Comparisons of Predicted and Observed AUCs for Both Acrylamide and Glycidamide Based on Hemoglobin Adduct Data--Implications for Model Parameter Adjustments and Model Behavior, Using the Original Kirman et al. (2003) Set of Partition Coefficients

Read Also

The basic dose response behavior of a variety of parameters in 24 hour simulations of the original Kirman et al. model is shown in Table 2-3.  It can be seen that this base model predicts that at low doses about 38% of the acrylamide is transformed to glycidamide.  As doses are increased, metabolic saturation is approached and the percent conversion to glycidamide declines to about 23% at 50 mg/kg—less than the 32.6% fraction observed in the urinary metabolite output in the Sumner et al. (1992).  Also, the original Kirman et al. (2003) model predicts that the low dose half lives for acrylamide and glycidamide are about 1.35 and slightly over 2 hours, respectively.  The last column of Table 2-3 also indicates that there is a modest degree of nonlinearity in the ratio of internal glycidamide AUC/external acrylamide dose in the 0.5 – 3 mg/kg dose range used for the chronic rat bioassay experiments.

Table 2-4 compares the expectations under the original Kirman et al. (2003) model with the adduct-based AUC observations presented in Table 2-2.  It can be seen that although the model conforms reasonably to the Bergmark et al. (1991) observations for the acrylamide AUC, there is an approximately 2-fold under prediction of the acrylamide AUCs derived from the Fennell and Sumner hemoglobin-valine adduct observations.  The discrepancies between predictions and observations for the adduct-based glycidamide AUCs are much more profound. The under- predictions of glycidamide AUCs are 3-7 fold for the Bergmark et al. (1991) adduct observations, and on the order of 10-fold or slightly more for the observations of Fennell and Sumner at low doses. 

We adjusted the model in a series of steps.  We will describe the steps and associated reasoning, but we will not document all of the intermediate results here for the sake of brevity. 

·         First, we removed the multiplier of 3.2 from the calculation of partition coefficients for glycidamide; substituting 1.  This directly increased the ratio of glycidamide in the blood relative to the tissues; and hence increased the AUC of glycidamide while making relatively modest changes in other aspects of model behavior.

·         The under prediction of acrylamide AUC by the model could only be rectified by reducing the rate of processing of acrylamide by the model.  The principal way we found to accommodate this without exacerbating the under prediction of glycidamide AUC was to reduce the rates of all non-P450 modes of acrylamide destruction—e.g. the glutathione transferase reaction and all the nonspecific reactions of acrylamide in the tissues.  In the final calibrated model (Table 2-5; presented in the same format as was used for the original model in Table 2-1), these non-P450 rates of reaction of acrylamide are reduced to approximately one quarter of their baseline values in the original Kirman et al. model.  This has the effect of increasing the proportion of acrylamide that is processed to glycidamide via P450 oxidation.

·         A more modest downward adjustment of 0.7 fold was made to both the Vmax and the Km for the P450-mediated metabolism of acrylamide to reduce the apparent dose dependence of the departures of predicted vs. observed acrylamide AUCs.  Changing both Vmax and Km  in parallel has the effect of decreasing the metabolism of acrylamide at high doses while leaving metabolism rates relatively unchanged at low doses.

·         Finally, to bring the model-“predicted” glycidamide AUCs into alignment with the observations, the rates of glycidamide metabolism by all routes were reduced to half their baseline values; increasing the glycidamide half-life.

The combined results of these changes for the dose response behavior various model parameters are shown in Table 2-6 (which is parallel to Table 2-3).   It can be seen that the low dose 1-2 hour half life of acrylamide has been increased from 1.35 to 2.46 hours (this slightly overstates the half life at later time points); the low dose half life of glycidamide has been increased  from about 2.1 hours to 3.8 hours and the low dose fraction of acrylamide processed via the glycidamide pathway has been increased from 39% to about 72%.  At 50 ppm, where the urinary metabolite data of Sumner et al. (1992) indicate a minimum of 34% must be processed by non-P450 pathways to soluble glutathione-derived metabolites, the model results show an expectation that 32.7% is processed by direct reaction with glutathione. 

Alternative model calibrations are probably possible that would reduce the fraction of acrylamide that is processed by the P450 pathway at low doses.  However in order to achieve compatibility with the glycidamide adduct data, such recalibrations would require further reduction in the metabolism rate of glycidamide—lengthening the internal half life of glycidamide in the system.  Choosing the calibration we have makes the smallest feasible modifications to the Kirman et al. rate constants for glycidamide metabolism.  This probably does not greatly affect the ultimate balance of metabolic processing derived for our human model.  As described in Section 3 below, the human model requires further reductions in the non-P450 rates of metabolism of acrylamide in order to both conform to the observation of relatively higher acrylamide AUC per external dose in humans compared to rats, while still generating sufficient amounts of glycidamide to produce the amounts of glycidamide adducts per unit dose that were observed in the human subjects.

The resulting fits to the Fennel/Sumner adduct based rat AUC data are reasonable although not ideal (Table 2-7) Except for the modest under prediction of the acrylamide AUC at the lowest dose used in the Fennell and Friedman (2004) dataset,  the remaining estimates do not depart by more than might be expected from the inherent variability of experimental results of this type.