High-dose naloxone, an experimental tool uncovering latent sensitisation: pharmacokinetics in humans

Abstract

Background

Naloxone, an opioid receptor antagonist, is used as a pharmacological tool to detect tonic endogenous activation of opioid receptors in experimental pain models. We describe a pharmacokinetic model linking naloxone pharmacokinetics to its main metabolite after high-dose naloxone infusion.

Methods

Eight healthy volunteers received a three-stage stepwise high-dose i.v. naloxone infusion (total dose 3.25 mg kg⁻¹). Naloxone and naloxone-3-glucuronide (N3G) plasma concentrations were sampled from infusion onset to 334 min after infusion discontinuation. Pharmacokinetic analysis was performed using non-linear mixed effect models (NONMEM). The predictive performances of Dowling's and Yassen's models were evaluated, and target-controlled infusion simulations were performed.

Results

Three- and two-compartment disposition models with linear elimination kinetics described the naloxone and N3G concentration time-courses, respectively. Two covariate models were developed: simple (weight proportional) and complex (with the shallow peripheral volume of distribution linearly increasing with body weight). The median prediction error (MDPE) and wobble for Dowling's model were –32.5% and 33.4%, respectively. For Yassen's model, the MDPE and wobble were 1.2% and 19.9%, respectively.

Conclusions

A parent–metabolite pharmacokinetic model was developed for naloxone and N3G after high-dose naloxone infusion. No saturable pharmacokinetics were observed. Whereas Dowling's model was inaccurate and over-predicted naloxone concentrations, Yassen's model accurately predicted naloxone pharmacokinetics. The newly developed covariate models may be used for high-dose TCI-naloxone for experimental and clinical practice.

Clinical trials registration

NCT01992146.

Editor's key points.

• •High-dose naloxone, an opioid receptor antagonist, is used as a tool to detect tonic endogenous activation of opioid receptors in experimental pain models, but its pharmacokinetics are not clear.
• •A model linking the pharmacokinetics of naloxone and its main metabolite was developed using data from eight healthy volunteers after high-dose naloxone infusion.
• •The model adequately described the concentration–time course without saturable pharmacokinetics for either naloxone or naloxone-3-glucuronide, which will inform future trials.

High-dose naloxone is used to study the constitutive activity of endogenous opioid receptor systems. These studies require very high doses of naloxone, and human pharmacodynamic results are inconsistent. A previous study including 12 healthy subjects used an experimental model of a cutaneous first-degree burn injury, followed by a 17 min infusion of naloxone (2 mg kg⁻¹) 168 h later. Naloxone, but not a saline placebo, caused recurrence of hyperalgesia at the post-inflammatory site in four subjects,¹ consistent with animal studies of latent sensitisation and endogenous opioid receptor system activation. However, the proportion of responders was much lower in the human study compared with animal studies. It is unclear whether the inconsistent results were attributable to the experimental model, or because the naloxone concentrations were not well controlled in the studies.

The pharmacokinetic profile of naloxone has been characterised for typical doses (5–30 μg kg⁻¹),2, 3, 4, 5 but not for high doses (>1500 μg kg⁻¹) such as those used in our previous study. The objective of this study was to characterise the pharmacokinetics of high-dose naloxone by developing a pharmacokinetic model suitable for target-controlled naloxone infusion (TCI) so that concentration–response curves can be established in the presence of stable naloxone concentrations.

Methods

Study design

This study presents the pharmacokinetic analysis of data collected during our basic pain study approved by the Committee of Health Research Ethics of the Capital Region (H-15002712), Danish Medicines Agency (2015-000793-36) and the Data Inspection Authority of the Capital Region (RH-2014-30-1150, I-Suite nr. 02815). The study was registered in EUDRACT (2015-000793-36) and ClinicalTrials.gov (NCT01992146). The protocol is shown in detail in Supplementary Data 1. All subjects received written and verbal information before providing written consent.

Subjects

Eligibility criteria were healthy male, ASA physical status 1, age 18–35 yr, BMI 18–30 kg m⁻², urine sample without traces of opioids, and signed informed consent. Detailed eligibility criteria can be found in Supplementary Data 2.

Procedure

Subjects were instructed to follow current guidelines for pre-anaesthesia fasting: no solid food intake was allowed from 00:00, whereas fluid intake was allowed until 06:00 am. Subjects reported to the laboratory from 8:00 to 9:00 am followed by a medical examination, urine sampling for opioid analysis, and signing of the informed consent. Monitoring was by continuous ECG and pulse oximetry, and intermittent non-invasive blood pressure and ventilatory frequency measurements. The Clinical Opiate Withdrawal Scale (COWS),⁶ an examiner-based scale grading potential symptoms of opioid withdrawal, was assessed at baseline and during naloxone infusion. Subjects rested comfortably in a bed in the supine position. An i.v. catheter (16 G [ID 1.7 mm]) was inserted bilaterally in antecubital veins, and a rehydration solution (Fresenius Kabi Denmark, Copenhagen, Denmark) was infused at a rate of 50 ml h⁻¹. At 95 min, subjects were allowed a standardised meal and begin to ambulate. Monitoring was continued until 100 min, and at 340 min the i.v. catheters were removed; when ‘street fitness’ was attained subjects were discharged. Subjects were contacted by phone the day after the study day to check on the occurrence of potential side-effects.

Drug administration

A total dose of 3.25 mg kg⁻¹ naloxone was administered i.v. in a step-wise approach for 75 min (time 0–75 min) with a step duration of 25 min (Table 1). Before trial initiation, we simulated with variability using Dowling's model⁴ to obtain the expected naloxone concentrations when administered under our infusion scheme (Table 1). Expected naloxone plasma concentrations (median [95% confidence interval, CI]) were 344 ng ml⁻¹ [130; 567], 1060 ng ml⁻¹ [400; 1750], and 3200 ng ml⁻¹ [1200; 5280] at steps one (15–25 min), two (40–50 min), and three (65–75 min), respectively (Table 1).⁷ A 50 ml syringe-based pump system was used for bolus administration and infusion (B. Braun Medical Perfusor Space Infusion Pump System; Model 8713030U; B. Braun Melsungen AG, Melsungen, Germany). Pump settings were manually adjusted during naloxone administration.

Table 1.

Intravenous administration of naloxone (4 mg ml⁻¹) using a three-step infusion scheme for bolus and infusion.

	Time (min)	Dose (mg kg−1)	Dose (mg 70 kg−1)	Injection volume (ml 70 kg−1)	Infusion rate (ml 70 kg−1 min−1)
Bolus 1	0–1	0.02	1.50	0.38	0.38
Infusion 1	1–25	0.23	17.25	4.31	0.18
Bolus 2	25–26	0.06	4.50	1.13	1.13
Infusion 2	26–50	0.69	51.75	12.94	0.54
Bolus 3	50–51	0.18	13.50	3.38	3.38
Infusion 3	51–75	2.07	155.25	38.81	1.62
Total	–	3.25	243.75	60.94	–

Blood sampling and bioanalytical method

Venous blood samples (n=22; 10 ml each) for determination of plasma concentrations of naloxone and naloxone-3-glucuronide (N3G) were assayed at 0, 17, 20, 23, 41, 44, 47, 67, 70, and 75 min during naloxone infusion and at 76, 77, 78, 79, 80, 82, 86, 94, 110, 142, 206, and 334 min after infusion discontinuation. Samples were drawn into EDTA tubes and immediately centrifuged (2200 g for 10 min), then plasma was stored in Eppendorf tubes at –20°C. At the end of the study day, tubes were stored at –80°C. Plasma concentrations of naloxone and N3G were determined using a fully validated liquid chromatography with tandem mass spectrometry (LC-MS/MS) method.⁸

Population pharmacokinetic analysis

Population pharmacokinetic analysis of data was performed using non-linear mixed-effects modelling (NONMEM version 7.3, ICON Development Solutions, Ellicott City, MD, USA).⁹ Parameter estimation was performed using the first-order conditional estimation method with interaction (FOCE-I). Pirana was used as a user interface for the model development procedure, and Perl speaks NONMEM (PsN) version 4.4.0 was used for controlling NONMEM runs.¹⁰ R version 3.1.1 (R Foundation for Statistical Computing) was used for dataset creation and graphical visualisation¹¹ using the plyr and ggplot2 packages.12, 13 The ADVAN6 library routine in NONMEM was used for all evaluated models.

Inter-individual variability (IIV) was assumed to be log-normally distributed and was parameterised as $P_i=\Theta \cdot \text{exp}\left({\eta }_i\right)$, where Θ is the population estimate for parameter P and ${\eta }_i\sim N(0,{\text{ω}}^2)$is a normally distributed random variability with a mean of 0 and a variance of ${\text{ω}}^2$. Additive, proportional, and combined additive and proportional residual error models were evaluated. The residual errors were assumed to be normally distributed with mean zero and variance σ.²

Model selection was based on scientific plausibility and the maximum likelihood criterion, that is, the difference in the objective function value (OFV), defined as minus two times the log likelihood. A decrease of more than 3.84 in OFV, which corresponds to a P— value of <0.05, was considered sufficient for inclusion of one model parameter. Standard goodness-of-fit plots (i.e. population and individual predicted values plotted against observed values, weighted residual plotted against time, and weighed residuals plotted against population predictions) were evaluated throughout the modelling procedure. Descriptive ability regarding central tendency and variability was evaluated using visual predictive checks (VPCs) based on 1000 Monte Carlo simulations. The 5th, 50th, and 95th percentiles of the simulated data alongside the observed data were plotted in the VPC. Parameter precision was evaluated during the modelling procedure using parameter relative standard errors (RSEs) as obtained by the covariance step. For the final model, log-likelihood profiling was used for obtaining parameter 95% CIs.

Structural model development was performed sequentially for parent and metabolite data. A pharmacokinetic model for naloxone alone was initially developed, which was followed by development of a pharmacokinetic model of N3G, where all naloxone-related parameters were kept fixed until a viable model was identified. Subsequently, all naloxone and N3G parameters were estimated simultaneously. One-, two-, and three-compartment models were evaluated for naloxone and N3G disposition, with clearance from the central compartment. Saturable elimination kinetics were explored, because of the large naloxone doses used in the study, using Michaelis–Menten kinetics for both naloxone and N3G.

Screening of covariates was undertaken by applying stepwise covariate model building (SCM) using the automated procedure in PsN.¹⁰ This method involved stepwise testing of linear and non-linear relationships in a forward inclusion based on a change in the OFV of 3.8, corresponding to a P-value <0.05 and backward elimination on a change in the OFV of 6.6, corresponding to a P-value <0.01. The covariates tested were age, height, weight, BMI, and body surface area (BSA; calculated using Mosteller's formula).¹⁴ In addition to the model obtained via this formal covariate searching routine (complex model), and in order to create a general model that can easily be used for creation of TCI algorithms based on body weight, we report a second, weight proportional model (simple model), where all volumes and clearances are assumed to be proportional to individual body weight.

Performance error

The predictive performance of our simple and complex models, and that of Dowling's and Yassen's models,⁵ was evaluated using the method proposed by Varvel and colleagues.¹⁵ For each time point that a blood sample was collected, the prediction error (PE) was calculated as:

$$PE\left(\%\right)=100\cdot \left(c_o-c_p\right)/c_p$$

where c_o is observed plasma concentration and c_p is the algorithm-predicted plasma concentration. To evaluate the predictive performance of the algorithms, a set of metrics was calculated for each individual i at each time point j. These were the median prediction error (MDPE) as a measure of bias.

$$\text{MDPE}=\text{Median}({\text{PE}}_{ij},j=1,\dots ,N_i)$$

The median absolute prediction error (MDAPE), as a measure of accuracy.

$$\text{MDAPE}=\text{Median}(\left|{\text{PE}}_{ij}\right|,j=1,\dots ,N_i)$$

The wobble as a measure of within-individual variation relative to the expected concentrations.

$$\text{Wobble}=\text{Median}(\left|{\text{PE}}_{ij}-{\text{MDPE}}_{ij}\right|,j=1,\dots ,N_i)$$

And divergence as a measure of time-related trends of over- or under-prediction as compared with the expected concentrations, which is defined as the slope of the linear regression of each individual's PE_ij against time.

Simulations

With our weight proportional pharmacokinetic model, we performed simulations of plasma concentration TCI using the PKPD Tools for Excel package for Microsoft Excel^® developed by T. Schnider and C. Minto (available from http://pkpdtools.com/; last accessed October 2018). The TCI algorithm comprised a 40 min naloxone infusion at a target plasma concentration of 1000 ng ml⁻¹. The maximum infusion rate was chosen so that the target naloxone concentration is obtained within 10 min for a 70 kg subject. The resulting infusion profiles were compared with those resulting from Dowling's and Yassen's pharmacokinetic models.

In addition to the TCI simulations, we used the weight proportional pharmacokinetic model to simulate naloxone and N3G plasma concentrations after the naloxone administration algorithms that we used in our two previous reports on latent sensitisation, where total naloxone doses of 21 and 2 mg kg⁻¹ was administered.1, 16

Results

Subjects

Nine subjects were eligible for the study (Supplementary Data 2). One subject was excluded because of a previous medical condition. Eight subjects completed the study as per protocol. The mean (range) age was 25.9 (24.4–28.9) yr, height 182 (177–194) cm, and weight 79.6 (61–92) kg. One subject had a BMI of 30.4 kg m⁻², slightly exceeding the stipulated limit of 30 kg m⁻².

Adverse events

No serious adverse events were experienced. One subject experienced unspecific general discomfort and anxiety during the beginning of the infusion. The symptoms disappeared after 10 min, and the subject expressed a desire to continue the infusion. The subject had experienced several psychosocial stressors before the study, and the subject himself attributed the distress mainly to these stressors.

Naloxone and N3G pharmacokinetic model

The observed plasma concentration profiles of naloxone and N3G after naloxone infusion are presented in Figure 1. The dataset consisted of 166 naloxone and 166 N3G plasma concentrations. No concentrations were below the lowest limit of quantification. The parameter estimates of the simple and complex pharmacokinetic models are presented in Table 2.

Fig 1.

Observed concentration–time profiles for naloxone and naloxone-3-glucuronide (N3G) after high-dose naloxone infusion. Solid lines present the population predicted trajectory for the simple body weight proportional model (red), Dowling's model (blue), and Yassen's model (orange).4, 5 Thin red lines present the individual post hoc trajectories based on the simple body weight proportional model. The inset presents the period after infusion discontinuation (i.e. time interval 76–334 min) where the distribution and elimination kinetics are most evident (note the log-concentration scale). Only predictions based on the simple PK model are presented for simplicity.

Table 2.

Population pharmacokinetic parameter estimates for naloxone and naloxone-3-glucuronide. Parameter 95% confidence intervals were obtained using log-likelihood profiling. Inter-individual variability approximate coefficient of variation (% CV) is reported as $\sqrt{{\omega }_{}^2}\cdot 100$

Parameter	Description	Simple model (weight proportional)				Complex model (weight as covariate)
		Unit	Estimate	95% confidence intervals		Unit	Estimate	95% confidence intervals
				Lower	Upper			Lower	Upper
Naloxone
CLNx	Naloxone systemic clearance	L kg−1 min−1	0.049	0.044	0.054	L min−1	3.84	3.48	4.23
V1Nx	Central volume of distribution	L kg−1	0.408	0.151	1.010	L	34.44	13.6	83.25
Q1Nx	Inter-compartmental clearance (V1Nx–V2Nx)	L kg−1 min−1	0.046	0.029	0.083	L min−1	3.93	2.98	4.83
V2Nx	Shallow peripheral volume of distribution	L kg−1	0.636	0.284	1.221	L	70.15	45.12	100.48
Q2Nx	Inter-compartmental clearance (V1Nx–V3Nx)	L kg−1 min−1	0.026	0.012	0.040	L min−1	0.92	0.41	1.71
V3Nx	Deep peripheral volume of distribution	L kg−1	1.637	1.118	2.041	L	82.12	54.31	154.1
Weight–V2	Weight–V2Nx linear relationship covariate	–	–	–	–	unitless	0.044	0.028	0.064
Naloxone-3-glucuronide
CLN3G	N3G systemic clearance	L kg−1 min−1	0.008 Naloxone-3-glucuronide	0.006	0.010	L min−1	0.6	0.49	0.73
V1N3G	Central volume of distribution	L kg−1	0.283	0.218	0.357	L	22.42	17.09	28.75
Q1N3G	Inter-compartmental clearance (V1N3G–V2N3G)	L kg−1 min−1	0.004	0.003	0.007	L min−1	0.33	0.21	0.54
V2N3G	Peripheral volume of distribution	L kg−1	0.254	0.171	0.444	L	20.88	14.29	39.34
Inter-individual variability		Variance (ω2)		%CV		Variance (ω2)		%CV
Naloxone
CLNx		0.013		11.4		0.012		11
V1Nx		1.26		112.2		1.22		110.5
V2Nx		0.16		40		–		–
Naloxone-3-glucuronide
CLN3G		0.079		28.1		0.059		24.2
V1N3G		0.06		24.5		0.08		27.8
Residual error
σNx,prop		0.051		22.6		0.051		22.6
σN3G,prop		0.029		17.1		0.03		17.4

Naloxone concentration–time profiles were best described by a three-compartment model. Inter-individual variability was estimated for clearance (CL_Nx), the central volume of distribution (V1_Nx), and the shallow peripheral volume of distribution (V2_Nx). Residual variability was described using a proportional residual error model. The goodness-of-fit plots (Fig. 2) and the VPC (Supplementary Data 2) show that the naloxone concentrations were well described for the different phases of naloxone infusion. Inclusion of saturable elimination kinetics did not improve the fit of the model, meaning that the hypothesis of linear kinetics cannot be rejected with the current data.

Fig 2.

Goodness-of-fit plots presenting the population and individual predicted naloxone concentrations for the simple body weight proportional model and the complex model with body weight linearly related to V2_Nxvs time and observed naloxone concentrations. The black line is the line of identity and the red line is a linear regression.

In addition to the simple weight proportional model, we used the SCM method for formal covariate search (complex model).¹⁰ The best model included only weight for the shallow peripheral volume of distribution via a linear relationship. No significant non-linear relationships were identified. Inclusion of weight via a linear relationship led to a reduction of V2_Nx IIV from 44.9% CV to 15.6% CV. Simplification of the complex model by removing V2_Nx IIV did not lead to significantly worse fit (dOFV=2.1, –1 parameter), and it was thus decided that it would not be included in the final complex model.

The formation of N3G was rapid, with no significant delay between naloxone and metabolite plasma concentrations. The disposition of N3G was best described using a two-compartment model. The goodness-of-fit plots (Fig. 3) and the VPC (Supplementary Data 2) show that both the simple and the complex models described the N3G concentration time-course well. The inclusion of saturable elimination kinetics did not improve the fit of the model, indicating linearity in elimination kinetics for N3G as well. Inter-individual variability was estimated for metabolite clearance (CL_N3G) and central volume of distribution (V1_N3G).

Fig 3.

Goodness-of-fit plots presenting the population and individual predicted naloxone-3-glucuronide (N3G) concentrations for the simple body weight proportional model and the complex model with body weight linearly related to V2_Nxvs time and observed N3G concentrations. The black line is the line of identity and the red line is a linear regression.

Performance error

The metrics of the predictive performance of the TCI are presented in Table 3. The overall bias (MDPE) for Dowling's model⁴ was –32.5%, meaning that the model-predicted concentrations were over-predicted for the entire observation period. Similarly, inaccuracy (MDAPE) and the within-subject variability in PE (wobble) were high (>25%). The overall predictive performance indices of Yassen's model⁵ were considerably better than those for Dowling's model⁴ and comparatively close to those for the simple and complex pharmacokinetic models, apart from divergence, indicating that Yassen's model⁵ led to overall very good predictions of the achieved naloxone concentrations, but suffered from time-dependent prediction errors.

Table 3.

Indices of model predictive performance. Results are presented as mean (range). MDPE, median performance error; MDAPE, median absolute performance error.

	MDPE	MDAPE	Wobble	Divergence
Dowling and colleagues4	–32.5 (–57.5 to –15.5)	32.5 (15.5–57.5)	33.4 (10–60.2)	0.34 (0.06–0.67)
Yassen and colleagues5	1.2 (–9.4 to 10.5)	7 (1–10.5)	19.9 (10.8–43.4)	1.46 (0.34–2.82)
Simple PK model	–0.1 (–3.4 to 6.4)	2.8 (0.8–6.4)	13.2 (8.0–19.4)	0.01 (–0.12 to 0.16)
Complex PK model	0.3 (–6.1 to 6.3)	3.1 (0.1–6.3)	12.3 (5.6–16.0)	0.02 (–0.16 to 0.10)

Simulations

A potential application of our pharmacokinetic model is for high-dose TCI-naloxone for experimental or clinical practice. Figure 4 presents the predicted infusion rate over time and the cumulative dose required to achieve a target naloxone plasma concentration of 1000 ng ml⁻¹. The infusion rate was simulated such that, when using the simple weight proportional model, a 70 kg subject reaches the target plasma concentration within 10 min of TCI initiation (8 mg min⁻¹). Infusion rates, cumulative dose, and predicted plasma concentration of the simple model were compared with those predicted by Dowling's and Yassen's models. Yassen's model required infusion rates that were remarkably similar to our new model, whereas Dowling's model differed substantially. The TCI simulation file can be found in Supplementary data 3.

Fig 4.

Comparison of infusion rate (left), target-controlled infusion (TCI) predicted naloxone concentrations (middle) and cumulative naloxone dose over time (right) for TCI driven by the simple body weight proportional model (NONMEM) (red), Dowling's model (blue) and Yassen's model (orange). A 40 min infusion of naloxone at a target plasma concentration of 1000 ng ml⁻¹ was simulated for a 70 kg subject for all models (solid line), and an 80 kg (dashed line) and 90 kg (medium dashed line) subject for the new model. The maximum infusion rate (8 mg min⁻¹) was chosen in order to obtain the target naloxone concentration within 10 min for a 70 kg subject when evaluated using the simple model. Only predictions based on the simple PK model are presented for simplicity.

In addition to TCI simulations, we performed simulations where naloxone was administered under the infusion algorithms used in our previous publications on latent sensitisation,1, 16 where plasma samples were not collected thus leading to a limited understanding of the naloxone plasma concentration–time course (Fig. 5). A simulated plasma concentration–time course based on Pereira and colleagues¹⁶ from 2013, which did not show latent sensitisation, yielded a C_max of 7.2 ng ml⁻¹. In contrast, the simulated plasma concentration–time course based on Pereira and colleagues⁷ from 2015, which did show unmasking of latent sensitisation, yielded a C_max of 1200 ng ml⁻¹.

Fig 5.

Simulated population profiles for a typical 70 kg individual based on the simple body weight proportional model for naloxone and naloxone-3-glucuronide. The dosing schemes are based on two published reports.1, 16 Prediction in red: i.v. bolus of naloxone (5 μg kg⁻¹, 2 min), followed by an initial infusion of 40 μg kg⁻¹ h⁻¹ over 20 min and a secondary infusion of 20 μg kg⁻¹ h⁻¹ over 8 min. Prediction in blue: i.v. bolus of naloxone (4 mg, 20 s), followed by a naloxone infusion of 7 μg kg⁻¹ h⁻¹ over 16.7 min.

Discussion

Opioid antagonists are used as pharmacological tools to detect the tonic activity of the endogenous opioid receptor systems in experimental pain models.¹⁷ Endogenous opioid receptor systems provide anti-nociception and analgesia and is key to the downstream modulation of nociceptive signal processing. Naloxone and naltrexone are opioid receptor antagonists that can suppress central endogenous opioid receptor systems, termed μ-opioid-receptor constitutive activity (MOR_CA).18, 19 Recent animal studies demonstrate that tissue inflammation produces a latent sensitisation that is masked by MOR_CA. This can last for months, even after re-establishment of normal nociceptive thresholds.20, 21, 22 High-dose naloxone (3–10 mg kg⁻¹) or naltrexone (1–3 mg kg⁻¹) disrupts MOR_CA and thus reinstates pain and precipitates nocifensive behaviour for approximately 120 min even when delivered 6 months after induction of inflammation.²¹ As latent sensitisation is considered to be a putative pathophysiological component in the development of persistent postsurgical pain, preliminary translational studies in humans have been carried out.1, 16

The main objective of the present study was to investigate the pharmacokinetics of naloxone and its main metabolite, N3G, after high-dose naloxone infusion in healthy subjects. The naloxone dose far exceeds the clinical dose range used to reverse severe opioid overdose,⁴ and the observed peak plasma concentrations of naloxone and N3G are more than 30 times higher than seen in previously published studies.2, 3, 4, 5 The rationale for the choice of this high dose is driven by the useful pharmacological properties of naloxone as a strict MOR antagonist, making it ideal for exploration of the physiological mechanisms of endogenous opioid receptor systems and their activation after injury.¹⁷

The densely sampled pharmacokinetic data and the large range of observed plasma concentrations enabled the development of a population pharmacokinetic model that simultaneously describes all individual concentration–time profiles. The model-based analysis of the pharmacokinetic data was performed for several reasons: first, to understand the disposition kinetics of both parent compound and metabolite; second, to explore whether naloxone in high concentrations follows a saturable elimination pathway; and, third, to evaluate the predictive performance of previously reported population pharmacokinetic models of naloxone.

The naloxone concentration–time profiles were well described using a three-compartment model with first-order elimination. The estimated clearance value for a 70 kg subject was 4.42 and 3.84 L min⁻¹ for the simple and complex pharmacokinetic models, respectively, which is in good agreement with a previously reported value estimated using non-linear mixed-effects modelling in pharmacokinetic data of healthy subjects (3.25 L min⁻¹).⁵ Interestingly, the estimated clearance value is 1.5 times higher than reported in Dowling's model,⁴ which was originally used for evaluating expected concentrations after our infusion algorithm. The estimated clearance value exceeds the blood flow of the liver (20 ml min⁻¹ kg⁻¹),²³ potentially because of extrahepatic naloxone metabolism. Indeed, extrahepatic glucuronidation likely contributes substantially to the systemic clearance of naloxone. For example, UGT2B7, the main metabolising enzyme of naloxone, is expressed in significant amounts in kidney²⁴ and intestine.25, 26 This high clearance is consistent with our observation that naloxone plasma concentrations decreased rapidly after discontinuation of the infusion, whereas naloxone plasma concentrations were observable for the entire observation period, mostly because of its re-distribution kinetics.

N3G pharmacokinetics were adequately described using a two-compartment model. Owing to the lack of data of i.v. administration of N3G alone, we could not determine its exact volumes of distribution. Instead, we assumed that glucuronidation to N3G was the sole clearance pathway for naloxone. To our knowledge, this study is the first to link the pharmacokinetics of naloxone to its glucuronide metabolite. Our simultaneous estimation modelling approach allows for information cross-talk between the pharmacokinetic parameters of the parent and the metabolite which is known to lead to less biased parameter estimates.²⁷

When evaluating the pharmacokinetic behaviour of a drug in concentrations far higher than previously reported, potential saturation of metabolic processes should be explored. The classical approach for investigating the presence of saturable pharmacokinetics is administration of multiple dose strengths, followed by comparison of the dose-corrected pharmacokinetic profiles.²⁸ In the present study, all dosing was performed with an i.v. infusion, with increasing increments over each infusion step. This means that residual naloxone concentrations from the previous step were already present. However, use of our modelling approach can overcome this and allowed us to test for saturable pharmacokinetics by comparing the statistical likelihood obtained by models with and without the inclusion of a saturable elimination pathway. Based on our analysis, no saturable pharmacokinetics were present, despite the very large range of concentrations observed. This supports the notion that naloxone can be used as a tool to detect activation of endogenous opioid receptor systems in experimental pain models.¹ If saturable pharmacokinetics had been demonstrated, recurrence of hyperalgesia in the post-inflammatory site seen in a restricted number of study subjects could be attributed to the unknown pharmacokinetic determinants and profiles. As this was not the case, the previously formulated hypothesis, that there may be a subgroup of the population that is more prone to latent sensitisation, may be accurate.¹ Nevertheless, the strong inter-individual variability seen in V1_NX, exceeding 100% CV, is a potential factor contributing to the low proportion of subjects in whom recurrence of hyperalgesia could be demonstrated. Future research should aim to directly correlate naloxone plasma concentrations to responder/non-responder rate via an integrated PK/PD modelling approach.

A target-controlled algorithm facilitates examination of concentration–response relationships and is a precautionary measure against individual overdosing, considering the very high dose range of naloxone. The predictive performance of the TCI algorithm using Dowling's model⁴ was overall poor with the predicted maximum concentrations being 3-fold higher than those observed. In contrast, Yassen's model⁵ led to considerably better predictive performance which was comparable with the performance measures of our newly developed model. It should be noted that our new model is intended for TCI simulations for subjects within the studied weight range, and care should be taken when extrapolating outside that range.

Comparison with our two previous studies1, 16 revealed that the simulated peak plasma concentrations of naloxone differed by 150-fold. This might explain the difference in pharmacodynamic response between the two studies (one yielded latent sensitisation, the other did not). Interestingly, data from a positron emission tomography (PET) study in humans demonstrated that naloxone 100 μg kg⁻¹ completely inhibited radionuclide labelled [¹¹C]-carfentanil binding to MOR.²⁹ Greater doses may be needed to sufficiently block MOR_CA, but an ‘off-target’ action of high-dose naloxone cannot be ruled out.¹

In conclusion, we developed a model that characterises the pharmacokinetics of naloxone and N3G after high-dose naloxone TCI in healthy subjects. The model adequately described the concentration–time course for both parent drug and metabolite. No saturable pharmacokinetics were found for either naloxone or N3G despite the very high concentrations observed. The developed pharmacokinetic model performed similarly to a previously proposed model,⁵ and will be used to design future trials to further investigate μ-opioid-receptor constitutive activity, endogenous opioid receptor systems, and latent sensitisation.

Authors' contributions

Study design and data analysis: TP, ADS, KK, DS, KM, BKT, TML, MUW.

Patient recruitment and data collection: ADS, MUW.

Writing of the first draft of the manuscript: TP.

Critical revisions and approval of the final version of the manuscript: all authors.

Declaration of interest

The authors declare that they have no conflicts of interest.

Funding

Aase og Ejnar Danielsens Fond, Brødrene Hartmanns Fond and Augustinus Fonden. US National Institutes of Health (DA37621 to BKT and MUW).

Handling editor: H.C. Hemmings Jr

Editorial decision: 13 December 2018

Appendix A. Supplementary data

The following are the Supplementary data to this article: