Present address: Department Internal Medicine VI, Clinical Pharmacology and Pharmacoepidemiology, University of Heidelberg, Heidelberg, Germany
Pharmacokinetic modelling of morphine, morphine-3-glucuronide and morphine-6-glucuronide in plasma and cerebrospinal fluid of neurosurgical patients after short-term infusion of morphine
Article first published online: 17 DEC 2002
British Journal of Clinical Pharmacology
Volume 54, Issue 6, pages 592–603, December 2002
How to Cite
Meineke, I., Freudenthaler, S., Hofmann, U., Schaeffeler, E., Mikus, G., Schwab, M., Prange, H. W., Gleiter, C. H. and Brockmöller, J. (2002), Pharmacokinetic modelling of morphine, morphine-3-glucuronide and morphine-6-glucuronide in plasma and cerebrospinal fluid of neurosurgical patients after short-term infusion of morphine. British Journal of Clinical Pharmacology, 54: 592–603. doi: 10.1046/j.1365-2125.2002.01689.x
- Issue published online: 17 DEC 2002
- Article first published online: 17 DEC 2002
- Received 12 February 2002, accepted 25 April 2002.
Aims Concentrations in the cerebrospinal fluid (CSF) are a useful approximation to the effect site for drugs like morphine. However, CSF samples, are available only in rare circumstances. If they can be obtained they may provide important insights into the pharmacokinetics/pharmacodynamics of opioids.
Methods Nine neurological and neurosurgical patients (age 19–69 years) received 0.5 mg kg−1 morphine sulphate pentahydrate as an intravenous infusion over 30 min. Plasma and CSF were collected for up to 48 h. Concentration time-course and interindividual variability of morphine (M), morphine-3-glucuronide (M3G) and morphine-6 glucuronide (M6G) were analysed using population pharmacokinetic modelling.
Results While morphine was rapidly cleared from plasma (total clearance = 1838 ml min−1 (95% CI 1668, 2001 ml min−1)) the glucuronide metabolites were eliminated more slowly (clearance M3G = 44.5 ml min−1 (35.1, 53.9 ml min−1), clearance M6G = 42.1 ml min−1 (36.4, 47.7 ml min−1)) and their clearance could be described as a function of creatinine clearance. The central volumes of distribution were estimated to be 12.7 l (11.1, 14.3 l) for morphine. Transfer from the central compartment into the CSF was also rapid for M and considerably slower for both glucuronide metabolites. Maximum concentrations were achieved after 102 min (M), 417 min (M3G) and 443 min (M6G). A P-glycoprotein exon 26 polymorphism previously found to be linked with transport activity could be involved in CSF accessibility, since the homozygous mutant genotype was associated (P < 0.001) with high maximum CSF concentrations of M but not M3G or M6G.
Conclusions From the population pharmacokinetic model presented, CSF concentration profiles can be derived for M, M3G and M6G on the basis of dosing information and creatinine clearance without collecting CSF samples. Such profiles may then serve as the link between dose regimen and effect measurements in future clinical effect studies.
Morphine is employed for the treatment of severe acute and chronic pain. Its mechanism of action, therapeutic and adverse effects as well as its distribution and biotransformation have been investigated by numerous researchers (for a recent review see ). Despite this large number of studies pivotal issues have remained unsolved. It is not entirely understood why in some patients sufficient pain relief is not achieved, or under which conditions the development of tolerance is to be expected [2, 3]. Since morphine exerts its effects in the CNS, one possible explanation for such interpatient variability is differing availability of morphine and its glucuronide metabolites, morphine-3-glucuronide (M3G) and morphine-6-glucuronide (M6G), in the CNS. This in turn could be caused by varying activity of drug transporters in the blood–brain barrier. The latter assumption is supported by more recent findings characterizing M and its pharmacologically active metabolite M6G as substrates of the transport protein P-glycoprotein (Pgp)  the MDR1 gene product, which in endothelial cells of the blood–brain barrier contributes to the regulation of the net transfer of drugs into the CNS.
The contributions of M3G and M6G to the opioid effects have been extensively investigated in recent years. Today, consensus has apparently been achieved that M3G is mostly devoid of activity , even if there was some evidence of excitatory behaviour in patients . With M6G the data are more complex. Recent studies have shown that M6G caused pain relief after i.v. adminstration in an ischaemic pain model , but was ineffective when given preemptively via the i.v. route for postoperative pain . In healthy volunteers M6G after short-term infusion did not produce EEG effects or clinical symptoms . On the other hand, excellent postoperative analgesia has been observed after intrathecal administration of M6G  with typical opiate side-effects. Similarly, M6G was reported to be a useful analgesic in 17/19 cases after i.v. infusion with no adverse effects . There is also evidence that M6G interacts with a specific opioid receptor .
To investigate further these issues the establishment of a pharmacokinetic-pharmacodynamic link is required, connecting effect site concentrations with drug effect measurements. An acceptable approximation to the effect site in the CNS is the analysis of CSF samples . In our study multiple samples of CSF could be obtained. However, the neurological diseases of the patients precluded the recording of effect data using visual analogue scales or similar methods . In an attempt to overcome this problem we proposed a two-step approach. The aim of the first step is the development of a population pharmacokinetic model for M, M3G and M6G in the CSF of neurosurgical patients. Such a model can then be used to predict CSF concentration-time profiles based on dose information in dose-effect studies and thus to establish the desired PK-PD link.
In this work we report the concentration-time course over 48 h of M, M3G and M6G in plasma as well as CSF after i.v. infusion of M. We have further examined interpatient variability in an attempt to identify covariates, including creatinine clearance and the MDR1 gene exon 26, C/T polymorphism. Recently this polymorphism has been found to correlate with Pgp expression and function in humans . Thus, we present for the first time a comprehensive population pharmacokinetic model as a contribution to a pharmacokinetic-pharmacodynamic modelling approach. The goal of predicting the clinical effects of morphine is another step towards future rational dose individualization.
The study was approved by the local Ethics Review Committee of the Medical Faculty of the University of Goettingen. Written informed consent was obtained for each patient from a relative. Patients were eligible for participation in the study if they were adults on the neurological or neurosurgical intensive care unit undergoing artificial respiration and analgo-sedation and fitted with a ventricular CSF catheter for therapeutic reasons. The CSF/plasma albumin-ratio (Q (mg l−1)/(g l−1)) had to be in the normal range (≤ 8.0), indicating an intact blood–brain barrier . CSF was collected over a period of 24 h 1 day before the start of the study. Eligibility for the study required a spontaneous CSF drain efflux of greater than 100 ml day−1. Analgesia had to be achieved with an opioid different from morphine (e.g. fentanyl, piritramide). Patients with meningeal or cerebral inflammation, persistent subarachnoidal haemorrhage, a pathological CSF/plasma albumin-ratio and patients without artificial respiration were excluded. Parameters for the cardiovascular, hepatic and renal functions had to be in the normal range. In addition, during the study the cerebral perfusion pressure was continuously monitored. Demographic data, concomitant medication and underlying diseases are given in Table 1. Creatinine clearance was calculated from serum creatinine according to Cockcroft & Gault .
|Patient||Gender||Age (years)||Weight (kg)||Underlying disease||Creatinine clearance (ml min −1 )||CSF/Plasma albumin ratio||Duration of intensive care prior to study (days)||Concomitant medication|
|1||F||19||55||Traumatic SAH with hydrocephalus occlusus||157||4.0||18||Clo, F, H, Lact, MP, Mer, Mid, N, P, U|
|2||F||60||70||Aneurysmatic SAH with hydrocephalus occlusus||110||4.7||9||Amo, Cis, Flu, H, Lact, Met, N|
|3||F||69||70||Aneurysmatic SAH with hydrocephalus occlusus||65||4.5||11||Acc, Cefurox, Flu, G, H, Lact, MP|
|4||M||41||70||Aneurysmatic SAH with hydrocephalus occlusus||137||7.3||19||Acc, Cis, Dim, Flu, Ket, Lact, N, Nim, O|
|5||M||63||92||Brain-stem haemorrhage with brain oedema and hydrocephalus occlusus||123||0.8||7||Acc, Cefurox, Dex, F, Flu G, H,Ins, MP, N, P, S|
|6||M||66||95||Intracerebral haemorrhage with brain oedema and hydrocephalus occlusus||98||6.6||5||Acc, Cef, Cer, Dex, Do, F, Flu, H,Ins,Lact, MP, Met, N, Ome|
|7||M||53||70||Cerebellar infarction with brain oedema and hydrocephalus occlusus||94||4.4||7||Amo, Cer, F, G, H, Ket, Lact, MP, Met, Mid, N, S|
|8||F||67||50||Intrathalamic haemorrhage with brain oedema and hydrocephalus occlusus||48||3.8||4||Dex, Do, F, H, Lac, MP, Mid, Mol, N, S, U|
|9||F||55||70||Aneurysmatic SAH and cerebellar infarction with hydrocephalus occlusus||117||4.0||12||Cer, F, Flu, G, H, Lact, MP, Met, Mid; N; S|
The patients received an i.v. infusion of morphine (0.5 mg kg−1 body weight morphine sulphate pentahydrate, Mundipharma GmbH, Limburg/Lahn, Germany) for 30 min. Regular analgesia was stopped during this interval. Blood samples (4 ml) were taken from an indwelling venous catheter before and 3, 15, 30, 60, 120, 180, 240, 360, 480, 720, 960, 1200, 1440, 1920, 2880 min after the start of morphine infusion.
CSF, obtained under strictly sterile conditions from the ventricular catheter using a 2 ml syringe, was taken before and 15, 30, 60, 120, 240, 360, 480, 720, 960, 1200, 1440, 1920, 2880 min after the start of the morphine infusion.
Blood samples were stored at 4 °C before centrifugation. To prevent possible adsorption losses of morphine, plasma and CSF were stored in polypropylene tubes at −20 °C until analysis.
The concentrations of M, M3G and M6G were determined in plasma and CSF by methods described previously [17, 18]. In brief, M was analysed by GC-MS-MS after liquid-liquid extraction with dichloromethane/2-propanol at alkaline pH followed by derivatization with pentafluoropropionic acid anhydride. For the analysis of M3G and M6G a h.p.l.c.-MS method was employed. The sample preparation was carried out using liquid-solid extraction on C2 columns. The deuterated analogues (morphine-d3, morphine-3-glucuronide-d3, morphine-6-glucuronide-d3) of the analytes were used as internal standards. The limits of quantification achieved with these methods were 0.37 ng ml−1 for M6G and 1.85 ng ml−1 for M3G with precisions of 7% and 9%, respectively, and 5.7 pg ml−1 for M with a precision of 16.9%. Concentrations are expressed as free base.
Genomic DNA was prepared from plasma samples of each patient using a QIAamp® DNA Mini Kit (Qiagen GmbH, Hilden, Germany). Genotype with respect to the C3435T polymorphism was determined by direct sequencing. Exon 26 of the MDR1 gene was amplified as previously described by Hoffmeyer et al.  and DNA sequencing was performed on an ABI 310 sequencer by using BigDye Terminator cycle sequencing reactions (Perkin-Elmer/Applied Biosystems®, Friedrichshafen, Germany).
The population pharmacokinetic model development was carried out with NONMEM version V . The morphine concentration-time data were analysed with a three-compartment model as implemented in NONMEM's PREDPP subroutine ADVAN11. All other structural models were written as systems of coupled differential equations, which were solved numerically with the subroutine ADVAN6. All estimations were performed with the first order conditional estimates method. Mass transfer constants and distribution volumes were used as variables in the differential equations and clearances were then calculated from these basic parameters. Only simple error models were used. Interindividual variability was generally implemented as an exponential term, while intraindividual variability was approximated by proportional error terms. Individual pharmacokinetic parameter estimates were obtained in a Bayesian fashion from the population estimates, their interindividual variability and the respective individual data set.
Since no information as to the mass balance in these patients was available, the fractions of the dose transformed into either metabolite and the fraction excreted unchanged could not be determined independently. Therefore the transformation ratios were taken from the literature  and set to 0.55 for M3G and 0.1 for M6G, respectively.
The concentration-time course of M in plasma was modelled by a three-compartment model with zero-order input. This choice was based on previous reports in the literature [20, 21] which describe the successful fit of both, two- and three-compartment models and suggest that the number of discernible compartments is clearly dependent on the observation time-interval. For the glucuronide metabolite M3G a two-compartment model was employed . In the case of M6G a one-compartment model was used since the data did not support a two-compartment model. For both, M3G and M6G, first-order input was assumed as given by the individual concentration-time function in the central compartment of the morphine model. The M pharmacokinetic parameters were kept fixed when the M3G or M6G parameters were estimated.
Separate CSF compartments were allocated for each compound assuming a total volume of 0.11 l for this biofluid . The respective CSF models were then linked via suitable differential equations to their plasma counterparts and the CSF kinetic parameters (maximum input and saturation constant) were estimated in further modelling steps. Again, the plasma pharmacokinetic parameters were kept fixed when the pharmacokinetic parameters for the CSF models were calculated.
Plots of measured concentrations vs individual predictions (identity plots) and of residuals vs predictions were also used in order to assess the quality of a model.
The individual predictions obtained from the final model were eventually used to estimate the relative mean prediction error for each patient  and to obtain rankings reflecting best and worst fits.
The age of the patients ranged from 19 to 69 years (median 60 years), their body weight from 55 to 95 kg (median 70 kg) and their creatinine clearances between 48 and 157 ml min−1 (mean 105 ml min−1) (Table 1). The morphine dose administered ranged from 18.8 mg to 35.8 mg expressed as free base (median 26.4 mg) (Table 2). A list of concomitantly administered drugs was recorded for each individual patient (Table 1).
|Patient||Morphine dose (mg free base)||Plasma Cmax M (ng ml−1)||Plasma Cmax M3G (ng ml−1)||Plasma Cmax M6G (ng ml−1)||CSF CmaxM (ng ml−1)||CSF CmaxM3G (ng ml−1)||CSF C max M6G (ng ml −1 )||MDR1* genotype|
Morphine concentrations rose rapidly after starting the infusion and reached their maximum between 201 and 421 ng ml−1 (Figure 1 plasma, Figure 2 CSF). Metabolite concentrations in plasma also increased rapidly, though slightly delayed, and reached peak concentrations of between 534 and 1743 ng ml−1 (M3G) and between 134 and 455 ng ml−1 (M6G). Morphine concentrations in CSF rose almost as fast as in plasma and reached maximum concentrations of between 13.3 and 31.4 ng ml−1. In contrast, the CSF concentrations of both M3G and M6G, increased and decreased more slowly with maximum values of between 4.4 and 42.2 ng ml−1 (M3G) and between 1.6 and 10.8 ng ml−1 (M6G) (Table 2). After 48 h plasma concentrations of morphine were below 0.1 ng ml−1, those of M3G were between 2 and 200 ng ml−1 and M6G concentrations between 0.5 and 50 ng ml−1. At that same time point morphine concentrations in CSF were below 1 ng ml−1 and metabolite concentrations ranged between 1 and 10 ng ml−1.
As shown in Table 2, one of the nine patients was homozygous for the wildtype allele of MDR1 (CC), five were heterozygous (CT) and three patients showed a homozygous mutant genotype (TT). The homozygous mutant genotype (TT) coincided with the highest maximum CSF concentrations of M (P < 0.001, nonparametric rank test ), which is consistent with previously published data  indicating that the TT genotype is associated with lower expression of P-glycoprotein. This would be compatible with high CSF concentrations due to a decreased efflux of M across the blood–brain barrier. However, no such effect could be seen for the glucuronides. This is consistent with the assumption that the transport of M and its glucuronides is mediated through different transporters [26, 27].
The complete pharmacokinetic model for M, M3G and M6G in plasma and CSF is shown in Figure 3. Plasma clearances for M, M3G and M6G were 1838 (1668–2001) ml min−1, 44.5 (35.1–53.9), ml min−1 and 42.1 (36.4–47.7) ml min−1, respectively. Corresponding CSF clearances were 2.31 (1.44–3.17) ml min−1, 0.45 (0.31–0.58) ml min−1 and 0.22 (0.17–0.27) ml min−1 for M, M3G and M6G, respectively. These values are population estimates, i.e. parameters for an average subject (70 kg body weight, dose 26.87 mg free base, and a creatinine clearance of 105 ml min−1, if applicable) in the population under investigation and their 95% confidence intervals. Clearances and distribution volumes of each compartment are given in Table 3.
|Central clearance||CL (ml min−1)||1838 (1668, 2001)||44.5 (35.1, 53.9) + 0.62* (CLCr-105.4)||42.1 (36.4, 47.7) + 0.66* (CLCr-105.4)|
|Central volume||V (l)||12.7 (11.1, 14.3)||7.81 (7.40, 8.21)||7.69 (7.38, 7.99)|
|Distributional clearance 1||Q1 (ml min−1)||2085 (1524, 2645)||40.6 (33.7, 47.5)|
|Peripheral volume 1||VP1 (l)||111 (78, 143)||4.0 (3.34, 4.65)|
|Distributional clearance 2||Q2 (ml min−1)||181 (135, 226)|
|Peripheral volume 2||VP2 (l)||179 (105, 252)|
|Clearance||CL (ml min−1)||2.31 (1.44, 3.17)||0.45 (0.31, 0.58)||0.22 (0.17, 0.27)|
|Maximum input||V max (µg min−1)||0.081 (0.055, 0.106)||0.011 (0.009, 0.012)||0.008 (0.003, 0.015)|
|Saturation constant||K m (ng ml)||24.0 (22, 26)||5.20 (4.80, 5.59)||39.5 (29.9, 49.1)|
The individual posthoc plasma M concentration predictions together with the population prediction are shown in Figure 1b. The concentration-time profile of M in CSF was different from the profiles in any of the peripheral kinetic compartments derived from the plasma concentrations. Whether the CSF compartment in the model was linked to the central M compartment or to the peripheral compartment did not have much effect on the fit of the model to the data. However, the best fit was seen when CSF was linked to the fast equilibrating peripheral compartment of the M plasma pharmacokinetic model. Assuming capacity limited transport of morphine into the CSF improved the fit of the model to the data greatly (χ2 test: P << 0.001). The maximum input rate for M was estimated to be 0.081 µg min−1. The CSF concentration-time course population predictions for M are shown in Figure 2b.
M3G pharmacokinetics in plasma as described by a two-compartment model are depicted in Figures 1c and 1d. The pharmacokinetics of M3G in CSF were modelled as a separate distributional space with capacity limited transfer from the central M3G compartment. The maximum input rates were 0.011 µg min−1 for M3G and 0.008 µg min−1 for M6G. Measured and predicted CSF concentrations are shown in Figures 2c and 2d. The typical predictions incorporate the influence of the different M doses and the individual elimination as determined by a creatinine clearance. The pharmacokinetics of M6G were found to be similar to those for M3G, despite the fact that a two-compartment model was not supported by the data. Measured plasma concentrations and individual predictions from the model are depicted in Figures 1e and 1f for plasma and in Figures 2e and 2f for CSF.
Overall, variability between our patients was small. In plasma the individual predictions for morphine showed little scatter around the population prediction (Figure 1b). Plasma and CSF concentrations of M3G and M6G had more variability and two patients (#3, #8) showed elevated plasma concentrations for both compounds whereas patients #1 and #9 had distinctively lower concentrations (Figures 1d and 1f). These interindividual differences could be explained by the inclusion of creatinine clearance as a covariate. This can be seen immediately from the close relationship between typical predictions and individual posthoc predictions in Figure 4. In CSF also, predicted morphine concentration-time courses were very similar for all patients (Figures 2b). The profiles of patient #3 were clearly different in terms of Cmax for M3G and M6G in this matrix and patient #8 displayed a different terminal slope. Again M3G and M6G were similar with respect to the overall shape of the profiles, with M3G concentrations consistently being slightly higher. The CSF predictions also document the variability due to different dose regimen and renal function.
Whereas extensive work has been devoted to the characterization of drug biotransformation in the past, much interest is now on the proteins mediating transcellular transport and pharmacokinetic models are required to consider these new findings. The data reported in this work have been obtained from a group of seriously ill patients, who each received a considerable number of comedications. Creatinine clearance appeared to be reduced in two patients (#3, #8) but was still within the normal range for subjects in this age group (60 ± 15 ml min−1). In addition, hepatic function was normal, cardiovascular conditions were stable and no substantial changes in blood flow were documented. The draining of 100 ml or more of CSF from an average daily production of 450 ml had no impact on the observed pharmacokinetics, because only a negligible fraction of the total amount of drug is present in the CSF at any given time. Secondly, elimination to an output compartment or return to the driving compartment cannot be distinguished on the basis of concentration observations in the CSF compartment.
To elucidate possible kinetic interactions between M and any of the concomitantly administered drugs a literature survey was undertaken. None of the agents given has been shown to alter the pharmacokinetics of morphine (Table 1). Moreover, no significant relationships between comedication and individual M, M3G and M6G clearances were observed.
The fit of the final model to the data resulted in good individual predictions in all patients for all compounds and both compartments studied (Figures 1 and 2). For M the mean prediction as described by the population estimates of the pharmacokinetic model parameters is a good representation of the group studied. Variability was particularly small in the recorded concentration-time data for M in plasma. In CSF data were more variable for M3G and M6G and could only partly be explained by differences in dose and creatinine clearance. The slow plasma elimination of M3G and M6G in patient #8 contributed to increased concentrations in CSF. In patient #3 a more rapid access of M to the CSF seemed to determine elevated concentrations, since there was a normal decline in concentration with time. Both M3G and M6G appeared rapidly in the plasma of patients after i.v. infusion due to the pronounced hepatic metabolism of M , but their transfer into the CSF was clearly delayed with respect to M, which had a short but detectable lag-time in its distribution into the CSF. It remains unclear whether M and its metabolites display differing permeability through the blood–brain barrier  or varying affinity for a specific transporter with limited capacity . The permeability of the morphine glucuronides through the blood–brain barrier is believed to be even poorer than that of M based on animal data  and this is reflected in our results. Possible mechanisms of active transport or facilitated diffusion into the CSF has been discussed [30–32], but remain to be established. The fact that CSF concentrations were better described by saturable transport seemed to exclude simple diffusion as the main mechanism of entry into the CSF.
The transporter glycoprotein (Pgp) may be involved in determining the transfer of M and M6G across the blood brain barrier. A genetic polymorphism in exon 2 of the MDR1 gene, which has functional relevance, has been described . The variability in M6G concentrations was up to 7 fold in our study population and subjects with the TT genotype, which is associated with low expression of Pgp, had the highest maximum M concentrations in CSF. In contrast, the patient with the highest maximum concentrations of M6G and M3G (#3: 10.8 ng ml−1 and 42.2 ng ml−1, respectively) had a wildtype genotype (CC). It is reasonable to assume that M3G and M6G are transported by other proteins like MRP1 or MRP2. Greater sample sizes are needed to confirm the clinical impact of M/M3G/M6G transport into the CSF by Pgp.
Numerous articles have been published in recent years on the pharmacokinetics of M and its metabolites. However, the methods used have been quite varied, in terms of route of administration, observation interval, assay technique, pharmacokinetic approach and statistical evaluation. Some consensus can be found with regard to the plasma clearance of M, which is reported to range from 533 ml min−1 to 1256 ml min−1 in patients with various diseases and from 805 ml min−1 to 2590 ml min−1 in healthy subjects.
Different study designs and methods of data analysis have been employed to determine volume in the central compartment, but have given similar results. Extreme values of 7 l for patients and 23 l for volunteers have been reported by Stanski and colleagues . Further estimates published for patients are 13 l [21, 37], 15 l  and 29 l . Our estimate of 12.6 l (Table 3) is therefore in good accordance with these findings.
Our data corroborate findings that, in addition to the central distributional space, a second fast equilibrating and a third slower equilibrating distributional space can be distinguished [20, 21]. Therefore, the steady-state volume of distribution (Vss) can be used, which is obtained as the sum of all compartmental volumes in a multicompartment model at steady-state or as the product of clearance and mean residence time. One study  gives a value of 129 l for Vss, which is considerably smaller than our value of 302 l (Table 3). Other authors have obtained intermediate values e.g. 196 l  and 201 l .
Less information is available with regard to the pharmacokinetics of M6G. This compound was originally considered to be inactive pharmacologically. When the analgesic properties of M6G had been discovered , further pharmacokinetic investigations followed. After direct administration of M6G, a two-compartment model was used by two authors [8, 32], while for the analysis of M6G concentrations occurring after M administration one compartment was sufficient .
There is very little published information on the pharmacokinetics of M3G in humans. Although several authors have reported M3G concentration data, for example in the form of plasma and CSF ratios and occasionally half-lives [20, 41, 42], there is only one study that presents a comprehensive pharmacokinetic model for M, M3G and M6G. However, parameters are only given for M . Recently, the pharmacokinetics of M3G have been investigated after iv administration of the compound .The authors report a clearance of 169 ± 48 ml min−1 and a volume of 23.1 ± 4.8 l.
The clearances of M3G and M6G derived from our data are clearly smaller than those quoted above. In our study we have used a considerably longer observation period (up to 48 h) than in previous studies and it is well known that too short a sampling span can result in an overestimation of elimination. A significant decrease of the clearance of glucuronides is seen in patients with renal impairment . However, the lower renal clearance of M metabolites cannot be explained by this factor, since the mean creatinine clearance in our patients was 105 ml min−1.
Due to the limited number of subjects in the study, the data set could not be split. Three studies from the large number of reports in the literature were used to test further the predictive ability of the model. Whereas simulations of plasma concentrations for model evaluation have been published , our comprehensive model allowed the prediction of concentrations of M, M3G and M6G in plasma and CSF after parenteral administration of morphine.
The first study was by Dahlström and coworkers  who showed that postoperative pain was relieved by patient-controlled administration (PCA) of 12 intravenous doses of M (3 mg/15 min) after a loading dose of 15 mg. Good agreement between predicted and measured concentrations was observed for plasma peak and trough concentrations after both the loading dose and repeated PCA (Table 4). A second study was chosen  because it reported concentrations of M, M3G and M6G in plasma and CSF. Here a mean daily dose of 57 mg was administered as a subcutaneous infusion (39.6 µg min−1). Drug concentrations were assessed after at least 72 h of unchanged treatment. The steady-state concentrations after subcutaneous infusion were less perfectly matched by the predictions (Table 4). Plasma concentrations in particular were clearly underestimated. In this example a mean dose taken from the literature was used for model input to generate population estimates of drug concentrations. These were then compared with experimental results. Considering the multiple error sources in this procedure, the agreement was judged reasonable.
|Loading dose 15 mg||Repeated dose 3 mg 15 min − 1|
|Predicted (ng ml − 1 )||Reported (ng ml − 1 )||Predicted (ng ml − 1 )||Reported (ng ml − 1 )|
|Case I (Dahlström et al.)|
|Predicted (ng ml − 1 )||Reported (ng ml − 1 )|
|Literature results||This work|
|Effect-site prediction||Observed in CSF 1||Model prediction|
|C max (ng ml − 1 )||t max (h)||C max (ng ml − 1 )||t max (h)||C max (ng ml − 1 )||t max [h]|
|Case III (Lötsch et al.)|
A further study by Lötsch et al. was selected because these authors reported effect-compartment modelling. Under the assumption that CSF is closely related to the effect-site concentration, predictions should be similar. A morphine dose of 23.4 mg infused over 30 min was simulated with both models, i.e. using the kinetic constants from the literature and from our data. While maximum concentrations and time-to-peak values are sufficiently close (Table 4), the time-course in the hypothetical effect compartment was different with a sharper peak and a monoexponetial decline in concentration with time (data not shown).
The number of discernible pharmacokinetic compartments depends on the duration of observation and the frequency of sampling. These are the main reasons for the diverging models proposed by others In the present work data up to 48 h were available, so that we can confidently assume that the elimination dominated phase has been achieved. The resulting three-compartment model has already been described by other groups.
The modelling of CSF concentrations has occasionally been described in the literature. Nevertheless no standard approach is presently defined on how to deal with the special requirements CSF imposes on the model, because drug does not instantaneously equilibrate in CSF, volume estimates are difficult to obtain. Thus CSF sometimes is treated like any other compartment [43, 44], while other scientists have used more complex physiological based pharmacokinetic models [12, 45].
CSF is different kinetically from plasma because it is a separate physiological distributional space. CSF has no influence on the overall pharmacokinetic behaviour, since it contains negligible amounts of drug, unless given by epidural or intrathecal administration . A pharmacokinetic volume of distribution for CSF cannot reliably be estimated, since either the amount in this compartment is unknown or after intrathecal administration the assumption of instantaneous equilibration is no longer valid. Therefore, the values between 0.1 and 0.14 l reported for this parameter are to be considered approximations only . CSF data were always added after completion of the corresponding plasma model.
The model presented here has not been applied to data after oral drug administration, since the marked first-pass metabolism of morphine or its enterohepatic recirculation are not incorporated. Experimental evidence for enterohepatic circulation in the dog has been presented by Garrett  and its role in humans has been discussed, e.g [20, 38]. However, no evidence of enterohepatic recirculation was found in our data, in the form of secondary peaks.
M, M3G and M6G have been quantified in patient plasma and CSF using the most sensitive and selective methodology presently available. A comprehensive pharmacokinetic model describing the concentration time-course of M, M3G and M6G in plasma and CSF as a function of dose and creatinine clearance has been presented and agrees well with the current knowledge base. The model can be applied to predict CSF concentrations for M, M3G and M6G. Further work will incorporate analgesic data using various pain models in an effort to define dose-concentration-effect relationships allowing optimal treatment of individual patients .
The work has been supported by the German ‘Bundesministerium für Bildung und Forschung’ (grants BMBF 01 EC 9405, 01 EC 0001 and 01 GG 9846) and the Robert Bosch Foundation, Stuttgart, Germany.
- 15Confounding factors and reference values for protein analysis in cerebrospinal fluid. In CNS barriers and modern CSF diagnostics, eds Felgenhauer K, Holzgraefe M, Prange HW. Weinheim: VCH, 1993; 305 – 312., , , ,
- 19NONMEM User Guides. University of California at San Francisco, San Francisco: NONMEM Project Group, 1992., eds.