Prediction of herb-drug interactions involving consumption of furanocoumarin-mixtures and cytochrome P450 1A2-mediated caffeine metabolism inhibition in humans.

Herb-drug interactions (HDI) has become important due to the increasing popularity of natural health product consumption worldwide. HDI is difficult to predict as botanical drugs usually contain complex phytochemical-mixtures, which interact with drug metabolism. Currently, there is no specific pharmacological tool to predict HDI since almost all in vitro-in vivo-extrapolation (IVIVE) Drug-Drug Interaction (DDI) models deal with one inhibitor-drug and one victim-drug. The objectives were to modify-two IVIVE models for the prediction of in vivo interaction between caffeine and furanocoumarin-containing herbs, and to confirm model predictions by comparing the DDI predictive results with actual human data. The models were modified to predict in vivo herb-caffeine interaction using the same set of inhibition constants but different integrated dose/concentration of furanocoumarin mixtures in the liver. Different hepatic inlet inhibitor concentration ([I]H) surrogates were used for each furanocoumarin. In the first (hybrid) model, the [I]H was predicted using the concentration-addition model for chemical-mixtures. In the second model, the [I]H was calculated by adding individual furanocoumarins together. Once [I]H values were determined, the models predicted an area-under-curve-ratio (AUCR) value of each interaction. The results indicate that both models were able to predict the experimental AUCR of herbal products reasonably well. The DDI model approaches described in this study may be applicable to health supplements and functional foods also.

Based on regulatory guidelines, new-drug-applicants are required to follow a step-wise protocol which includes the investigation of major metabolizing enzymes inhibition, most notably cytochrome P450 (CYP) enzymes, in an effort to reduce unnecessary clinical trials and post-marketed drug-withdrawals. Several DDI prediction models have been developed by researchers and adopted by regulatory bodies. Until recently, DDI prediction is based mainly on reversible CYP enzyme inhibition mechanisms and has been carried out routinely as part of drug preclinical studies (Fahmi et al., 2008). If irreversible inhibition were to occur, this would result in underestimation of the magnitude of DDI (Einolf, 2007). Co-administration of a pharmaceutical drug and an herbal product, with bioactive constituents that interfere with drug metabolite(s) formation, might significantly alter the pharmacokinetics of the victim-drug. The outcome of such alterations may result in serious clinical consequence and deaths (Ebbesen et al., 2001).
Predictive DDI models have been developed with increased accuracy in prediction not only for reversible DDI but also for irreversible DDI. For example, Mayhew et al. model (Safe, 1998;ATSDR, 2004;Alehaideb et al., 2019) to calculate an integrated dose/concentration for the furanocoumarin-mixture in the liver. The integrated dose is then used to calculate the area-under-curve-ratio (AUCR) with the Mayhew et al.

model.
The "victim-drug" in this study is caffeine. A popular drug with adverse-health-effects upon abusive consumption (Dews, 1982). Caffeine is also an ideal probe to measure in vivo CYP1A2 activity (Doehmer et al., 1992;Miners et al., 1996). The pharmacokinetics of caffeine in humans has been studied extensively (Kot and Daniel, 2008). The "perpetrators" in the present study are the linear furanocoumarins ( Figure   1) which are chemical isomers and congeners found in Apiaceae, Leguminosae, Moraceae, and Rutaceae plant families (Diawara and Trumble, 1997). Previous in vitro liver microsomal studies have shown that the main metabolic pathway of 8-MOP and 5-MOP is the oxidative ring-opening of the furan structure to form an epoxide, or an intermediate electrophilic reactive metabolite, which binds covalently to human CYP proteins (Fouin-Fortunet et al., 1986;Tinel et al., 1987;Mays et al., 1990;John et al., 1992). Other linear furanocoumarins such as ISOP (Kang et al., 2011) andpsoralen (Zhuang et al., 2013) also have been shown to be TDI, or MBI, in human liver microsomes (HLM) or recombinant human CYP1A2 expressed in yeast.
Our previous studies have shown that detectable levels of 8-MOP, 5-MOP, and ISOP in nine traditional herbal medicine (Alehaideb et al., 2017). We also have shown four of the furanocoumarin-containing herbs significantly reduced the oral clearance of caffeine in human volunteers (Alehaideb et al., 2021). Furthermore, the aforementioned furanocoumarins were found to be potent TDI or MBI of CYP1A2 isozyme (Alehaideb et al., 2021). The objectives of this study were: (1)

Source of plant products
The plant products used in this study were obtained from North American commercial

Using DDI models to predict herb-caffeine interactions.
In the present study, caffeine is the victim-drug and the furanocoumarin bioactive in the herbs are the perpetrators or inhibitors as mentioned earlier. As both caffeine and furanocoumarin inhibitors are metabolized by the same CYP1A2 enzyme, metabolic inhibition may occur in humans after co-administration. Indeed, the furanocoumarins have been shown previously to be TDI of CYP1A2. We have used two different DDI

Using Wang et al. model to predict caffeine-herb interaction.
Instead

In vitro caffeine metabolism inhibition data.
The in vitro experimental inhibition parameters we obtained using radiolabeled caffeine and pooled human liver microsomes as described in detail in our previous study (Alehaideb et al., 2021). Briefly, the IC50 values were obtained using serial dilutions of pure furanocoumarin incubated with radiolabeled caffeine, microsomes, and NADPH cofactor. The radiolabeled metabolites were collected by solid-phase extraction and counted by scintillation. The inactivation constants were experimentally measured using the two-step dilution with preincubation time points from 0.5 to 4 min.   (b) Cmax,PU was derived from Cmax,PT by multiplying the latter with the unbound fractions (fup) of 8-MOP or 5-MOP in the plasma; they were 17.0 (±7.4) and 3.6 (±2.2) percent respectively (Veronese et al., 1978;Artuc et al., 1979;Pibouin et al., 1987;Makki et al., 1991;Muret et al., 1993a). No information was found for ISOP. Therefore, an average of 8-MOP and 5-MOP values was used.
(c) The Cmax,LT was derived by multiplying Cmax,PT with the liver:plasma partition coefficient (Pt:p) which was calculated as follows: Equation 4: Non-adipose tissue:plasma partition model of Poulin and Theil (2002).
where Po/w is the n-octanol/water partition for non-ionized inhibitor, Vnt is the fraction weight of neutral lipids in liver tissue, Vpht is the fraction weight of phospholipids in liver tissue, Vwt is the fraction weight of water in liver tissue, Vnp is the fraction weight of neutral lipids in plasma, Vwp is the fraction weight of water in plasma, Vphp is the fraction weight of phospholipids in plasma, fup is the fraction unbound in plasma, and fut is the fraction unbound in liver tissue.
(d) The Cmax,LU was derived by multiplying the Cmax,LT by the calculated unbound fraction (fut) in tissue. The fut was determined using the following equation: Equation 5: Unbound tissue fraction model of Poulin and Theil (2000).

Human caffeine pharmacokinetic studies
The in vitro pharmacokinetic data for caffeine metabolism inhibition due to consumption of furanocoumarin-containing herbal products were reported in our previous publications in great detail (Alehaideb et al., 2021). Briefly, eligible volunteers were dosed with 200 mg caffeine twice: with and without prior treatment with an aqueous extract of one selected herbal medicine. Saliva samples were collected at timepoints ranging between 0 to 48 hours. Caffeine and internal standard were separated and measured chromatographically using an isocratic method with an ultra-violet detector. The salivary caffeine concentrations were converted into plasma concentrations using a conversion factor of 0.79. The human plasma caffeine AUC from zero to infinity (AUC0-inf) for each volunteer was measured twice and the experimental AUCR was calculated using the PKsolver software.

Data and statistical analysis.
Data plotting and extrapolation were performed using GraphPad Prism Software version 5.04 (San Diego, CA). Statistical analysis was performed using Microsoft Excel software. Model-predicted AUCR were reported as mean ± standard deviation (SD).
Herb-caffeine interaction occurred when the mean AUCR was equal or greater than 2.0.
No interaction occurred when the mean AUCR was less than 2.0 (Einolf, 2007). The geometric mean-fold error (GMFE) (Equation 6) was also used to assess the accuracy of model prediction by equal weighting under-predictions and over-predictions. The model that predicted perfectly would give a GMFE value of 1; GMFE value between 1 and 2-fold is considered to be accurate. where n is the number of predictions for each herb.

Results and Discussion
Tables 2  Relatively high experimental AUCR were observed in volunteers pre-treated by A.
These results are consistent with the relatively high levels of furanocoumarins found in the herbal extracts (Table 1) predictability of the IVIVE model is also affected by the kinetic parameters as follows: (a) the kdeg value of hCYP1A2. The 0.00030 min -1 kdeg used in the present study is based on the 38 h t1/2 of a tobacco smoking cessation study (Faber and Fuhr, 2004). Mayhew et al. has reported a kdeg value of 0.00083 min -1 which is derived from rats. If this kdeg was used in the present study, both predictive models would underestimate the AUCR of caffeine, (b) in vitro kinetic inhibition parameters such as IC50, KI and kinact were derived using pooled HLM from multiple donors. This also appears to improve the accuracy of our predictions, and (c) caffeine as the victim drug has simplified model prediction by eliminating the need to account for parallel metabolic pathways by other CYP isoforms and urinary excretion of unchanged caffeine. As a result, the uncertainty involved in the AUCR calculation is greatly reduced.
The following are some of the uncertainties or limitations of the present study: (a) the data used to establish the 8-MOP dose-response curves for allometric extrapolation are based on psoriasis patients and these raised concern that plasma 8-

Conclusions
Using IVIVE models to predict HDI is an ongoing research program in our laboratories.
Our goal is to develop a reliable and simple prediction tool for HDI. Results of this study suggest that linear furanocoumarins such as 8-MOP, 5-MOP and ISOP are responsible for the inhibition of caffeine metabolism in humans after consuming herbal extracts containing furanocoumarin derivatives. The described modeling approaches in this study may also be applicable to health supplements and functional food.

List of Equations
Equation 1 Figure 1 Psoralen and derivatives in this study.

Figure 2
Flow-chart summary of experimental procedures and validation steps in the use of drug-drug interaction models to predict herb-drug interaction.