2.7. Metabolism Models
Mammalian metabolism of drugs is classically divided into two stages: Phase I metabolism involves oxidation and Phase II involves conjugation. Cytochrome P450s (CYPs) account for the majority of first-pass drug transformations taking place in the intestinal epithelium and in the liver (Williams 2004; Wienkers 2005; Guengerich 2008), but metabolism by cytosolic aldehyde oxidase (AOX) can also be important for some nitrogen heterocycles (Montefiori 2017). UDP-glucopyranosyl transferases (UGTs), on the other hand, are among the most important Phase II metabolic enzymes. They glucuronidate hydroxy groups, carboxylates, amines, heterocyclic nitrogens and other nucleophilic centers in xenobiotics.
Such polar groups generally interfere with intestinal absorption. Hence most glucuronidation of drugs takes place at sites that were introduced or unmasked by Phase I oxidation - hence the term "Phase II." Non-drug xenobiotics like pesticides, on the other hand, are often good UGT substrates even without first being activated by Phase I enzymes.
Another kind of important xenobiotic "unmasking" involves hydrolytic breakdown, which often takes place before a compound is exposed to Phase I and Phase II enzymes. Such "Phase 0" metabolism can occur within the gut lumen, in the bloodstream, in the cytosol and in the endoplasmic reticulum. The enzymes responsible - hydrolases, peptidases, carboxylesterases and alkaline phosphatase, among others - are remarkably myopic: they are quite particular as regards to the immediate molecular environment of the bond that they cleave but generally don't "see" molecular features more than a few bonds away.
Knowing that a compound is subject to metabolism by one or more of these enzymes is key to understanding its bioavailability and, if it is a prodrug, its bioactivation. Knowing which metabolites are most likely is critical to understanding the toxicology, efficacy, and pharmacokinetic properties of drug and agrochemical candidates. Sometimes all three sets of enzymes work together, with hydroxylated or dealkylated CYP products of cleaved esters serving as substrates for one or more UGTs.
Scope
The Metabolism Module provides substrate classification models for nine human UGT isoforms. Much more data is available for CYPs, and correspondingly more detailed models are available, including models for estimating kinetic parameters as well as ones for substrate and inhibitor classification.
Substrate classification and atomic site of metabolism models cover nine CYP isoforms that, taken together, account for over 90% of CYP-mediated reactions in humans: 1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1, and 3A4 (Guengerich 2008). Estimates for the Michaelis-Menten constants Km and Vmax are available for oxidation in recombinant CYP (rCYP) assay systems for isoforms 1A2, 2C9, 2C19, 2D6 and 3A4. Kinetic parameters are estimated for the molecule as a whole, which is essential for estimating overall metabolism, but they are also estimated for each predicted site of metabolism. Having site-specific kinetic information makes it possible to predict not only which metabolites are likely to be generated but also which are likely to dominate the oxidative metabolic profile.
The rCYP3A4 models are complemented by models based on CYP3A4 assay data for human liver microsomes (HLMs). Each pair of Michaelis-Menten parameter predictions comes with a matching intrinsic clearance (CLint) prediction that has been scaled to reflect the average abundance of the corresponding isoform in human liver. There is also a model for overall intrinsic clearance by HLM CYPs (CYP_HLM_CLint).
Identification of compounds that are likely to inhibit CYPs is critical to anticipating drug-drug interactions and potential metabolic complications. Classification models for inhibition of CYPs 1A2, 2C9, 2C19, 2D6 and 3A4 are provided, along with predictions specifically related to inhibition of CYP3A4's metabolism of two of its archetypal substrates, midazolam and testosterone.
Substrate classification models are provided for human UGTs 1A1, 1A3, 1A4, 1A6, 1A8, 1A9, 1A10, 2B7 and 2B15. Unfortunately, the combination of complex kinetics (Hyland 2009) and scarcity of suitable data means that we are currently unavailable to provide kinetic models for these enzymes. As of ADMET Predictor-X, however, we have added the ability to generate the metabolites most likely to be generated for each UGT for which a compound is predicted to be a substrate.
Hydrolytic enzyme kinetics are usually simpler, but good kinetic data is still limited. The combination of enzymatic diversity and substrate overlap also hinders quantitative modeling. Nonetheless, the focus of these enzymes on specific substructures combined with the extensive ester prodrug literature (Rautio 2018) has made it possible to construct query-based models that predict which metabolites will be generated by hydrolytic metabolism.
Caveats
The absence of a predicted clearance rate for a metabolic transformation should not be taken as a prediction that the rate is negligible. Despite being hard to quantitate, clearance by aldehyde oxidase and UGTs are often substantial.
The same is true for the four "minor" CYPs for which we do not make kinetic predictions. Spot checks on compounds in the CYP_HLM_CLint training pool for which appreciable clearance rates have been reported but which are predicted to not be a substrate for any of the "big five" CYPs has shown that they are, in fact, substrates for minor isoforms. In the few cases where oxidatively metabolized compounds are predicted to not be a substrate for any of the nine, some other CYP was responsible most of the time.
Also bear in mind, however, that the CYP_HLM_CLint is of necessity only trained on things that are substrates for some CYP. Unlike the recombinant CYP models, this one does not have an associated substrate/nonsubstrate classification filter. It is difficult to gather enough of the negative (i.e., nonsubstrate) instances needed to construct such a model because the absence of CYP metabolism is very rarely reported as such.