What is the study of physiochemical properties of drugs and how they influence the body called?

Maria Karlgren and Christel A. S. Bergström, CHAPTER 1:How Physicochemical Properties of Drugs Affect Their Metabolism and Clearance , in New Horizons in Predictive Drug Metabolism and Pharmacokinetics, 2015, pp. 1-26 DOI: 10.1039/9781782622376-00001 eISBN: 978-1-78262-237-6

From Book Series: Drug Discovery

In this chapter the transport proteins and enzymes of importance for drug clearance are discussed. The primary organ for drug metabolism is the liver and to reach the intracellular compartment of hepatocytes, orally administered drugs must cross both the intestinal wall and the cell membrane of the liver cells. Transport proteins present in the cellular membrane may facilitate or hinder the compounds crossing these cellular barriers and hence will influence to what extent compounds will reach the enzymes. Here, the enzymes and transport proteins of importance for drug clearance are discussed. The molecular features of importance for drug interactions with transport proteins and enzymes are analyzed and the possibility to predict molecular features vulnerable to enzymatic degradation is discussed. From detailed analysis of the current literature it is concluded that for interaction, both with transport proteins and enzymes, lipophilicity plays a major role. In addition to this property, molecular properties such as hydrogen bond acceptors and donors, charge, aromaticity and molecular size can be used to distinguish between routes of clearance.

The physicochemical properties of compounds have been used for more than a century to predict or estimate pharmacokinetic processes. The most well known property is lipophilicity, often defined as the partition coefficient between octanol and water. This property is related to passive diffusion across cell membranes, solubility, interaction with receptors, metabolism and toxicity. To activate proteins, e.g. receptors and enzymes, the compound needs to bind to a binding pocket. Besides lipophilicity, physicochemical properties of importance for binding include molecular size, hydrogen bond acceptors/donors and charge. This chapter discusses the physicochemical properties of importance for drug metabolism. The primary organ for drug metabolism is the liver and to reach the liver the compound must cross cellular barriers. Absorption from the gastrointestinal tract (GIT) is therefore of critical importance for orally administered drugs, before distribution into and out of the liver can occur. We introduce the GIT in this chapter, and all of these processes are discussed in detail in other chapters. Thereafter we describe the enzymes responsible for drug metabolism in different tissues; the biology of these enzymes is further discussed in later chapters. Finally, the role of the enzymes and that of transporters in drug clearance is presented together with an analysis of the structural features of molecules of importance for binding to enzymes and transporters.

Only free molecules can pass through cell barriers and, hence, only the unbound fraction of drugs can pass over the intestinal epithelium. Solubility governs the concentration reached in the intestinal fluid and is therefore a major driving force for the absorption. Hydrophilic and small molecules may be absorbed by diffusing through the paracellular route. This transport route has limited capacity as its total surface area is much smaller than that of the transcellular (membrane related) pathway. Furthermore, the tight junction reduces the pore size. In the small intestine compounds greater than 4 Å have limited permeability through this pathway whereas those greater than 15 Å are excluded from permeation.1 To cross the cell membrane compounds have several options. The two most common pathways are passive diffusion through the lipoidal membrane or active transport mediated by transport proteins. The impact of these pathways is heavily debated. Kell and coworkers have challenged the theory that the majority of drugs use lipoidal passive diffusion to pass through cells.2–4 Their hypothesis is that most of the transport across cells involves active processes and transport proteins. This debate has spurred research to determine to what extent the two pathways are involved in drug distribution.5,6

Once the compound has traversed the luminal membrane, it may either diffuse through the cytosol and cross the serosal membrane, or interact with enzymes, intracellular organelles (lysosomes, endoplasmic reticulum) or the cell nuclei. It has been proposed that there is a substrate overlap between cytochrome P450 3A4 (CYP3A4) and the efflux protein P-glycoprotein [P-gp; also known as multidrug resistance protein 1 (MDR1)]. Hence, these two different pathways may have synergistic effects in the clearance and detoxification of certain compounds.

When the compound has crossed the intestinal epithelium it reaches the portal vein from where the systemic circulation transports it to the main metabolic organ in the body—the liver. The compound can then reach the cytosol by either passive or active transport mechanisms across the basolateral membrane facing the bloodstream. The capacity of the liver as a detoxification organ is remarkable. Even compounds with high protein binding can be extracted to a large extent by the liver. This can be exemplified by atorvastatin, the cholesterol-lowering compound marketed as LIPITOR®. A high fraction of atorvastatin is absorbed from the intestine but is also highly bound (98%) to proteins in the blood. Therefore, only 2% is available in an unbound free form that can permeate the cell membrane. In spite of this, the absolute bioavailability after oral administration is only 14%. This low number is a result of the cooperation between active influx transporters [mainly organic anion-transporting polypeptides (OATP) 1B1 and 1B3] in the basolateral membrane and CYP3A4 in the cytosol. In addition to these processes, atorvastatin is thereafter cleared from the hepatocytes through canalicular efflux by P-gp. The drug transporters and metabolic enzymes in the gut and liver that are crucial for the first-pass effect are shown in Figure 1.1. The metabolic capacity of the gut and liver are discussed in more detail below.

The intestine is the most important extrahepatic site of drug metabolism and its involvement in the first-pass metabolism of orally administered drugs makes it a major determinant of drug bioavailability. The most abundant CYP in the small intestine is CYP3A, constituting 50–82% of the intestinal CYP content.7–10 However, compared with the liver, the total mass of CYP3A in the small intestine corresponds to only about 1% of the hepatic CYP3A levels.9,11 Other CYPs expressed in the small intestine, as determined using immunoblotting or liquid chromatography-tandem mass spectrometry (LC-MS/MS)-based protein quantification, are CYP2C9, CYP2C19, CYP2D6 and CYP2J2 (Table 1.1).7,8 Their expression levels vary in the different regions of the intestine. CYP3A, CYP2C and CYP2D6 show highest expression in the proximal intestinal region and decreasing levels in the distal regions.9,12 For CYP2J2, the expression is constant throughout the GIT.13

More CYPs are expressed in human liver and at higher expression levels than in the intestine. In 1994 Shimada et al. determined the expression levels of the major drug metabolizing CYPs in human liver using P450-spectra of total CYP content and SDS-PAGE and immunoblotting in 60 people (30 Caucasians and 30 Japanese).15 Although the expression levels displayed both interindividual and interethnic variations, the average expressions levels in comparison with total CYP content were: CYP3A (28.8%) > CYP2C (18.2%) > CYP1A2 (12.7%) > CYP2E1 (6.6%) > CYP2A6 (4.0%) > CYP2D6 (1.5%) > CYP2B6 (0.2%). Since then the methodological development of more sophisticated methods, e.g. different types of LC-MS/MS-based proteomics, has allowed quantification of CYPs.7,14,16,17 Although the quality of the LC-MS/MS analyses may vary due to the level of method validation etc., the results are consistent with those obtained by Shimada et al. that identified the CYP3A and CYP2C families as the most abundant hepatic CYPs (Table 1.2). All tissues have enzymatic activity to some extent; however, the gut and the liver are the two most important metabolic tissues. An overview of the metabolic profile of different tissues is provided in Table 1.3. It should be noted that not only the type of enzymes differs between tissues; the expression levels of these enzymes differ as well.

Although the abundance of CYPs is of major interest it does not provide the full picture of the importance of the specific CYP enzymes for drug metabolism. One striking example showing discrepancy between expression levels and importance is the CYP2D6 enzyme. This enzyme is only expressed in low levels in human liver (1.5–2% of the total CYP content). However, it is one of the major drug-metabolizing enzymes and metabolizes up to 25% of clinically used drugs.31–34 Another example is CYP1A2. It constitutes approximately 18% of the human hepatic CYP content,15 but its relative importance in drug metabolism is only 3–9% (Table 1.4).32–34

In addition to interindividual variation in expression levels, many drug-metabolizing enzymes, and especially some of the CYPs, are highly polymorphic. Approximately 40% of CYP-dependent phase I metabolism is performed by polymorphic CYPs, including CYP2D6, CYP2C9, CYP2C19 and CYP2B6.36 For CYP2D6 more than 100 different alleles and suballeles have been identified. These include alleles where the entire CYP2D6 gene is deleted, alleles with duplicated or multiduplicated CYP2D6 genes, and alleles containing single-nucleotide polymorphisms (SNPs).36 Such gene variants may of course have a major impact on the pharmacokinetics and may result in adverse effects of drugs that are CYP2D6 substrates (cf.37,38). A classic example of this is the prodrug codeine, which is activated by CYP2D6 into the active drug morphine. For people who are poor CYP2D6 metabolizers, i.e. their CYP2D6 genes are deleted or contain mutations leading to non-functional enzymes, codeine does not give the desired analgesic effect.39 On the contrary, for individuals with alleles with duplicated or multiduplicated CYP2D6 genes, i.e. ultra-rapid metabolizers, codeine is activated rapidly, which can lead to codeine toxicity and central nervous system depression.40 There are also a few cases of infant mortality where ultra-rapid metabolizer mothers treated with codeine transferred fatally high morphine concentrations to their breastfed infants.41–43

For other CYPs, e.g. CYP2C9, CYP2C19 and CYP2B6, many variant alleles and suballeles have been described, some of which have significant clinical impact. The most well known clinical CYP2C examples are the CYP2C9 polymorphisms involved in warfarin metabolism44,45 and CYP2C19 polymorphisms associated with clopidogrel activation.46,47 Both of these were highlighted in 2011 as important pharmacogenomics biomarkers.37 The warfarin (COUMADIN®) and the clopidogrel (PLAVIX®) Food and Drug Administration (FDA) drug labels have been updated to contain recommendations for initial doses based on e.g. CYP2C9 genotype48 and a warning about diminished effectiveness in CYP2C19 poor metabolizers,49 respectively. A complete and updated overview of CYP and CYP oxidoreductase (POR) polymorphisms can be found on the home page of The Human Cytochrome P450 (CYP) Allele Nomenclature Committee (http://www.cypalleles.ki.se).

Many substrates and inhibitors have been identified for transporters that are of importance for drug distribution into and out of cells. For a selection of these, see Table 1.5. While a substrate of the transporter can also be an inhibitor of the transport protein and block transport of other compounds, compounds that have been identified as inhibitors may not be transported. The latter is related to the inactivation of the transport protein by binding to sites other than the one crucial for mediating transport. Interaction with the transport-mediating site allows the drug compound (or its metabolite) to traverse the lipophilic membrane. Hence, the molecular requirements of the different transporters have been studied to better understand what physicochemical properties of a compound will result in them being actively transported by a particular transport protein. While metabolism is a chemical reaction that turns a substrate into a product that is chemically different, the substrates of transport proteins remain the same; no chemical reaction occurs. However, the terminology of transporters and experimental procedures to study transport have been inspired by those in the metabolism field. So, for example, the Michaelis–Menten equation is often used to describe the efficiency of transporters to flux compounds across the membrane.

The structural requirements for transport by influx and efflux transport proteins have been heavily studied. The majority of studies have been directed towards investigation of transport protein inhibition. The reason for this is mainly methodological issues associated with substrate assays. While analyses of molecular features of substrates require determination of the intracellular concentration of a large number of compounds, inhibition assays rely on screening a large number of compounds for their inhibition of the transport of one substrate. Hence, analytical demands for the latter are reduced and a higher throughput mode is possible. The most important transport proteins for clearance are discussed below.

One of the most studied transport proteins is the efflux protein P-gp since it is important for drug distribution to several tissues, including the gut and liver. Drug–drug interactions (DDIs) have also been identified that are mediated by P-gp. Among the most well known are those that occur between digoxin and the P-gp inhibitors amiodarone, cyclosporin A, quinine and verapamil.50 Seelig and coworkers were pioneers in the study of the recognition pattern of P-gp (cf.51,52). Based on studies of ∼100 compounds, they suggested that a special spatial separation of electron donor groups is required for compounds to be transported by P-gp. Their work was followed by a number of structure–activity relationship (SAR) studies in which P-gp substrates are predicted on the basis of chemical information calculated from the molecular structure. The SAR models are typically classification models used to distinguish compounds that are substrates from those that are not transported by the P-gp. One classification model used the sum of atomic electrotopological states (MolES), a descriptor of molecular bulkiness, to predict substrates.53 Compounds with a MolES >110 are regarded as substrates for P-gp whereas a MolES <49 indicates non-substrates. For compounds with a value between 49 and 110 other descriptors are needed to identify whether they would be substrates.53 A similar study using a classification approach established the rule of four.54 This rule states the following: compounds with (N + O) ≥ 8, molecular weight > 400 and acid pKa > 4 are likely to be P-gp substrates. Compounds with (N + O) ≤ 4, molecular weight < 400 and base pKa < 8 are likely to be non-substrates. Both of these SAR studies identified that P-gp transports larger molecules. Furthermore, it seems that compounds with many hydrogen bonds, and to some extent negative charges, are transportable by P-gp. The non-substrates have fewer hydrogen bond acceptors and are neutral, or at least not highly positively charged. The importance of N and O demonstrated by this study confirms the work by Seelig and colleagues. Finally, P-gp substrates are amphipathic and lipophilic.55 It has been suggested that the substrate binding pocket sits inside the cellular membrane and needs to be accessed by distribution into the lipid bilayer.56–58 Based on this, the lipophilic and amphiphilic nature of the substrates is to be expected.

While it is important to understand molecular features that result in substances being substrates to efflux proteins, it is also of interest to look at which molecular features lead to inhibition of transport. Inhibition may result in severe DDIs. Inhibitors may be competitive (they bind to the same binding site as the substrate) or non-competitive (they bind to another site on the transport protein and thereby block the transport). Therefore, a substrate may inhibit the transport of another substrate, and an inhibitor is not necessarily transported by the protein. Artursson and colleagues have explored large compound series to identity inhibitors of the transport proteins most important for drug disposition. They identified specific molecular requirements of the different transporters and the extent to which the molecular requirements for inhibition of these transporters overlap. For example, the ABC transporters P-gp, breast cancer resistance protein (BCRP), multidrug resistance-associated protein 2 (MRP2) and bile salt export pump (BSEP), all of which are expressed in the canalicular membrane of the hepatocyte, have a significant overlap of inhibitors, i.e. the same compound may block several of these transporters at the same time. The impact on drug clearance, for instance from hepatocytes to bile, may therefore be greatly affected. Such inhibition may also result in reduced enterohepatic recycling of endogenous substances such as bile acids and bilirubin, which can result in, among others, fatal cholestasis.59 In a study of 122 compounds, all tested for their inhibition of P-gp, BCRP and MRP2, molecular features of specific inhibitors (interacting with only one of the transporters) and of those that interacted with all three transporters were identified.60 The inhibitors of P-gp were lipophilic, non-polar and had higher structure connectivity. BCRP inhibitors were also more lipophilic than non-inhibitors and the number of aromatic rings correlated positively with inhibition. Inhibitors of MRP2 had similar properties; lipophilicity and unsaturated bonds (double bonds) positively correlated with inhibition, as did shape. Thus, inhibitors of P-gp, BCRP and MRP2 are all lipophilic and aromatic, but to different degrees. The specific inhibitors of P-gp are less aromatic than those of MRP2 and BCRP, and the BCRP inhibitors generally have more aromatic nitrogens than the P-gp inhibitors. P-gp inhibitors are the most lipophilic (logDpH7.4 of 2.3) followed by BCRP (logDpH7.4 of 1.9) and MRP2 (logDpH7.4 of 1.2). By contrast, multi-specific inhibitors, i.e. compounds that inhibit all three proteins, are 100- to 1000-fold more lipophilic (logDpH7.4 of 4.5).60

Another study investigated 250 compounds for their inhibition of BSEP. Of the 86 inhibitors identified, 58% were neutral at physiological pH, 36% were negatively charged and only 6% were positively charged. By contrast, BSEP substrates are typically monovalent, negatively charged bile acids. BSEP inhibition is also favored by lipophilicity, hydrophobicity and number of halogens. Reciprocally, hydrophilicity and hydrogen bond acceptors negatively correlate with inhibition.61

There are a number of studies on the inhibition of OATP uptake transporters, particularly OATP1B1 (which is the most important hepatic OATP). Two studies by Karlgren et al. investigated the inhibition of OATP1B1 by 146 compounds and the inhibition of OATP1B1, OATP1B3 and OATP2B1 by 225 compounds.62,63 In both studies, a significantly larger proportion of the inhibitors were negatively charged compounds compared with the non-inhibitors. This is not surprising given that OATPs are known to primarily transport anionic drugs. Furthermore, these studies showed that compared with the non-inhibitors, the OATP inhibitors had a significantly higher lipophilicity (mean NNLogP of 3.6–4.0 vs. 2.3–2.7), larger molecular weight (mean weight of 481–514 vs. 325–336 g mol−1) and a larger polar surface area (PSA; mean PSA of 115–142 vs. 66–74 Å2).62,63 OATP1B1 inhibitors also displayed a lower mean square distance index (MSD), a topological distance descriptor normalized for size.63 Inhibitors of OATP1B3—but not of OATP1B1 and OATP2B1—had more hydrogen bond donors than the non-inhibitors, whereas the OATP2B1 inhibitors were less dependent on polarity than those of OATP1B1 and OATP1B3.62 These findings were confirmed by an in vitro study of 2000 compounds on OATP1B1 and OATP1B3.64 It was also found that a low number of aromatic bonds (<7) correlated positively with OATP1B1 inhibition but negatively with OATP1B3 inhibition, whereas a logD value of >7.5 and 3–4 hydrogen bond donors correlated positively with OATP1B3 inhibition. Interestingly, due to the high number of compounds investigated, they could also identify substructures that favored inhibition of a specific transporter or favored inhibition of both OATP1B transporters.

The three OATP transporters share many inhibitors. Two examples are atazanavir and ritonavir, which are considered general OATP inhibitors.62 In one study of 91 identified inhibitors, 42 were common for OATP1B1 and OATP1B3. Of these 42 inhibitors, 16 did not inhibit OATP2B1. By contrast, only 9 of the inhibitors were identified as inhibitors of OATP1B1 and OATP2B1 but not OATP1B3. Only one compound, nefazodone, interacted with both OATP1B3 and OATP2B1 but did not inhibit OATP1B1.

Many of the compounds identified as inhibitors of the OATP transporters are also inhibitors or substrates of other transporters or metabolizing enzymes. For example, the FDA and/or the European Medicines Agency (EMA) list that 67 of the 225 compounds included in the studies above are substrates, inhibitors or inducers of CYP enzymes. Of these 67 compounds, 21 compounds were also identified as inhibitors of one or more OATP transporters.62 The largest overlap was for OATPs and CYP2C8, followed by OATPs and CYP3A4. Previously it was suggested that there was a substrate overlap between OATP1B1 and the efflux transporter MRP2.65 However, an investigation of common inhibitors of OATP1B1 and MRP2 found no such corresponding overlap of inhibitors.63

Organic cation transporter 1 (OCT1) is the major cationic uptake transporter in the liver. An investigation of 191 compounds identified 62 as inhibitors of OCT1.66 These inhibitors tended to be positively charged (66%) or neutral (32%) at physiological pH. They were more lipophilic (mean ClogP of 3.50 vs. 1.43), had a lower PSA (mean PSA of 42.9 vs. 95.5), and a lower number of both hydrogen bond donors (1.07 vs. 2.66) and acceptors (3.38 vs. 5.09) than the non-inhibitors.66 These results agree with a previous study of OCT1 inhibition that used a more homogeneous dataset (n = 30).67 The results also support previous observations that a positive charge is important for interactions with the OCT1 transporter.68,69

Table 1.6 presents a representative sample of the many substrates and inhibitors of the enzymes responsible for drug metabolism. The liver is the organ with the highest metabolic capacity (Table 1.4). The enzymes of highest importance for drug metabolism in this tissue are CYP3A4, CYP2C9, CYP2C19 and CYP2D6. Below we discuss the molecular features of the substrates and inhibitors of these four enzymes. We focus on CYP3A4 as this enzyme is of the greatest importance for drug metabolism and therefore the most studied.

The binding pocket of CYP3A4 is quite large. Pharmacophore modeling has been used to reveal the molecular requirements of compounds that bind and activate the enzyme.74 Information extracted from 38 compounds and the software Catalyst showed that the large binding pocket required interaction with a hydrophobic fragment and hydrogen bond interactions through a hydrogen bond donor and a hydrogen bond acceptor. These different features require a particular spatial distribution in the molecule to interact with the binding pocket. The pharmacophore was later regenerated in a study by Norinder, who also identified another pharmacophore with similar accuracy.75 Norinder included more hydrophobic interaction points (three hydrophobic fragments) and only one hydrogen bond acceptor to achieve the same quality of pharmacophore as the previous one. This shows the complexity in identifying the molecular features that characterize substrates. A more complex approach, also using pharmacophore modeling, identifies structures vulnerable to metabolism and the reactive site. This methodology calculates the fingerprint of both the enzyme and the substrate. The calculations are based on GRID methodology, i.e. the enzyme fingerprints are calculated by the GRID flexible molecular interaction fields and the substrate fingerprints are obtained through GRID probe pharmacophore recognition. The latter calculates hydrophobicity, hydrogen bond donors and acceptors, and charge to obtain a fingerprint of each atom in the molecule. These descriptors are then assessed for their capacity to interact with the reactive heme atom of the enzyme. This is performed through assessment of how accessible they are for interactions with the heme. This fingerprint method has resulted in the software MetaSite for prediction of vulnerable sites for CYP metabolism of CYP3A4, CYP2C9 and CYP2D6 (see Section 1.7.1).76–78

Another method to predict the site of metabolism (SOM) in a drug molecule was developed at the University of Copenhagen. Their approach uses quantum chemical calculations with the density functional method B3LYP to estimate the activation energy required for different atoms to become the SOM. The group has also tested less time-consuming calculations by making use of the semi-empirical AM1 method.79 Using the two methods together, the following descriptors were calculated for the substrate and the radical obtained after dehydration: the Mulliken charges on the carbon and hydrogen atoms involved in the reaction; the spin of the carbon atom in the radical; the energies of the highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO); and the energy difference between these two orbitals. The coefficients for the hydrogen 1s and carbon 2p atomic orbitals in the HOMO and LUMO were also calculated. Their analyses of computational models of different complexity revealed that simpler and less computationally demanding methods could be used to identify SOM. This spurred many articles (cf.80–82) as well as the development of the software SMARTCyp, further discussed in Section 1.7.2.

Inhibitors of CYP3A4 have been studied extensively. Datasets in earlier studies often had fewer than 30 compounds and, based on these, typically linear SAR models were developed. These studies identified lipophilicity as an important descriptor for achieving inhibition.75,83,84 Later studies used datasets with several hundreds of compounds. The aim of these was to find molecular motifs of importance for CYP3A4 inhibition or to develop global computational models for predicting the risk that a new compound might be a CYP3A4 inhibitor. AstraZeneca studied 463 compounds for their CYP3A4 inhibition and the response data were analyzed by either partial least squares or regression tree methodology.85 The modeling used molecular descriptors based on atoms and fragments as input. The greater the aromaticity and lipophilicity, the more potently the compound inhibited CYP3A4. Furthermore, neutral compounds and bases inhibited CYP3A4 whereas negatively charged compounds bound to other isozymes. Another study of 741 compounds developed a classification model to distinguish inhibitors from non-inhibitors. The model was then validated with a test set of 186 compounds. The recognition rate of the model was relatively high and 73% of the compounds were correctly identified as either inhibitors or non-inhibitors. The final model was based on constitutional, electrostatic and geometric descriptors. Examples of these are molecular weight, flexibility (number of flexible bonds, rigid bonds and rings), charge, lipophilicity and van der Waals surface area. This model identified that inhibitors are larger than non-inhibitors and extracted a cut-off value of 354 g mol−1. Inhibitors are also more hydrophobic and have fewer chargeable groups, the latter being <5% of the molecular composition for inhibitors. Another molecular property that discriminates inhibitors from non-inhibitors is the number of nitrogens; a compound with more than two nitrogens does not inhibit CYP3A4.86 A study of 1756 compounds on the inhibition of CYP3A4 concluded that the models for prediction of CYP3A4 inhibition must be based on algorithms that can handle the complexity of enzymatic inhibition and the resulting non-linear data.87

Computational analyses of pharmacophore modeling, protein conformation analyses and multivariate data analyses have identified that substrates to CYP2C9 are hydrophobic (up to two functions), and include at least one hydrogen bond donor and one hydrogen bond acceptor.88–91 Substrates of CYP2C9 are favored by being negatively charged but metabolism by CYP2C19 enzyme is not. Substrates of CYP2D6 often contain overlapping hydrophobic features, a hydrogen bond donor function well separated from the hydrophobic features and negative molecular electrostatic potential.92,93 The hydrophobic domain near the oxidation site interacts with a large, flat and lipophilic region of the CYP2D6 that contains residues Leu121, Leu213, Ala305, Val370 and Thr309. The hydrogen bond donor group can have two different spatial locations from the hydrophobic region. Substrates with the nitrogen atom positioned 10 Å from the oxidation center interact with CYP2D6 through hydrogen bonds between the nitrogen and Glu216 and Gln117. Other substrates have the nitrogen atom positioned 5 or 7 Å away from the oxidation center and the nitrogen interacts only with the Asp301 residue of CYP2D6.93

Gleeson and colleagues studied a dataset of 457 compounds to predict CYP2C9 inhibitors.85 Molecular features that correlated with CYP2C9 inhibition were lipophilicity, aromaticity and non-ionizability. The enzyme was also found to be inhibited by negatively charged compounds, a finding also confirmed by Manga et al.94 Pharmacophore models based on three different datasets (n = 9, 29 and 13, respectively) each generated a different pharmacophores that inhibited CYP2C9. The pharmacophores differed from each other spatially and in other important molecular features.95 All three pharmacophore models included one hydrophobic pocket and one hydrogen bond acceptor but they differed in number of hydrophobic pockets (1–2), hydrogen bond acceptors (1–2) and whether they had hydrogen bond donors. There is a large overlap in the physicochemical properties of inhibitors of CYP2C9 and CYP2C19. For example, Gleeson et al., who used a dataset of 369 compounds, found that CYP2C19 is also inhibited by lipophilic and aromatic compounds.85 However, CYP2C19 has a preference for neutral compounds. The overlap between these two enzymes is understandable as they share 95.7% homology—only 43 of their 490 amino acids differ from each other.96 Studies of CYP2C19 have also revealed that stronger inhibitors are more lipophilic at the N-3 position. The binding affinity of the inhibitors also increases with the degree of steric bulk. This is a result of the general entropic effect associated with solvation where the increased order of the bulk water for larger compounds favors binding of such molecules to the enzyme.

In contrast to the CYP3A4, CYP2C9 and CYP2C19 inhibitors, the role of lipophilicity for inhibition of CYP2D6 is less clear. Using a dataset of 170 compounds, Gleeson et al. identified that inhibitors of this enzyme are aromatic structures with weak basic functions, but not lipophilic per se.85 Groot et al. reviewed different models for the prediction of CYP2D6 inhibitors (and substrates).92 Based on 3500 compounds, they extracted the following rules for inhibitors: (i) inhibitors are weak bases (92% of the compounds with CYP2D6 IC50 < 1 mM were weak bases); and (ii) decreasing polarity, as measured by the total PSA (TPSA), increases CYP2D6 inhibition (e.g. 73% of the weak bases had CYP2D6 IC50 < 10 mM when the TPSA was <50 Å2). In comparison, this number was 37% when TPSA was >100 Å2. In contrast to the study by Gleeson et al. Groot and coworkers found that lipophilicity was positively related to inhibition. The majority (73%) of the weak bases with a calculated logP of 3–5 had CYP2D6 IC50 < 10 mM, whereas this number was 45% for the weak bases with a logP of 1–3.

A number of different methods and software are available for the prediction of metabolism. These enable predictions of metabolic sites, metabolic reactions and products, mechanisms, and enzyme dynamics. More traditional structure–activity approaches are also used and models have been developed to predict CYP inducers and inhibitors. An extensive list of software can be found in the article published by Kirchmair and coworkers.97 Three commonly used pieces of software that make use of a combination of different in silico approaches are MetaSite, SMARTCyp and StarDrop. These are briefly described below.

MetaSite (Molecular Discovery, Italy) predicts phase I metabolism mediated by CYP enzymes and flavin-containing monooxygenases. It predicts the binding between substrates and enzymes (a thermodynamic factor) and the chemical transformation (a kinetic factor). The predictions are obtained by a combination of molecular interaction fields that analyze ligand and enzyme properties, together with quantum mechanics and knowledge-based components that relate to the kinetics of metabolism (i.e. the reactivity). The software enables identification of molecular sites vulnerable to metabolism so that medicinal chemists can redesign structures that are rapidly cleared through metabolism. To further improve predictions of metabolism patterns and reactivity, Molecular Discovery has formed a human CYP consortium with some pharmaceutical companies. The overarching goal of this consortium is to produce high-quality experimental data to improve predictions for metabolic rate, site and reaction pathways, and the likelihood of a compound being a substrate/inhibitor for a specific enzyme. Information about the program can be found on the MetaSite webpage (http://www.moldiscovery.com/software/metasite).

The software SMARTCyp originates from the University of Copenhagen. It includes a fragment-based database for which the density functional theory activation energy has been calculated. This database is then used to match structural fragments of drug molecules to estimate CYP3A4-, CYP2D6- and CYP2C9-mediated transformation. Data from 211 transitions were used to develop the fragment-based energy rules. To rank the SOM in the molecules, an accessibility descriptor is used.

Figure 1.2 shows clopidogrel, an antithrombosis drug discussed in the pharmacogenomics Section (1.4). SMARTCyp provides probability estimations of a particular atom being the site of enzymatic activation. For clopidogrel, which is composed of 18 heteroatoms, the atoms are numbered 1 to 18. The estimation is based on three factors related to the molecular structure: activation energy, accessibility and solvent-accessible surface area. The activation energy is the approximate energy required for the reaction of the catalytic site of the enzyme (i.e. CYP3A4, CYP2C9 or CYP2D6) to occur at this atom. Accessibility is a measure of the distance of the particular atom from the center of the molecule and always has a number between 0.5 (atom is positioned in the center) and 1.0 (atom is positioned at the far end of the molecule). The solvent-accessible surface area is the total surface area of that particular atom exposed to e.g. water and thereby accessible for interaction with the enzyme. This surface area is predicted from 2D molecular topology descriptors. CYP3A4 and CYP3A5 are the most important enzymes for hepatic metabolism of clopidrogel.98 Clopidogrel is a prodrug and requires enzymatic activation in vivo, the atom of importance for this is the carbon next to the sulfur in the thiophene group (Figure 1.2).99 It should be noted that in vivo about 85% of clopidogrel is inactivated by hydrolysis of the ester group, i.e. this ester is necessary to obtain active inhibition of platelet aggregation. This type of reaction is not predictable by the SMARTCyp software.

Similar to SMARTCyp, the software Stardrop (Optibrium, UK) predicts reaction sites. It makes use of quantum mechanics calculations to predict metabolic sites and their vulnerability to different CYPs. This software has also been suggested as a useful tool in the redesign of enzymatically liable molecules. The accuracies of the predictions of SMARTCyp and StarDrop are similar.80

This chapter has reviewed the physicochemical properties of compounds that determine their interaction with transport proteins and the enzymes involved in drug clearance. Lipophilicity is an important physicochemical property resulting in interaction and, in particular, inhibition of both transport proteins and enzymes. Other specific features of the substrates and inhibitors are summarized below. Substrates to OATP1B1, OATP2B1 and OATP1B3 are negatively charged. Inhibitors of these transport proteins are lipophilic (∼logP 4), large (∼500 g mol−1) and polar ∼120 Å2, although OATP2B1 inhibitors are less dependent on polarity than those of OATP1B1 and OATP1B3. Inhibitors of OATP1B3 have a larger number of hydrogen bond donors (3–4). A low number of aromatic bonds (<7) increases the risk of OATP1B1 inhibition, but reduces the risk of OATP1B3 inhibition. Substrates to OCT1 are cationic and, hence, many of the inhibitors also carry a positive charge. Furthermore, inhibitors are lipophilic (logP of 3.50) and less polar (PSA of 43), with a lower number of hydrogen bond donors (1) and acceptors (∼3) than the non-inhibitors. P-gp substrates require specific hydrogen bond interactions and compounds with (N + O) ≥ 8, molecular weight > 400 g mol−1 and acid pKa > 4 are likely to be P-gp substrates. Inhibitors of P-gp and other efflux proteins (BCRP and MRP2) are lipophilic and aromatic. P-gp is the most lipophilic (logDpH7.4 of 2.3) followed by BCRP (logDpH7.4 1.9) and MRP2 (logDpH7.4 of 1.2). Multi-specific inhibitors are 100- to 1000-fold more lipophilic (logDpH7.4 of 4.5). Specific inhibitors of P-gp are less aromatic than those of MRP2 and BCRP. In addition, BCRP inhibitors have a greater number of aromatic nitrogens than P-gp inhibitors. Studies of CYP substrates have identified the importance of an acidic function, hydrogen bond donors and hydrogen bond acceptors for CYP2C9, a weak basic function (i.e. cationic charge), and aromatic ring features for CYP2D6 and hydrophobic features for CYP3A4. Inhibition of CYP2C9, CYP2C19, CYP2D6 and CYP3A4 is highly related to the lipophilicity and aromaticity of the drug—inhibition increases with lipophilicity for all four enzymes. Inhibitors of CYP2C9 and CYP2C19 have common features; however, CYP2C9 prefers negatively charged compounds whereas CYP2C19 is inhibited by neutral compounds. Inhibition of CYP3A4, which is the most studied enzyme, is favored by aromatic structures, larger molecules (molecular weight > 354 g mol−1) and compounds with a low number of nitrogens (<2).

The literature on substrates and inhibitors of the transporters and CYP enzymes suggests that different computational approaches are required to arrive at reliable and robust predictions of interactions. Increasing the prediction accuracy of SOM and the development of quantitative models, for the prediction of e.g. IC50 values, will require elucidation of the pathway (competitive/non-competitive inhibition and binding site of the enzyme for substrates) based on large, high-quality datasets. In addition, the interplay between transporters and enzymes needs further attention to understand how drug-like compounds access the intracellular enzymes.

1. J. L. Madara and K. Dharmsathaphorn , J. Cell Biol., 1985, 101 , 2124 CrossRef CAS PubMed .
2. P. D. Dobson and D. B. Kell , Nat. Rev. Drug Discovery, 2008, 7 , 205 CrossRef CAS PubMed .
3. D. B. Kell Trends Pharmacol. Sci., 2015, 36 , 15 CrossRef CAS PubMed .
4. D. B. Kell , P. D. Dobson and S. G. Oliver , Drug Discovery Today, 2011, 16 , 704 CrossRef CAS PubMed .
5. L. Di , P. Artursson , A. Avdeef , G. F. Ecker , B. Faller , H. Fischer , J. B. Houston , M. Kansy , E. H. Kerns , S. D. Kramer , H. Lennernas and K. Sugano , Drug Discovery Today, 2012, 17 , 905 CrossRef CAS PubMed .
6. D. Smith , P. Artursson , A. Avdeef , L. Di , G. F. Ecker , B. Faller , J. B. Houston , M. Kansy , E. H. Kerns , S. D. Kramer , H. Lennernas , H. van de Waterbeemd , K. Sugano and B. Testa , Mol. Pharmaceutics, 2014, 11 , 1727 CrossRef CAS PubMed .
7. C. Groer , D. Busch , M. Patrzyk , K. Beyer , A. Busemann , C. D. Heidecke , M. Drozdzik , W. Siegmund and S. Oswald , J. Pharm. Biomed. Anal., 2014, 100 , 393 —401 CrossRef CAS PubMed .
8. M. F. Paine , H. L. Hart , S. S. Ludington , R. L. Haining , A. E. Rettie and D. C. Zeldin , Drug Metab. Dispos., 2006, 34 , 880 CrossRef CAS PubMed .
9. M. F. Paine , M. Khalighi , J. M. Fisher , D. D. Shen , K. L. Kunze , C. L. Marsh , J. D. Perkins and K. E. Thummel , J. Pharmacol. Exp. Ther., 1997, 283 , 1552 CrossRef CAS .
10. P. B. Watkins , S. A. Wrighton , E. G. Schuetz , D. T. Molowa and P. S. Guzelian , J. Clin. Invest., 1987, 80 , 1029 CrossRef CAS PubMed .
11. J. Yang , G. T. Tucker and A. Rostami-Hodjegan , Clin. Pharmacol. Ther., 2004, 76 , 391 CrossRef CAS PubMed .
12. I. de Waziers , P. H. Cugnenc , C. S. Yang , J. P. Leroux and P. H. Beaune , J. Pharmacol. Exp. Ther., 1990, 253 , 387 CrossRef CAS .
13. D. C. Zeldin , J. Foley , S. M. Goldsworthy , M. E. Cook , J. E. Boyle , J. Ma , C. R. Moomaw , K. B. Tomer , C. Steenbergen and S. Wu , Mol. Pharmacol., 1997, 51 , 931 CrossRef CAS .
14. K.Nakamura, M.Hirayama, S.Ito and S.Ohtsuki, Poster presented at the 19th ISSX conference in San Francisco, 2014.
15. T. Shimada , H. Yamazaki , M. Mimura , Y. Inui and F. P. Guengerich , Pharmacol. Exp. Ther., 1994, 270 , 414 CrossRef CAS .
16. H. Kawakami , S. Ohtsuki , J. Kamiie , T. Suzuki , T. Abe and T. Terasaki , J. Pharm. Sci., 2011, 100 , 341 CrossRef CAS PubMed .
17. A. Vildhede , J. R. Wisniewski , A. Noren , M. Karlgren and P. Artursson , J. Proteome Res., 2015, 14 , 3305 CrossRef CAS PubMed .
18. J. M. Lasker , M. R. Wester , E. Aramsombatdee and J. L. Raucy , Arch. Biochem. Biophys., 1998, 353 , 16 CrossRef CAS PubMed .
19. M. Ingelman-Sundberg Naunyn-Schmiedeberg’s Arch. Pharmacol., 2004, 369 , 89 CrossRef CAS PubMed .
20. S. C. Sim , R. J. Edwards , A. R. Boobis and M. Ingelman-Sundberg , Pharmacogenet. Genomics, 2005, 15 , 625 CrossRef PubMed .
21. E. Hedlund , J. A. Gustafsson and M. Warner , Curr. Drug Metab., 2001, 2 , 245 CrossRef CAS PubMed .
22. M. Karlgren , M. Backlund , I. Johansson , M. Oscarson and M. Ingelman-Sundberg , Biochem. Biophys. Res. Commun., 2004, 315 , 679 CrossRef CAS PubMed .
23. S. Miksys and R. F. Tyndale , J. Psychiatry Neurosci., 2013, 38 , 152 —163 CrossRef PubMed .
24. S. L. Miksys and R. F. Tyndale , J. Psychiatry Neurosci., 2002, 27 , 406 Search PubMed .
25. V. Ravindranath and H. W. Strobel , Expert Opin. Drug Metab. Toxicol., 2013, 9 , 551 CrossRef CAS PubMed .
26. K. M. Knights , A. Rowland and J. O. Miners , Br. J. Clin. Pharmacol., 2013, 76 , 587 CrossRef CAS .
27. K. R. Chaudhary , S. N. Batchu and J. M. Seubert , IUBMB Life, 2009, 61 , 954 CrossRef CAS PubMed .
28. N. Ahmad and H. Mukhtar , J. Invest. Dermatol., 2004, 123 , 417 CrossRef CAS PubMed .
29. L. Du , S. M. Hoffman and D. S. Keeney , Toxicol. Appl. Pharmacol., 2004, 195 , 278 CrossRef CAS PubMed .
30. X. Ding and L. S. Kaminsky , Annu. Rev. Pharmacol. Toxicol., 2003, 43 , 149 CrossRef CAS PubMed .
31. W. E. Evans and M. V. Relling , Science, 1999, 286 , 487 CrossRef CAS PubMed .
32. L. C. Wienkers and T. G. Heath , Nat. Rev. Drug Discovery, 2005, 4 , 825 CrossRef CAS PubMed .
33. C. R. Wolf and G. Smith , Br. Med. Bull., 1999, 55 , 366 CrossRef CAS PubMed .
34. R. J. Bertz and G. R. Granneman , Clin. Pharmacokinet., 1997, 32 , 210 CrossRef CAS PubMed .
35. U. M. Zanger and M. Schwab , Pharmacol. Ther., 2013, 138 , 103 —141 CrossRef CAS PubMed .
36. M. Ingelman-Sundberg , M. Oscarson and R. A. McLellan , Trends Pharmacol. Sci., 1999, 20 , 342 —349 CrossRef CAS PubMed .
37. S. C. Sim and M. Ingelman-Sundberg , Trends Pharmacol. Sci., 2011, 32 , 72 CrossRef CAS PubMed .
38. S. C. Sim , M. Kacevska and M. Ingelman-Sundberg , Pharmacogenomics J., 2013, 13 , 1 CrossRef CAS PubMed .
39. M. Eichelbaum and B. Evert , Clin. Exp. Pharmacol. Physiol., 1996, 23 , 983 CrossRef CAS PubMed .
40. P. Dalen , C. Frengell , M. L. Dahl and F. Sjoqvist , Ther. Drug Monit., 1997, 19 , 543 CrossRef CAS PubMed .
41. G. Koren , J. Cairns , D. Chitayat , A. Gaedigk and S. J. Leeder , Lancet, 2006, 368 , 704 CrossRef .
42. P. Madadi , G. Koren , J. Cairns , D. Chitayat , A. Gaedigk , J. S. Leeder , R. Teitelbaum , T. Karaskov and K. Aleksa , Can. Fam. Physician, 2007, 53 , 33 Search PubMed .
43. P. Madadi , C. J. Ross , M. R. Hayden , B. C. Carleton , A. Gaedigk , J. S. Leeder and G. Koren , Clin. Pharmacol. Ther., 2009, 85 , 31 CrossRef CAS PubMed .
44. H. Furuya , P. Fernandez-Salguero , W. Gregory , H. Taber , A. Steward , F. J. Gonzalez and J. R. Idle , Pharmacogenetics, 1995, 5 , 389 CrossRef CAS PubMed .
45. A. K. Daly Arch. Toxicol., 2013, 87 , 407 CrossRef CAS PubMed .
46. J. L. Mega , T. Simon , J. P. Collet , J. L. Anderson , E. M. Antman , K. Bliden , C. P. Cannon , N. Danchin , B. Giusti , P. Gurbel , B. D. Horne , J. S. Hulot , A. Kastrati , G. Montalescot , F. J. Neumann , L. Shen , D. Sibbing , P. G. Steg , D. Trenk , S. D. Wiviott and M. S. Sabatine , JAMA, 2010, 304 , 1821 CrossRef CAS PubMed .
47. L. Wallentin , S. James , R. F. Storey , M. Armstrong , B. J. Barratt , J. Horrow , S. Husted , H. Katus , P. G. Steg , S. H. Shah , R. C. Becker and P. Investigators , Lancet, 2010, 376 , 1320 —1328 CrossRef CAS .
48. FDA, Label and Approval History–Coumadin, 2015.
49. FDA, Label and Approval History–Plavix, 2015.
50. G. Englund , P. Hallberg , P. Artursson , K. Michaelsson and H. Melhus , BMC Med., 2004, 2 , 8 CrossRef PubMed .
51. A. Seelig Eur. J. Biochem., 1998, 251 , 252 CrossRef CAS .
52. A. Seelig Int. J. Clin. Pharmacol. Ther., 1998, 36 , 50 CrossRef CAS .
53. V. K. Gombar , J. W. Polli , J. E. Humphreys , S. A. Wring and C. S. Serabjit-Singh , J. Pharm. Sci., 2004, 93 , 957 CrossRef CAS PubMed .
54. R. Didziapetris , P. Japertas , A. Avdeef and A. Petrauskas , J. Drug Targeting, 2003, 11 , 391 CrossRef CAS PubMed .
55. A. H. Schinkel and J. W. Jonker , Adv. Drug Delivery Rev., 2003, 55 , 3 CrossRef CAS PubMed .
56. E. Gatlik-Landwojtowicz , P. Aanismaa and A. Seelig , Biochemistry, 2006, 45 , 3020 CrossRef CAS PubMed .
57. L. Homolya , Z. Hollo , U. A. Germann , I. Pastan , M. M. Gottesman and B. Sarkadi , J. Biol. Chem., 1993, 268 , 21493 —21496 CrossRef CAS .
58. A. B. Shapiro , A. B. Corder and V. Ling , Eur. J. Biochem., 1997, 250 , 115 CrossRef CAS .
59. B. L. Shneider Pediatr. Transplant., 2004, 8 , 609 CrossRef PubMed .
60. P. Matsson , J. M. Pedersen , U. Norinder , C. A. S. Bergström and P. Artursson , Pharm. Res., 2009, 26 , 1816 CrossRef CAS PubMed .
61. J. M. Pedersen , P. Matsson , C. A. S. Bergström , J. Hoogstraate , A. Noren , E. L. LeCluyse and P. Artursson , Toxicol. Sci., 2013, 136 , 328 CrossRef CAS PubMed .
62. M. Karlgren , A. Vildhede , U. Norinder , J. R. Wisniewski , E. Kimoto , Y. Lai , U. Haglund and P. Artursson , J. Med. Chem., 2012, 55 , 4740 CrossRef CAS PubMed .
63. M. Karlgren , G. Ahlin , C. A. S. Bergström , R. Svensson , J. Palm and P. Artursson , Pharm. Res., 2012, 29 , 411 CrossRef CAS PubMed .
64. T. De Bruyn , G. J. van Westen , A. P. Ijzerman , B. Stieger , P. de Witte , P. F. Augustijns and P. P. Annaert , Mol. Pharmacol., 2013, 83 , 1257 CrossRef CAS PubMed .
65. L. Liu , Y. Cui , A. Y. Chung , Y. Shitara , Y. Sugiyama , D. Keppler and K. S. Pang , J. Pharmacol. Exp. Ther., 2006, 318 , 395 CrossRef CAS PubMed .
66. G. Ahlin , J. Karlsson , J. M. Pedersen , L. Gustavsson , R. Larsson , P. Matsson , U. Norinder , C. A. S. Bergström and P. Artursson , J. Med. Chem., 2008, 51 , 5932 CrossRef CAS PubMed .
67. D. Bednarczyk , S. Ekins , J. H. Wikel and S. H. Wright , Mol. Pharmacol., 2003, 63 , 489 CrossRef CAS PubMed .
68. H. Koepsell , K. Lips and C. Volk , Pharm. Res., 2007, 24 , 1227 CrossRef CAS PubMed .
69. J. W. Jonker and A. H. Schinkel , J. Pharmacol. Exp. Ther., 2004, 308 , 2 CrossRef CAS PubMed .
70. G. Englund , F. Rorsman , A. Ronnblom , U. Karlbom , L. Lazorova , J. Grasjo , A. Kindmark and P. Artursson , Eur. J. Pharm. Sci., 2006, 29 , 269 —277 CrossRef CAS PubMed .
71. C. Hilgendorf , G. Ahlin , A. Seithel , P. Artursson , A. L. Ungell and J. Karlsson , Drug Metab. Dispos., 2007, 35 , 1333 CrossRef CAS PubMed .
72. K. M. Giacomini , S. M. Huang , D. J. Tweedie , L. Z. Benet , K. L. Brouwer , X. Chu , A. Dahlin , R. Evers , V. Fischer , K. M. Hillgren , K. A. Hoffmaster , T. Ishikawa , D. Keppler , R. B. Kim , C. A. Lee , M. Niemi , J. W. Polli , Y. Sugiyama , P. W. Swaan , J. A. Ware , S. H. Wright , S. W. Yee , M. J. Zamek-Gliszczynski and L. Zhang , Nat. Rev. Drug Discovery, 2010, 9 , 215 CrossRef CAS PubMed .
73. C. Groer , S. Bruck , Y. Lai , A. Paulick , A. Busemann , C. D. Heidecke , W. Siegmund and S. Oswald , J. Pharm. Biomed. Anal., 2013, 85 , 253 CrossRef CAS PubMed .
74. S. Ekins , G. Bravi , J. H. Wikel and S. A. Wrighton , J. Pharmacol. Exp. Ther., 1999, 291 , 424 CrossRef CAS .
75. U. Norinder SAR QSAR Environ. Res., 2005, 16 , 1 CrossRef CAS PubMed .
76. L. Afzelius , I. Zamora , C. M. Masimirembwa , A. Karlen , T. B. Andersson , S. Mecucci , M. Baroni and G. Cruciani , J. Med. Chem., 2004, 47 , 907 CrossRef CAS PubMed .
77. G. Cruciani , E. Carosati , B. De Boeck , K. Ethirajulu , C. Mackie , T. Howe and R. Vianello , J. Med. Chem., 2005, 48 , 6970 CrossRef CAS PubMed .
78. I. Zamora , L. Afzelius and G. Cruciani , J. Med. Chem., 2003, 46 , 2313 CrossRef CAS PubMed .
79. L. Olsen , P. Rydberg , T. H. Rod and U. Ryde , J. Med. Chem., 2006, 49 , 6489 CrossRef CAS PubMed .
80. P. Rydberg , D. E. Gloriam , J. Zaretzki , C. Breneman and L. Olsen , ACS Med. Chem. Lett., 2010, 1 , 96 CrossRef CAS PubMed .
81. P. Rydberg and L. Olsen , ChemMedChem, 2012, 7 , 1202 CrossRef CAS PubMed .
82. P. Rydberg and L. Olsen , ACS Med. Chem. Lett., 2012, 3 , 69 CrossRef CAS PubMed .
83. D. F. Lewis Chem.-Biol. Interact., 1987, 62 , 271 CrossRef CAS PubMed .
84. M. T. Cronin and T. W. Schultz , J. Mol. Struct., 2003, 622 , 39 CrossRef CAS .
85. M. P. Gleeson , A. M. Davis , K. K. Chohan , S. W. Paine , S. Boyer , C. L. Gavaghan , C. H. Arnby , C. Kankkonen and N. Albertson , J. Comput.-Aided Mol. Des., 2007, 21 , 559 —573 CrossRef CAS PubMed .
86. I. Choi , S. Y. Kim , H. Kim , N. S. Kang , M. A. Bae , S. E. Yoo , J. Jung and K. T. No , Eur. J. Med. Chem., 2009, 44 , 2354 CrossRef CAS PubMed .
87. S. Ekins , J. Berbaum and R. K. Harrison , Drug Metab. Dispos., 2003, 31 , 1077 CrossRef CAS PubMed .
88. B. C. Jones , G. Hawksworth , V. A. Horne , A. Newlands , J. Morsman , M. S. Tute and D. A. Smith , Drug Metab. Dispos., 1996, 24 , 260 CrossRef CAS .
89. A. Mancy , P. Broto , S. Dijols , P. M. Dansette and D. Mansuy , Biochemistry, 1995, 34 , 10365 CrossRef CAS PubMed .
90. M. J. de Groot , A. A. Alex and B. C. Jones , J. Med. Chem., 2002, 45 , 1983 CrossRef CAS PubMed .
91. M. Ramesh and P. V. Bharatam , J. Mol. Model., 2012, 18 , 709 CrossRef CAS PubMed .
92. M. J. de Groot , F. Wakenhut , G. Whitlock and R. Hyland , Drug Discovery Today, 2009, 14 , 964 CrossRef PubMed .
93. S. Sciabola , I. Morao and M. J. de Groot , J. Chem. Inf. Model., 2007, 47 , 76 CrossRef CAS PubMed .
94. N. Manga , J. C. Duffy , P. H. Rowe and M. T. Cronin , SAR QSAR Environ. Res., 2005, 16 , 43 CrossRef CAS PubMed .
95. S. Ekins , G. Bravi , S. Binkley , J. S. Gillespie , B. J. Ring , J. H. Wikel and S. A. Wrighton , Drug Metab. Dispos., 2000, 28 , 994 CrossRef CAS .
96. J. A. Goldstein and S. M. de Morais , Pharmacogenetics, 1994, 4 , 285 CrossRef CAS PubMed .
97. J. Kirchmair , M. J. Williamson , J. D. Tyzack , L. Tan , P. J. Bond , A. Bender and R. C. Glen , J. Chem. Inf. Model., 2012, 52 , 617 CrossRef CAS PubMed .
98. T. A. Clarke and L. A. Waskell , Drug Metab. Dispos., 2003, 31 , 53 CrossRef CAS .
99. P. Savi , J. M. Pereillo , M. F. Uzabiaga , J. Combalbert , C. Picard , J. P. Maffrand , M. Pascal and J. M. Herbert , Thromb. Haemostasis, 2000, 84 , 891 CrossRef CAS .

© The Royal Society of Chemistry 2016 (2015)