Pharmacokinetics and pharmacodynamics of intravenous and oral (S)‐ketamine: Investigating metabolite contribution to subjective effects

Abstract

Aims

Oral administration of (S)‐ketamine for treatment‐resistant depression (TRD), as alternative to the registered intranasal or off‐label intravenous administrations, has high potential. However, it is characterized by an extensive first‐pass metabolism, resulting in low (S)‐ketamine exposure and high levels of active metabolites, including (S)‐norketamine and (S)‐hydroxynorketamine. The relative contribution of the parent and metabolites to the resulting antidepressant effects remains unclear. Therefore, this study aimed to first characterize the pharmacokinetics (PK) of (S)‐ketamine and its metabolites after oral and intravenous administration in healthy participants and secondly quantify the pharmacokinetic/pharmacodynamic (PKPD) relationship of (S)‐ketamine and (S)‐norketamine to the subjective effects measured on the visual analogue scale (VAS) ‘Feeling High’.

Methods

Data from a previously conducted clinical study was used, where 17 healthy participants received oral (0.20 and 0.45 mg/kg) and intravenous (0.4 mg/kg over 40 min) (S)‐ketamine in a randomized, placebo‐controlled, crossover clinical trial. A semi‐physiological population PK model was developed to describe the first‐pass metabolism and (S)‐ketamine and subsequently (S)‐norketamine concentrations were linked to the VAS ‘Feeling High’ using a bounded integer modelling approach.

Results

A significant (S)‐norketamine PKPD relationship was determined alongside (S)‐ketamine, but estimated variance of the bounded integer model was high warranting further investigation.

Conclusions

Our analysis suggests that not (S)‐ketamine, but its metabolite, is the main driver of subjective effects after oral administration, and as such may also contribute to antidepressant effects in TRD patients receiving oral (S)‐ketamine.

What is already known about this subject

• Oral (S)‐ketamine as treatment for depression is currently investigated as alternative for intranasal and intravenous administrations.

• First‐pass metabolism of oral (S)‐ketamine results in relatively high exposure of active metabolites such as (S)‐norketamine.

• The relative contribution of (S)‐ketamine and metabolites to any resulting antidepressant effects after oral (S)‐ketamine remains unclear.

What this study adds

• Oral and intravenous (S)‐ketamine administration data in healthy participants was used to develop a population PKPD model, linking (S)‐ketamine and metabolite exposure to subjective effects on the VAS ‘Feeling High’

• A significant metabolite contribution was identified, which must be considered when determining effective oral (S)‐ketamine doses for future clinical trials.

1. INTRODUCTION

(S)‐ketamine demonstrates rapidly emerging antidepressant effects that persist for several weeks following single dose administration in patients with treatment‐resistant depression (TRD). ¹ , ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ Its mood improving effects in TRD may result from inhibition of N‐Methyl‐D‐Aspartate receptors (NMDARs) expressed on postsynaptic glutamatergic neurons as well as gamma‐aminobutyric acid (GABA) interneurons, leading to increased presynaptic glutamate release and a shift to postsynaptic α‐amino‐3‐hydroxy‐5‐methyl‐4‐isoxazolepropionic acid receptor (AMPAR)‐mediated glutamatergic signalling. ¹ , ⁸ , ⁹ As such, (S)‐ketamine represents a novel antidepressant drug in terms of both mechanism of action and onset and duration of therapeutic effects, as conventional drugs that predominantly act on monoaminergic neurotransmitter pathways typically require daily administration for several weeks to achieve efficacy.

(S)‐ketamine is currently registered for the treatment of TRD via intranasal (IN) administration, and can also be administered off‐label through intravenous (IV) infusion. However, both applications limit the administration to be performed in specialized facilities, in order to monitor for side effects of hypertension and dissociation. To improve clinical scalability, as well as reduce costs and patient burden, orally (PO) administered (S)‐ketamine has been put forward as a more viable alternative. ¹⁰ , ¹¹ , ¹² In contrast to IV administration, PO (S)‐ketamine is associated with a significant first‐pass effect as it undergoes extensive hepatic metabolism, where it is primarily converted into (S)‐norketamine by cytochrome P450 (CYP) enzymes CYP2B6 and CYP3A4, ¹³ , ¹⁴ , ¹⁵ and subsequently transformed into (S)‐hydroxynorketamine, predominantly (2S,6S) and (2S,6R)‐hydroxynorketamine by CYP2A6 and CYP2B6, although other metabolites also exist. ⁹ , ¹³ , ¹⁶ , ¹⁷ CYP3A4 is also expressed in the gut wall, ¹⁸ , ¹⁹ where active metabolism would be pertinent to the oral absorption process. However, the extent of its contribution to the total first‐pass effect remains ambiguous and requires further investigation, with one study suggesting it is limited. ²⁰ As a result of the first‐pass metabolism, systemic exposure of PO (S)‐ketamine is lower, while that of its metabolites (S)‐norketamine and (S)‐hydroxynorketamine is relatively higher compared with either IN or IV administration.

(S)‐norketamine and (S)‐hydroxynorketamine are pharmacologically active metabolites that demonstrate relatively distinct in vitro pharmacological profiles compared with (S)‐ketamine. (S)‐norketamine does act as NMDAR antagonist, albeit with an approximately fivefold lower binding affinity. ²¹ , ²² In contrast, (S)‐hydroxynorketamine demonstrates even lower NMDAR binding affinity yet does act as inhibitor of the α7‐nicotinic acetylcholine (α7nACh) receptor. ⁹ , ²³ Both metabolites have demonstrated antidepressant‐like effects in preclinical depression models, ²⁴ , ²⁵ indicating they may contribute to mood improvement in patients suffering from depression, possibly via mechanisms other than NMDAR antagonism. Since PO administration alters the relative systemic exposure of (S)‐ketamine and its active metabolites as compared to IN or IV administration, the contributions of parent compound and active metabolites may alter (S)‐ketamine's clinical safety and/or PD profile, and as a consequence, its efficacy and safety in TRD. Therefore, a clear understanding of the pharmacology of both parent and active metabolites is required prior to its application as alternative to IN or IV administration.

Disentangling the pharmacodynamics of (S)‐ketamine and its metabolites after oral dosing is challenging due to their simultaneous presence related to first‐pass metabolism. This can only be achieved by comparing effects to conditions where exposure of one compound predominates, either via direct metabolite administration or by bypassing first‐pass metabolism through IV administration. The latter was performed in a recent study, where IV and PO (S)‐ketamine were administered to healthy participants in a crossover fashion. ²⁶ As antidepressant effects were naturally absent in this healthy population, the Visual Analogue Scale (VAS) ‘Feeling High’ provided an alternative to assess subjective drug effects. ²⁷ This scale has previously demonstrated to be sensitive to a wide range of central nervous system (CNS) active drugs, including ketamine. ²⁸ , ²⁹ Alongside other pharmacodynamic measures, it quantifies participants' psychotomimetic effects (i.e., altered perception, dissociation and hallucination). The presence of such effects is by some regarded as essential for subsequent antidepressant effects in TRD. ³⁰ , ³¹ This notion on imperative psychotomimetic effects is also supported by a recent meta‐analysis indicating a modest correlation between ketamine's subjective and therapeutic effects. ³² The results of the clinical study suggested the psychotomimetic drug effects to be predominantly mediated by (S)‐ketamine and not its metabolites. ²⁶ However, pharmacokinetic‐pharmacodynamic (PKPD) modelling would be required to specifically quantify the differential contributions of (S)‐ketamine and its pharmacologically active metabolites. To this end, this study aimed to characterize the PK and PD of (S)‐ketamine and its pharmacologically active metabolites following PO and IV administration using a population PKPD modelling approach, with the goal of quantitatively exploring their relative contributions to the observed subjective drug effects.

2. METHODS

2.1. Clinical trial data

A placebo‐controlled, blinded, double‐dummy crossover clinical trial investigating the acute and transient effects of PO and IV (S)‐ketamine in healthy participants was performed at the Centre for Human Drug Research (CHDR), Leiden, the Netherlands (Dutch Trial Registry ID: 28424, EudraCT: EUCTR2020‐002083‐31‐NL). Study details have been published previously. ²⁶ In short, 16 healthy participants received 0.4 mg/kg (S)‐ketamine IV infusion over 40 min, 0.2 mg/kg (S)‐ketamine solution PO, 0.45 mg/kg (S)‐ketamine solution PO, each with matching PO or IV placebo, or a double placebo for both IV and PO, randomized over four visits with a washout period of 14 to 21 days. Doses were selected such that maximum (S)‐norketamine levels were expected to be similar for the 0.4 mg/kg IV and 0.2 mg/kg PO regimens, while (S)‐ketamine exposures would be opposite, that is, high and low, respectively. ²⁶ PO administration occurred 15 min after start of IV infusion, to ensure similar times of maximum concentrations (t_max). One participant dropped out and was replaced after the first dosing event having received 0.45 mg/kg PO, but data from this participant was included for this analysis, resulting in a sample size of 17 for the goals of the present analysis. Participants had a mean ± SD (range) age of 27.2 ± 4.6 (21–35) years, weight of 67.6 ± 9.3 (51.5–81.6) kg and included eight males and nine females.

Ketamine, norketamine and hydroxynorketamine (i.e., [2,6]‐hydroxynorketamine) levels were measured in venous plasma samples using LC/MS–MS with a lower limit of quantification of 1.00 ng/mL for all analytes (details in Supporting Information). As only (S)‐ketamine and no (R)‐ketamine was administered and as previous studies have shown no enantiomeric conversion occurs in vivo, the measured analyte levels were assumed to represent the (S)‐enantiomers (i.e., (S)‐ketamine, (S)‐norketamine and (S)‐hydroxynorketamine (i.e., [2S,6R;2S,6S]‐hydroxynorketamine)). ³³ , ³⁴ , ³⁵ Subjective effects were measured through the participants' response on the VAS ‘Feeling High’, where response values for the VAS are discrete integers, ranging between 0 and 100. Placebo data was not included for model development due to a lack of response.

2.2. Semi‐physiological PK model development

The semi‐physiological population PK model for (S)‐ketamine and (S)‐norketamine as developed by Ashraf et al. was used as a framework for model development ²⁰ (Figure S1). This model includes physiological compartments for the gut wall, portal vein and liver, to describe (S)‐ketamine's first‐pass metabolism. First, the predictive performance of the Ashraf model for the available data was evaluated through visual predictive checks (VPCs). If deemed necessary, further development was done through re‐evaluation of the number of central distribution compartments and parameter re‐estimation, while maintaining the values of the semi‐physiological components (Ashraf et al. for a detailed description ²⁰ ). Finally, the model was extended to also describe (S)‐hydroxynorketamine data.

Model development was performed sequentially for parent and metabolites, using the individual PK parameter estimates and their uncertainty, also known as the IPPSE approach. ³⁶ The parent compound was assumed to be fully metabolized to the next metabolite to circumvent issues with parameter identifiability. Distribution volumes and intercompartmental clearances were scaled allometrically. Intrinsic gut wall clearance and hepatic clearance were already related to body weight through their assumed blood flows. Data below the limit of quantification was included using the M3 method. ³⁷ A proportional residual error structure was used. Interindividual and between‐occasion variability (IIV and BOV, respectively) were assumed to be log‐normally distributed and included in a stepwise manner. Trends in variability potentially related to the treatment route or dose were investigated by plotting the distribution of the empirical Bayes estimates (EBEs) and by estimation of separate parameters.

2.3. (S)‐Ketamine and (S)‐norketamine PKPD relationship with VAS ‘feeling high’

A bounded integer modelling approach was used to account for the integer nature of the VAS ‘Feeling High’ data and its boundaries (0 and 100). A more detailed description of this method is provided by Wellhagen et al. ³⁸ In short, the approach assumes the data to be ordered categorical with $n$ categories, where $n=101$ for this study. The area under the standard normal distribution (mean of 0, variance of 1) is divided into $n$ equal‐sized areas using the probit function, resulting in $n-1$ cutoff values represented as $Z_{1/n}$ to $Z_{\left(1-n\right)/n}$. Based on fixed effects ( $\mathrm{\theta },\mathrm{\sigma }$), random effects for individual $i$ ( ${\eta }_i$), time ( $t$) and covariates ( $X_i$), functions to describe the mean ( $f\left(\mathrm{\theta },{\eta }_{i,f},t,X_{i,f}\right)$) and variance ( $g\left(\mathrm{\sigma },{\eta }_{i,g},t,X_{i,g}\right)$) of a normal distribution can be defined. Together with the $Z$ values, the probability for a certain category or score $k$ ( $P_{i,j}\left(k\right)$) can then be defined as

$$P_{i,j}\left(k\right)=\Phi \left(\frac{Z_{\frac{k}{n}}-f\left(\theta ,{\eta }_{i,f},t,X_{i,f}\right)}{\mathrm{g}\left(\mathrm{\sigma },{\eta }_{i,g},t,X_{i,g}\right)}\right)-\Phi \left(\frac{Z_{\frac{k-1}{n}}-f\left(\theta ,{\eta }_{i,f},t,X_{i,f}\right)}{\mathrm{g}\left(\mathrm{\sigma },{\eta }_{i,g},t,X_{i,g}\right)}\right)$$ (1)

where $\Phi$ is the cumulative distribution of the normal distribution function.

Linear (Equation 2), exponential (Equation 3) and power (Equation 4) functions were tested to describe the change in distribution mean related to the drug concentration:

$$\mathrm{f}\left(\mathrm{\theta },\mathrm{C}\right)={\mathrm{\theta }}_0+{\mathrm{\theta }}_1*\mathrm{C}$$ (2)

$$\mathrm{f}\left(\mathrm{\theta },\mathrm{C}\right)={\mathrm{\theta }}_0+{\mathrm{\theta }}_1^{\mathrm{C}}$$ (3)

$$\mathrm{f}\left(\mathrm{\theta },\mathrm{C}\right)={\mathrm{\theta }}_0+{\left(\frac{\mathrm{C}}{{\mathrm{\theta }}_1}\right)}^{{\mathrm{\theta }}_2}$$ (4)

where ${\theta }_0$ is the estimated baseline mean of the distribution on the continuous scale (f(), Equation (1)), ${\theta }_1$ and ${\theta }_2$ are slope and shape parameters, and $C$ is the drug concentration. IIV was added in a stepwise manner and tested on both the $\theta$ parameters and estimated variance of the distribution ( $\mathrm{\sigma }$ in g(), Equation (1)) and was added on a log‐normal scale if applicable.

First, the PKPD model linking (S)‐ketamine concentrations to the VAS ‘Feeling High’ was developed. The presence of a metabolite effect was then investigated by incorporating an additive term for metabolite concentration, for which linear, exponential and power functions were also tested. Ultimately, the functions for (S)‐ketamine and the metabolite PKPD relationships were re‐evaluated.

Data exploration showed similar changes in concentration ratio's over time for (S)‐ketamine to (S)‐norketamine as to (S)‐hydroxynorketamine. As a result, distinguishing the contribution of each metabolite to the effect was not feasible due to parameter identifiability. Since (S)‐norketamine is (S)‐ketamine's primary metabolite and has higher affinity for the NMDAR than (S)‐hydroxynorketamine, this metabolite was selected to evaluate the metabolite contribution.

2.4. Model selection criteria and diagnostics

Model selection was driven by drop in objective function value (dOFV ≤ −6.64, α = 0.01) for nested models or the lower Akaike Information Criterion (AIC) for non‐nested models, low relative standard errors (RSE < 50%), low condition number (<1000), well distributed goodness‐of‐fit (GOF) plots and adequate predictive performance as judged from the confidence interval visual predictive checks (ciVPC's), based on 1000 simulations of the clinical trial data. Visual inspection of the predictions for the VAS ‘Feeling High’ data was done using conditional individual weighted residual for categorical and count data (CIWRESlike). ³⁹

2.5. PKPD model simulation and external evaluation

The final PKPD model was used to simulate the expected VAS ‘Feeling High’ for a typical individual (70 kg) for the 0.4 mg/kg IV infusion over 40‐min and 0.45 mg/kg PO doses, as well as a third, higher PO dosing regimen to demonstrate which dose would be required to reach similar effects on the VAS ‘Feeling High’ as compared to the IV dosing regimen.

The PKPD relationship of the model was externally evaluated using data published by Simons et al., where the VAS ‘High’ was measured after sublingual and buccal administration of oral (S)‐ketamine thin films (50 and 100 mg). ⁴⁰ The reported mean (S)‐ketamine and (S)‐norketamine levels were used as input for the PKPD relationship to simulate the average expected response, which was compared to the actual observed response.

2.6. Software

PKPD model development was done with NONMEM (version 7.5) with the Laplacian estimation method. ⁴¹ Data transformation, analysis, simulation and visualization was done in R (version 4.4.1) ⁴² using the dplyr, tidyr, ggplot2 and mrgsolve packages. ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶

2.7. Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org and are permanently archived in the Concise Guide to PHARMACOLOGY 2023/24. ⁴⁷

3. RESULTS

3.1. Semi‐physiological PK model development

Assessment of the predictive performance of the model by Ashraf et al. ²⁰ for the clinical data showed overpredicted concentrations after IV administration, during infusion and over the whole time profile for (S)‐ketamine and (S)‐norketamine, respectively (Figures S1, S2). Predictions for PO doses were closer to the observed data as compared to IV; however, C_max of both compounds were slightly underpredicted, especially for the 0.45 mg/kg PO dose. Due to the inadequate predictive capability of the literature model, it was decided to re‐estimate model parameters based on the available data.

The re‐estimated semi‐physiological PK model consisted of two distribution compartments for both (S)‐ketamine and (S)‐norketamine and one for (S)‐hydroxynorketamine (Figure 1). Absorption kinetics were described with a first order rate constant, but as no gut wall metabolism was identifiable, (S)‐ketamine gut wall bioavailability (F_g,k) was set to 100%. A complete overview of estimated parameter values is provided in Table 1.

FIGURE 1.

Model structure of the final semi‐physiological pharmacokinetic model. F_g, gut bioavailability; F_h, hepatic bioavailability; IV, intravenous; PO, oral; Q_g, gut wall blood flow, see Table 1 for all other abbreviations. Footnotes depict whether parameters are shown for (S)‐ketamine (X_k), (S)‐norketamine (X_nk) or (S)‐hydroxynorketamine (X_hk). The model structure as previously presented by Ashraf et al. ²⁰ on which the final model structure was based is shown in Figure S1.

TABLE 1.

System and drug specific parameters as originally reported by Ashraf et al. ²⁰ and estimated empirical pharmacokinetic parameters of the final semi‐physiological pharmacokinetic model.

	Abbreviation	Unit
System specific parameters 20			Literature value
Blood flow
Hepatic	Q h	L/h	3.75*WT0.75
Portal vein	Q pv	L/h	0.75*Qh
Hepatic artery	Q ha	L/h	0.25*Qh
Intestinal	Q int	L/h	0.40*Qh
Mucosal	Q mu	L/h	0.80*Qint
Villous	Q vi	L/h	0.60*Qmu
Volume a
Hepatic	V h	L	1
Gut wall b	V gw	L	1
Portal vein	V pv	L	1

Drug specific parameters 20			Literature value
			(S)‐ketamine	(S)‐norketamine	(S)‐hydroxynorketamine
Unbound drug fraction in the blood	fu		0.70	0.5	0.5
Unbound drug fraction in the gut wall b	fu, gw		1
Permeability clearance through enterocyte	CLperm	L/h	4.1
Blood to plasma ratio	BPratio		0.5	1	1

Estimated pharmacokinetic parameters b			Estimate (RSE%)
			(S)‐ketamine	(S)‐norketamine	(S)‐hydroxynorketamine
Population parameters
Central distribution volume	V d	L	150.8 (13.3)	105.7 (4.1)	126.52 (8.6)
Intrinsic hepatic clearance	CL int,h	L/h	441.2 (17)	101.6 (6.3)	120.56 (6.3)
Peripheral distribution volume	V p	L	209.4 (6.6)	104.65 (10.9)
Intercompartmental clearance	Q	L/h	40.0 (12.4)	23.06 (14.9)
Absorption rate	K a	h−1	3.35 (23.8)
Interindividual variability
Central distribution volume	V d	CV%	17.4 (65.1)		28.6 (73)
Intrinsic hepatic clearance	CL int,h	CV%	43.2 (43.3)	31.1 (32.0)	18.1 (48.9)
Peripheral distribution volume	V p	CV%		17.1 (64.5)
Between‐occasion variability
Absorption rate	K a	CV%	55.5 (42.6)
Residual variability
Proportional		CV%	29.0 (5.4)	30.3 (8.9)	22.6 (3.8)

Note: Coefficient of variation calculated as $\sqrt{{\mathrm{e}}^{{\mathrm{\omega }}^2}-1}*100\%$ for the interindividual and between‐occasion variability and as $\sqrt{{\mathrm{\sigma }}^2}$ for the proportional residual variability, where ${\mathrm{\omega }}^2$ and ${\mathrm{\sigma }}^2$ are the respective estimated variances.

Abbreviations: CV, coefficient of variation; RSE, relative standard error; WT, weight (kg).

^a As the semi‐physiological model uses a simplified quasi‐steady‐state approximation approach, volumes are fixed to 1, see Ashraf et al.for details. ²⁰

^b In the final model, no gut wall metabolism could be quantified for (S)‐ketamine.

Variability in PK was high, with the highest estimated coefficients of variation for parameters related to the first‐pass metabolism and oral absorption, at 43.2% for IIV on CL_int,h,k and 55.5% for BOV on K_a (Table 1). Addition of BOV was initially significant for both CL_int,h,k and K_a, which revealed potential non‐linearities in the data when comparing distributions of individual parameter estimates (i.e., BOV EBE's) between treatment groups. If estimated as separate population parameters per treatment, K_a significantly reduced by 51% for 0.45 mg/kg PO vs. 0.2 mg/kg PO and CL_int,h,k increased with 30% for PO vs. IV administration, although not significantly. Therefore, ultimately only BOV for K_a was included in the model, also because intraindividual day‐to‐day differences are more typical for drug absorption than for metabolism related processes. The distributions of EBE's for included variability, showing the trend for K_a, and for the IPPSE approach are depicted in Figure S3.

Evaluation of the model through GOF plots (Figure S4), individual prediction plots (Figure S5) and VPC's (Figure 2 and Figure S6) showed that the model generally described the data well. No absorption delay could be quantified, which may explain the minor underprediction of the (S)‐ketamine C_max for PO administration (Figure 2 and S6), particularly for the higher PO dose, a difference between the PO doses that was not completely resolved by the inclusion of BOV for K_a. Additionally, the predicted formation of the metabolites may be too slow for IV administration but too fast for PO dosing (Figures 2, S4 and S6), presumably related to trends observed when BOV for CL_int,h,k was tested. Despite these trends, the model's predictive performance was deemed adequate to apply the individual predicted concentrations for PKPD model development.

FIGURE 2.

Confidence interval visual predictive checks (ciVPC) based on the final semi‐mechanistic pharmacokinetic model for the first 4 h. Results are shown for intravenous (I.) and oral (II. And III.) (S)‐ketamine administration. (a) (S)‐ketamine, (b) (S)‐norketamine, (c) (S)‐hydroxynorketamine.

3.2. (S)‐Ketamine and (S)‐norketamine PKPD relationships with VAS ‘feeling high’

Visual inspection of the PD data prior to analysis revealed several outliers that were excluded for model development (Figure S7). A relationship between the individual predicted (S)‐ketamine concentrations and mean of the distribution or f() was characterized using the bounded integer modelling approach to transform the VAS ‘Feeling High’ data to a continuous scale (Equation 1). First, a linear concentration–effect relationship between (S)‐ketamine and f() was established (Equation 2), with IIV on baseline and without the presence of an (S)‐norketamine effect. IV treatment data was adequately predicted with this model but failed to capture the observed effects after PO administration.

Besides the (S)‐ketamine PKPD relationship, a significant linear relationship for individual predicted (S)‐norketamine with f() was identified (Equation 2), revealing metabolite contribution to the VAS ‘Feeling High’ effects observed after (S)‐ketamine administration. Re‐evaluation of the PKPD relationships of both compounds resulted in a final model with an exponential and linear concentration–effect relationship for (S)‐ketamine (Equation 3) and (S)‐norketamine, respectively, with IIV on baseline (Table 2). Our analysis identified (S)‐ketamine as the main driver for observed effects on VAS ‘Feeling High’ during IV treatment, whereas the metabolite exposure is almost solely responsible for observed effects during PO treatment (Figure S8, Figure 3). The PKPD model revealed that the probabilities for responses of 0 and 100 for the VAS ‘Feeling High’ remain relatively high throughout the studied range of (S)‐ketamine exposure (Figure 3). This is due to the high estimated SD or g() (Table 2, Equation 1), which in general increases the probabilities for scores closer to the boundaries regardless of the mean of the distribution or f() (Equation 1).

TABLE 2.

Parameter estimates for the final PKPD model describing the relationship of (S)‐ketamine and (S)‐norketamine concentrations with VAS ‘feeling high’.

Parameter	Estimate (RSE%)
Baseline ( ${\theta }_0$, f()) Corrected for (S)‐ketamine exponential base of 1	−5.47 (5.5) −4.47
SD (σ, g())	1.15 (8.4)
(S)‐ketamine exponential slope ( ${\theta }_1$, f())	1.013 (3.1)
(S)‐norketamine linear slope ( ${\theta }_1$, f())	0.019 (17.1)
IIV Baseline (CV%) ( ${\theta }_0$, f())	61.6 (48.5)

Note: coefficient of variation calculated as $\sqrt{{\mathrm{\omega }}^2}$ where ${\mathrm{\omega }}^2$ is the estimated variance.

Abbreviations: ${\mathrm{\theta }}_1$, PKPD relationship slope parameters (Equations (2) and (3)); g(), variance of the distribution (Equation 1); CV, coefficient of variation; f(), mean of the distribution (Equation 1); IIV, interindividual variability; RSE, relative standard error; SD, standard deviation defined.

FIGURE 3.

Final PKPD relationship between (S)‐ketamine and (S)‐norketamine concentrations as represented by the expected value on the VAS ‘feeling high’ (A) and the probability of observing an extreme values of 0 or 100 for the VAS ‘feeling high’ (B). The response of a typical individual was simulated (i.e., no IIV included) over a continuous range of (S)‐ketamine concentrations and three levels of (S)‐norketamine concentration (0, 75 and 150 ng/mL), the expected value is the sum of all probabilities multiplicated with their respective category at a certain (S)‐ketamine concentration.

Model evaluation through GOF plots (Figure S9) and VPCs (Figure 4) demonstrated adequately predicted effects after IV dosing, but high predicted variability with a large range towards lower values for measurements taken directly post‐dose. Effects observed after PO dosing were underpredicted, especially for the first observations right after dosing, where (S)‐ketamine and (S)‐norketamine concentrations are still too low to elicit an increase in the underlying mean of the distribution or f() (Figures S10–S12).

FIGURE 4.

Confidence interval visual predictive checks (ciVPC's) of the final pharmacodynamic model with combined effects for (S)‐ketamine and (S)‐norketamine for VAS ‘feeling high’ over time. Panels in the top right corners zoom in on the first two timepoints. Solid lines represent the median (thick) and 80% interval (thin) of the observed data, shaded areas represent the 95% confidence interval for the median (pink) and 80% interval (purple) based on the simulated data.

3.3. PKPD model simulation and external evaluation

Assuming that responses on the VAS ‘Feeling High’ after PO dosing should be similar to the IV regimen to elicit antidepressant effects in patients, then the studied 0.2 and 0.45 mg/kg PO doses would not have been sufficient (Figure 4). ²⁶ Instead, simulations with the final PKPD model indicated that a PO dose of approximately 1.0 mg/kg would be required to achieve similar subjective VAS ‘Feeling High’ effects (Figure 5).

FIGURE 5.

Simulated expected response for a typical individual (70 kg) on the VAS ‘feeling high’ using the final PKPD model.

Lastly, the final PKPD model was externally evaluated through comparison of model simulations to observed data published by Simons et al., where the VAS ‘High’ was measured after sublingual and buccal administration of oral (S)‐ketamine thin films (50 and 100 mg) ⁴⁰ (Figure 6). Based on the observed concentrations of both parent and metabolite, it seems the developed model overpredicts the VAS response. However, if it is assumed that the observed effects should only be attributed to (S)‐ketamine concentrations without contribution of (S)‐norketamine, our model would not have predicted effects sufficient to match the observed responses seen for the 50 mg treatment as well as the observations after t_max for the 100 mg treatment. This shows that some contribution of (S)‐norketamine is necessary to explain the observed results but that the current (S)‐norketamine PKPD relationship may be overestimated.

FIGURE 6.

(A) Observed (S)‐ketamine and (S)‐norketamine concentrations and (B) observed (solid line) and predicted (dashed lines) response on the VAS ‘feeling high’ after sublingual or buccal administration of oral (S)‐ketamine thin film as reported by Simons et al. ⁴⁰ reported mean concentrations were used as input for the predicted response of a typical participant. The observed VAS responses (solid green line) represent the median reported VAS ‘high’ transformed from a 10‐cm scale to a 0–100 range. The dashed lines represent the predicted responses using the final PKPD model.

4. DISCUSSION

This report presents the development of a population PKPD model that adequately describes plasma concentrations of (S)‐ketamine and its pharmacologically active metabolites (S)‐norketamine and (S)‐hydroxynorketamine, as well as their relation to subjective VAS ‘Feeling High’ effects in healthy participants after both IV and PO administration of (S)‐ketamine. With this quantitative modelling approach, a significant contribution of (S)‐norketamine to subjective drug effects was identified, which is relevant to dose selection for and design of future clinical trials investigating PO (S)‐ketamine in TRD.

To adequately describe IV and PO (S)‐ketamine PK data, a semi‐physiological population PK model previously published (2018) was refitted, resulting in increased estimated central distribution volume (V_d). ²⁰ Although this model included crossover data with PO and IV bolus administrations, it treated the data as separate individuals, whereas the current model required individuals to have consistent distribution estimates across administration routes, which may explain the disparate V_d. However, their data did allow for identification of gut wall metabolism, which could not be estimated in the present study, likely due to their larger total sample size (n = 56) and earlier scheduled PK samples, which may have captured the oral absorption kinetics more accurately than in the current model.

Development of the semi‐physiologic PK model revealed potential non‐linear PK for (S)‐ketamine, with slower absorption at the higher PO dose. However, this was not reported previously for other (S)‐ketamine PK models based on more extensive PO data or reaching higher exposures ²⁰ , ⁴⁸ Therefore, no definitive conclusions can be drawn on the presence of non‐linear PK, for which further investigation over a larger range of PO doses would be required. Nonetheless, as slow absorption potentially results in prolonged t_max at higher PO doses, this may necessitate a different safety monitoring strategy compared to lower PO doses and thus is considered a relevant observation when designing future studies in TRD.

Although (S)‐norketamine contributed significantly to the (S)‐ketamine PKPD relationship for acute effects on the VAS ‘Feeling High’, an additional contribution by (S)‐hydroxynorketamine or other pharmacologically active metabolites cannot be ruled out. (S)‐norketamine's affinity for NMDAR is significantly higher than (S)‐hydroxynorketamine, while (S)‐hydroxynorketamine acts on the α7nACh receptor and may in part promote antidepressant effects through a different mechanism of action. ⁹ , ²¹ , ²² , ²³ Still, in the current report only (S)‐norketamine was selected for PKPD modelling, as similar changes in metabolite to (S)‐ketamine ratio over time precludes disentanglement of the two metabolites. Furthermore, as no antidepressant efficacy can be measured in healthy participants, acute subjective effects were used as endpoint for this analysis. Therefore, these results may only cautiously translate to clinical antidepressant effects, as this relies on the assumption that psychotomimetic effects are necessary for antidepressant efficacy in TRD. ³⁰ , ³¹ , ³² To conclude, although (S)‐norketamine was considered the most informative in the context of the current study, future studies would be needed to clarify its effects through administration of IV (S)‐norketamine.

Previous PKPD analyses have been unsuccessful at identifying metabolite contributions to (S)‐ketamine's subjective drug effects. These studies ignored the bounded and integer nature of subjective questionnaire data and treated them as being continuous, which may have obscured metabolites PKPD relationships. ⁴⁰ , ⁴⁹ By using a bounded integer modelling approach, a metabolite contribution could be identified in the current model. However, the estimated SD or $g\left(\right)$ was higher than 1, and as this function represents consistency in responded values, it likely resulted in overpredicted variability in the data. Explanations could be the substantial amount of data at the boundary, especially at zero, or an intrinsic feature of the scale, where values closer to the extremes are more likely to be responded with. ³⁸ Still, when the same approach was applied to study drug intensity ratings of the 5‐hydroxytryptamine 2A (5‐HT2A) receptor agonist N,N‐dimethyltryptamine (DMT) in healthy participants, the estimated SD was 0.275, indicating a high SD estimate is not necessarily associated with scales measuring subjective psychomimetic effects. ⁵⁰ As a high frequency of zero's is inherent to a scale measuring subjective effects of psychoactive substances in healthy participants, this requires additional investigation and possibly alternative approaches to bounded integer modelling such as beta distributions or zero inflation may need to be explored. ⁵¹ , ⁵²

VAS ‘Feeling High’ effects were underpredicted in the period directly following PO dosing and therefore well in advance of (S)‐ketamine or (S)‐norketamine peak concentrations being achieved. As no effects would have been expected at this timepoint, this possibly indicates the presence of an anticipatory and/or placebo effect. In contrast to these findings, very few responses higher than zero on the VAS ‘Feeling High’ were observed during the placebo occasion of the clinical study; thus, no placebo modelling was performed. ²⁶ Potentially, these effects only occur after some threshold level is experienced, leading to an exaggeration of the participants' experience. This may be related to the crossover aspect of the study if this effect is dependent on the treatment sequence due to functional unblinding, where the participants' anticipation for later treatment periods is increased if no or limited subjective effects were experienced during the first visit. Visual inspection of the data did not point towards such a trend, although this is hard to assess with a small dataset (n = 17). Regardless, this may have influenced the SD or g() estimate or alternatively, the PKPD relationship for (S)‐norketamine is overestimated and partly also represents this placebo and/or anticipation effect, in line with the overprediction of effects as observed by Simons et al. ⁴⁰ As such, more placebo‐controlled data for PO dosing would be necessary in order to rule out a placebo‐induced overestimation of the metabolite effect.

Based on the current model, a 1.0 mg/kg PO (S)‐ketamine dose is expected to induce subjective drug effects similar to 0.4 mg/kg administered IV over 40 min, which has repeatedly demonstrated antidepressant effects following a single dose in TRD. ⁷ However, taking the potential overestimation of the metabolite effect into account, the PO dose required for antidepressant efficacy may actually be higher. This is arguably supported by a recent study investigating the antidepressant efficacy of 30‐mg PO (S)‐ketamine (0.43 mg/kg for a 70‐kg individual) administered thrice daily for 6 weeks in TRD. Although no significant antidepressant effects were reported following the randomized, placebo‐controlled, double blind 6‐week treatment period, these were in fact observed in a consecutive open‐label part following the initial 6‐week period where PO doses were titrated individually, ranging from 0.5 to 3.0 mg/kg. ⁵³ This highlights the utility of the current PKPD model, where its clinical trial simulations can be used to guide dose ranging studies in TRD and improve their efficiency. Notably, this statement assumes comparable PK and PD of oral (S)‐ketamine between TRD patients and healthy participants. While a neuropsychiatric disorder such as TRD is unlikely to affect PK, which is supported by the nonsignificant effect of disease status in an intranasal PK study, the validity of this assumption for PD remains uncertain and warrants direct comparison of VAS ‘Feeling High’ responses between healthy and patient populations. ⁴⁸ Therefore, the safety and efficacy of the proposed 1.0 mg/kg PO (S)‐ketamine dose in TRD would still need to be further established, preferably in a RCT setting.

To conclude, the development of a novel (S)‐ketamine PKPD model has led to new insights into its clinical pharmacology. Subjective NMDAR‐mediated drug effects were attributed to both (S)‐ketamine and its pharmacologically active metabolite (S)‐norketamine. In addition, the model revealed the potential presence of potential non‐linear PK and placebo and/or anticipation effects. Finally, the PKPD model has shown to be useful for interpretation of recent clinical trial data, and, when aware of the underlying assumptions, can be applied for guidance of future clinical study designs through the simulation of suitable PO doses to investigate for treatment of TRD.

AUTHOR CONTRIBUTIONS

Marije E. Otto, Johan G. C. van Hasselt and Linda B S. Aulin designed the research. Marije E. Otto performed the data analysis and wrote the manuscript. Joost C. van Mechelen, Laura G.J.M. Borghans, Gabriël E. Jacobs, Johan G. C. van Hasselt and Linda B S. Aulin reviewed the manuscript. All authors read and approved the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Figure S1: Model structure of the previously presented semi‐physiological pharmacokinetic model by Ashraf et al. (2018) on which the final redeveloped model structure was based.

Figure S2: Confidence interval visual predictive checks (ciVPC) based on the model of Ashraf et al. (2018). Results are shown for intravenous (I.) and oral (II. and III.) (S)‐ketamine administration. 1) (S)‐ketamine, 2) (S)‐norketamine, a) ciVPC over the whole studied time period, b) ciVPC up to 4 h, c) ciVPC of the percentage predicted to be below the limit of quantification (BLQ). Solid lines represent the median (thick) and 80% interval (thin) of the observed data, shaded areas represent the 95% confidence interval (pink) for the median (a,b) or percentage (c) and 80% interval (purple) based on the simulated data.

Figure S3: Empirical Bayes Estimates distributions for (A) IIV and BOV (i.e. for Ka only) and (B) standard errors (SE) from the IPPSE step for individual parameters.

Figure S4: Goodness of fit plots based on the final semi‐mechanistic pharmacokinetic model. A) Conditional weighted residuals with interaction (CWRESI) over time, B) CWRESI over population predictions.

Figure S5: Individual observed (black) and predicted (red) concentrations of (S)‐ketamine, (S)‐norketamine and (S)‐hydroxynorketamine up to 10 h.

Figure S6: Confidence interval visual predictive checks (ciVPC) based on the final semi‐mechanistic pharmacokinetic model. Results are shown for intravenous (I.) and oral (II. and III.) (S)‐ketamine administration. 1) (S)‐ketamine, 2) (S)‐norketamine, 3) (S)‐hydroxynorketamine, a) ciVPC over the whole studied time period, b) ciVPC up to 4 h, c) ciVPC of the percentage predicted to be below the limit of quantification (BLQ). Solid lines represent the median (thick) and 80% interval (thin) of the observed data, shaded areas represent the 95% confidence interval (pink) for the median (a,b) and percentage (c) and 80% interval (purple) based on the simulated data.

Figure S7: Observed responses to the VAS ‘Feeling High’ over time after placebo, oral or intravenous (S)‐ketamine administration. The shape represents outlier datapoints that, next to the placebo data, were excluded from analysis. The color represents the occasion or order in which participants received the specific treatment.

Figure S8: Individual predicted change in latent variable (φ) from baseline over time as caused by (S)‐ketamine, (S)‐norketamine, and both together (‘Total effect’). The latent variable represents the VAS ‘Feeling High’ in the bounded integer model.

Figure S9: Conditional individual weighted residuals for categorical and count data (CWRESlike) for the final pharmacodynamic model with combined effects for (S)‐ketamine and (S)‐norketamine.

Figure S10: Confidence interval visual predictive checks (ciVPC's) of the final pharmacodynamic model with combined effects for (S)‐ketamine and (S)‐norketamine for VAS ‘Feeling High‘over individual predicted (S)‐ketamine concentration. Solid lines represent the median (thick) and 80% interval (thin) of the observed data, shaded areas represent the 95% confidence interval for the median (pink) and 80% interval (purple) based on the simulated data.

Figure S11: Confidence interval visual predictive checks (ciVPC's) of the final pharmacodynamic model with combined effects for (S)‐ketamine and (S)‐norketamine for VAS ‘Feeling High‘over individual predicted (S)‐norketamine concentration. Solid lines represent the median (thick) and 80% interval (thin) of the observed data, shaded areas represent the 95% confidence interval for the median (pink) and 80% interval (purple) based on the simulated data.

Figure S12: Individual confidence interval visual predictive checks (ciVPC's) of the final pharmacodynamic model with combined effects for (S)‐ketamine and (S)‐norketamine for VAS ‘Feeling High’ over time. Solid lines and dots represent observed values, the pink shaded area represents the 95% confidence interval based on the simulated data.

Otto ME, Jacobs GE, van Mechelen JC, Borghans LGJM, van Hasselt JGC, Aulin LBS. Pharmacokinetics and pharmacodynamics of intravenous and oral (S)‐ketamine: Investigating metabolite contribution to subjective effects. Br J Clin Pharmacol. 2026;92(7):2260‐2271. doi: 10.1002/bcp.70503

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request.