A positron emission tomography study in healthy volunteers to estimate mGluR5 receptor occupancy of AZD2066 — Estimating occupancy in the absence of a reference region

NeuroImage 82 (2013) 160–169

Contents lists available at ScienceDirect

NeuroImage journal homepage: www.elsevier.com/locate/ynimg

A positron emission tomography study in healthy volunteers to estimate mGluR5 receptor occupancy of AZD2066 — Estimating occupancy in the absence of a reference region Matts Kågedal a, c,⁎, Zsolt Cselényi a, Svante Nyberg a, Patrick Raboisson a, Lars Ståhle a, Per Stenkrona b, Katarina Varnäs b, Christer Halldin b, Andrew C. Hooker c, Mats O. Karlsson c a b c

AstraZeneca R&D, SE-151 85 Södertälje, Sweden Karolinska institutet, Department of Clinical neuroscience, Psychiatry Section, Karolinska Hospital, S-171 76 Stockholm, Sweden Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden

a r t i c l e

i n f o

Article history: Accepted 6 May 2013 Available online 11 May 2013 Keywords: Nonlinear mixed effects modelling Positron emission tomography Receptor occupancy [11C]-ABP688 mGluR5 receptor

a b s t r a c t AZD2066 is a new chemical entity pharmacologically characterized as a selective, negative allosteric modulator of the metabotropic glutamate receptor subtype 5 (mGluR5). Antagonism of mGluR5 has been implicated in relation to various diseases such as anxiety, depression, and pain disorders. To support translation from preclinical results and previous experiences with this target in man, a positron emission tomography study was performed to estimate the relationship between AZD2066 plasma concentrations and receptor occupancy in the human brain, using the mGluR5 radioligand [11C]-ABP688. The study involved PET scans on 4 occasions in 6 healthy volunteers. The radioligand was given as a tracer dose alone and following oral treatment with different doses of AZD2066. The analysis was based on the total volume of distribution derived fro m each PET-assessment. A non-linear mixed effects model was developed where ten delineated brain regions of interest from all PET scans were included in one simultaneous fit. For comparison the analysis was also performed according to a method described previously by Lassen et al. (1995). The results of the analysis showed that the total volume of distribution decreased with increasing drug concentrations in all regions with an estimated Kipl of 1170 nM. Variability between individuals and occasions in non-displaceable volume of distribution could explain most of the variability in the total volume of distribution. The Lassen approach provided a similar estimate for Kipl, but the variability was exaggerated and difficult to interpret. © 2013 Elsevier Inc. All rights reserved.

Introduction Estimation of receptor occupancy can significantly improve the translation between compounds targeting the same receptor or from preclinical efficacy models to man (Kagedal et al., 2012). The present study of the relationship between AZD2066 concentrations in plasma and occupancy in the human brain was performed in order to improve predictions of therapeutic exposure levels and aid dose-selection for following clinical studies. AZD2066 is a new chemical entity pharmacologically characterized as a potent and selective, negative allosteric modulator (NAM, aka non-competitive antagonist) of the metabotropic glutamate receptor subtype 5 (mGluR5) (Raboisson et al., unpublished). The mGluR5 receptor is widely distributed in the central nervous system (Bear et al., ⁎ Corresponding author at: Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden. Fax: +46 8 553 288 96. E-mail address: [email protected] (M. Kågedal). 1053-8119/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.neuroimage.2013.05.006

2004; Berg et al., 2011; Brodkin et al., 2001; Chiamulera et al., 2001; Johnson et al., 2009; Spooren et al., 2000; Tatarczynska et al., 2001; Varney and Gereau, 2002; Walker et al., 2001a,b) and has received attention as a potential therapeutic target in various diseases such as anxiety, depression, and pain disorders. Using in vitro functional assays AZD2066 has been shown to inhibit human mGluR5 with mean IC50 values of 5.7 and 7.5 nM (calcium mobilization assay and phosphatidyl inositol hydrolysis assay, respectively). AZD2066 also competitively inhibited (Ki of 21 nM) [3H]MPEP, a prototypic mGluR5 NAM, from binding to human mGluR5 (Gasparini et al., 1999). AZD2066 displayed >100 to >1000-fold selectivity in vitro for mGluR5 vs. 147 other molecular targets, including all other mGluR subtypes. In vivo, AZD2066 penetrated the blood brain barrier easily and displaced the known mGluR5 radioligands [3H]methoxymethyl-MTEP (Anderson et al., 2002) and [11C]-ABP688 (Ametamey et al., 2006, 2007; Hintermann et al., 2007) in rat (Raboisson et al., Neuropharmacology submitted) and non-human primate brains (unpublished data, manuscript in preparation).

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

AZD2066 safety and efficacy have been investigated in clinical studies in healthy volunteers as well as in patients with painful diabetic neuropathy, neuropathic pain with mechanical hypersensitivity and major depressive disorder. In patients suffering from neuropathic pain with mechanical hypersensitivity, AZD2066 demonstrated a clinically and statistically significant reduction in pain and hypersensitivity (Jonzon et al., 2012), while no efficacy conclusion could be drawn from the studies in the other two indications. In healthy volunteers, AZD2066 was found to be tolerable with mainly dose-dependent CNS adverse events (AEs) observed. The development of AZD2066 was terminated since some AEs observed with AZD2066, and also previously with other mGluR5 antagonists, suggested that AZD2066 risk-benefit profile was not favourable. Detailed reports of the phase 2a data and of AZD2066 safety profile will be reported elsewhere (Karlsten et al., 2012; Jonzon et al., 2012; Ståhle et al., 2012). The study was carried out using the radioligand [ 11C]-ABP688, a highly selective radioligand for in vivo imaging of mGluR5 receptors in human using PET. The radioligand has favourable kinetics allowing for quantification of central mGluR5 in a reasonable short, 45–60 min acquisition (Ametamey et al., 2006, 2007; Treyer et al., 2007). The binding on human mGluR5 of ABP688, a chemical derivative of MPEP in which the aromatic ring is replaced by a functionalized cyclohexenone moiety, is also known to be fully displaceable by MPEP (Hintermann et al., 2007), indicating that ABP688, MPEP and AZD2066 share a common binding site on mGluR5. In humans, no reference region known to be devoid of mGluR5 receptors exists (Patel et al., 2007; Treyer et al., 2007), making estimation of occupancy more difficult. It is theoretically possible to derive the binding potential (BPND) as an index of specific binding based on the estimated rate parameters in each individual PET scan (Innis et al., 2007). In the absence of a reference region, this method however suffers from identifiability problems. Obviously, when occupancy becomes high, identifying rate-parameters relating to specific binding becomes futile. While the individual rate-constants often are highly correlated, a robust estimate of the total volume of distribution (VT) is however often possible to derive (Cunningham et al., 2004). Lassen et al. (1995) proposed an approach allowing estimation of occupancy, based on VT, leveraging the difference in receptor density between regions. The Lassen approach was subsequently further developed (Cunningham et al., 2010). In a situation when information is sparse based on each individual subject, nonlinear mixed effects modelling (NLME) has been shown to be useful (Aarons, 1996). By integrating all data into one simultaneous fit, a model that is un-identifiable based on individual subjects may become identifiable. This method may thus improve the ability to separate specific from nonspecific brain uptake in the analysis and hence allow estimation of occupancy. Increasingly the analysis of PET-data is being performed by population methods (Liefaard et al., 2005; Lim et al., 2007; Syvanen et al., 2011; Zamuner et al., 2012). Recently it was proposed that simultaneous modelling of radioligand kinetics of all PET scans in two regions of interest (ROIs) allowed quantification of the relationship between exposure and occupancy applying NLME methodology (Kagedal et al., 2012). This approach made use of all the data in an efficient way, since the radioligand kinetics as well as the difference in specific uptake between regions informed the model in one simultaneous fit. The drawback with this approach was that only two regions of interest were included and that it had rather long runtimes and more complex model-building process. In the present analysis VT for ten different regions of interest was determined in a first step for each of the PET scans. The relationship between plasma drug concentration during the PET scan and occupancy was subsequently estimated using NLME modelling where VT from 10 brain regions in six subjects were included in one simultaneous fit. A similar approach has been proposed previously by Berges et al. (2008). For comparison, the analysis was also performed by the method proposed by Lassen et al. (1995).

161

The aim of the present work was to quantify the relationship between AZD2066 plasma concentrations and displacement of [11C]-ABP688 from mGluR5 binding-sites in the CNS and to compare the results obtained with NLME modelling to the Lassen approach. Methods Overall study design This was an open-label, non-randomised, single-centre, exploratory PET study in 6 healthy male volunteers aged between 23 and 40 years. The study comprised 2 panels, Panel 1 and Panel 2, with 3 subjects each. An interim analysis of preliminary data from Panel 1 including PET data and pharmacokinetic data was performed for dose selection for Panel 2. Four PET scans were performed in each subject. The first PET scan was performed in the absence of AZD2066 (baseline PET). In the subsequent occasions the PET scan was preceded by administration of an oral solution of AZD2066. Venous blood samples for determination of AZD2066 concentration in plasma were collected regularly before, during and after the PET scan. In Panel 1, 3 single oral doses (3.5 mg, 6.9 mg and 13.5 mg) of AZD2066 were administered at separate occasions with a dosing interval of at least one week. In Panel 2, the subjects received 13.5 mg, 6.9 mg and 0 mg (a second baseline scan). These doses were selected to provide informative data for the estimation of the concentration–occupancy relationship. The repeated baseline measurement was performed since it was judged informative for the estimation of the concentration–occupancy relationship and in addition would provide some (albeit limited) test–retest data. Study drugs The oral solution of AZD2066 was prepared by the Karolinska Hospital Pharmacy. [11C]-ABP688 was manufactured extempore at the PET centre at Karolinska University Hospital, Solna. At each PET-scan, approximately 300 MBq of [11C]-ABP688 in an aqueous solution was administered as an intravenous bolus injection. The total radioactivity amount for each volunteer including all four PET-scans was approximately 1200 MBq. The injected radioligand had high specific radioactivity and the total mass administrated was less than 0.3 μg per injection. Subcutaneous lidocaine, 1 to 2 mL, was given as a local anaesthesia prior to the insertion of the arterial cannula. PET related measurements Prior to PET scans, two anatomical 3D MRI examinations were made in one session. The first examination was T2-weighted and was used for clinical evaluation and exclusion of pathology. The second examination was T1-weighted and was used for delineation of anatomically defined ROIs. MR images were acquired using a 1.5 T General Electric Signa Unit (Milwaukee, WI, USA). Imaging parameters included a repetition time of 23 ms, echo time of 4 ms, matrix of 256 × 192 × 156, and voxel size of 1.02 × 1.02 × 1.0 mm. The PET system was a Siemens Medical Solutions High Resolution Research Tomograph (HRRT) which follows radioactivity in 207 sections of the brain with three dimensional acquisition (Varrone et al., 2009). The spatial resolution in the reconstructed image is on average 2.3 mm full-width half-maximum (FWHM) in all directions. In each PET scan the subject was placed supine and the head was fixed to the positron camera by the use of an individualized plaster helmet as described previously (Bergström et al., 1981). Radioactivity in the brain was measured for 63 min, following radioligand injection. The duration of the first 9 frames was 10 s each, followed by 2 × 15 s, 3 × 20 s, 4 × 30 s, 4 × 60 s, 4 × 180 s, and finally 7 × 360 s. A 63 minute acquisition was judged sufficient for obtaining stable results using [ 11C]-ABP688 according to prior results (Ametamey et al., 2007).

162

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

Arterial blood samples were obtained from the arteria radialis. During the first 5 min after intravenous bolus injection of [11C]-ABP688 an automatic blood sampling system (ABSS) was used (Eriksson et al., 1988). Discrete manual arterial samples were drawn at scheduled time points for measurement of both radioactivity and radioligand metabolites, 6 mL (4, 10, 20, 30, 40, 50 and 60 min) and for measurement of radioactivity only, 2 mL (3, 6.5, 7.5, 8.5, 13.5, 16.5, 24 and 36 min). The samples were collected and analysed using standard methodology in order to derive a metabolite corrected plasma input function, as described before (Farde et al., 1989; Halldin et al., 1995). Blood samples (2 mL) were collected at frequent intervals for 47 h after each dose of AZD2066 to determine the pharmacokinetic profile (Fig. 2). It was ensured that samples were taken at the beginning, midpoint and end of the PET-scan. Plasma AZD2066 concentrations were determined by using liquid chromatography and mass spectrometry (quantification limit: 1.00 nmol/L). The average plasma concentration during the PET-scan was used in the subsequent estimation of occupancy. PET data-analysis Image pre-processing consisted of spatial reorientation and alignment of MR images to the anterior–posterior commissural plane, resampling, and cropping in order to generate a 256 × 256 × 144 mm image array with 1 mm3 voxels. The T1-weighted MR images were coregistered to each PET image of the subject and the coregistration parameters were saved using SPM5 software (SPM5; Wellcome Department of Imaging Neuroscience, University College, London, UK). The delineations of anatomical ROIs were made manually on the spatially reoriented MRI images in the three orthogonal projections using software developed for the Human Brain Atlas (Roland et al., 1994). The ROIs were resliced in the space of the PET image using the MR-to-PET coregistration parameters and radioactivity concentration in each ROI (CROI) was calculated for each sequential PET time-frame, corrected for 11C decay, and plotted versus time. The ROIs were pooled for each bilateral region. The results were stored as time-activity (TAC) data for brain tissue, given as radioactive concentration (nCi/mL). The ROIs included known grey matter regions

Baseline

3.5 mg

Fig. 2. Geometric mean plasma concentrations (nmol/L) of total AZD2066 versus planned time. Linear and log linear plot. Shaded area indicates time of PET scan.

of high mGluR5 density (anterior cingulate cortex, ACC; prefrontal cortex, PFC; temporal cortex, TC; caudate nucleus, CAU; putamen, PUT; ventral striatum, VST; hippocampus, HIP; and thalamus, THA) as well as the 2 regions with the lowest density (cerebellum, CER; and pons, PONS). The VT was determined using Logan's linear graphical approach based on the radioactivity concentrations over time for the delineated brain regions and the arterial plasma input function (Ametamey et al., 2006, 2007; Logan et al., 1990).

6.9 mg

13.5 mg

3.5 3

BPND

2.5

2

1.5 1 0.5 0

Fig. 1. BPND parametric images showing the effect of AZD2066 on binding of [11C]-ABP688. Data from subject 101. BPND images in colour scale are overlaid on anatomical MR images in greyscale.

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

Relationship between AZD2066 plasma concentrations and occupancy The analysis was based on the VT derived from each PET-assessment. The VT, sometimes referred to as the brain-plasma partition coefficient, correspond to the ratio of radioactivity concentration in the brain and concentration of unchanged radioligand in plasma at equilibrium. The VT corresponds to the sum of the non-displaceable volume of distribution (VND) and the specific, displaceable volume of distribution VS. The binding potential (BPND) corresponds to the ratio VS/VND and is frequently used as an index of receptor binding when a reference region representing VND is available (Innis et al., 2007). The analysis included data from 10 delineated brain regions of interest as well as the average plasma concentration of AZD2066 assessed during the PET experiment (Cp). A model was developed where this data was included in one simultaneous fit. In addition, for comparison the analysis was performed according to the Lassen approach (Lassen et al., 1995) described below. Model implementation The population analysis programme NONMEM 7, a widely used nonlinear regression software package in population pharmacokinetic–pharmacodynamic data analysis and the first order conditional estimation method (FOCE), was employed (Beal et al., 1989–2006). The post-processor Xpose implemented in R and S-plus was used for model diagnostic purposes (Jonsson and Karlsson, 1999). Fixed effects. The relationship between AZD2066 exposure and binding potential for ROI i (BPND,i) defined in Eq. (1), was assessed based on the saturation function shown in Eq. (2). BPND;i ¼ VS;i =VND
BPND;i ¼ BPBL;i ⋅ 1−

ð1Þ Cp Cp þ Kipl

 ð2Þ

163

The following proportional residual error model was used:   VTobsijk ¼ VTijk 1 þ εijk :

ð6Þ

In which VTijk is the model predicted VT for region i, subject j and occasion k, and VTobsijk is the corresponding observed VT. The deviations of the model from the observations are represented by εijk. The values for εijk are assumed to be normally distributed with a mean of zero and a variance σ2. In the final model, inter individual and inter-occasion variability in VND was included. The observed VT for region i, subject j and occasion k, including fixed and random effects can then be described by: VTobsijk ¼

BPBL;i ⋅ 1−

Cpj;k Cpj;k þ Kipl

!

! þ1

   VNDj;k ⋅ 1 þ εi;j;k :

ð7Þ

Model evaluation. Models were evaluated by NONMEM objective function values (OFV), goodness of fit plots, and standard errors of parameter estimates. The 95% confidence interval of the Kipl parameter was evaluated by means of likelihood profiling. Since the analysis is based on few subjects, the difference in objective function value corresponding to the 5% level needs to be calibrated. The OFV increase defining the confidence interval was therefore determined using simulation and re-estimation similarly as described previously (Vong et al., 2012). This assessment was based on simulation of 1000 studies from the final model followed by re-estimation with and without Kipl being fixed to the estimate of the final model. The OFV difference between the estimation models was calculated for each of the simulated studies. The difference in OFV corresponding to the 5% significance level was subsequently determined to 5.9. The likelihood profiling was thereafter performed where the 95% confidence interval for Kipl was defined by an increase in the objective function value of 5.9.

Where Kipl is the plasma concentration corresponding to 50% occupancy, Cp is the average plasma concentration of AZD2066 during the PET-experiment and BPBL,i is the baseline BPND for region i. To remove correlation between parameters, the model was parameterized such that BPBL for caudate (BPBL,CAU) was estimated as a parameter. For other regions, the BPBL relative to caudate (BPRELi) was estimated (Eq. (3)).

Evaluation of the cerebellum as reference region The possibility of using the cerebellum as a reference region has been considered for [ 11C]-ABP688 (DeLorenzo et al., 2011a,b; Treyer et al., 2007). When BPND in the reference region (BPr) is non-zero, the apparent occupancy (Oapp) will be lower than the true occupancy as described previously (Olsson et al., 2004):

BPBL;i ¼ BPBL;CAU  BPRELi

Oapp ¼

ð3Þ

The VND was assumed to be the same for all regions and the model predicted VT for ROI i was then:   VTi ¼ BPND;i þ 1  VND :

ð4Þ

Occupancy : 1 þ ð1−OccupancyÞ  BPr

Based on this equation and the relationship between drug concentration in plasma and occupancy (Eq. (10)) it can be shown that the apparent Ki is dependent on BPr according to the following based on the analysis of one region of interest and one reference region: Ki apparent ¼ Kið1 þ BPrÞ:

Random effects. The modelling process included estimation of inter individual variability (IIV) of parameters in a first step. For parameters where IIV was significant, inter-occasion variability (IOV) was tested. After inclusion of IOV (if significant) the importance of IIV was re-evaluated on that parameter. The IIV and IOV were tested on the parameters VND, BPBL and Kipl with lognormal variance models according to the following. ηj þηjκ

Pjκ ¼ θ⋅e

ð5Þ

where Pjκ is the parameter value in the model for subject j at occasion k, θ is the typical parameter value in the population and ηs are zero-mean, normally distributed variables with standard deviation ωP,IIV and ωP,IOV which are estimated as part of the population model.

ð8Þ

ð9Þ

The expected bias on Ki can thus be predicted from the BPND in the reference region. The effect of fixing the BPND in the cerebellum to zero was also evaluated in the analysis. Lassen method Occupancy was also estimated by the Lassen approach. With this method the difference between VT at the baseline PET (VTBL) and the VT obtained after pretreatment (VTPT) of the 10 regions included in the analysis is plotted versus VTBL. Linear regression is then performed to estimate the slope and intercept for all pre-treatment PET experiments. Occupancy corresponds to the slope and the intercept on the X-axis corresponds to the VND. This analysis is based on the assumption that VND and occupancy is the same in all regions

164

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

included in the analysis. The Kipl was subsequently estimated by fitting the following hyperbola to the derived occupancy values. Occupancy% ¼

Cp  100 Cp þ Kipl

ð10Þ

Simulation study In order to evaluate whether the results of the Lassen approach was consistent with the results of the NLME model, a simulation study was performed. Based on the final NLME model, 300 studies of identical design as the executed study was simulated. The occupancy was subsequently estimated in each of the simulated studies with the Lassen approach as well as with the NLME model. Parametric image generation VT parametric images based on Logan were obtained using waveletaided parametric imaging (Cselényi et al., 2006). From these BPND parametric images were calculated for visual illustration using the estimated VND for each occasion in the nonlinear mixed effects model. Results AZD2066 pharmacokinetics in plasma The plasma concentration during the PET scan was relatively constant and proportional to the dose of AZD2066 administered, Fig. 2. The time of maximum concentration (Tmax) was in the range 20 min to 2 h and always occurred prior to the PET scan performed at 3 to 4 h post dose. The half-life (t1/2) in the terminal phase was approximately 20 h and oral plasma clearance (CL/F) was around 1.3 L/h. The maximum plasma concentration geometric mean (CV%) was 325 (25%), 651 (18%) and 1267 (11%) nmol/L for the doses 3.5, 6.9 and 13.5, respectively. The corresponding concentration during the PET scan were 242 (15%), 459 (13%) and 869 (15%) nmol/L. Radioligand concentration in blood The radioligand [ 11C]-ABP688 has fast initial metabolism. Only 50% of the parent compound was unchanged in plasma within 4–5 min after injection. The parent fractions reached a plateau after approximately 20 min leaving 10–30% of the parent compound intact. Plots of the mean time activity curves for each dose group and the baseline occasions for the whole blood and metabolite corrected plasma suggest that AZD2066 has no significant effect on the blood concentrations of [ 11C]-ABP688 (data not shown). After correction for metabolite, it is clear that equilibration of the radioligand between red blood cells and plasma is fast and appears to yield uniform concentrations in both compartments (data not shown). Regional brain radioactivity and estimation of VT Administration of AZD2066 diminished the total uptake of the radioligand in a dose-dependent, gradual, manner. This is evident based on plots of metabolite corrected plasma TAC (Fig. 3). The Logan method provided stable, consistent estimates of VT, which are in agreement with regional values reported previously (Hintermann et al., 2007; Patel et al., 2007; Treyer et al., 2007). Relationship between exposure and occupancy Non-linear mixed effects modelling The final model, based on VT values included fixed effects corresponding to the binding potential for ROI, VND and Kipl. The random

effects included inter individual and inter-occasion variability in VND and a proportional residual error. The parameter estimates are shown in Table 1. Briefly, the VND was estimated to be 1.6 (RSE = 23%). The BPND for CAU was estimated to be 2.46 (RSE = 23%). HIP, PUT and VST also had high uptake with a BPND relative to CAU close to 1. The lowest BPND was seen in CER and PONS with a BPND of around 0.75 i.e. 30% of that in CAU. The Kipl i.e. the plasma concentration corresponding to 50% occupancy, was estimated to be 1170 nM (95% CI = 874– 1620). In a sensitivity analysis, the model was estimated including only one base-line scan per volunteer. Two estimations including either first or the second base-line scan from volunteers who had two of them were performed. The estimates of Kipl based on these estimations were 1060 nM and 1400 nM respectively, which is fairly similar to the estimate obtained based on all data. The estimated specific binding in the cerebellum relative to that in the caudate is slightly higher compared to the value reported in literature of 20% (Patel et al., 2007). When fixing the ratio to 20% based on literature data, and re-fitting the model, the estimated Kipl value was only slightly higher at 1350 nM. However when fixing the BPND in the pons and cerebellum to zero, the Kipl estimate increased to 1900, a 60% increase relative to the final model. This is fairly close to the expected increase of 75% based on the estimated BPND in the cerebellum and pons of 0.75 (Eq. (9)). Goodness of fit plots suggest that the model can describe the saturation of uptake and the variability well (Figs. 4, 5 and 6). Variability in VND between individuals and occasions could explain most of the variability in VT for all regions with a proportional residual error (SD) of only around 0.05. When first introducing IIV on VND, BPBL and Kipl one by one, the estimated variability (SD) was 0.24, 0.29 and 5.7 respectively. The largest drop in OFV was seen with IIV on VND which resulted in a 16 and a 175 point larger drop in OFV compared to IIV on BPND and Kipl respectively. After the introduction of IOV on VND, the OFV dropped by a further 274 points. Based on the final model including IIV and IOV on the VND parameter, the IIV on BPND and KI were reassessed with resulting drops in OFV of 0 and 2 respectively. The estimated IIV (SD) tended to zero for BPBL and was estimated to be 0.29 for Kipl with a rather large standard error of 55%. All other parameters, including random effects remained essentially the same. The goodness of fit expressed in terms of BPND (Fig. 5) illustrates that the residual error scatters around the population prediction displaying that no apparent unexplained inter individual variability in BPND is present. Lassen The analysis was also performed by the Lassen approach for comparison. The occupancy and VND was estimated for each PET scan based on the same regions that were included in the analysis based on the nonlinear mixed effects modelling. In three of the subjects two PET scans were performed without pre-treatment with AZD2066. For these subjects the occupancy was calculated based on the mean VT of the two visits without pre-treatment. Examples of plots are shown in Fig. 7. While variability in occupancy is large, the analysis based on the Lassen approach support the conclusion that AZD2066 saturates the uptake of [11C]-ABP688 with all pre-treatment PET scans having an estimated occupancy larger than zero. Based on the individual occupancy values, the Kipl was estimated to be 1165 nM i.e. similar to the estimated result of the NLME analysis (Fig. 8). The median VND was 1.4 with individual values varying between −1.2 and 4.2 (Fig. 9). Simulation study Simulation of 300 studies based on the final NLME model with a Kipl of 1169 was performed. Estimation of Kipl based on the Lassen approach for each of the simulated studies resulted in a mean Kipl of 1105 (SD = 166). The corresponding mean for the NLME model

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

165

Fig. 3. [11C]AZ12713580 tissue radioactivity concentration normalized for injected radioactivity over time by AZD2066 dose for CAU and CER subject 101.

was 1207 (SD = 170). Both methods thus resulted in fairly unbiased estimates of Kipl. The rather large variability in occupancy estimates based on the Lassen approach on simulated data was consistent with the result based on the observed data (Fig. 8).

general of mild to moderate intensity and resolved spontaneously and completely. At the two highest doses one instance each was reported for illusion, dizziness, feeling drunk, asthenia and disturbance in attention and there were two instances of somnolence.

Parametric images Fig. 1 demonstrates the effect of AZD2066 on uptake and binding of [ 11C]-ABP688 in orthogonal cross sections through the BPND parametric images of subject 101 (baseline, 3.5 mg, 6.9 mg and 13.5 mg doses) visually confirming the dose dependent, homogenous reduction of [ 11C]-ABP688 binding across receptor-rich grey matter areas.

Discussion and conclusions

Adverse events There were few and mild AEs reported with 2, 2, 3 and 5 subjects reporting AEs on the baseline PET (n = 9 occasions), 3.5 mg (n = 3), 6.9 mg (n = 6) and 13.5 mg (n = 6) respectively. The AEs were in

Table 1 PK–PD parameter estimates. Parameter

Parameter RSE%1 BPBL,i VTBL,i Comment estimate

BPBL,CAU2 BPRELi2

2.46

ACC CER HIP PFC PONS PUT TC THA VST

0.88 0.31 1.05 0.771 0.301 0.961 0.861 0.663 0.991 1.61 1170

VND Kipl (nM)

Proportional error (CV) 5% 9% IOV in VND (CV) 23% IIV in VND (CV)

23.1%

3.3% 18.3% 3.4% 1.8% 18.4% 1.7% 2.7% 4.1% 1.5% 14.1% 20.4%

7.7% 12.6% 26.6%

2.46

2.15 0.76 2.58 1.90 0.74 2.36 2.12 1.63 2.44

5.57

5.08 2.84 5.77 4.66 2.80 5.42 5.02 4.24 5.53

Binding potential relative to CAU 88% 31% 105% 77% 30% 96% 86% 66% 99% Plasma concentration corresponding to 50% occupancy Shrinkage = 13% Shrinkage = −8%

1 The relative standard errors for omega and sigma are reported on the approximate standard deviation scale (SE/variance estimate)/2. 2 Regions: CAU, caudate nucleus; ACC, anterior cingulate cortex; CER, cerebellum; HIP, hippocampus; PFC, prefrontal cortex; PUT, putamen; TC, temporal cortex; THA, thalamus; VST, ventral striatum.

This is according to our knowledge the first time that occupancy in human subjects is reported based on displacement of the radioligand [ 11C]-ABP688. Since no reference region known to be devoid of mGluR5 exists, identification of the specific uptake and hence occupancy becomes more difficult. The current analysis was performed using a non-linear mixed effects modelling (NLME) approach where all VT data were included in one simultaneous fit to make the analysis as informative as possible. For comparison the analysis was also performed by the method proposed by Lassen et al. (1995). The results from the present study demonstrate that AZD2066 can displace [ 11C]-ABP688 from mGluR5 binding sites in the human brain. The estimated Kipl was around 1200 nM suggesting that approximately 50% occupancy was achieved at Cmax with the highest dose (13.5 mg) in the study. It is suggested that most of the variability in VT could be explained as the variability between subjects and between occasions in VND. Inclusion of IIV in BPND or Kipl did not further improve the description of data. Due to the low number of subjects, difficulty of characterizing multiple sources of variability may not be surprising, in particular given the rather low occupancies achieved (ca 15–50%). It is difficult to know whether this situation is unusual for a radioligand or not, since VND rarely is estimated for radioligands with a reference region and for radioligands that lacks a reference region, attempts to distinguish between variability in nonspecific and specific uptake is not usually made. The variability in VND can be related to variability in plasma protein binding or variability in nonspecific binding to brain tissue. Also uncertainties in the generation of the input function can contribute to this variability. If ABP688 VT is to be used as a marker of MgluR5 receptor density, the variability in VND also should be considered as a potential explanation for differences between individuals or populations. Analyses, including both high and low uptake regions can help in this assessment. While the Lassen approach provided a similar estimate of Kipl, a simultaneous NLME analysis based on data from all PET scans allows a better understanding of the sources of variability. Considering for example Fig. 8 where occupancy based on the Lassen approach is plotted versus concentration one may easily be misled to believe

166

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

Fig. 4. Population predicted VT (black) and VT observed (grey) versus AZD2066 plasma concentration.

Fig. 5. Population predicted BPND (black) and BPND based on observed VT and individual model prediction of VND for each occasion (grey) plotted versus plasma concentration. (BPND calculated as VT/VND − 1).

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

167

Fig. 6. Basic goodness-of-fit plots.

that the variability seen in this figure represents variability in occupancy when in a more thorough analysis using NLME no inter-individual variability in occupancy is identifiable. Simulation from the NLME model supports the notion that the observed variability in occupancy based on the Lassen approach may well be a result of variability in VND. The individual empirical Bayes estimates of VND from the NLME displayed low shrinkage suggesting that they may be used to represent the underlying VND in each PET scan (Savic and Karlsson, 2009). This

allowed generation of parametric images that illustrate the distribution of binding potential in the brain in Fig. 1. The estimated VND with the Lassen method will be more sensitive to noise in the data and was clearly more variable as compared to the NLME results (Fig. 9). The specific uptake in the cerebellum and its potential use as a reference region have been discussed in the literature (Ametamey et al., 2007; DeLorenzo et al., 2011a,b; Treyer et al., 2007). In the current analysis of specific uptake was also estimated to be present in the

Fig. 7. Lassen plots for subject 101 based on the base-line PET and three pre-treatment PET-scans after different doses of AZD2066. Circles correspond to estimates in individual brain regions and the line is a result of linear regression where the slope is the estimated occupancy and the intercept on the x-axis is the VND.

168

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

Fig. 8. Occupancy derived using the Lassen approach. The analysis was based on observed (left) and simulated (right) VT data. Circles represents the occupancy estimates from each pre-treatment PET, the black line is the regression curve based on these estimates. The grey line is the true occupancy used in the simulation.

cerebellum with a BPND of 0.75 corresponding to around 30% of the specific uptake in the caudate. This is in line with results in vitro (Patel et al., 2007). Thus if the cerebellum is used as a reference region, ignoring the specific uptake, the Kipl value is predicted to be overestimated by 75% (see Eq. (9)). Treyer et al. (2007) report that a two tissue compartment model is superior to a one tissue compartment model for 11C-ABP688, suggesting that there is a kinetically distinguishable compartment that could be related to specific binding. It has previously been proposed that information on specific uptake present in the radioligand kinetics can be useful in the estimation of occupancy and Kipl (Kagedal et al., 2012). The present analysis being based on VT ignored this information. While a model, based directly on radioligand concentrations, can be more informative, it also requires the additional

assumption that the more slowly equilibrating binding is related to binding to the receptor. In addition it is more challenging to include many regions with this approach. Two important assumptions in the present analysis are that the occupancy and VND are the same in the different brain regions. Graphical exploration and statistical testing in the present analysis support that these assumptions are valid. Despite the lack of a reference region, the relationship between AZD2066 plasma concentration and MgluR5 occupancy could be estimated, providing a basis for translation between the preclinical species and man and between different mGluR5 receptor antagonists. The present NLME method provides a way to incorporate many brain regions of interest in a relatively simple model that can provide a better understanding of sources of variability compared to the Lassen approach.

Fig. 9. VND estimates based on the Lassen approach and the NLME model plotted versus Cp.

M. Kågedal et al. / NeuroImage 82 (2013) 160–169

Conflict of interest statement There are no conflicts of interest.

References Aarons, L., 1996. Population approaches/sparse data analysis for human variability in kinetics and dynamics. Environ. Toxicol. Pharmacol. 2, 197–199. Ametamey, S.M., Kessler, L.J., Honer, M., Wyss, M.T., Buck, A., Hintermann, S., Auberson, Y.P., Gasparini, F., Schubiger, P.A., 2006. Radiosynthesis and preclinical evaluation of 11C-ABP688 as a probe for imaging the metabotropic glutamate receptor subtype 5. J. Nucl. Med. 47, 698–705. Ametamey, S.M., Treyer, V., Streffer, J., Wyss, M.T., Schmidt, M., Blagoev, M., Hintermann, S., Auberson, Y., Gasparini, F., Fisher, U.C., Buck, Alfred, 2007. Human PET studies of metabotropic glutamate receptor subtype 5 with 11C-ABP688. J. Nucl. Med. 48, 247–252. Anderson, J.J., Rao, S.P., Rowe, B., Giracello, D.R., Holtz, G., Chapman, D.F., Tehrani, L., Bradbury, M.J., Cosford, N.D., Varney, M.A., 2002. [3H]methoxymethyl-3-[(2-methyl-1,3-thiazol-4-yl)ethynyl]pyridine binding to metabotropic glutamate receptor subtype 5 in rodent brain: In vitro and in vivo characterization. J. Pharmacol. Exp. Ther. 303, 1044–1051. Beal, S., Sheiner, L.B., Boeckmann, A. (Eds.), 1989–2006. NONMEM User's Guide. Icon Development Solutions, Ellicott City, Maryland, USA. Bear, M.F., Huber, K.M., Warren, S.T., 2004. The mGluR theory of fragile X mental retardation. Trends Neurosci. 27 (7), 370–377 (Jul). Berg, D., Godau, J., Trenkwalder, C., Eggert, K., Csoti, I., Storch, A., Huber, H., Morelli-Canelo, M., Stamelou, M., Ries, V., Wolz, M., Schneider, C., Di Paolo, T., Gasparini, F., Hariry, S., Vandemeulebroecke, M., Abi-Saab, W., Cooke, K., Johns, D., Gomez-Mancilla, B., 2011. AFQ056 treatment of levodopa-induced dyskinesias: results of 2 randomized controlled trials. Mov. Disord. 26 (7), 1243–1250. Berges, A., Zamuner, S., Laurijssens, B., Gunn, R., Cunningham, V., Chen, C., 2008. Using VT (Total Volume of Distribution from PET) in Estimating the PK-Receptor Occupancy Relationship in the Absence of Reference Regions. 17 (Abstr 1398 [www.page-meeting.org/? abstract=1398]). Bergström, M., Boethius, J., Eriksson, L., Greitz, T., Ribbe, T., Widén, L., 1981. Head fixation device for reproducible position alignment in transmission CT and positron emission tomography. J. Comput. Assist. Tomogr. 5, 136–141. Brodkin, J., Busse, C., Sukoff, S.J., Varney, M.A., 2001. Anxiolytic-like activity of the mGluR5 antagonist MPEP: a comparison with diazepam and buspirone. Pharmacol. Biochem. Behav. 73 (2), 359–366. Chiamulera, C., Epping-Jordan, M.P., Zocchi, A., Marcon, C., Cottiny, C., Tacconi, S., Corsi, M., Orzi, F., Conquet, F., 2001. Reinforcing and locomotor stimulant effects of cocaine are absent in mGluR5 null mutant mice. Nat. Neurosci. 4 (9), 873–874 (Sep). Cselényi, Z., Olsson, H., Halldin, C., Gulyás, B., Farde, L., 2006. A comparison of recent parametric neuroreceptor mapping approaches based on measurements with the high affinity PET radioligands [11C]FLB 457 and [11C]WAY 100635. NeuroImage 32, 1690–1708. Cunningham, V.J., Rabiner, E.A., Matthews, J., Gunn, R., Zamuner, S., Gee, A., 2004. Kinetic analysis of neuroreceptor binding using PET. Int. Congr. Ser. 1265, 12–24. Cunningham, V.J., Rabiner, E.A., Slifstein, M., Laruelle, M., Gunn, R., 2010. Measuring drug occupancy in the absence of a reference region: the Lassen plot re-visited. J. Cereb. Blood Flow Metab. 30 (1), 46–50 (1). DeLorenzo, C., Mathew, S.M., Brennan, K., Mann, J., Parsey, R., 2011. In vivo positron emission tomography imaging with [11C] ABP688: binding variability and specificity for the metabotropic glutamate receptor subtype 5 in baboons. Eur. J. Nucl. Med. Mol. Imaging 38 (6), 1083–1094. DeLorenzo, C., Kumar, J.S., Mann, J.J., Parsey, Ramin V., 2011. In vivo variation in metabotropic glutamate receptor subtype 5 binding using positron emission tomography and [11C] ABP688. J. Cereb. Blood Flow Metab. 31, 2169–2180. Eriksson, L., Holte, S., Bohm, C., Kesselberg, M., Hovander, B., 1988. Automated blood sampling systems for positron emission tomography. IEEE Trans. Nucl. Sci. 35, 703–707. Farde, L., Eriksson, L., Blomquist, G., Halldin, C., 1989. Kinetic analysis of central [11C] raclopride binding to D2-dopamine receptors studied by PET — a comparison to the equilibrium analysis. J. Cereb. Blood Flow Metab. 9 (5), 696–708. Gasparini, F., Lingenhohl, K., Stoehr, N., et al., 1999. 2-methyl-6-(phenylethynyl)-pyridine (MPEP), a potent, selective and systemically active mGlu5 receptor antagonist. Neuropharmacology 38, 1493–1503. Halldin, C., Farde, L., Högberg, T., Mohell, N., Hall, H., Suhara, T., Karlsson, P., Nakashima, Y., Swahn, C.G., 1995. Carbon-11-FLB 457: a radioligand for extrastriatal D2 dopamine receptors. J. Nucl. Med. 36 (7), 1275–1281 (Jul). Hintermann, S., Vranesic, I., Allgeier, H., Brülisauer, A., Hoyer, D., Lemaire, M., Moenius, T., Urwyler, S., Whitebread, S., Gasparini, F., Auberson, Y.P., 2007. ABP688, a novel selective and high affinity ligand for the labeling of mGlu5 receptors: Identification, in vitro pharmacology, pharmacokinetic and biodistribution studies. Bioorg. Med. Chem. 15, 903–914. Innis, R.B., Cunningham, V.J., Delforge, J., Fujita, M., Gjedde, A., Gunn, R.N., Holden, J., Houle, S., Huang, S.C., Ichise, M., Iida, H., Ito, H., Kimura, Y., Koeppe, R.A., Knudsen, G.M., Knuuti, J., Lammertsma, A.A., Laruelle, M., Logan, J., Maguire, R.P., Mintun, M.A., Morris, E.D., Parsey, R., Price, J.C., Slifstein, M., Sossi, V., Suhara, T., Votaw, J.R., Wong, D.F., Carson, R.E., 2007. Consensus nomenclature for in vivo imaging of reversibly binding radioligands. J. Cereb. Blood Flow Metab. 27, 1533–1539. Johnson, K.A., Conn, P.J., Niswender, C.M., 2009. Glutamate receptors as therapeutic targets for Parkinson's disease. CNS Neurol. Disord. Drug Targets 8 (6), 475–491 (Dec).

169

Jonsson, E.N., Karlsson, M.O., 1999. Xpose — an S-PLUS based population pharmacokinetic/ pharmacodynamic model building aid for NONMEM. Comput. Methods Prog. Biomed. 58, 51–64. Jonzon, B., Butler, S., Karlsten, R., Malamut, R., Ståhle, L., Huizar, K., Stacey, B.R., 2012. Efficacy and safety of the mGluR5 antagonist AZD2066 in peripheral neuropathic pain patients with mechanical hypersensitivity (NPMH): results of a Phase IIa randomised, doubleblind, placebo-controlled study. Abstract of the 14thWorld Congress on Pain, Milan PT427. Kågedal, M., Cselényi, Z., Nyberg, S., Jönsson, S., Raboisson, P., Stenkrona, P., Hooker, A.C., Karlsson, M.O., 2012. Non-linear mixed effects modelling of positron emission tomography data for simultaneous estimation of radioligand kinetics and occupancy in healthy volunteers. NeuroImage 61, 849–856. Karlsten, R., Malamut, R., Jonzon, B., Ståhle, L., Huizar, K., Argoff, C.E., 2012. Efficacy and safety of the mGluR5 antagonist AZD2066 in painful diabetes neuropathy (PDN): Results of a phase IIa randomised, double-blind, placebo-controlled study. Abstract of the 14th World Congress on Pain, Milan PT429. Lassen, N.A., Bartenstein, P.A., Lammertsma, A.A., Prevett, M.C., Turton, D.R., Luthra, S.K., Osman, S., Bloomfield, P.M., Jones, T., Patsalos, P.N., O'Connell, M.T., Duncan, J.S., Andersen, Vanggaard, 1995. Benzodiazepine receptor quantification in vivo in humans using [llC]flumazenil and PET: application of the steady-state principle. J. Cereb. Blood Flow Metab. 15, 152–165. Liefaard, L.C., Ploeger, B.A., Molthoff, C.F.M., Boellaard, R., Lammertsma, A.A., Meindert, Danhof, Voskuyl, R.A., 2005. Population pharmacokinetic analysis for simultaneous determination of Bmax and KD in vivo by positron emission tomography. Mol. Imaging Biol. 7, 411–421. Lim, K.S., Kwon, J.S., Jang, I.-J., Jeong, J.M., Lee, J.S., Kim, H.W., Kang, W.J., Kim, J.-R., Cho, J.-Y., Kim, E., Yoo, S.Y., Shin, S.-G., Yu, K.-S., 2007. Modeling of brain D2 receptor occupancyplasma concentration relationships with a novel antipsychotic, YKP1358, using serial PET scans in healthy volunteers. Clin. Pharmacol. Ther. 81 (2). Logan, J., Fowler, J.S., Volkow, N.D., Wolf, A.P., Dewey, S.L., Schlyer, D.J., MacGregor, R.R., Hitzemann, R., Bendriem, B., Gatley, S.J., et al., 1990. Graphical analysis of reversible radioligand binding from time-activity measurements applied to [N-11C-methyl](−)-cocaine PET studies in human subjects. J. Cereb. Blood Flow Metab. 10, 740–747. Olsson, H., Halldin, C., Farde, L., 2004. Differentiation of extrastriatal dopamine D2 receptor density and affinity in the human brain using PET. NeuroImage 22, 794–803. Patel, S., Hamill, T.G., Connolly, B., Jagoda, E., Li, W., Gibson, R.E., 2007. Species differences in mGluR5 binding sites in mammalian central nervous system determined using in vitro binding with [18F]F-PEB. Nucl. Med. Biol. 34, 1009–1017. Roland, P.E., Graufelds, C.J., Wåhlin, L., Ingelman, L., Andersson, M., Ledberg, A., Pedersen, J., Åkerman, S., Dabringhaus, A., Zilles, K., 1994. Human brain atlas: for high-resolution functional and anatomical mapping. Hum. Brain Mapp. 1, 173–184. Savic, R.M., Karlsson, M.O., 2009. Importance of shrinkage in empirical Bayes estimates for diagnostics: problems and solutions. AAPS J. 11 (3), 558–569. Spooren, W.P., Vassout, A., Neijt, H.C., Kuhn, R., Gasparini, F., Roux, S., Porsolt, R.D., Gentsch, C., 2000. Anxiolytic-like effects of the prototypical metabotropic glutamate receptor 5 antagonist 2-methyl-6-(phenylethynyl)pyridine in rodents. J. Pharmacol. Exp. Ther. 295 (3), 1267–1275. Ståhle, L., Karlsten, R., Jonzon, B., Eriksson, B., Kågedal, M., Dominicus, A., 2012. Safety evaluation of the mGluR5 antagonists AZD9272, AZD2066 and AZD 2516 in healthy volunteers and patients with neuropathic pain or major depressive disorder. Abstract of the 14th World Congress on Pain, Milan PT444. Syvanen, S., de Lange, E.C., Tagawa, Y., Schenke, M., Molthoff, C.F., Windhorst, A.D., Lammertsma, A.A., Voskuyl, R.A., 2011. Simultaneous in vivo measurements of receptor density and affinity using [11C]flumazenil and positron emission tomography: comparison of full saturation and steady state methods. NeuroImage 57, 928–937. Tatarczyńska, E., Klodzińska, A., Chojnacka-Wójcik, E., Palucha, A., Gasparini, F., Kuhn, R., Pilc, A., 2001. Potential anxiolytic- and antidepressant-like effects of MPEP, a potent, selective and systemically active mGlu5 receptor antagonist. Br. J. Pharmacol. 132 (7), 1423–1430. Treyer, V., Streffer, J., Wyss, M.T., Bettio, A., Ametamey, S.M., Fischer, U., Schmidt, M., Gasparini, F., Hock, C., Buck, A., 2007. Evaluation of the metabotropic glutamate receptor subtype 5 using PET and 11C-ABP688: assessment of methods. J. Nucl. Med. 48, 1207–1215. Varney, M.A., Gereau IV, R.W., 2002. Metabotropic glutamate receptor involvement in models of acute and persistent pain: prospects for the development of novel analgesics. Curr. Drug Targets CNS Neurol. Disord. 1 (3), 283–296. Varrone, A., Sjöholm, N., Eriksson, L., et al., 2009. Advancement in PET quantification using 3D-OP-OSEM point spread function reconstruction with the HRRT. Eur. J. Nucl. Med. Mol. Imaging 36, 1639–1650. Vong, C., Bergstrand, M., Nyberg, J., Karlsson, M.O., 2012. Rapid sample size calculations for a defined likelihood ratio test-based power in mixed-effects models. AAPS J. 14 (2), 176–186. Walker, K., Bowes, M., Panesar, M., Davis, A., Gentry, C., Kesingland, A., Gasparini, F., Spooren, W., Stoehr, N., Pagano, A., Flor, P.J., Vranesic, I., Lingenhoehl, K., Johnson, E.C., Varney, M., Urban, L., Kuhn, R., 2001. Metabotropic glutamate receptor subtype 5 (mGlu5) and nociceptive function. I. Selective blockade of mGlu5 receptors in models of acute, persistent and chronic pain. Neuropharmacology 40 (1), 1–9. Walker, K., Reeve, A., Bowes, M., Winter, J., Wotherspoon, G., Davis, A., Schmid, P., Gasparini, F., Kuhn, R., Urban, L., 2001. mGlu5 receptors and nociceptive function II. mGlu5 receptors functionally expressed on peripheral sensory neurones mediate inflammatory hyperalgesia. Neuropharmacology 40 (1), 10–19. Zamuner, S., Rabiner, E.A., Fernandes, S.A., Bani, M., Gunn, R.N., Gomeni, R., Ratti, E., Cunningham, V.J., 2012. A pharmacokinetic PET study of NK(1) receptor occupancy. Eur. J. Nucl. Med. Mol. Imaging 39, 226–235.