Molecular determinants of corticosteroid and anti-cancer drug pharmacokinetics and pharmacodynamics
This thesis dissertation seeks to identify the molecular determinants of the pharmacokinetics (PK) and pharmacodynamics (PD) of corticosteroids and selected anti-cancer agents (erbB2 antagonists and camptothecin analogues) using both empirical and mechanistic modeling of quantitative structure-property relationships (QSPR). Several contemporary algorithms and methods were utilized, and in addition, a novel approach for directly predicting the time-course of plasma drug concentrations from molecular descriptors was developed. Empirical QSPR models . Empirical QSPR models are largely statistical in nature and aim to correlate molecular properties of compounds with their PK/PD properties. Such models can be useful in drug discovery and development; however, they can be limited owing to a focus on errorless point estimates of PK/PD parameters and the disregard of mechanisms underlying physiological and pharmacological processes. Several limitations of empirical QSPR models are addressed in this thesis to improve their predictive performance. Chapter 1 highlights the multidimensional nature of QSPR data sets and the challenges of selecting training and test or validation data sets for model development. A new visualization tool developed for pharmacogenomic data was evaluated for use in data splitting of QSPR data sets. In Chapter 2, a traditional partial least squares regression technique was used to establish QSPR models for a congeneric series of camptothecin anti-cancer agents. Understanding the PK/PD properties of these drugs is complicated by a pH-dependent hydrolysis of their lactone structure to an inactive form. The construction of several QSPR models for multiple PK/PD properties of camptothecins provided a unique perspective for suggesting favorable physicochemical properties. For example, drug lipophilicity emerged as a critical descriptor influencing net systemic exposure and interactions with known efflux pumps (e.g., breast cancer resistance protein). Chapter 3 describes a new approach for the direct prediction of in vivo disposition profiles from molecular descriptors. A time-dependent neural network model was developed to map the relationships between drug dose and structure to the PK profiles of erbB2 antagonists in rat. To our knowledge, this is the first time this has been reported without specifying a physiologically-based structural model. Mechanism-based QSPR models . Whole-body physiologically based pharmacokinetic (PBPK) models emulate the physiological processes controlling the disposition of drugs. PBPK models have been coupled with in silico predicted tissue:plasma partition coefficients and in vitro measured intrinsic clearances for simulation of in vivo drug disposition. In preparation for in vitro metabolism experiments, a modified LC/MS assay was developed in Chapter 4 with improved sensitivity over standard UV detection assays. A QSPR model was developed for the intrinsic clearances of corticosteroids in rat liver microsomes in Chapter 5, and this relationship was incorporated into a mechanistic PBPK model. Predicted PK profiles were then integrated into a PD model of lymphocyte trafficking for anticipating corticosteroid responses in humans (Chapter 6). This QSPR-PBPK-PD approach represents a useful scheme, coupling empirical and mechanistic PK/PD models, for the a priori prediction of drug disposition and dynamics. Conclusions . Both empirical and mechanistic QSPR models were evaluated for corticosteroids and selected anti-cancer agents in this dissertation. In addition to applying a number of contemporary modeling approaches, a new empirical approach was developed for predicting the time-course of drug disposition directly from molecular properties. Integration of both empirical and mechanistic approaches can improve understanding of the molecular properties and physiological factors controlling the intensity and time-course of pharmacological effects.