Caution with PBPK Modeling Based on the In Vitro Kinetics for Bilirubin Hepatic Uptake and Glucuronidation: Commentary on Dong et al.

PBPK modeling is increasingly used to predict clearance and pharmacokinetic interactions, and to understand their molecular mechanisms. The hepatic uptake (OATP1B1 and OATP1B3) and glucuronidation (UGT1A1) of bilirubin can be inhibited by various drugs, resulting in hyperbilirubinemia. Predicting the relative contributions of these molecular pathways to the pharmacokinetics of bilirubin and its interactions is made difficult by the highly variable in vitro kinetic parameters available for PBPK modeling.

Hyperbilirubinemia is a common adverse effect in patients treated with atazanavir, alone or in combination with ritonavir or cobicistat. Bilirubin is considered to be glucuronidated exclusively by UGT1A1, and atazanavir is a potent inhibitor of this enzyme in vitro. Thus, inhibition of glucuronide formation is believed to contribute significantly to the increased exposure of UGT1A1 substrates in subjects treated with atazanavir. ¹ Importantly, however, atazanavir is additionally known to potently inhibit the transporters OATP1B1 and OATP1B3, which play a pivotal role in the hepatocellular uptake of bilirubin glucuronide and bilirubin itself. ¹ The relative contributions of UGT1A1 and OATP1B1/3 inhibition to the atazanavir–bilirubin interaction have remained unknown for some time. Dong et al. ² recently reported in this Journal the results of a physiologically based pharmacokinetic (PBPK) analysis that assessed the contributions of UGT1A1 and OATP1B1/3 inhibition to the atazanavir–bilirubin interaction. It was concluded that UGT1A1 inhibition plays a modest role (< 33%) in the increase in bilirubin exposure.

The PBPK model incorporated published kinetic parameters for bilirubin glucuronidation and for the transport of bilirubin and its glucuronide conjugates by OATP1B1/1B3 in vitro. As described below, however, the measurement of kinetic parameters for bilirubin glucuronidation and for the uptake of bilirubin and its glucuronides is associated with numerous technical challenges, and experimental conditions differ markedly between studies. It is conceivable that the parameter estimates employed for model development and underlying assumptions will influence the outcome of PBPK modeling.

The PBPK model of Dong et al. used input parameters (K _m and “optimized” J _max) for the uptake of unconjugated bilirubin derived from kinetic studies that employed recombinant human OATP1B1 and OATP1B3 expressed in Xenopus laevis oocytes; respective K _m values were 0.0076 and 0.039 μM. ³ The K _m for OATP1B1 is approximately one‐twentieth that (approximately 0.16 μM) reported by Cui et al. ⁴ using the recombinant human protein expressed in human embryonic kidney (HEK293) cells, illustrating the potential bias that may arise from inter‐laboratory differences. Cui et al. further highlighted the difficulties estimating unbound bilirubin concentrations given both studies included human serum albumin (HSA) in the incubation medium. The K _m values for conjugated bilirubin (i.e., mono‐ and di‐glucuronidated bilirubin) uptake by OATP1B1 and OATP1B3 used in model development were as reported by Cui et al., although J _max values appear to have been determined indirectly from J _max/K _m ratios for unconjugated bilirubin. Of particular note, the liver was modeled as a permeability‐limited tissue, which biases toward the importance of transport. In this respect, it is widely believed that the uptake of bilirubin into hepatocytes involves both OATP1B1/3‐mediated transport and passive diffusion, although the relative contribution of each is yet to be established conclusively.

The measurement of kinetic parameters for bilirubin uptake and glucuronidation is associated with numerous technical challenges. In particular, (i) bilirubin and its glucuronide conjugates are chemically unstable, being especially prone to photo‐degradation; (ii) bilirubin binds avidly to numerous proteins, for example human liver microsomal proteins and albumin that may be added to incubations, reducing the concentration of substrate that is available for metabolism and/or transport; and (iii) with respect to glucuronidation kinetic studies, establishing initial rate conditions is problematic due to sequential glucuronidation. To overcome these problems, Zhou et al. ⁵ proposed that incubations should be performed in a reduced light environment over a short incubation time (5 min) at a very low recombinant UGT1A1 cell lysate protein concentration (0.05 mg/mL). Under these conditions, the K _m value reported for bilirubin glucuronidation by recombinant UGT1A1 was 0.2 μM. As noted by the authors, this value may well be an over‐estimate of the unbound K _m. Given these limitations, β‐estradiol is commonly employed as an alternate substrate for the assessment of UGT1A1 activity in vitro. ⁶

The kinetic parameters for bilirubin glucuronidation employed by Dong et al. for PBPK model development were taken from a study that used UGT1A1 genotyped HLM as the enzyme source. ⁷ Experimental conditions differed markedly from those proposed by Zhou et al.; incubations were conducted over 45 min and contained 0.5 mg/mL of HLM protein. Further, the K _m and V _max values used for PBPK model development were derived using hepatic microsomes from a single donor homozygous for the reduced function UGT1A1*28 allele (Dong et al.; Supplementary table B1). The K _m, 3.52 μM, was almost 20‐fold higher than the value (viz. ≤ 0.2 μM) generated using recombinant UGT1A1 as the enzyme source. This discrepancy almost certainly arises from the non–“optimized” incubation conditions employed in the study with HLM. ⁷ At least for glucuronidation reactions catalyzed by UGT1A1, concordance of K _m values for the two enzyme sources is anticipated when incubations are performed under identical conditions. ⁶ The V _max was 38 pmol/min.mg microsomal protein (table 5, reference 7). Since the liver was from a donor homozygous for UGT1A1*28, it was converted to the V _max for wild‐type UGT1A1 (i.e., UGT1A1*1) using Simcyp (Dong et al.; Supplementary table B1); the value obtained was 90.5 pmol/min.mg microsomal protein. It should be noted that while the UGT1A1*28 allele does not affect the K _m of bilirubin and other UGT1A1 substrates, it significantly reduces V _max. Owing to the reduction in “free” substrate concentrations arising from degradation and bilirubin binding in the human liver microsomal study, the measured and extrapolated V _max values are expected to be markedly lower than the “true” values. Interestingly, a V _max of 391 pmol/min.mg microsomal protein was reported for hepatic microsomes from a donor homozygous for wild‐type UGT1A1, ⁷ but this value was overlooked.

It is therefore apparent that hepatic intrinsic clearance, calculated as V _max/K _m, for bilirubin glucuronidation will be underestimated, possibly by as much as 2 orders of magnitude, due to the higher K _m and lower V _max values used by Dong et al. Also of interest, the unbound K_i value for atazanavir inhibition of bilirubin glucuronidation employed for PBPK modeling was 0.77 μM (Dong et al., Supplementary table C1). IC₅₀ values down to 0.31 μM have been reported for atazanavir inhibition of human liver microsomal β‐estradiol 3‐glucuronidation, which is considered a reliable marker of UGT1A1 activity in vitro. ⁶ This suggests that the K _i for atazanavir inhibition of human liver microsomal UGT1A1 could be as low as 0.16 μM. ¹ Collectively, these data suggest that the contribution of UGT1A1 inhibition to the bilirubin–atazanavir interaction may be greater than reported by Dong and colleagues.

Caution is warranted when interpreting the results of PBPK modeling based on in vitro kinetics for bilirubin hepatic uptake transport and glucuronidation. More broadly, we note a trend in PBPK modeling where in vitro kinetic parameters are selected subjectively by modelers from a range of published data or back‐extrapolated from clinical studies to achieve simulation results that best fit the observed data, thus compromising in vitro–in vivo extrapolation of enzyme intrinsic clearance and/or clearance by transporters. Given that many input parameters are required for PBPK modeling, multiple combinations of these parameters, either optimized or unoptimized, may produce results within model acceptance criteria. ⁸ Superior justification of parameter selection and optimization is warranted when the so‐called top‐down or middle‐out PBPK approaches are used to formulate conclusions regarding molecular mechanisms of drug disposition and metabolism, including those responsible for pharmacokinetic drug–drug interactions and altered pharmacokinetics with organ impairment. Artificial intelligence may be particularly useful for identifying and optimizing the in vitro kinetic data and other parameters used for model development. ⁸ Automation could then explore parameter spaces within the known ranges, strengthening the conclusions from simulations regarding molecular mechanisms. ⁹

The generalization of Dong et al. that “UGT enzymes are well‐known as a low‐affinity and high‐capacity system” also requires comment. As described in recent reviews from this laboratory, previously published K _m values of glucuronidated substrates have frequently been over‐estimated due to the use of non‐optimized experimental conditions. ¹ , ⁶ Bilirubin is a notable example in this regard, with an unbound K _m ≤ 0.2 μM (see above). Importantly, however, it is the unbound inhibition constant K _i,u, along with fractional metabolism by the inhibited enzyme (f _m) and the hepatic inlet maximum unbound concentration of the precipitant drug in blood, that is relevant to the assessment of drug–drug interaction liability in vitro. ¹ , ⁶ There are many inhibitors of UGT enzymes that have K _i,u values in the low micromolar to nanomolar range. By way of example, the K _i,u values for regorafenib and sorafenib inhibition of human liver microsomal UGT1A1 activity using β‐estradiol as the substrate are 0.022 and 0.033 μM, respectively, ¹⁰ indicating that ligands may bind with high affinity. Thus, we caution that the likelihood of a drug–drug interaction arising from UGT inhibition should not be discarded based on a preconceived notion that UGTs are “low‐affinity” enzymes.

FUNDING

No funding was received for this work.

CONFLICTS OF INTEREST

The authors declared no competing interests for this work.