Prediction of transport, pharmacokinetics, and effect of drugs

Prediction and modulation of pharmacokinetics and the effects of drugs is a major concern in drug discovery, drug safety, and clinical practice. The projects in this thesis span the entire range of levels upon which these issues can be explored, ranging from in silico to in vivo evaluations and molecular, cellular, organ, and systemic phenomena. Drug discovery has come a long way, from a mostly serendipitous endeavor to the highly focused process in today’s global pharmaceutical companies, where large numbers of candidate substances are routinely weeded out to arrive at a few promising drug leads.
Effective wet lab screening tools (such as high-throughput screening (HTS)) exist to evaluate compounds that exist physically. Increasingly, however, chemical libraries are designed using combinatorial chemistry – by applying a pre-defined set of permissible reactions to a few scaffold structures, by creating databases of compounds with a certain degree of similarity to members of a successful drug class, or any other such method. These virtual libraries require virtual screening tools. The first set of projects in this thesis presents such tools. They assess pharmacokinetic and toxicological profiles using machine learning methods. The second set of projects deals with the clinical implications of individual pharmacogenetic differences and ways of predicting or circumventing them.
The first study outlines the development of decision tree induction (DTI) models to evaluate interactions with P-glycoprotein (P-gp, MDR1, ABCB1). This efflux pump protein is a major factor in the elimination of xenobiotics from cells. Substrate, inhibitor, and inducer properties are predicted by these models with an accuracy of 77.7 % (substrates), 86.9 % (inhibitors), and 90.3 % (inducers). Furthermore, the study shows the superiority of the DTI algorithms CHAID and CART over the more widely used work-horse algorithm C4.5, and the utility of lipophilic distribution coefficients that take into account differing states of ionization depending on the pH of the environment.
In the second study, interactions with the Cytochrome P450 (CYP) family of enzymes are predicted. These enzymes are involved in phase I metabolism of most drugs, during which compounds are not only detoxified and prepared for elimination but also active metabolites are formed. Because of the promiscuity and cross-selectivity of the different CYP isoforms, they are heavily involved in complications arising from co-administration of multiple drugs or seemingly innocuous natural substances (e.g. St. John’s wort, grapefruit juice). Models were built for substrate, inhibitor, and inducer activities for CYP 1A2, CYP 2D6, and CYP 3A4 using DTI, the k-nearest neighbor algorithm, random forests, artificial neural networks, and support vector machines. Predictive accuracies were very high (81.7 to 91.9 % for CYP 1A2, 89.2 to 92.9 % for CYP 2D6, and 87.4 to 89.9 % for CYP3A4). The commonly held hypothesis that P-gp substrates are CYP 3A4 substrates was evaluated using data from the first study. Both datasets overlap for only 84 compounds and agreement was moderate at 45%.
The third study takes a more global approach and discusses the development of models for broad classes of adverse drug reactions (in the central nervous system, liver, kidney, and allergic potential). Here, the adverse drug reactions of over 500 drugs were determined from the drug register of Switzerland and classified according to the categories mentioned above. The resulting models perform very well (88.0 to 89.7 % for the CNS, 87.1 to 90.2 % for the liver, 84.7 to 88.6 % for the kidney, and 78.4 to 78.9 % for allergic potential) and can serve as valuable tools not only in early drug discovery but also in pharmacovigilance. A comparison with the data on P-gp obtained in the first study of this thesis indicates that compounds extruded by P-gp do not cause less CNS related adverse drug reactions.
The fourth project deals with the potential consequences of dysfunction in the efflux pumps P-gp and breast cancer resistance protein (BCRP, ABCG2). Several genetic polymorphisms are known for these proteins, some of which alter phenotypes. Because both proteins are expressed at crucial barriers such as the luminal wall of the intestine or the blood brain barrier, decrease in activities is thought to influence disease susceptibility and severity in inflammatory bowel disease (IBD, i.e. Crohn’s disease and ulcerative colitis). Furthermore, they influence success of pharmacotherapy because many drugs used in the treatment of IBD are transported by or alter activity of these pumps. Genetic constitution also influences activity and knowledge the individual pharmacogenomic profile could help predict response to pharmacotherapy. In this study, the prevalence of two single nucleotide polymorphisms (SNPs) of P-gp (C3435T, G2677T) and one SNP of BCRP (C421A) were assessed in peripheral blood in healthy volunteers and patients newly diagnosed with IBD, both from the Swiss population. All three SNPs are known to result in a decrease in activity. The rationale behind this study was therefore that such constitutional changes in efflux pump activities may predict disease susceptibility for IBD. While no statistically significant results could be obtained, there are discernible trends towards BCRP 421A (p < 0.18), MDR1 2677T (p < 0.27), and the wild type allele MDR1 C3435 (p < 0.46) in patients with ulcerative colitis. Also in these patients, the haplotypes MDR1 3435CC / BCRP 421CC (Χ2: 1.0142, p < 0.30) and MDR1 2677G / BCRP 421A (Χ2: 1.5615, p < 0.22) were more prevalent, although, again, with no statistical significance. IBDs are complex multifactorial diseases and any single SNP is unlikely to serve in reliably predicting susceptibility. However, the study showed the promise of these SNPs and haplotypes derived from them, and a repeat in a larger sample from the Swiss population may reveal stronger associations. Also, because the genotype analysis was performed in peripheral blood (obtained by venipuncture instead of more costly and complex intestinal biopsies), such a study may yield an easily applicable tool for predicting response to pharmacotherapy with P-gp and BCRP substrates.
The fifth study deals with a rare genetic variation in the phase II metabolizing enzyme uridine 5'-diphospho-glucuronosyltransferase 1A1 (UGT 1A1) and the ensuing Crigler-Najjar syndrome type II (CN II) in a Caucasian patient. The patient was suffering from the consequences of hyperbilirubinaemia, which are essentially cosmetic in nature. CN II patients still have residual UGT 1A1 activity (compared to CN I, which, prior to the availability of phototherapy often resulted in neonatal death due to kernicterus). Currently, patients are treated with phenobarbital which binds to the phenobarbital-responsive enhancer module of UGT 1A1. Phenobarbital is a barbiturate with a small therapeutic window and many adverse effects such as drowsiness, impaired motor function, sexual dysfunction, and dependency. Less toxic alternatives are therefore desirable. The residual activity in CN II could be enhanced in a variety of other ways, for example by translational activation via the pregnane X receptor (PXR), the glucocorticoid receptor (GR), or the aryl hydrocarbon receptor (AhR). It is therefore possible to predict that since some degree of UGT 1A1 activity is present in the patient, translational activation can increase UGT-1A1-mediated metabolism. This study tested this prediction for the PXR activator hyperforin (one of the constituents of St. John’s Wort (SJW)). SJW is available in standardized extracts, generally well tolerated (photo-toxicity, fatigue, and gastrointestinal discomfort are the most common adverse reactions), and cost-efficient. The study showed a significant decrease in plasma bilirubin levels in the patient and suggests that SJW is an interesting therapeutic alternative.
The sixth project is a proof-of-principle study showing the feasibility of transdermal pulsatile administration of nicotine in heavy male smokers. Nicotine substitution is an important therapeutic tool in weaning patients from tobacco and it is predicted that mimicking the pharmacokinetic profile of cigarette smoking (short bursts of nicotine exposure resulting in acute peaks in plasma and fast elimination) may give better outcomes (abstinence from tobacco) than conventional modes of administration (e.g. reservoirs in transdermal patches which give long-lasting nicotine plasma levels or nicotine gum which, while giving the patient more control over release than patches, still continuously releases fractions of its content). For the study, volunteers were subjected to three increasing doses of nicotine from a prototype computer controlled device and showed statistically significant peaks in nicotine plasma levels for the highest dose. Adverse events, esp. on the skin, were minimal and receded shortly after the device was removed. The device is a promising way of nicotine substitution and could also be applied to other settings where transdermal patient-controlled delivery is desirable (e.g. opioids in pain control).
In an isolated seventh project, a fully automated system was established for a general unknown screening for toxic substances in serum and urine using liquid chromatography / mass spectrometry technology. A library of over 350 compounds along with spectra is presented as well as a computer program which uses this library to identify substances on-line.