Simplified quantification of 5-HT2A receptors in the human brain with [11C]MDL 100,907 PET and non-invasive kinetic analyses

NeuroImage 50 (2010) 984–993

Contents lists available at ScienceDirect

NeuroImage j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / y n i m g

Simplified quantification of 5-HT2A receptors in the human brain with [11C]MDL 100,907 PET and non-invasive kinetic analyses Philipp T. Meyer a,⁎, Zubin Bhagwagar b, Philip J. Cowen c, Vincent J. Cunningham d, Paul M. Grasby e, Rainer Hinz f a

Department of Nuclear Medicine, University Hospital Freiburg, Hugstetter Strasse 55, 79106 Freiburg, Germany Bristol-Myers Squibb, Wallingford, and Yale University, USA Department of Psychiatry, University of Oxford, Oxford, UK d Clinical Imaging Centre, GlaxoSmithKline, Imperial College, London, UK e Medical Research Council Clinical Sciences Centre, London, UK f Wolfson Molecular Imaging Centre, University of Manchester, Manchester, UK b c

a r t i c l e

i n f o

Article history: Received 29 October 2009 Revised 8 January 2010 Accepted 9 January 2010 Available online 18 January 2010 Key words: [11C]MDL 100,907 5-HT2A receptor Positron emission tomography Tracer kinetic modelling Reference region

a b s t r a c t Background: [11C]MDL100,907 is a promising positron emission tomography (PET) ligand for 5-HT2A receptor quantification in vivo. Studies suggest that [11C]MDL100,907 PET may be quantified by non-invasive reference tissue analyses using cerebellum as reference region. We systematically investigated the validity of such analyses. Methods: Five healthy volunteers underwent [11C]MDL100,907 PET at baseline and after mirtazapine pre-treatment. Regional time–activity curves of 10 regions of interest (ROI) were analyzed for binding potential (BPND) and mirtazapine receptor occupancy (Occ) using: simplified reference tissue model (SRTM), multi-linear reference tissue model (MRTM), their two-parameter versions (SRTM2/MRTM2), noninvasive graphical analysis (NIGA) and a tissue activity concentration ratio. NIGA was also applied voxel-wise to generate BPND maps. These methods were compared with a two-tissue compartment model with arterial input function (2TCM) Results: SRTM and MRTM frequently failed to yield reliable results. SRTM2 and MRTM2 gave virtually identical estimates of BPND, which were highly correlated with 2TCM analyses (R2 ≥ 0.86) although with negative bias (− 29 ± 27% at baseline across all ROI). NIGA was less biased (− 19 ± 16%) and better correlated with 2TCM (R2 ≥ 0.93). Regarding Occ, NIGA and SRTM2/MRTM2 showed comparable mean biases (− 11 ± 27% vs. − 7 ± 47%) but correlation with 2TCM was higher for NIGA (R2 = 0.90 vs. 0.77). NIGA parametric maps (analysed using identical ROI) resulted in moderate bias in BPND (− 26 ± 22%; R2 ≥ 0.88) and Occ (−17 ± 36%; R2 = 0.78). Estimates obtained from tissue ratios performed least favourably. Conclusions: NIGA is well suited for analysis of [11C]MDL100,907 PET studies, yielding estimates of 5-HT2A receptor availability and changes that are highly correlated with results from invasive 2TCM analyses. This should greatly enhance the applicability of 5-HT2A receptor PET studies. © 2010 Elsevier Inc. All rights reserved.

Introduction The G-protein coupled serotonin receptor subtype 2A (5-HT2A receptor) modulates cortical GABAergic, glutamatergic and dopaminergic neurotransmission by various mechanisms and interactions. A proper balance of 5-HT2A receptor activity at inhibitory and excitatory neurons appears to be required for normal neuronal functioning (Leysen, 2004). Consequently, the 5-HT2A receptor has been implicated in various physiological functions and pathological conditions, including schizophrenia, major depression, anxiety and sleep disorders (Jones and Blackburn, 2002; Landolt and Wehrle, 2009). Positron emission tomography (PET) allows a quantification of 5-HT2A

⁎ Corresponding author. Fax: +49 761 2703930. E-mail address: [email protected] (P.T. Meyer). 1053-8119/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2010.01.037

receptor availability in the living human brain, and thus provides an invaluable technique for the investigation of the 5-HT2A receptor in different subject and patient populations and its interventional interference. Various 5-HT2A receptor ligands have been proposed as PET radiopharmaceutical for 5-HT2A receptor quantification, most notably [11C]N-methylspiperone ([11C]NMSP) (Wong et al., 1984), [18F]altanserin (Biver et al., 1994), [18F]setoperone (Blin et al., 1988) and [11C]MDL 100,907 (Lundkvist et al., 1996). Compared with the former, [11C]MDL 100,907 appears to be the most promising 5-HT2A radiopharmaceutical because of its high brain uptake with high specific-to-non-specific uptake contrast, prototypical 5-HT2A receptor selectivity and absence of blood–brain-barrier penetrating radiolabelled metabolites interfering with 5-HT2A receptor quantification. Several methods of analysis have been considered for [11C]MDL 100,907 PET quantification so far: The theoretically best validated approach is the use of a pharmacokinetic two-tissue compartment

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

model (2TCM) analysis with an arterial plasma input function, which yields significantly better fits to kinetic [11C]MDL 100,907 data than an one-tissue compartment model (1TCM), especially in regions with no or little 5-HT2A receptor binding (Ito et al., 1998; Watabe et al., 2000a; Hinz et al., 2007). The total volume of distribution (VT) and the binding potential relative to non-displaceable radioligand in tissue (BPND) (Innis et al., 2007) were proposed as outcome measures related to 5-HT2A receptor availability. The 2TCM has also been successfully used recently in praxi to demonstrate increased 5-HT2A receptor binding in patients recovered from depression (Bhagwagar et al., 2006). As a simplification, albeit still relying on arterial blood samples, various groups have also used Logan's graphical analyses (GA) (Logan et al., 1990) to estimate VT or BPND (Ito et al., 1998; Watabe et al., 2000a; Ichise et al., 2002; Kakiuchi et al., 2000). Although 2TCM and GA agree well, simulations studies suggest a negative VT bias in GA in presence of high noise levels (Watabe et al., 2000a; Ichise et al., 2002). This is in line with the well-known noisedependent bias of GA (Carson et al., 1993; Slifstein and Laruelle, 2000). In addition, however, Ichise et al. (2002) also noted a small negative VT bias (about 7%) at relatively low noise, which can be attributed to difficulties in identifying the linear part of the GA plot (its slope being equal to VT). This appears to be particularly problematic in the case of the slow kinetics of [11C]MDL 100,907, where the asymptomatically linear portion of the GA occurs at relatively late times (Ichise et al., 2002). The reliance of the kinetic analyses (1TCM, 2TCM and GA) on an arterial plasma input function represents a substantial drawback in regard of a broad application of this technique in basic and applied neurosciences. On one hand, arterial blood sampling and plasma analyses are invasive, labour and cost intensive. On the other, they can be a major source of error in quantitative PET analyses. Thus, the application of entirely non-invasive reference tissue analyses, which avoid the need for an arterial input function, is of great interest and the crucial question, whether there is a valid reference region devoid of detectable specific binding, has been a matter of debate. In their initial study, Lundkvist et al. (1996) found no effect of ketanserin preblocking or displacement on [11C]MDL 100,907 binding in the cerebellum of a cynomolgus monkey, which suggested that the cerebellum is a valid reference region. However, Watabe and colleagues noted that there is a small fraction of displaceable [11C]MDL 100,907 binding in cerebellum in preblocking experiments in rhesus monkeys (Watabe et al., 2000a). Hinz et al. (2007) performed preblocking experiments in humans using a clinical dose of the antidepressant drug mirtazapine (30 mg), which has a nanomolar affinity for 5-HT2A receptors (Anttila and Leinonen, 2001). They found a substantial and fairly homogenous 5-HT2A occupancy in all regions except the cerebellum, in which mirtazapine exhibited no significant effect on [11C]MDL 100,907 uptake. In line with this, the cerebellum was also found to be virtually devoid of 5-HT2A receptors on human post mortem autoradiography (Hall et al., 2000). However the use of reference tissue analyses for [11C]MDL 100,907 PET quantification has not been appropriately validated so far, although different reference tissue approaches have been used in [11C]MDL 100,907 studies in humans (Talvik-Lotfi et al., 2000; Turkheimer et al., 2003; Perani et al., 2008) and rats (Hirani et al., 2003). Although only in abstract form, the work of Watabe et al. (2000b) requires particular attention: They proposed a reference tissue model assuming a 1TCM and 2TCM configuration for the regions of interest and the reference region cerebellum, respectively. This model performed well on simulated [11C]MDL 100,907 data but no actual experimental data were given. Hinz and colleagues tested this model on their human [11C]MDL 100,907 data but no reliable parameter and occupancy estimates could be obtained (Hinz et al., 2007) probably because of too many free parameters in this particular reference tissue model configuration (two rate constants for the

985

target region and four rate constants in the reference region) and the very slow kinetics of [11C]MDL 100,907 in tissue. Thus, the purpose of the present study was to evaluate the use of standard reference tissue analyses for quantification of 5-HT2A receptor availability using [11C]MDL 100,907 PET. We re-analyzed the paired human [11C]MDL 100,907 PET studies with and without mirtazapine pre-treatment previously published by Hinz et al. (2007) to evaluate the use of the following widely used kinetic analysis methods: the simplified reference tissue model (Lammertsma and Hume, 1996), the two-step simplified reference tissue model (Wu and Carson, 2002), the multi-linear reference tissue model (MRTM) and the multi-linear reference tissue model 2 (MRTM2) (Ichise et al., 2003) and a non-invasive graphical analysis (NIGA) (Logan et al., 1996). In addition, we considered a simple late [11C]MDL 100,907 tissue activity concentration ratio as a surrogate marker of receptor availability and the use of NIGA for generation of quantitative 5-HT2A maps. Materials and methods Subjects In the present study we re-analyzed the data of a previous [11C] MDL 100,907 PET study which included arterial blood sampling and comprehensive metabolite analyses for input function measurement and subsequent tracer kinetic modelling in healthy normal volunteers (Hinz et al., 2007). Five normal volunteers (four males aged 37, 55, 62 and 63 years and one female of 59 years) were included after giving written informed consent and careful screening to exclude any relevant medical, psychiatric or neurological disease as previously described (Hinz et al., 2007). Each subject underwent two [11C]MDL 100,907 PET scans at the same time of the day with an interval 6 to 83 days between the two scans. For blocking experiments, subjects were given a single oral dose of mirtazapine 30 mg approximately 90 min before the start of the PET acquisition on one of the two occasions in a randomized open label design (the first scan was the blocking and baseline scan in two and three subjects, respectively). The scans were timed to coincide with the time of peak plasma concentration of mirtazapine (Timmer et al., 2000). All subjects had an indwelling venous cannula inserted into the median cubital vein for tracer injection and an indwelling arterial cannula in the radial artery for arterial blood sampling. The Research Ethics Committee of the Hammersmith Hospitals Trust and the Administration of Radioactive Substances Advisory Committee (ARSAC) of the United Kingdom approved the study. Data acquisition The PET acquisition, input function generation and volume of interest (VOI) analysis have been described in detail previously (Hinz et al., 2007). In brief: [11C]MDL 100,907 was synthesized according to the method previously described by Lundkvist et al. (1996). The injected radioactivity dose (given as a intra-venous slow bolus over 30 s) ranged between 337 and 379 MBq (mean ± standard deviation, 361 ± 14 MBq; specific activity: 65 ± 48 GBq/mmol, injected dose ranging between 0.021 to 0.293 nmol/kg body weight, i.e., tracer dose in all subjects). All PET scans were performed on the high-sensitivity Siemens/CTI scanner ECAT EXACT3D with an axial field of view of 23.4 cm and 95 reconstructed transaxial image planes (Spinks et al, 2000). A 5-min transmission scan using a 137Cs point source was performed for attenuation and scatter correction before each 94-min three-dimensional (3D) emission scan acquisition in list mode. The emission data was rebinned into a sequence of 28 time frames as follows: 30 s background frame before injection, then 3 × 10 s, 3 × 20 s, 3 × 30 s, 3 × 60 s, 4 × 120 s, 6 × 300 s and 5 × 600 s. Images were reconstructed using the reprojection algorithm proposed by Kinahan

986

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

and Rogers (1998) with ramp and Colsher filters set to Nyquist frequency, yielding a spatial resolution of 5.1 mm full-width at halfmaximum (FWHM) transaxially and 5.9 mm FWHM axially (Spinks et al, 2000). Arterial whole-blood activity was sampled continuously for the first 15 min with a bismuth germanate coincidence detector (Ranicar et al, 1991) and manually afterwards at 5, 10, 15, 20, 30, 40, 50, 60, 75 and 95 min. Metabolite analyses were performed in eight plasma samples per scan (at 5, 15, 30, 40, 50, 60, 75 and 95 min) and used for generation of arterial plasma input functions as previously described (Hinz et al., 2007). The following 11 VOI were chosen for generation of side-averaged regional time–activity curves (TACs) using the probabilistic brain atlas template of Hammers et al. (2002) and each individual's T1-weighted RF-spoiled gradient echo magnetic resonance imaging scan (1 T Philips Medical Systems HPQ+ Scanner): frontal lobe (575 cm3), anterior cingulate gyrus (24.3 cm3), insula (41.3 cm3), hippocampus (6.9 cm3), amygdala (4.8 cm3), anterior medial temporal lobe (20.2 cm3), parietal lobe (364 cm3), occipital lobe (155 cm3), putamen (11.9 cm3) and thalamus (21.4 cm3). The cerebellum (206 cm3) served as a reference region with negligible concentration of HT2A receptors. Kinetic analyses The standard five-parameter (K1 to k4 and fractional blood volume term) two-tissue compartment model (2TCM) relying on an arterial plasma input function measurement was previously validated as the most suited invasive model for analyzing [11C]MDL 100,907 PET studies (Hinz et al., 2007). Consequently, this model was used as the reference method for evaluation of the non-invasive reference tissue models contemplated in the present study. The following standard kinetic reference tissue models (as implemented in PMOD Version 3.0, PMOD Technologies Ltd., Adliswil, Switzerland) were employed for kinetic analyses of VOI-based regional [11C]MDL 100,907 PET data using the cerebellum TAC as reference region input (Hinz et al., 2007): (1) Simplified reference tissue model (SRTM) by Lammertsma and Hume (1996). This model estimates three parameters for each regional TAC by non-linear regression analysis: the binding potential (BPND; for parameter definitions and model assumption see below) of the ligand in the target region, the ratio of ligand delivery to the target and the reference region (R1) and the efflux rate constant of the ligand from the reference region (k2′). (2) Two-step simplified reference tissue model (SRTM2) by Wu and Carson (2002). This method takes advantage of the fact that k2′ should be equal for all regional analyses in the same subject. To apply this model, k2′ is first estimated by using the SRTM analyses on one or multiple TACs with low noise. The number of parameters of the SRTM is then reduced to two (BPND, R1) by fixing k2′ in all regions to the value estimated by the first step. The SRTM2 is particularly suited for noisy data. In the present study, k2′ was estimated by employing a low-noise volume-weighted average TAC of neocortical regions with high [11C]MDL 100,907 5-HT2A receptor binding (i.e., anterior medial temporal lobe, insula, anterior cingulate gyrus, occipital lobe, frontal lobe, parietal lobe). (3) Multi-linear reference tissue model (MRTM) by Ichise et al. (2003). The MRTM represents a linearized reference tissue model for ligands with reversible tissue kinetics. It provides estimates of three linear regression parameters (γ1 to γ3), which are used to calculate BPND, R1 and k2′. (4) Multilinear reference tissue model 2 (MRTM2) by Ichise et al. (2003). Following the same rational as the SRTM2, the number of parameters of the MRTM can be reduced to two by fixing k2′ in all regions to the individual's k2′ value gained from a preceding MRTM analysis of regions with low noise (volume-weighted neocortical TAC in the present study).

(5) Non-invasive Logan's graphical analysis (NIGA) (Logan et al., 1996). This is also a linearized reference tissue model for reversible ligands. It allows the estimation of the distribution volume ratio (DVR) of a ligand in a target region relative to a reference region, which is equal to BPND + 1. The DVR is given by the slope of the linear part of the NIGA plot. The NIGA operational equation requests that k2′ is provided in advance. As suggested by Logan et al. (1996), we used the population mean k2′ value, which was calculated from the individual rate constant estimates gained from 2TCM analyses of cerebellar TACs (Hinz et al., 2007) according to the following equation: k2′ = k2/ (1 + k3/k4) (Koeppe et al., 1991). This yielded a population mean k2′ value of 0.02 min− 1. BPND is proportional to the density of 5-HT2A receptors available for binding (Bavail) (Innis et al., 2007): BPND = fND


Bavail KD

ð1Þ

with fND and KD being the free fraction (i.e., unbound) of the tracer of non-displaceable tissue uptake at equilibrium and the dissociation constant, respectively. Consequently, the fraction of 5-HT2A receptor occupancy due to the mirtazapine dose was calculated as follows (Hinz et al., 2007): ! blocked BPND ð2Þ Occ = 1 − baseline BPND Occ of 0 corresponds to no occupancy, and a value of 1 is equal to full receptor blockade. The parameter R1 (given by SRTM, SRTM2, MRTM and MRTM2) is the ratio of the rate constant for tracer transfer from plasma to the target region (K1) to the rate constant for tracer transfer from plasma to the reference region (K1′) (if model assumptions are met, see below). All reference tissue methods assume that the equilibrium distribution volume of non-displaceable tissue uptake (VND) is the same across the reference region and the target regions. The distribution volume of specifically bound [11C]MDL 100,907 in tissue (VS) is then given by the difference between the total distribution volume (VT) in a target region and the VT of the cerebellum (=VND). This assumption was confirmed by our previous blocking studies using absolute quantification relative to the arterial plasma input function (Hinz et al., 2007). Furthermore, SRTM and SRTM2 rely on the assumption that the ligand kinetics in both the target region and the reference region can be adequately described by a one-tissue compartment model (1TCM). If this assumption is violated, these models may result in biased parameter estimates. In contrast, in this situation MRTM and MRTM2 should still provide accurate BPND estimates if the start time of linearization (t⁎) is set to an appropriate value (t⁎ = 0 in case of 1TCM) (Ichise et al., 2003). Like MRTM and MRTM2, NIGA does also not assume a particular compartment model structure (Logan et al., 1996), however, it requires the a priori choice of the clearance rate constant k2′ for the reference tissue, e.g. a population average value for k2′. The inclusion of this k2′-dependent term in the NIGA equation reduces the parameter bias in particular for slowly equilibrating tracers, however may result in additional variability of the DVR estimates (Holden et al., 2001). In addition to the above kinetic analyses, we also evaluated the use of a simple uptake ratio given by [11C]MDL 100,907 uptake in the target region to [11C]MDL 100,907 uptake in the reference region cerebellum during the final 30 min of the total scan duration (i.e., average target region/cerebellum uptake ratio of the final three 10-min frames). At true distribution equilibrium, this ratio would be equal to the distribution volume ratio (DVR) of the target region to the reference region (i.e., DVR = VT,target / VT,reference) and, thus, equal to BPND + 1 (assuming a valid reference region and homogenous VND; see above).

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

Although true equilibrium conditions are not reached after singlebolus injection of [11C]MDL 100,907, we employed target region/ cerebellum uptake ratio – 1 as an estimate of an apparent BPND. Finally, NIGA was also applied at the voxel-level to evaluate the use of this method for generation of parametric maps. Because of the high-resolution native image reconstruction settings (see above) that were used for the VOI analysis, all dynamic image data sets were smoothed using a 3D Gaussian filter with a kernel of 5 × 5 × 5 mm3 full width at half maximum (FWHM) prior to the generation of parametric images. For comparison with the 2TCM (reference method) and the VOI-based NIGA, parametric maps were read out using the aforementioned VOI sets. Statistics All results are expressed as mean ± standard deviation. Results of different analyses were compared by calculating the mean value of the parameter difference expressed as percent (ΔBPND, ΔR1, Δk2′ and ΔOcc) and by linear regression analyses. Differences between analyses were checked for statistical significance using a paired Student's t test, considering a p value b 0.05 significant (a correction for multiple comparisons was not performed since tests were planned a priori and given the limited number of subjects). The coefficient of variation (%CV; standard error of the estimate expressed as percentage of its mean) of individual model estimates was used as a measure of parameter precision.

987

Results Simplified reference tissue models and multilinear reference tissue models Representative TACs are shown in Figs. 1A and B under baseline conditions and after blockade with 30 mg mirtazapine, respectively. Note the pronounced effect of mirtazapine on [11C]MDL 100,907 uptake in cortical and subcortical target regions, while cerebellar uptake remained unaffected which underlines the validity of the cerebellum as a reference region (Hinz et al., 2007). SRTM and MRTM yielded reasonable fits to regional TAC in most instances. All data points were included into MRTM analyses (t⁎ = 0 min) since residuals did not reveal a systematic deviation. However, these models failed to converge or showed a %CV of individual BPND estimates N50% (i.e., considered unreliable) in 17 and 20 regions using SRTM and MRTM, respectively (100 regions in total, 5 subjects × 2 scans × 10 VOI). The majority of failed analyses occurred in the blocked state (16 each) and in small regions with relatively low specific binding and thus high noise level like amygdala, hippocampus, anterior medial temporal lobe, putamen and thalamus (14 and 15, respectively). Regional estimates of BPND from baseline studies are tabulated in Table 1. SRTM and MRTM gave virtually identical estimates of BPND with a mean difference (ΔBPND) of only 0.3 ± 1.8% (p = 0.24). Thus, for the sake of conciseness, only MRTM will be referred to further (all statements made about MRTM

Fig. 1. Representative regional [11C]MDL 100,907 PET time–activity curves, model fits and non-invasive graphical analyses. A and B, regional time–activity curves of occipital lobe, putamen, anterior medial temporal lobe (ant. temp. lobe) and cerebellum in subject number 1 before (A) and after (B) pre-treatment with mirtazapine. The solid lines give the results of fits by the multi-linear reference tissue model 2 (MRTM2). C and D, non-invasive graphical analyses (NIGA) of regional time–activity curves shown in A and B, using the cerebellum as reference region. The solid lines represent linear regression lines to the last 10 data points. CVOI(t) and CRef(t), time–activity curve of the volume of interest and the reference region (cerebellum), respectively; k′, 2 average constant for tracer efflux from the reference region.

988

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

Table 1 Regional binding potential estimates under baseline condition given by the 2TCM plasma input function model and the contemplated reference tissue analyses. Region

2TCM

SRTM

MRTM

SRTM2

MRTM2

NIGA VOI-based

NIGA voxel-wise

Ratio

Frontal lobe Anterior gyrus cinguli Insula Hippocampus Amygdala Ant. MTL Parietal lobe Occipital lobe Putamen Thalamus

1.68 ± 0.38 2.59 ± 0.75 1.47 ± 0.33 0.73 ± 0.33 0.84 ± 0.26 1.38 ± 0.41 1.55 ± 0.54 2.06 ± 0.23 0.40 ± 0.23 0.25 ± 0.14

1.07 ± 0.38 1.48 ± 0.52 1.05 ± 0.38 0.56 ± 0.33 0.57 ± 0.30 0.68 ± 0.43 1.00 ± 0.39 1.33 ± 0.28 0.39 ± 0.17 0.28 ± 0.11⁎

1.07 ± 0.38 1.48 ± 0.53 1.05 ± 0.38 0.52 ± 0.37 ⁎ 0.57 ± 0.29 0.52 ± 0.29⁎ 1.00 ± 0.39 1.33 ± 0.28 0.45 ± 0.14⁎ 0.28 ± 0.11⁎

1.07 ± 0.39 1.53 ± 0.62 0.93 ± 0.25 0.49 ± 0.18 0.51 ± 0.23 0.57 ± 0.25 0.99 ± 0.39 1.35 ± 0.30 0.40 ± 0.18 0.29 ± 0.10

1.07 ± 0.39 1.53 ± 0.62 0.93 ± 0.25 0.49 ± 0.18 0.51 ± 0.23 0.57 ± 0.26 0.99 ± 0.39 1.35 ± 0.30 0.40 ± 0.18 0.29 ± 0.10

1.33 ± 0.44 1.89 ± 0.63 1.17 ± 0.28 0.58 ± 0.29 0.63 ± 0.27 0.89 ± 0.34 1.22 ± 0.48 1.65 ± 0.36 0.38 ± 0.20 0.26 ± 0.12

1.19 ± 0.36 1.64 ± 0.50 1.08 ± 0.26 0.48 ± 0.19 0.51 ± 0.21 0.59 ± 0.19 1.17 ± 0.35 1.53 ± 0.29 0.41 ± 0.18 0.24 ± 0.08

0.84 ± 0.26 1.13 ± 0.32 0.77 ± 0.19 0.41 ± 0.17 0.40 ± 0.19 0.31 ± 0.18 0.78 ± 0.28 1.09 ± 0.21 0.39 ± 0.18 0.27 ± 0.11

Data are given as mean ± 1 standard deviation of regional binding potential (BPND) estimates (n = 5 normal subjects). For method abbreviations see text. Ant. MTL, anterior medial temporal lobe. ⁎n = 4 due to model failure.

also apply to SRTM): Compared to the 2TCM, estimates of BPND provided by MRTM show a considerable negative bias (ΔBPND) of − 24 ± 27% (p b 0.00001; − 30 ± 25% and − 15% ± 27% in baseline and blocked studies, respectively; note that there was no bias in the regions with lowest specific binding, i.e., putamen and thalamus). Furthermore, even in those regions with successful fits, the correlation between MRTM and the 2TCM (R2 = 0.86 and 0.73 for baseline and blocked studies, respectively) was lower than for the other analyses except for the ratio method under baseline condition (see below) (Fig. 2). Finally, the high failure rate in blocked studies enabled us to calculate the regional 5-HT2A receptor occupancy only in 27 of 50 regions. Occupancy given by MRTM and 2TCM showed

only a relatively weak correlation (R2 = 0.65; for SRTM: R2 = 0.49 for 32 regions). For implementation of SRTM2 and MRTM2, the efflux rate of [11C] MDL 100,907 from cerebellar tissue was estimated by fitting a lownoise volume-weighted average TAC of neocortical regions with high [11C]MDL 100,907 uptake and SRTM and MRTM. Both models yielded virtually identical estimates of k2′ (p = 0.48) which did not differ between the baseline and the blocked state (SRTM: baseline 0.030 ± 0.008 min− 1 vs. blocked 0.032 ± 0.015 min− 1; MRTM: baseline 0.030 ± 0.007 min− 1 vs. blocked 0.032 ± 0.013 min− 1, p = 0.88). These estimates of k2′ were higher than those given by the 2TCM (baseline, 0.020 ± 0.002 min− 1; blocked, 0.021 ± 0.002 min− 1)

Fig. 2. Linear regression analyses between the binding potential estimates given by volume of interest-based reference tissue analyses and the 2-tissue compartment model analysis. Results are given separately for baseline (open circles) and blocked (closed circles; pre-treatment with 30 mg mirtazapine) studies. Solid lines represent linear regression lines with regression parameters (± standard error) given in the upper left corner of each plot (a, regression slope; b, y-axis intercept). The broken line represents unity. For method abbreviations see text; BPND, binding potential.

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

989

Table 2 Regional occupancy estimates given by the 2TCM plasma input function model and the contemplated reference tissue analyses. Region

2TCM

SRTM2

MRTM2

NIGA VOI-based

NIGA voxel-wise

Ratio

Frontal lobe Anterior gyrus cinguli Insula Hippocampus Amygdala Ant. MTL Parietal lobe Occipital lobe Putamen Thalamus

0.69 ± 0.08 0.64 ± 0.09 0.56 ± 0.15 0.64 ± 0.13 0.68 ± 0.15 0.79 ± 0.12 0.67 ± 0.16 0.70 ± 0.03 0.27 ± 0.22 0.48 ± 0.24

0.60 ± 0.12 0.52 ± 0.12 0.48 ± 0.14 0.56 ± 0.17 0.58 ± 0.18⁎ 0.63 ± 0.12§ 0.62 ± 0.14 0.60 ± 0.09 0.17 ± 0.17 0.38 ± 0.12

0.60 ± 0.11 0.52 ± 0.12 0.48 ± 0.14 0.55 ± 0.17 0.59 ± 0.17⁎ 0.64 ± 0.12§ 0.62 ± 0.13 0.60 ± 0.09 0.17 ± 0.17 0.38 ± 0.12

0.62 ± 0.14 0.56 ± 0.11 0.52 ± 0.14 0.57 ± 0.23 0.66 ± 0.15 0.75 ± 0.14 0.62 ± 0.22 0.64 ± 0.08 0.22 ± 0.23 0.42 ± 0.31

0.57 ± 0.11 0.54 ± 0.11 0.51 ± 0.13 0.57 ± 0.14 0.60 ± 0.13 0.62 ± 0.11 0.59 ± 0.13 0.62 ± 0.07 0.29 ± 0.19 0.39 ± 0.21

0.54 ± 0.10 0.42 ± 0.13 0.40 ± 0.15 0.53 ± 0.19 0.61 ± 0.09 0.87 ± 0.35 0.57 ± 0.12 0.53 ± 0.06 0.21 ± 0.18 0.42 ± 0.27

Data are given as mean ± 1 standard deviation of regional occupancy values, calculated according to Eq. (2) (see text) from five paired [11C]MDL 100,907 PET studies with and without pre-treatment with 30 mg mirtazapine. For method abbreviations see text. Ant. MTL, anterior medial temporal lobe. ⁎n = 4 and §n = 3 due to model failure in the blocked condition.

(p = 0.014). Using these estimates, SRTM2 and MRTM2 provided reasonable fits to regional TAC as shown in Figs. 1A and B. Both models failed to converge in the same three regions (2× anterior medial temporal lobe and 1 × amygdala) in the blocked condition. Thus, SRTM2 and MRTM2 could be successfully applied to regional [11C] MDL 100,907 data in 97% of regions, yielding occupancy data in 94% of regions. Regional BPND and occupancy values given by the SRTM2 and MRTM2 are listed in Tables 1 and 2. Again, SRTM2 and MRTM2 provided virtually identical results (ΔBPND = − 0.1% ± 1.3%; ΔR1 = − 0.3 ± 1.0%; ΔOcc = 0.7 ± 2.4%; all p N 0.05), so that additional detailed results are given for MRTM2 only: Compared to the 2TCM, BPND estimates provided by MRTM2 suffered from a considerable negative bias of − 19 ± 33% across all studies (ΔBPND; p b 0.00001), which was higher in the baseline studies (−29 ± 27%) compared to the blocked studies (−9 ± 35%). Nevertheless, correlation between 2TCM and MRTM2 in terms of BPND was high (R2 = 0.89 and 0.86 under baseline and blocked conditions, respectively; Fig. 2B). Consequently, occupancy values yielded by MRTM2 are also negatively biased compared to 2TCM (ΔOcc = − 7 ± 47%; p b 0.00001), whereby overall correlation between 2TCM and MRTM2 in terms of Occ was moderate (R2 = 0.77; Fig. 3A). Finally, estimates of R1 obtained from by MRTM2 and those calculated from regional K1 values obtained from the 2TCM showed a very high correlation without a significant bias across all studies (linear regression: R 1 -MRTM2 = 0.89 ⁎ R 1 -2TCM + 0.13, R2 = 0.90; ΔR1 = 1.8 ± 6.2%; p = 0.06). Non-invasive graphical analysis Representative NIGA plots are shown in Fig. 1C (baseline study) and 1D (blocked study). Based on visual inspection of the residuals, linearization occurred after about 19 min (t⁎) so that regional BPND was estimated by using the last 10 PET frames for linear regression analyses. Regional estimates of BPND given by NIGA are tabulated in Table 1, and regional occupancy values in Table 2. Compared to 2TCM, NIGA estimates of BPND were significantly negatively biased by on average −14 ±17% across all studies (p b 0.00001). Again, this bias was larger in baseline studies (ΔBPND = −19 ±16%) than in blocked studies (ΔBPND =−9 ±17%). However, this bias was significantly smaller than the bias associated with MRTM2 analyses (ΔBPND = − 19 ± 33%; pb 0.00001). BPND values given by NIGA showed a very high correlation with those from 2TCM (R2 = 0.93 and 0.94 for baseline and blocked studies, respectively; Fig. 3C). Regarding occupancy, NIGA was negatively biased compared to 2TCM (ΔOcc = −11± 27%; pb 0.00004), but gave significantly higher Occ estimates than MRTM2 (p= 0.017; see Table 2). As shown in Fig. 3B, correlation with 2TCM was very high (R2 = 0.90) in terms of occupancy, comparing favourable with MRTM2.

the 2TCM of all analyses investigated (ΔBPND = −33 ± 32% across all studies; p b 0.00001; ΔBPND = −41 ± 28% and −25 ± 34% in baseline and blocked studies, respectively). Underestimation of the regional 5HT2A receptor occupancy was also pronounced (ΔOcc = − 19 ± 21%; p b 0.00001). The correlation between the ratio method and 2TCM was moderate to high in terms of BPND (Fig. 2D; R2 = 0.79 and 0.83 for baseline and blocked studies, respectively) and regional occupancy (Fig. 3C; R2 = 0.69). Of note, estimates of BPND and Occ (Tables 1 and 2) given by the ratio method were significantly smaller than those given by NIGA (p b 0.00001 and p b 0.003, respectively). Parametric Imaging Fig. 4 shows the parametric images of BPND of the subject whose regional TACs and NIGA plots are given in Fig. 1. BPND maps of comparable high visual quality could also be generated in all other subjects, and the pronounced effect of mirtazapine pre-treatment was clearly visible in all subjects. As shown in Fig. 4, few hot spot voxels with implausible high BPND estimates were visually apparent in blocked studies. Compared to 2TCM, voxel-wise estimates of BPND from NIGA were significantly negatively biased by −15 ± 27% across all studies (p b 0.00001; see Table 1 for regional baseline BPND values). However, this bias was primarily due to the difference in baseline studies (ΔBPND = − 26 ± 22%), while the average difference between voxel-based NIGA and 2TCM was small in blocked studies (ΔBPND = −4 ± 28%). Fig. 5A illustrates the correlation between voxel-wise NIGA and 2TCM in terms of BPND, which was high under baseline and blocked conditions (R2 = 0.88 and 0.89, respectively). Compared to the VOI-based NIGA, estimates of BPND provided by voxel-based NIGA were significantly lower (ΔBPND = − 1 ± 29% across all regions; p b 0.00001). Again, this can be attributed to the difference between voxel- and VOI-based NIGA in baseline studies (ΔBPND = − 9 ± 13%), whereas the average difference between voxel- and VOI-based NIGA was actually positive although fairly variable (ΔBPND = 6 ± 37%) in blocked studies. Nevertheless, Fig. 5B depicts the very high correlation between voxel- and VOI-based BPND estimates (R2 = 0.96 for baseline and blocked studies). Regional occupancy values gained by voxel-wise NIGA are listed in Table 2. These were significantly lower than those given by the 2TCM (ΔOcc = −17 ± 36%; p b 0.00001). Correlation between both methods in terms of occupancy was moderate (R2 = 0.78; Fig. 5C) and less strong than the correlation between VOI-based NIGA and 2TCM (R2 = 0.90; Fig. 3C). The difference in regional occupancy estimates between VOI- and voxel-based NIGA (ΔOcc = − 7 ± 34%) was significant (p b 0.02). Discussion

Ratio method Apparent estimates of BPND derived from late scan time tissue activity concentration ratios showed the largest bias in comparison to

The 5-HT2A receptor antagonist [11C]MDL 100,907 holds great promise as a PET radiopharmaceutical for in vivo 5-HT2A receptor quantification. It binds to 5-HT2A receptor with high affinity

990

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

Fig. 4. Parametric binding potential maps generated by voxel-wise non-invasive graphical analyses. Upper row, baseline study; lower row, blocked study after pretreatment with 30 mg mirtazapine. PET studies stem from subject number 1, whose regional data is given in Fig. 1. Identical colour scaling was used for both images (see below; BPND, binding potential).

Fig. 3. Linear regression analyses between the fractional occupancy estimates given by volume of interest-based reference tissue analyses and the 2-tissue compartment model analyses. Solid lines represent linear regression lines with regression parameters (± standard error) given in the upper left corner of each plot (a, regression slope; b, yaxis intercept). The broken line represents unity. For method abbreviations see text; Occ, fractional occupancy, calculated according to Eq. (2) (see text).

(Ki = 0.2 nM) und high selectivity, rendering [11C]MDL 100,907 a prototypical 5-HT2A receptor antagonist. In contrast, other 5-HT2A PET tracers like [11C]NMSP and [18F]setoperone also exhibit significant binding to dopamine receptors (Lyon et al., 1986; Blin et al., 1988). Furthermore, [11C]MDL 100,907 quantification is not hampered by blood-brain-barrier penetrating radiolabelled metabolites as is the case for the other frequently used 5-HT2A receptor ligand [18F] altanserin. The resulting difficulties in 5-HT2A receptor quantification

with [18F]altanserin have been successfully solved by using a bolus/ infusion [18F]altanserin administration protocol (Pinborg et al., 2003). However, such a protocol relies on a prolonged [18F]altanserin infusion and plasma metabolite analyses which can be fairly complex and error-prone. In contrast, [11C]MDL 100,907 PET studies should be amendable to reference tissue analyses with the cerebellum serving as a reference region. Thus, we evaluated reference tissue analyses for quantifying single-bolus injection [11C]MDL 100,907 PET studies to take full advantage of the favourable properties of [11C]MDL 100,907. Aside from two methodological studies primarily relying on simulated [11C]MDL 100,907 PET data (Watabe et al., 2000b; Turkheimer et al., 2003), very few groups have employed noninvasive reference tissue analyses so far. Talvik-Lotfi and colleagues (2000) estimated frontal cortex 5-HT2A receptor occupancy in two patients with schizophrenia who underwent [11C]MDL 100,907 PET in the course of pharmacological doses of MDL 100,907. Both a simple uptake ratio of specific frontal uptake to non-specific cerebellar uptake (interval 42–63 min) and the SRTM, indicated high 5-HT2A receptor occupancy (86–92%) (Talvik-Lotfi et al., 2000). Recently, Perani et al. (2008) also used the SRTM to demonstrate a wide-spread reduction in 5-HT2A receptor availability in obsessive-compulsive disorder patients compared to healthy controls. Finally, in an [11C] MDL 100,907 PET study in rats (Hirani et al., 2003), cold MDL 100,907 and ketanserin pre-treatment greatly reduced the apparent 5-HT2A receptor availability as given by an uptake ratio (20–60 min; cerebellum as reference), while the serotonin releasing agent fenfluramine had no effect. The plausibility of the aforementioned results strongly suggest that non-invasive reference tissue analyses can provide meaningful BPND estimates in case of [11C]MDL 100,907. Our results indicate that estimates of BPND gained from SRTM and MRTM analyses are subject to a considerable negative bias of about −30% in our baseline studies. This can be attributed to the underlying assumption of the SRTM of a 1TCM for the target and the reference region, which is not fulfilled for the reference region and for target regions with low 5-HT2A receptor density (Ito et al., 1998; Watabe et al., 2000a; Hinz et al., 2007). Theoretically, MRTM should be validly applicable also to a non-1TCM situation if t⁎ is selected appropriately late (Ichise et al., 2003). However, inspection of residuals revealed no systematic deviations (see MRTM2 fits in Figs. 1A and B) and choosing a later t⁎ did not diminish the aforementioned bias but increased BPND variability (data not shown). Consequently, with t⁎ set to 0, MRTM implies a 1TCM situation and gave very similar results to SRTM as

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

Fig. 5. Linear regression analyses: voxel-wise non-invasive graphical analysis. (A) Linear regression between binding potential (BPND) estimates given by voxel-wise noninvasive graphical analysis (NIGA) and volume of interest (VOI)-based two-tissue compartment model (2TCM) analyses. (B) Linear regression between BPND estimates given by voxel-wise and VOI-based NIGA. (C) Linear regression between fractional occupancy values gained from voxel-wise NIGA and VOI-based 2TCM analyses. Solid lines represent linear regression lines with regression parameters (± standard error) given in the upper left corner of each plot (a, regression slope; b, y-axis intercept). The broken line represents unity.

previously noted for other ligands (Ichise et al., 2003; Meyer et al., 2008). Interestingly, the magnitude of bias detected in the present study based on measured data perfectly agrees with the bias predicted by simulation for the SRTM (− 29%) (Watabe et al., 2000b). In that

991

simulation study, Watabe et al. (2000b) demonstrated that a reference tissue model with two tissue compartments for the reference regions should effectively reduce the BPND bias to about −5%, but at the cost of higher noise. In fact, the later model did not converge or yielded only unreliable parameter estimates in the majority of regions when applied to actually measured [11C]MDL 100,907 data and was therefore not pursued further (Hinz et al., 2007). Another drawback of SRTM and MRTM analyses is relatively high susceptibility to noise, resulting in frequent non-converging fits or unreliable BPND estimates in small regions with low binding and after mirtazapine pre-treatment. Therefore, we also evaluated SRTM2 and MRTM2 for [11C]MDL 100,907 analyses, which have not been previously employed for this purpose. A low-noise volume-weighted average TAC of neocortical regions was used for k2′ estimation, which did not differ between both approaches and states. However, the non-invasive k′ estimates from the SRTM and MRTM were about 50% higher than those obtained from the calculation k2′ = k2/(1 + k3/k4) using the estimates from the 2TCM analyses with plasma input function (0.03 min− 1 vs. 0.02 min− 1; p = 0.014). Biases in SRTM parameter estimates have been previously reported in similar cases when the reference region is best described by a two-tissue compartment model rather than a one-tissue compartment model, for example for [18F]FCWAY (Wu and Carson, 2002) and [11C]WAY-100635 (Slifstein et al., 2000). Furthermore, parameter bias in reference tissue model estimates has been reported because, in contrast to the original plasma input function analysis (Hinz et al., 2007), the contribution of the activity in the vasculature to the tissue activity is not accounted for (Gunn et al., 1988; Tomasi et al., 2008). Therefore the use of k2′ estimated with the SRTM or MRTM for the correction in NIGA cannot be recommended. Except for three regions in blocked studies, MRTM2 and SRTM2 yielded reliable and quasi-identical parameter estimates. Thus, correlation with the reference method 2TCM was higher for SRTM2 and MRTM2 than for SRTM and MRTM, particularly after mirtazapine pre-treatment (Figs. 2A and B), and enabled use to calculate regional 5-HT2A receptor occupancy in 47 of 50 regions. The later correlated fairly well (R2 = 0.77) with occupancy rates estimated invasively by 2TCM analyses, which suggests that SRTM2 and MRTM2 can be used to assess 5-HT2A receptor changes despite the still considerable negative BPND bias (−29 ± 27% and −9 ± 35% for baseline and blocked studies, respectively). In contrast, the BPND bias associated with NIGA was considerably smaller and less variable than for SRTM2 and MRTM2, particularly in baseline studies (−19 ± 16%; blocked studies −9% ± 17%). This led to a very high correlation with regression slopes closer to unity between NIGA and 2TCM in terms of BPND (R2 N 0.93) and 5-HT2A receptor occupancy (R2 = 0.90). As NIGA does not assume a specific compartmental model structure and only requires reversibility of the binding, the observed bias stems from other sources than discussed for SRTM/ 2 and MRTM/2. As mentioned before for the invasive GA (Ichise et al., 2002), the slow kinetics of [11C]MDL 100,907 also represent a challenge for correct identification of the asymptomatically linear part of the NIGA plot. By visual inspection of individual NIGA plots and the residuals, we found that the linear part of the NIGA plot is reached after about t⁎ = 19 min (i.e., final 10 PET frames). This impression is corroborated by systematically analyzing the low-noise volumeweighted average TACs of neocortical regions with high [11C]MDL 100,907 uptake in which t⁎ is largest. Under baseline condition, linearization occurred after 12 min (n = 2), 14 min, 19 min and 24 min (n = 1 each). As would be expected, linearization was reached considerably earlier under 5-HT2A receptor blockade (between 4 min and 10 min in all subjects). Thus, using a fixed t⁎ = 19 min for all regions and both conditions appears to be reasonable, especially if voxel-wise analyses are to be performed. The well known noise-dependent negative bias is a limitation of NIGA. Simulation studies on [11C]MDL 100,907 by Turkheimer et al.

992

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993

(2003) suggest that NIGA is negatively biased by about 5% even in presence of little noise (5%) and despite rather long scan duration of 120 min. Consequently, we tested if the NIGA bias can be reduced by choosing a later t⁎ of 44 min. The resulting mean bias was comparable to the NIGA with t⁎ = 19 min, albeit much more variable (−14 ± 28% vs. −14 ± 17%). This implies that using a later t⁎ is unlikely to reduce the bias with the present scan duration. However, the higher variability also suggests that the NIGA with t⁎ = 44 min (i.e., more than one carbon-11 half-life later, only 5 instead of 10 data points) is actually affected by a stronger noise-dependent bias which most likely just balances with the improvement in bias due to the later t⁎. Nevertheless, the benefit of increasing the scan duration is questionable given the half-life of carbon-11 and increasing subject discomfort. The selection of the population average value for k2′ is another factor that may cause a negative bias in NIGA (Holden et al., 2001). In the present study, we chose the average value of k2′ gained from 2TCM analyses (Hinz et al., 2007) as the best validated k2′ available. These 2TCM k2′ estimates showed a fairly low variability among subjects and was not affected by 5-HT2A blockade (baseline k2′ = 0.020 ± 0.002 min− 1 [range: 0.016 to 0.023 min− 1]; blockade k2′ = 0.021 ± 0.002 min− 1 [range: 0.017 to 0.023 min− 1]; p = 0.78), as would be expected in case of a valid reference region. Given the structure of the NIGA operational equation (average k2′ being a denominator in the enumerator of the x-axis term; see Fig. 1), a small average k2′ has major impact on the GA slope and thus DVR estimate. Albeit deviations of one individual's k2′ from the assumed average k2′ usually have only a small impact on NIGA accuracy (Logan et al., 1996), this needs proper consideration if different patient populations are contemplated. Population differences in k2′ as a possible source of error may be ruled out by invasive 2TCM analyses in a representative subset of patients or, as a pragmatic albeit less accurate choice, by a direct comparison of cerebellar TAC. However, using k2′ estimates provided by MRTM (or SRTM) for NIGA (NIGA-k2′MRTM2) does not allow an accurate BPND estimation, since k2′ estimates from MRTM (SRTM) analyses are biased due to the violation of underlying model assumptions (for MRTM2: baseline k2′ = 0.030 ± 0.007 min− 1, blockade k2′ = 0.032 ± 0.013 min− 1). The magnitude of the resulting error of NIGA-k2′MRTM2 is comparable to the bias of MRTM2 itself, although the bias variability is slightly smaller (− 21 ± 25% vs. − 19 ± 33%; across all studies). This translates into a slightly better correlation of NIGA-k2′MRTM2 than MRTM2 with 2TCM in terms of 5-HT2A receptor occupancy (R2 = 0.83 vs. 0.77). Of note, k2′ estimates provided by MRTM (SRTM) not only show a higher variability then those given by 2TCM (see above) but also a weak inverse association (not significant for n = 5) with BPND of the volume-weighted average neocortical VOI used for k2′ estimation (Pearson’s correlation coefficient r = − 0.59 and − 0.35 under baseline and blocked conditions, respectively). Because of the superior performance of NIGA on VOI-based data, we also explored this method for generation of parametric BPND maps as prerequisite for voxel-based statistical analyses. Compared to its VOI-based counterpart, voxel-based NIGA is subject to a slightly higher BPND bias (−26 ± 22% and −9 ± 13% compared to 2TCM and VOI-based NIGA, respectively, for baseline studies), which was expected because of the higher data noise on voxel-level. This agrees well with a bias of −22% predicted by simulations for a high noise level of 20% and a scan duration of 120 min (Turkheimer et al., 2003). The positive bias for blocked studies (+ 9 ± 44% compared to 2 TCM) is somewhat surprising but can be explained by the lower limit of possible values BPND being set to zero. With BPND being fairly variable and close to zero in the blocked state (note the hot spots in Fig. 4), the mean bias will tend towards positive values, if negative values are truncated. Nevertheless, voxel-based NIGA shows a high correlation with 2TCM in terms of BPND (R2 N 0.88) and regional

occupancy (R2 = 0.78), which clearly indicates that voxel-wise NIGA is a versatile tool for voxel-based investigations of changes in 5-HT2A receptor availability across populations and conditions. Because of the high-resolution but noise-permissive reconstruction filter settings (ramp and Colsher filters set to Nyquist frequency) we used a Gaussian filter (5 mm FWHM) for noise reduction before applying NIGA. This only slightly decreases spatial resolution (from to about 5.1 mm to 7.1 mm FWHM transaxially), but has been shown to effectively decrease the noise-associated bias in graphical analyses (Meyer et al., 2006). Particularly in the setting of voxel-based statistical analyses, this does not represent a drawback since even higher spatial smoothing is commonly required. Alternatively, numerous other procedures have been proposed for bias reduction in graphical analyses (Logan et al., 2001; Cselényi et al., 2002; Varga and Szabo, 2002; Parsey et al., 2003; Zhou et al., 2009), whose exploration for use in case of [11C]MDL 100,907 is beyond the scope of this report. Of note, however, Turkheimer et al. (2003) demonstrated by their simulations and one preliminary patient study that rank-shaping regularization of exponential spectral analyses may yield greatly improved parametric images (bias of only − 8% at 20% noise). Finally, we also explored the use of a simple VOI to cerebellum uptake ratio (minus one) as an estimate of apparent BPND. Lundkvist et al. (1996) reported that specific [11C]MDL 100,907 binding peaks between 40 and 60 min in anaesthetized cynomolgus monkeys, which supports the use of a peak-equilibrium analysis (Farde et al., 1986). This approach was followed in human [11C]MDL 100,907 PET studies (Talvik-Lofti et al., 2000). However, we found that specific cortical binding and, consequently, the cortex to cerebellum ratio increase continuously. That is, neither a peak- nor a pseudo-equilibrium is attained until the end of our baseline studies. In line with this, the tissue activity concentration ratio considerably underestimated the true BPND (−41 ± 28% in baseline studies) although BPND (R2 N 0.79) and occupancy (R2 N 0.83) values of the ratio method correlated fairly well with those given by 2TCM analyses. Therefore, the ratio method may represent a pragmatic simplified method for selected applications, e.g. in demented patients or acute psychiatric patients, in which scan durations should be as short as possible. However, if done so, one has to carefully consider that all factors governing [11C]MDL 100,907 uptake and retention in tissue (i.e., cerebral blood flow, receptor availability and peripheral tracer metabolism) (Carson et al., 1993) will interfere with the ratio method in a hardly controllable manner. In conclusion, the present study demonstrates that reference tissue kinetic analyses can be employed for non-invasive measurement of 5-HT2A receptor binding in the living human brain. NIGA is the method of choice, yielding estimates of 5-HT2A receptor availability and its changes that are very highly correlated with results gained from invasive 2TCM analyses. Furthermore, NIGA can easily be applied on a voxel-level for generation of quantitative 5HT2A receptor maps. This should greatly enhance the applicability of 5-HT2A receptor PET studies in basic and applied neurosciences.

References Anttila, S.A., Leinonen, E.V., 2001. A review of the pharmacological and clinical profile of mirtazapine. CNS Drug Rev. 7, 249–264. Bhagwagar, Z., Hinz, R., Taylor, M., Fancy, S., Cowen, P., Grasby, P., 2006. Increased 5-HT (2A) receptor binding in euthymic, medication-free patients recovered from depression: a positron emission study with [(11)C]MDL 100,907. Am. J. Psychiatry 163, 1580–1587. Biver, F., Goldman, S., Luxen, A., Monclus, M., Forestini, M., Mendlewicz, J., Lotstra, F., 1994. Multicompartmental study of fluorine-18 altanserin binding to brain 5HT2 receptors in humans using positron emission tomography. Eur. J. Nucl. Med. 21, 937–946. Blin, J., Pappata, S., Kiyosawa, M., Crouzel, C., Baron, J.C., 1988. [18F]setoperone: a new high-affinity ligand for positron emission tomography study of the serotonin-2 receptors in baboon brain in vivo. Eur. J. Pharmacol. 147, 73–82. Carson, R.E., Channing, M.A., Blasberg, R.G., Dunn, B.B., Cohen, R.M., Rice, K.C., Herscovitch, P., 1993. Comparison of bolus and infusion methods for receptor

P.T. Meyer et al. / NeuroImage 50 (2010) 984–993 quantitation: application to [18F]cyclofoxy and positron emission tomography. J. Cereb. Blood Flow. Metab. 13, 24–42. Cselényi, Z., Olsson, H., Farde, L., Gulyás, B., 2002. Wavelet-aided parametric mapping of cerebral dopamine D2 receptors using the high affinity PET radioligand [11C]FLB 457. Neuroimage 17, 47–60. Farde, L., Hall, H., Ehrin, E., Sedvall, G., 1986. Quantitative analysis of D2 dopamine receptor binding in the living human brain by PET. Science 231, 258–261. Gunn, R.N., Sargent, P.A., Bench, C.J., Rabiner, E.A., Osman, S., Pike, V.W., Hume, S.P., Grasby, P.M., Lammertsma, A.A., 1988. Tracer kinetic modeling of the 5-HT1A receptor ligand [carbonyl-11C]WAY-100635 for PET. Neuroimage 8, 426–440. Hall, H., Farde, L., Halldin, C., Lundkvist, C., Sedvall, G., 2000. Autoradiographic localization of 5-HT(2A) receptors in the human brain using [(3)H]M100907 and [(11)C]M100907. Synapse 38, 421–431. Hammers, A., Koepp, M.J., Free, S.L., Brett, M., Richardson, M.P., Labbé, C., Cunningham, V.J., Brooks, D.J., Duncan, J., 2002. Implementation and application of a brain template for multiple volumes of interest. Hum. Brain Mapp. 15, 165–174. Hinz, R., Bhagwagar, Z., Cowen, P.J., Cunningham, V.J., Grasby, P.M., 2007. Validation of a tracer kinetic model for the quantification of 5-HT(2A) receptors in human brain with [(11)C]MDL 100,907. J. Cereb. Blood Flow. Metab. 27, 161–172. Hirani, E., Sharp, T., Sprakes, M., Grasby, P., Hume, S., 2003. Fenfluramine evokes 5HT2A receptor-mediated responses but does not displace [11C]MDL 100907: small animal PET and gene expression studies. Synapse 50, 251–260. Holden, J.E., Sossi, V., Chan, G., Doudet, D.J., Stoessl, A.J., Ruth, T.J., 2001. Effect of population k2 values in graphical estimation of DV ratios of reversible ligands. In: Gjedde, A., Hansen, S.B., Knudsen, G.M., Paulson, O.B. (Eds.), Physiological imaging of the brain with PET. Academic Press, Inc., San Diego, pp. 127–129. Ichise, M., Toyama, H., Innis, R.B., Carson, R.E., 2002. Strategies to improve neuroreceptor parameter estimation by linear regression analysis. J. Cereb. Blood Flow. Metab. 22, 1271–1281. Ichise, M., Liow, J.S., Lu, J.Q., Takano, A., Model, K., Toyama, H., Suhara, T., Suzuki, K., Innis, R.B., Carson, R.E., 2003. Linearized reference tissue parametric imaging methods: application to [11C]DASB positron emission tomography studies of the serotonin transporter in human brain. J. Cereb. Blood Flow. Metab. 23, 1096–1112. 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. Ito, H., Nyberg, S., Halldin, C., Lundkvist, C., Farde, L., 1998. PET imaging of central 5HT2A receptors with carbon-11-MDL 100,907. J. Nucl. Med. 39, 208–214. Jones, B.J., Blackburn, T.P., 2002. The medical benefit of 5-HT research. Pharmacol. Biochem. Behav. 71, 555–568. Kakiuchi, T., Nishiyama, S., Sato, K., Ohba, H., Nakanishi, S., Tsukada, H., 2000. Agerelated reduction of [11C]MDL100,907 binding to central 5-HT(2A) receptors: PET study in the conscious monkey brain. Brain Res. 883, 135–142. Kinahan, P.E., Rogers, J.G., 1998. Analytic 3D image reconstruction using all detected events. IEEE Trans. Nucl. Sci. 36, 964–968. Koeppe, R.A., Holthoff, V.A., Frey, K.A., Kilbourn, M.R., Kuhl, D.E., 1991. Compartmental analysis of [11C]flumazenil kinetics for the estimation of ligand transport rate and receptor distribution using positron emission tomography. J. Cereb. Blood Flow. Metab. 11, 735–744. Lammertsma, A.A., Hume, S.P., 1996. Simplified reference tissue model for PET receptor studies. Neuroimage 4, 153–158. Landolt, H.P., Wehrle, R., 2009. Antagonism of serotonergic 5-HT2A/2C receptors: mutual improvement of sleep, cognition and mood. Eur. J. Neurosci. 29, 1795–1809. Leysen, J.E., 2004. 5-HT2 receptors. Curr. Drug. Targets CNS Neurol. Disord. 3, 11–26. 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., Christman, D.R., 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. Logan, J., Fowler, J.S., Volkow, N.D., Wang, G.J., Ding, Y.S., Alexoff, D.L., 1996. Distribution volume ratios without blood sampling from graphical analysis of PET data. J. Cereb. Blood Flow. Metab. 16, 834–840. Logan, J., Fowler, J.S., Volkow, N.D., Ding, Y.S., Wang, G.J., Alexoff, D.L., 2001. A strategy for removing the bias in the graphical analysis method. J. Cereb. Blood Flow. Metab. 21, 307–320.

993

Lundkvist, C., Halldin, C., Ginovart, N., Nyberg, S., Swahn, C.G., Carr, A.A., Brunner, F., Farde, L., 1996. [11C]MDL 100907, a radioligand for selective imaging of 5-HT(2A) receptors with positron emission tomography. Life Sci. 58, 187–192. Lyon, R.A., Titeler, M., Frost, J.J., Whitehouse, P.J., Wong, D.F., Wagner Jr., H.N., Dannals, R.F., Links, J.M., Kuhar, M.J., 1986. 3H-3-N-methylspiperone labels D2 dopamine receptors in basal ganglia and S2 serotonin receptors in cerebral cortex. J. Neurosci. 6, 2941–2949. Meyer, P.T., Elmenhorst, D., Matusch, A., Winz, O., Zilles, K., Bauer, A., 2006. 18F-CPFPX PET: on the generation of parametric images and the effect of scan duration. J. Nucl. Med. 47, 200–207. Meyer, P.T., Sattler, B., Winz, O.H., Fundke, R., Oehlwein, C., Kendziorra, K., Hesse, S., Schaefer, W.M., Sabri, O., 2008. Kinetic analyses of [123I]IBZM SPECT for quantification of striatal dopamine D2 receptor binding: a critical evaluation of the single-scan approach. Neuroimage 15, 548–558. Parsey, R.V., Ogden, R.T., Mann, J.J., 2003. Determination of volume of distribution using likelihood estimation in graphical analysis: elimination of estimation bias. J. Cereb. Blood Flow. Metab. 23, 1471–1478. Perani, D., Garibotto, V., Gorini, A., Moresco, R.M., Henin, M., Panzacchi, A., Matarrese, M., Carpinelli, A., Bellodi, L., Fazio, F., 2008. In vivo PET study of 5HT(2A) serotonin and D(2) dopamine dysfunction in drug-naive obsessive–compulsive disorder. Neuroimage 42, 306–314. Pinborg, L.H., Adams, K.H., Svarer, C., Holm, S., Hasselbalch, S.G., Haugbøl, S., Madsen, J., Knudsen, G.M., 2003. Quantification of 5-HT2A receptors in the human brain using [18F]altanserin-PET and the bolus/infusion approach. J. Cereb. Blood Flow. Metab. 23, 985–996. Ranicar, A.S., Williams, C.W., Schnorr, L., Clark, J.C., Rhodes, C.G., Bloomfield, P.M., Jones, T., 1991. The on-line monitoring of continuously withdrawn arterial blood during PET studies using a single BGO/photomultiplier assembly and non-stick tubing. Med. Prog. Technol. 17, 259–264. Slifstein, M., Laruelle, M., 2000. Effects of statistical noise on graphic analysis of PET neuroreceptor studies. J. Nucl. Med. 41, 2083–2088. Slifstein, M., Parsey, R.V., Laruelle, M., 2000. Derivation of [11C]WAY-100635 binding parameters with reference tissue models: effect of violations of model assumptions. Nucl. Med. Biol. 27, 487–492. Spinks, T.J., Jones, T., Bloomfield, P.M., Bailey, D.L., Miller, M., Hogg, D., Jones, W.F., Vaigneur, K., Reed, J., Young, J., Newport, D., Moyers, C., Casey, M.E., Nutt, R., 2000. Physical characteristics of the ECAT EXACT3D positron tomograph. Phys. Med. Biol. 45, 2601–2618. Talvik-Lotfi, M., Nyberg, S., Nordström, A.L., Ito, H., Halldin, C., Brunner, F., Farde, L., 2000. High 5HT2A receptor occupancy in M100907-treated schizophrenic patients. Psychopharmacology (Berl) 148, 400–403. Timmer, C.J., Sitsen, J.M., Delbressine, L.P., 2000. Clinical pharmacokinetics of mirtazapine. Clin. Pharmacokinet. 38, 461–474. Tomasi, G., Edison, P., Bertoldo, A., Roncaroli, F., Singh, P., Gerhard, A., Cobelli, C., Brooks, D.J., Turkheimer, F.E., 2008. Novel reference region model reveals increased microglial and reduced vascular binding of 11C-(R)-PK11195 in patients with Alzheimer's disease. J. Nucl. Med. 49, 1249–1256. Turkheimer, F.E., Hinz, R., Gunn, R.N., Aston, J.A., Gunn, S.R., Cunningham, V.J., 2003. Rank-shaping regularization of exponential spectral analysis for application to functional parametric mapping. Phys. Med. Biol. 48, 3819–3841. Varga, J., Szabo, Z., 2002. Modified regression model for the Logan plot. J. Cereb. Blood Flow. Metab. 22, 240–244. Watabe, H., Channing, M.A., Der, M.G., Adams, H.R., Jagoda, E., Herscovitch, P., Eckelman, W.C., Carson, R.E., 2000a. Kinetic analysis of the 5-HT2A ligand [11C]MDL 100,907. J. Cereb. Blood Flow. Metab. 20, 899–909. Watabe, H., Carson, R.E., Iida, H., 2000b. The reference tissue model: three compartments for the reference region. Neuroimage 11, S12. Wong, D.F., Wagner Jr., H.N., Dannals, R.F., Links, J.M., Frost, J.J., Ravert, H.T., Wilson, A.A., Rosenbaum, A.E., Gjedde, A., Douglass, K.H., Petronis, J.D., Folstein, M.F., Toung, J.K.T., Burns, H.D., Kuhar, M.J., 1984. Effects of age on dopamine and serotonin receptors measured by positron tomography in the living human brain. Science 226, 1393–1396. Wu, Y., Carson, R.E., 2002. Noise reduction in the simplified reference tissue model for neuroreceptor functional imaging. J. Cereb. Blood Flow. Metab. 22, 1440–1452. Zhou, Y., Ye, W., Brasić, J.R., Crabb, A.H., Hilton, J., Wong, D.F., 2009. A consistent and efficient graphical analysis method to improve the quantification of reversible tracer binding in radioligand receptor dynamic PET studies. Neuroimage 44, 661–670.