Using a PBPK Model Incorporating Lymphatic Absorption to Predict Food Effect, Multiple Dosing, and Hepatic Impairment of Cannabidiol

ABSTRACT

Cannabidiol (CBD) is one of the most extensively studied cannabinoids and is used for myriad conditions. Its oral pharmacokinetics are complex, exhibiting non‐linear absorption, significant food effects, and variable exposure in hepatic impairment. Existing physiologically based pharmacokinetic (PBPK) models for oral CBD have largely relied on fitted first‐order absorption or fitted dissolution profiles, limiting their mechanistic and predictive capabilities and extrapolation, particularly regarding the mechanistic details of its absorption. This study developed and verified the first PBPK model for oral CBD with mechanism‐based oral absorption based on our prior published PBPK model. It incorporated mechanism‐based absorption using the multi‐layer gut wall within the advanced dissolution, absorption, and metabolism (M‐ADAM) model within Simcyp. Pharmacokinetic parameters for CBD or population parameters related to absorption were obtained from the literature or optimized. Some physiological parameters (e.g., luminal bile salt and lymph flow rate) were adjusted mechanistically to account for CBD's sesame oil formulation and meal characteristics. The model well captured the CBD concentration‐time profiles and key pharmacokinetic parameters, including area under the concentration‐time curve (AUC), peak concentration (C _max), and time to maximum concentration (t _max) across diverse doses, fed/fasted states, and in patients with hepatic impairment (mild, moderate, severe). Predictions were consistently within two‐fold of observed data. This work offers a robust foundation for the mechanistic understanding of CBD's complex oral absorption. The verified models can optimize CBD dosing strategies, predict potential drug–drug interactions, and evaluate CBD pharmacokinetics in various specific populations, ultimately contributing to safer and more effective clinical use.

Study Highlights.

• What is the current knowledge on this topic?

  • ○
    CBD is a widely used cannabinoid with complex oral pharmacokinetics, exhibiting nonlinear absorption and a significant food effect. CBD in specific conditions like hepatic impairment is also impacted by the complex absorption. However, existing pharmacokinetic models have not adequately incorporated a mechanistic description of these complex absorption processes, limiting their predictive capabilities under varying clinical conditions.
• What question did this study address?

  • ○
    Can a novel PBPK model, incorporating a mechanism‐based absorption model including lymphatic absorption, accurately predict the food effect, multiple‐dose pharmacokinetics, and impact of hepatic impairment on oral CBD pharmacokinetics?
• What does this study add to our knowledge?

  • ○
    This study provides the first comprehensive PBPK model for CBD that mechanistically describes the CBD oral absorption, successfully explaining its complex absorption, substantial food effect, and altered disposition in hepatic impairment.
• How might this change drug discovery, development, and/or therapeutics?

  • ○
    Our verified CBD PBPK enables rational CBD dosing optimization (especially for food effects), potentially reducing clinical food effect trials. It supports dose modifications for specific populations and mechanistic in silico prediction of drug–drug interactions. Our model can also be used as an example when modeling other lipophilic drugs with lymphatic absorption and to predict the food effect on new lipophilic drugs prior to clinical trials.

1. Introduction

The therapeutic use of cannabidiol (CBD), a non‐psychotoxic cannabinoid derived from the plant Cannabis sativa (cannabis), has seen a significant increase over the past decade [1]. Its purified form, Epidiolex (or Epidyolex in some countries), has been approved by the Food and Drug Administration (FDA) and other regulators to treat several types of epilepsy (daily dose from 5 to 25 mg/kg) [2]. Interest in CBD as a treatment for other conditions like psychotic disorders [3], cancers, and opioid use disorder continues to increase [4].

Despite its therapeutic potential, oral CBD presents complex pharmacokinetic challenges. As a Biopharmaceutics Classification System (BCS) Class II drug, CBD is highly lipophilic, with high permeability and low aqueous solubility. This is evident from an oral CBD formulation in sesame oil, where the drug is reported to be neither dissolved in simulated fasted gastric nor intestinal fluids [5]. To overcome this, many CBD oral formulations, including Epidiolex, contain high proportions of oil excipients [6]. CBD in Epidiolex is dissolved in sesame oil, which consists mostly of long‐chain triglycerides (LCT), to significantly enhance CBD bioavailability [6, 7, 8]. Orally administered CBD showed nonlinear absorption in clinical trials, a characteristic that remains incompletely understood in existing pharmacokinetic models [9].

A key factor contributing to CBD's complex absorption is its high intestinal lymphatic absorption, which allows it to partially bypass the first‐pass metabolism [7, 10]. CBD can be readily incorporated into chylomicrons within enterocytes and transported to the systemic circulation via the lymphatic system [10]. In a rat study, CBD concentrations have been detected in the lymph fluid, exhibiting an earlier time to maximum concentration (t _max) and higher concentrations when compared to plasma [7]. These findings indicate the significance of this pathway for CBD absorption. Therefore, a comprehensive understanding of CBD's nonlinear pharmacokinetics, particularly for oral administration, critically depends on accurately describing its lymphatic absorption.

The Epidiolex label recommends administration with meals [11]. The influence of food on CBD exposure is significant; the bioavailability of CBD can increase up to 4‐fold with food [11], a magnitude higher than observed for most other FDA‐approved drugs [6]. Interestingly, the t _max of CBD is almost the same between fasted and fed states when administered as Epidiolex [11]. This unusual finding may be attributed to lymphatic absorption, which was previously investigated in a PBPK study of halofantrine, a highly lipophilic drug [12]. The food content has a significant impact on CBD exposure. Both fat and calories can increase the bioavailability of CBD, with the high‐fat/high‐calorie meal having the greatest impact [11]. This may be mechanically linked to changes in lymph flow rate, a critical determinant of lymphatic drug absorption, which is known to increase with the intake of lipids or glucose [12, 13, 14]. Therefore, incorporating the dynamics of lymphatic absorption is critical for predicting how different types of meals or food intake affect CBD exposure and bioavailability.

Moreover, hepatic impairment poses another layer of complexity to CBD's pharmacokinetics. While hepatic impairment typically reduces hepatic and first‐pass metabolism [15], potentially increasing systemic exposure, it also notably increases the gastric lymph flow rate to approximately three times higher than in healthy individuals [16]. Without accounting for lymphatic absorption, the first‐pass metabolism of CBD could be overpredicted, and the overall impact of hepatic impairment on CBD exposure might be mischaracterized.

These complexities, both related to meal effects and altered disease states like hepatic impairment, highlight the need for a more comprehensive and mechanistic modeling approach than currently available. Although several quantitative models, including population pharmacokinetics models and physiologically‐based pharmacokinetic (PBPK) models, have been developed for oral CBD, all of them have primarily relied on fitting‐based first‐order absorption models or fitted dissolution profiles [17, 18, 19, 20, 21]. Moreover, these models have not adequately incorporated a mechanistic description of lymphatic absorption. This limitation significantly restricts their application in accurately predicting CBD exposure under varying clinical conditions, such as different meal compositions or in patients with hepatic impairment. While understanding the clinical implications of lymphatic absorption for meal effects and hepatic impairment is critical, mechanistic pharmacokinetic modeling of this pathway remains relatively underexplored. PBPK modeling, with its robust structure for describing oral absorption, offers a powerful framework to integrate lymphatic pathways. This approach has demonstrated success in predicting food effects for other highly lipophilic drugs, like halofantrine, where over 50% of the dose was absorbed via the lymphatic system in the fed state [12].

Based on these critical knowledge gaps, the study aimed to develop and verify a novel CBD PBPK model incorporating mechanism‐based absorption. This model was subsequently used to accurately predict the impact of food (including single and multiple doses), and to simulate the effect of hepatic impairment on CBD pharmacokinetics, under both single‐ and multiple‐dose regimens, considering the interplay with food effects.

2. Method

2.1. PBPK Software, Parameter Assessment, and Data Acquisition

PBPK models were developed using the Simcyp PBPK Simulator (version 24, Certara, Sheffield, UK). The Livermore ordinary differential equation solver was used to solve the stiff equations. Simulations used the virtual healthy adult population, virtual North European Caucasian population, and virtual populations with hepatic impairment (Cirrhosis Child–Pugh grade A [CP‐A], CP‐B, CP‐C) to accurately simulate the population in each trial. For each clinical trial, 10 virtual trials, each comprising 10 subjects, were simulated using study designs that closely matched the corresponding clinical trials, ensuring consistency in the population demographics, age range, proportion of females, dosing regimen, 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 ), peak concentration (C _max), and t _max, were sourced from the published literature. Concentration‐time profiles were digitized using WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/).

2.2. Datasets for Model Establishment and Verification

Inclusion criteria for CBD clinical studies used for advanced oral absorption and food effect PBPK model development and verification were (1) CBD was administered orally as Epidiolex in either fasted or fed states, once or daily; (2) the studies involved healthy young adult participants (age: 18–65); and (3) pharmacokinetic parameters (AUC_0−t , C _max, t _max) and/or concentration‐time profiles were provided. Exclusion criteria were (1) co‐administration of drugs or drinks with potential drug–drug interactions with CBD; (2) lack of reported CBD exposures; (3) undocumented meal events during the study; and (4) inconsistent dosing conditions, for example, when the drug was not always administered in a consistent fasted or fed state during multiple‐doses administration were given.

In total, the observed data used to develop the CBD model with advanced absorption were derived from five published clinical trials comprising 17 dosing regimens [9, 11, 22, 23, 24]. The food effect analysis included seven dosing regimens with various fed states, such as whole milk, low‐fat meal, regular fed (fat percentage between low‐fat and high‐fat meals), and high‐fat meal [11, 23]. Studies used for CBD PBPK model training and verification, and their observed pharmacokinetic parameters are presented in Table 1. For the hepatic impairment model, CBD data from subjects with moderate hepatic impairment were used in model development, while data for mild and severe hepatic impairments were used for verification [22].

TABLE 1.

Physiologically based pharmacokinetic (PBPK) model‐predicted and observed cannabidiol (CBD) exposure.

Dose	Model/Verification	Absorption rate scalars	Stomach MRT (h)	Observed AUC0−t (h·ng/mL)	Predicted AUC0−t (h·ng/mL)	AUC Predicted: Observed	Observed C max (ng/mL)	Predicted C max (ng/mL)	C max Predicted: Observed	Observed t max (h)	Predicted t max (h)	t max Predicted: Observed	References
Single dose
200 mg oral	V	1.00	Default	449.00	436.36	0.97	148.00	166.56	1.13	2.30	2.35	1.02	22
750 mg oral, study 1	M	1.00	Predicted	1070.00	907.75	0.85	290.80	287.32	0.99	5.00	2.65	0.53	9
750 mg oral, study 2	V	1.00	Predicted	1077.00	1085.01	1.01	187.00	271.49	1.45	4.00	2.70	0.68	11
1500 mg oral	V	1.00	Predicted	1517.00	1301.72	0.86	292.40	321.20	1.10	4.00	2.65	0.66	9
3000 mg oral	M	2.00	Predicted	2669.00	2568.61	0.96	533.00	490.39	0.92	5.00	2.70	0.54	9
4500 mg oral	M	2.80	Predicted	3216.00	3195.69	0.99	723.10	528.47	0.73	5.00	2.80	0.56	9
6000 mg oral	V	2.80	Predicted	3696.00	3771.57	1.02	782.00	540.26	0.69	5.00	2.85	0.57	9
Food effect
750 mg oral, low‐fat meal	M	1.00	Default	3202.00	3376.11	1.05	722.00	624.73	0.87	4.51	3.05	0.68	11
750 mg oral, high‐fat meal	M	2.80	Default	4584.00	5150.51	1.12	1050.00	889.84	0.85	3.00	3.10	1.03	11
750 mg oral, whole milk	V	1.00	Predicted	2450.00	2585.06	1.06	527.00	529.00	1.00	5.00	3.05	0.61	11
1500 mg oral, high‐fat meal	V	2.80	Default	8347.00	8017.29	0.96	1628.00	1250.35	0.77	3.00	3.18	1.06	9
5 mg/kg oral, high‐fat food	V	2.80	Default	1793.00	2403.59	1.34	248.00	447.34	1.80	4.00	2.75	0.69	23
10 mg/kg oral, high‐fat food	V	2.80	Default	4025.00	4968.95	1.23	626.00	875.98	1.40	4.00	2.85	0.71	23
20 mg/kg oral, high‐fat food	V	2.80	Default	7618.00	8947.22	1.17	1003.00	1370.20	1.37	4.00	3.20	0.80	23
Multiple doses
750 mg oral, fed, arm 1	V	1.00	Default	3500.00	3601.24	1.03	840.00	716.65	0.85	5.00	3.05	0.61	24
750 mg oral, fed, arm 2	V	1.00	Default	3690.00	3723.58	1.01	852.00	738.35	0.87	5.00	3.05	0.61	24
750 mg oral, fed, arm 3	V	1.00	Default	3560.00	3626.92	1.02	838.00	718.54	0.86	5.00	3.05	0.61	24
Hepatic impairment (HI)
200 mg oral, mild HI, fast	V	1.00	Default	648.00	771.15	1.19	233.00	291.20	1.25	2.80	2.25	0.80	22
200 mg oral, moderate HI, fast	M	1.00	Default	1054.00	1077.17	1.02	354.00	340.76	0.96	2.00	2.30	1.15	22
200 mg oral, severe HI, fast	V	1.00	Default	1855.00	1303.52	0.70	381.00	349.89	0.92	2.50	2.35	0.94	22

Note: The observed area under the concentration‐time curve from time zero to time of the last observed concentration (AUC_0−t ) and the peak concentration (C _max) were presented as the reported geometric mean values, and the time to maximum concentration (t _max) was presented as the reported median values. HI, hepatic impairment.

Abbreviations: HI, hepatic impairment; M, dataset was used for model development; V, dataset was used for model verification.

2.3. PBPK Model Development

The CBD PBPK model incorporating the lymphatic absorption was developed based on our previously published CBD PBPK model [18]. Physicochemical properties, distribution, and elimination parameters not related to oral absorption (e.g., LogP, pKa, Vss, CL_int) were adopted unchanged from our previously published CBD PBPK model [18].

The oral CBD PBPK model for CBD with sesame oil was improved using the multi‐layer gut wall within the advanced dissolution, absorption, and metabolism (M‐ADAM) model to describe the CBD absorption via the lymph system after oral administration in sesame oil. The M‐ADAM model mechanistically incorporates a lymphatic absorption route from the intestinal interstitial fluid (ISF) to the systemic circulation. Dissolution was predicted using the diffusion layer model (DLM) within Simcyp based on the Simcyp predicted solubility and micelle: buffer partition coefficients, including the drug solution mediated by bile micelle. CBD permeability was predicted using the mechanistic passive permeability (Mech P_eff) model with Log P_o:w (method 2). Capillary bed permeability‐surface area product and lymphatic reflection coefficient, two absorption parameters influencing the percentage of drug transferred to the systemic circulation via the intestinal lymphatic system, were optimized manually. This optimization aimed to approximate the uptake of CBD by human chylomicrons derived from an in vitro study [10]. The parameter optimization of the model was guided by the model's ability to accurately describe the observed data and by local sensitivity analysis (Figures S1 and S2). Colon absorption was set to the minimum (absorption rate scalar: 1E⁻⁶) based on evidence of poor CBD absorption from the colon in a rat study [25]. We assumed that the CBD taken by chylomicrons inside the enterocytes was entirely absorbed into the systemic circulation through the lymph pathway to simplify the model. CBD is highly enriched in micelles when administered with sesame oil, and micelles can be absorbed by the enterocyte [10]. Therefore, we also assumed that the total concentration is the driving concentration in the unstirred boundary layer for epithelial permeability. Additionally, the absorption rate scalars were optimized for the high‐dose CBD, given that increased fat intake enhances lipid absorption, likely in the form of micelles [14].

The population parameters, including small intestine mean residence time (MRT), luminal bile salts, luminal fluid volume, and lymph flow rate, were modified to reflect study conditions and CBD‐specific characteristics. Small intestine MRTs for both fasted and fed states were increased by 45% from default values, accounting for the rich oleic acid content in sesame oil [26]. Luminal bile salts for the fasted state were set to fed state values due to the presence of sesame oil, and the fed state values were further increased by 60%, calculated using the Simcyp default fed gallbladder residual volume (59.9%) and a reported gallbladder residual volume (36%) [27]. Luminal fluid volume in the fasted state was also set to the values in the fed state because the sesame oil intake and the strawberry flavor in Epidiolex can stimulate gastrointestinal fluid secretion [28]. However, the quantitative effects of sesame oil versus the strawberry flavor on gastrointestinal fluid secretion are not fully known [2, 29]. Lymph flow rates were optimized within a reference range (0.01–0.2 L/kg/h) to reflect the gastrointestinal lymph flow rate [30]. When simulating a clinical trial where glucose was included during food intake, the lymph flow rate was set to 1.9 times that of the fasted state, based on a rat study [13]. An absorption lag time of 1.5 h with 30% variability from two human stomach emptying studies was included to reflect the long t _max of CBD [31, 32]. An additional custom Lua model was applied to account for the 1.5‐h lag time between undissolved CBD and the gastric fluid [31, 32]. When simulating the multiple doses, all the lag times (1.5 h) were modeled with another custom Lua model, as the default absorption lag time was only applied to the first CBD dose. No variability was applied to the custom lag time due to the software settings.

Additionally, when the details of the calories and fat mass information for a meal or a CBD dosing are available, the stomach MRT is predicted using the gastric MRT function in Simcyp (Equation 1):

$$\begin{array}{c}\text{Stomach}\mathrm{MRT}\left(h\right)=a\times \mathrm{Fat}\left(\%\text{of Meal calories}\right)+b\times \hfill \\ \text{Meal calories}\left(\mathit{\text{kcal}}\right)+c\hfill \end{array}$$ (1)

where a = −0.0024853, b = 0.0016963, c = 0.77209. The fat and calorie content of different CBD doses is listed in Table S1. Each mL of Epidiolex contains 736 mg of sesame oil [33], and each gram of sesame oil provides 8.84 kcal [34]. Sesame oil was assumed to be entirely lipid (fat) [34], and the calories from other ingredients in Epidiolex were not considered. As detailed in Table S1, CBD doses ranging from 750 to 6000 mg correspond to Epidiolex solution volumes between 7.5 and 60 mL. These volumes contain between 5.51 and 44.10 g of sesame oil and provide 48.73–389.84 kcal.

Model performance was assessed by comparing the observed clinical data with the mean values and the 5^th to 95^th percentiles from the simulated pharmacokinetic profiles. Models were considered acceptable if the predicted AUC_0−t , C _max, or t _max were within two‐fold of observed values, a widely applied standard in PBPK model verification [35].

2.4. CBD PBPK Model in Hepatic Impairment Patients

For simulations involving hepatic impairment, the software's built‐in population files were used with modifications on the hepatic abundances of UGT 1A9, 2B7, and 2B15. The specific numerical values for abundances for mild, moderate, and severe hepatic impairment were directly adopted from an in vitro study in human liver samples from donors with different levels of hepatic impairment [36]. Since CBD in the hepatic impairment clinical trial was administered 2 h post‐meal, this condition was simulated as a fasted state, and the corresponding absorption parameters of the M‐ADAM model for the fasted state were applied. The lymph flow rate for hepatic impairment simulations was set to 3 times the value for the fasted state, based on a rat study [16]. Predicted results were compared with observations for single‐dose CBD exposures to verify the hepatic impairment model. Once verified, the model was used to simulate 750 mg CBD single or multiple doses in hepatic impairment, both with and without food. To ensure appropriate comparisons, the minimum age in the trial design of healthy subjects was modified to match the age distribution in subjects with hepatic impairment (Sim‐Cirrhosis CP‐A and CP‐B) for pharmacokinetic simulation in subjects with normal hepatic function. The lymph flow rate for hepatic impairment simulations was kept at the 3× fasted value even for “fed state” simulations, as there is no evidence to suggest a higher lymph flow rate for patients with hepatic impairment in the various fed states.

3. Result

The overall workflow of the CBD PBPK modeling with mechanism‐based absorption is shown in Figure 1. The final model parameters are presented in Table 2. Model performance, comparing model‐simulated versus observed AUC_0−t , C _max, and t _max is summarized in Table 1 and presented in Figure 2. This includes data for CBD single doses in the fasted state and different fed states, multiple doses in a regular fed state, and single doses in fasted patients with hepatic impairments. The predicted and observed concentration‐time profiles of CBD following a dose range of 200–6000 mg in the fasted state are shown in Figure S3. The predictions well captured the observed CBD concentrations.

FIGURE 1.

General workflow for cannabidiol (CBD) model development. HI, hepatic impairment; M‐ADAM, simcyp multi‐layer gut wall within the advanced dissolution, absorption, and metabolism model; MD, multiple doses; PBPK, physiologically based pharmacokinetic; SD, single dose.

TABLE 2.

Final population and absorption input parameters for the CBD PBPK model.

Parameter	Value	References
Population
Stomach lag time (h)	1.5	From literature [31, 32]
Stomach MRT [for fluid, dissolved drug, and fine particles] (h)	CBD dose < 750 mg, or the fat and calorie intake are unknown: default values	Simcyp default value. See Table 1 for details
	CBD dose ≥ 750 mg, and the fat and calorie intake can be calculated: predicted	Individually predicted with Simcyp; see Table 1 for details
Small intestine MRT [for fluid, dissolved drug, and fine particles] (h)	Fasted: 4.93	1.45 times of the Simcyp default value [26]
	Fed: 6.86	1.45 times of the Simcyp default value [26]
Luminal bile salts (mM)	Fasted: default values of fed	Simcyp default value
	Fed: 1.6 times of default	Calculated a
Luminal fluid volume, stomach (mL)	Fasted: 1000	Simcyp default value for fed b
	Fed: 1000	Simcyp default value
Lymph flow rate (L/kg/h)	Fasted or whole milk: 0.170	Optimized from reference range (0.01–0.2) [30]
	Regular fed, low‐fat meal, and high‐fat meal: 0.323	1.9 times the fasted value (rat study) [13]
	Hepatic impairment: 0.510	Three times the fasted value (rat study) [16]
Absorption
Model type	Simcyp multi‐layer gut wall within the advanced dissolution, absorption, and metabolism (M‐ADAM) model
Formulation	Suspension with 0% API dissolved
fugut	0.025479	Simcyp predicted
Dissolution	Diffusion layer model
Intrinsic solubility (mg/mL)	0.000181	Simcyp predicted
logKm:w	Neutral: 6.974	Simcyp predicted
	Ion: 4.974	Simcyp predicted
Ptrans,0 (10−6 cm/s)	3036.5	Mechanistic Peff (Mech Peff) model, method 2, global
Absorption rate scalars	1.0, 2.0, 2.8	Simcyp default value; optimized; optimized; see Table 1 for details
Capillary bed permeability‐surface area product (L/h)	0.35	Optimized
Epithelial permeability	Total concentration
Lymphatic reflection coefficient	0.000001	Optimized

Note: All other general physicochemical, distribution, and elimination input parameters for CBD were adopted unchanged from our previously published PBPK model [18].

Abbreviations: fu_gut, unbound fraction of drug in enterocytes; logK_m:w, logarithm of bile micelle: buffer partition coefficients; MRT, mean residence times; P_trans,0, intrinsic transcellular permeability.

^a Calculated using the Simcyp default fed gallbladder residual volume (59.9%) and the reported gallbladder residual volume (36%) with the following equation [27]: $\text{Luminal bile salts}\left(\mathrm{mM}\right)=\text{Simcyp}\text{default Luminal bile salts}\left(\mathrm{mM}\right)\times \frac{\left(100\%-36\%\right)}{\left(100\%-59.9\%\right)}$.

^b The luminal fluid volume in the stomach for the fasted state was set to the Simcyp default value for the fed state to simulate the effect of sesame oil taken from Epidiolex.

FIGURE 2.

Observed versus predicted (A) area under the concentration‐time curve from time zero to time of the last observed concentration (AUC_0−t ) and (B) the peak concentration (C _max) for CBD.

For food effect assessment, the model‐simulated CBD concentration‐time profiles following a 750 mg single dose with whole milk, a low‐fat meal, and a high‐fat meal, a 1500 mg single dose with a high‐fat meal, 750 mg multiple doses with regularly fed states, and 5, 10, and 20 mg/kg CBD single doses with high‐fat meals are shown in Figure S4. All the predicted AUC_0−t , C _max, and t _max values are within the two‐fold range of the corresponding observed data. The model accurately captured the magnitude of the food effect for both AUC_0−t and C _max, as well as t _max across low‐fat and high‐fat meal conditions.

Model performance for hepatic impairment simulations is presented in Table 1, where predicted hepatic impairment data are compared with the observed values. The predicted AUC_0−t , C _max, and t _max values for patients with mild, moderate, and severe hepatic impairment were in good agreement with the observed data, consistently falling within the two‐fold range of the observations. The predicted CBD concentration‐time profiles after a 200 mg single dose in patients with different levels of hepatic impairment in the fasted state are shown in Figure S5. Building on the verified model, simulated CBD concentration‐time profiles in patients with different levels of hepatic impairment (CP‐A, CP‐B, and CP‐C) for both single dose and multiple doses, under various fasted/fed states, are presented in Figure 3. Simulations showed that the exposure of CBD increased by a range of 38%–91% with food intake in patients with different levels of hepatic impairment, and the food effect is most significant in patients with severe hepatic impairment in our single and multiple dose simulations.

FIGURE 3.

PBPK model‐simulated CBD plasma concentrations in mild, moderate, and severe hepatic impairment after single or multiple doses in various diets. The orange, blue, and pink lines represent the simulated concentrations for patients with mild, moderate, and severe hepatic impairments, respectively.

4. Discussion

Several groups, including ours, have developed PBPK for CBD to improve the understanding of its pharmacokinetics [17, 18, 19]. However, the oral absorption of CBD was not well described, and significant gaps remained concerning its nonlinear absorption, food effect, and the impact of hepatic impairment on CBD exposure. Accordingly, this study addresses these critical gaps by presenting, to our knowledge, the first comprehensive PBPK model for CBD that mechanistically incorporates lymphatic oral absorption to accurately predict the effect of dose, food, and hepatic impairment on CBD exposure and bioavailability.

The M‐ADAM model in Simcyp was used to simulate the lymphatic oral absorption of CBD. This same absorption model has successfully described the food effect on halofantrine tablets, another BCS Class II drug known for its higher lipophilicity than CBD [12]. For halofantrine, AUC in the fed state was reported as 12‐fold compared to fasting, with over 50% of the dose absorbed via lymphatic absorption in the fed state, compared to 1.3% in the fasted state. The optimization of lymph flow rate, capillary bed permeability‐surface area product, and lymphatic reflection coefficient effectively captured the lymphatic absorption and food effect of high‐fat meals for halofantrine [12]. In comparison with halofantrine, the modeling of absorption and the food effects for CBD has more challenges. The CBD in Epidiolex was formulated with sesame oil, and clinical trials were conducted in various fed states, including high‐fat meals, low‐fat meals, regular meals, and whole milk. Therefore, additional parameters related to the stomach and the intestinal physiology were carefully considered in our model. Studies have shown that even small amounts of fat intake (e.g., approximately 2 g) can trigger bile release, with the extent of gallbladder emptying directly related to the amount of fat intake [37]. To mechanistically reflect the effect of fat intake on bile release, the concentrations of luminal bile salts for the fasted states were adjusted to the fed values. Furthermore, fed‐state luminal bile salt concentrations were increased by 60%, based on calculations assuming a higher gallbladder emptying percentage to mimic the effect of high‐fat intake [27]. Similarly, stomach fluid volume for the fasted state was also set to the value of the fed state to account for gastric secretion stimulated by sesame oil intake and the strawberry flavor added. The impact of sesame oil and flavoring in Epidiolex on gastrointestinal fluid secretion and lymphatic absorption remains to be quantitatively determined. Further clinical and physiological studies are needed to clarify the relative contributions of taste versus fat content, which would inform future model refinements for different CBD formulations.

The model provided valuable mechanistic insights into the fraction of CBD reaching the lymph node (f_a,lymph) and the fraction of CBD reaching the systemic circulation (f_sc). The mean model‐simulated f_sc following a 750 mg dose markedly increased with meal intake, 10% for the fasted state, 29% for a low‐fat meal, 31% for a regular fed meal, and 42% for a high‐fat meal. Regarding lymphatic absorption, the model predicted that the f_a,lymph decreased with increasing CBD doses in the fasted state, from 0.12 for 200 mg CBD to 0.04 for 6000 mg CBD. Despite this dose‐dependent decrease in f_a,lymph, the model‐predicted percentage of lymphatic transport relative to total CBD reaching the interstitial fluid (CBD in chylomicrons + CBD out of chylomicrons, assuming all CBD in chylomicrons were absorbed via lymph) for all fasted doses (ranging from 59.9% to 65.2%) was consistent with the reported observation in human plasma‐derived chylomicrons of 67.7% [10]. This indicates that while the CBD amount absorbed via lymph may exhibit saturation at higher doses, the relative proportion of overall absorption occurring via the lymphatic route remains significant and consistent with observed data.

The model successfully simulated oral CBD AUC_0−t , C _max, and t _max across different doses, food conditions, and in varying degrees of hepatic impairment, showing close concordance with observed data. However, despite the inclusion of a relatively long stomach emptying lag time reported (1.5 h) [31, 32], the simulated t _max values for different doses and food effects were mostly under‐predicted. This discrepancy may be attributed to several factors. In a rat study, the lymph flow rate increases during the initial hours following food intake and reaches the maximum value at around 3 h post‐meal [14]. While lymphatic absorption is expected to increase with lymph flow rate, the current PBPK model structure does not allow for simulation of this time‐dependent lymph flow rate increase, thus representing a modeling limitation. Another possible reason for the under‐predicted t _max is that CBD with sesame oil has been reported to have early and delayed absorption, with observed t _max ranging from 1.5 to 8 h [38]. The sensitivity analysis showed that the t _max is highly correlated with stomach MRT (Figure S1). However, there is currently no evidence of a delayed stomach MRT specifically after CBD administration, and thus, default or calculated values based on calories and fat were applied in our model. The highly variable absorption profiles can result in a lower peak concentration on the concentration‐time profile compared to the reported C _max value in clinical studies. Although with limitations, our model successfully captured the AUC_0−t , C _max, and t _max changes for CBD after the low‐fat meal, regular fed, high‐fat meal, and whole milk intake. Specifically, the lymph flow rate was set to 1.9 times that of fasted for most food effects, except for whole milk, based on a rat study that showed glucose alone was shown to stimulate the lymph flow rate [13]. The simulated CBD exposure when taken with whole milk accurately captured the observations without including the effect of glucose on lymph flow rate, suggesting that the glucose in milk may not significantly impact the lymph flow rate. Glucose exists as part of lactose in whole milk. Given that the glycemic index of lactose (46) is notably lower than that of carbohydrates in meals [39], the slower absorption and metabolism of glucose in milk may not be sufficient to change lymph flow rate [13]. Furthermore, the absorption rate scalars were increased for high‐fat meals. This adjustment is justified by the principle that the high lipid concentrations in the intestine can increase the overall absorption rate of lipids, and given CBD's high lipophilicity, its absorption rate would similarly increase. The increase in absorption rate scalars was also applied to high fasted CBD doses of 4500 and 6000 mg, as these doses involve the co‐administration of approximately 33 and 44 g of fat, respectively, which is a substantial amount approaching the 55–65 g of fat in a typical high‐fat meal [40].

Our model also robustly captured the AUC_0−t , C _max, and t _max changes of CBD in different degrees of hepatic impairment. While the model generally performed well, AUC_0−t was slightly over‐predicted in patients with mild hepatic impairment and under‐predicted in patients with severe hepatic impairment. This may be because we use the same lymph flow rate for different degrees of hepatic impairment. A lymph flow rate of three times that of the healthy was used in patients with mild, moderate, or severe hepatic impairment, as reported in a rat study [16]. However, the level of hepatic impairment was not classified in this rat study. They also mentioned that intestinal lymph flow increases with the acute elevation of portal pressure. Given that portal pressure increases with Child‐Pugh grade, the lymph flow rate increase could be lower than three times for CP‐A patients and potentially higher than three times for CP‐C patients [41]. Incorporating this gradation of lymph flow increase based on hepatic impairment severity could lead to improved predictions in these specific populations.

There are limitations to our study. First, the under‐prediction of simulated t _max values persists, which may be due to the time‐dependent increase in lymph flow rate after drug and food administration, or the complex early and delayed absorption of CBD as discussed above. The current model supports only a fixed lymph flow rate, and variabilities were not applied to the lag times in the custom Lua models. Second, the M‐ADAM model currently supports only fixed fasted or fed states when simulating multiple doses. This meant that clinical trials with CBD administered twice daily in dynamic fasted/fed states were excluded from our model development, thereby limiting the model's direct applications to more complex fasted/fed state regimens. Third, while stomach MRT in patients with severe hepatic impairment is reported to be increased, the exact ratio of this increase was not reported in the literature; thus, we could not modify it accordingly in the model. Finally, although the general lymph flow rate changes used in this study are from the literature, further refinement of the specific lymph flow rate dynamics after different meals, including different amounts and sources of glucose, and in patients with varying degrees of hepatic impairment, is required. In this study, a single factor for lymph flow rate change from an animal study using rat cirrhosis is applied for all levels of hepatic impairment; however, changes in cirrhosis in rats may be different from humans with hepatic impairments, and such changes are likely to be heterogeneous in different levels of hepatic impairment. Currently, this is the most justified approach we can take given the available data. Future studies will focus on improving the mechanism‐based lymphatic absorption within the PBPK models by including a time‐dependent lymph flow rate increase and a time‐dependent lipid absorption rate increase. Additional studies are needed to explore how lymph flow might change with age and consequently impact the CBD oral absorption.

5. Conclusion

In summary, our study successfully developed and verified a novel PBPK model incorporating mechanism‐based lymphatic absorption, which provides a comprehensive mechanistic explanation for the observed nonlinear absorption, significant food effect, and impact of hepatic impairment on oral CBD pharmacokinetics. This is particularly relevant for CBD in sesame oil, where guidelines recommend dosing with food, and dose modifications for patients with hepatic impairment. To our knowledge, this work represents the first PBPK study to mechanistically assess these complex effects. The verified model offers a robust foundation to optimize dosing strategies, predict potential drug–drug interactions, and evaluate CBD pharmacokinetics in various specific populations, ultimately contributing to safer and more effective clinical use of CBD.

Author Contributions

L.Q. and Z.Z. wrote the manuscript. Z.Z. designed the research. L.Q. and Z.Z. performed the research. L.Q. analyzed the data.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: cts70401‐sup‐0001‐DataS1.pdf.

Qian L. and Zhou Z., “Using a PBPK Model Incorporating Lymphatic Absorption to Predict Food Effect, Multiple Dosing, and Hepatic Impairment of Cannabidiol,” Clinical and Translational Science 18, no. 11 (2025): e70401, 10.1111/cts.70401.