ADME is an abbreviation of "absorption, distribution, metabolism, and excretion", and is a pillar of pharmacokinetics in pharmacology. Together, these characteristics affect a drug’s performance and dosage, and are therefore of utmost importance in drug design. Absorption describes a drug’s ability to traverse from the gastrointestinal tract into the bloodstream after oral ingestion; poor absorption means a drug might require intravenous administration, which is less desirable. Distribution encompasses the drug’s dispersal across different bodily tissues, while metabolism is the transformation of the original drug into “metabolites” catalyzed by enzymes. If the metabolites are inactive, a high rate of metabolic conversion would lower a drug’s performance. Finally, excretion is the drug’s ultimate fate as it is eliminated from the body; for nontoxic drugs, a longer elimination half-life is better. The ADME concept can also be expanded to include a drug’s toxicity, ADMET or ADME-Tox, such as liver, cardiac, and acute toxicity.
As illustrated briefly in the introduction, ADME profoundly affects a drug’s properties and has long been a cornerstone of pharmacology and drug discovery. Recent years, however, have seen an evolution in ADME concepts. An impetus for precision medicine has fostered personalized ADME approaches through pharmacogenomics, the role the genome plays in drug response, and even through chronobiology, the role of circadian rhythms on drug efficacy and toxicity. Like many fields, ADME is also embracing computational approaches to optimize a drug’s pharmacological properties. The recent surge in biologics, such as antibodies, is also stimulating research into ADME specific to biologics.
ADME for personalized medicine
Pharmacogenomics is the study of inter-individual variation to drug response through gene copy-number variants, i.e., variability in the number of gene copies, or single nucleotide variants (SNPs), i.e., differences in gene nucleotide sequence at a single base that translate to differences in protein expression or function. Specifically, in the context of ADME, pharmacogenomics examines genetic variants in drug metabolizing enzymes, such as cytochrome P450s (CYPs), and drug transporters. CYPs oxidize up to 70 to 80% of drugs in clinical use and can render them inactive. SNPs to CYPs can affect how fast the enzymes metabolically deactivate drugs or cause adverse reactions in drug-drug interactions.
“Previously, most of the focus was on known common genetic variants to CYPs, which metabolize pharmaceuticals, or drug transporters, such as efflux transporters, that prevent drugs from penetrating certain tissues or actively pump drugs out of their target cells,” explained Volker M. Lauschke, Associate Professor in Personalized Medicine and Drug Development, Director of Micro- and Nanofabrication Core facility at the Karolinska Institutet in Stockholm, Sweden. He and his lab research how an individual’s or a population’s genetic makeup affects their ADME response to drugs. “We are using next generation sequencing (NGS) to examine all variations in genes that participate in ADME in individuals and populations, not just the common known SNPs. What we’re discovering is fascinating. We’re finding that rare SNPs account for over 99% of the number of genetic variants and explain 20-40% of inter-individual variation.” These findings have important clinical implications and suggest that testing patients for common SNPs may not suffice. Rather, a patient’s genome could be profiled by NGS for all SNPs, including rare ones, to determine how they will affect the individual’s response to a drug and adjust its administration, e.g., raise or lower dosage or select a therapeutic alternative, in a personalized approach to the patient’s treatment.
When asked about future prospects for ADME pharmacogenomics, Lauschke answered: “We need randomized controlled clinical trials that are highly powered to determine the added value of personalized ADME using NGS. We are participating in the Ubiquitous Pharmacogenomics (U-PGx) trial consortium, which was launched in 2016 across several European countries and institutions, to evaluate the cost-benefit of pharmacogenetically-guided prescribing. It’s concluding this year, and we’re anticipating the results shortly.”
Lauschke and his team are also pioneers of in vitro models of “miniature livers”, generated from cells obtained directly from patients, so called primary cells. “The liver is a major metabolic center of drug metabolism and potential toxicity. However, traditional 2D cultures grown as a single layer of cells in a dish are not good mimics of the liver’s complexity. Instead, we are using 3D liver cell cultures, either as “spheroids”, which are liver cells cultured in a roughly spherical mass, or “microfluidic” cultures, which are cells grown in a complex pattern of microchannels.”
Lauschke continued: “Most importantly, it is long established that 2D liver cultures rapidly dedifferentiate and lose the proteomic, transcriptomic, microRNA, and metabolomic profile of mature, differentiated liver cells, which limits their utility for drug testing. However, when we grow liver cells as 3D spheroids, they retain a differentiated phenotype with molecular signatures consistent with the original donor patient liver tissue.” Indeed, in a multi-center study using liver cells from three donors, hepatotoxicity from six drugs was more consistent across testing centers in 3D than in 2D cultures, demonstrating the suitability of 3D cultures for toxicity testing. The 3D liver spheroid cultures also predicted toxic versus nontoxic compounds in 123 tested drugs with 69% sensitivity and 100% specificity. “We’re very excited about the liver spheroid toxicity studies and are expanding our pharmacological studies into other 3D models of pancreas, adipose, and skeletal muscle tissue,” explained Lauschke.
ADME by artificial intelligence
Computational tools are increasingly becoming an important facet of ADME research. The drug development process suffers from high attrition rates, starting from in vitro screening, to cell culture testing, preclinical animal and safety models, and ultimately human clinical trials. Yet despite the time-consuming and costly process, a promising drug candidate can fail early phase clinical trials due to a lack of appropriate ADME properties and toxicity safety concerns. Therefore, implementing computational tools for ADME-Tox earlier and throughout the drug discovery pipeline can lower the failure rate later down the line. “There is a real need for reliable predictive models to improve drug discovery,” explained John P. Overington, Chief Informatics Officer at Catapult Medicines Discovery, a national facility in the United Kingdom for accelerating drug discovery. “This area of research is really timely, due to combined recent progress in deep learning/artificial intelligence, training data availability and high-performance computing.”
There are numerous types of computation to consider. First, there is the pharmaceutical’s chemical structure itself, which imparts certain molecular properties. These physicochemical properties, e.g., aqueous solubility, lipophilicity (the partitioning from water to fatty tissue), ionization state, etc, affect a molecule’s ADME properties. A drug intended to penetrate the blood–brain barrier and distribute into the brain, e.g., to treat a brain tumor, is a particular challenge. Moreover, it should not be a substrate of efflux transporters, which actively expel drugs from the brain. This highlights another important consideration; the interaction of the drug itself within the human body and with “off-targets”, interactions with proteins or biomolecules other than the drug’s intended target. Aside from its primary target, a drug may bind to efflux transporters (limiting tissue penetration), to metabolizing enzymes (deactivating and clearing the drug), or to targets that cause toxicity.
“Drug discovery is very much a highly complex multiparameter optimization problem, which are derived from both the molecule structure and the genes, cells, tissues and organs of a human,” explained Overington. One method for predicting how a molecule will interact with a target is quantitative structure–activity relationship (QSAR) studies. QSAR modeling begins with a curated dataset of chemicals, which is used to identify “descriptors”, i.e., a molecule’s physicochemical properties, and how they relate to “response-variables”, i.e., a biological activity. These identified relationships between drug structure and biological activity are used to build the QSAR model, which can then predict the properties of drugs with new structures. “We’ve also extended QSAR to a technique called proteochemometric modeling (PCM),” elaborated Overington of his collaborative research. “In this approach, rather than modeling chemical structure to the biological activity of one target, we model the activity to a family of targets.”
Overington was also involved in developing a web-based, and freely-available ADME tool called ADME SARfari. This resource itself used a database called PharmaADME, which contains genes coding for proteins important to ADME properties, such as metabolizing enzymes. The ADME genes from PharmaADME were mapped to a database called ChEMBL, which contains bioactive molecules with drug-like properties. Linking the two databases generated an ADME gene-chemical structure relationship, which can now be used to predict ADME properties of new molecules. The field is rapidly shifting and newer machine learning algorithms will lead to ever better predictions.
“We are still data limited however, so the quest for novel data, covering more of drug-like chemical space will never end,” Overington concluded.