- Email: [email protected]
Overview of strategies for addressing BRIs in drug discovery: Impact on optimization and design
Chemico-Biological Interactions 192 (2011) 56–59
Contents lists available at ScienceDirect
Chemico-Biological Interactions journal homepage: www.elsevier.com/locate/chembioint
Overview of strategies for addressing BRIs in drug discovery: Impact on optimization and design W. Griffith Humphreys ∗ Department of Biotransformation, Bristol-Myers Squibb Research & Development, P.O. Box 4000, Princeton, NJ 08543, United States
a r t i c l e
i n f o
Article history: Available online 13 January 2011 Keywords: Cytochrome P450 (CYP) Drug design Reactive metabolites
a b s t r a c t The sensitive and specific detection of adducts derived from reactive intermediates during discovery metabolite profiling has been made feasible by advances in LC–MS/MS instrumentation. Many companies employ screens with nucleophilic trapping agents as a routine part of early screening efforts. Although certainly not as straightforward as initial adduct detection, the positives in the profiling experiment can be followed-up with determination of exact adduct structure. This information feeds naturally into drug design efforts as the structural motifs responsible for reactive metabolite formation can be altered to reduce the property. While the process of generation of reactive metabolite data has become more straightforward, the conversion of that data into an optimization paradigm remains challenging. Recent studies have shown a very loose correlation between extent of reactive metabolite formation and observed toxicity, so setting stringent criteria likely leads to discarding compounds that would not have problems. On the other hand, the central role of reactive metabolites in most accepted mechanisms of drug-induced toxicity points to the fact that there is value in minimizing the property. Decision making based on information on reactive metabolite formation remains a difficult process in all phases of drug discovery and development. Decisions on compounds in discovery can be made based on a fixed threshold value or relative to a reference point within a chemical series, but should be made with a firm understanding of the limitation of the data. © 2011 Published by Elsevier Ireland Ltd.
1. Introduction The metabolism of drugs to reactive intermediates followed by covalent binding to cellular components is generally considered to be the basis for the acute or idiosyncratic toxicities caused by some drugs [1–4]. It is also thought to be related to acute or long term toxicity seen in animal testing protocols [5]. The testing of new drug candidates for their potential to form reactive metabolites and the challenges associated with data interpretation from those experiments has been reviewed [5–8]. Many pharmaceutical companies examine new drug candidates for the potential to form reactive metabolites and if present make attempts to design the property out through targeted structural modification [9,10]. Reactive intermediates are most commonly thought to arise through the generation of high energy intermediates during the oxidation of drugs by CYP or other enzymes [11–13]. Examples of these inter-
Abbreviations: CYP, cytochrome P450; DIT, drug-induced toxicity; GSH, reduced glutathione; LC–MS/MS, liquid chromatography–tandem mass spectrometry; PCB, protein covalent binding. ∗ Corresponding author. Tel.: +1 609 252 3636; fax: +1 609 252 6802. E-mail address: [email protected] 0009-2797/$ – see front matter © 2011 Published by Elsevier Ireland Ltd. doi:10.1016/j.cbi.2011.01.005
mediates are epoxides, oxirenes, arene oxides and quinoid species. Reactive esters formed by the conjugation of carboxylic acids with glucuronic acid or acyl coenzyme A are also thought to be a source of reactive metabolites. The mechanisms by which reactive intermediates produce observed toxicity remains controversial. The proposed downstream impacts of reactive metabolite formation are generally proposed as: (1) damage through generalized oxidative or inflammatory stress [14,15] or (2) damage through alteration of specific protein function or formation of specific protein adducts that lead to neoantigen formation [16]. The two general mechanisms may also work in combination [17]. Multiple recent publications have linked genetic polymorphisms in genes for immune system components to DIT [18]. These associations have been seen with abacavir [19], flucloxacillin [20], ximelgatran [21] and even acetaminophen [22] and underscore the fact that while covalent binding may serve as an initiating event for idiosyncratic toxicity, the downstream responses are very complicated and difficult to predict based on early steps in the process. This manuscript will give an overview of how data can be generated and strategies for using that data as part of optimization and design efforts in drug discovery.
W.G. Humphreys / Chemico-Biological Interactions 192 (2011) 56–59
1.1. Why study the formation of reactive metabolites? The concept that reactive metabolites are the basis for many drug induced toxicities has been built over many years of research on drugs that demonstrated some type of toxicity, either in laboratory animals or in human. The positive link of compounds that caused toxicity also being compounds that formed reactive intermediates led to the conclusion that minimization of the formation of reactive metabolites would also minimize the risk of toxicity. While a conclusion that reactive metabolites are the basis for most cases of DIT still seems valid [3,4], the need to reduce the concept to practice has resulted in several recent publications that have described the PCB properties of drugs known to cause clinical drug induced toxicity and have importantly included data on drugs that are not associated with clinical toxicity. This type of data on “safe drugs” has not been available previously and is essential in judging how to use information from reactive metabolite studies. The results from the three major reports from the Pfizer, Daiichi and Dianippon groups were similar. They all conclude that: (1) some compounds not associated with DIT do form measurable amounts of reactive intermediate, (2) there is a weak correlation between reactive intermediates and whether a compound is considered generally safe or associated with DIT, (3) there are multiple factors that complicate the simple extrapolation of protein covalent binding/GSH adduct data to a prediction of toxicology liabilities and (4) there is a large “grey zone” between positives and negatives [23–27]. A study with a broad range of positive and negative compounds that measured thiol adduct levels after metabolic activation also reached similar conclusions [28] (Fig. 1, example of the results illustrating the correlation between total adduct burden and DIT potential [28]). Although, the conclusions of these studies certainly make the use of reactive metabolite data in a prospective manner more challenging, they do not alter the fact that central to all hypotheses regarding DIT is the initial formation of reactive species. An important consideration in the discovery arena is that while the studies listed above did measure the reactive metabolite potential of “positives”, i.e., drugs that caused DIT, the dynamic range of the response was not as great as might be seen in a typical discovery setting. This is possibly because the drugs selected as positives had all progressed through clinical development with associated safety studies in animals and man. Drug candidates in the discovery setting can be much more proficient at generating reactive species in in vitro experiments and the extent of the formation can easily exceed 50% of the total drug [29]. The most proficient true positives studied in the in vitro experiments described above produced
57
PCB values equivalent to conversion of ca. 10–20% of total drug, while the majority of compounds studied gave values in the range of 1–5% of total drug converted to trapped reactive intermediate. Whether there is any better correlation between observed toxicity and compounds predicted to produce a very high flux of reactive metabolites is unknown. Perhaps compounds that form very high levels of reactive intermediates in vitro are underrepresented in the true positive compound libraries because these types of compound fail during development. Failure in long term toxicology studies is another potential risk in advancing a compound that forms reactive metabolites. Whether in a rodent carcinogenicity bioassay or a multiple dose study in rodent or higher species, many toxicity-related failures are likely linked to downstream events subsequent to the formation reactive metabolites [5]. The prospective nature of reactive metabolite data likely has many limitations in the prediction of toxicology failures similar to those pointed out above for the prediction of idiosyncratic toxicities. Findings in these exploratory toxicity studies are often complicated by uncertainties around the target pharmacology and whether the observed toxicity could be somehow related to the pharmacological target. In these cases it may be beneficial to try to de-link the toxicity from the target by demonstrating that the toxicity is or is not metabolism-related and thus likely compound specific. 1.2. Methods to study the formation of reactive metabolites The study of the interaction of drugs with cellular components generally involves either measurement of covalent binding after reaction with nucleophilic sites on cellular macromolecules or studies with small molecule nucleophiles [7]. Newer assays attempt to use generic endpoint readouts to measure the impact of reactive metabolite formation [30]. For some classes of reactive metabolites such as acyl glucuronides, the reactive species can be synthesized and studied to determine their intrinsic reactivity and compared directly to historical reference compounds [31]. Finally, mechanistic studies seek to modulate formation of reactive species to attempt to alter observed toxicology. 1.3. Reactive metabolite studies – in vitro Trapping experiments are most often done with unlabeled drug and glutathione with detection by LC–MS/MS. Because unlabeled drug is used, these experiments can be done earlier in the discovery cycle than the covalent binding experiments [10,32]. The 1000
Estimated Total Daily Burden of Reactive Metabolites (mg)
% Thiol Adduct Formation
(a)
0.2%
LOQ
(b) 100
10
1
0.1
0.01
Non-DIT
DIT
Non-DIT
DIT
Fig. 1. Plot of % dansylGSH adduct formation and estimated total daily burden for drugs known to cause DIT (circles) or drugs not associated with DIT (triangles). Plot A shows adduct levels for all drugs assayed (50 total) including those below lowest limit of quantitation (LOQ). Plot B shows drugs where adduct levels could be measured, but includes an estimation of total body burden of adduct for a particular drug after administration of the recommended dose. A horizontal dotted line is plotted at 0.2% adduct level in panel a, and another is plotted at the 1 mg level in panel b to illustrate the best discrimination point between groups (with permission from [28]).
58
W.G. Humphreys / Chemico-Biological Interactions 192 (2011) 56–59
detection of glutathione adducts is aided by the characteristic fragmentation pattern of glutathione. Other analytical strategies have been developed to provide qualitative and/or quantitative determination of reactive metabolite formation at the screening stage [33–37]. These assays provide only qualitative information or utilize a non-physiological trapping agent and are thus somewhat limited in utility. The approach most often employed to circumvent this problem is through the use of radiolabeled trapping agents, typically tritiated glutathione. Protein covalent binding studies require radiolabeled drug and are thus usually carried out late in the discovery phase or in early development. A typical experiment is done either with labeled drug in microsomes or by administering the labeled drug to rodents. Both types of studies involve isolation of cellular proteins through precipitation, followed by extensive washing steps to remove noncovalently bound drug. Residual radioactivity bound to proteins is then determined by scintillation counting. The use of this type of data has received much recent attention as a potential predictor of toxicity, especially of idiosyncratic toxicities, although this remains a controversial subject. As described above, multiple recent publications have determined the PCB properties of drugs known to cause clinical DIT and have importantly included data on drugs that are not associated with clinical toxicity. This type of data on “safe drugs” has not been available previously and is essential in judging how to use information from reactive metabolite studies and is further discussed below. Many additional types of screens for reactive metabolite generation have been suggested in the literature but have not gained wide acceptance. Most notably are cell-based screens that utilize toxicity readouts in metabolically competent cells [38] or alternatively measure activation of cellular stress signals [30]. 1.4. Reactive metabolite studies – in vivo Questions regarding the formation of metabolites that covalently bind proteins can be addressed in vivo with experiments similar in nature to the experiments described in the in vitro section. Animal studies are often focused on the liver proteins as a target for drug binding, while human studies are usually focused on blood components due to obvious ethical limitations. Baillie and colleagues have proposed the use of extent of covalent modification as a measure of risk of adverse events for new chemical entities. In this proposal, a level of less than 50 pmol compound bound/mg microsomal protein has been put forth as a level of binding that would provide a safety margin (approximately 20-fold) over compounds that have shown liver toxicities in the clinic and were thought to produce toxicity through formation of reactive intermediates [39]. The determination of whether an observed toxicology is associated with reactive metabolites can often be an important question in the interpretation of toxicology results. The experimental approaches utilized often revolve around the modulation of metabolism thought to be involved in bioactivation. This can be done with a number of tools including chemical inhibitors or inducers, knock-out or humanized mouse models, etc. Results from the studies could potentially yield data that would feed into drug design efforts. 1.5. Utilization of data The information generated from the screening approaches outlined above can be utilized in a number of ways: (1) after determination of the formation of adduct and/or quantitation of adduct level, the adduct structure can be characterized and lead to medicinal chemistry approaches to design new analogs with the goal of limiting the amount of adduct formed. (2) As a trigger to perform
more advanced covalent binding studies. (3) As a trigger to do toxicology studies in a cell-based assay or in animals. The first way of utilizing data outlined above is directed solely at improving a chemotype through designing out the liability while the second and third make an attempt to contextualize the liability for the compound in question. There are multiple literature examples of the use of reactive metabolite data in aiding in the design of drug candidates [10]. These studies follow a minimization paradigm and have been successful in many cases in finding structure based solutions that greatly reduce the reactive metabolite formation of compounds from a given chemotype. Drug design efforts that seek to minimize the property of reactive metabolite formation naturally lead to questions such as “how low is low enough?”, “is there a threshold limit that needs to be achieved?”, etc. There are several strategies that can be employed to aid in decision making: (1) set a threshold for acceptance, this approach would include setting the threshold at the detection limits of the assay, in other words a zero tolerance policy, (2) minimization of the property relative to a set of standards (should include known positives and negatives), (3) minimization relative to a local reference standard, i.e., a compound from the chemotype being examined. All of these approaches, except for the zero tolerance approach, require some type of quantitation of reactive metabolite formation. While trying to put reactive metabolite data into perspective it is important to understand the limitations. As mentioned above, the recent studies examining positives and negatives in PCB and GSH trapping experiments found a very loose correlation of reactive metabolite formation and toxicity endpoint. All studies conclude that it is important to consider the total adduct burden rather than simply the level of adduct formed in the in vitro incubation, as the correlation of covalent binding measurement with DIT improved in all cases. The important parameters to consider for calculation of adduct burden are total dose and some measure of the fraction of drug predicted to proceed down the pathway to reactive metabolite formation (e.g. Total burden of adduct = dose × Fa × Fm × Fadduct, where Fa is fraction absorbed, Fm is fraction metabolized and Fadduct is the ratio of covalent adduct/total metabolite). A class of compounds where these considerations may be especially important are compounds that are slowly metabolized in vitro yet predicted based on animal studies to be completely cleared via metabolism in vivo. Although the experiments are very challenging in these cases, the results may be important as a simple determination of in vitro rate will underpredict that total burden of adduct seen in vivo. Another important conclusion from these studies was the fact that there were both false positives and false negatives found in each of the studies. The false positives are not that surprising in that it is very difficult to find highly metabolized compounds that are completely devoid of covalent binding. False negatives in the experiments can be explained by bioactivation by alternate pathways in some but not all cases.
2. Conclusion Even with the full consideration of all factors that may be involved in the prediction of toxicity based on reactive metabolite formation as discussed above, it must be understood that there are large gaps in the basic understanding of the science involved that add a great deal of uncertainty to the prediction process. While a paradigm of screening for evidence of reactive species and modifying structure based on the results is fairly straightforward to implement, it comes with a great deal of risk that the property being optimized is not a true liability. For this reason, the degree with which reactive metabolite information is used in drug design
W.G. Humphreys / Chemico-Biological Interactions 192 (2011) 56–59
and decision-making in early development must be carefully considered on a program by program and compound by compound basis. References [1] D.E. Amacher, Reactive intermediates and the pathogenesis of adverse drug reactions: the toxicology perspective, Current Drug Metabolism 7 (3) (2006) 219–229. [2] J.L. Walgren, M.D. Mitchell, D.C. Thompson, Role of metabolism in druginduced idiosyncratic hepatotoxicity, Critical Reviews in Toxicology 35 (4) (2005) 325–361. [3] F.P. Guengerich, Mechanisms of drug toxicity and relevance to pharmaceutical development, Drug Metabolism and Pharmacokinetics (2010), doi:10.2133/dmpk.DMPK-10-RV-062. [4] B.K. Park, H. Laverty, A. Srivastava, D.J. Antoine, D. Naisbitt, D.P. Williams, Drug bioactivation and protein adduct formation in the pathogenesis of drug-induced toxicity, Chemico-Biological Interactions (2010), doi:10.1016/j.cbi.2010.09.011. [5] F.P. Guengerich, J.S. MacDonald, Applying mechanisms of chemical toxicity to predict drug safety, Chemical Research in Toxicology 20 (3) (2007) 344–369. [6] J. Uetrecht, Screening for the potential of a drug candidate to cause idiosyncratic drug reactions, Drug Discovery Today 8 (18) (2003) 832–837. [7] T.A. Baillie, Metabolism and toxicity of drugs. Two decades of progress in industrial drug metabolism, Chemical Research in Toxicology 21 (1) (2008) 129–137. [8] A.S. Kalgutkar, M.T. Didiuk, Structural alerts, reactive metabolites, and protein covalent binding: how reliable are these attributes as predictors of drug toxicity? Chemistry and Biodiversity 6 (11) (2009) 2115–2137. [9] G.A Doss, T.A. Baillie, Addressing metabolic activation as an integral component of drug design, Drug Metabolism Reviews 38 (4) (2006) 641–649. [10] S. Kumar, K. Mitra, K. Kassahun, T.A. Baillie, Approaches for minimizing metabolic activation of new drug candidates in drug discovery, Handbook of Experimental Pharmacology 196 (2010) 511–544. [11] F.P. Guengerich, Generation of reactive intermediates, Journal of Biochemical and Molecular Toxicology 19 (3) (2005) 173–174. [12] F.P. Guengerich, Cytochrome p450 and chemical toxicology, Chemical Research in Toxicology 21 (1) (2008) 70–83. [13] A.S. Kalgutkar, I. Gardner, R.S. Obach, C.L. Shaffer, E. Callegari, K.R. Henne, A.E. Mutlib, D.K. Dalvie, J.S. Lee, Y. Nakai, J.P. O’Donnell, J. Boer, S.P. Harriman, A comprehensive listing of bioactivation pathways of organic functional groups, Current Drug Metabolism 6 (3) (2005) 161–225. [14] P.J. Shaw, P.E. Ganey, R.A. Roth, Idiosyncratic drug-induced liver injury and the role of inflammatory stress with an emphasis on an animal model of trovafloxacin hepatotoxicity, Toxicological Sciences 118 (1) (2010) 7–18. [15] D.P. Jones, J.J. Lemasters, D. Han, U.A. Boelsterli, N. Kaplowitz, Mechanisms of pathogenesis in drug hepatotoxicity putting the stress on mitochondria, Molecular Interventions 10 (2) (2010) 98–111. [16] D.H. Adams, C. Ju, S.K. Ramaiah, J. Uetrecht, H. Jaeschke, Mechanisms of immune-mediated liver injury, Toxicological Sciences 115 (2) (2010) 307–321. [17] B.K. Gunawan, N. Kaplowitz, Mechanisms of drug-induced liver disease, Clinics in Liver Disease 11 (3) (2007) 459–475. [18] A.K. Daly, C.P. Day, Genetic association studies in drug-induced liver injury, Seminars in Liver Disease 29 (4) (2009) 400–411. [19] A.R. Hughes, C.H. Brothers, M. Mosteller, W.R. Spreen, D.K. Burns, Genetic association studies to detect adverse drug reactions: abacavir hypersensitivity as an example, Pharmacogenomics 10 (2) (2009) 225–233. [20] A.K. Daly, P.T. Donaldson, P. Bhatnagar, Y. Shen, I. Pe’er, A. Floratos, M.J. Daly, D.B. Goldstein, S. John, M.R. Nelson, J. Graham, B.K. Park, J.F. Dillon, W. Bernal, H.J. Cordell, M. Pirmohamed, G.P. Aithal, C.P. Day, HLA-B*5701 genotype is a major determinant of drug-induced liver injury due to flucloxacillin, Nature Genetics 41 (7) (2009) 816–819. [21] A. Kindmark, A. Jawaid, C.G. Harbron, B.J. Barratt, O.F. Bengtsson, T.B. Andersson, S. Carlsson, K.E. Cederbrant, N.J. Gibson, M. Armstrong, M.E. LagerstromFermer, A. Dellsen, E.M. Brown, M. Thornton, C. Dukes, S.C. Jenkins, M.A. Firth, G.O. Harrod, T.H. Pinel, S.M. Billing-Clason, L.R. Cardon, R.E. March, Genomewide pharmacogenetic investigation of a hepatic adverse event without clinical signs of immunopathology suggests an underlying immune pathogenesis, The Pharmacogenomics Journal 8 (3) (2008) 186–195.
59
[22] A.H. Harrill, P.B. Watkins, S. Su, P.K. Ross, D.E. Harbourt, I.M. Stylianou, G.A. Boorman, M.W. Russo, R.S. Sackler, S.C. Harris, P.C. Smith, R. Tennant, M. Bogue, K. Paigen, C. Harris, T. Contractor, T. Wiltshire, I. Rusyn, D.W. Threadgill, Mouse population-guided resequencing reveals that variants in CD44 contribute to acetaminophen-induced liver injury in humans, Genome Research 19 (9) (2009) 1507–1515. [23] H. Takakusa, H. Masumoto, H. Yukinaga, C. Makino, S. Nakayama, O. Okazaki, K. Sudo, Covalent binding and tissue distribution/retention assessment of drugs associated with idiosyncratic drug toxicity, Drug Metabolism and Disposition 36 (9) (2008) 1770–1779. [24] R.S. Obach, A.S. Kalgutkar, J.R. Soglia, S.X. Zhao, Can in vitro metabolismdependent covalent binding data in liver microsomes distinguish hepatotoxic from nonhepatotoxic drugs? An analysis of 18 drugs with consideration of intrinsic clearance and daily dose, Chemical Research in Toxicology 21 (9) (2008) 1814–1822. [25] S. Nakayama, R. Atsumi, H. Takakusa, Y. Kobayashi, A. Kurihara, Y. Nagai, D. Nakai, O. Okazaki, A zone classification system for risk assessment of idiosyncratic drug toxicity using daily dose and covalent binding, Drug Metabolism and Disposition 37 (9) (2009) 1970–1977. [26] J.N. Bauman, J.M. Kelly, S. Tripathy, S.X. Zhao, W.W. Lam, A.S. Kalgutkar, R.S. Obach, Can in vitro metabolism-dependent covalent binding data distinguish hepatotoxic from nonhepatotoxic drugs? An analysis using human hepatocytes and liver S-9 fraction, Chemical Research in Toxicology 22 (2) (2009) 332–340. [27] T. Usui, M. Mise, T. Hashizume, M. Yabuki, S. Komuro, Evaluation of the potential for drug-induced liver injury based on in vitro covalent binding to human liver proteins, Drug Metabolism and Disposition 37 (12) (2009) 2383–2392. [28] J. Gan, Q. Ruan, B. He, M. Zhu, W.C. Shyu, W.G. Humphreys, In vitro screening of 50 highly prescribed drugs for thiol adduct formation–comparison of potential for drug-induced toxicity and extent of adduct formation, Chemical Research in Toxicology 22 (4) (2009) 690–698. [29] D.C. Evans, T.A. Baillie, Minimizing the potential for metabolic activation as an integral part of drug design, Current Opinion in Drug Discovery and Development 8 (1) (2005) 44–50. [30] S.O. Simmons, C.Y. Fan, R. Ramabhadran, Cellular stress response pathway system as a sentinel ensemble in toxicological screening, Toxicological Sciences 111 (2) (2009) 202–225. [31] R. Sawamura, N. Okudaira, K. Watanabe, T. Murai, Y. Kobayashi, M. Tachibana, T. Ohnuki, K. Masuda, H. Honma, A. Kurihara, O. Okazaki, Predictability of idiosyncratic drug toxicity risk for carboxylic acid-containing drugs based on the chemical stability of acyl glucuronide, Drug Metabolism and Disposition 38 (10) (2010) 1857–1864. [32] Y.Z. Shu, B.M. Johnson, T.J. Yang, Role of biotransformation studies in minimizing metabolism-related liabilities in drug discovery, The AAPS Journal 10 (1) (2008) 178–192. [33] C.M. Dieckhaus, C.L. Fernandez-Metzler, R. King, P.H. Krolikowski, T.A. Baillie, Negative ion tandem mass spectrometry for the detection of glutathione conjugates, Chemical Research in Toxicology 18 (4) (2005) 630–638. [34] Z. Yan, N. Maher, R. Torres, G.W. Caldwell, N. Huebert, Rapid detection and characterization of minor reactive metabolites using stable-isotope trapping in combination with tandem mass spectrometry, Rapid Communications in Mass Spectrometry 19 (22) (2005) 3322–3330. [35] A. Mutlib, W. Lam, J. Atherton, H. Chen, P. Galatsis, W. Stolle, Application of stable isotope labeled glutathione and rapid scanning mass spectrometers in detecting and characterizing reactive metabolites, Rapid Communications in Mass Spectrometry 19 (23) (2005) 3482–3492. [36] J. Gan, T.W. Harper, M.M. Hsueh, Q. Qu, W.G. Humphreys, Dansyl glutathione as a trapping agent for the quantitative estimation and identification of reactive metabolites, Chemical Research in Toxicology 18 (5) (2005) 896–903. [37] D. Argoti, L. Liang, A. Conteh, L. Chen, D. Bershas, C.P. Yu, P. Vouros, E. Yang, Cyanide trapping of minimum ion reactive intermediates followed by detection and structure identification using liquid chromatography–tandem mass spectrometry (LC–MS/MS), Chemical Research in Toxicology 18 (10) (2005) 1537–1544. [38] G. Tuschl, B. Lauer, S.O. Mueller, Primary hepatocytes as a model to analyze species-specific toxicity and drug metabolism, Expert Opinion on Drug Metabolism and Toxicology 4 (7) (2008) 855–870. [39] D.C. Evans, A.P. Watt, D.A. Nicoll-Griffith, T.A. Baillie, Drug–protein adducts: an industry perspective on minimizing the potential for drug bioactivation in drug discovery and development, Chemical Research in Toxicology 17 (1) (2004) 3–16.
Contents lists available at ScienceDirect
Chemico-Biological Interactions journal homepage: www.elsevier.com/locate/chembioint
Overview of strategies for addressing BRIs in drug discovery: Impact on optimization and design W. Griffith Humphreys ∗ Department of Biotransformation, Bristol-Myers Squibb Research & Development, P.O. Box 4000, Princeton, NJ 08543, United States
a r t i c l e
i n f o
Article history: Available online 13 January 2011 Keywords: Cytochrome P450 (CYP) Drug design Reactive metabolites
a b s t r a c t The sensitive and specific detection of adducts derived from reactive intermediates during discovery metabolite profiling has been made feasible by advances in LC–MS/MS instrumentation. Many companies employ screens with nucleophilic trapping agents as a routine part of early screening efforts. Although certainly not as straightforward as initial adduct detection, the positives in the profiling experiment can be followed-up with determination of exact adduct structure. This information feeds naturally into drug design efforts as the structural motifs responsible for reactive metabolite formation can be altered to reduce the property. While the process of generation of reactive metabolite data has become more straightforward, the conversion of that data into an optimization paradigm remains challenging. Recent studies have shown a very loose correlation between extent of reactive metabolite formation and observed toxicity, so setting stringent criteria likely leads to discarding compounds that would not have problems. On the other hand, the central role of reactive metabolites in most accepted mechanisms of drug-induced toxicity points to the fact that there is value in minimizing the property. Decision making based on information on reactive metabolite formation remains a difficult process in all phases of drug discovery and development. Decisions on compounds in discovery can be made based on a fixed threshold value or relative to a reference point within a chemical series, but should be made with a firm understanding of the limitation of the data. © 2011 Published by Elsevier Ireland Ltd.
1. Introduction The metabolism of drugs to reactive intermediates followed by covalent binding to cellular components is generally considered to be the basis for the acute or idiosyncratic toxicities caused by some drugs [1–4]. It is also thought to be related to acute or long term toxicity seen in animal testing protocols [5]. The testing of new drug candidates for their potential to form reactive metabolites and the challenges associated with data interpretation from those experiments has been reviewed [5–8]. Many pharmaceutical companies examine new drug candidates for the potential to form reactive metabolites and if present make attempts to design the property out through targeted structural modification [9,10]. Reactive intermediates are most commonly thought to arise through the generation of high energy intermediates during the oxidation of drugs by CYP or other enzymes [11–13]. Examples of these inter-
Abbreviations: CYP, cytochrome P450; DIT, drug-induced toxicity; GSH, reduced glutathione; LC–MS/MS, liquid chromatography–tandem mass spectrometry; PCB, protein covalent binding. ∗ Corresponding author. Tel.: +1 609 252 3636; fax: +1 609 252 6802. E-mail address: [email protected] 0009-2797/$ – see front matter © 2011 Published by Elsevier Ireland Ltd. doi:10.1016/j.cbi.2011.01.005
mediates are epoxides, oxirenes, arene oxides and quinoid species. Reactive esters formed by the conjugation of carboxylic acids with glucuronic acid or acyl coenzyme A are also thought to be a source of reactive metabolites. The mechanisms by which reactive intermediates produce observed toxicity remains controversial. The proposed downstream impacts of reactive metabolite formation are generally proposed as: (1) damage through generalized oxidative or inflammatory stress [14,15] or (2) damage through alteration of specific protein function or formation of specific protein adducts that lead to neoantigen formation [16]. The two general mechanisms may also work in combination [17]. Multiple recent publications have linked genetic polymorphisms in genes for immune system components to DIT [18]. These associations have been seen with abacavir [19], flucloxacillin [20], ximelgatran [21] and even acetaminophen [22] and underscore the fact that while covalent binding may serve as an initiating event for idiosyncratic toxicity, the downstream responses are very complicated and difficult to predict based on early steps in the process. This manuscript will give an overview of how data can be generated and strategies for using that data as part of optimization and design efforts in drug discovery.
W.G. Humphreys / Chemico-Biological Interactions 192 (2011) 56–59
1.1. Why study the formation of reactive metabolites? The concept that reactive metabolites are the basis for many drug induced toxicities has been built over many years of research on drugs that demonstrated some type of toxicity, either in laboratory animals or in human. The positive link of compounds that caused toxicity also being compounds that formed reactive intermediates led to the conclusion that minimization of the formation of reactive metabolites would also minimize the risk of toxicity. While a conclusion that reactive metabolites are the basis for most cases of DIT still seems valid [3,4], the need to reduce the concept to practice has resulted in several recent publications that have described the PCB properties of drugs known to cause clinical drug induced toxicity and have importantly included data on drugs that are not associated with clinical toxicity. This type of data on “safe drugs” has not been available previously and is essential in judging how to use information from reactive metabolite studies. The results from the three major reports from the Pfizer, Daiichi and Dianippon groups were similar. They all conclude that: (1) some compounds not associated with DIT do form measurable amounts of reactive intermediate, (2) there is a weak correlation between reactive intermediates and whether a compound is considered generally safe or associated with DIT, (3) there are multiple factors that complicate the simple extrapolation of protein covalent binding/GSH adduct data to a prediction of toxicology liabilities and (4) there is a large “grey zone” between positives and negatives [23–27]. A study with a broad range of positive and negative compounds that measured thiol adduct levels after metabolic activation also reached similar conclusions [28] (Fig. 1, example of the results illustrating the correlation between total adduct burden and DIT potential [28]). Although, the conclusions of these studies certainly make the use of reactive metabolite data in a prospective manner more challenging, they do not alter the fact that central to all hypotheses regarding DIT is the initial formation of reactive species. An important consideration in the discovery arena is that while the studies listed above did measure the reactive metabolite potential of “positives”, i.e., drugs that caused DIT, the dynamic range of the response was not as great as might be seen in a typical discovery setting. This is possibly because the drugs selected as positives had all progressed through clinical development with associated safety studies in animals and man. Drug candidates in the discovery setting can be much more proficient at generating reactive species in in vitro experiments and the extent of the formation can easily exceed 50% of the total drug [29]. The most proficient true positives studied in the in vitro experiments described above produced
57
PCB values equivalent to conversion of ca. 10–20% of total drug, while the majority of compounds studied gave values in the range of 1–5% of total drug converted to trapped reactive intermediate. Whether there is any better correlation between observed toxicity and compounds predicted to produce a very high flux of reactive metabolites is unknown. Perhaps compounds that form very high levels of reactive intermediates in vitro are underrepresented in the true positive compound libraries because these types of compound fail during development. Failure in long term toxicology studies is another potential risk in advancing a compound that forms reactive metabolites. Whether in a rodent carcinogenicity bioassay or a multiple dose study in rodent or higher species, many toxicity-related failures are likely linked to downstream events subsequent to the formation reactive metabolites [5]. The prospective nature of reactive metabolite data likely has many limitations in the prediction of toxicology failures similar to those pointed out above for the prediction of idiosyncratic toxicities. Findings in these exploratory toxicity studies are often complicated by uncertainties around the target pharmacology and whether the observed toxicity could be somehow related to the pharmacological target. In these cases it may be beneficial to try to de-link the toxicity from the target by demonstrating that the toxicity is or is not metabolism-related and thus likely compound specific. 1.2. Methods to study the formation of reactive metabolites The study of the interaction of drugs with cellular components generally involves either measurement of covalent binding after reaction with nucleophilic sites on cellular macromolecules or studies with small molecule nucleophiles [7]. Newer assays attempt to use generic endpoint readouts to measure the impact of reactive metabolite formation [30]. For some classes of reactive metabolites such as acyl glucuronides, the reactive species can be synthesized and studied to determine their intrinsic reactivity and compared directly to historical reference compounds [31]. Finally, mechanistic studies seek to modulate formation of reactive species to attempt to alter observed toxicology. 1.3. Reactive metabolite studies – in vitro Trapping experiments are most often done with unlabeled drug and glutathione with detection by LC–MS/MS. Because unlabeled drug is used, these experiments can be done earlier in the discovery cycle than the covalent binding experiments [10,32]. The 1000
Estimated Total Daily Burden of Reactive Metabolites (mg)
% Thiol Adduct Formation
(a)
0.2%
LOQ
(b) 100
10
1
0.1
0.01
Non-DIT
DIT
Non-DIT
DIT
Fig. 1. Plot of % dansylGSH adduct formation and estimated total daily burden for drugs known to cause DIT (circles) or drugs not associated with DIT (triangles). Plot A shows adduct levels for all drugs assayed (50 total) including those below lowest limit of quantitation (LOQ). Plot B shows drugs where adduct levels could be measured, but includes an estimation of total body burden of adduct for a particular drug after administration of the recommended dose. A horizontal dotted line is plotted at 0.2% adduct level in panel a, and another is plotted at the 1 mg level in panel b to illustrate the best discrimination point between groups (with permission from [28]).
58
W.G. Humphreys / Chemico-Biological Interactions 192 (2011) 56–59
detection of glutathione adducts is aided by the characteristic fragmentation pattern of glutathione. Other analytical strategies have been developed to provide qualitative and/or quantitative determination of reactive metabolite formation at the screening stage [33–37]. These assays provide only qualitative information or utilize a non-physiological trapping agent and are thus somewhat limited in utility. The approach most often employed to circumvent this problem is through the use of radiolabeled trapping agents, typically tritiated glutathione. Protein covalent binding studies require radiolabeled drug and are thus usually carried out late in the discovery phase or in early development. A typical experiment is done either with labeled drug in microsomes or by administering the labeled drug to rodents. Both types of studies involve isolation of cellular proteins through precipitation, followed by extensive washing steps to remove noncovalently bound drug. Residual radioactivity bound to proteins is then determined by scintillation counting. The use of this type of data has received much recent attention as a potential predictor of toxicity, especially of idiosyncratic toxicities, although this remains a controversial subject. As described above, multiple recent publications have determined the PCB properties of drugs known to cause clinical DIT and have importantly included data on drugs that are not associated with clinical toxicity. This type of data on “safe drugs” has not been available previously and is essential in judging how to use information from reactive metabolite studies and is further discussed below. Many additional types of screens for reactive metabolite generation have been suggested in the literature but have not gained wide acceptance. Most notably are cell-based screens that utilize toxicity readouts in metabolically competent cells [38] or alternatively measure activation of cellular stress signals [30]. 1.4. Reactive metabolite studies – in vivo Questions regarding the formation of metabolites that covalently bind proteins can be addressed in vivo with experiments similar in nature to the experiments described in the in vitro section. Animal studies are often focused on the liver proteins as a target for drug binding, while human studies are usually focused on blood components due to obvious ethical limitations. Baillie and colleagues have proposed the use of extent of covalent modification as a measure of risk of adverse events for new chemical entities. In this proposal, a level of less than 50 pmol compound bound/mg microsomal protein has been put forth as a level of binding that would provide a safety margin (approximately 20-fold) over compounds that have shown liver toxicities in the clinic and were thought to produce toxicity through formation of reactive intermediates [39]. The determination of whether an observed toxicology is associated with reactive metabolites can often be an important question in the interpretation of toxicology results. The experimental approaches utilized often revolve around the modulation of metabolism thought to be involved in bioactivation. This can be done with a number of tools including chemical inhibitors or inducers, knock-out or humanized mouse models, etc. Results from the studies could potentially yield data that would feed into drug design efforts. 1.5. Utilization of data The information generated from the screening approaches outlined above can be utilized in a number of ways: (1) after determination of the formation of adduct and/or quantitation of adduct level, the adduct structure can be characterized and lead to medicinal chemistry approaches to design new analogs with the goal of limiting the amount of adduct formed. (2) As a trigger to perform
more advanced covalent binding studies. (3) As a trigger to do toxicology studies in a cell-based assay or in animals. The first way of utilizing data outlined above is directed solely at improving a chemotype through designing out the liability while the second and third make an attempt to contextualize the liability for the compound in question. There are multiple literature examples of the use of reactive metabolite data in aiding in the design of drug candidates [10]. These studies follow a minimization paradigm and have been successful in many cases in finding structure based solutions that greatly reduce the reactive metabolite formation of compounds from a given chemotype. Drug design efforts that seek to minimize the property of reactive metabolite formation naturally lead to questions such as “how low is low enough?”, “is there a threshold limit that needs to be achieved?”, etc. There are several strategies that can be employed to aid in decision making: (1) set a threshold for acceptance, this approach would include setting the threshold at the detection limits of the assay, in other words a zero tolerance policy, (2) minimization of the property relative to a set of standards (should include known positives and negatives), (3) minimization relative to a local reference standard, i.e., a compound from the chemotype being examined. All of these approaches, except for the zero tolerance approach, require some type of quantitation of reactive metabolite formation. While trying to put reactive metabolite data into perspective it is important to understand the limitations. As mentioned above, the recent studies examining positives and negatives in PCB and GSH trapping experiments found a very loose correlation of reactive metabolite formation and toxicity endpoint. All studies conclude that it is important to consider the total adduct burden rather than simply the level of adduct formed in the in vitro incubation, as the correlation of covalent binding measurement with DIT improved in all cases. The important parameters to consider for calculation of adduct burden are total dose and some measure of the fraction of drug predicted to proceed down the pathway to reactive metabolite formation (e.g. Total burden of adduct = dose × Fa × Fm × Fadduct, where Fa is fraction absorbed, Fm is fraction metabolized and Fadduct is the ratio of covalent adduct/total metabolite). A class of compounds where these considerations may be especially important are compounds that are slowly metabolized in vitro yet predicted based on animal studies to be completely cleared via metabolism in vivo. Although the experiments are very challenging in these cases, the results may be important as a simple determination of in vitro rate will underpredict that total burden of adduct seen in vivo. Another important conclusion from these studies was the fact that there were both false positives and false negatives found in each of the studies. The false positives are not that surprising in that it is very difficult to find highly metabolized compounds that are completely devoid of covalent binding. False negatives in the experiments can be explained by bioactivation by alternate pathways in some but not all cases.
2. Conclusion Even with the full consideration of all factors that may be involved in the prediction of toxicity based on reactive metabolite formation as discussed above, it must be understood that there are large gaps in the basic understanding of the science involved that add a great deal of uncertainty to the prediction process. While a paradigm of screening for evidence of reactive species and modifying structure based on the results is fairly straightforward to implement, it comes with a great deal of risk that the property being optimized is not a true liability. For this reason, the degree with which reactive metabolite information is used in drug design
W.G. Humphreys / Chemico-Biological Interactions 192 (2011) 56–59
and decision-making in early development must be carefully considered on a program by program and compound by compound basis. References [1] D.E. Amacher, Reactive intermediates and the pathogenesis of adverse drug reactions: the toxicology perspective, Current Drug Metabolism 7 (3) (2006) 219–229. [2] J.L. Walgren, M.D. Mitchell, D.C. Thompson, Role of metabolism in druginduced idiosyncratic hepatotoxicity, Critical Reviews in Toxicology 35 (4) (2005) 325–361. [3] F.P. Guengerich, Mechanisms of drug toxicity and relevance to pharmaceutical development, Drug Metabolism and Pharmacokinetics (2010), doi:10.2133/dmpk.DMPK-10-RV-062. [4] B.K. Park, H. Laverty, A. Srivastava, D.J. Antoine, D. Naisbitt, D.P. Williams, Drug bioactivation and protein adduct formation in the pathogenesis of drug-induced toxicity, Chemico-Biological Interactions (2010), doi:10.1016/j.cbi.2010.09.011. [5] F.P. Guengerich, J.S. MacDonald, Applying mechanisms of chemical toxicity to predict drug safety, Chemical Research in Toxicology 20 (3) (2007) 344–369. [6] J. Uetrecht, Screening for the potential of a drug candidate to cause idiosyncratic drug reactions, Drug Discovery Today 8 (18) (2003) 832–837. [7] T.A. Baillie, Metabolism and toxicity of drugs. Two decades of progress in industrial drug metabolism, Chemical Research in Toxicology 21 (1) (2008) 129–137. [8] A.S. Kalgutkar, M.T. Didiuk, Structural alerts, reactive metabolites, and protein covalent binding: how reliable are these attributes as predictors of drug toxicity? Chemistry and Biodiversity 6 (11) (2009) 2115–2137. [9] G.A Doss, T.A. Baillie, Addressing metabolic activation as an integral component of drug design, Drug Metabolism Reviews 38 (4) (2006) 641–649. [10] S. Kumar, K. Mitra, K. Kassahun, T.A. Baillie, Approaches for minimizing metabolic activation of new drug candidates in drug discovery, Handbook of Experimental Pharmacology 196 (2010) 511–544. [11] F.P. Guengerich, Generation of reactive intermediates, Journal of Biochemical and Molecular Toxicology 19 (3) (2005) 173–174. [12] F.P. Guengerich, Cytochrome p450 and chemical toxicology, Chemical Research in Toxicology 21 (1) (2008) 70–83. [13] A.S. Kalgutkar, I. Gardner, R.S. Obach, C.L. Shaffer, E. Callegari, K.R. Henne, A.E. Mutlib, D.K. Dalvie, J.S. Lee, Y. Nakai, J.P. O’Donnell, J. Boer, S.P. Harriman, A comprehensive listing of bioactivation pathways of organic functional groups, Current Drug Metabolism 6 (3) (2005) 161–225. [14] P.J. Shaw, P.E. Ganey, R.A. Roth, Idiosyncratic drug-induced liver injury and the role of inflammatory stress with an emphasis on an animal model of trovafloxacin hepatotoxicity, Toxicological Sciences 118 (1) (2010) 7–18. [15] D.P. Jones, J.J. Lemasters, D. Han, U.A. Boelsterli, N. Kaplowitz, Mechanisms of pathogenesis in drug hepatotoxicity putting the stress on mitochondria, Molecular Interventions 10 (2) (2010) 98–111. [16] D.H. Adams, C. Ju, S.K. Ramaiah, J. Uetrecht, H. Jaeschke, Mechanisms of immune-mediated liver injury, Toxicological Sciences 115 (2) (2010) 307–321. [17] B.K. Gunawan, N. Kaplowitz, Mechanisms of drug-induced liver disease, Clinics in Liver Disease 11 (3) (2007) 459–475. [18] A.K. Daly, C.P. Day, Genetic association studies in drug-induced liver injury, Seminars in Liver Disease 29 (4) (2009) 400–411. [19] A.R. Hughes, C.H. Brothers, M. Mosteller, W.R. Spreen, D.K. Burns, Genetic association studies to detect adverse drug reactions: abacavir hypersensitivity as an example, Pharmacogenomics 10 (2) (2009) 225–233. [20] A.K. Daly, P.T. Donaldson, P. Bhatnagar, Y. Shen, I. Pe’er, A. Floratos, M.J. Daly, D.B. Goldstein, S. John, M.R. Nelson, J. Graham, B.K. Park, J.F. Dillon, W. Bernal, H.J. Cordell, M. Pirmohamed, G.P. Aithal, C.P. Day, HLA-B*5701 genotype is a major determinant of drug-induced liver injury due to flucloxacillin, Nature Genetics 41 (7) (2009) 816–819. [21] A. Kindmark, A. Jawaid, C.G. Harbron, B.J. Barratt, O.F. Bengtsson, T.B. Andersson, S. Carlsson, K.E. Cederbrant, N.J. Gibson, M. Armstrong, M.E. LagerstromFermer, A. Dellsen, E.M. Brown, M. Thornton, C. Dukes, S.C. Jenkins, M.A. Firth, G.O. Harrod, T.H. Pinel, S.M. Billing-Clason, L.R. Cardon, R.E. March, Genomewide pharmacogenetic investigation of a hepatic adverse event without clinical signs of immunopathology suggests an underlying immune pathogenesis, The Pharmacogenomics Journal 8 (3) (2008) 186–195.
59
[22] A.H. Harrill, P.B. Watkins, S. Su, P.K. Ross, D.E. Harbourt, I.M. Stylianou, G.A. Boorman, M.W. Russo, R.S. Sackler, S.C. Harris, P.C. Smith, R. Tennant, M. Bogue, K. Paigen, C. Harris, T. Contractor, T. Wiltshire, I. Rusyn, D.W. Threadgill, Mouse population-guided resequencing reveals that variants in CD44 contribute to acetaminophen-induced liver injury in humans, Genome Research 19 (9) (2009) 1507–1515. [23] H. Takakusa, H. Masumoto, H. Yukinaga, C. Makino, S. Nakayama, O. Okazaki, K. Sudo, Covalent binding and tissue distribution/retention assessment of drugs associated with idiosyncratic drug toxicity, Drug Metabolism and Disposition 36 (9) (2008) 1770–1779. [24] R.S. Obach, A.S. Kalgutkar, J.R. Soglia, S.X. Zhao, Can in vitro metabolismdependent covalent binding data in liver microsomes distinguish hepatotoxic from nonhepatotoxic drugs? An analysis of 18 drugs with consideration of intrinsic clearance and daily dose, Chemical Research in Toxicology 21 (9) (2008) 1814–1822. [25] S. Nakayama, R. Atsumi, H. Takakusa, Y. Kobayashi, A. Kurihara, Y. Nagai, D. Nakai, O. Okazaki, A zone classification system for risk assessment of idiosyncratic drug toxicity using daily dose and covalent binding, Drug Metabolism and Disposition 37 (9) (2009) 1970–1977. [26] J.N. Bauman, J.M. Kelly, S. Tripathy, S.X. Zhao, W.W. Lam, A.S. Kalgutkar, R.S. Obach, Can in vitro metabolism-dependent covalent binding data distinguish hepatotoxic from nonhepatotoxic drugs? An analysis using human hepatocytes and liver S-9 fraction, Chemical Research in Toxicology 22 (2) (2009) 332–340. [27] T. Usui, M. Mise, T. Hashizume, M. Yabuki, S. Komuro, Evaluation of the potential for drug-induced liver injury based on in vitro covalent binding to human liver proteins, Drug Metabolism and Disposition 37 (12) (2009) 2383–2392. [28] J. Gan, Q. Ruan, B. He, M. Zhu, W.C. Shyu, W.G. Humphreys, In vitro screening of 50 highly prescribed drugs for thiol adduct formation–comparison of potential for drug-induced toxicity and extent of adduct formation, Chemical Research in Toxicology 22 (4) (2009) 690–698. [29] D.C. Evans, T.A. Baillie, Minimizing the potential for metabolic activation as an integral part of drug design, Current Opinion in Drug Discovery and Development 8 (1) (2005) 44–50. [30] S.O. Simmons, C.Y. Fan, R. Ramabhadran, Cellular stress response pathway system as a sentinel ensemble in toxicological screening, Toxicological Sciences 111 (2) (2009) 202–225. [31] R. Sawamura, N. Okudaira, K. Watanabe, T. Murai, Y. Kobayashi, M. Tachibana, T. Ohnuki, K. Masuda, H. Honma, A. Kurihara, O. Okazaki, Predictability of idiosyncratic drug toxicity risk for carboxylic acid-containing drugs based on the chemical stability of acyl glucuronide, Drug Metabolism and Disposition 38 (10) (2010) 1857–1864. [32] Y.Z. Shu, B.M. Johnson, T.J. Yang, Role of biotransformation studies in minimizing metabolism-related liabilities in drug discovery, The AAPS Journal 10 (1) (2008) 178–192. [33] C.M. Dieckhaus, C.L. Fernandez-Metzler, R. King, P.H. Krolikowski, T.A. Baillie, Negative ion tandem mass spectrometry for the detection of glutathione conjugates, Chemical Research in Toxicology 18 (4) (2005) 630–638. [34] Z. Yan, N. Maher, R. Torres, G.W. Caldwell, N. Huebert, Rapid detection and characterization of minor reactive metabolites using stable-isotope trapping in combination with tandem mass spectrometry, Rapid Communications in Mass Spectrometry 19 (22) (2005) 3322–3330. [35] A. Mutlib, W. Lam, J. Atherton, H. Chen, P. Galatsis, W. Stolle, Application of stable isotope labeled glutathione and rapid scanning mass spectrometers in detecting and characterizing reactive metabolites, Rapid Communications in Mass Spectrometry 19 (23) (2005) 3482–3492. [36] J. Gan, T.W. Harper, M.M. Hsueh, Q. Qu, W.G. Humphreys, Dansyl glutathione as a trapping agent for the quantitative estimation and identification of reactive metabolites, Chemical Research in Toxicology 18 (5) (2005) 896–903. [37] D. Argoti, L. Liang, A. Conteh, L. Chen, D. Bershas, C.P. Yu, P. Vouros, E. Yang, Cyanide trapping of minimum ion reactive intermediates followed by detection and structure identification using liquid chromatography–tandem mass spectrometry (LC–MS/MS), Chemical Research in Toxicology 18 (10) (2005) 1537–1544. [38] G. Tuschl, B. Lauer, S.O. Mueller, Primary hepatocytes as a model to analyze species-specific toxicity and drug metabolism, Expert Opinion on Drug Metabolism and Toxicology 4 (7) (2008) 855–870. [39] D.C. Evans, A.P. Watt, D.A. Nicoll-Griffith, T.A. Baillie, Drug–protein adducts: an industry perspective on minimizing the potential for drug bioactivation in drug discovery and development, Chemical Research in Toxicology 17 (1) (2004) 3–16.