Physiologically Based Pharmacokinetic Modeling of Cannabidiol, Delta‐9‐Tetrahydrocannabinol, and Their Metabolites in Healthy Adults After Administration by Multiple Routes

ABSTRACT

The two most extensively studied cannabinoids, cannabidiol (CBD) and delta‐9‐tetrahydrocannabinol (THC), are used for myriad conditions. THC is predominantly eliminated via the cytochromes P450 (CYPs), whereas CBD is eliminated through both CYPs and UDP‐glucuronosyltransferases (UGTs). The fractional contributions of these enzymes to cannabinoid metabolism have shown conflicting results among studies. Physiologically based pharmacokinetic (PBPK) models for CBD and THC and for drug–drug interaction studies involving CBD or THC as object drugs were developed and verified to improve estimates of these contributions. First, physicochemical and pharmacokinetic parameters for CBD, THC, and their metabolites (7‐OH‐CBD, 11‐OH‐THC, and 11‐COOH‐THC) were obtained from the literature or optimized. Second, PBPK base models were developed for CBD and THC after intravenous administration. Third, beginning with the intravenous models, absorption models were developed for CBD after oral and oromucosal spray administration and for THC after oral, inhalation, and oromucosal spray administration. The full models well‐captured the area under the concentration–time curve (AUC) and peak concentration (C _max) of CBD and THC from the verification dataset. Predicted AUC and C _max for CBD and 7‐OH‐CBD were within two‐fold of the observed data. For THC, 11‐OH‐THC, and 11‐COOH‐THC, 100%, 100%, and 83% of the predicted AUC values were within two‐fold, respectively, of the observed values; 100%, 92%, and 94% of the predicted C _max values, respectively, were within two‐fold of the observed values. The verified models could be used to help address critical public health needs, including assessing potential drug interaction risks involving CBD and THC.

Summary.

• What is the current knowledge on the topic?

  • ○
    CBD and THC are widely used cannabinoids, and several research groups have developed physiologically based pharmacokinetic (PBPK) models to improve the understanding of their pharmacokinetics. However, several knowledge gaps remain.
• What question did this study address?

  • ○
    Can PBPK models for CBD and THC involving different formulations, including oromucosal sprays, administered to healthy adults be developed and verified? Can key metabolites of CBD and THC be incorporated successfully into the CBD and THC PBPK models while optimizing enzyme contributions to CBD and THC metabolism?
• What does this study add to our knowledge?

  • ○
    Our comprehensive PBPK models encompass a variety of CBD and THC formulations and routes of administration. We successfully integrated 11‐COOH‐THC into a complete PBPK model of THC. Our results substantiate that the contribution by cytochrome P450 enzymes to CBD metabolism is higher than that by UDP‐glucuronosyltransferases.
• How might this change clinical pharmacology or translational science?

  • ○
    These PBPK models can be used for other applications, including predicting changes in systemic exposure to CBD, THC, and metabolites in combination with cytochrome P450 inhibitors or inducers, as well as in specific populations, to assess potential safety concerns.

1. Introduction

Cannabidiol (CBD) and delta‐9‐tetrahydrocannabinol (THC), two prominent cannabinoids isolated from the plant Cannabis sativa (cannabis), have been used for various medical purposes since the first century A.D. [1]. Both CBD and THC are psychoactive, mainly by acting at the cannabinoid receptor type 1, G protein‐coupled receptor 55, peroxisome proliferator‐activated receptor, and transient receptor potential vanilloid 1 channel in the central nervous system [2]. THC, but not CBD, is psychotoxic and can lead to psychotic disorder and addiction [3, 4]. The major primary metabolites of CBD and THC are 7‐hydroxy‐CBD (7‐OH‐CBD) and 11‐hydroxy‐THC (11‐OH‐THC), respectively. Both are psychoactive, and 11‐OH‐THC is psychotoxic [3, 5]. 11‐nor‐9‐carboxy‐delta‐9‐THC (11‐COOH‐THC) is an inactive secondary metabolite of THC and has been proposed as a biomarker for inhaled THC [6].

Cannabinoids continue to be used medically as prescription drugs. For example, the US Food and Drug Administration (FDA) approved a purified form of CBD (Epidiolex) for certain types of epilepsy, as well as synthetic THC (dronabinol) and a synthetic analog of THC (nabilone) for anorexia and cancer‐related nausea and vomiting [3]. Sativex, a combination of purified CBD and THC administrated via oromucosal spray, is approved in the United Kingdom and other countries for the treatment of multiple sclerosis [3]. Interest in CBD and THC as treatments for neurodegenerative disorders, gastrointestinal diseases, depression, pain, anxiety, and addiction continues to increase [7, 8].

CBD and THC are considered Biopharmaceutics Classification System (BCS) Class II drugs, which are characterized by high permeability and low aqueous solubility, resulting in low and variable bioavailability when administered orally (CBD: 6%–23%; THC: 4%–20%) [9]. Because of low oral bioavailability caused by extensive first‐pass metabolism, alternative formulations with higher bioavailability have been developed. These formulations include an oral solution for CBD (e.g., Epidiolex), an inhalation formulation for THC, and an oromucosal spray containing both CBD and THC (e.g., Sativex).

As aforementioned, both CBD and THC are extensively metabolized, particularly by the cytochromes P450 (CYPs), including CYP2C9, CYP2C19, and CYP3A. Additionally, CBD is metabolized by a lesser extent by the UDP‐glucuronosyltransferases (UGTs), including UGT1A9, UGT2B7, and UGT2B17 [10]. One in vitro study reported the fraction metabolized (f _m) for CBD to be 70% and 30% for CYPs and UGTs, respectively, whereas another study reported an f _m of 21% and 79%, respectively [10, 11]. Both CBD and THC inhibit CYP2C9 and 2C19 in vitro, and CBD also inhibits CYP3A4 [3]. More recently, 7‐OH‐CBD and 11‐OH‐THC were shown to inhibit some CYPs, including CYP2C9, 2C19, and 2D6 [12]. In vivo, marijuana smoking was suggested to induce CYP1A2 in adults [13]; a brownie containing a CBD‐dominant extract inhibited CYP2C19, CYP2C9, CYP3A, and CYP1A2 in healthy adults [14]; and a CBD oral solution inhibited CYP2C19 in patients with epilepsy [15].

Given the therapeutic and toxic potential of cannabinoids, a thorough understanding of their pharmacokinetics is critical. However, the large interindividual variability (Figures S1 and S2), along with the different formulations and administration routes, pose difficulties in integrating and characterizing their pharmacokinetics. Robust physiologically based pharmacokinetic (PBPK) models can provide estimates of drug exposure from different formulations and routes of administration in different populations. Several groups have developed PBPK models for CBD or THC that include different routes of administration [10, 11, 16, 17, 18, 19, 20, 21]. Some full‐PBPK models included 7‐OH‐CBD or 11‐OH‐THC. However, unresolved issues remain in describing the pharmacokinetics of CBD and THC. First, the metabolic pathways of CBD and THC need further verification [10, 11]. Only one study evaluated the f _m of CYPs and UGTs for CBD using drug–drug interaction (DDI) studies during the PBPK modeling process [10]. An analogous evaluation for THC has not been reported. Second, a PBPK model for the oromucosal spray administration of Sativex has not been reported. Finally, the potential use of 11‐COOH‐THC as a biomarker for THC pharmacodynamics warrants further characterization of 11‐COOH‐THC pharmacokinetics, and a full‐PBPK model for THC that includes this metabolite has not been reported.

Based on these knowledge gaps, the overarching objective of this study was to establish PBPK models for CBD, THC, and their metabolites. The aims were to (1) develop and verify PBPK models for CBD in healthy adults that incorporate 7‐OH‐CBD following intravenous (IV), oral, and oromucosal spray administration and (2) develop and verify PBPK models for THC in healthy adults that incorporate 11‐OH‐THC and 11‐COOH‐THC following IV, oral, inhalation, and oromucosal spray administration. The novelties of this study include the first PBPK model for Sativex, the first THC PBPK model that was verified with DDI studies, and the first full‐PBPK model for THC that incorporates 11‐COOH‐THC.

2. Methods

2.1. PBPK Software, Data Acquisition, and Parameter Assessment

PBPK models were developed using the Simcyp PBPK Simulator (version 22, Certara, Sheffield, UK) and the virtual healthy adult population. Ten trials were simulated using study designs that closely matched the corresponding clinical trials, ensuring consistency in the dosing regimen, number of subjects, age range, proportion of females, and sampling window. All accessible clinical data, including concentration‐time profiles, area under the concentration‐time curve from time zero to time of the last observed concentration (AUC_0‐t), and peak concentration (C _max), were sourced from the published literature and digitized using WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/). If not reported, AUC_0‐t was determined using the trapezoidal method. If C _max was not reported, the highest observed concentration was used. Simulated AUC_0‐t values were determined using the trapezoidal method.

2.2. Datasets for Model Establishment and Verification

Inclusion criteria for CBD and THC clinical studies used for PBPK model development and verification were (1) the studies involved healthy adult participants; (2) systemic exposure (AUC_0‐t, C _max) and/or concentration‐time profiles were provided; and (3) CBD was administered via the IV, oral, or oromucosal spray route and/or THC was administered via the IV, oral, inhalation, or oromucosal spray route. Exclusion criteria were (1) CBD studies involving the inhalation route reported limited exposure data and different inhalation methods [22, 23, 24, 25]; (2) clinical trial data for verification purposes were not available for studies involving CBD or THC formulations administered by the oral or oromucosal spray routes; and (3) studies that measured THC using thin‐layer chromatography, which cannot distinguish between THC isomers [19]. In total, the observed data used to develop the CBD models were derived from five published clinical trials comprising 16 dosing regimens, whereas data for the THC models were derived from 30 published clinical trials comprising 64 dosing regimens. The observed pharmacokinetic parameters for CBD, THC, and their metabolites are detailed in Tables S1–S3. Studies used for PBPK model training and verification are presented in Tables S1–S3.

2.3. PBPK Modeling of CBD and Metabolites

A “middle‐out” strategy was used to establish the CBD and 7‐OH‐CBD model. Figure 1 summarizes the general workflow for CBD and 7‐OH‐CBD model development. Model verification was applied at each step. The PBPK model for 7‐OH‐CBD was linked to the CBD model. Physicochemical and in vitro metabolic parameters were used as inputs (Table 1).

FIGURE 1.

General workflow for cannabidiol (CBD) and delta‐9‐tetrahydrocannabinol (THC) model development. PBPK, physiologically based pharmacokinetic; IV, intravenous; V _ss, volume of distribution at steady state; CL, clearance; DDI, drug–drug interaction; PK, pharmacokinetic.

TABLE 1.

Final input parameters for the cannabidiol (CBD) and 7‐hydroxy‐CBD (7‐OH‐CBD) model.

Parameter	CBD		7‐OH‐CBD
	Value	Reference	Value	Reference
Physicochemical and blood binding
MW (g/mol)	314.47	ChEMBL	330.5	PubChem
Log Po:w	6.33	ChEMBL	5.94	ChemSpider
Compound type	Monoprotic Acid	Yeung 2023 [10]	Monoprotic Acid	Same as CBD
pKa	9.13	ChEMBL	9.13	Same as CBD
B/P	0.67	Samara 1988 [46]	0.67	Same as CBD
f u,plasma	0.18	Vuong 2020 [47]	0.55	Optimized
Absorption
Model type	First‐order	/
f a	0.1 a , 0.8 b , 0.4 c , 0.3 d	Optimized	/	/
k a (h−1)	0.55 a , e	Bansal 2023 [11]	/	/
Lag time (h)	0.5 a , 1.5 e	Optimized; Bansal 2023 [11]	/	/
F inh	0.1 a	Optimized	/	/
Lung k a (h−1)	2 a	Optimized	/	/
Proportion of dose inhaled (%)	8 a	Same as THC	/	/
Distribution
Model type	Full PBPK	Full PBPK
V ss (L/kg)	5.2711	Simcyp predicted, Method 2	1.6007	Simcyp predicted, Method 1
Lipid‐binding scalar (Ψ)	0.0001516	Simcyp predicted	/	/
K p scalar	1.25	Optimized	0.1	Optimized
Elimination
Model type	Enzyme kinetics	In vivo clearance
f u,mic	1	Default value by Reverse Translational Tools (RTT)	/	/
CYP2C9 CLint (μL/min/pmol enzyme) metabolite formed: 7‐OH‐CBD	0.845	Retrograde from CLtot of 60 L/h (Ohlsson 1986 [22]), 10% of CLtot (f m from Beers 2021 [48])	/	/
CYP2C19 CLint (μL/min/pmol enzyme) metabolite formed: 7‐OH‐CBD	29.201	Retrograde from CLtot of 60 L/h, 21% of CLtot (f m from Beers 2021 [48])	/	/
CYP3A4 CLint (μL/min/pmol enzyme)	1.524	Retrograde from CLtot of 60 L/h, 37% of CLtot (f m from Beers 2021 [48])	/	/
UGT1A9 CLint (μL/min/pmol enzyme)	1.479	Retrograde from CLtot of 60 L/h, 14.6% of CLtot (f m from Mazur 2009 [49])	/	/
UGT2B7 CLint (μL/min/pmol enzyme)	0.604	Retrograde from CLtot of 60 L/h, 9.1% of CLtot (f m from Mazur 2009 [49])	/	/
UGT2B15 CLint (μL/min/pmol enzyme)	1.076	Retrograde from CLtot of 60 L/h, 7.3% of CLtot (f m from Mazur 2009 [49])	/	/
Additional clearance, Liver, HLM (μL/min/mg)	5.354	Retrograde from CLtot of 60 L/h, residual clearance	/	/
CLiv (L/h)	/	/	64.34	Optimized
CLR (L/h)	0		0
CLbile (L/h)	0		0
Interaction
f u,mic	1		1
CYP2C9 K i (μM)	0.19	Nasrin 2021 [12]	/	/
CYP2C19 K i (μM)	0.092	Nasrin 2021 [12]	/	/
CYP2D6 K i (μM)	0.31	Nasrin 2021 [12]	/	/
CYP3A4 K i (μM)	0.22	Nasrin 2021 [12]	/	/

Abbreviations: /, not applied or reported; B/P, blood‐to‐plasma partition ratio; CL_bile, biliary clearance; CL_int, in vitro intrinsic clearance; CL_iv, intravenous clearance; CL_R, renal clearance; CL_tot, total clearance; f _a, fraction absorbed from dosage form; F _inh, fraction of drug absorbed from the inhalation; f _m, relative contribution of a metabolic enzyme; f _u,mic, fraction of unbound drug in the in vitro microsomal incubation; f _u,plasma, fraction unbound in plasma; HLM, human liver microsome; k _a, first‐order absorption rate constant; K _i, inhibition constant; K _p scalar, scalar applied to all predicted tissue K _p values; Log P_o:w, neutral species octanol: buffer partition coefficient; MW, molecular weight; pK _a, negative base‐10 logarithm of the acid dissociation constant; THC, delta‐9‐tetrahydrocannabinol; V _ss, volume of distribution at steady state.

^a Oromucosal spray.

^b 750 mg oral solution (Epidiolex).

^c 1500 and 3000 mg oral solution (Epidiolex).

^d 4500 and 6000 mg oral solution (Epidiolex).

^e Oral solution (Epidiolex).

First, a base model for CBD incorporating clearance and volume of distribution was developed using clinical data from a single IV dose (20 mg) [22]. Elimination parameters for CBD, including in vitro intrinsic clearance (CL_int) for each enzyme and additional hepatic clearance, were calculated from the in vitro clearance fraction of CYPs and UGTs using the reverse‐translational tools within Simcyp. Because CBD is not a substrate for the bile salt export pump in vitro [24] and CBD was not detected in the urine in clinical trials [26, 27, 28], biliary clearance (CL_bile) and renal clearance (CL_R) were set to zero.

Second, absorption models were developed for oral administration, and the PBPK model for 7‐OH‐CBD was established and linked to the CBD model. Published clinical data from the literature were used to refine these models (Table S2). The CBD PBPK model with oral absorption was developed assuming first‐order absorption. Because orally administrated CBD showed nonlinear absorption in clinical trials [29], the fraction absorbed (f _a) was optimized to capture the observed t _max and C _max for low (750 mg), medium (1500 and 3000 mg), and high (4500 and 6000 mg) doses of oral CBD. Verification of the oral model included a single dose of CBD (1500 and 4500 mg). For the 7‐OH‐CBD model, fraction unbound in plasma (f _u,plasma) and volume of distribution at steady state (V _ss) were optimized.

Third, absorption by the oromucosal spray route was modeled as a blend of oral and lung absorption using the Simcyp inhalation absorption module. The absorption rate constant (k _a) for the oral part of the oromucosal spray model was the same as for the oral model. For the lung absorption part, f _a and k _a for the lung were optimized to capture the observed t _max and C _max. CBD and THC were co‐administered in the oromucosal sprays. To predict the inhibitory effect of CBD on THC metabolism, competitive inhibition of CYP2C9, 2C19, 2D6, and 3A4 by CBD was modeled using the in vitro inhibition constant (K _i) (Table 1). Finally, for the oromucosal spray route, the f _m for CBD metabolism was validated with three precipitants: ketoconazole (CYP3A4 inhibitor), rifampicin (CYP3A4 inducer), and omeprazole (CYP2C19 inhibitor). PBPK models for these precipitants were applied from the Simcyp library or published models. The parameter optimization of the model for CBD and its metabolites was guided by the model's ability to accurately describe the observed data and by local sensitivity analysis (Figures S3 and S4).

2.4. PBPK Modeling of THC and Metabolites

A similar middle‐out strategy was used to develop PBPK models for THC, 11‐OH‐THC, and 11‐COOH‐THC (Figure 1). First, an initial THC model was built, and clearance and volume of distribution were refined using clinical IV data from single doses of THC (2 and 5 mg). The maximum rate of metabolism (V _max) and Michaelis constant (K _m) for THC metabolism by relevant metabolizing enzymes were derived from published literature and models [19, 30]. The CL_int of CYP3A4 was optimized because it contributes to the formation of multiple THC metabolites and K _m values were unavailable. The fraction of unbound drug in the in vitro microsomal incubation (f _u,mic) for THC metabolism was estimated using the Simcyp prediction toolbox. CL_bile and CL_R for THC were set to zero because < 5% of THC is excreted through the bile and kidney [19, 24, 31].

Second, absorption models for the oral and inhalation routes were developed for THC, and the PBPK models for 11‐OH‐THC and 11‐COOH‐THC were established simultaneously and linked to their parent compounds. Absorption models and the metabolic parameters for THC were optimized using data from published clinical trials (Table S2). The THC PBPK model with oral absorption was developed assuming first‐order absorption. Absorption via inhalation was modeled using the Simcyp inhalation module. The f _a and k _a for the oral part of the inhalation models were consistent with the oral model. Parameters including f _a and k _a for lung absorption, and the proportion of dose inhaled were optimized to capture the observed t _max and C _max. The fraction of drug absorbed by inhalation (F _inh) was individually optimized because of the varying requirements of inhalation in clinical trials, such as smoking duration and breath‐holding time (Table S2). For 11‐COOH‐THC, oral clearance was obtained from a previous study [32]. A full PBPK model for the secondary metabolite is not supported by the Simcyp version used in this study; therefore, the distribution of 11‐COOH‐THC was modeled using the minimal PBPK model. An in vitro study reported that approximately 4% of 11‐OH‐THC is converted to 11‐COOH‐THC by CYP2C9, which accounts for approximately 8% of the CYP2C9‐mediated metabolites [33]. However, under this assumption, the predicted exposure to 11‐COOH‐THC was significantly lower than the observed exposure, even at the lowest V _ss (0.05 L/kg) allowed by Simcyp. Accordingly, we assumed that the CYP2C9‐mediated metabolism of 11‐OH‐THC was exclusively 11‐COOH‐THC formation.

Third, THC by the oromucosal spray route was modeled co‐administered with CBD. f _a and k _a for the oral absorption of the oromucosal spray were consistent with the oral model, and the parameters for lung absorption were optimized. To predict the inhibitory effect of THC on CBD metabolism, competitive inhibition of CYP2C9, 2C19, and 2D6 by THC and competitive inhibition of CYP2C9 and 2D6 by 11‐OH‐THC were modeled using relevant K _i values (Table 1). Finally, clinical DDI studies involving THC oromucosal spray as the object drug were used to validate the f _m for THC metabolism (Tables S1–S3). The parameter optimization of the model for THC and its metabolites was guided by the model's ability to accurately describe the observed data and by local sensitivity analysis (Figures S5 and S6).

2.5. Model Verification and Performance Assessment

The PBPK models were verified with observed data during the modeling process (Tables S1–S3). For each study included in the model verification, 10 trials were simulated using study designs that closely matched the corresponding clinical trials, ensuring consistency in the dosing regimen, number of subjects, age range, proportion of females, and sampling window. Model performance was assessed by comparing the observed clinical data with the mean values and the 5th to 95th percentile from the simulated pharmacokinetic profiles. Models were considered acceptable if the predicted AUC_0‐t or C _max were within two‐fold of observed values, a widely applied standard [34, 35, 36]. This criterion was not applied for inhaled THC because of the individual optimization of F _inh mentioned above. Instead, model performance was based on the performance of the models for the primary and secondary THC metabolites. The DDI model was considered acceptable if the predicted inhibition or induction effect on the AUC_0‐t and C _max of the object drug was within a two‐fold range of the observed effect.

3. Results

The f _m for CBD by CYPs and UGTs of 70% and 30%, respectively, was applied to the elimination model of CBD. The schematic diagram of the whole‐body PBPK model for CBD and 7‐OH‐CBD is presented in Figure S7. The final input parameters for the linked CBD and 7‐OH‐CBD model are listed in Table 1. CBD doses in the verification dataset ranged from 1.6 to 4500 mg (Table S1). Clinical trials using ascending doses showed that the oral bioavailability of CBD decreased as the doses increased [37]. Therefore, f _a of the CBD oral solution was optimized to 0.8 for the low dose (750 mg), 0.4 for the medium doses (1500 and 3000 mg), and 0.3 for the high doses (4500 and 6000 mg). The majority (78%) of the observed plasma CBD and 7‐OH‐CBD concentrations were within the 5th and 95th percentiles of the simulated concentration‐time profiles (examples in Figures S10 and S11). All AUC_0‐t and C _max predictions for CBD and 7‐OH‐CBD were within two‐fold of observations (Figure 2; Table S1).

FIGURE 2.

Predicted vs. observed AUC_0‐t (A, C) and C _max (B, D) for CBD and 7‐hydroxy‐CBD (7‐OH‐CBD) (A, B) and for THC, 11‐hydroxy‐THC (11‐OH‐THC), and 11‐nor‐9‐carboxy‐delta‐9‐THC (11‐COOH‐THC) (C, D) after administration of CBD or THC to healthy adults by multiple routes of administration.

The final input parameters for the linked THC, 11‐OH‐THC, and 11‐COOH‐THC models are listed in Table 2. The schematic diagrams of the PBPK model for THC, 11‐OH‐THC, and 11‐COOH‐THC are presented in Figures S8 and S9. THC doses in the verification dataset ranged from 1.25 to 86 mg (Table S2). All predicted AUC_0‐t and C _max values were within two‐fold of the observed values. The verification dataset for 11‐OH‐THC included studies involving IV, inhalation, and oromucosal spray administration of THC. All predicted 11‐OH‐THC plasma/blood AUC_0‐t values were within two‐fold of the observed values, and 92% of the predicted C _max were within two‐fold of the observations. For the 11‐COOH‐THC verification dataset consisting of IV and inhalation studies, 83% and 94% of the predicted AUC_0‐t and C _max, respectively, were within two‐fold of the observed values. Predicted and observed AUC_0‐t and C _max for all studies are presented in Table S2. Examples of predicted versus observed concentration‐time profiles for THC, 11‐OH‐THC, and 11‐COOH‐THC are shown in Figures S12–S14.

TABLE 2.

Final input parameters for the delta‐9‐tetrahydrocannabinol (THC), 11‐hydroxy‐THC (11‐OH‐THC), and 11‐nor‐9‐carboxy‐delta‐9‐THC (11‐COOH‐THC) model.

Parameter	THC		11‐OH‐THC		11‐COOH‐THC
	Value	Reference	Value	Reference	Value	Reference
Physicochemical and blood binding
MW (g/mol)	314.5	PubChem	330.5	PubChem	344.4	PubChem
Log Po:w	6.97	Patilea‐Vrana 2021 [19]	5.33	Patilea‐Vrana 2021 [19]	5.24	DrugBank
Compound type	Neutral	Patilea‐Vrana 2021 [19]	Neutral	Patilea‐Vrana 2021 [19]	Monoprotic Acid	DrugBank
pKa	/	/	/	/	4.02	DrugBank
B/P	0.667	Patilea‐Vrana 2021 [19]	0.625	Patilea‐Vrana 2021 [19]	0.667	Same as THC
f u,plasma	0.0022448	Simcyp predicted	0.01209	Simcyp predicted	0.101	Calculated based on Schwilke 2011 [50]
Absorption
Model type	First‐order		/	/
f a	0.45	Optimized	/	/	/	/
k a (h−1)	0.7	Optimized	/	/	/	/
Lag time (h)	0.75	Optimized	/	/	/	/
F inh	0.22 a , 0.3 b	Patilea‐Vrana 2021 [19]; Optimized	/	/	/	/
Lung k a (h−1)	12 a , 5 b	Patilea‐Vrana 2021 [19]; Optimized	/	/	/	/
Proportion of dose inhaled (%)	97.5 c , 8 b	Optimized	/	/	/	/
Distribution
Model type	Full PBPK		Full PBPK		Minimal PBPK Model
V ss (L/Kg)	5.1281	Wolowich 2019 [41], Method 1	4.5376	Simcyp predicted, Method 2	0.37	Optimized
Lipid Binding Scalar (Ψ)	/	/	0.059	Simcyp predicted	/	/
K p scalar	0.266	Optimized	0.5	Optimized	/	/
V sac (L/kg)	/	/	/	/	0.28	Optimized
Q (L/h)	/	/	/	/	9.64	Optimized
Elimination
Model type	Enzyme kinetics		Enzyme kinetics		In vivo clearance
f u,mic,CYP	0.38067	Simcyp predicted based on Watanabe 2007 [30]	0.06	Patilea‐Vrana 2021 [19]	/	/
CYP2C9 V max (pmol/min/pmol enzyme)	19.2	Watanabe 2007 [30]	/	/	/	/
CYP2C9 V max (pmol/min/mg protein)	/	/	59.23 d	Patilea‐Vrana 2021 [19]	/	/
CYP2C9 K m (K s) (μM)	0.07	Patilea‐Vrana 2019 [33]	0.5	Patilea‐Vrana 2021 [19]	/	/
CYP2C19 V max (pmol/min/pmol enzyme)	0.22	Watanabe 2007 [30]	/	/	/	/
CYP2C19 K m (K s) (μM)	0.01	Optimized	/	/	/	/
CYP2D6 V max (pmol/min/pmol enzyme)	0.01	Watanabe 2007 [30]	/	/	/	/
CYP2D6 K m (K s) (μM)	0.01	Optimized	/	/	/	/
CYP3A4 V max (pmol/min/mg protein)	/	/	1826	Patilea‐Vrana 2021 [19]	/	/
CYP3A4 K m (K s) (μM)	/	/	12.8	Patilea‐Vrana 2021 [19]	/	/
CYP3A4 CLint (μL/min/pmol enzyme)	100	Optimized based on IV and DDI studies	/	/	/	/
f u,mic,UGT	/	/	1	Mazur 2009 [49]	/	/
UGT1A9 V max (pmol/min/mg protein)	/	/	140	Mazur 2009 [49]	/	/
UGT1A9 K m (μM)	/	/	7.3	Mazur 2009 [49]	/	/
UGT1A10 V max (pmol/min/mg protein)	/	/	330	Mazur 2009 [49]	/	/
UGT1A10 K m (μM)	/	/	72	Mazur 2009 [49]	/	/
Clpo (L/h)	/	/	/	/	6.27	Sempio 2022 [32]
CLR (L/h)	0	Patilea‐Vrana 2021 [19]	0	Patilea‐Vrana 2021 [19]	0
CLbile (L/h)	0	Patilea‐Vrana 2021 [19]	0	Patilea‐Vrana 2021 [19]	0
Interaction
f u,mic	1		1		/
CYP2C9 K i (μM)	0.17	Nasrin 2021 [12]	0.21	Nasrin 2021 [12]	—	Nasrin 2021 [12]
CYP2C19 K i (μM)	0.21	Nasrin 2021 [12]	/	/	—	Nasrin 2021 [12]
CYP2D6 K i (μM)	0.28	Nasrin 2021 [12]	0.32	Nasrin 2021 [12]	—	Nasrin 2021 [12]

Abbreviations: —, no inhibition or induction; /, not applied or reported; B/P, blood‐to‐plasma partition ratio; CL_bile, biliary clearance; CL_int, in vitro intrinsic clearance; CL_po, oral clearance; CL_R, renal clearance; f _a, fraction absorbed from dosage form; F _inh, fraction of drug absorbed from the inhalation; f _m, relative contribution of a metabolic enzyme; f _u,mic, fraction of unbound drug in the in vitro microsomal incubation; f _u,plasma, fraction unbound in plasma; k _a, first‐order absorption rate constant; K _i, inhibition constant; K _m (K _s), Michaelis constant; K _p scalar, scalar applied to all predicted tissue K _p values; Log P_o:w, neutral species octanol: buffer partition coefficient; MW, molecular weight; pK_a, negative base‐10 logarithm of the acid dissociation constant; Q, intercompartmental clearance; V _max, maximum rate of metabolism; V _sac, volume of single adjusting compartment; V _ss, volume of distribution at steady state.

^a The THC F _inh for inhalation was obtained from Patilea‐Vrana GI et al. If the predicted plasma or whole blood AUC_0‐t or C _max was not within the two‐fold interval of the observations when using the reference THC F _inh of 0.22, THC F _inh was optimized to ensure that both the AUC_0‐t and C _max were within the two‐fold interval.

^b Oromucosal spray.

^c Inhaled.

^d The V _max of CYP2C9 is the sum of the reported V _max of CYP2C9 for the formation of 11‐COOH‐THC and the reported V _max of CYP2C9 for the formation of unknown metabolites.

4. Discussion

Several PBPK models have been developed for CBD and THC to improve our understanding of their disposition in the body. However, gaps remain with respect to routes of administration and incorporation of metabolites. Accordingly, we developed and verified the most comprehensive PBPK models to date for CBD, THC, and their key metabolites after administration of different formulations CBD and THC by various routes. Notably, these models provide more accurate estimates of AUC_0‐t and C _max for these cannabinoids while enhancing our understanding of the enzymatic contributions to their metabolism.

The fractional contributions of CYP and UGT enzymes for CBD were verified with DDI studies to improve estimates of these contributions. Three full‐PBPK models for CBD have been established that incorporated two combinations of f _m for CYPs and UGTs for CBD metabolism [10, 11, 17]. Only one PBPK model used clinical data involving inhibition of CBD metabolism by itraconazole (CYP3A4 inhibitor), fluconazole (CYP2C9 inhibitor), and rifampicin (CYP3A4 inducer) as verification datasets [10, 38]. The authors concluded that reducing f _m for CYPs improved DDI prediction accuracy. However, study design details, concentration–time profiles, and AUC calculation details were not provided [38]. As such, we used data obtained from another phase I DDI clinical trial to evaluate the f _m of CYPs and UGTs [39]. Specifically, ketoconazole (CYP3A4 inhibitor), rifampicin, and omeprazole (CYP2C19 inhibitor) were tested as precipitants. Our results support that 70% of CBD metabolism is mediated by CYPs and 30% by UGTs.

Comprehensive PBPK models for THC in healthy adults that incorporate 11‐OH‐THC and 11‐COOH‐THC following IV, oral, inhalation, and oromucosal spray administration have been developed and verified in this study. Four full‐PBPK models have been developed [18, 19, 20, 21]. Existing THC PBPK models typically include only CYP2C9 and CYP3A4 in the metabolic pathway of THC [19, 20, 21]. Regarding our THC PBPK model, we used the in vitro V _max for CYPs reported by Watanabe et al. [30] Because the K _m for CYP2C19‐ and CYP2D6‐mediated metabolism was not reported, this value was optimized to 0.01 μM, which aligns with the range of unbound K _m values determined in a recent in vitro study (0.005–0.014 μM, when f _u,mic,CYP = 0.38067) [40]. Notably, while only one minimal THC PBPK study included 11‐COOH‐THC [41], our THC full‐PBPK model included both 11‐OH‐THC and 11‐COOH‐THC for the first time. Incorporating 11‐COOH‐THC is a valuable component that refines the elimination pathways of 11‐OH‐THC, verifies the models of THC and 11‐OH‐THC, enhances predictive accuracy among various formulations and routes of administration. Concentrations of THC and its metabolites are critical for identifying cannabis‐impaired drivers. A meta‐regression analysis indicated that blood concentrations of 11‐COOH‐THC were associated with subjective impairments in driving and related cognitive skills following THC inhalation [6]. A robust PBPK model for 11‐COOH‐THC could help establish the relationship between THC dose and subjective impairment, predicting the risk of driving performance impairment because of THC.

CBD and THC were modeled simultaneously for the oromucosal spray because (1) both cannabinoids are contained in the marketed product (Sativex) and (2) a potential DDI between CBD and THC was considered. The oromucosal spray absorption model consisted of oral and lung absorption components, which have been used for other oromucosal spray models [42]. Our results suggested limited DDI potential between CBD and THC at low doses. When co‐administered in a 1:1 ratio, the predicted AUC_0‐t ratios with or without the precipitant were approximately unity for either CBD or THC as the object drug (5–20 mg CBD and 5.4–21.6 mg THC). These results are consistent with results from published clinical trials involving similar doses [43, 44]. A clinical trial involving 5.4 mg CBD and 10 mg THC showed no DDI effect [43]. Another DDI clinical trial also reported that 10 mg CBD did not significantly increase exposure to 9 mg THC; however, higher CBD doses (30 mg or 450 mg) significantly increase THC exposure [44]. A recent DDI clinical trial reported that CBD appeared to inhibit the metabolism of THC to 11‐OH‐THC at a higher dose when comparing the THC profile between a CBD‐dominant brownie (640 mg CBD/20 mg THC) and a THC‐dominant brownie (no CBD/20 mg THC) [45]. Collectively, we anticipate that the interaction may have clinical significance with higher doses of CBD or THC.

There are limitations to our study. First, the relative enzyme contributions for CBD and THC were extracted from published studies. Although DDI studies verified these contributions in our model, in vitro studies are still needed to comprehensively characterize the metabolites formed by various enzymes, especially the UGTs. DDI studies using UGT inhibitors and inducers are essential to verify the contribution of UGTs to CBD metabolism. Second, pharmacogenetic differences in the various drug metabolizing enzymes were not considered because only one THC clinical trial reported the CYP2C9 genotype for all subjects [41]. Because of the lack of actual distributions in clinical trials combined with small sample sizes, using the distribution of genotype from the simulated population may introduce more bias. Third, first‐order processes were used to describe the absorption of CBD and THC via the oral, inhalation, and oromucosal spray routes. Despite testing several mechanistic absorption models, we found that they could not fully capture the concentration profiles without optimizing multiple parameters. We anticipate that this limitation is because of the limited understanding of the absorption mechanisms for both CBD and THC. Some key parameters have not yet been characterized, such as intestinal solubility, bile‐partition coefficient, supersaturation ratio, and precipitation rate constant. Future studies will focus on developing more mechanistic absorption models for CBD and THC, applying the CBD and THC PBPK models in specific populations, including patients with hepatic impairments, and investigating the pharmacodynamic effects of CBD and THC based on their target organ exposure.

In summary, our study verified the fractional contributions of CYPs and UGTs to CBD and THC clearance using PBPK modeling. Our models included the most comprehensive routes of administration for medical use, demonstrating high predictive performance for CBD and THC. To our knowledge, this study is the first to include oromucosal spray administration of both CBD and THC into the development of full PBPK models for these two cannabinoids. The THC PBPK model was the first to be verified with DDI studies and the first full‐PBPK THC model that incorporates 11‐COOH‐THC. These models are expected to advance the understanding of the pharmacokinetics of CBD and THC and provide a more robust foundation for future studies, including evaluating their pharmacokinetics in specific populations and making DDI predictions. Ultimately, these models provide a more robust foundation for improving our understanding of cannabinoid exposure‐response relationships.

Author Contributions

J.D., L.Q., M.F.P., T.Z., and Z.Z. wrote the manuscript. Z.Z. designed the research. L.Q., T.Z., and Z.Z. performed the research. L.Q. analyzed the data.

Conflicts of Interest

J.D. is an employee of Simcyp Ltd/Certara, a company that provides PBPK Software. M.F.P. is a member of the Scientific Advisory Board for Simcyp Ltd/Certara. All other authors declared no competing interests for this work.

Supporting information

Data S1. Supporting Information.