Abstract
Conducting clinical trials to understand the exposure risk/benefit relationship of cannabis use is not always feasible. Alternatively, physiologically based pharmacokinetic (PBPK) models can be used to predict exposure of the psychoactive cannabinoid (−)-Δ9-tetrahydrocannabinol (THC) and its active metabolite 11-hydroxy-Δ9-tetrahydrocannabinol (11-OH-THC). Here, we first extrapolated in vitro mechanistic pharmacokinetic information previously quantified to build a linked THC/11-OH-THC PBPK model and verified the model with observed data after intravenous and inhalation administration of THC in a healthy, nonpregnant population. The in vitro to in vivo extrapolation of both THC and 11-OH-THC disposition was successful. The inhalation bioavailability (Finh) of THC after inhalation was higher in chronic versus casual cannabis users (Finh = 0.35 and 0.19, respectively). Sensitivity analysis demonstrated that 11-OH-THC but not THC exposure was sensitive to alterations in hepatic intrinsic clearance of the respective compound. Next, we extrapolated the linked THC/11-OH-THC PBPK model to pregnant women. Simulations showed that THC plasma area under the curve (AUC) does not change during pregnancy, but 11-OH-THC plasma AUC decreases by up to 41%. Using a maternal-fetal PBPK model, maternal and fetal THC serum concentrations were simulated and compared with the observed THC serum concentrations in pregnant women at term. To recapitulate the observed THC fetal serum concentrations, active placental efflux of THC needed to be invoked. In conclusion, we built and verified a linked THC/11-OH-THC PBPK model in healthy nonpregnant population and demonstrated how this mechanistic physiologic and pharmacokinetic platform can be extrapolated to a special population, such as pregnant women.
SIGNIFICANCE STATEMENT Although the pharmacokinetics of cannabinoids have been extensively studied clinically, limited mechanistic pharmacokinetic models exist. Here, we developed and verified a physiologically based pharmacokinetic (PBPK) model for (−)-Δ9-tetrahydrocannabinol (THC) and its active metabolite, 11-hydroxy-Δ9-tetrahydrocannabinol (11-OH-THC). The PBPK model was verified in healthy, nonpregnant population after intravenous and inhalation administration of THC, and then extrapolated to pregnant women. The THC/11-OH-THC PBPK model can be used to predict exposure in special populations, predict drug-drug interactions, or impact of genetic polymorphism.
Introduction
As of November 2020, 36 states in the United States allow either medical (21 states) and/or recreational (15 states) use of marijuana (cannabis) (https://www.ncsl.org/research/health/state-medical-marijuana-laws.aspx).In 2018, an estimated 40.3 million adults, which corresponds to 15.9% of the population in the United States, used cannabis (SAMHSA, 2018). Concentrations of (−)-Δ9-tetrahydrocannabinol (THC), the psychoactive constituent of cannabis, have been increasing, with estimated mean THC concentrations of 17.1% (Chandra et al., 2019). THC is eliminated from the body primarily by hepatic cytochrome P450 enzyme 2C9 metabolism, and this pathway also results in the production of its main psychoactive metabolite, 11-hydroxy-Δ9-tetrahydrocannabinol (11-OH-THC) (Patilea-Vrana and Unadkat, 2019). 11-OH-THC is cleared from the body by CYP3A, CYP2C9, and UGT metabolism (Patilea-Vrana and Unadkat, 2019).
Cannabis is consumed by a spectrum of individuals spanning from those taking Food and Drug Administration-approved drugs or supplements, pregnant women, to those with hepatic or renal impairment. Therefore, it is possible that the disposition of THC/11-OH-THC in special populations (e.g., pregnant women or patients with hepatic impairment) may be altered, or THC/11-OH-THC exposure may be impacted by drug-drug interactions. The impact of altered THC/11-OH-THC disposition on safety and efficacy under the aforementioned scenarios needs to be further studied. However, conducting studies, especially in pregnant women, to address these questions is ethically and logistically challenging. One approach to overcoming this dilemma is to first develop a physiologically based pharmacokinetic model (PBPK) of THC/11-OH-THC disposition using the bottom-up approach. Then, this PBPK model can be interrogated as to how changes in physiology in special populations or drug interactions will impact the disposition of THC/11-OH-THC. As such, the aims of this study were 1) to build a linked THC/11-OH-THC PBPK model using a bottom-up approach by extrapolating previously quantified in vitro enzyme kinetics of THC and 11-OH-THC (Patilea-Vrana et al., 2018), 2) to verify the THC/11-OH-THC PBPK model after intravenous and inhalation administration of THC in a healthy nonpregnant population, and finally 3) to extrapolate the verified THC/11-OH-THC PBPK model in healthy nonpregnant population to pregnant women and predict maternal and fetal cannabinoid serum concentrations. The final THC and 11-OH-THC PBPK models were verified after intravenous and inhalation but not oral administration of THC, since mechanistic data on extrahepatic metabolism is missing. Importantly, smoking cannabis is the most popular and therefore relevant route of administration (Schauer et al., 2016).
We chose to extrapolate the THC/11-OH-THC PBPK model to pregnant women since pregnant women are a special population often not studied. There are also concerns surrounding fetal exposure to THC/11-OH-THC. Epidemiologic studies have shown that maternal cannabis use leads to poorer neonatal outcomes (Grzeskowiak et al., 2020) and subtle but persistent neurodevelopmental consequences (Stickrath, 2019). Prospective clinical studies to quantify the risk of maternal cannabis use are not ethical. Alternatively, if the maternal and fetal THC and 11-OH-THC exposure during pregnancy is known, then the risk associated with maternal cannabis use can be anticipated. Measuring maternal and fetal drug exposure clinically is not ethical. Therefore, the only safe alternative to predict THC/11-OH-THC exposure during pregnancy is via PBPK modeling and simulation (M&S). This can be achieved by extrapolating the THC/11-OH-THC PBPK model verified in healthy, nonpregnant population to pregnant women by accounting for gestational age physiologic changes, such as changes to the activity of drug metabolizing enzymes and expression of drug binding proteins. Our laboratory has previously demonstrated the validity of using PBPK M&S to successfully predict maternal and fetal exposure of a variety of drugs (Ke et al., 2012, 2014; Zhang and Unadkat, 2017).
Materials and Methods
Meta-analysis of THC and 11-OH-THC Exposure after Intravenous and Inhalation of Either THC or 11-OH-THC to Identify Optimal Datasets for PBPK Model Training and Verification
The inclusion criteria for clinical studies used for PBPK model development included single dose studies in healthy nonpregnant subjects where THC and 11-OH-THC were administered either intravenous or via inhalation with no exclusions made for sample size, gender, frequency of cannabis use, or bioanalytical methodology. Where available, plasma/serum or blood concentration-time profiles were digitized using the online semiautomatic tool WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/). If not reported, the plasma/serum or blood area under the concentration-time curve (AUC) was estimated via noncompartmental analysis using Phoenix 8.1, Certara (Princeton, NJ). The details and observed PK parameters after intravenous administration of THC or 11-OH-THC and after inhalation of THC can be found in Supplemental Tables 1 and 2, respectively. Unless otherwise specified, plasma/serum concentrations and PK parameters are reported. Throughout this entire manuscript, observed or simulated plasma concentrations are assumed to be similar to serum concentrations.
To assess the quality of the observed data and identify any potential outliers, the observed THC and 11-OH-THC concentration-time profiles were dose normalized (Supplemental Fig. 1). As shown in Supplemental Fig. 1A, studies that used thin-layer chromatography (TLC) had remarkably different plasma concentration-time profiles compared with all other studies, particularly at timepoints >12 hours postdose. For example, THC plasma clearance (CL) estimated via TLC (Wall et al., 1983) was 12–15 l/h, whereas studies that used high performance liquid chromatography-UV (Hunt and Jones, 1980) or gas chromatograph–mass spectrometry (GC-MS) (Ohlsson et al., 1982; Kelly and Jones, 1992; Naef et al., 2004) estimated THC plasma clearance between 36 and 59 l/h. Similarly, a much longer terminal half-life was observed in the TLC studies (Lemberger et al., 1970; Lemberger et al., 1971, 1972b; Wall et al., 1983) when compared with the remaining studies (Supplemental Table 1). It should be noted that the TLC studies are from two different groups, so it is likely that the analytical methodology of TLC versus high performance liquid chromatography-UV or GC-MS contributed to the difference. Furthermore, it has been noted that less selective methodologies, such as TLC, are unable to distinguish THC isomers (Wall et al., 1972). This may be a potential reason for the higher Δ9-THC concentrations measured at later timepoints via TLC when compared with other methodologies. Since characterization of the terminal phase kinetics was crucial during model development, all TLC studies were removed and not used during THC or 11-OH-THC PBPK model training and verification. Lastly, the study from Brenneisen et al., 2010 was removed from the THC inhalation datasets since both THC and 11-OH-THC plasma concentrations were at least one order of magnitude smaller than all other studies. The reason for this discrepancy is unknown.
Details of the studies used for PBPK model training and development are shown in Table 1. All the studies had small sample sizes (n = 3–22), had mostly male subjects, and were of younger age. There is large interindividual as well as interstudy variability (see Supplemental Tables 1 and 2 for further details). A designation of naïve, casual, or chronic cannabis users was given for no previous cannabis use, cannabis use less than twice a week, and cannabis use greater than twice a week, respectively. The designation for chronic or casual cannabis use was chosen arbitrarily but guided by the general categorization of cannabis use frequency in the studies shown in Table 1.
Estimation of Population THC Volume of Distribution at Steady State and Inhalation Absorption Kinetics via Nonlinear Mixed Effects Modeling
Due to its high lipophilicity, THC volume of distribution at steady state (Vss) was severely overestimated in the initial PBPK model. Hence, the THC PBPK model–predicted Vss needed to be fixed to the observed THC Vss. As shown in Supplemental Table 1, the reported THC Vss ranged from 0.6–8.9 l/kg (Hunt and Jones, 1980; Kelly and Jones, 1992; Naef et al., 2004). However, the smaller Vss values are likely underestimates of the true Vss because of shorter total blood sampling times (e.g., 8 versus 24 hours). Additionally, there is no mechanistic information on lung disposition of THC in humans. Because of these limitations, the observed THC Vss as well as the absorption rate constant (ka) and bioavailability (Finh) after inhalation of THC were estimated by simultaneously fitting a three-compartment model without and with absorption compartment to the digitized plasma/serum concentration-time profiles after intravenous and inhalation administration of THC, respectively, using a nonlinear mixed effects (NLME) model in Phoenix 8.1, Certara (Princeton, NJ) (Supplemental Fig. 2). In using this approach, we assumed that the mean data set from each study constituted an individual subject. We recognize that this is not usual for NLME analyses of data, but our approach suffices for our goal, that is to obtain mean population parameters for THC Vss, Finh, and ka to use in our PBPK model.
THC PBPK Model Development and Verification in Healthy Nonpregnant Population
The linked THC/11-OH-THC PBPK model development in a healthy nonpregnant population using Simcyp (Simcyp population-based simulator version 16, Simcyp Limited, Sheffield, UK) is outlined in Fig. 1. Final input values for the THC PBPK model are listed in Table 2. Physicochemical properties for THC were collected from literature. We have previously measured the in vitro kinetics (Vmax and Km) of THC via CYP2C9 [major pathway, fraction metabolized (fm) = 0.91] and CYP2D6 (fm = 0.09) (Patilea-Vrana and Unadkat, 2019). These values, which are listed in Table 2, were extrapolated to in vivo using enzyme expression levels as listed in the Simcyp Virtual Healthy Population. Renal and biliary clearance were set to 0 l/h, since less than 5% of administered THC dose is excreted unchanged in urine and feces (Wall and Perez-Reyes, 1981). Within Simcyp, the Rodgers and Rowland method (Simcyp method 2) was selected for predicting tissue distribution because both THC and 11-OH-THC are neutral lipophilic compounds. The Rodgers and Rowland method accounts for binding to extracellular neutral lipids and neutral phospholipids and the affinity constant for binding to extracellular lipoproteins is predicted from the observed plasma unbound fraction (fup) and the blood-to-plasma (B/P) ratio. To match the predicted PBPK Vss to the NLME estimated Vss for THC, we elected to use a manually identified Kp scaler of 0.004. Such a Kp scaler was necessary because the Rodgers and Rowland method tends to overestimate Vss for highly lipophilic compounds (Haddad et al., 2000; Rodgers and Rowland, 2007).
Data after intravenous administration of THC from Ohlsson et al., 1982 was used as the training dataset for the THC PBPK model. This dataset was chosen because it had the longest sampling time of 48 hours. As described above, and shown in Supplemental Fig. 2F, sampling times greater than 12 and 56 hours are necessary to accurately estimate the THC CL and Vss, respectively. The THC intravenous PBPK model was verified using the remaining datasets after intravenous administration of THC.
Simulations of THC plasma/serum concentrations after inhalation of THC were conducted using the THC intravenous PBPK model with NLME estimated absorption parameters (ka and Finh). To note, no change was made in the NLME model for chronic versus casual cannabis users. Furthermore, optimization of THC Finh in the PBPK model for casual and chronic users was conducted via manual sensitivity analysis. Identification of the optimum THC Finh was driven primarily by the observed THC plasma/serum AUC0-t and not Cmax, since AUC0-t after THC inhalation was deemed a more robust parameter during model optimization.
To note, no THC parameters required optimization during the initial THC PBPK model development. During the development of the 11-OH-THC metabolite PBPK model, the 11-OH-THC fup was lowered via manual sensitivity analysis (further explanation below). We believed THC fup may have also been overpredicted, since THC fup was measured via the same methodology as 11-OH-THC. Since THC AUC0-t after intravenous administration or inhalation of THC was not sensitive to changes to fup (because THC clearance is hepatic blood flow limited), we used the Simcyp prediction toolbox to estimate THC fup. This change was incorporated in the final THC PBPK model.
11-OH-THC PBPK Model Training and Verification in Healthy Nonpregnant Population
Development of 11-OH-THC metabolite PBPK model is shown in Fig. 1. Input values for the final 11-OH-THC PBPK model are listed in Table 1. As for THC, physicochemical properties were collected from literature. Due to similar high lipophilicity, we assumed the same percentage of 11-OH-THC bound to lipoprotein (63%) as THC. The Rodgers and Rowland method (Simcyp method 2), using a lipid binding scaler, was used to predict 11-OH-THC distribution. The lipid binding scaler (ψ) back calculates the maximum extent of cellular lipid binding by correlating neutral lipid and neutral phospholipid binding in red blood cells to that in tissue. Since there was no available observed quality data on 11-OH-THC Vss to enable a similar strategy as for THC Vss, and the Rodgers and Rowland prediction method greatly overestimates Vss for compounds with logPo:w > 4 (Haddad et al., 2000; Rodgers and Rowland, 2007), the lipid binding scaler was a reasonable alternative adjustment for predicting 11-OH-THC Vss.
We have previously quantified the enzyme kinetics for 11-OH-THC clearance in vitro (Table 2) and found that UGT enzymes (fm = 0.60), CYP3A4 (fm = 0.18), and CYP2C9 (fm = 0.15) are the drug metabolizing enzymes responsible for hepatic metabolism of 11-OH-THC (Patilea-Vrana and Unadkat, 2019). Renal and biliary clearance were set to 0 l/h for 11-OH-THC. There is negligible excretion of unchanged 11-OH-THC in the urine, but up to 20% of radiolabeled 11-OH-THC dose is excreted unchanged in feces (Lemberger et al., 1972, 1973). However, the unchanged 11-OH-THC in feces may be 11-OH-THC deconjugated by bacterial β-glucuronidases. Furthermore, there is no current information on the extent (if any) of transport of 11-OH-THC by biliary efflux transporters, such as P-gp or BCRP.
The only datasets where 11-OH-THC was administered (as a parent) used TLC to quantify 11-OH-THC plasma concentrations(Lemberger et al., 1972a, 1973). For the reasons outlined above, these datasets were not considered further. There was only one remaining dataset with measured 11-OH-THC concentrations after intravenous administration of THC (Naef et al., 2004) available for 11-OH-THC PBPK model training. Due to this limitation, the 11-OH-THC PBPK metabolite model was optimized using observed 11-OH-THC data after intravenous administration of THC, and then verified using the observed 11-OH-THC plasma/serum concentration-time data after THC inhalation. Since the THC inhalation PBPK model was unable to be verified, THC Finh was optimized for each inhalation dataset so that the observed THC plasma/serum AUC0-t and Cmax approximated the observed THC values. An optimized THC Finh was necessary to ensure that the AUC and concentration-time profile of the parent compound (THC) was well characterized to verify the disposition of the metabolite (11-OH-THC).
During model development, 11-OH-THC AUC0-t after administration of THC was initially underestimated. To recapitulate observations, we considered the following parameters for optimization to increase the simulated 11-OH-THC AUC0-t after administration of THC: increase formation (fm) via CYP2C9, decrease 11-OH-THC intrinsic clearance (CLint), increase fraction unbound in human liver microsome incubation (fuinc), and decrease 11-OH-THC fup. We chose to optimize 11-OH-THC fup because this was a sensitive parameter that was experimentally derived and one for which we had the lowest confidence.
PBPK Model Verification and Performance Assessment
For each verification dataset, 10 trials were run, and the trial design was set to match the number of subjects, age range, and proportion of females as shown in Table 1. The simulated PK parameters needed to be within the 95% confidence interval of the observed value to meet our success criteria. Since all the observed studies had small sample size (<30 subjects), a t-distribution was used to calculate the 95% confidence interval. THC and 11-OH-THC PBPK model verification was performed by comparing observed and simulated THC AUC0-t after intravenous administration of THC, and 11-OH-THC Cmax and 11-OH-THC AUC0-t after THC inhalation. Of note, no success criteria were assigned to THC AUC0-t and Cmax after THC inhalation because Finh was individually optimized to verify the 11-OH-THC PBPK metabolite model. Bias and precision were calculated via the mean relative error (MRE) and relative root mean square error (rRMSE) as shown in Eq. 1 and 2, respectively (Sheiner and Beal, 1981).
Extrapolation of THC and 11-OH-THC PBPK Model from Healthy Nonpregnant Population to Pregnant Women
The THC/11-OH-THC PBPK model verified in a healthy nonpregnant population (Fig. 1) was extrapolated to pregnant women by incorporating gestational age–dependent physiologic changes. Based on phenytoin (a CYP2C9 substrate), pharmacokinetic data in pregnant and nonpregnant women, as well as PBPK modeling, our laboratory has determined that CYP2C9 activity increases 40%, 50%, and 60% during the first (T1), second (T2), and third trimester (T3), respectively (Ke et al., 2014). Based off midazolam (CYP3A4 substrate) pharmacokinetics in pregnant and nonpregnant women as well as in vitro studies and PBPK modeling, our laboratory has shown that CYP3A4 activity increases 100% throughout pregnancy (T1–T3) (Hebert et al., 2008; Ke et al., 2012; Zhang et al., 2015). There is no significant change in zidovudine (a UGT2B7 substrate) pharmacokinetics during pregnancy, suggesting there is no increase in UGT2B7 activity during pregnancy (O'Sullivan et al., 1993). Because the change, if any, in UGT1A9 activity during pregnancy is unknown, it was assumed to be unaffected during pregnancy (Anderson, 2005; Tasnif et al., 2016).
The final linked THC/11-OH-THC PBPK model developed for a healthy nonpregnant population was used to simulate disposition of THC and 11-OH-THC after inhalation of THC in pregnant women using the Simcyp virtual pregnancy population. Within the virtual pregnancy population, CYP2C9 expression was increased by 40%, 50%, and 60%, and CYP3A4 expression was increased 100%, 100%, and 100% for T1, T2, and T3, respectively. The trial design consisted of 10 trials with 10 subjects per trial of women age 20-45 at gestational weeks (GWs) 12, 26, and 40 to represent T1, T2, and T3, respectively.
To our knowledge, only one human clinical study has reported serum concentrations of THC and 11-nor-9-carboxy-Δ9-THC (COOH-THC), the main secondary and nonpsychoactive metabolite of THC, in maternal and umbilical cord blood at delivery (Blackard and Tennes, 1984). Concentrations of 11-OH-THC were not reported by Blackard and Tennes, 1984. This study used GC-MS to measure THC concentrations, and per our criteria defined earlier, we considered the data to be analytically robust. Ten subjects who smoked cannabis daily during the third trimester participated in the study. Information on the time since last smoking cannabis was provided, but the THC dose was not. The THC dose the subjects may have inhaled was calculated by multiplying the average percentage of THC (weight/weight) (ElSohly et al., 2000) in 1984 (the year the study was conducted) with the average weight of a joint (Mariani et al., 2011). The THC dose was approximated to be 21.8 mg. Simulations were performed using Finh of 0.35, a value estimated for chronic cannabis use during the PBPK model development after inhalation of THC.
Fetal serum concentrations of THC were simulated using a maternal-fetal PBPK (m-f-PBPK) model previously described (Zhang et al., 2017; Zhang and Unadkat, 2017). Since maternal THC serum concentrations drive fetal serum concentrations, adjustments were made to the THC dose for each subject, so the simulated maternal THC serum concentrations matched the observed data. Due to high lipophilicity, it was assumed THC is placental blood flow limited and, as such, the placental passive diffusion clearance was set to 500 l/h. Placental or fetal THC metabolism was assumed to be negligible. The simulations were first run with no placental efflux transport, and simulated fetal-to-maternal (F/M) serum concentrations were compared with observed data. Due to an overprediction of THC F/M ratio, we explored the impact of placental efflux transport required to recapitulate the observed fetal data. To match the observed fetal concentrations, the fraction transported (ft) via active placental efflux (expressed as a fraction of active efflux to active efflux plus passive diffusion clearance) was adjusted so that the simulated fetal THC serum concentrations matched the observed values.
Results
THC Vss and Inhalation Absorption Kinetics Estimates via NLME Model
THC Vss, estimated via an NLME model using the mean concentration data from 11 datasets after intravenous administration of THC was 6.5 l/kg (39% RSE). The NLME model estimates for THC ka and Finh, estimated using the mean concentration data from 29 datasets after inhalation of THC, was 12 (12% RSE) min−1 and 0.22 (33% RSE), respectively. The goodness-of-fit plot in Supplemental Fig. 2 shows good model fit of the observed THC plasma/serum concentrations after intravenous and inhalation of THC.
THC PBPK Model Development and Verification in Healthy Nonpregnant Population
The in vitro to in vivo extrapolation of THC and 11-OH-THC enzyme kinetics based on our previous quantification was successful (Patilea-Vrana and Unadkat, 2019). The only PBPK model optimization performed was adjustment of THC fup from 0.011 to 0.0022 and 11-OH-THC fup from 0.012 to 0.0050. The final PBPK model input parameters used are shown in Table 2.
During model verification after intravenous administration of THC, the PBPK model–predicted THC plasma AUC0-t met our success criteria for seven out of eight datasets (Table 3). For the Lindgren_1981a dataset, the simulated plasma AUC0-t was underestimated, however, it was close to the 95% confidence interval limit. Nevertheless, all simulated THC plasma AUC0-t were within 2-fold of the observed value (Table 3). The bias (MRE) and precision (rRMSE) for the predicted THC AUC0-t after intravenous administration of THC were −6% and 20%, respectively. Overall, the predicted THC plasma concentration-time profiles were similar to the observed profiles, with 70% of the simulated values within 2-fold of the observed values (Supplemental Fig. 3). Given the success criteria, bias and precision, the THC PBPK intravenous model performed well.
Using the verified THC intravenous PBPK model and the NLME estimates for THC inhalation absorption (ka = 12 hour−1 and Finh = 0.22, see Supplemental Fig. 2), THC plasma/serum AUC0-t and Cmax after inhalation of THC were simulated and compared with observed data (Supplemental Fig. 4). Of the 27 THC inhalation datasets, 13 did not meet the success criteria for AUC0-t (Supplemental Table 3). Interestingly, there was very little bias (MRE = 1%) but poor precision (rRMSE = 88%). This was in part because THC plasma/serum AUC0-t and Cmax were generally overpredicted for casual cannabis users and underpredicted for chronic users (Fig. 2, A and B). For these reasons, the THC inhalation PBPK model could not be verified. For any simulated THC plasma/serum AUC0-t that did not initially fall within the 95% confidence interval of the observed value, Finh was manually optimized to approximate the observed THC AUC0-t value (Supplemental Table 3). As shown in Fig. 2C, chronic users had a higher Finh compared with casual users (mean ± S.D. Finh = 0.35 ± 0.19 and 0.19 ± 0.14, respectively).
11-OH-THC PBPK Model Development and Verification in Healthy Nonpregnant Population
The 11-OH-THC PBPK model was developed using the only available dataset that measured 11-OH-THC plasma concentrations after THC intravenous administration (Naef et al., 2004) (Supplemental Fig. 3C) and then verified using datasets that measured 11-OH-THC plasma/serum concentrations after inhalation of THC using optimized THC Finh (Supplemental Fig. 4; Supplemental Table 3; Table 4). Twelve out of the 16 simulated 11-OH-THC plasma/serum Cmax met the success criteria, whereas 12 out of 14 simulated 11-OH-THC plasma/serum AUC0-t met the success criteria (Table 4). To note, the two datasets from Newmeyer et al., 2017 only reported 11-OH-THC blood Cmax and not AUC0-t. For the prediction of plasma/serum concentrations of 11-OH-THC, the bias and precision were as follows: 87% and 151% for Cmax and 15% and 60% for AUC0-t, respectively. As shown in Fig. 2D, although 11-OH-THC Cmax tended to be overestimated, all the values (except for the Newmeyer_2017a dataset) were within 2-fold of the observed Cmax. Furthermore, as shown in Fig. 2E, except for the Huestis_1992b dataset, all the simulated AUC0-t were within 2-fold of the observed values. Due to these outliers, the precision for 11-OH-THC Cmax and AUC0-t was worsened. Further inspection showed good agreement between the simulated versus observed mean ± S.D. 11-OH-THC/THC plasma AUC ratio (M/P AUCR) of 0.29 ± 0.05 and 0.29 ± 0.15, respectively (Fig. 2F). Overall, the simulated 11-OH-THC plasma/serum concentrations were comparable to the observed values, however, there was variability among the studies. Indeed, 80% and 79% of the simulated THC and 11-OH-THC plasma/serum concentrations, respectively, were within 2-fold of the observed value (Fig. 2, G and H). Pooling together the success criteria, bias, precision, and the additional PBPK model performance, the 11-OH-THC PBPK model performed well.
Using the final linked THC/11-OH-THC PBPK model, sensitivity analysis was performed to identify parameters that impact THC CL, THC Vss, and M/P AUCR (Fig. 3). THC predicted plasma CL was 57 l/h (blood CL of 86 l/h), making THC a high clearance compound. Since THC clearance is blood flow limited, THC CL was not sensitive to changes in total CLint, fup, or fuinc. Due to high lipophilicity, THC Vss was most sensitive to fup and distribution into the adipose. As anticipated, the metabolite (11-OH-THC) to parent (THC) AUC ratio (M/P AUCR) was sensitive to changes in 11-OH-THC formation (fm and THC CLplasma) and changes to 11-OH-THC elimination (11-OH-THC CLint). A sensitivity analysis was performed for fuinc since the CLint values used in the PBPK model are in vitro observed values adjusted for microsomal binding and, therefore, errors in fuinc will reflect in error in total CLint values. To reflect potential errors in CLint due to fuinc measurements, fuinc was changed by the same proportion for both THC and 11-OH-THC in the sensitivity analysis, which effectively changed the THC and 11-OH-THC CLint by the same proportion. As shown in Fig. 3, fuinc impacts THC and 11-OH-THC disproportionally, since fuinc and therefore CLint was a sensitive parameter to M/P AUCR but not THC CL. To further explain the impact on M/P AUCR, a simulation was run where the fupCLint of THC and 11-OH-THC were decreased by 10 or 100-fold, whereas fm and 11-OH-THC CLint formation-to-elimination ratio was maintained the same (Supplemental Table 4). 11-OH-THC plasma AUC decreased proportional to its formation CLint, whereas THC plasma AUC did not change proportionally to its CLint.
Extrapolation of THC and 11-OH-THC PBPK Model From Healthy Nonpregnant Population to Pregnant Women
Gestational age–dependent changes were applied to the final linked THC/11-OH-THC PBPK model verified in healthy nonpregnant population, and THC/11-OH-THC disposition was simulated at the end of T1 (GW = 12), T2 (GW = 26), and T3 (GW = 40). As shown in Fig. 4, A and B, the overall simulated THC plasma concentration-time profile or AUC did not change during pregnancy. The simulated plasma 11-OH-THC AUC decreased 28%, 37%, and 41% during T1, T2, and T3, respectively.
To investigate the impact of the gestational age–dependent physiologic changes on the observed THC and 11-OH-THC PK changes during pregnancy, a systematic sensitivity analysis was performed. As shown in Table 5, THC plasma AUC was not sensitive to any of the gestational age–dependent changes. 11-OH-THC plasma AUC was most sensitive to the increase in 11-OH-THC fup and increase in the cytochrome P450 enzyme expression (represented by increase in CYP3A4 and CYP2C9 Vmax). Even though 50% increase in CYP2C9 activity led to 50% increase in the intrinsic formation clearance of 11-OH-THC (represented by THC CYP2C9 Vmax), the overall fm to 11-OH-THC increased from 0.91 to 0.94. As such, the increase in intrinsic formation clearance of 11-OH-THC was not a sensitive parameter. Collectively, the simulated decrease in 11-OH-THC AUC during pregnancy was due to an increase in unbound elimination clearance of 11-OH-THC.
To test whether the extrapolation of THC from healthy nonpregnant to pregnant women was successful, THC plasma/serum concentrations were simulated after inhalation of an average joint with THC concentrations in cannabis reflective of that available in 1984 and compared with the observed data in pregnant women (Fig. 4C). Although dosing information is missing from Blackard and Tennes (1984), the simulated serum concentrations after inhalation of THC are comparable to the observed data. Although we cannot verify the THC PBPK model in pregnant women, we have confidence in the simulated output during pregnancy for THC and 11-OH-THC, since the methodology of extrapolating from healthy nonpregnant population to pregnant women has been verified by us for multiple drugs (Ke et al., 2012, 2014; Zhang and Unadkat, 2017).
THC fetal serum concentrations were simulated using a m-f-PBPK model. The dose for each subject was adjusted in order for the simulated THC maternal serum concentration to approximate the observed values. This was performed to ensure that maternal concentrations that drive fetal THC concentrations were well predicted. The observed THC F/M ratio for the three subjects with available data were 0.17, 0.25, and 0.38 (Supplemental Table 5). This suggests fetal exposure of THC is limited by placental efflux, placental metabolism, fetal metabolism, and/or placental sequestration. Indeed, simulations run in the absence of any placental and fetal metabolism/transport, the simulated F/M values overpredicted the observed values. Placental efflux (presumably by P-gp and/or BCRP) ft of 0.88, 0.92, and 0.93 was necessary to recapitulate the observed fetal THC serum concentrations (Supplemental Table 5).
Discussion
Several THC PBPK models in healthy nonpregnant populations have been previously developed. Wolowich et al., 2019 developed a linked THC/11-OH-THC/COOH-THC minimal PBPK (mPBPK) model for THC intravenous and oral administration using a hybrid bottom-up and top-down approach. The authors attributed the CLint of THC and 11-OH-THC entirely to CYP2C9, and they estimated these parameters by fitting the mPBPK model data after intravenous and oral (PO) administration of THC. Methaneethorn et al., 2020 developed a THC PBPK model for intravenous, PO, smoking, and vaporization of THC by extrapolating a previously developed PBPK model in mice, rats, and pigs to humans. The authors used observed values for systemic THC CL, and such, the PBPK model is not mechanistic. As detailed below, our THC/11-OH-THC PBPK model adds to the above contributions in several important ways. First, our PBPK model is built using a bottom-up approach that integrates full mechanistic PK information (such as metabolism of 11-OH-THC via CYP2C9, CYP3A, and UGTs). Second, we performed extensive verification of the THC and 11-OH-THC PBPK models in healthy nonpregnant population after intravenous and inhalation administration of THC. Finally, we are the first to use a THC/11-OH-THC PBPK model to predict the disposition of THC/11-OH-THC in pregnant women.
The meta-analysis of available THC data after intravenous administration and inhalation of THC provides insights previously not recognized. First, studies that used TLC to quantify THC and 11-OH-THC (Lemberger et al., 1970, 1971, 1972b; Wall et al., 1983) were different from the remaining studies, likely due to the nonspecific assay used to quantify cannabinoid plasma concentrations. Therefore, caution should be exercised when interpreting THC/11-OH-THC PK parameters from these studies. Second, NLME modeling of the mean THC concentrations from 25 different studies (40 datasets) that vary widely in sampling time window and sampling density allowed for the most comprehensive analysis of the population mean Vss, Finh, and ka of THC. Lastly, we showed that at least 12 and 56 hours are necessary to capture enough of THC PK profile to accurately calculate its CL and Vss. That is, the large variability in THC reported PK parameters, especially Vss, can be attributed, in part, to insufficient sampling time windows.
The foundation of our bottom-up approach was the extrapolation of in vitro mechanistic PK information previously quantified in vitro (Patilea-Vrana and Unadkat, 2019). The only parameters that required optimization during PBPK model building were fup of THC and 11-OH-THC, which were previously measured by us using ultracentrifugation (Patilea-Vrana and Unadkat, 2019). For drugs that are highly protein bound, it is possible that THC binding to lipoproteins (Klausner et al., 1975) and contamination of the unbound fraction from these lipoproteins led to an overestimation of fup (Brockman et al., 2015). Since THC CLint exceeds hepatic blood flow, THC plasma CL is blood flow limited and not sensitive to changes in its CLint. Therefore, the in vitro estimated THC CLint could be several folds off, and the THC PBPK model would still perform well. The mean simulated and observed M/P AUCR after inhalation of THC were the same (Fig. 2F). This means that the combination of fm, 11-OH-THC CLint, and fup estimates must be accurate. When considering the impact of CLint on both THC and 11-OH-THC disposition as outlined above, we conclude that the in vitro to in vivo extrapolation of the mechanistic PK information previously quantified (Patilea-Vrana and Unadkat, 2019) was successful. Further PBPK model development using THC and 11-OH-THC disposition after oral administration of THC will help identify the accuracy of the extrapolated THC and 11-OH-THC CLint values.
Initially, using the NLME estimates for Finh of 0.22 and ka of 12 hour−1, our PBPK model over- and underestimated plasma/serum THC AUC0-t and Cmax for the casual and chronic users after THC inhalation, respectively. After optimization of Finh, the average estimated Finh for casual and chronic users was 0.19 and 0.35, respectively (Fig. 2C). Indeed, studies have observed an increase in THC plasma AUC after inhalation of THC in chronic versus casual users (Toennes et al., 2008; Desrosiers et al., 2014). Some have hypothesized that chronic users have an increased THC CL (Lemberger et al., 1971), but this is unlikely due to THC CL being blood flow limited. History of regular cannabis use is correlated with average volume inhaled per puff and puff volumes (McClure et al., 2012). Furthermore, chronic users develop a tolerance to the psychoactive effects of cannabis and require higher levels of THC to achieve the desired effect (Lee et al., 2013). Therefore, the different smoking topography and tolerance but not increase in THC CL between chronic and casual users likely leads to the difference in THC Finh.
A seemingly paradoxical situation is observed regarding the formation kinetics of 11-OH-THC. The half-life of 11-OH-THC is similar to that of THC, suggesting formation-rate limited kinetics, however, the in vitro formation clearance of 11-OH-THC is much greater than its elimination clearance (Patilea-Vrana and Unadkat, 2019). Interestingly, Wolowich et al. (2019) invoked a hepatic sinusoidal diffusional barrier for THC to help explain this discrepancy in their mPBPK model. A sinusoidal diffusional barrier is unlikely given the high lipophilicity of THC. In our sensitivity analysis, we showed that proportional changes to CLint of THC and 11-OH-THC impact THC CL and M/P AUCR disproportionally (Supplemental Table 4). This phenomenon can be explained by the M/P relationship, defined as: (Pang et al., 2008). The formation clearance of the metabolite, defined as fm*CLparent, can become blood/plasma-flow limited, whereas the metabolite intrinsic elimination clearance, defined by fup*CLintmetabolite, cannot. As such, even though the in vitro intrinsic formation of 11-OH-THC is much greater than its elimination, 11-OH-THC has apparent formation-rate limited kinetics because THC hepatic CL is limited by hepatic blood/plasma flow. This helps explain why the observed M/P AUCR is less than unity, but THC and 11-OH-THC have similar in vivo half-lives.
To our knowledge, this is the first example of extrapolation of a PBPK model of THC/11-OH-THC during pregnancy. Previous studies from our laboratory demonstrated successful prediction of drugs, such as midazolam, indinavir, and glyburide, by extrapolating from healthy nonpregnant population to pregnant women using PBPK M&S (Ke et al., 2012, 2014). Using the same approach, exposure of THC and 11-OH-THC after inhalation of THC was extrapolated from the THC/11-OH-THC PBPK model verified in nonpregnant population to pregnant women by accounting for gestational age–dependent changes. Although we were unable to verify the THC predictions during pregnancy since the study by Blackard and Tennes (1984) lacked THC dosing information, the current extrapolation using average THC content in joints in 1984 was comparable with observed data (Fig. 4C). Simulations shown in Fig. 4 predict that during pregnancy, THC AUC does not change, whereas 11-OH-THC AUC decreases by up to 41% by T3. As predicted by sensitivity analysis in Table 5, changes to 11-OH-THC are mainly driven by the overall increase in 11-OH-THC fupCLint, which is driven by an increase in both CYP2C9 and CYP3A4 activity and increase in plasma protein binding. Although the increase in CYP2C9 did not increase the THC fm to 11-OH-THC (0.91 to 0.94 for nonpregnant and T3, respectively), the impact may be more pronounced if this fm value was smaller. This may be true for subjects with CYP2C9 genetic polymorphism that confer decreased activity and therefore lower fm. However, it is unknown whether induction of polymorphic CYP2C9 during pregnancy is similar to that of the wild-type CYP2C9.
The observed THC F/M concentration ratios in humans ranged from 0.17 to 0.38 (Blackard and Tennes, 1984). A similar THC F/M ratio was observed in macaques (Bailey et al., 1987), indicating placental/fetal metabolism, placental efflux, and/or placental sequestration may limit exposure of THC to the fetus. Indeed, initial simulations assuming no placental and fetal metabolism or transport overestimated the THC F/M concentration ratio. Placental enzymes, including CYP1A1, UGT1A9, and UGT2B7, were observed to turn over THC and 11-OH-THC (UGTs only) in recombinant enzyme studies (Patilea-Vrana et al., 2018). However, there was minimal depletion of THC and 11-OH-THC in human placental microsomes (data not shown). Since THC is a substrate of P-gp and BCRP (Bonhomme-Faivre et al., 2008; Spiro et al., 2012), we explored the theoretical active placental efflux needed to recapitulate the observed F/M ratios in the m-f-PBPK model. An average placental efflux ft of 0.91 was estimated via sensitivity analysis (Supplemental Table 5). Placental efflux ft ranging from 0.33 for tenofovir to 0.94 for paclitaxel have been previously observed for drugs that are substrates of P-gp or BCRP (Han et al., 2018). These data and our simulations suggest that placental efflux transport may play a role in limiting fetal exposure to THC. The magnitude of THC placental active efflux as well as additional mechanisms that may contribute to THC fetal exposure needs to be experimentally characterized. Since the placental expression of P-gp and BCRP varies with gestational age (Anoshchenko et al., 2020), and since prenatal use of THC at different stages of pregnancy can differentially impact neonatal outcomes (Grzeskowiak et al., 2020), it is important to understand the fetal exposure of THC throughout pregnancy. Further investigation into THC and 11-OH-THC placental metabolism and transport (or other reasons for this low F/M ratio) are ongoing in our laboratory to further refine the fetal THC and 11-OH-THC predictions.
Overall, a linked THC/11-OH-THC PBPK model was built and verified after intravenous administration and inhalation of THC in a healthy nonpregnant population. This PBPK model provides the mechanistic foundation that can be used to extrapolate and then predict THC/11-OH-THC exposure after oral administration of THC, in special populations (e.g., maternal-fetal dyad, disease, genetic polymorphism), or in the presence of drug-drug interactions.
Acknowledgments
We would like to acknowledge the University of Washington (UW) Pharmacokinetics of Drug Abuse during Pregnancy team members for insightful discussion.
Authorship Contributions
Participated in research design: Patilea-Vrana, Unadkat.
Conducted experiments: Patilea-Vrana.
Contributed new reagents or analytic tools: Patilea-Vrana.
Performed data analysis: Patilea-Vrana.
Wrote or contributed to the writing of the manuscript: Patilea-Vrana, Unadkat.
Footnotes
- Received November 27, 2020.
- Accepted April 6, 2021.
↵1 Current affiliation: Seagen, Inc., Bothell, Washington.
This work was supported by the National Institutes of Health National Institute on Drug Abuse [Grant P01-DA032507] (to G.P-V. and J.D.U.) and the Rene Levy Fellowship to G.P-V.
The authors report no conflicts of interest.
This manuscript constituted part of G.P-V.’s dissertation: Patilea-Vrana GI (2019) Predicting Maternal-Fetal Cannabinoid Exposure During Pregnancy Using Physiologically-Based Pharmacokinetic Modeling and Simulation. Doctoral dissertation, University of Washington, Seattle, WA.
↵This article has supplemental material available at dmd.aspetjournals.org.
Abbreviations
- AUC
- area under the curve
- AUC0-t
- area under the curve from 0 to time t
- AUCR
- AUC ratio
- BCRP
- breast cancer resistance protein
- CL
- clearance
- CLint
- intrinsic clearance
- COOH-THC
- 11-nor-9-caroboxy-Δ9-THC
- Finh
- inhalation bioavailability
- F/M
- fetal to maternal
- fm
- fraction metabolized
- ft
- fraction transported
- fuinc
- fraction unbound in human liver microsome incubation
- fup
- fraction unbound in plasma
- GC-MS
- gas chromatograph–mass spectrometry
- GW
- gestational week
- Kp
- tissue:plasma partition coefficient
- m-f-PBPK
- maternal-fetal PBPK
- M/P
- metabolite to parent ratio
- mPBPK
- minimal PBPK
- MRE
- mean relative error
- M&S
- modeling and simulation
- NLME
- nonlinear mixed effects
- 11-OH-THC
- 11-hydroxy-Δ9-tetrahydrocannabinol
- PBPK
- physiologically based pharmacokinetic modeling
- P-gp
- P-glycoprotein
- PK
- pharmacokinetics
- rRMSE
- relative root mean square error
- RSE
- relative S.E.
- T1
- first trimester
- T2
- second trimester
- T3
- third trimester
- THC
- (−)-Δ9-tetrahydrocannabinol
- TLC
- thin-layer chromatography
- UGT
- UDP-glucuronosyltransferase
- Vss
- volume of distribution at steady state
- Copyright © 2021 by The American Society for Pharmacology and Experimental Therapeutics