Mechanistic Model-based Drug Oral Absorption Analysis

Abstract

The thesis research consists of two projects: the major focus is using population approach to account for the systemic availability variability of Mycophenolate Mofetil (MMF) in human and providing insights into developing a predictive bioequivalence (BE) test. A separate small project on text mining via natural language processing (NLP) is included for drug Biopharmaceutics Classification System (BCS) classification.

The prodrug MMF, which is pre-systemically hydrolyzed into the pharmacologically active compound Mycophenolic Acid (MPA), has been widely used for the prophylaxis of acute allograft rejection in solid organ transplantation. However, the huge variability in plasma level makes MMF development difficult due to the great challenge of meeting the traditional BE limits. Numerous models have been developed in the past decade to explain the variability with the emphasis on characterizing the enterohepatic circulation (EHC), while the variability arising from absorption, can also contribute to the remarkable MPA variability to a great extent, but has been ignored for this BCS class 2 drug.

Two population pharmacokinetics (PK) models of MMF focusing on the absorption process were developed based on the plasma concentrations of MPA and its major metabolite MPAG in a long-term MMF treatment on liver transplant patients. The MPA PK profiles were best characterized by a two-compartment disposition model with zero inter-individual variability (IIV) of elimination (K20), lag time (Tlag) but considerable inter-occasion variability (IOV) regarding systemic appearance rate (Ka), K20 and volume of distribution (V2). The second model took into consideration the EHC by including MPAG profiles as well. The results from both models showcased that the within-subject variability (WSV) of the MMF’s systemic appearance played a much more significant role than the IIV. The large WSV can be mechanistically explained by the gastrointestinal (GI) physiological dynamics, especially gastric emptying (GE) in the fasted state regulated by migrating motor complex (MMC) and in the fed state by the caloric content with irregular patterns of GI motility and secretion. The results implied that dosing under fed conditions was recommended for the in vivo clinical BE study of MMF to reduce the WSV and that developing a predictive in vitro dissolution test with sufficient simulation of the GI physiological dynamics would be a good surrogate.

The second project explored the application of NLP in drug BCS classification. NLP, a confluency of artificial intelligence and computational linguistics, has gained widespread popularity in tech companies for machine translation, chatbot system, etc. In biotech and pharmaceutical industry, NLP-based text mining has been utilized to transform text information for decision support in multiple areas, including gene disease mapping, biomarker discovery, drug-drug interaction, and pharmacovigilance.

The BCS system, designed to recommend a waiver of in vivo bioavailability and BE studies for immediate-release (IR) solid oral drug products, classifies drugs based on their aqueous solubility under physiological pHs and intestinal permeability. However, there’s no complete summarization of drug BCS classification information published to date. This project extracted solubility, permeability and/or BCS class information of IR solid oral dosage forms in the human GI tract or simulated in vitro experiments from the FDA orange book, drug labels, FDA review documents, along with some selected literatures, and identified drug BCS class via NLP technology. The text mining results can be one of the key components in building up a database containing drug oral absorption information for drug discovery and development.