The estimation of external exposure and intake from biomonitoring data utilising Physiologically Based BioKinetic (PBBK) models is an ongoing research field known as exposure reconstruction or reverse dosimetry. The major benefit of this retrospective interpretation is that it can be used to quantify both aggregate and cumulative exposure. Although PBBK models are able to translate external exposure from multiple routes into internal exposure metrics addressing the effects of exposure route in the overall bioavailability, uncertainties of their use for exposure reconstruction rely on the validity of the PBBK models, the analytical error, the fitness-for-purpose of the biomonitoring data, as well as the exposure reconstruction algorithms used. Assimilation of human biomonitoring data and their translation into intake distribution accounts to a computational inversion problem, assessed by a distribution of exposures possibly associated with specific blood or urine levels.
The current study aims at the development and evaluation of an integrated framework for exposure reconstruction of environmental and consumer product chemicals. The developed framework consist of 3 basic steps. The first step aims at specifying and modifying the prior parameter distribution, the joint probability distribution, the population model and the determination of the measurement model. At the second step, the biomonitoring data is imported to the procedure and the exposure reconstruction algorithms is employeed; two different algorithms have been applied. Finally, Monte Carlo simulation is realized on the results applying a comparison of prior and posterior distribution as well as parameter independence. In principle, based on the identification of a global minimum of the error metric, the input parameters and the model setup the computational inversion techniques for exposure reconstruction can be classified as deterministic or stochastic. In this framework a stochastic exposure reconstruction method was applied and the selected algorithms were the Markov Chain Monte Carlo (MCMC) and the Differential Evolution Monte Carlo (DEMC). DEMC is the integration of the Differential Evolution (DE) and MCMC and it is a population MCMC algorithm characterized by (a) the parallel run of multiple chains and (b) the ability to choose the appropriate scale and orientation for the jumping distribution during the simulation. DE is a genetic algorithm for numerical optimization in real parameter spaces. In addition, MCMC is a stochastic process in which future states are independent of past states, given the present state. During iterative simulations, the random variables of interest are drawn from a chain of distribution that eventually converge to a stable posterior distribution. The generic PBBK model captures multi-route exposure (including inhalation, skin absorption and gastro-intestinal absorption), and incorporates detailed compartmental description, plasma protein/red cell binding and estimates the concentration of parent compounds and metabolites in several biological fluids including breast milk. The generic model was applied for the assessment of a highly controversial industrial chemical with widespread applicability in consumer goods, namely bisphenol-A (BPA). BPA is chemical of great interest cause it is an estrogen characterized by endocrine disrupting activities that are mediated via multiple molecular mechanisms as well as it is associated to neurodevelopmental defects. Parameterization of exposure scenarios were based on an extensive literature review of BPA exposure data and the generic PBBK model was properly parameterized. BPA is considered to be rapidly conjugated into BPA glucuronide (BPA-Glu), resulting in rapid elimination from the human body due to the water solubility of this metabolite. Glucuronidated bisphenol-A (BPA-Glu) is a metabolite product of BPA that can be detected in human urine samples. Exposure to BPA was reconstructed based on real-life HBM European data. Average urinary glucuronidated bisphenol-A (BPA-Glu) was 2.8 μg/L across Europe, covering different age groups. In particular the algorithm has been tested under the assumption that the average amount of 2.8 μg/L BPA-Glu to human’ s urine is the results of an ordinary adult dietary schedule that includes 3 different meals. The prior knowledge of the distribution of the exposure time is based on the actual dietary schedule of the generic population. The time of the three basic dietary meals follows normal distribution and in particular, the breakfast is at 7:00 AM and the dinner is at 7:00 PM with standard deviation of 30 minutes for both and the lunch is at 2:00 PM with standard deviation of 30 minutes.
The selected exposure reconstruction algorithms (MCMC as well as DEMC) were tested for 1000 iterations. All the prior distributions converged to the actual exposure dose. The results of the exposure reconstruction illustrated that the posterior distributions include the actual exposure doses. Moreover, the posterior distributions had a reduced standard deviation and a mean value closer to the real one. However, MCMC model using the prior knowledge could not achieve a posterior distribution with a sufficient confidence interval (CI) towards the actual exposure value. The exposure dose of second and third meal were more accuratelly predicted. In practice, the results indicated that the overall daily intake is very low, approximately 1 μg/kg_bw/d, which is far below the Tolerable Daily Intake (TDI) of 50 μg/kg_bw/d proposed by the European Food Safety Authority (EFSA). Comparing the computation needs of the algorithms, the DEMC is 3 to 4 time slower than MCMC but the predictions converged faster when the algorithm was tested against synthetic data for validation purpose. The high frequency of exposure interval representing the real life exposure scenario of this study increases the computation needs. In a statistical context, the combination of the genetic algorithm and the stochastic process of MCMC that are coupled by DEMC results in more accurate predictions reducing the overall uncertainty.
This study demonstrated how an integrated framework based on a PBBK model could effectively combine biomonitoring data in order to estimate exposure, absorbed, metabolized and extracted dose on a chemical compound, since biomonitoring data reflect an internal dose associated with external exposures from all potential sources and routes. However, the exposure routes and sources (natural or synthetic; endogenous or exogenous) have to be appropriately identified and characterized before the procedure. A key issue on reverse dosimetry is the bio-persistence of a compound, since the measured biomonitoring data might reflect a recent or an older exposure regime.
The algorithm of DEMC is used for first time in exposure reconstruction models providing clear benefits in terms of biomonitoring data assimilation; however, additional optimization of the code will result in reduced computational time. Further work is expected on developing and using sufficient genetic algorithms appropriate for the exposure reconstruction models in order to reduce the computational time as well as to reduce the uncertainty of the estimated results. A generic and reliable generic (covering a large chemical space) exposure reconstruction scheme, could be a very useful tool for modern risk assessment of chemicals, utilizing the large amount of existing human biomonitoring data. In addition, higher frequency of biomonitoring sampling is needed for rapidly metabolized compounds. Overall, exposure reconstruction has the potential to provide reliable estimates of total exposure by interpreting biomonitoring data. This gives us the capability to re-run forward our model to estimate the actual biologically effective dose in the target tissue and to compare this to toxicity results obtained from in vitro testing, significantly advancing environmental health science and chemical safety regulation.
See more of this Group/Topical: Computing and Systems Technology Division