Cytosolic Binding Proteins and P450–P450 Contacts as Missing Mechanisms in Hepatic Clearance IVIVE

Question: How have physiologically based pharmacokinetic models incorporated cytosolic binding proteins and protein-protein interactions to improve in vitro-to-in vivo extrapolation of hepatic clearance for low-clearance CYP substrates?

Answer

Across the retrieved literature, the answer is narrow and largely programmatic: cytosolic binding proteins enter PBPK-scale liver models as an explicit intracellular binding term for the free substrate concentration, while protein-protein interactions enter as a change in the functional form of the intrinsic clearance equation, and the two are usually implemented together because the same protein (FABP1) both sequesters ligand and contacts CYP enzymes. The strongest evidence is for CYP2C9: FABP1 alters metabolite formation kinetics in human liver microsomes, a model including FABP1–CYP2C9 protein-protein interaction best fitted the kinetic data, and CYP2C9 and FABP1 expression jointly best predicted hydroxylation across 61 human liver S9 fractions, with incorporation of both into novel physiologically based liver models "significantly improved the quantitative accuracy of in vivo hepatic clearance prediction" [1]. A companion liver distribution model showed the same binding term is required on the distribution side: liver Kp values for FABP1-binding drugs were accurately predicted only by the model incorporating lipid partitioning, albumin distribution and FABP1 binding, and FABP1 abundance variability across 61 human livers produced interindividual variability and concentration-dependent distribution [4]. An independent line of evidence shows that P450–P450 interactions can decouple activity from abundance, which challenges abundance-scaled extrapolation generally [3]. Notably, no retrieved paper frames its analysis specifically in terms of low-clearance substrates or reports accuracy stratified by extraction ratio.

FABP1 is the only cytosolic binding protein parameterised into a PBPK-scale liver model, and it operates through both ligand sequestration and CYP contact

The models that quantify intracellular binding do so through a liver distribution model built to test four alternative mechanistic structures, in which known distribution processes were combined with intracellular binding to liver FABP1 [4]. Accuracy was structure-dependent: Kp was accurately predicted by the model combining lipid partitioning, albumin distribution and FABP1 binding, and not by the models lacking FABP1 binding [4]. The same binding protein was quantified across a 61-liver panel and its interindividual variability propagated into Kp predictions, generating concentration-dependent tissue distribution [4]. On the clearance side, the presence of FABP1 altered the kinetics of metabolite formation in human liver microsomes, and the model that best fitted those kinetics included FABP1–CYP2C9 protein-protein interactions rather than FABP1 binding alone [1]. The claim shared by both papers is that intracellular binding is not a free-fraction scalar that can be lumped into microsomal or plasma binding; it requires its own compartment and its own affinity term.

Protein-protein interaction terms change the functional form of intrinsic clearance rather than rescaling it

In the CYP2C9 work, protein-protein interaction was introduced as a mechanistic element of the kinetic model, and it was the model containing it that best described the data — implying that a binding-only correction is insufficient [1]. Independent structural and crosslinking evidence establishes that such CYP–CYP contacts are physical and functionally consequential: tag-transfer chemical crosslinking mass spectrometry on Supersomes co-incorporating CYP2E1 and CYP3A4 confirmed direct protein-protein contacts, identified CYP3A4 residues at the interface, and supported structural models of the complex, invoked to explain accelerated CYP3A-mediated drug elimination in alcohol-exposed livers despite unaltered CYP3A protein levels [3]. That is a stronger version of the claim than a scaling correction: the interacting partner can change catalytic output while the nominal enzyme abundance stays constant [3].

Proteomics across donor panels shows partner-protein abundance carries predictive information that CYP abundance alone does not

Statistical modelling of proteomic expression against drug hydroxylation in 61 human liver S9 fractions showed that CYP2C9 and FABP1 expression together best predicted probe substrate hydroxylation, i.e. the non-CYP binding protein added predictive value beyond the enzyme itself [1]. The same pattern appears for a different isoform and a different partner: in 23 proteomically characterised human liver microsome preparations, the correlation of CYP1A2-linked Vmax with CYP2E1 abundance was further improved by combining it with the abundances of CYP2A6 and CPR (cytochrome P450 reductase) [2]. Together these results argue that the scaling factor in IVIVE should be built from the abundances of the enzyme, its redox partner, and its cytosolic binding partner, all measured in the same donor panel.

Binding-protein abundance variability is a source of interindividual and concentration-dependent error, not a constant

Because FABP1 was quantified in 61 human livers and its variability propagated into tissue Kp predictions, the simulations showed that intracellular FABP1 binding can cause interindividual variability in Kp values and produce concentration-dependent tissue distribution [4]. Concentration-dependence is the critical implication for extrapolation: a single donor-averaged FABP1 term cannot be valid across the exposure range, and the error it introduces differs between compounds according to their FABP1 affinity [4]. The same donor-panel logic underlies the S9 analysis, where the joint CYP2C9-plus-FABP1 model was preferred over enzyme-only models across a 61-sample panel [1].

Conflicts and gaps

The retrieved papers do not agree on, or in most cases even address, the direction and magnitude of the correction. The CYP2C9 abstract states that microsomal clearance is "systematically over- or underpredicted" but reports no fold-error values, no confidence intervals, and no per-compound accuracy for diclofenac or tolbutamide, so the size of the improvement attributed to FABP1 and to the interaction term cannot be assessed [1]. The two mechanistic accounts compete rather than reconcile: one attributes altered hepatic metabolism to FABP1–CYP binding and intracellular sequestration [1], the other to physical CYP–CYP heterocomplexation that changes catalysis at unchanged enzyme abundance [3]; neither abstract reports a design that discriminates between the two mechanisms. Relatedly, the Kp work separates FABP1 binding from lipid partitioning and albumin distribution [4], whereas the clearance work bundles binding with protein-protein interaction [1], so the two model structures are not nested and their parameter sets are not directly comparable. All of the FABP1 modelling derives from a single group and a single liver bank, with no independent replication in the retrieved set. Most importantly for the question as posed, no retrieved paper defines a low-clearance or low-extraction substrate set, reports extrapolation accuracy stratified by extraction ratio, or demonstrates that the benefit of these terms is larger for low-clearance compounds — the abstracts concern CYP2C9 probe substrates and liver Kp, not low-clearance IVIVE as a class. Whether the improvement generalises beyond FABP1-binding, CYP2C9-metabolised compounds, and beyond the redox-partner effects documented for CYP1A2 [2], remains untested here.

What would change the answer

A blinded external validation in a dedicated low-clearance substrate set — compounds with low hepatic extraction and described largely by unbound intracellular concentration — would settle whether FABP1 and protein-protein interaction terms materially reduce bias, provided the analysis reports fold error and average fold error with and without each term and stratifies by extraction ratio. Causation, as opposed to correlation, requires a binding-deficient or interface-mutated FABP1 variant expressed in the same liver model, so that the FABP1–CYP2C9 contact is broken while ligand affinity is retained, or vice versa. Independent quantification of the intracellular unbound fraction — for example equilibrium dialysis or ultracentrifugation of liver cytosol, or direct Kd determination for the probe substrates — would replace the fitted binding term with a measured one, and would allow the sequestration and heterocomplexation mechanisms to be tested as separate parameters rather than combined [1] [4]. Extending the crosslinking-mass-spectrometry interface mapping used for CYP2E1–CYP3A4 [3] to CYP2C9-containing systems, combined with proteomics across a liver bank [2], would establish which P450–P450 contacts occur at physiological abundance ratios. Finally, formal identifiability and sensitivity analysis of the PBPK structure is needed to show that improved accuracy stems from the FABP1 and interaction parameters themselves rather than from the added degrees of freedom [4].

References

[1] Wen, Y. W., Friedman, E., Zelter, A., Yabut, K. C., Nath, A., MacCoss, M. J., Isoherranen, N.. "Liver Fatty Acid Binding Protein (FABP1) Affects Cytochrome P450 2C9 (CYP2C9)-Mediated Metabolism in Human Liver Models". bioRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.10.12.758325

[2] Gaither, K. A., Davydova, N., Ponraj, K., Singh, D. K., Prasad, B., Davydov, D. R.. "A New Fluorogenic Substrate for CYP1A2 and Its Application in Studying the Effects of Alcohol Exposure on Liver Drug Metabolism". bioRxiv. 2026-02-23T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.21.703381

[3] Davydov, D. R., Ponraj, K., Davydova, N., Yue, G., Singh, D. K., Neogi, A. G., Gaither, K. A., Prasad, B.. "The role of CYP3A-CYP2E1 interactions in activation of CYP3A enzymes by chronic alcohol exposure". bioRxiv. 2026-03-20T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.06.703602

[4] Wen, Y. W., Isoherranen, N.. "Development of Physiologically Based Liver Distribution Model that Incorporates Intracellular Lipid Partitioning and Binding to Fatty Acid Binding Protein 1 (FABP1)". bioRxiv. 2026-01-21T00:00:00.000Z. URL: https://doi.org/10.64898/2026.01.17.700130