 0.1. Cited from Table 2 in Ref. [55]
Abbreviation
Descriptor Name
NO
w52
w101
w105
The number of oxygens
-1.6
2.0
NN
The number of nitrogens
-4.8
2.6
NS
The number of sulfurs
-1.5
NF
The number of fluorines
-1.2
Ncl
The number of chlorines
-3.2
NSB
The number of single bond
-1.1
NBR
The number of basis rings
-1.0
WTPT3
Sum of path lengths starting from heteroatoms
7.9
logP
Calculate of logP
MDE 34
Molecular distance edge between all tert quat C
2SP2
Doubly bound carbon bound to two other carbons
-2.2
2SP3
Singly bound carbon bound to two other carbons
-1.4
SSS(-C)
Count of substructure
1.1
SSS(-O)
Count of substructure
-1.6
1.6
SSS(-O-)
Count of substructure
-1.3
1.1
SSS(-C(O)-)
Count of substructure
-1.8
SSS(12-1_=CH-)
Count of substructure
4.0 1.2
1.0
y
Output v1
Hidden
vA
t1
θ
tA
w11 w12
bias θ1
θA
Input
bias 1
2
-5.6
d
Fig. (10). Three-layer neural networks. Cited from Fig. (1) in Ref. [56].
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 395
where D is the number of input nodes and Wap is the connection weight between hidden node a and input node p. a is the bias connecting hidden node a to the input layer bias node. T is an adjustable parameter determining the shape of the sigmoidal curve and referred to as a computational temperature. The output node produces the calculated ADME property by a linear equation. A
A
a =1
a =1
y = Va t a + = Va f ( s a ) +
(10)
where Va is the connection weight between the output node and hidden node a. is the bias connecting the output node to the hidden layer bias node. Applications with ANN method are cited in literatures [58-66]. The back-propagation (BP) algorithm has been widely used for ANN optimization [56]. The BP algorithm is based on a gradient descent method to minimize an error function E with respect to the connection weights (Wap, Va) and biases (a, ) from the top output layer toward the bottom input layer. McElroy et al. have presented QSPR model to predict aqueous solubility of heteroatom-containing organic compounds [67]. Before descriptor generation and model building, compounds were placed into three subsets: a training set, a CV set, and an external prediction set. The training set compounds are used in nonlinear model training phase. CV set compounds are used in nonlinear modeling phase to prevent over-training of the ANN. The prediction set compounds are never used in model building and are used to demonstrate predictive ability of a model. Data set contained a 298-member training set, a 50-member cross-validation set, and a 51member prediction set. A total of 229 descriptors were calculated for each compound. Of those, 141 are topological, 30 are geometric, 10 are electronic, and 48 are hybrid descriptors. After pre-selection, 98 descriptors remained for subsequent ANN analysis. The 11-5-1 architecture was selected as the final model with GA and ANN routine. The best model gave a training set error was 0.576 (R2 = 0.90) log units. Cross-validation set error was 0.587 (Q2 = 0.88) log units. The prediction set error was 1.223 (Rpred2 = 0.53) log units. These results show that this model is effective at predicting aqueous solubility values for both oxygen- and nitrogen-containing compounds. A list of the descriptors can be seen in Table 4. Many of the descriptors in the model give some insight into the solute-solvent interaction when placing organic compounds in water. Branching information held in indices (KAPA-6), weighted path (WTPT-3), and distance-edge descriptors (MDE-44) allow for more topological detail. Molecular shape, which affects packing and solvent interactions, can be described through geometry dependent descriptors (GEOM-1, SYMM-25). The charged partial surface areas of the molecule, which play an important role in solvent-solute interactions, are described with DPSA-2 and SAAA-3 descriptors. The authors have stressed that aqueous solubility for a wide range of compounds could be predicted accurately based solely on molecular structure, with no corrective factor for physical state or the use of other data. Classical ANN can handle non-linear problems, but are prone to over-training, have problems with network optimization and model selection and are not efficient in dealing with high-dimensional data without pre-selection of descriptors. Bayesian neural network (BNN) represent a special type of neural net which overcome the problems of conventional ANN [57]. BNN is based on a probabilistic interpretation of network training. Network weights are found by Bayesian inference that gives an objective solution to the
396 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
network optimization. This approach also provides the model predictions as probability distributions and therefore permits evaluation of the confidence in prediction [68-70]. Lee et al. have developed highly predictive classification models for human liver microsomal (HLM) stability using the intrinsic clearance (Clint) as the end point [71]. They Table 4.
a
Descriptors of the Nonlinear Type III Model for the Combined Data Set. Cited from Table 7 in Ref. [67]
Descriptor
Type
Range
Explanationa
KAPA-6
topo
0.0-18.1
Kappa index – atom corrected
NO
topo
0-16
number of oxygens
NN
topo
0-6
number of nitrogens
NDB
topo
0-6
number of double bonds
WTPT-3
topo
2.35-43.2
Sum of path wts from heteroatoms
MDE-44
topo
0.0-83.3
distance edge between 4o and 4o carbons
SYMM-25
topo
0.05-1.0
geometrical symmetry
GEOM-1
geom
0.66-48.4
first geometric moment
DPSA-2
comb
64.3-4334
difference in partial surface areas
CHDH-1
comb
0.0-2.32
charge on donatable hydrogens
SAAA-3
comb
0.10-62.1
surface area of acceptor atoms 34
KAPA-6, Kappa index of three-bond counts, corrected for atom type: NO, the number of oxygens in the molecule; NN, the number of nitrogens in the molecule; NDB, number of double bonds in the molecule; WTPT-3, sum of all path weights starting from heteroatoms;36 MDE-44, distance-edge between all quaternary carbons;39 SYMM-25, geometric consideration of (number of unique atoms/total atoms); [(-SAi)];47 CHDH-1, sum of charges on donatable hydrogen; SAAA-3, (SAacc)/SAtot .
only used HLM data that were measured using an internally harmonized in vitro assay protocol. The total 14557 molecules were assigned as HLM stable and unstable compounds according to a cutoff of Clint = 20 ul/min/mg. The data set was partitioned into training (11646 compounds) and test (2911 compounds) sets randomly. Another data set was collected to validate the predictive power of the model (276 compounds). As chemical descriptors, the extended connectivity fingerprints of maximum diameter 6 (ECFP_6) were used. The ECFP_6 fingerprints and the HLM clearance values were used to build a classification model based on BNN algorithm. BNN model was able to correctly classify 78% of the 2911 compounds in the test set. For the validation set, BNN model was able to correctly classify 69% of the 276 compounds in the test set. The fact that predictions are worse for the validation set than the test set may suggest that the model should be updated continuously with new data. The ability of BNN classification model to discriminate between stable and unstable compounds was evaluated with a bimodal histogram of the test dataset, shown in Fig. (11). This histogram shows how BNN scores obtained for stable compounds are distributed along the positive range, while unstable compounds tend to have negative values. It can also be seen that a grey area
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 397
could be defined between -10 and 0, where there is some overlap between both classes of compounds. An advantage of the ECFP_6 fingerprints is that they are easy to translate into 2D sub-structural sketches. This allows for a structural analysis of the molecular 18% 16%
% of samples
14% 12% 10% 8% 6% 4% 2% 0% -100
-80
-20 -40 0 -60 HLM Bayesian score
20
40
Fig. (11). Bimodal histogram representing the distribution of Bayesian scores for stable (squares) and unstable (diamonds) compounds in the test set. Cited from Fig. (2) in Ref. [71].
features that contribute most to HLM stability. Fig. (12) shows the ECFP_6 fragments that the BNN classifier found most frequently in the stable molecular set. Similarly, Fig. (13) summarizes the most common fragments in the unstable set. By comparing Figs. (12) and (13), it can be seen for example that amide groups in stable compounds are surrounded by non-aromatic cyclic systems, while in unstable compounds those chemical groups are not sterically hindered. If comparisons would be extended to the hundreds of fingerprint features obtained for each class of compounds, the atomic environments most favorable for HLM stability could be easily identified. The authors have stressed that this classification model can be used in the design of new compounds with stable property. 2.4. SVM Support vector machines (SVM) are based on the structural risk minimization principle from statistical learning theory. SVM constructs a hyperplane, which separates the two classes of vectors with a maximum margin. Separating the classes with a large margin minimizes a bound on the expected generalization error. In many cases, SVM has been found to be consistently superior to other supervised learning methods and less prone to over-fitting [72, 73]. In linearly separable cases, SVM searches vector w and parameter b that minimizes ||w||2 and satisfies the following conditions:
wxi + b +1 , for y i = +1 Class 1 (positive)
(11)
398 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
A
A
NH
A
A
NH
NH
O N
N
NH
A
N
O
N
A A
A
A
O
A
A
NH
NH
N A
A
A
A
A
A
A
A
A
O
A
A
N
A
N A
A
A
O
A
A
A
N N
O
A
A
A
NH
A
N A
A
O
A
A
A
NH
O
A
A
A
A
A A
A
O OH N
A
N A
A A
N
A
A
OH
O
N
A
OH
O
A A
O
A
OH
OH
A
A
O
A
A
A
A
A A
A
A A
N N
A
N
N
A
A
A
A
A
A
A
A
A A
A
A
A N
O
N A
A
O
A
A
N
N
A
O A A
A
N
A N
A
O A
A
A
A N
N
A
A
A
A
A A
O
N
A
N
A
OH
A
O
A
A
O
A
A OH
A
A A
N
O
A
A
A
O
A
O A
A
Fig. (12). ECFP_6 sub-structural features most frequently observed in stable compounds. Cited from Fig. (3) in Ref. [71].
wxi + b 1 , for y i = 1 Class 2 (negative)
(12)
where yi is the class index, w is a vector normal to the hyperplane, |bj|/||w|| is the perpendicular distance from the hyperplane to the origin, and ||w||2 is the Euclidean norm of w. These relationships are schematically shown in Fig. (14). After the determination of w and b, a given vector xi can be classified by decision function ‘sign[(wix) + b]’. In nonlinear cases, SVM maps the input variable into a high dimensional feature space using a kernel function such as K(xi,xj) = exp(-||xj-xi||2/2ó2). Linear SVM is then applied to this feature space, and then, the decision function is given by
L f ( x) = sign i0 y i K ( x, xi ) + b i =1
(13)
where L is the number of support vectors. The coefficients i0 and b are determined by maximizing the following Langrangian expression:
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 399
A
A A A
N
N
N
N
A
O
A
A
A
A
A
A
A
A A
A A A A A
A
A A A
A N
NH
A
A
N
A N
NH
O
A A
A
N
N
A
A
O
A N
A
O
A
A A
A
N
A
A
N
A
A
A
A
A
A
A
A A
A O
A
O
A
A
F
F
F
A
F
O A
A
A
A A
AA N
A
F
F
O
A
AA N
A
A
A
NH O
A
N
A
N
NH
A
A N
N
A
A
A A
O
O
A
A
A
A
A
A O
O A
A
A A
O
A
A
A
A
A
A
A
A
Fig. (13). ECFP_6 sub-structural features most frequently observed in unstable compounds. Cited from Fig. (4) in Ref. [71]. L
i i =1
1 L L i j yi y j K ( xi , x j ) 2 i =1 j =1
(14) L
under the following conditions: 0 < i < C and
i =1
i
y i = 0 . A positive or negative
value from Eq. 13 indicates that the vector x belongs to the positive or negative classes, respectively. Applications with SVM method are cited in literatures [74-76]. Yap et al. have developed filters for predicting substrates/non-substrates of three CYP isoenzymes, CYP3A4, CYP2D6 and CYP2C9 [77]. Classification for inhibitors/non-inhibitors was not described here for avoiding redundancy. GA based descriptor selection method was used to select relevant chemical descriptors for SVM classification of the substrates. The consensus SVM classification system by using multiple descriptor
400 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
Fig. (14). Schematic illustration of SVM algorithm.
sets generated from GA was employed. The dataset were composed of 368 substrates and 334 non-substrates for CYP3A4, 198 substrates and 504 non-substrates for CYP2D6, and 144 substrates and 558 non-substrates for CYP2C9, respectively. Substrates of an isoenzyme were denoted as belonging to positive class (P+) of the isoenzyme, and non-substrates were denoted as belonging to the negative class (P-) of the isoenzyme. Representative training and validation sets were constructed from the datasets according to their distribution in the chemical space. Any pair of compounds of similar chemical features was evenly assigned into separate datasets. As chemical descriptors, a total 1497 1D, 2D, and 3D molecular descriptors were derived from the 3D structure of each compound. The SVM classification system was composed from the 81 SVM classification models. The accuracies for classification of substrates and non-substrates of CYP3A4, CYP2D6, and CYP2C9 were 98.2 and 90.9%, 96.6 and 94.4%, and 85.7 and 98.8%, respectively. Because composite descriptors encode multiple physicochemical and structural aspects of the molecule, it is difficult to extract from these descriptors information about which specific molecular characteristics are important for the substrates of these CYP isoenzymes. The authors have tried to roughly distinguish between substrates/non-substrates from the values of six selected descriptors, S, nHAcc, nHDon, MLOGP, MW, and SPH. S is the combined dipolarity/polarizability, nHAcc and nHDon are the number of acceptor and donor atoms for hydrogen bonds, respectively, MLOGP is the Moriguchi’s LogP, MW is the molecular weight, and SPH is the spherosity. The average values of these six descriptors for P+ and P- compounds of all of the various datasets are given in Table 5. Substrates of CYP3A4 are generally larger in size, less spherical in shape, more hydrophobic, and have more hydrogen bonding sites than nonsubstrates. Substrates of CYP2D6 are generally smaller in size, more hydrophobic than non-substrates, and contain one hydrogen bond donor. Substrates of CYP2C9 generally are more hydrophobic than non-substrates but are smaller in size and have lesser hydro-
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 401
gen bonding capacity. The consensus SVM model generally gives better accuracies than single SVM classification models. This model is potentially useful for developing filters for the prediction of substrates of CYP isoenzymes. Table 5.
Differences in the Values of Descriptors Importance for Distinguishing between P+ and P- Compounds. Cited from Table 8 in Ref. [77] Average Valuea
Dataset
CYP
Descriptor P+
P-
S
2.56 (1.15)
2.29 (1.17)
nHAcc
5.53 (3.45)
4.91 (3.14)
nHDon
1.72 (1.99)
1.44 (1.75)
MLogP
2.20 (1.99)
1.60 (2.06)
MW
379 (157)
315 (137)
SPH
0.76 (0.13)
0.78 (0.13)
S
2.19 (1.08)
2.53 (1.18)
nHAcc
4.10 (2.13)
5.68 (3.58)
nHDon
1.15 (1.22)
1.76 (2.07)
MLogP
2.51 (1.74)
1.68 (2.11)
MW
320 (100)
360 (166)
SPH
0.78 (0.14)
0.77 (0.13)
S
2.52 (1.26)
2.41 (1.14)
nHAcc
4.69 (2.52)
5.38 (3.48)
nHDon
1.03 (1.14)
1.73 (2.01)
MLogP
2.05 (2.04)
1.88 (2.05)
MW
326 (112)
354 (160)
SPH
0.75 (0.14)
0.78 (0.13)
3A4
Substrates/ Nonsubstrates
2D6
2C9
a
Values in parentheses are the standard deviations.
Xue et al. have modeled the binding affinities to HSA with the chemical descriptors [78]. This data set is the same as used in Colmenarejo’s study [37]. The training and test sets are also same. About 600 descriptors were calculated for each compound. After the heuristic reduction, the pool of the descriptors was reduced to 243. A variety of subsets sizes were investigated to determine the optimum number of the descriptors in model. The influences of the number of the descriptors on R2, Q2, and s2 are shown in Fig. (15).
402 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
From Fig. (15), it can be seen that seven descriptors appear to be sufficient for a successful regression model. The list of the selected seven descriptors is summarized in Table 6. The obtained model had R2 = 0.86 with Q2 of 0.63. With the test set, the prediction result 0.9 2
R 0.8
R s
R
2
0.7
2 cv
2
0.6
R
2 cv
0.5 0.4
s
2
0.3 0.2 0.1
0
1
2
3
4
5
6
7
8
9
10
number of descriptors Fig. (15). Influence of the number of descriptors on R2, R2CV, and s2 of the regression models. Cited from Fig. (1) in Ref. [78]. Table 6.
a
Seven-Descriptor Linear Model for the Binding Affinitya. Cited from Table 2 in Ref. [78]
Descriptor
Chemical Meaning
Coefficient
t-Test
(constant)
intercept
-2.513 ± 0.388
-6.472
HDCA-2
HA dependent HDCA-2 [Zefirov’s PC]
-0.401 ± 0.078
-5.136
MSA
molecular surface area
0.007 ± 0.001
12.801
NO
number of O atoms
-0.149 ± 0.017
-8.877
RNR
relative number of rings
9.210 ± 1.395
6.605
RNN
relative number of N atoms
-3.945 ± 0.663
-5.950
BI
Balaban index
0.403 ± 0.097
4.147
RNCS
relative negative charged SA (SAMNEG*RNCG) [quantum-chemical PC]
-0.045 ± 0.013
-3.392
R2 = 0.86; s2 = 0.050; rms = 0.212; n = 84; F = 63.89; R2cv = 0.63.
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 403
was obtained, Rpred2=0.71. By interpreting the descriptors in the regression model, it is possible to gain some insight into factors that are likely to govern the binding affinities to HSA. According to the analysis of the corresponding regression coefficient (Table 6), molecular surface area (MSA), relative number of rings (RNR), and the Balaban index (BI) present positive contributions for binding affinity, whereas HA dependent HDCA-2, number of O atoms (NO), relative number of N atoms (RNN), and relative negative charges surface area (RNCS) present negative contributions. From MSA, binding is favored for the molecules with large molecular surface area. From RNR, molecules with a larger number of rings are expected to bind more tightly to HSA. From BI, the large degree of branching for molecules is in favor of the binding. From HDCA-2, NO, RNN, and RNCS, the hydrogen bonding might not be favorable in protein binding. Heuristic method (HM) was not sufficiently accurate and the prediction ability was not satisfactory, showing the factors influencing the binding affinities of these compounds are complex and not all of them are linear correlations with the binding affinity. So, after the establishment of the linear model by HM, the authors have built the nonlinear model based on the same seven descriptors by SVM. The final SVM model through optimization of the parameters, gave R2 of 0.94 for the training set, Rpred2 of 0.89 for the prediction set, respectively. The authors have reported that the successive application of HM and SVM is good modeling strategy for ADME properties. 2.5. Other Regression Techniques k-Nearest Neighbors (kNN) kNN is a non-parametric approach and does not require any concrete function in advance. In this approach, a single data point is left out of the training set and its ADME property is predicted as a weighted average of the ADME values of its nearest neighbors. k equal to 1 or can be a larger number when looking for a set of k nearest neighbors. The metric used to search the nearest neighbors is usually the Euclidian distance [79-82]. Gaussian Process (GP) GP is another new promising method based on a Bayesian approach. GP is equivalent to ANN with a single hidden layer containing an infinite number of nodes. There are four main advantages in GP. (a) GP does not require subjective a priori determination of parameters such as variable importance or network architectures. (b) Since the algorithm minimize the log marginal likelihood, which directly prevents the model from overtraining, GP does not need internal CV. (c) GP does work well for a big pool of descriptors. (d) GP can easily identify domain of applicability. This means that GP can estimate the reliability of a given prediction. The disadvantage of GP is the black box. Then, the model is difficult to be interpreted [83, 84]. Random Forest (RF) RF is an ensemble of decision trees (DT) using bootstrap samples of the training data set and random variable selection in the tree induction. Each tree provides a classification. RF chooses the classification having the most votes. Compared to single DT, RF generally performs better in terms of prediction accuracy. Additional nice features of RF are its out-of-bag performance estimate that can be used in place of CV, and a measure of descriptor importance. [85-87].
404 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
Classification and Regression Trees (CART) CART is a non-parametric statistical method, which uses decision trees (DT) to solve classification and regression problems. A CART analysis generally consists of three steps. In a first step an over-grown tree is build. This tree is called the maximum tree and is grown using a binary split-procedure. In a next step the over-grown tree, which shows over-fitting, is pruned. During this procedure, a series of less complex trees is derived from the maximum tree. In the final step, the tree with the optimal tree size is selected using CV. In CART, the variable selection is part of the methodology. This means that data modeling can be started with an extended set of descriptors [88-91]. 3. RULE MINING 3.1. RST Recently, for the discovery of the particular patterns or rules from data set, what we call rule mining method has been developed and has been widely used. Rough set theory (RST) has been introduced in chemometrics [92]. The essential part of RST is construction of minimal subset of descriptors, “reducts”, which can distinguish samples belonging to different class, and extraction of rules using these reducts. For easily understanding, we will show small data set as an example. In Table 7, a1, a2 and a3 are variables and d is ADME property, each of which is discretized into three categories (1, 2 and 3). The procedure to extract the reducts and rules from Table 7 are explained step by step. First, we construct the discernibility matrix from Table 7. Discernibility matrix is the matrix that has the same numbers of columns and rows as the number of samples. In discernibility matrix, (i,j) element is filled with the variables which are needed to discern the sample Xi and Xj (Table 8). For example, X1 and X2 are discernible each other by a1 or a2 or a3, so (i, j) element is filled with ‘a1 a2 a3’. Samples labeled same class attributes Table 7.
Decision Table Using All Variables. Cited from Ref. [93] U
a1
a2
a3
d
X1
2
1
3
1
X2
3
2
1
2
X3
2
1
3
1
X4
2
2
3
2
X5
1
1
4
3
X6
1
1
2
3
X7
3
2
1
2
X8
1
1
4
3
X9
2
1
3
1
X10
3
2
1
2
Data Modeling and Chemical Interpretation
Table 8.
Frontiers in Drug Design & Discovery, 2009, Vol. 4 405
Discernibility Matrix. Cited from Ref. [93]
1
2
3
4
5
6
7
8
9
1
-
2
a1a2a3
-
3
-
a1a2a3
-
4
a2
-
a2
-
5
a1a3
a1a2a3
a1a3
a1a2a3
-
6
a1a3
a1a2a3
a1a3
a1a2a3
-
-
7
a1a2a3
-
a1a2a3
-
a1a2a3
a1a2a3
-
8
a1a3
a1a2a3
a1a3
a1a2a3
-
-
a1a2a3
-
9
-
a1a2a3
a2
a2
a1a3
a1a3
a1a2a3
a1a3
-
10
a1a2a3
-
-
-
a1a2a3
a1a2a3
-
a1a2a3
a1a2a3
10
-
each other cannot be distinguished, so the value in the element is missing. In addition, (n, m) and (m, n) should have the same value, so the upper triangle matrix is not necessary. Second, we can obtain reducts from this matrix. Reduct means that the minimal subset of variables to discern all samples each other which are not classified to the same class. In order to obtain reducts, we have to define a special function f(D), the kind of Boolean equations. The function f(D) is calculated from the discernibility matrix by multiplying all elements. (Eq. 15) After Boolean’s manipulations, we can obtain two reducts, {a1, a2} and {a2, a3}. (Eq. 16) This means that we can discern all samples by using either {a1, a2} or {a2, a3}. f(D) = (a1+a2+a3)a2(a1+a3)(a1+a3)… (a1+a3)(a1+a2+a3)(a1+a2+a3) =a1a2+a2a3
(15) (16)
Finally, we can extract rules by using either of these reducts. The procedure is the same as above. Preparing decision table, calculating discernibility matrix, and extraction rules by discernibility function. In this case, we focus on one reduct {a1, a2}. Table 9 represents the decision table based on reduct {a1, a2}. Table 10 represents the discernibility matrix derived from reduct {a1, a2}. Table 10 is just the rule: If a1 = 2 and a2 = 1 then d = 1 If a2 = 2 then d = 2 If a1 = 1 then d = 3
(17)
Koyama et al. have applied RST to the oxidation sites data in CYP2C9 and extracted useful rules which atoms are oxidized [93]. The data set is the same as used in Sheridan’s study [85]. As for 50 drugs, the primary and secondary oxidation sites were experimentally identified. The total number of heavy atoms was 1031 and among them, 96
406 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Table 9.
Hasegawa and Funatsu
Decision Table Using the Reduct {a1, a2}. Cited from Ref. [93]
U
a1
a2
d
X1
2
1
1
X2
3
2
2
X3
2
1
1
X4
2
2
2
X5
1
1
3
X6
1
1
3
X7
3
2
2
X8
1
1
3
X9
2
1
1
X10
3
2
2
Table 10. Discernibility Matrix Using the Reductu {a1, a2}. Cited from Ref. [93] U
a1
a2
d
X1
2
1
1
X2
*
2
2
X3
2
1
1
X4
*
2
2
X5
1
*
3
X6
1
*
3
X7
*
2
2
X8
1
*
3
X9
2
1
1
X10
*
2
2
atoms were assigned as oxidation atoms. For each heavy atom, descriptor variables were calculated taking account of the nearest environments. Four types of descriptors were used. 1) SS descriptors: substructure descriptors to describe local chemical environments; 2) PE descriptors: physicochemical descriptors through bond distance to atom property; 3) SPAN descriptors: whether an atom is at the end or the middle of a molecu-
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 407
le in a topological sense; 4) HYDROGENAREA descriptor: surface areas attached to hydrogens. Fig. (16) shows the substructure descriptors in a small example molecule. For example, a symbol ‘CX2sp3-NX3sp3’ means that central sp3 carbon is connected to sp3 nitrogen with three neighbors. Totally, 1981 descriptors were used for describing each heavy atom. RST was applied to the data matrix comprising of 1031 samples and 1981 descriptors. RST could successfully produce many useful rules. Two major rules for CYP2C9 oxidation are described here: Rule 1: If CX1sp3 = 1 & 6_11 = 0 & 6_3 = 2 & SPAN_ DISTFURTHESTNEIGHBOR = 3 then oxidation_site
(18)
An atom is metabolized if the molecule is somewhat long and does not have an atom labeled ‘6’ at the 11 bond distance from the sp3-carbon and has two atoms labeled ‘6’ at the 3 bond distance from the sp3-carbon. The representative examples obeying Rule 1 are shown in Fig. (17). This rule corresponds to demethylation of methoxy group frequently observed in CYP2C9 oxidation. Rule 2: If CX2sp3 = 0 & 3_4 = 1 & 6_6 = 2 & HYFROGENAREA_ SUMAREA = 2 then oxidation_site
(19)
An atom are metabolized if the molecule has somewhat small solvent accessible surface area and has an atom labeled ‘3’ at the 4 bond distance from the focused atom and has two atoms labeled ‘6’ at the 6 bond distance from the focused atom. The representative examples obeying Rule 2 are shown in Fig. (18). This rule corresponds to epoxidation on aromatic ring frequently observed in CYP2C9 oxidation. The rule mining method such as RST is useful for chemical interpretation. Because only RST cannot achieve high statistical performance, the combination use of Random forest or ADAboost is recommended. 3.2. Rules of Thumb Gleeson has generated a set of simple and interpretable rules for 15 different ADMET assays [94]. (solubility, permeability, oral BA, volume of distribution, plasma protein binding, central nervous system penetration, brain tissue binding, P-gp efflux, in vivo clearance, hERG inhibition, and CYP 1A2/2C9/2C19/2D6/3A4 inhibitions) More than 30000 diverse molecules were collected from in-house database and PCA was performed with 12 commonly used physicochemical descriptors. Result of PCA indicated that three essentially orthogonal molecular descriptors (molecular weight, ionization state, ClogP) are enough to describe chemical space. The likelihood of a molecule having a particular ADMET parameter above average, average, or less than average is reported in Table 11. Only result of neutral compound is shown in Table 11. If molecular weight of compound increases, the solubility would decrease, the permeability would decrease, and an increase in protein binding and increase CYP inhibition on average would be observed. (Table 11) These rules are consistent with chemists’ intuition and can be used to supplement the more complex, predictive in silico models. Martin has also proposed similar approach called BA score and applied it to the real projects in Abbott Laboratories [95]. 4. VISUALIZATION Numerous mapping or projection methods have been investigated in chemometrics [96]. Only the commonly used methods are described here.
408 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
6 6 6 3
N
7
6 N
6
6
1
6
6
6
6
SPAN descriptor Longest through-bond distancc in molecule MAXDISTMOL=8 Longest through-bond distancc from indicatcd atom DISTFURTHESTNEIGHBOR=5 RATIO= 5/8= 0.625 SS descriptor CX2sp3 CX2sp3-CX3sp2a5 CX2sp3-NX3sp3A6 CX2sp3-CX3sp2a5-CX2sp2a5 CX2sp3-CX3sp2a5-NX2sp3a5 CX2sp3-NX3sp3A6-CX2sp3A6 CX2sp3-CX3sp2a5-CX2sp2a5-CX2sp2a5 CX2sp3-CX3sp2a5-NX2sp3a5-CX2sp2a5 CX2sp3-NX3sp3A6-CX2sp3A6-CX2sp3A6
Frequency 1 1 1 1 1 2 1 1 2
SS-A descriptor CX2sp3 CX2sp3-CX3sp2 CX2sp3-NX3sp3 CX2sp3-CX3sp2-CX2sp2 CX2sp3-CX3sp2-NX2sp3 CX2sp3-NX3sp3-CX2sp3 CX2sp3-CX3sp2-CX2sp2-CX2sp2 CX2sp3-CX3sp2-NX2sp3-CX2sp2 CX2sp3-NX3sp3-CX2sp3-CX2sp3
1 1 1 1 1 2 1 1 2
SS-B descriptor n n-A6 n-A6-A6 n-A6-A6-A6 n-a5 n-a5-a5 n-a5-a5-a5
1 1 2 2 1 2 2
PE descriptor 6_0 1_1 7_1 3_2 6_2 6_3 6_4 6_5
1 1 1 1 3 4 1 1
Fig. (16). The topological substructure descriptors (SS, SS-A, SS-B) and physicochemical environment (PE) descriptors for an atom (indicated by arrow) in an example molecule. The number near each atom is the physicochemical type (1 = cation, 3 = H-bond donor, 6 = hydrophobe, 7 = other). Also indicated is the ratio for the SPAN descriptor that determines whether an atom is at the end or middle of a molecule based on its topology. Cited from Fig. (1) in Ref. [85].
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 409
O Cl
Cl
O Cl
O
Cl NH
N
O
O
O
N
OH
methoxychlor harmine
indomethacin
Fig. (17). Examples derived from rule 1. Red circle means the experimentally determined oxidation site. Cited from Ref. [93]. O
HO O
O
O
H N O
NH NH HN
O
OH
Cl
Cl
O
etodolac
phenytoin aceclofenac
Fig. (18). Examples derived from rule 2. Red circle means the experimentally determined oxidation site. Cited from Ref. [93].
4.1. KNN Kohonen neural networks (KNN) are based on the idea that human brains tend to compress and organize sensory data, spontaneously. KNN can be used to generate a projection of objects from a higher-dimensional space onto a two dimensional space. In other words, this method enables a decrease in dimension while conserving the topology of the information as much as possible [97-99]. KNN is typically made up from two layers of neurons (input and output layers). The input layer contains m neurons corresponding to m variables describing objects. The output layer is a two-dimensional geometri-
410 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
cal arrangement of n neurons and the topology is usually defined as a ‘torus’. ‘Torus’ means that the right or top edge of map is continued to its left on low edge, respectively, and vice versa. The m neurons of the input layer are all connected to each of the n neurons of the output layer as shown in Fig. (19). The network is trained by adjustment of the connection weight in two phases, the competitive learning and self-organization phases. After two phases, the objects that are close to each other in the multi-dimensional descriptor space remain neighbors on the KNN map. Applications with KNN method are cited in literatures [100, 101]. Table 11. Indication of How Changes in Key Molecular Properties will Affect a Range of ADMET Parametersa. Cited from Table 3 in Ref. [94] Neutral Molecules
MWT < 400 and clogP < 4
MWT > 400 and/or clogP > 4
solubility
average
lower
permeability*
higher
average/higher
bioavailability
average
lower
volume of Dist.**
average
average
plasma protein binding
average
higher
CNS penetration***
higher/average
average/ lower
brain tissue binding
lower
higher
P-gp efflux
average
higher/average
in-vivo clearnce
average
average
hERG inhibition
lower
lower
P450 inhibition****
lower 2C9, 2C19, 2D6 & 3A4 inhibition
higher 2C9, 2C19 & 3A4 inhibition
P450 inhibition****
higher 1A2 inhibition
lower 1A2 inhibition
P450 inhibition****
average 2D6 inhibition
a
Expressed relative to the mean value of the data sets. MWT and clogP cut-offs of 400 and 4, respectively, are used. *Optimum clogP bin is 3-5 with respect to permeability. **Average to high volumes rather than high, low, or average generally considered optimum. *** Low CNS considered optimum, although for targets in the brain, this will be reversed. **** Some isoforms show a nonlinear relationship with clogP and/or MWT. These are guides only. For greater detail, look at the individual ADMET ANOVA graphs found in the text or the tables reported in the Supporting Information.
Balakin et al. have developed KNN model for P-glycoprotein (P-gp) substrates and non-substrates [99]. As chemical descriptors, a wide range of molecular descriptors of different types was calculated. These descriptors included electronic, topological, spatial, structural, and thermodynamic descriptors. A total of more than 1000 descriptors were calculated for each compound. To reduce the number of descriptors that could contain redundant information, PCA was performed. Five descriptors maximally contributing to
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 411
the first significant PC, were selected and were used as input parameters in KNN model. (BELp3: eigenvalues of burden matrix/weighted by atomic polarization, H3u: H autocorrelation/unweighted, CIC0: complementary information content, RDF025v: radial distribution function/weighted by atomic van der waals volumes, HTu: H total index/unweighted) The KNN map was generated for the entire P-gp training dataset including 96
m input Iayer
xi
x
w ij * weight vector w j*
r
output layer j*
n
n Fig. (19). Architecture of KNN. Cited from Fig. (2) in Ref. [97].
P-gp substrates and 79 non-substrates. Fig. (20) separately shows the sites of distribution of these compound categories. Obviously, their positions on the KNN map are quite different. The results of a leave-10%-out experiment indicated that the developed model is general enough and can be used for prediction purposes. On average, 77.4% of P-gp substrates (+) and 80.6% of P-gp non-substrates (-) compounds were correctly classified with this model. The authors have reported other ADME models using KNN method and their classification performances are relatively good. Korolev et al. have developed KNN models for CYP substrates and reaction products [102]. The substrates consisted of 485 compounds and the products consisted of 523 compounds. Sixty molecular descriptors describing the important molecular properties, such as lipophilicity, charge distribution, topological features, steric and surface parameters were explored. The number of descriptors was reduced to 26 by the omission of the low variable and highly correlated descriptors. To further reduce the descriptor space, PCA was performed. Eventually, seven descriptors were selected as the most relevant. (logD7.4: logP at pH 7.4, HOMO: highest occupied molecular orbital, Jurs-PPSA-1: partial negative surface area, Jurs-TPSA: total solvent-accessible surface area, MW: mole-
412 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
cular weight, HBA: number of H-bond acceptors, HBD: number of H-bond donors) The KNN map for CYP substrates was generated using the selected seven descriptors. (Fig. (21a)) The CYP substrates are distributed throughout the map as the irregularly shaped islands, with a clearly defined trend toward the right side of the map. The area occupied by the substrates is relatively large, which reflects the broad substrate specificity against CYPs. For the comparison, the CYP products were mapped into the same KNN map. (Fig. (21b)) This data set occupies distinct areas on the map substantially different from the regions of the substrates localization. On the basis of these distributions, the authors 6
6 0 1 2 3 4 5 6 7 8
5
5
4
4
3
3
2
2
1
0 1 2 3 4 5 6 7 8
1 2
1
3
5
4
6
2
1
3
(a)
5
4
6
(b)
Fig. (20). Kohonen map generated for the entire P-gp training set (167 compounds). The areas of substrates (a) and non-substrates (b) are shown separately. Cited from Fig. (8) in Ref. [99]. 10
10 0 0.5 1 1.5 2
8
8
6
6
4
4
2
2
2
6
4
(a)
8
10
0 0.5 1 1.5 2 2.5 3
2
4
6
8
10
(b)
Fig. (21). (a) 10*10 Kohonen network trained with seven selected descriptors for cytochrome substrates (485 compounds). (b) Final cytochrome reaction products (523 compounds) processed within the same map. The data have been smoothed. Cited from Fig. (1) in Ref. [102].
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 413
have built the smoothed contour plots of the occurrences of these two compound categories within the KNN map. (Fig. (22)) The area of substrates is marked in green, the area of products is in blue, and the low-populated area is in brown. The model correctly classified 76.7% of substrates and 62.7% of products, as defined by their localization in the corresponding areas of the KNN map. Although the general classification power of the model is moderate, it reasonably discriminates between CYP substrates and nonsubstrates. The enhancement factors for the areas of substrates and non-substrates were equal to 7.17 and 4.21, respectively. The enhancement factor is a ratio between the fractions of correctly and incorrectly classified compounds within the corresponding areas on the map. This number exceeds the random expectation. The authors have concluded that the developed model is useful in assessment of compound’s ability to be a CYP substrate.
10
Substrates Products
8
6
4
2
2
4
6
8
10
Fig. (22). Smoothed contour plots of the occurrences of substrates and final products within the Kohonen map. The area of substrates is depicted in green, the area of products is in blue, and the low-populated area is in brown. The contours correspond to at least 1.5% of compounds, from a particular category, per node. Cited from Fig. (2) in Ref. [102].
4.2. DT Decision trees (DT), also called recursive partitioning, create a branching structure. The branch taken at each node is determined by a rule that is derived from the molecular descriptors [103]. Thus, the data set is iteratively split into smaller and more homogeneous subsets. If a leaf in the tree contains predominantly compounds of one category, the path to this particular node provides rules for molecular properties that are associated with this specific category. The Student’s t-test is used as the split criterion. The Student’s t-test is computed according to the following formula:
414 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
t = ( X - Y ) / ( (1 / M + 1 / N ) * ( SSX + SSY ) /( M + N 2))
(20)
M
SSX = ( X i X ) 2
(21)
i =1 N
SSY = (Yi Y ) 2
(22)
i =1
where X1, X2, …., XM are the ADME property values of the compounds in the first subset, and Y1, Y2, …., YN are the ADME property values of the compounds in the second subset. M and N are the numbers of the compounds in these two subsets, respectively.
X and Y are the averaged ADME property value in each subset, respectively. SSX and SSY are the variance of X and y, respectively. The molecular descriptor that gives the largest t value is chosen as the descriptor for the splitting. Applications with DT method are cited in literatures [104-108]. Burton et al. have developed two DT models for the prediction of CYP2D6 and CYP1A2 inhibitions [109]. Dataset was extracted from in-house structured databases that contain precise measures and detailed experimental protocols. For CYP2D6, the number of inhibitors and non-inhibitors were 78 and 85 as the class thresholds of Ki = 10 uM, respectively. For CYP1A2, the number of inhibitors and non-inhibitors were 41 and 40 as the class thresholds of Ki = 30 uM, respectively. Two types of descriptors were used for building the DT models. First, 114 two-dimensional descriptors were calculated for all the compounds. This parameter set contains information about the molecules such as molecular weight, number of a given atom, number of H-bond acceptors/donors, lipophilicity, etc. A second type of descriptors was created with 32 P_VSA parameters. They are based on the approximation at atomic level of the molecular van der Waals surface area, VSAi, along with several other molecular properties, Pi. Pi considered are the molar refractivity, logP(o/w), and the electrostatic properties or pharmacophore characteristics. The DT model for CYP2D6 inhibitors with 2D and P_VSA descriptors is shown in Fig. (23a). The + and – signs mean that the class assigned to a leaf is inhibitor or non-inhibitor, respectively. The model accuracy was 90% for CYP2D6 inhibition. The DT model for CYP1A2 inhibitors with P_VSA descriptors is shown in Fig. (23b). The model accuracy was 89% for CYP1A2 inhibition. To confirm the performance of two models, two external test sets of 34 and 58 molecules related to the CYP2D6 and CYP1A2 Ki experiments were collected. The validation result for CYP2D6 inhibition was 89% accuracy. The validation result for CYP1A2 inhibition was 81% accuracy. The average values of the selected descriptors for inhibitors and non-inhibitors of CYP2D6 dataset are gathered in Table 12. These descriptors indicate the hydrophobicity, shape, and electrostatic contributions. The average values of the selected descriptors for inhibitors and non-inhibitors of CYP1A2 dataset are gathered in Table 12. It is found that the influence of hydrophilicity expressed by logP is large for CYP1A2 inhibition. Yamashita et al. have proposed multi-objective DT model [110]. The metabolic stability values for 161 drugs involving 6 CYP isoforms (1A2, 2C9, 2C19, 2D6, 2E1, and 3A4) was used. The averaged information gains for multiple objective variables were
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 415
CYP2D6
Training set: 163 compounds
CYP1A2
Training set: 81 compounds SMR_VSA6
a_hyd bpol VDisEq
PEOE_RPC+
chi l SMR_VSA5
SlogP_VSA7 PEOE_ VSA-3
PEOE_PC+ PEOE_VSA _EPNEG
(4/0)
(0/0) (0/0) (0/0)
(4/0) (4/0)
SlogP_ VSA5
SlogP_VSA9 PEOE_VSA+4
SlogP_VSA9
(1/1)
(16/3)
(10/1) (0/1) (0/0)
(3/4) (8/1)
(6/0) (4/2)
(4/1) (19/1)
Fig. (23). Two of the best models from the CYP2D6 (Ki dataset, 10 uM threshold, 2D and P_VSA descriptors) and CYP1A2 (Ki dataset, 30 uM threshold, P_VSA descriptors) datasets. Both models have been validated with an external dataset of 34 and 58 compounds for CYP2D6 and VYP1A2, respectively. + and – signs mean that the class assigned to a leaf is inhibitor or non-inhibitor, respectively. The distribution of molecules of the test set in each leaf is positioned between brackets (number of correctly classified compounds/number of misclassified compounds). Cited from Fig. (4) in Ref. [109]. Table 12. Differences in the Values of the Selected Descriptors for CYP2D6 and CYP1A2 Inhibitors Classification. Cited from Table 6 in Ref. [109] CYP2D6
CYP1A2
Average Value Descriptor
Average Value Descriptor
Inhibitor
Noninhibitor
Inhibitor
Noninhibitor
A_hydr
18.0
14.2
SMR_VSA6
15.1
56.2
bpol
30.4
25.9
SlogP_VSA7
124.6
78.4
Chil
11.8
9.7
SlogP_VSA9
63.3
79.4
VdistEq
3.3
3.0
PEOE_VSA4
2.3
3.8
PEOP_RPC+
0.2
0.3
SMR_VSA5
166.0
127.9
used as a quality-of-split criterion instead of the standard Student’s t test. This is like to find the best balanced-compounds taking account of various ADME properties including biological activities. After the tree was fully grown, pruning of the tree was performed with reference to the misclassification rate determined by the leave-some-out procedure. The number of terminal groups giving the minimum misclassification rate was regarded as optimal. Molecular descriptors of each compound were calculated by ADMET Predictor, which include constitutional descriptors, topological and electrotopological descriptors, and descriptors relating to hydrophobicity, electronic properties, hydrogen bon-
416 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
ding, and molecular ionization. When leave-some-out prediction was performed, the misclassification rate was minimal at a terminal group number of 9. Fig. (24) shows a classification tree model. The model consists of 8 splitting rules and 9 terminal groups (misclassification rate = 9.63%). Fig. (24) also presents the distribution of CYP metabolism levels in each terminal group. To obtain intuitive understanding of the trends of the tree structure, a visual image of the hierarchically structured data was presented. (Fig. (25)) The 6-colored square icons indicate the compounds studied, where the color and its brightness represent metabolic susceptibility toward each CYP isoform. The trends found were as follows: (a) CYP2C9 and CYP2E1 substrates mostly belong to Groups 1 and 2, respectively; (b) CYP2D6 substrates belong to Groups 5-8; and (c) CYP3A4 substrates are detected in almost all groups, while CYP3A4 substrates belonging to Groups 3 and 4 are highly susceptible to the enzyme. The QAvgNeg descriptor is the population average across all ionized species of the net formal negative charge calculated at pH 7.4. Considering that most of the CYP2C9 substrates belong to Group 1, it seems that CYP2C9 preferentially metabolizes anionizable compounds. On the other hand, many CYP2D6 substrates belong to Groups 5-8, in which the QAvgPos, i.e., the population average across all ionized species of the net formal positive charge calculated at pH 7.4, is greater than 0.7912. In contrast to CYP2C9 substrates, many CYP2D6 substrates appear to be cationic compounds. CYP2E1 substrates belong to Group 2, for which the QA vgNeg >0.133
<0.133
QA vgPos 1A2 2C9 2C19 2D6 2E1 3A4
>0.7912
<0.7912
VMcGowan
VMcGowan
strng modr none 0 22 2 20 2 2 22 0 2 24 0 0 24 0 0 7 15 2
Group 1 >182.6
<182.6
1A2 2C9 2C19 2D6 2E1 3A4
strng modr none 2 2 5 0 0 9 9 0 0 0 0 9 7 0 2 0 1 8
Group 2
1A2 2C9 2C19 2D6 2E1 3A4
<487.1
strng modr none 2 76 2 75 3 2 71 3 6 73 5 2 78 2 0 69 8 8
>487.1
HBAo >1.5
<1.5
MaxQ
HBDn
1A2 2C9 2C19 2D6 2E1 3A4
Group 3
Group 4 >0.12209
<0.12209
1A2 2C9 2C19 2D6 2E1 3A4
strng modr none 9 0 1 8 1 1 10 0 0 2 6 2 0 10 0 10 0 0
strng modr none 0 3 9 12 0 0 0 3 9 12 0 0 0 12 0 2 6 3
Group 5
<0.5
SsssCH <0.076
>0.076
1A2 2C9 2C19 2D6 2E1 3A4
strng modr none 0 0 7 0 0 7 0 7 0 7 0 0 7 0 0 0 0 7
Group 6 0 strng modr none 0 2 2 1A2 2C9 4 1 0 4 0 1 2C19 0 1 2D6 4 5 2E1 0 0 5 3A4 0 0
Group 8
1A2 2C9 2C19 2D6 2E1 3A4
strng modr none 0 0 5 0 0 5 0 4 1 0 0 5 0 0 5 0 4 1
>0.5
1A2 2C9 2C19 2D6 2E1 3A4
strng modr none 0 5 3 9 0 0 0 8 1 9 0 0 9 0 0 7 0 2
Group 7
Group 9
Fig. (24). Decision tree model for classifying CYP substrates. The decision tree was constructed by using the multi-objective recursive partitioning method. Each value indicates the count of compounds belonging to each category. Cited from Fig. (2) in Ref. [110].
Data Modeling and Chemical Interpretation
root ‘QAvgNeg’ <0.133 ‘QAvgPos’<0.7912 ‘VMcGowan’<182.6
Frontiers in Drug Design & Discovery, 2009, Vol. 4 417
‘QAvgPos’>0.7912 ‘VMcGowan’<487.1 ‘HBAo’ <1.5 ‘MaxQ’ <0.12209
#2
‘VMcGowan’ >487.1 ‘MaxQ’ >0.12209 ‘SsssCH’ <0.076
#8 #3
‘VMcGowan’>182.6
‘SsssCH’ >0.076
#5
#4 #9
‘HBAo’ >0.5 ‘HBDn’ <0.5
#6 ‘HBDn’ >0.5
#7
‘QAvgNeg’>0.133
#1
Fig. (25). Extended Heiankyo View image of hierarchically structured data involving CYPmediated drug metabolism. Each 6-colored rectangular icon indicates compounds where the brightness of each color represents the metabolic susceptibility toward each CYP inform: CYP1A2 (red), CYP2C9 (yellow), CYP2C19 (green), CYP2D6 (cyan), CYP2E1 (blue), and CYP3A4 (magenta). Rectangular borders represent hierarchically organized group structures based on the splitting rules of the decision tree. Cited from Fig. (3) in Ref. [110].
splitting rule is a McGowan molecular volume (VMcGowan) of less than 182.6. This suggests that CYP2E1 substrates are smaller compounds. In contrast, CYP3A4 substrates appear to be larger compounds, taking into account the fact that many of them belong to Group 3 and 4. The authors have reported that hierarchically visualization is useful for chemists in drug design. 4.3. Other Visualization Techniques Sammon Non-Linear Mapping (NLM) Sammon NLM is well suited for data visualization. NLM represent all relative distances between all pairs of compounds in the descriptor space. Therefore, the distance between two points on the map directly reflects the similarity of the compounds. NLM are unique due to their conceptual simplicity and ability to reproduce the topology and structure of the data space in a faithful and unbiased manner. In contrast, KNN shows a
418 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
significant speed gain compared to NLM. KNN is favorable for visualization and analysis of larger databases than NLM [99]. Embedded Method Ridder et al. have collected metabolic reactions and compiled reaction rules covering a broad range of human phase 1 and phase 2 from Metabolite database [111]. The reaction similarity was evaluated on the basis of reaction fingerprints. They used the calculated Soergel distance in combination with 2D projection based on stochastic proximity embedding to visualize the contents of the reaction database. This method optimizes the distances between points on 2D plane to correspond as much as possible to the distances calculated in the fingerprint space between all pairs of metabolic reactions. The resulting scatter plot provides 2D map of the metabolic reactions in which similar reactions are clustered together. This is like the classification of chemical reactions with KNN exemplified by Funatsu et al. [112, 113]. Two interesting visualization methods are introduced. Boyer et al. have calculated the relative frequency of a reaction occurring at a specific site, “occurrence ratio”, and have represented it in colors [114]. Mandagere et al. have proposed graphical model for estimating oral BA in human from Caco-2 permeability and in vitro liver enzyme metabolic stability [115]. 5. WEB APPLICATION Rapid developments in chemistry, screening and automation in the last decade have led to the generation of large amount of data in drug discovery process. As the amount of chemical and biological information increases, the success of drug discovery depends on how well information is integrated. The important step is to build an adaptable system to access and analyze the data. These data accessing and analysis tools must be disseminated to the research scientists in a timely and efficient manner. The ultimate goal is to associate data with empirical rules and to provide novel ideas to scientists. Among many communication tools, web application would be the most convenient way from both sides of provider and user. We will introduce two web applications done in industries. Amgen’s group has developed their own system inside, called as Amgen’s Data Access Analysis Prediction Tools (ADAAPT) [116]. Their system can provide a number of standard computational tools to perform property calculation, QSPR and statistical analyses. It also provides in-house models to calculate a drug-likeness score and absorption index. Property calculations can be performed on compounds in the data warehouse as well as new compounds. The structure can be copied to the table in ADAAPT client via copying and pasting ISIS/draw structures directly. User also have an option to input structures as SMILES, and SD files can be also read in directly. Users have an option to select and unselect types of properties to be calculated. The properties of selected structures are displayed and are color-coded according to pre-determined rules (Fig. (26)); green, yellow, magenta, and red colors are used to indicate good, warning, unreliable, and bad, respectively. Fig. (27) depicts how the absorption index is calculated. Compounds located inside of the 99% confidence ellipse are given a value of 0. Compounds located between the 99% and 95% confidence ellipse are given a value of 1, and compounds located outside of the 95% confidence ellipse are given a value of 2. The higher absorption index means that compound is less likely absorbed in oral administration.
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 419
Color Legend Good
Warning
Pad
Unreliable
PSA:
< 130 2 to 4 5
LogP [CLogP. ACDLogP . ALogP98]:
-2 to 3
ACDLogD [pH=2,4,6,5,7,4,10]:
<2
No of Acids: No of Bases:
<3
-3 to -2 Or 4.5 to 5.5 < -30, > 55 < -30, > 4 -3 to -2 Or 3 to 4 2 >2 >3 3
No of H-Acceptors:
< 450
MW:
>5 > 10
5 10 450 to 550
<5 < 10
No of H-Donors:
> 140
130 to 140
> 550
(a) ADAPT- Amgen’s Data Analysis Prediction Tools (4.08.10) File Fdit Table Data Access Analysls Visualization Prediotion Tools Window Help R
name
MW
structure
PSA
cLogP
NheavyAtoms
Rolbon
N
1
N
MDDR327280
N
S
N N
S
N
450.6186
4.2930
50.0800
31.0000
237.2943
2.3730
30.7100
16.0000
528.4254
3.7520
119.3900
35.0000
C1
MDDR327280
N
2
MDDR147076
S
F
N N
MDDR147076
N
S
3
MDDR197003
O F
N F
N
F
O
F
N
F
F
O O
O
MDDR197003 F N S
(b) Fig. (26). Color Legend window (a) and a table containing structures and colored calculated properties (b). Cited from Fig. (3) in Ref. [116].
We have developed web application for medicinal chemists [117]. Two software are mainly used for web application. For calculation of ADME properties, ADMET Predictor is employed as the best quality of product [118, 119]. ADMET Predictor has the own global models such as pKa, ClogP, human intestinal absorption, and water solubility etc. Furthermore, recently, ADMET Predictor can provide CYP kinetics values, Km and Vmax against five CYP isoenzymes. Because there are many statistical tools implemented, the local model limiting chemical skeleton can be easily constructed. When chemists
420 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
ADAPT - Amgen’s Data Analysis Prediction Tools (4.08.10) File Edit Table Data Access Analysis Visualization Prediction Tools Window Help
Similazrity Se
Absorption Index
name
name
psa
clogp
index
1
MFCD03645423
1
MFCD03645423
43.860001
4.093000
2
MFCD03645423
2
MFCD03645423
43.860001
4.093000 4.619000
PSA vs. Clogp
3.906000
ClogP 8.00
3.258000
7.14
2.692000
6.29
2.996000
5.43
2.246000
4.57
3.536000
3.750000
4.480000
3.71
1.292000
2.86
1 Selected
2.00 1.14 0.29 -0.57 -1.43
N
-2.29
N
-3.14 -4.00 0.00
N
133.33
66.67 PSA
O
O S O
2 MFCD01936602
Fig. (27). Absorption index calculation example using a set of ACD compounds. PSA vs. ClogP plot is shown, and the structure of a data point colored in red is displayed. Cited from Fig. (7) in Ref. [116].
submitting ISIS/Draw structure, ChimePro automatically translates 2D image structure into 2D mol file. ADMET Predictor receives 2D mol file, and then calculates several ADME properties based on global or local models. The necessary values are extracted from the output file using Perl script. These values are displayed in web page using CGI protocol. (Fig. (28)) For prediction of CYP oxidation sites, MetaSite is employed [120]. (Fig. (29)) Submitting chemical structure from ISIS/Draw, MetaSite calculates a metabolism score to each heavy atom. The higher score value means that the corresponding heavy atom is more likely oxidized. The output page displays the 3D structure with atom
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 421
labels and scores. Now, web page supports the prediction of metabolites against three specific organs (liver, skin, brain) that user can choose freely.
Query Structure
Prediction
O
F
Results
molname
NoName
Acid pKa
4.03
Base pKa 3.67405 >
logP (caution: not ClogP) Human jejunal permeability (cm/s*10 4) >
O
9.35438
MDCK permeability (cm/s* 10 7)
163.902
Water solubility (mg/ml)
0.0148024
BBB penetration
Undecided
Percent unbound to blood plasma proteins (%) 1.35346 Volume of distribution (L/kg)
0.159123
Km value against CYP1A2
39.0677
Km value against CYP2C19
64.7945
Km value against CYP2C9
14.0065
Km value against CYP2D6
87.6499
Km value against CYP3A4
192.301
Fig. (28). Web page of ADMET Predictor.
6. CONCLUSION Concerning in silico ADME prediction, there are three main issues that we have to be addressed. 1) to improve the quality and quantity of data; 2) to develop predictive model with high accuracy 3) to develop visualization tool for chemical interpretation. There is little doubt that significant improvements in the quality and size of ADME database will lead to significant improvements in the quality of ADME predictions. In pharmaceutical companies, many kinds of ADME databases are integrated and the prediction tools are being developed routinely [2]. Furthermore, recent progress of in vitrobased technologies corresponding to each ADME process would facilitate PBPK modeling [10]. Absorption of drugs from the gastrointestinal tract is complex and can be influenced by many factors. The factors would fundamentally be classified into three categories; physicochemical factors (pKa, solubility, stability, diffusivity, lipophilicity, and salt forms), physiological factors (gastroinstetinal pH, gastric emptying, small and large bowel transit times, active transport and efflux, and gut wall metabolism), and formulation factors (particle size and crystal form, and dosage forms such as solution, tablet, capsule, suspension, emulsion, gel, and modified release). Kinetic parameters for drug metabolism such as Km, Vmax etc. are obtained from in vitro experiments using liver
422 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
microsomes and hepatocytes. Physicochemical properties of compound could be predicted from molecular structure alone. The in vitro data and physicochemical properties can serve as inputs for simulation to predict PK parameters, such as BA, CL, Vss. The PBPK approach will help to realize the goal of fully characterizing PK of a compound early in the drug discovery process. GastroPlus seems to be a promising product as PBPK tool. GastroPlus has unique physiological models of gastrointestinal tract for human, beagle dog, rat, mouse, rabbit, cat, and monkey. Furthermore, it is flexibly linked to ADME calculation product ‘ADMET Predictor’ [121, 122].
F
O
Input molecule:
O
Atom number Atom name Atom type Score C16 C1 C2 C3 C4
2
C2
C.ar 11.000
1
C1
C.ar 5.500
3
C3
C.ar 5.500
16
C16
C.3 3.667
4
C4
C.ar 2.517
All data (csv file)
Fig. (29). Web page of MetaSite.
Combinatorial QSAR (CQSAR) has been used for developing the predictive models [123-125]. There exists no gold standard QSAR approach that guarantees the best model for every data set. CQSAR explores various combinations of optimization methods and descriptor types and includes rigorous and consistent validation. The exploratory nature of CQSAR helps in identifying highly predictive models for particular data set, whereas a conventional approach to QSAR studies using only one method and one type of descriptors has a higher chance to fail. CQSAR allows full automation, and this is highly suited to the recent drug discovery situation. The domain of applicability is another useful concept for developing the predictive models [126, 127]. This allows one can use the distance to model space to estimate the reliability of a given prediction. This is particular important for local model since it is relatively easy to stray outside the predictive domain of the model with a small number of chemistry iterations.
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 423
With virtual ADME there is a growing need for significantly improved data handling. Given that many screening and ADME predictors generate dozens of data points for each molecule, there is a clear need to develop far more sophisticated data reduction and visualization. Spotfire has become popular since it has many unique visualization modules such as heat maps, hierarchical clustering diagrams and color-coded symbols [128, 129]. We can imagine future of in silico ADME embedded in drug discovery. A dream system is presented in Fig. (30) [130-133]. A drug candidate is first presented to the QSAR and metabolic fate prediction modules. The system also predicts the structure of the first-pass metabolites and transition states. Predicted structures are again fed to the QSAR module. Metabolites are prioritized based on relative predicted reaction rates. Valuable information about metabolic fate and major metabolites is then placed into the context of pathways analysis. Information on predicted structures, their biological activity and the enzymes involved is merged with information on pathways. Simultaneously, metabonomic and toxicogenomic data are mapped onto the pathways. After going through many filters, we can select most promising compounds virtually. This kind of infrastructure system has been developed in pharmaceutical companies and ADME models play more important roles in drug discovery process.
QSAR module
Predict Binder and/or non-binder, Substrate and/or non-substrate, Inhibitor and/or Inducer
Prioritize metabolites
CH3 NH
NCEs
NH O
Model reaction module
O CH3
CH3
Select most promising compounds
Visualize and analyze Interference with normal metabolism. (possible bottlenecks, side-effects, important SUPs)
Visualization tools
CH3
Predict metabolic routes and metabolite structures
Pathway database and/or maps Toxicity arrays
Fig. (30). Proposed functionality of the next-generation ADME/Tox platform. Cited from Fig. (4) in Ref. [130].
424 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Hasegawa and Funatsu
ACKNOWLEDGEMENT We would like to thank Dr. Arakawa and Mr. Kaneko, and Mr. Koyama at The University of Tokyo for their helps when preparing this review article. ABBREVIATIONS ADAAPT
=
Amgen’s Data Access Analysis Prediction Tools
ADME
=
Absorption, distribution, metabolism and excretion
ANN
=
Artificial neural networks
BA
=
Bioavailability
BBB
=
Blood brain barrier
BNN
=
Bayesian neural networks
BP
=
Back propagation
CART
=
Classification and regression trees
CI
=
Contribution index
CQSAR
=
Combinatorial QSAR)
CV
=
Cross-validation
CYP
=
Cytochrome P450
DT
=
Decision tress
GA
=
Genetic algoritm
GP
=
Gaussian process
HLM
=
Human liver microsome
HM
=
Heuristic method
HSA
=
Human serum albumin
ICA
=
Independent component analysis
kNN
=
k-Nearest neighbors
KNN
=
Kohonen neural networks
LOO
=
Leave-one-out
MLR
=
Multiple linear regressions
NLM
=
Non-linear mapping
Nmt inhibitor
=
N-Myristoyltransferase inhibitor
ORMUCS
=
Ordered multi-categorical classification
PBPK modeling
=
Physiologically based pharmacokinetic modeling
PCA
=
Principal component analysis
P-gp
=
P-Glycoproteins
Data Modeling and Chemical Interpretation
Frontiers in Drug Design & Discovery, 2009, Vol. 4 425
PK
=
Pharmacokinetics
PLS
=
Partial least squares
QSAR
=
Quantitative structure-pharmacokinetic relationship
QSPKR
=
Quantitative structure-pharmacokinetic relationship
QSPR
=
Quantitative structure-property relationship
RF
=
Random forest
RMSE
=
Root mean squares error
RST
=
Rough set theory
SVM
=
Support vector machines
t1/2
=
Elimination half-life
Vss
=
Volume of distribution
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31]
Waterbeemd, H. van de In Molecular Drug Properties in Methods and Principles in Medicinal Chemistry; Mannhold, R., Ed.; Wiely: Amsterdam, 2007; Vol. 37, pp. 25-52. Wishart, D.S. Drugs R. D., 2007, 8, 349. Avdeef, A.; Bendels, S.; Di, L.; Faller, B.; Kansy, M.; Sugano, K.; Yamauchi, Y. J. Pharm. Sci., 2007, 96, 2893. Yu, H.; Adedoyin, A. Drug Discov. Today, 2003, 8, 852. Beresford, A.P.; Selick, H.E.; Tarbit, M.H. Drug Discov. Today, 2002, 7, 109. Francois, P.; Olivier, B.; Elodie, D.; Ismail, I.; Mary, D.; Sophie, O.; Andre, M. Exp. Opin. Drug Dis., 2006, 1, 737. Lipinski, C.A.; Lombardo, F.; Dominy, W.B.; Feeney, P.J. Adv. Drug Deliv. Rev., 1997, 23, 3. Yamashita, F.; Hashida, M. Drug Metab. Pharmacokin., 2004, 19, 327. Arimoto, R. Curr. Top. Med. Chem., 2006, 6, 1609. Subramanian, K. Exp. Opin. Drug Metab. Tox., 2005, 1, 555. Tingjun, H.; Junmei, W.; Wei, Z.; Wei, W.; Xiaojie, X. Curr. Med. Chem., 2006, 13, 2653. Gola, J.; Obrezanova, O.; Champness, E.; Segall, M. QSAR, 2006, 25, 1172. Thomas, F.; Kriegl, J.M. Curr. Top. Med. Chem., 2006, 6, 1579. Pelkonen, O.; Raunio, H. Exp. Opin. Drug Metab. Tox., 2005, 1, 49. Stefan, W.; Jörg, L.; Walter, S. Exp. Opin. Drug Metab. Tox., 2005, 1, 159. Gombar, V.K.; Silver, I.S.; Zhao, Z. Curr. Top. Med. Chem., 2003, 3, 1205. George, M.G.; Patrick, J.S. Drug Discov. Today, 2001, 6, s54. Refsgaard, H.H.F.; Jensen, B.F.; Christensen, I.T.; Hagen, N.; Brockhff, P.B. Drug Develop. Res., 2006, 67, 417. Tetko, I.V.; Bruneau, P.; Mewes, H.-W.; Rohrer, D.C.; Poda, G.I. Drug Discov. Today, 2006, 11, 700. Hansch, C.; Leo, A.; Mekapati, S.B.; Kurup, A. Bioorg. Med. Chem., 2004, 12, 3391. Hasegawa, K.; Funatsu, K. SAR QSAR Env. Res., 2000, 11, 189. Norinder, U.; Haeberlein, M. Adv. Drug Deliv. Rev., 2002, 54, 291. Livingstone, D.J. Curr. Top. Med. Chem., 2003, 3, 1171. Butina, D.; Gola, J.M.R. J. Chem. Inf. Comput. Sci., 2003, 43, 837. Narayanan, R.; Gunturi, S.B. Bioorg. Med. Chem., 2005, 13, 3017. Abraham, M.H.; Ibrahim, A.; Zissimos, A.M.; Zhao, Y.H.; Comer, J.; Reynolds, D.P. Drug Discov. Today, 2002, 7, 1056. Klopman, G.; Zhu, H. J. Chem. Inf. Comput. Sci., 2001, 41, 439. Rose, K.; Hall, L.H.; Kier, L.B. J. Chem. Inf. Comput. Sci., 2002, 42, 651. Hall, L.M.; Hall, L.H.; Kier, L.B. J. Chem. Inf. Comput. Sci., 2003, 43, 2120. Zhao, Y.H.; Le, J.; Abraham, M.H.; Hersey, A.; Eddershaw, P.J.; Luscombe, C.N.; Boutina, D.; Beck, G.; Sherborne, B.; Cooper, I.; Platts, J.A. J. Pharm. Sci., 2001, 90, 749. Raevsky, O.A.; Schaper, K.-J. QSAR, 2003, 22, 926.
426 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80]
Hasegawa and Funatsu
Schaper, K.-J.; Kunz, B.; Raevsky, O.A. QSAR, 2003, 22, 943. Hou, T.J.; Zhang, W.; Xia, K.; Qiao, X.B.; Xu, X.J., J. Chem. Inf. Comput. Sci., 2004, 44, 1585. Hou, T.; Wang, J.; Li, Y. J. Chem. Inf. Model., 2007, 47, 2408. Wichmann, K.; Diedenhofen, M.; Klamt, A. J. Chem. Inf. Model., 2007, 47, 228. Hasegawa, K.; Shindoh, H.; Shiratori, Y.; Ohtsuka, T.; Aoki, Y.; Ichihara, S.; Horii, I.; Shimma, N. J. Chem. Inf. Comput. Sci., 2002, 42, 968. Colmenarejo, G.; Alvarez-Pedraglio, A.; Lavandera, J.-L. J. Med. Chem., 2001, 44, 4370. Yoshida, F.; Topliss, J.G. J. Med. Chem., 2000, 43, 2575. Hasegawa, K.; Arakawa, M.; Funatsu, K. Curr. Comput. -Aided Drug Des., 2005, 1, 129. Susanne, W.; Fredrik, A. Hans, L.; Anders, H.; Curt, P.; Anders, K. J. Mol. Graph. Model., 2003, 21, 273. Bergström, C.A.S.; Norinder, U.; Luthman, K.; Artursson, P. Pharm. Res., 2002, 19, 182. Milletti, F.; Storchi, L.; Sforna, G.; Cruciani, G. J. Chem. Inf. Model., 2007, 47, 2172. Moda, T.L.; Montanari, C.A.; Andricopulo, A.D. Bioorg. Med. Chem., 2007, 15, 7738. Crivori, P.; Zamora, I.; Speed, B.; Orrenius, C.; Poggesi, I. J. Comput. -Aided Mol. Des., 2004, 18, 155. Wolohan, P.R.N.; Clark, R.D.; J. Comput. Aided Mol. Des., 2003, 17, 65. Kratochwil, N.A.; Huber, W.; Müller, F.; Kansy, M.; Gerber, P.R. Biochem. Pharmacol., 2002, 64, 1355. Winiwarter, S.; Ax, F.; Lennernäs, H.; Hallberg, A.; Pettersson, C.; Karlén, A. J. Mol. Graph. Model., 2003, 21, 273. Norinder, U.; Österberg, T. J. Pharm. Sci., 2001, 90, 1076. Sun, H. J. Chem. Inf. Comput. Sci., 2004, 44, 748. Gleeson, M.P. J. Med. Chem., 2007, 50, 101. Afzelius, L.; Masimirembma, C.M.; Karlen, A.; Andersson, T.B.; Zamra, I. J. Chomput. Aided Mol. Des., 2007, 16, 443. Gleeson, M.P.; Waters, N.J.; Paine, S.W.; Davis, A.M. J. Med. Chem., 2006, 49, 1953. Crivori, P.; Cruciani, G.; Carrupt, P.-A.; Testa, B. J. Med. Chem., 2000, 43, 2204. Singh, S.B.; Shen, L.Q.; Walker, M.J.; Sheridan, R.P. J. Med. Chem., 2003, 46, 1330. Kaneko, H.; Arakawa, M.; Funatsu, K. J. Chem. Inf. Model., 2008, 48, 534. Hasegawa, K.; Deushi, T.; Yaegashi, O.; Miyashita, Y.; Sasaki, S. Eur. J. Med. Chem., 1995, 30, 569. Winkler, D.A. Drugs of the Future, 2004, 29, 1043. Huuskonen, J.; Rantanen, J.; Livingstone, D. Eur. J. Med. Chem., 2000, 35, 1081. Liu, R.; So, S.-S. J. Chem. Inf. Comput. Sci., 2001, 41, 1633. Goeller, A.H.; Hennemann, M.; Keldenich, J.; Clark, T. J. Chem. Inf. Model., 2006, 46, 648. Yaffe, D.; Cohen, Y.; Espinosa, G.; Arenas, A.; Giralt, J. Chem. Inf. Comput. Sci., 2002, 42, 162. Catana, C.; Gao, H.; Orrenius, C.; Stouten, P.F.W. J. Chem. Inf. Comput. Sci.,2005, 45, 170. Tetko, I.V.; Tanchuk, V.U.; Kasheva, T.N.; Villa, A.E.P. J. Chem. Inf. Comput. Sci., 2001, 41, 1488. Tetko, I.V. J. Chem. Inf. Comput. Sci., 2002, 42, 717. Garg, P.; Verma, J. J. Chem. Inf. Model, 2006, 46, 289. Tetko, I.V.; Tanchuk J. Chem. Inf. Comput. Sci., 2002, 42, 1136. McElroy, N.R.; Jurs, P.C. J. Chem. Inf. Comput. Sci., 2001, 41, 1237. Klon, A.E.; Lowrie, J.F.; Diller, D.J. J. Chem. Inf. Model., 2006, 46, 1945. Pelletier, D.J.; Gehlhaar, D.; Tilloy-Ellul, A.; Johnson, T.O.; Greene, N. J. Chem. Inf. Model., 2007, 47, 1196. Bruneau, P.; McElroy, N.R. J. Chem. Inf. Comput. Sci., 2006, 46, 1379. Lee, P.H.; Cucurull-Sanchez, L.; Lu, J.; Du, Y.J. J. Comput. Aided Mol. Des., 2007, 21, 665. Arakawa, M.; Hasegawa, K.; Funatsu, K. Chemom. Intell. Lab. Syst., 2008, 92, 145. Matthew, W.B. Trotter, S.B.H. QSAR, 2003, 22, 533. Kriegl, J.M.; Arnhold, T., Beck, B.; Fox, T. J. Comput. Aided. Mol. Des., 2005, 19, 189. Huang, J.; Ma, G.; Muhammad, I.; Cheng, Y. J. Chem. Inf. Model., 2007, 47, 1638. Li, H.; Yap, C.W.; Ung, C.Y.; Xue, Y.; Cao, Z.W.; Chen, Y.Z. J. Chem. Inf. Model., 2005, 45, 1376. Yap, C.W.; Chen, Y.Z. J. Chem. Inf. Model., 2005, 45, 982. Xue, C.X.; Zhang, R.S.; Liu, H.X.; Yao, X.J.; Liu, M.C.; Hu, Z.D.; Fan, B.T. J. Chem. Inf. Comput. Sci., 2004, 44, 1693. Sitarama, B.G.; Ramamurthi, N. QSAR, 2007, 26, 653. Refsgaard, H.H.F.; Jensen, B.F.; Brockhoff, P.B.; Padkjar, S.B.; Guldbrandt, M.; Christensen, M.S. J. Med. Chem., 2005, 48, 805.
Data Modeling and Chemical Interpretation [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [120] [121] [122]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 427
Jensen, B.F.; Vind, C.; Padkjar, S.B.; Brockhoff, P.B.; Refsgaard, H.H.F. J. Med. Chem., 2007, 50, 501. Nigsch, F.; Bender, A.; van Buuren, B.; Tissen, J.; Nigsch, E.; Mitchell, J.B.O. J. Chem. Inf. Model., 2006, 46, 2412. Obrezanova, O.; Csanyi, G.; Gola, J.M.R.; Segall, M.D. J. Chem. Inf. Model., 2007, 47, 1847. Obrezanova, O.; Gola, J.M.R.; Champness, E.J.; Segall, M.D. J. Comput. Aided Mol. Des., 2008, 22, 431. Sheridan, R.P.; Korzekwa, K.R.; Torres, R.H.; Walker, M.J. J. Med. Chem., 2007, 50, 3173. Sakiyama, Y.; Yuki, H.; Moriya, T.; Hattori, K.; Suzuki, M.; Shimada, K.; Honma, T. J. Mol. Graph. Model., 2008, 26, 907. DeLisle, R.K.; Dixon, S.L. J. Chem. Inf. Comput. Sci., 2004, 44, 862. Deconinck, E.; Zhang, M.H.; Coomans, D.; Heyden, Y.V. J. Chem. Inf. Comput. Sci., 2006, 46, 1410. Deconinck, E.; Xu, Q.S.; Put, R.; Coomans, D.; Massart, D.L.; Vander Heyden, Y. J. Pharm. Biomed. Anal., 2005, 39, 1021. Deconinck, E.; Hancock, T.; Coomans, D.; Massart, D.L.; Vander Heyden, Y. J. Pharm. Biomed. Anal., 2005, 39, 91. Bai, J.P.F.; Utis, A.; Crippen, G.; He, H.-D.; Fischer, V.; Tullman, R.; Yin, H.-Q.; Hsu, C.-P.; Jiang, L.; Hwang, K.-K. J. Chem. Inf. Comput. Sci., 2004, 44, 2061. Hasegawa, K.; Koyama, M.; Arakawa, M.; Funatsu, K. J. Comput. Aided Des., 2008, 9, 1. Koyama, M. 8th ICCS conference 2008, P60, University of Hamburg. Gleeson, M.P. J. Med. Chem., 2008, 51, 817. Martin, Y.C. J. Med. Chem., 2005, 48, 3164. Daszykowski, M.; Walczak, B.; Massart, D.L. Chemom. Intell. Lab. Syst., 2003, 65, 97. Hasegawa, K.; Morikami, K.; Shiratori, Y.; Ohtsuka, T.; Aoki, Y.; Shimma, N. Chemom. Intell. Lab. Syst., 2003, 69, 51. Arakawa, M.; Hasegawa, K.; Funatsu, K. Curr. Comput. -Aided Mol. Des., 2007, 3, 341. Balakin, K.V.; Ivanenkov, Y.A.; Savchuk, N.P.; Ivashchenko, A.A.; Ekins, S. Curr. Drug Dis. Tech., 2005, 2, 99. Yan, A.; Gasteiger, J. J. Chem. Inf. Comput. Sci., 2003, 43, 429. Maniyar, D.M.; Nabney, I.T.; Williams, B.S.; Sewing, A. J. Chem. Inf. Model, 2006, 46, 1806. Korolev, D.; Balakin, K.V.; Nikolsky, Y.; Kirillov, E.; Ivanenkov, Y.A.; Savchuk, N.P.; Ivashchenko, A.A.; Nikolskaya, T. J. Med. Chem., 2003, 46, 3631. Chen, X.; Rusinko III, A.; Yong, S.S. J. Chem. Inf. Comput. Sci., 1998, 38, 1054. Terfloth, L.; Bienfait, B.; Gasteiger, J. J. Chem. Inf. Model., 2007, 47, 1688. Lamanna, C.; Bellini, M.; Padova, A.; Westerberg, G.; Maccari, L.C. J. Med. Chem., 2008, 51, 2891. Matthew G.H.; Jeffrey P.J. J. Med. Chem., 2006, 49, 4367. Mente, S.R.; Lombardo, F. J. Comput. Aided Mol. Des., 2005, 19, 465. Bursi, R.; de Gooyer, M.E.; Grootenhuis, A.; Jacobs, P.L.; van der Louw, J.; Leysen, D. J. Mol. Graph. Model., 2001, 19, 552. Burton, J.; Ijjaali, I.; Barberan, O.; Petitet, F.; Vercauteren, D.P.; Michel, A. J. Med. Chem., 2006, 49, 6231. Yamashita, F.; Hara, H.; Ito, T.; Hashida, M. J. Chem. Inf. Model., 2008, 48, 364. Ridder, L.; Wagener, M. Chem. Med. Chem., 2008, 3, 821. Sato, H.; Sacher, O.; Nakata, T.; Chen, L., Gasieiger, J.; Funatsu, K. J. Chem. Inf. Comput. Sci., 1998, 38, 210. Sato, H.; Itono, S.; Funatsu, K.; Takano, K.; Nakata, T. J. Chem. Inf. Comput. Sci., 1999, 39, 671. Boyer, S.; Arnby, C.H.; Carlsson, L.; Smith, J.; Stein, V.; Glen, R.C. J. Chem. Inf. Model., 2007, 47, 583. Mandagene, A.K.; Thompson, T.N.; Hwang, K.-K. J. Med. Chem., 2002, 45, 304. Cho, S.J.; Sun, Y.; Harte, W. J. Comput. Aided Mol. Des., 2006, 20, 249. Hasegawa, K. The inter-discipline between chemoinformatics and bioinformatics, CICSJ Bulletin., 2008, 26, 34. Dearden, J.C. Exp. Opin. Drug Dis., 2006, 1, 31. Fraczkiewicz, R. In Comprehensive Medicinal Chemistry II; Testa, B.; van de Waterbeemd, H., Eds., Elsevier: Oxford, 2006; Vol5, pp. 603-626. Cruciani, G.; Carosati, E.; De Boeck, B.; Ethirajulu, K.; Mackie, C.; Howe, T.; Vianello, R. J. Med. Chem., 2005, 48, 6970. Tubic-Grozdanis, M.; Bolger, M.B.; Langguth, P. AAPS Journal, 2008, 10, 213. Kuentz, M.; Nick, S.; Parrott, N.; Rothlisberger, D. Eur. J. Pharm. Sci., 2006, 27, 91.
428 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133]
Hasegawa and Funatsu
Yap, C.W.; Li, Z.R.; Chen, Y.Z. J. Mol. Graph. Model., 2006, 24, 383. Lima, P.C.; Golbraikh, A.; Oloff, S.; Xiao, Y.; Tropsha, A. J. Chem. Inf. Model., 2006, 46, 1245. Votano, J.R.; Parham, M.; Hall, L.M.; Hall, L.H.; Kier, L.B.; Oloff, S.; Tropsha, A. J. Med. Chem., 2006, 49, 7169. Weaver, S.; Gleeson, M.P. J. Mol. Graph. Model., 2008, 26, 1315. Bidault, Y. Expert Opin. Drug Metab. Toxicol., 2006, 2, 157. Stoner, C.L.; Gifford, E.; Stankovic, C.; Lepsy, C.S.; Brodfuehrer, J.; Vara Prasad, J.V.N.; Surenfram, N. J. Pharm. Sci., 2004, 93, 1131. Jenkins, K.M.; Angeles, R.; Quintos, M.T.; Xu, R. J. Pharm. Biomed. Anal., 2004, 34, 989. Bugrim, A.; Nikolskaya, T.; Nikolsky, Y. DDT, 2004, 9, 127. Jeffrey A. Drug Discov. Today, 2002, 7, 315. Jennifer, F. Expert Opin. Drug Metab. Toxicol., 2005, 1, 565. Segall, M.D; Beresford, A.P; Gola, J.M.R.; Hawksley, D.; Tarbit, M.H. Expert Opin. Drug Metab. Toxicol., 2006, 2, 325.
Frontiers in Drug Design & Discovery, 2009, 4, 429-453
429
A Review on Virtual Reality and Haptics Approaches in Drug Design and Discovery Susana K. Lai-Yuen* Department of Industrial & Management Systems Engineering, University of South Florida, Tampa, FL 33620-5350, USA Abstract: Virtual reality interfaces and haptics are rapidly becoming a powerful technology to enable researchers to interactively manipulate and evaluate potential drug molecules in an immersive virtual environment to accelerate the drug design process. Virtual reality refers to a computer-generated and interactive three-dimensional environment that immerses people into a virtual world while haptic devices are electromechanical devices that exert forces on users giving the illusion of touching something in the simulated environment. As molecular forces play a major role in determining the successful docking of drug molecules, virtual reality and haptics can provide researchers with invaluable human-computer interface tools for visualizing, manipulating, and “feeling” complex molecular systems in real time. The force feedback provided by haptic devices can direct researchers towards favorable drug molecule positions and orientations increasing the understanding of key forces during molecular interactions and enabling new kinds of drug design exploration. However, the main difficulty of modeling molecular systems through virtual reality and haptics is that visualization models and simulations need to be processed rapidly to satisfy the update requirements needed for real-time visualization and sense of touch. Any time delay between a user action and the corresponding update of the virtual object can lead to unrealistic visualization, unstable force response, and simulation sickness. This paper reviews some of the research advances for addressing these computational challenges ranging from new graphical representations of molecules for effective haptic force feedback calculation to virtual reality algorithms and devices for modeling complex molecular systems in a real-time virtual environment.
1. INTRODUCTION The design and discovery of new drugs is one of the most complex, time-consuming, and cost-intensive process in the pharmaceutical industry [1, 2]. Advances in computer technology and scientific visualization techniques have been vital in speeding the development of new drugs while reducing development costs. Computational tools increase current molecular modeling capabilities providing new insights into the properties of
*Corresponding Author: E-mail:
[email protected] Gary W. Caldwell / Atta-ur-Rahman / Z. Yan / M. Iqbal Choudhary (Eds.) All rights reserved – © 2009 Bentham Science Publishers.
430 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
macromolecules and their interactions with ligands. Molecular interactions can be accurately represented by computer simulation methods such as molecular dynamics. These methods simulate very detailed molecular motions based on physical laws. However, current tools are computationally intensive and cannot account for every part of a problem as complex as drug design without the knowledge and intuition of the researcher. Moreover, these tools mainly provide a two-dimensional visual graphic display limiting the visualization and understanding of three-dimensional molecular structures and their interactions. These shortcomings can be addressed by developing new computational approaches that incorporate immersive sensory experiences to speed molecular interaction simulations and to enable researchers to visualize, feel, and interact with the simulation in real-time. An immersive and interactive virtual environment can provide researchers with a more meaningful modeling process to identify key molecular forces based on physical and chemical principles and can be an invaluable design tool for drug design and discovery. Virtual reality is rapidly becoming a powerful technology that enables people to visualize, manipulate, and interact with a simulated environment for solving today’s realworld problems. Virtual reality can be defined as a “high-end user-computer interface that involves real-time simulation and interactions through multiple sensorial channels. These sensorial modalities are visual, auditory, tactile, smell, and taste” [3]. The combination of these sensorial modalities provides an immersive and interactive environment for the user where his/her instantaneous position and orientation are tracked. This enables the user to manipulate and interact with scientific data and complex engineering systems. The most common applications of virtual reality can be found in the areas of medicine, education, entertainment, design, manufacturing, and military [4-8]. For example, virtual surgical simulators are being developed to train medical students to operate on realistic models of biological tissues and organs. One of the emerging areas of virtual reality application is in drug design and discovery where virtual reality has shown to enhance the ability to understand properties of proteins and molecular docking [9-11]. Fig. (1) shows a research chemist at the National Institute of Standards and Technology (NIST) immersed in a 3D environment for studying the behavior of “smart gels” and for understanding the binding of their components [12].
Fig. (1). A virtual reality environment for studying molecular interactions at the National Institute of Standards and Technology (NIST) [12]. Reprinted with permission. Copyright Robert Rathe.
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 431
The majority of virtual reality simulations have focused on incorporating visual and auditory modalities in the form of 3D stereo displays and sound while smell and taste feedback are still at the early stage of research. In recent years, there is a growing interest to incorporate tactile modality in virtual environments by enabling users to touch and feel virtual objects using haptic devices. Haptic devices are electromechanical devices that exert forces on users giving the illusion of touching or feeling something in the simulated environment as shown in Fig. (2). The incorporation of haptics into a virtual environment enables a more immersive experience that can greatly increase the effectiveness of real-world based simulation. In drug design and discovery, haptics can enable users to manipulate virtual molecules and feel the molecular forces providing an important design and visualization tool to speed the drug design process. The development of new mechanisms for haptics and their demonstrated potential to make the virtual experience more realistic and interactive have made these devices a key component of virtual reality systems and an emergent area of research.
Fig. (2). Example of a haptic device for molecular docking.
In drug design, recent studies have shown that interaction between molecules can be better understood with the aid of a haptic device than by having a visual display alone [11, 13]. Experiments showed that chemists could perform the positioning of a rigidbody ligand inside the binding site of a receptor up to twice as fast with haptic feedback compared to visual display alone [14]. Scientists have also stated that the forces provided by a haptic device increased their understanding of how the drug fits into the receptor. Below is a list of the benefits of using virtual reality and haptics in drug design and discovery: • Increase understanding of molecular systems and their interactions through 3dimensional visualization and interactive force feedback. • Accelerate molecular simulations through the incorporation of researcher’s knowledge and intuition via haptic devices. The researcher can steer a simulation towards better molecular conformations or specify alternate starting points for the molecular docking process.
432 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
• Facilitate an interaction of the researcher with a molecular system by manipulating molecules as if they were solid physical models. This achieves better solutions than either the computer or the researcher could obtain alone. • Provide an immersive and intuitive human-computer interaction system to bring the molecular system to life. • Enable multiple researchers to work together. • Improve productivity in the pharmaceutical industry. Although there have been remarkable advances in virtual reality and haptics applications in drug design and discovery, there still remains a lot of challenges to achieve a complete virtual reality system for interacting with a molecular simulation. A convincing immersive and interactive virtual environment requires an extensive combination of high-performance computer hardware and software. An effective virtual reality display requires a minimum display rate of 10 frames per second although frame rates of 30 per second are preferred [15]. Consequently, “there is at most 0.1 second of time available to compute an instance of a molecular simulation, send the data over the network, and display the molecular model” [9]. Most current computational tools for modeling molecular interactions are unsuitable for real time applications as they are computationally intensive. Therefore, the main challenge is the need for modeling and simulation techniques to represent molecular interactions and behavior in a physically correct manner while satisfying the strict time constraints of virtual reality and haptic rendering. This paper presents an overview of the state-of-the-art in virtual reality and haptics for drug design and discovery. It begins with a brief description of the different types of commercially available virtual reality systems and haptic devices in Section 2. Section 3 provides the current issues and challenges for incorporating virtual reality and haptics in the drug design and discovery process. Section 4 presents an overview of representative research advances on virtual reality and haptics for facilitating the understanding of molecular interactions and for speeding the design and discovery of pharmaceutical drugs. Section 5 concludes with a discussion on some areas of future research work for achieving a complete immersive and interactive virtual system for drug design and discovery. 2. OVERVIEW OF VIRTUAL REALITY AND HAPTIC DEVICES Virtual reality interfaces and haptics are rapidly becoming a powerful technology to enable researchers to interactively manipulate and evaluate potential drug molecules in an immersive virtual environment to accelerate the drug design process. Virtual reality refers to a computer-generated and interactive three-dimensional environment that immerses people into a virtual world while haptic devices are electromechanical devices that exert forces on users giving the illusion of touching something in a simulated environment. A virtual reality system requires four crucial technologies [3]: a visual, aural and haptic display, a graphics rendering system to generate the constantly changing images, a tracking system to collect data on the user’s position and orientation, and a database construction and maintenance system to generate realistic virtual models. Fig. (3) shows the overall system using virtual reality and haptic devices for drug design and discovery. The user utilizes one of the several types of virtual reality devices that provides him/her with a 3-dimensional image of the molecular simulation. Virtual reality devices track the user’s position and orientation to provide adequate images to the
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 433
user. At the same time, the user uses a haptic device to manipulate molecules in a simulation (i.e., dock a ligand into a receptor, etc.) while feeling the molecular forces in real time.
Fig. (3). Overview of the application of virtual reality and haptic devices in drug design and discovery. * Picture courtesy of Dr. Christoph Sensen, Sun Center of Excellence for Visual Genomics [16]. ** Reproduced by permission of Immersion Corporation. Copyright © 2008 Immersion Corporation. All rights reserved [17].
The following sections briefly introduce different types of virtual reality and haptics devices that are currently commercially available to familiarize the reader with these types of devices. These devices will be referred throughout the paper as their specific applications in drug design and discovery are described in Section 4. This review does not intend to be a thorough description of all available virtual reality and haptic devices as the main focus of this paper is on their applications in drug design and discovery. For more detailed information, the reader is referred to the provided references. 2.1. Virtual Reality Devices Virtual reality is a powerful technology involving computers and various peripherals that enables users to visualize, manipulate and interact with simulated environments to
434 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
solve complex problems. Although there are many different types of virtual reality systems, they can be categorized based on the level of immersion provided. The major component for a virtual reality system is the display as it provides images and immersive presence within the environment. Users can look at a display such as that provided by a computer or be surrounded by several displays to experience a semi or fully immersive environment. Tracking devices are necessary to determine the position and orientation of the user(s) to display graphics accordingly. Users also utilize haptic devices to feel and interact with the virtual environment. Following is a description of the most commonly used displays for virtual environments. More detailed information on virtual reality hardware and software can be found in [3, 4, 18-20]. Head-Mounted Displays (HMD) A HMD is a set of goggles worn on the head or mounted on a helmet that generate images through small monitors placed in front of each eye as shown in Fig. (4). Separate views of the image are provided to the left and right eye creating a fully immersive virtual environment. A tracking device is attached to the HMD to generate a stereoscopic view that changes as the user’s head position and orientation change.
Fig. (4). Example of a head-mounted display. VR1280 picture courtesy of Virtual Research Systems, Inc. [21].
Projection Displays Projection displays provide field sequential stereoscopic images to a static screen. These displays allow more than one person to see the virtual environment enabling collaborative work. Field sequential stereoscopic systems sequentially display the left and right eye views of an object while the user wears special stereo-glasses to view images. The stereo-glasses are synchronized with the display screen to ensure that each eye of the user receives the correct view of the image. As shown in Fig. (5), there are three types of projection-based display systems: workbench systems, wall systems, and immersive rooms [20]. As shown in Fig. (5a), a workbench has the image projected towards the surface of a table where users stand around the table using shutter glasses and interact with the images using haptic devices. In wall systems, images are projected onto a large vertical screen where many users can see the screen as shown in Fig. (5b). Finally, immersive room systems are a multiple-person theater with surround view and sound. The images are rear projected to the walls of a theater by multiple projectors as shown in Fig. (5c). These systems provide full immersion into scientific data as they enclose the user.
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 435
(a) Example of a workbench display. Picture courtesy of Dr. Russell M. Taylor II, University of North Carolina at Chapel Hill [22].
(b) Example of a wall system display. Picture courtesy of Barco, Inc. [23].
(c) Example of an immersive room system. Picture courtesy of Dr. Christoph Sensen, Sun Center of Excellence for Visual Genomics [16]. Fig. (5). Categories of virtual reality displays.
436 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
2.2. Haptic Devices “Haptics” comes from the Greek word haptesthai that means “to touch”. Haptics is a technology that enables users to touch, feel and manipulate virtual objects in a simulated environment. It is a computer-human interface that enables bidirectional communication between human and machine through touch in response to user movements. Haptics is being increasingly used as a key part of virtual reality systems as it adds tactile and force feedback to the visualization making virtual experience more realistic and interactive. Tactile feedback enables the user to feel surface geometry, smoothness, and slippage while force feedback reproduces the object’s hardness, weight, and inertia [24, 25]. In drug design, force feedback is useful in providing the user with a feeling of molecular forces created between molecules during molecular docking. It also prevents the user from penetrating virtual objects such as atoms. Haptic devices measure the position of a user’s hand or other parts of the body to provide force and torque feedback to the user. There are different types of haptic devices available in the market as well as those developed by laboratories. The devices vary based on structural frame, workspace, stiffness, maximum vs. sustained force and torque, among others. This section briefly describes the different types of haptic devices that are commercially available. More detailed information on various commercial and laboratory haptic devices can be found in [25-28]. Haptic devices can be ground-based (grounded to a desk or floor) or body-based (attached to parts of the body): Ground-Based Haptic Devices Haptic devices that are grounded to a desk (or desktop devices) are the most widely used nowadays due to their easiness for installation and commercial availability. Ground-based devices are used with elbow or wrist support and controlled through the fingers, hand, arm or a handle. Fig. (6) shows an example of these devices that consists
Fig. (6). Example of a desktop haptic device. SensAble PHANTOM® Desktop™ haptic device [29]. © Copyright SensAble Technologies, Inc. PHANTOM, PHANTOM Desktop, SensAble, and SensAble Technologies, Inc. are trademarks or registered trademarks of SensAble Technologies, Inc. (Reprinted with permission).
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 437
of a handle interface with a frame. Some devices can provide force feedback in six degrees of freedom (6DOF) providing the user with both force and torque feedback, which are useful in molecular docking for facilitating proper alignment of molecules. Users control virtual objects through the handle or fingertip interface and receive force or force and torque feedback at a single point through the devices.
Fig. (7). Example of a wearable haptic device. Reproduced by permission of Immersion Corporation. Copyright © 2008 Immersion Corporation. All rights reserved. [17].
Body-Based Haptic Devices This type of haptic devices can be attached to parts of the body so the user has a larger work volume. An example of these devices is shown in Fig. (7), which is a handcentric device for animating hand movements and controlling virtual objects through hands. This type of devices is suitable when the virtual reality simulation requires the use of multiple contact points, freedom of motion, and data on texture and grasping. 3. VIRTUAL REALITY AND HAPTICS FOR DRUG DESIGN AND DISCOVERY: ISSUES AND CHALLENGES Virtual environments have found acceptance in drug design and discovery as they enable scientists to have a visual and tactile experience with a molecular system while manipulating the simulation in real-time. This overcomes the limitation of most traditional molecular modeling environments that provide static 3D models by immersing the scientist in a realistic and dynamic representation of a molecular system [18]. In order to provide a convincing virtual reality and haptics experience, the system requires a high degree of performance for computer hardware and software. An effective virtual reality display requires a minimum display rate of 10 frames per second although frame rates of 30 per second are preferred to avoid any flicker and image jumps [9].
438 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
Consequently, there is at most 0.1 second to compute a particular instance of a molecular simulation and to display the molecular model. On the other hand, haptic feedback requires much faster refresh rates than visual feedback as the human sense of touch is more sensitive than the sense of sight. For this reason, while an image needs to be updated at 30 frames per second, the sensation provided to the user needs to be updated 1000 times per second or more to provide a realistic tactile experience [24, 27]. The strict time constraints on visual and tactile response impose high demands on the performance of the hardware and software for simulating molecular systems. Any time delay between the user action and the corresponding update of the virtual object in the simulation leads to unrealistic visualization, unstable force feedback, and simulation sickness [30]. Limited time is available to compute changes in the molecules, display them, and provide the corresponding force feedback to the user. Several of the current virtual reality and haptic systems for drug design and discovery have represented the molecules as rigid bodies as this simplifies the calculation of forces. However, as molecular flexibility is critical for understanding the principles that govern in the binding of molecules, it is necessary to represent molecules as flexible bodies. Modeling flexible molecules in real-time is computationally expensive due to the exponential explosion in complexity as large numbers of variables are considered to represent the molecule’s conformations. This computational complexity is the main challenge in modeling flexible molecules in real-time to achieve a fully virtual immersive and interactive environment. In addition, to provide a haptic feedback, the forces acting on the molecules need to be calculated as the molecules change conformations. Modeling the physical changes and interactions between molecules require effective graphical molecular representations and efficient force computation to achieve visual and tactile feedback for a virtual environment [25]. 3.1. Graphical Representation of Molecules Given the refresh rate requirements for virtual reality and haptics, the time for rendering molecules and their motions is essential. The realistic physical modeling of molecules significantly contributes to the user’s sense of immersion and interactivity. Molecules are represented graphically in the form of stick model, ball-and-stick model, and CPK space-filling model as shown in Fig. (8a). The simulation program that updates the atom positions sends the new data for updating the models for visualization. Some programs use different number of polygons to represent the atoms at different levels of detail. When the molecules are far from the user, less polygons are used to speed the rendering process. Other systems use solvent-accessible surface (SAS) [31, 32] to make the rendering more efficient and to omit the display of interior atoms as shown in Fig. (8b) [33]. SAS representation can also be used for surface matching between the ligand and the receptor. The molecular representation should also facilitate the display and change of molecular conformations in real time. One of the main challenges in modeling molecules in real-time is the requirement for rapid identification and display of feasible molecular conformations. Most of the research work so far has modeled both the ligand and receptor molecules as rigid bodies in order to achieve the virtual reality and haptic refresh rates. However, as molecules are very “flexible” in nature and can adopt different conformations, new approaches are necessary to identify feasible conformations and to enable their display in real time.
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 439
(a) Ligand and receptor represented as spheres with van der Waals radii.
(b) Molecule represented by the solvent-accessible surface (SAS). Reprinted from [33], Journal of Molecular Graphics and Modelling, Vol. 17, Anderson, A. and Weng, Z., VRDD: Applying virtual reality visualization to protein docking and design, pp. 180-189, 1999, with permission from Elsevier. Fig. (8). Examples of molecular representations used in virtual reality and haptic applications for drug design and discovery.
The increase in power of computer hardware has led to the development of methods for modeling the flexibility of molecules. Some of the reviewed work has modeled the molecule as an articulated body or kinematic chain with torsional bonds. Based on the location of the torsional bonds, atoms are clustered into groups and their relationship is established to reduce the computation time and to decrease inaccuracies in update calculations during conformational search [34]. Wriggers and Birmanns [35] presented a new multiresolution visualization approach to facilitate the fitting of molecules into lowresolution density maps. A vector quantization method is used to replace the molecule with a number of codebook vectors in order to reduce the modeling of molecular data as
440 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
shown in Fig. (9). This molecular representation is also used to calculate the haptic force feedback in real time. Although there have been many great advances towards the display of molecules and their conformations in real time, further research is still needed to incorporate the receptor’s flexibility, which increases the problem complexity exponentially.
Fig. (9). Molecule represented by a set of 10 codebook vectors. Picture courtesy of Dr. Willy Wriggers, D.E. Shaw Research, and Dr. Stefan Birmanns, University of Texas Health Science Center at Houston [36].
3.2. Real-Time Force and Torque Computation Interactions between molecules are represented by the potential energy generated between them. This energy consists of the electrostatic potential, van der Waals potential, hydrogen bond potential, etc. As users manipulate a ligand molecule to dock it into a receptor in a virtual reality environment, the interactive molecular forces need to be calculated in real time to provide realistic force feedback through the haptic device. If the haptic device provides 6DOF force feedback, then it is also necessary to calculate the torque feedback. Both the modeling and force calculation need to satisfy the virtual reality and haptic refresh rates. However, molecular forces are computationally expensive to determine as many atom pairs are considered for the calculation. The total force acting on the ligand is determined by adding all the forces acting on each ligand atom by the receptor atoms.
(a) Potential force calculation Fig. (10). Haptic force calculation.
(b) Collision force calculation
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 441
Fig. (10a) shows that as a ligand approaches the receptor, it is subject to the intermo lecular force Fpotential acting from the receptor. This force needs to be provided to the user through the haptic device anytime the force is created between the ligand and the receptor. Once the user moves the ligand towards the receptor’s surface, a force that is high enough needs to be transmitted through the haptic device. This is necessary to provide a sense of touching and to avoid molecular interpenetration. This “collision” force Fcollision will depend on the atoms in contact and those in the neighborhood that will determine the magnitude and direction of the force vector as shown in Fig. (10b). In order to calculate the torque feedback, a “pivot point” Ppvt can be defined at the center of the ligand as shown in Fig. (11). The accumulated force is assumed to be ap plied through the pivot point of the molecule so the torque is around an axis that pas ses the pivot point. In Fig. (11), the torque 1 is induced at the pivot point by a collision point P1 when a force is applied on it.
Fig. (11). Haptic torque calculation.
Moreover, as the forces and torques are created, both the ligand and the receptor should be allowed to change conformations leading to a continuous change in force magnitude and direction as the user manipulates the ligand. The new forces and torques need to be updated satisfying the haptic refresh rates to provide a continuous and realistic feedback. Therefore, approaches have been proposed and used to speed the calculation of force and torque to achieve real-time visualization and haptic feedback. One approach to compute molecular forces in real time is the 3D grid method [37, 38]. This method consists of enclosing a molecule such as the protein in a 3D grid where each grid point stores the potential energy generated by surrounding protein atoms at that point. For each 3D grid, all the protein atoms located within a predefined range from the current grid center are used in the potential energy calculation. In this way, when a ligand is introduced in the vicinity of the protein, all the grids occupied by the ligand atoms are used in calculating the total interaction energy between the ligand and the protein as shown in Fig. (12). Atoms that fall outside the reasonable range 3D grid are assigned a large value to avoid incorrect selection of molecules that are far away from the receptor. The resultant energy is then converted into a force vector that is transmitted to the user through the haptic device.
442 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
Fig. (12). 3D grid method for real-time energy and force calculation.
Another approach is to use parallel computing to distribute and perform the calculations for the potential energy function among several processors. The function evaluation is performed in parallel using the master/slave model thus allowing the simulation of the physical and chemical properties of the molecular system in real time. For example, atom coordinates needed to calculate the potential energy can be stored in each processor so that calculations are carried out independently and in parallel. Results are then exchanged between processors to determine the molecular forces and provide them through the haptic device [9]. Real-time force and torque computation is an ongoing area of research in haptics as further factors need to be incorporated in the simulation. Scaling of force feedback based on the orientation of the ligand as it collides with the receptor, continuous force feedback as molecules change conformations, and consideration of receptor’s flexibility are some of the factors being considered. All these need to be calculated subject to the haptic update rate to provide realistic and continuous force and torque feedback to the user. 4. STATE-OF-THE-ART The first application of haptic devices for molecular docking was presented by OuhYoung et al. from the University of North Carolina at Chapel Hill [11, 14]. They designed and built a haptic device system called the “Docker” to simulate the interaction between a rigid drug molecule and a rigid receptor as shown in Fig. (13). The manipulator provides forces and torques acting on the drug to the user as he/she docks the drug into the receptor’s binding site. The interactive molecular energy and force are approximated by the electrostatic and van der Waals energies and are calculated in real-time using a 3D grid method [37, 38]. As the user moves the drug with the manipulator, the
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 443
user can feel the interactive molecular forces pulling the drug towards local energy minima. The researchers also carried out experiments with chemists to analyze the effectiveness of having both a visual and haptic feedback compared to having a visual feedback only. Experiments showed that chemists could identify the lowest energy configuration of the rigid ligand up to twice as fast with the haptic feedback compared to the visual display alone.
Fig. (13). A user guiding a drug molecule towards the receptor’s binding site with haptic force feedback. Reprinted with permission from [22], Taylor II, R. M., “Scientific applications of force feedback: molecular simulation and microscope control,” Course Notes of ACM SIGGRAPH’99, Los Angeles, CA, 1999.
A Virtual Biomolecular Environment (VIBE) for interactive molecular modeling was developed by researchers from the University of Illinois at Chicago, Florida State University and the Argonne National Laboratory [9]. VIBE uses parallel computing to simulate a molecular system that is displayed through a Cave Automatic Virtual Environment (CAVE) [39] while the data is exchanged through a high-speed network. The CAVE is a type of immersive room that surrounds users with stereoscopic images while the users wear shutter glasses and tracking devices so the images are displayed based on the users’ position and orientation. The user interacts with the molecular model and the simulation by manipulating the drug molecule around the receptor with a wand-like device. The simulation is based on molecular dynamics, which provide specific atomic interaction. In this way, a molecular scientist can visualize and control the molecular dynamics simulation in a real-time virtual environment and obtain immediate quantitative and qualitative information about a particular molecular system. Researchers at Argonne National Laboratory introduced a virtual reality system for molecular docking called Stalk [10]. The Stalk system combines parallel and distributed computing, genetic algorithms, high-speed networking, and virtual reality to simulate the interaction of a rigid-body ligand docking into a rigid receptor. Parallel computing and high-speed networking enable the rapid display of the molecular simulation while genetic algorithm is used to search for ligand conformations with minimum interaction ener-
444 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
gy. As shown in Fig. (14), the molecular simulation is displayed through a CAVE and the user is able to manipulate the molecules’ position to guide the energy minimization search while observing the docking process.
Fig. (14). A user docking a ligand into a receptor using Stalk. Reprinted with permission from [10], Levine, D.; Facello, M.; Hallstrom, P.; Reeder, G.; Walenz, B. and Stevens, F., “Stalk: An interactive system for virtual molecular docking,” Computing in Science and Engineering, 4(2), April-June 1997, pp. 55-65. © 1997 IEEE.
Researchers at the Fraunhofer Institute for Computer Graphics in Germany developed a virtual environment for interactive molecular dynamics simulation called RealMol [40]. RealMol can run on different virtual reality hardware such as CAVE, head mounted displays, and wall systems. The user wears a cyberglove to select a molecule and move it to a different location, which is used as the starting point for the molecular dynamics simulation. The simulation is run through a molecular dynamics program called NAMD [41] and any change in the energy of the molecular system triggers a sound to the user. The user can interactively start/stop the simulation by using his/her hand with the cyberglove to make modifications to the simulation while looking for the best binding site of the drug molecule. Anderson and Weng [33] developed an interactive molecular docking program called VRDD. Molecules are represented by solvent-accessible surface models and displayed through an Immersakesk™ workbench system that provides a multiviewer, semiimmersive virtual reality experience. The user utilizes a wand to manipulate a ligand around the receptor while a Monte Carlo algorithm is used to perform an energy local search around the user-specified ligand orientation. The binding energies are computed in real-time using a previously calibrated energy function. VRDD also provides auditory feedback to indicate to the user when atoms overlap and their interaction energy changes.
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 445
Researchers at the University of North Carolina at Chapel Hill developed the Steered Molecular Dynamics (SMD) system to interactively steer forces (or place restraints) in a running molecular dynamics simulation for observing particular molecular behaviors [42, 43]. The SMD system was applied to the extraction of small ligands from proteins and can be used with a virtual 3D environment. The virtual environment was provided through the Protein Interactive Theater (PIT) built at UNC Chapel Hill [44]. The PIT is a dual-screen, stereo display system for two users who are seated at a table as shown in Fig. (15). The users wear shutter glasses with a tracking sensor and observe the image displayed across the table. The user steers the dynamics simulation by selecting an atom with a pointer device and moving it. Thus, the PIT system simplifies the placement of the restraints and facilitates the visualization of the dynamics by multiple users.
Fig. (15). Protein Interactive Theater (PIT). Reprinted with permission from [44], Arthur, K.; Preston, T.; Taylor II, R. M.; Brooks Jr., F. P.; Whitton, M. C. and Wright, W. V., “Designing and building the PIT: a head-tracked stereo workspace for two users,” Technical Report TR98-015, University of North Carolina at Chapel Hill, 1998.
Stone et al. [45] implemented a system called Interactive Molecular Dynamics (IMD) that allows the manipulation of molecules in molecular dynamics simulations with real-time visual and force feedback. IMD consists of a haptic device, a molecular dynamics simulation program called NAMD [41], and a visualization program named VMD [46] as shown in Fig. (16). NAMD is a fast and scalable program that implements the Charmm force field for molecular dynamics. VMD provides the user graphic interface for displaying the molecular dynamic simulation and is connected to the haptic device, which provides the force feedback. The use of NAMD parallel molecular dynamics program and high-speed computers were key to overcome the challenge of incorporating force feedback to molecular dynamics simulations. In subsequent work, IMD was applied to study the membrane channel protein GlpF and the unbinding pathways of enzyme glycerol kinase [47]. The IMD simulation incorporates the researcher’s intuition to adjust molecular simulations in order to significantly accelerate processes that are otherwise too slow to model within reasonable time.
446 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
Fig. (16). Interactive Molecular Dynamics (IMD) framework. Reprinted from [45], Stone, J. E.; Gullingsrud, J. and Schulten, K., “A system for interactive molecular dynamics simulation,” ACM Symposium on Interactive 3D Graphics, 2001, pp. 191-194. http://doi.acm.org/10.1145/364338. 364398. © 2001 ACM, Inc. Reprinted by permission.
Bayazit et al. at Texas A&M University [48] presented a framework called OBPRM that combines a path planning approach with haptic devices to locate the ligand’s binding site on the protein. OBPRM uses a fully automated motion planner that generates ligand configurations close to the protein’s surface to create a roadmap for identifying potential binding sites in the protein. The generated roadmap indicates accessible potential binding sites to the ligand from an outside location. As the motion planner has difficulty sampling configurations between high potential energy areas, a haptic device is introduced to enable the user to explore the energy landscape and to identify key ligand configurations to aid the motion planner. Nagata et al. [49] presented a protein-ligand docking simulator using haptic devices. This system allows the user to feel the electrostatic forces of a protein in real time through a globular probe that has electrostatic charge. The probe is controlled by a haptic device and is used to search for regions with highly attractive forces that can indicate potential binding sites. Researchers from the National Institute of Standards and Technology proposed a new method for smoothing haptic forces between rigid molecules as they are docked or assembled [50]. Spatial constraints were introduced to provide stable forces to the user as molecules approach each other until they are in contact using the Lennard-Jones forcefield as shown in Fig. (17). The darker and lighter color arrows indicate the resultant force and torque vectors, respectively. A virtual wall is presented that impede atoms from penetrating beyond the wall providing the user with a smoother haptic force feedback.
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 447
Fig. (17). Resultant force and torque vectors experienced by a ligand as it approaches a receptor. Reprinted from [50], Computer-Aided Design, Vol. 36, No. 1, Lee, Y.-G. and Lyons, K. W., “Smoothing Haptic Interaction using Molecular Force Calculations,” pp. 75-90, 2004, with permission from Elsevier.
Birmanns and Wriggers from the University of Texas Health Science Center at Houston introduced a new real-time interactive fitting strategy to reduce the search complexity during molecular docking [51]. The proposed strategy applies virtual reality and haptic devices to assist users in orienting and positioning molecules in 3D space relative to each other. Fig. (18) shows the graphical user interface of the system, which Sculptor File
Docking
View
Data
Help
Sculptor Scene ICLL.pdb ICLL.situs
Graphics
Dynamics
Docking
Info
Graphics Mode All
Atoms:
LEU
Graphics Mode Color Mapping:
Add Transformation Translation X:
Translation Y:
Translation Z:
Label: Graphics Mode St All - CPK - Color: Resname ( HoH 0
0
Rotation X:
Rotation Y:
0
Rotation Z:
Add Apply
0
0
Apply
0
Close
Fig. (18). Graphical user interface for fitting molecules using SenSitus. Picture courtesy of Dr. Willy Wriggers, D.E. Shaw Research, and Dr. Stefan Birmanns, University of Texas Health Science Center at Houston [51].
448 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
was implemented in a virtual reality visualization program called SenSitus. The generated haptic force and torque feedback guide the user towards the best fit of molecules into low-resolution electron microscopy (EM) density maps. The force and torque calculation complexity is reduced by vector quantization to achieve the necessary refresh rates for realistic haptic perception. This work was extended to model molecular docking at various levels of resolution through a hybrid modeling technique [36]. The proposed hybrid model consists of a dynamic mesh simplification algorithm to balance the computer processor load between the haptic and visual rendering for achieving both haptic and visual refresh rates. The virtual reality environment is a self-built, low-cost back-projection system with projectors, mirrors, tracking system, glasses, and a 6DOF haptic device. Researchers at the Lawrence Berkeley National Laboratory developed a tool for interactive molecular docking called DockingShop [52]. DockingShop enables users to manipulate a ligand or protein to an approximated binding site of a receptor with real-time visual feedback, and side chain and backbone flexibility as shown in Fig. (19). The main objective is to develop a graphical interface for integrating human knowledge and intuition and accelerating the molecular docking process. DockingShop provides adjustable scoring functions that indicate the quality of the molecular configuration and guide the docking process. The resultant ligand-protein or protein-protein complex can be input into a more detailed molecular docking algorithm or optimization process for further refinement. DockingShop File
Server
Edit
View
Modes
Tools
Windows
Help
LEU(5)
Psi: 0.373863 Phi: -1.726796
pi
0
PHI LEU
-pi
Energy: 64775.9
Residue index P
64775.9
5
Residue name
Dihedral angles
Rad/Deg
Energy
U_Dir
Clamp
Draw atom names
Visualize atom collisions 0
pi
Draw hydrogen bond sites
cages
Fig. (19). Graphical user interface for DockingShop. Reprinted with permission from [52], Lu, T.C.; Max, N. L.; Ding, J.; Bethel, E. W. and Crivelli, S. N., “DockingShop: A tool for interactive molecular docking,” Lawrence Berkeley National Laboratory, Paper LBNL-58170, April 24, 2005.
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 449
(a) Graphical user interface and 5DOF haptic device for molecular docking.
(b) Ligand’s force and torque vectors provided by the haptic device as the ligand approaches the receptor. Fig. (20). Haptic system for molecular docking with flexible ligand and rigid receptor.
Lai-Yuen and Lee [53, 54] presented a system for docking flexible ligand molecules into a rigid receptor using a 5DOF haptic device as shown in Fig. (20a). As the user manipulates the ligand around the surface of the receptor, an energy minimization technique searches for a ligand conformation with lower energy while the user feels the force and torque feedback in real-time. Fig. (20b) shows the force (black arrow) and torque (blue arrow) vectors that are transmitted to the user through the haptic device as the user manipulates the ligand around the receptor. During the exploration, the user can apply a simulator called NanoDAS (Nano-scale Docking and Assembly Simulator) that explores the surroundings of the ligand to identify accessible paths to potential binding sites of the receptor [55]. NanoDAS uses potential field analysis and local search techniques to construct a search tree to “explore” and “exploit” the search space of the ligand towards various potential binding sites. Daunay et al. [56] proposed a new method for a six degrees of freedom (DOF) haptic device that allows real-time virtual interaction with molecular docking simulations as
450 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Susana K. Lai-Yuen
shown in Fig. (21). The method consists of applying wave variables to achieve stable simulation and force and torque feedback as the user manipulates a flexible ligand around the binding site of a flexible receptor. As molecular simulators used in pharmaceutical applications do not achieve real-time performance but are very accurate, the proposed method enables the user to feel the forces and torques during molecular docking using any molecular simulator. Subsequent work introduced an energy minimization process based on an energy equation with known analytic derivation [57]. The energy obtained from the minimization process is approached by a potential containing two distinct terms for the interaction force and torque. This provides forces and torques that look stable in a particular case regardless of the molecule’s displacement value.
Fig. (21). A ligand being manipulated around a receptor using virtual reality and a 6DOF haptic device. Reprinted with permission from [56], Daunay, B.; Micaelli, A. and Régnier, S., “6 DOF haptic feedback for molecular docking using wave variables,” IEEE International Conference on Robotics and Automation, Roma, Italy, April 10-14, 2007, pp. 840-845. © 2007 IEEE.
Subasi and Basdogan at Koç University in Turkey [58] presented a new humancomputer interaction approach for docking a rigid ligand molecule into a rigid receptor using an Active Haptic Workspace (AHW) visualization technique. The AHW enables the user to explore the receptor’s surface in high resolution with haptic force feedback allowing the visualization of complex 3D surfaces and overcoming the workspace limitations of commercially available haptic devices. Once the user identifies the true binding site with the haptic device, the final ligand configuration is calculated off-line using time-stepping molecular dynamics simulations and a proposed distance error minimization approach. The system creates a visual copy of the ligand to inform the user when a low-energy configuration is found. Virtual reality and haptics have also been used to teach students chemistry and molecular biology. Researchers at the University of Washington and The Scripps Research Institute used augmented reality on top of physical molecular models to facilitate the visualization of molecular structures of different sizes and complexity [59]. The approach uses haptic devices and a user display such as a head-mounted display to provide electrostatic force feedback and force field visualization on top of the physical molecular models as a new intuitive learning interface. The physical molecular models were fabricated through 3D printing, which is a type of rapid prototyping process that builds physical models in a layer-by-layer process [60]. The models are then integrated into the
A Review on Virtual Reality
Frontiers in Drug Design & Discovery, 2009, Vol. 4 451
augmented reality environment for teaching, research, and collaboration in molecular biology. 5. DISCUSSION This paper presented an overview of recent advances in virtual reality and haptics applications for drug design and discovery. Current research work has demonstrated that a virtual reality environment can be a powerful human-computer interface for visualizing and manipulating complex molecular systems in real time. This immersive and interactive computational tool can significantly increase researchers’ understanding of molecular docking by providing a 3-dimensional view of the system and enabling the researcher to feel the molecular forces acting on the ligand through haptic devices. Moreover, it can accelerate molecular simulations that are otherwise too slow to model within a reasonable time by facilitating the incorporation of the researcher’s knowledge and intuition into the simulation. However, the strict update requirements needed to provide a realistic visualization and tactile experience still limit the scope of the molecular simulations developed so far. It is expected that as computational power increases and becomes available, more thorough and accurate simulations can be achieved in real time. New approaches will be necessary for the realization of a complete and realistic immersive and interactive environment for drug design and discovery. Below are some of the areas of future work needed to achieve this goal: • Modeling the physical changes and interactions of molecules in real time is computationally expensive, particularly as both the size of the molecule and the number of interacting molecules increase. As both the ligand and receptor are allowed to change conformations, new algorithms are needed to speed the display of molecular conformations and to effectively identify collisions between the molecules for virtual reality and haptics applications. • The effect of the solvent in the molecular system is also required for a more complete modeling and for providing the adequate force feedback to the user. • New techniques for parallel computing will be useful for distributing the various processes involved in the display and force calculation of the molecular system in order to achieve the virtual reality and haptic update requirements. • Further experiments on the effect of the virtual reality environment on scientists are necessary as new approaches are developed. This can lead to changes in graphical user interfaces and to novel virtual reality and haptics equipment to make the environment more intuitive to the user. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8]
Kuhlman, J. Int. J. Clin. Pharm. Th., 1997, 35, 541. Ooms, F. Curr. Med. Chem., 2000, 7, 141-158. Burdea, G.; Coiffet, P. Virtual Reality Technology, 2nd ed., John Wiley & Sons: New York, 2003. Brooks, Jr., F.P. IEEE Comput. Graphics Appl., 1999, 19(6), 16-27. Ren, Y.; Lai-Yuen, S. K.; Lee, Y.-S. Virtual and Physical Prototyping, 2006, 1(1), 3-18. Salisbury, Jr., J.K. Commun. ACM, 1999, 42(8), 75-81. Sharma, G.; Mavroidis, C.; Ferreira, A. In Handbook of Theoretical and Computational Nanotechnology; M. Rieth and W. Schommers, Eds.; American Scientific Publishers, 2005. van Dam, A.; Laidlaw, D.H.; Simpson, R.M. Comput. Graph., 2002, 26, 535-555.
452 Frontiers in Drug Design & Discovery, 2009, Vol. 4 [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26]
[27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49]
Susana K. Lai-Yuen
Cruz-Neira, C.; Langley, R.; Bash, P.A. Comput. Chem., 1996, 20(4), 469-477. Levine, D.; Facello, M.; Hallstrom, P.; Reeder, G.; Walenz, B.; Stevens, F. Comput. Sci. Eng., 1997, 4(2), 55-65. Ouh-Young, M.; Pique, M.; Hughes, J.; Srinivasan, N.; Brooks, F. P. Proceedings of the IEEE International Conference on Robotics and Automation, 1988, 1824-1829. National Institute of Standards and Technology (NIST) http://www.nist.gov/public_affairs/ newsfromnist_smartgels.htm Baxter, C.A.; Murray, C.W.; Clark, D.E.; Westhead, D.R.; Eldridge, M.D. Proteins, 1998, 33, 367382. Ouh-Young, M. Technical report #90-004, Computer Science, University of North Carolina at Chapel Hill, 1990. Cruz-Neira, C. Course Notes #23, ACM SIGGRAPH, 1993. Sun Center of Excellence for Visual Genomics: http://www.visualgenomics.ca/ Immersion Corporation: http://immersion.com/ Ihlenfeldt, W.-D. J. Mol. Model., 1997, 3, 386-402. Jayaram, S.; Vance, J.; Gadh, R.; Jayaram, U.; Srinivasan, H. Transact. ASME, 2001, 1, 72-83. Lanzagorta, M.; Rosenblum, L.; Kuo, E.; Rosenberg, R. Proceedings of the Scientific Visualization Conference DAGSTUHL, 1997, 161-170. Virtual Research Systems, Inc. http://www.virtualresearch.com/products/vr1280.htm Taylor II, R.M. Course Notes of ACM SIGGRAPH’99, Los Angeles, CA, 1999. Barco, Inc.: http://www.barco.com/ Burdea, G.C. Invited paper in Proceedings of Computer Graphics International, Geneva, 18-24, 2000, 295-302. Burdea, G.C. Int. J. Des. Innovation Res., 2000, 2(1), 17-29. Fernando, T.; Murray, N.; Gautier, G.; Mihindu, S.; Loupos, K.; Gravez, P.; Hoffmann, H.; Blondelle, J.; Di Marca, S.; Fontana, M.; Cellary, W.; Aked, R.; Garcia-Berdones, C.; Reyes-Lecuona, A.; Marin, M.; Perez, J.; Fabbri, M.; Lichau, D. Technical Report, Intuition project, http://www.intuitioneunetwork.net. Fisch, A.; Mavroidis, C.; Bar-Cohen, Y.; Melli-Huber, J. Invited Chapter in Biologically-Inspired Intelligent Robots; Y. Bar-Cohen and C. Breazeal, Eds.; SPIE Press, 2003. Hayward, V.; Astley, O.R.; Cruz-Hernandez, M.; Grant, D.; Robles-De-La-Torre, G. Sensor Review, 2004, 24(1), 16-29. SensAble Technologies, Inc.: http://sensable.com/ Laycock, S.D.; Day, A.M. Comput. Graph. Forum, 2007, 26(1), 50-65. Connolly, M. L. Science, 1983, 221, 709-713. Sanner, M. F.; Olson, A. J.; Spehner, J. C. Biopolymers, 1996, 38, 305-320. Anderson, A.; Weng, Z. J. Mol. Graphics Modell., 1999, 17, 180-189. Zhang, M.; Kavraki, L.E. J. Chem. Inf. Comput. Sci., 2002, 42, 64-70. Wriggers, W.; Birmanns, S. J. Struct. Biol., 2001, 133, 193-202. Birmanns, S.; Boltes, M.; Zilken, H.; Wriggers, W. Proceedings of the IEEE International Conference on Mechatronics and Robotics, 2004, 4, 1351-1356. Meng, E. C.; Shoichet, B. K.; Kuntz, I. D. J. Comput. Chem., 1992, 13(4), 505-524. Pattabiraman, N.; Levitt, M.; Ferrin, T. E.; Langridge, R. J. Comput. Chem., 1985, 6, 432-436. Cruz-Neira, C.; Sandin, D.; DeFanti, T. Proceedings of SIGGRAPH, 1993, 135-142. Ai, Z.; Fröhlich, T. EUROGRAPHICS, 1998, 17(3), 267-273. Nelson, M.; Humphrey, W.; Gursoy, A.; Dalke, A.; Kalé, L.; Skeel, R. D.; Schulten, K. Int. J. Supercomput. App., 1996, 10, 251-268. Leech, J.; Prins, J. F.; Hermans, J. IEEE Comput. Sci. Eng., 1996, 38-45. Prins, J. F.; Hermans, J.; Mann, G.; Nyland, L. S.; Simons, M. Future Gener. Comp. Sy., 1999, 485495. Arthur, K.; Preston, T.; Taylor II, R. M.; Brooks Jr., F. P.; Whitton, M. C.; Wright, W. V. Technical Report TR98-015, University of North Carolina at Chapel Hill, 1998. Stone, J. E.; Gullingsrud, J.; Schulten, K. ACM Symposium on Interactive 3D Graphics, 2001, 191194. http://doi.acm.org/10.1145/364338.364398. Humphrey, W.; Dalke, A.; Schulten, K. J. Mol. Graphics, 1996, 14, 33-38. Grayson, P.; Tajkhorshid, E.; Schulten, K. Biophys. J., 2003, 85, 36-48. Bayazit, O.B.; Song, G.; Amato, N.M. Proceedings of the IEEE International Conference on Robotics and Automation, Seoul, Korea, May 21-26, 2001, 954-959. Nagata, H.; Mizushima, H.; Tanaka, H. Bioinformatics, 2002, 18(1), 140-146.
A Review on Virtual Reality [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60]
Frontiers in Drug Design & Discovery, 2009, Vol. 4 453
Lee, Y.-G.; Lyons, K. W. Comput. Aided Des., 2004, 36(1), 75-90. Birmanns, S.; Wriggers, W. J. Struct. Biol., 2003, 144, 123-131. Lu, T.C.; Max, N.L.; Ding, J.; Bethel, E.W.; Crivelli, S.N. Lawrence Berkeley National Laboratory, Paper LBNL-58170, 2005. Lai-Yuen, S.K.; Lee, Y.S. Comput. Aided Des. Appl., 2006, 3(6), 701-709. Lai-Yuen, S.K.; Lee, Y.S. Proceedings of the 14th Symposium Haptic Interfaces for Virtual Environment and Teleoperator Systems, Alexandria, VA, 2006, 34. Lai-Yuen, S.K.; Lee, Y.-S. Transactions NAMRI/SME, 2006, Vol. 34, 357-364. Daunay, B.; Micaelli, A.; Régnier, S. IEEE International Conference on Robotics and Automation, Roma, Italy, 2007, 840-845. Daunay, B.; Micaelli, A.; Régnier, S. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, San Diego, CA, October 29-November 2, 2007, 2704-2709. Subasi, E.; Basdogan, C. Presence-Teleop. Virt., MIT Press, 2008, 17(1), 73-90. Sankaranarayanan, G.; Weghorst, S.; Sanner, M.; Gillet, A.; Olson, A. Proceedings of the 11th Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems, 2003, 363-366. Gillet, A.; Sanner, M.; Stoffler, D.; Goodsell, D.; Olson, A. IEEE Visualization, 2004, 235-241.
454 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Contributors
Contributors Alamo, C.
Department of Pharmacology, Faculty University of Alcalá, Madrid, Spain
of
Medicine,
Axarli, I.
Laboratory of Enzyme Technology, Department of Agricultural Biotechnology, Agricultural University of Athens, 75 Iera Odos Street, GR-11855-Athens, Greece
Bause, A.S.
University of Arizona, College of Pharmacy, Arizona Cancer Center, Tucson, AZ, USA
Bramuglia, G.F.
Cátedra de Farmacología, Facultad de Farmacia y Bioquímica, Universidad de Buenos Aires, Junín 956, (C1113AAD) Buenos Aires, Argentina
Cal, K.
Department of Pharmaceutical Technology, Medical University, of Gdansk, Hallera 107, 80-416 Gdansk, Poland
Campillo, N.E.
Instituto de Química Médica (CSIC), Juan de la Cierva, nº 3, 28006-Madrid, Spain
Charcosset, C.
Laboratoire d’Automatique et de Génie des Procédés, Université de Lyon, UMR CNRS 5007, ESCPE-Lyon, 43 Bd du 11 Novembre 1918, 69 622 Villeurbanne Cedex, France
Coluzzi, F.
I.C.O.T. – Polo Pontino, Dept. Anaesthesiology, Intensive Care Medicine and Pain Therapy, University of Rome “La Sapienza”, Rome, Italy
Escobar-Chávez, J.J.
División de Estudios de Posgrado (Tecnología Farmacéutica), Facultad de Estudios Superiores CuautitlánUniversidad Nacional Autónoma de México, Cuautitlán Izcalli, Estado de México, México 54740 AND Departamento de Sistemas Biológicos, Universidad Autónoma Metropolitana-Xochimilco, Calzada del Hueso 1100, Colonia Villa Quietud, México D.F. 04960, México
Fessi, H.
Laboratoire d'Automatique et de Génie des Procédés, Université de Lyon, UMR CNRS 5007, ESCPE-Lyon, 43 Bd du 11 Novembre 1918, 69622 Villeurbanne Cedex, France
Fischer, R.
Medice chem.-phar., Fabrik Pütter GmbH & Co.KG, Iserlohn, Germany
Contributors
Frontiers in Drug Design & Discovery, 2009, Vol. 4 455
Fogel, G.B.
Natural Selection, Inc., 9330 Scranton Rd., San Diego, CA 92121, USA
Funatsu, K.
Kamakura Research Laboratories, Chugai Pharmaceutical Co., LTD, 200 Kajiwara, Kamakura, Kanagawa, 247-8530, Japan
Ganem-Quintanar, A.
División de Estudios de Posgrado (Tecnología Farmacéutica), Facultad de Estudios Superiores CuautitlánUniversidad Nacional Autónoma de México, Cuautitlán Izcalli, Estado de México, México 54740
García-García, P.
Department of Pharmacology, Faculty University of Alcalá, Madrid, Spain
Hasegawa, K.
Department of Chemical System Engineering, The University of Tokyo, Hongo 7-3-1, Bunkyo-ku, Tokyo, 1138656, Japan
Hecht, D.
Southwestern College, 900 Otay Lakes Rd., Chula Vista, CA 91910, USA
Höcht, C.
Cátedra de Farmacología, Instituto de Fisiopatología y Bioquímica Clínica, Facultad de Farmacia y Bioquímica, Universidad de Buenos Aires, Junín 956, (C1113AAD) Buenos Aires, Argentina
Hsieh, Y.
Department of Drug Metabolism and Pharmacokinetics, Schering-Plough Research Institute, 2015 Galloping Hill Road, K-15-3700, Kenilworth, NJ 07033, USA
Ishiguro, H.
Kyoto Univercity Hospital, Translational Research Center, Department of Clinical Trial Management / Outpatient Oncology Unit, 54 Shogoinkawaharara-cho, Sakyo-Ku, Kyoto-City, 606-8507, Japan
Khan, M.T.H.
Department of Pharmacology, Institute of Medical Biology, Faculty of Medicine, University of Tromsø, 9037 Tromsø, Norway
Khanna, I.
Reddy US Therapeutics Inc., 3065 Northwoods Circle, Norcross GA 30071, USA
Labrou, N.E.
Laboratory of Enzyme Technology, Department of Agricultural Biotechnology, Agricultural University of Athens, 75 Iera Odos Street, GR-11855-Athens, Greece
of
Medicine,
456 Frontiers in Drug Design & Discovery, 2009, Vol. 4
Contributors
Lai-Yuen, S.K.
Department of Industrial & Management Systems Engineering, University of South Florida, Tampa, FL 33620-5350, USA
Lamore, S.D.
University of Arizona, College of Pharmacy, Arizona Cancer Center, Tucson, AZ, USA
López-Cervantes, M.
División de Estudios de Posgrado (Tecnología Farmacéutica), Facultad de Estudios Superiores CuautitlánUniversidad Nacional Autónoma de México, Cuautitlán Izcalli, Estado de México, México 54740
López-Muñoz, F.
Department of Pharmacology, Faculty University of Alcalá, Madrid, Spain
Mattia, C.
I.C.O.T. – Polo Pontino, Dept. Anaesthesiology, Intensive Care Medicine and Pain Therapy, University of Rome “La Sapienza”, Rome, Italy
Mayer, M.
Cátedra de Farmacología, Instituto de Fisiopatología y Bioquímica Clínica, Facultad de Farmacia y Bioquímica, Universidad de Buenos Aires, Junín 956, (C1113AAD) Buenos Aires, Argentina
Melgoza-Contreras, L.M.
Departamento de Sistemas Biológicos, Universidad Autónoma Metropolitana-Xochimilco, Calzada del Hueso 1100, Colonia Villa Quietud, México D.F. 04960, México
Molina, J.D.
Acute Inpatients Unit, Dr. Lafora Psychiatric Hospital, Madrid, Spain
Opezzo, J.A.W.
Cátedra de Farmacología, Instituto de Fisiopatología y Bioquímica Clínica, Facultad de Farmacia y Bioquímica, Universidad de Buenos Aires, Junín 956, (C1113AAD) Buenos Aires, Argentina
Páez, J.A.
Instituto de Química Médica (CSIC), Juan de la Cierva, 3, 28006-Madrid-Spain
Pillarisetti, S.
Reddy US Therapeutics Inc., 3065 Northwoods Circle, Norcross GA 30071, USA
Platis, D.
Laboratory of Enzyme Technology, Department of Agricultural Biotechnology, Agricultural University of Athens, 75 Iera Odos Street, GR-11855-Athens, Greece
of
Medicine,
Contributors
Frontiers in Drug Design & Discovery, 2009, Vol. 4 457
Quintanar-Guerrero, D.
División de Estudios de Posgrado (Tecnología Farmacéutica), Facultad de Estudios Superiores CuautitlánUniversidad Nacional Autónoma de México, Cuautitlán Izcalli, Estado de México, México 54740
Skopelitou, K.
Laboratory of Enzyme Technology, Department of Agricultural Biotechnology, Agricultural University of Athens, 75 Iera Odos Street, GR-11855-Athens, Greece
Taira, C.A.
Cátedra de Farmacología, Instituto de Fisiopatología y Bioquímica Clínica, Facultad de Farmacia y Bioquímica, Universidad de Buenos Aires, Junín 956, (C1113AAD) Buenos Aires, Argentina
Toi, M.
Kyoto Univercity Hospital, Breast Surgery Department, 54 Shogoinkawaharara-cho, Sakyo-ku, Kyoto-city, 606-8507, Japan
Wondrak, G.T.
University of Arizona, College of Pharmacy, Arizona Cancer Center, Tucson, AZ, USA
Yano, I.
Kyoto Univercity Hospital, Department of Pharamcy, 54 Shogoinkawaharara-cho, Sakyo-ku, Kyoto-city, 606-8507, Japan