Reconstruction of Exposure to m-Xylene from Human Biomonitoring Data Using PBPK Modelling, Bayesian Inference, and Markov Chain Monte Carlo Simulation

Abstract

There are numerous biomonitoring programs, both recent and ongoing, to evaluate environmental exposure of humans to chemicals. Due to the lack of exposure and kinetic data, the correlation of biomarker levels with exposure concentrations leads to difficulty in utilizing biomonitoring data for biological guidance values. Exposure reconstruction or reverse dosimetry is the retrospective interpretation of external exposure consistent with biomonitoring data. We investigated the integration of physiologically based pharmacokinetic modelling, global sensitivity analysis, Bayesian inference, and Markov chain Monte Carlo simulation to obtain a population estimate of inhalation exposure to m-xylene. We used exhaled breath and venous blood m-xylene and urinary 3-methylhippuric acid measurements from a controlled human volunteer study in order to evaluate the ability of our computational framework to predict known inhalation exposures. We also investigated the importance of model structure and dimensionality with respect to its ability to reconstruct exposure.

Figure 1

(a) Venous blood concentrations of m-xylene. Data from three volunteers were prepared and measured on a different day than other four. This set of data has the expected appearance and was considered acceptable for use in reverse dosimetry. (b) Venous blood concentrations of m-xylene. Data from four volunteers were prepared and measured on a different day than other three. This set of data does not have the expected appearance and was considered unacceptable for use in reverse dosimetry.

Figure 2

Exhaled m-xylene. Data from eight volunteers used in reverse dosimetry. The data points enclosed within the grey bar were excluded from the final exposure reconstruction simulations.

Figure 3

Urinary 3-methylhippuric acid (MHA). Urinary excretion of MHA expressed against creatinine for eight volunteers used in reverse dosimetry.

Figure 4

Schematic of the PBPK model for m-xylene with a bladder compartment, to simulate fluctuations in the concentration of the main metabolite, methylhippuric acid.

Figure 5

A comparison of 5 CV biomarker profiles corresponding to parameter sets sampled from the priors and reliable CV measurements.

Figure 6

Lowry plot of the eFAST quantitative measure. The total effect of a parameter S _Ti comprised the main effect S _i (black bar) and any interactions with other parameters (grey bar) given as a proportion of variance. The ribbon, representing variance due to parameter interactions, is bounded by the cumulative sum of main effects (lower bold line) and the minimum of the cumulative sum of the total effects (upper bold line), (a) CV at 3 hours, (b) CXPPM at 3 hours, (c) C_urine at 5 hours.

Figure 7

Comparison of estimated posterior distributions for 4-hour inhalation exposure to m-xylene. Posterior distributions were estimated by updating the entire set of parameters, most influential parameters, or by fixing the measured parameters and updating the remaining most influential: (a) C_urine, full parameter set, (b) C_urine, most influential, (c) C_urine, most influential and measured spot urine production rates and creatinine concentrations.

Figure 8

Comparison of estimated posterior distributions for 4-hour inhalation exposure to m-xylene. Posterior distributions were estimated by updating the most influential parameters or by fixing the measured parameters and updating the remaining most influential: (a) CV, most influential, (b) CV, fixed, measured, and remaining most influential, (c) CV, most influential, using unreliable data.

Figure 9

Comparison of estimated posterior distributions for 4-hour inhalation exposure to m-xylene. Posterior distributions were estimated by updating the most influential parameters or by fixing the measured parameters and updating the remaining most influential: (a) CXPPM, full parameter set, (b) CXPPM, most influential.

Figure 10

Model predictions for three volunteers from one iteration of the Markov chain and the associated measurements: (a) urine predictions and data, (b) CV predictions and data, (c) CX predictions and data.