You are here:
Developing a Predictive Model for Chemical Excretion in Urine (SOT 2023)
Citation:
Stanfield, Z., Woody Setzer, J. Sobus, AND J. Wambaugh. Developing a Predictive Model for Chemical Excretion in Urine (SOT 2023). SOT, Nashville, TN, March 19 - 23, 2023. https://doi.org/10.23645/epacomptox.22280641
Impact/Purpose:
N/A
Description:
Pharmacokinetic (PK) models encode physical, chemical, and biological processes that govern chemical adsorption, distribution, metabolism, and elimination. Many man-made chemical substances are eliminated in urine following phase I and II metabolic processes. Urinary biomarker measurements are therefore often used to evaluate PK model predictions and refine model parameters. Most PK models assume that urinary elimination occurs via glomerular filtration. Concentrations of chemicals excreted via glomerular filtration are directly affected by changing urine output (ml/min). As such, models must directly account for changing flow rates (using a defined model variable) or adjust urinary biomarker concentrations to negate the effects of changing urine flow. Urinary creatinine concentration is commonly used for this adjustment, as it is assumed inversely proportional to urine output. Recent work has highlighted two potential limitations of using creatinine correction as a default correction approach. First, creatinine excretion is not perfectly constant for any given person, making creatinine concentration not exactly (inversely) proportional to urine output. Second, some chemicals are excreted (at least in part) by passive diffusion rather than glomerular filtration; urinary concentrations of these chemicals are independent of changing urine output and therefore should not be corrected using creatinine concentrations. Here, we examined the correlation of creatinine concentration and urine output. We used metabolite concentrations from all individuals across 9 cohorts of the CDC’s National Health and Nutrition Examination Survey (NHANES). Based on these correlations, chemicals were clustered into 3 urine elimination groups (glomerular filtration, passive diffusion, and an intermediate class). This cluster labeling was used to build a training set for a random forest model that was constructed using molecular descriptors (Mordred) for 145 of the NHANES metabolites. The preliminary model achieved an out-of-bag (OOB) error of 34.48% with clear distinction between the two primary elimination routes, glomerular filtration and passive diffusion, and strongly outperformed a y-randomized (null) model (OOB error of around 75%). The model was then used to classify the expected primary route of elimination for multiple chemical lists of interest, including chemicals occurring in consumer products and chemicals in the Toxic Substances Control Act (TSCA) active list. Additionally, the majority of the chemicals in these lists (about 80% on average) were shown to fall within the domain of applicability of the model. Predictions for a subset of chemicals were also validated through literature sources and metabolism databases. This model can serve as a high-throughput, first-pass method to determine how to handle urine concentrations of various chemicals in pharmacokinetic modeling, which may improve our ability to estimate chemical risk. As there is currently no broadly accepted guidance on how to treat urine data, this model may also serve as a first step toward establishing such guidance. This abstract does not reflect U.S. EPA policy.
URLs/Downloads:
DOI: Developing a Predictive Model for Chemical Excretion in Urine (SOT 2023)
POSTER.PDF (PDF, NA pp, 719.971 KB, about PDF)